Modeling and Parametric Study for CO2CH4 Separation using Membrane Processes

合集下载

基于表达式树的Modelica单位检查和推导

基于表达式树的Modelica单位检查和推导赵建军;陈闻豪【摘要】为提高多领域物理系统统一建模中复杂模型的质量和可靠性,介绍Modelica对单位的支持方式以及单位之间的换算和运算规则,基于对表达式树的遍历,给出单位检查和推导机制的实现方案.通过分析该方案的时间复杂度,优化影响其效率的主要因素.在MWorks上实现该方案,并通过Modelica标准库中的模型证明该方案可行.【期刊名称】《计算机辅助工程》【年(卷),期】2013(022)004【总页数】6页(P76-81)【关键词】Modelica;表达式树;多领域物理系统;统一建模;模型可靠性;MWorks 【作者】赵建军;陈闻豪【作者单位】华中科技大学机械科学与工程学院,武汉430074;华中科技大学机械科学与工程学院,武汉430074【正文语种】中文【中图分类】TP391.90 引言在工程物理系统建模中，物理量通常由数值和单位组成，单位赋予数值以物理意义用于描述物理规律．［1］一般地，程序语言只支持单纯的数值表示，但不支持带单位的物理运算，因此要求建模者必须保证变量单位的可靠性，导致物理建模的复杂度和模型调试的难度增加．在物理模型中，变量的单位不仅使物理模型便于理解，而且为模型的调试和诊断提供有力依据，增加模型的可靠性并提高模型的质量．Modelica对不同领域物理系统模型进行统一表述，实现统一建模．目前，Modelica已经在很多领域得到应用，赵建军等［2］通过 Modelica模型与Simulink模块的转换机理，实现在S-Function联合仿真框架下的联合仿真．邓忠生等［3］对异步电机涉及的机械、电气与控制等多个领域的整体设计进行优化和性能分析，并对其进行统一建模仿真;为增加模型的可靠性，方便对模型的调试，用Modelica定义物理量单位的描述方式．MATTSSON等［4］阐述物理量单位对Modelica模型的意义以及Modelica对单位表示和运算的支持，并描述Dymola对单位检查和推导功能的支持．BROMAN等［5］从单位检查和推导原理的角度给出方案的合理化建议，并在Dymola平台上通过一些示例证明方案的可行性．目前，未见文献从方案实现和模型编译效率的角度阐述Modelica的单位检查和推导的具体实现方案．本文着重针对Modelica单位检查和推导机制实现方案以及效率分析优化进行初步探讨．1 Modelica的单位检查和推导1.1 物理模型中的单位物理模型中的单位既可以是SI单位，又可以是非SI单位，其中，SI单位定义7种基础单位(m，kg，s，A，K，mol和 cd)和一系列导出单位，这些导出单位都可以由这7种基础单位表示．［6］非SI单位虽然没有在SI中定义，但也可以由这7种SI基础单位表示或导出，如min=60 s，h=60 min等．在物理模型中，物理量的单位可以表示成一个或多个单位的n次幂的乘积形式，如kJ·m2/N，等效于kJ·m2·N－1．因此，单位可以表示为式(1)［7］的形式．式中:［］表示其中内容为可选项;{}表示为列表，包含一个或多个括号中内容．由于每个单位都可以由7个基础单位表示，因此，单位最终可以转换为一系列的基础单位和有理数指数的集合，表示为式中:prefix 为单位的前缀因数;m，kg，s，A，K，mol和 cd 为7 个基础单位;a1，a2，a3，a4，a5，a6和 a7分别为指数，可以为任意整数．Modelica中变量单位是物理模型中实型变量的一个名为unit的属性，通过单位表达式对unit赋值，单位表达式通过EBNF(Extended Backus-Naur Form)范式的形式描述．［8］Modelica 的代码为model TestReal m(unit=“kg”)=1“质量”;Real F(unit=“N”)=10“动力”;Real f=5“阻力”;//此处f无单位，将被推导Real a(unit=“m/s2”)“加速度”;Real s(unit=“km”)“位移”;Real W(unit=“N·m”)“做功”;Real Q(unit=“J”)“内能”;equationF－f=m·a;1 000·s=0．5·a·tim e2;W=1 000·F·s;Q=1 000·f·s;end Test1.2 单位检查和推导规则对等式方程中运算表达式的单位检查与推导基于各个运算表达式的单位运算规则．单位检查，指检查运算表达式的单位及其子节点单位是否满足该运算表达式的单位运算规则;单位推导由单位检查触发，即在单位检查过程中，如果运算表达式的子节点单位未知，那么根据该运算表达式的单位运算规则，将该子节点的单位运算出来．Modelica中需要进行单位检查推导的运算表达式主要包括等式方程、加减运算、乘除运算、幂运算、比较运算、条件表达式和函数调用表达式等7种．针对不同类型的运算表达式的单位检查和推导的规则，见表1．表1中的幂运算和函数调用的单位推导方向为单向，即只能通过幂底的单位运算得到幂运算表达式的单位，只能通过函数调用的形参单位推导出函数调用的实参单位．表1 表达式单位检查和推导规则Tab．1 Rules of expression unit check and derivation表达式类型单位检查和推导规则等式方程要求等式两边单位匹配加减运算要求左、右两边的单位匹配，并与表达式本身的单位匹配乘除运算要求左、右变量的单位运算结果与表达式本身的单位匹配幂运算要求其幂底的单位的n次方与表达式本身的单位匹配比较运算要求两边单位匹配条件表达式要求true/false分支的单位匹配，并与表达式本身的单位匹配函数调用要求函数的输入实参与输入形参单位匹配，输出实参与输出形参匹配2 基于表达式树的单位检查和推导的实现2.1 单位的解析2．1．1 单位表达式树为便于进行符号操作，表达式在分析时通常采用二叉树表示，称为表达式树．在表达式树中，非叶子节点为运算符、调用函数名或者子句标志，如+，－，×，÷，sin，cos，log，exp，if和 when 等;叶子节点可以是常量、参量或变量等．由式(1)可知，单位表达式中只包含乘法运算和幂运算2种．根据Modelica以EBNF范式的形式对单位表达式的描述，利用词法自动生成工具，可以将Modelica代码文本中的单位表达式生成表达式树的数据结构，以变量属性的形式与变量绑定储存．单位表达式树见图1，变量W的单位经过词法解析后，以表达式树数据结构形式描述变量单位．图1 单位表达式树Fig．1 Unit expression tree2．1．2 单位转换由式(2)可知，所有单位都可转换为7种基础单位表示［6］，因此，在系统初始化时导入映射表，用于换算各种SI单位前缀(见表2中列出的部分常见SI单位前缀)，以及将各种非基础单位转化为7种基础单位组成的表达式(见表3的示例)．表2 常见的SI词头换算Tab．2 Conversion of familiar SI prefixes前缀 G M k m μ n p换算 109 106 103 10－3 10－6 10－9 10－12表3 部分非基本单位到单位的转换映射Tab．3 Conversion mapping fromnon-basic unit to basic unit非基础单位转化为基础单位表示N kg·m·s －2 J kg·m2·s －2 km 103m根据表2和3，将图1的原始单位N·m解析为由7种基础单位表示的形式，见表4．表4 单位转换结果Tab．4 Unit conversion results初始单位单位因数基础单位集合单位指数kg m s N 1 1 1 －2 m m 1 1 1 12．1．3 单位约简完成单位转换后，需要根据乘法运算和幂运算规则，对表4中的单位进行单位约简运算，结果见表5．最终将单位N·m转化为由7个基础单位表示的形式kg·m2·s－2．表5 单位约简结果Tab．5 Results of unit reduction初始单位单位因数基础单位集合kg m s N·m 1 1 2 －22.2 单位检查和推导流程在模型编译［9］完成后，遍历编译生成的方程，依次对每个方程进行单位检查和推导．编译生成的方程通过表达式树的形式描述，因此对等式方程的单位检查和推导的根本依据，就是表1中的单位检查和推导的规则以及基于表达式树的对运算表达式节点的遍历操作．［10］针对某一运算表达式节点的单位检查和推导的流程见图2．图2 单位检查和推导流程Fig．2 Process of unit check and derivation以图1中的Modelica模型为例，具体阐述该单位检查和推导流程．模型编译完成后，生成以下7个方程，遍历方程集合，依次对每个方程进行单位检查和推导．以式(6)为例，其等式方程的表达式树形式见图3．图3 式(6)的表达式树形式Fig．3 Expression tree of equation(6)图3 中，(1)～(7)号节点为运算表达式节点，其他节点为单位表达式节点．依据图2的流程描述，对其进行单位检查和推导，过程如下．(1)对等式方程的运算表达式根节点(1)进行单位检查和推导，遍历其子节点．(2)对等式左节点(2)进行单位检查和推导，遍历其子节点．①对减号表达式左节点(4)进行单位检查和推导:获取变量单位为N;解析单位得到kg·m·s－2;返回节点(4)，单位推导成功．②对减号表达式右节点(5)进行单位检查和推导:变量无单位属性;返回节点(5)，推导失败．③存在子节点单位推导失败，对减号表达式(2)进行单位运算:运算得到节点(5)的单位为kg·m·s－2;再次对节点(5)进行单位推导，将单位推导至变量的单位属性，然后返回节点(5)，单位推导成功．④运算得到减号表达式节点(2)的单位为kg·m·s－2．⑤返回节点(2)，单位推导成功．(3)对等式右节点(3)进行单位检查和推导，遍历其子节点．①对乘号表达式左节点(6)进行单位检查和推导:获取变量单位为kg，是基础单位，无须单位解析;返回节点(6)，单位推导成功．②对乘号表达式右节点(7)进行单位检查和推导:获取变量单位为m·s－2，是基础单位，无须单位解析;返回节点(7)，单位推导成功．③子节点单位推导都成功，运算得到表达式节点(3)的单位为kg·m·s－2．④返回节点(3)，单位推导成功．(4)等式方程节点(1)所有子节点的单位推导都成功，且都为kg·m·s-2．(5)返回节点(1)，单位推导成功．经过上述流程分析，并结合流程图可知，对某一表达式节点的单位检查和推导的过程可分为3类情况:(1)变量单位一致，且单位完整．(2)变量单位一致，但单位不完整，需要进行单位推导．(3)变量单位不一致．需要注意的是，式(7)通过上述流程的单位运算后，可以得到等式方程左边表达式的单位为km，右边表达式的单位为m．二者虽然属于长度单位，具有相同的量纲，但是存在不同的单位前缀因数．如果对该种情况判定为单位一致而不给用户以警告，那么错误的单位转换会影响求解结果，必然留下求解错误的隐患．因此，对于单位量纲相同但存在转换因数差别的情况，都给用户以警告．3 单位检查和推导的效率分析3.1 时间复杂度计算由上述的实现流程可知，单位检查和推导是基于对表达式树的遍历操作、依次对表达式节点进行单位运算的过程．因此，单位检查和推导会给模型编译带来效率损失，本文就该问题对过程的时间复杂度进行分析．在对单位表达式进行解析过程中，其时间消耗取决于单位约简的效率．单位约简通过模式匹配的方式进行，其时间复杂度为式中:a，b和c分别表示进行单位运算的单位所包含基础单位的个数，体现模型中单位的复杂程度．单位的检查和推导过程就是对等式方程的每个表达式进行单位运算的过程，即该过程是对等式方程的表达式树遍历的过程．假设模型中有n个变量，模型中某等式方程最多由这n个变量组成，则该等式方程表达式树的末节点个数可以表示为k×n(如果每个变量在方程中只出现一次，那么k=1，满足k≥1)．考虑表达式节点总数最多的情况，该表达式树的节点总数可表示为S(n)也可以表示为对某一方程进行单位检查和推导过程所需单位运算最多的次数．对于整个模型，包含n个变量，必然也是n个方程，因此，对整个模型进行单位检查和推导所需的单位运算次数最多可表示为由式(10)可知，单位运算与单位的复杂程度有关，如果取单位的复杂程度为期望值，那么单位运算一次所需的时间可视为常数．因此，对模型进行单位检查和推导的时间复杂度［11］可表示为3.2 影响效率的主要因素根据对单位检查和推导的时间复杂度的计算，可以分析得到影响单位检查和推导效率的主要因素就是单位运算，其优化措施主要包括提高单次单位运算的效率和减少单位运算的次数．因此，本文从数据结构的调整和流程的优化2个方面进行分析，对整体效率进行优化．(1)调整单位的数据结构以提高单位运算效率．由式(10)可知，如果单位运算采用模式匹配算法，那么面对较复杂单位的运算，其效率十分低下，对单位解析以及单位检查和推导的效率造成巨大影响，而之所以采用模式匹配算法，是基于储存单位的数据结构因素．对该数据结构的调整方案如下．单位都可以表示成7种基础单位的指数集合的形式，则可以用一个长度为7的整型数组顺序储存7个基础单位的指数．以单位运算 J/N为例，单位 J解析为kg·m2·s－2，对应的 baseUnit={2，1，－2，0，0，0，0};单位N解析为kg·m·s－2，对应的baseUnit={1，1，－2，0，0，0，0}．单位运算 J/N 就分别对baseUnit对位相减，得到 baseUnit={1，0，0，0，0，0，0}，即将J/N运算得到m．因此，该方案的时间复杂度为常数时间O(1)．(2)选择性进行单位解析以减少单位运算次数．单位解析是将所有单位转化为基础单位的表示形式，如 J/N，转化为基础单位的表示形式，为kg·m2·s－2/(kg·m·s－2)，最后才能得到单位m．对于常见单位的运算，如果编译器能够识别，直接得到结果，可省去繁琐的转换到7种基础单位的过程．因此，可以通过查找哈希映射表的方式直接找到一些常见单位的运算，如果无法查找到运算结果，再进行原方案中的单位解析流程．其中，查找哈希映射表的时间为常数时间O(1)，对整体流程的效率影响可以忽略．因此，通过先在哈希映射表中查找对应的单位运算，然后根据查找结果选择性地进行单位解析运算，可以有效地减少单位运算的次数．3.3 工程示例在多领域物理建模仿真平台MWorks上，以Modelica标准库中多体库的双摆模型为例进行研究．双摆模型的组件见图4．该模型包含1 776个变量和方程，在本地计算机上(计算环境:CPU为Intel酷睿 i3 2100，3 100 MHz;内存为 2 GB，1 333 MHz)分别对不做单位检查、优化前做单位检查和优化后做单位检查的编译时间进行比对，见表6．由表6可知，优化前的单位检查时间为0．23 s，占总编译时间的8．68%，优化后的单位检查时间为0．11 s，占总编译时间的4．35%．可见，优化方案有效．图4 双摆模型的组件Fig．4 Components of double pendulum model表6 编译时间比较Tab．6 Comparison of compile time s检查类型不做单位检查优化前做单位检查优化后做单位检查编译时间2．42 2．65 2．534 结束语从单位运算规则的角度描述Modelica对物理单位的支持，并基于表达式树，在多领域统一建模仿真平台MWorks上实现单位检查和推导机制，最后从效率的角度进行分析，找到影响单位检查和推导的主要因素，对其进行针对性的效率优化．通过MWorks平台对标准库模型的测试证明方案实现和优化的可行性，在工程示例中效果显著．参考文献:【相关文献】［1］吴义忠，吴民峰，陈立平．基于Modelica语言的复杂机械系统统一建模平台研究［J］．中国机械工程，2006，17(22)：2391-2396．WU Yizhong，WU Minfeng，CHEN Liping．Study on the hybrid modeling platform based on Modelica language for complex machinery system［J］．China Mech Eng，2006，17(22)：2391-2396．［2］赵建军，吴紫俊．基于Modelica的多领域建模与联合仿真［J］．计算机辅助工程，2011，20(1)：168-172．ZHAO Jianjun，WU Zijun．Multi-domain modeling and co-simulation based on Modelica［J］．Comput Aided Eng，2011，20(1)：168-172．［3］邓忠生，刘文华，何相良，等．基于Modelica的异步电机建模仿真［J］．计算机辅助工程，2011，20(1)：177-180．DENG Zhongsheng，LIU Wenhua，HE Xiangliang，et al．Modeling and simulation for induction motor based on Modelica［J］．Comput Aided Eng，2011，20(1)：177-180．［4］ MATTSSON S E，ELMQVIST H．Unit checking and quantity conservation［C］//Proc 6th Int Modelica Conf，Germany，2008：13-20．［5］ BROMAN D，ARONSSON P，FRITZSON P．Design considerations for dimensional inference and unit consistency checking in Modelica［C］//Proc 6th Int Modelica Conf，Germany，2008：3-12．［6］ TAYLOR B N，THOMPSON A．NIST Special Publication 330，The international system of units(SI)［S］．［7］ Modelica Association．Modelica Language Specification 3．2［EB/OL］．(2010-03-24)［2012-10-30］．https：//www．modelica．org/documents/ModelicaSpec32．pdf．［8］徐玉坤，刘淑芬，李兵．建模语言中的文本表面语法分析方法研究［J］．计算机应用研究，2008，25(3)：791-794．XU Yukun，LIU Shufen，LI Bing．Research of text concrete syntax analys is methods in modeling language［J］．Application Res Computers，2008，25(3)：791-794．［9］周凡利．工程系统多领域统一模型编译映射与仿真求解研究［D］．武汉：华中科技大学，2011．［10］何志宏，毛志军．表达式与二叉树的相互转换［J］．电脑知识与技术，2010，6(5)：1201-1203．HE Zhihong，MAO Zhijun．The mutual conversion between an expression and a binary tree［J］．Comput Knowledge ＆ Technol，2010，6(5)：1201-1203．［11］严蔚敏，吴伟民．数据结构：C语言版［M］．3版．北京：清华大学出版社，2007：13-17.。

利用作物生长模型和时序信号甄别水稻镉胁迫

第37卷第4期农业工程学报 V ol.37 No.42021年2月Transactions of the Chinese Society of Agricultural Engineering Feb. 2021 249 利用作物生长模型和时序信号甄别水稻镉胁迫孔丽，刘美玲※，刘湘南，邹信裕（中国地质大学（北京）信息工程学院，北京 100083）摘要：在自然农田生态系统中，农作物的生长通常受到各类环境胁迫（如重金属胁迫、病虫害、水分、营养）的影响，如何区分重金属胁迫与其他胁迫有待进一步研究。

该研究选取了湖南省株洲为试验区，收集2017—2019年的Sentinel-2卫星影像数据，结合野外实测数据，开展水稻重金属镉（Cd）胁迫识别研究。

首先，利用作物生长模型World Food Studies （WOFOST）同化时序遥感数据获取每年的叶面积指数（Leaf Area Index，LAI）时间序列曲线；然后运用集合经验模态分解（Ensemble Empirical Mode Decomposition，EEMD）方法对LAI时间序列进行多尺度分解，得到不同的时序信号分量（Intrinsic Mode Function，IMF）；最后使用动态时间规整（Dynamic Time Warping，DTW）方法计算受胁迫水稻分解后的时间序列与健康水稻分解后的时间序列之间的DTW距离，即归一化胁迫指数。

结果表明：归一化胁迫指数是水稻重金属胁迫敏感的参数，与土壤重金属含量的相关系数为0.851，水稻受到的胁迫程度越高，归一化胁迫指数值越大，反之越低；在试验区中，水稻重度重金属胁迫的分布面积比例相对较低，且主要集中在西部、东北部以及偏东南地区。

融合集合经验模态分解和动态时间规整方法能有效地甄别并定量分析水稻重金属胁迫状况，从而为作物重金属污染胁迫监测提供重要参考。

关键词：遥感；模型；重金属；镉胁迫；时序信号分解；WOFOSTdoi：10.11975/j.issn.1002-6819.2021.04.030中图分类号：S127 文献标志码：A 文章编号：1002-6819(2021)-04-0249-08孔丽，刘美玲，刘湘南，等. 利用作物生长模型和时序信号甄别水稻镉胁迫[J]. 农业工程学报，2021，37(4)：249-256.doi：10.11975/j.issn.1002-6819.2021.04.030 Kong Li, Liu Meiling, Liu Xiangnan, et al. Identifying heavy metal (Cd) stress in rice using time-series signals and crop growth model[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2021, 37(4): 249-256. (in Chinese with English abstract) doi：10.11975/j.issn.1002-6819.2021.04.030 0 引言随着经济的不断发展，工业化不断推进，土壤重金属污染成为当今世界面临的重大生态环境问题之一[1]。

页岩气储层水力压裂复杂裂缝导流能力实验研究

页岩气储层水力压裂复杂裂缝导流能力实验研究王雷;王琦【摘要】为研究页岩气储层水力压裂后复杂裂缝导流能力,运用FCES-100裂缝导流仪,选取页岩地面露头岩心,加工成符合实验要求尺寸岩心板,将页岩复杂裂缝简化为转向裂缝和分支裂缝两种形式,用陶粒和覆膜砂两种类型支撑剂进行导流能力实验测试.实验结果表明:裂缝形态对导流能力影响较大,裂缝转向后导流能力明显低于单一裂缝,低闭合压力条件下转向裂缝与单一裂缝导流能力相差35％～ 40％,随闭合应力增大,差距逐渐增大;低闭合压力下陶粒导流能力高于覆膜砂,而当闭合压力增大后覆膜砂的导流能力反超陶粒,低铺砂浓度下反超趋势更加明显;分支裂缝存在时,等量支撑剂多条分支裂缝的等效导流能力小于单一裂缝,高闭合压力下分支裂缝中不同分支铺砂浓度的差异越大,导流能力与单一裂缝越接近.%In order to study the seepage capacity of complex fracture after fracturing of shale gas well,the outcrop shale being processed into the core plates whose size meets the requirements of the experiments,the complex fractures in the shale being simplified to two types:turning fractures and branching fractures,and ceramsite and coated sand being used as proppant,the seepage capacity of 2 kinds of complex fractures was tested by FCES-100 fracture flow deflector.The experimental results show that:the fracture morphology has a great influence on its seepage capacity,the seepage capacity of turning fracture is lower 35％～40％ than that of single fracture under low closing pressure,and the difference between both increases gradually with the increase of closing pressure;under low closure pressure,the seepage capacity of the ceramic proppant fracture is higherthan that of the coated proppant fracture,but with the increase of the closure pressure,the seepage capacity of the coated proppant fracture increases gradually and exceeds that of the ceramic proppant fracture,and the exceeding trend becomes more obvious under low sand concentration;the equivalent seepage capacity of branching fracture is lower than that of single fracture under the same amount of proppant,the difference between both dwindles with the increase of the difference in the sand concentration of different branch cracks under high closure pressure.【期刊名称】《西安石油大学学报（自然科学版）》【年(卷),期】2017(032)003【总页数】5页(P73-77)【关键词】页岩气井;水力压裂;裂缝导流能力;支撑剂;复杂裂缝【作者】王雷;王琦【作者单位】中国石油大学(北京)石油工程教育部重点实验室,北京102249;中国石油勘探开发研究院,北京100083【正文语种】中文【中图分类】TE357.1王雷,王琦.页岩气储层水力压裂复杂裂缝导流能力实验研究[J].西安石油大学学报(自然科学版),2017,32(3):73-77.WANG Lei,WANG Qi.Experimental research on seepage capacity of complex fracture in shale gas reservoir after hydraulic fracturing[J].Journalof Xi′an Shiyou University (Natural Science Edition),2017,32(3):73-77.页岩气储层渗透率低、物性差，不采取增产改造措施一般没有工业产能[1-2]，而水力压裂是提高页岩气井生产能力的有效措施[3]。

Geometric Modeling

Geometric ModelingGeometric modeling is a fundamental aspect of computer-aided design (CAD) and computer graphics, playing a crucial role in the creation of virtual 3D models and simulations. This technology has revolutionized various industries, including architecture, engineering, animation, and gaming, by enabling designers and developers to visualize and manipulate complex geometric shapes with precision and efficiency. In this response, we will explore the historical background, different perspectives, case studies, and critical evaluation of geometric modeling, as well as its future implications and recommendations. The development of geometric modeling can be traced back to the early 1960s when Ivan Sutherland created Sketchpad, a revolutionary computer program that allowed users to draw and manipulate basic geometric shapes on a screen. This marked the beginning of computer-aided design (CAD) and laid the foundation for the development of geometric modeling as we know it today. Over the years, geometric modeling has evolved significantly, with the introduction of advanced algorithms, rendering techniques, and modeling tools that have expanded its applications across various industries. From a historical perspective, the evolution of geometric modeling has been driven by the increasing demand for more sophisticated and realistic 3D models in fields such as architecture, automotive design, industrial engineering, and entertainment. As technology has advanced, so too has the complexity and realism of geometric models, leading to a greater emphasis on precision, detail, and interactivity in the design and visualization process. From a technological perspective, geometric modeling has undergone a paradigm shift with the advent of parametric and non-parametric modeling techniques. Parametric modeling allows designers to create models based on a set of parameters, enabling them to make changes and updates to the design easily. On the other hand, non-parametric modeling focuses on creating freeform shapes and surfaces, providing greater flexibility and creativity in the design process. These different perspectives have led to debates within the industry about which approach is more effective for specific applications and design requirements. One example of the impact of geometric modeling can be seen in the field of architecture. With the use of CAD software and advanced geometric modeling tools, architects can create highlydetailed and realistic 3D models of buildings and structures, allowing them to visualize the final product and make necessary adjustments before construction begins. This not only improves the design process but also helps in conveying the design intent to clients and stakeholders, leading to better communication and decision-making. In the automotive industry, geometric modeling hasrevolutionized the design and manufacturing process, enabling engineers and designers to create complex 3D models of vehicles and their components with precision and accuracy. This has led to significant improvements in vehicle safety, performance, and aesthetics, as well as streamlined production processes and reduced time-to-market for new vehicle models. While geometric modeling has brought about numerous benefits, it also poses certain challenges and drawbacks. One of the main drawbacks is the steep learning curve associated with advanced modeling tools and techniques, which can be a barrier for newcomers to the field. Additionally, the complexity of geometric models can lead to performance issuesand computational challenges, especially when dealing with large-scale models or simulations. Looking ahead, the future implications of geometric modeling arevast and promising. As technology continues to advance, we can expect to see even more realistic and interactive 3D models that push the boundaries of visualfidelity and immersion. Furthermore, the integration of geometric modeling with other emerging technologies such as virtual reality (VR) and augmented reality (AR) holds great potential for creating new and innovative applications in fields such as education, training, and entertainment. In conclusion, geometric modeling has played a pivotal role in shaping the way we design, visualize, and interact with3D models across various industries. While it has brought about significant advancements and benefits, there are also challenges and considerations that need to be addressed. By understanding the historical background, different perspectives, and case studies related to geometric modeling, we can better appreciate its impact and potential for the future. As technology continues to evolve, it is essential to stay informed and adaptable to the latest trends and developments in geometric modeling, in order to harness its full potential and drive innovation in design and visualization.。

【机械类文献翻译】CAD或CAE在塑料模具设计中的应用

Fundamentals of Mechanical DesignAs a plastic polymer chemistry and materials science and the development of the important achievements already familiar, Plastic products have become productive human life and an indispensable important component. Over the years, the plastic products industry has been developing rapidly, As the global scope of plastic instead of metal trend has been further accelerated the pace of development. Plastic products are generally used molding methods of production, thereby plastic mold has become an important production technology and equipment, in the national economy plays an increasingly important role. With plastic products in the home appliances, electronic products and other daily necessities to the increasingly widespread application on the plastic mold design and manufacturing requirements are also increasing. Traditional manual design and manufacturing has long since failed to meet production needs. CAD and CAM development is adapted to the actual requirements of this objective. CAD/CAM be significantly enhanced plastic products, and plastic mold design and manufacturing efficiency and raise the quality of design and manufacture. reduce tryout repair mode, thus shortening from plastic product design, mold design, Mold manufacture of molded products for the entire production cycle.1，plastic mold CADCAD design and computer organic combination play their respective expertise to the new design methods. That the design staff in the design process, giving full playto a powerful computer computing, information storage and the ability to rapidly identify, complete information management, numerical calculation, simulation analysis, Optimum design and drafting tasks designers concentrate on effective creative thinking, in order to better accomplish the program from the design, evaluation, simulation and analysis of specific modifications to the design of the whole process.CAD development generally follows : electronic drawing board system (2-D computer graphics technology), surface modeling system, solid modeling, parametric technology and variable technology. Electronic drawing board system is the traditional view of the three ways to express parts of the media to the drawings for technical exchange. Surface modeling system is developed in the two-dimensional mapping system, which is based on Bessel algorithm, introduction of 3D surface modeling system, it is the first CAD revolution. Solid modeling technology can convey accurately the quality of parts and the center, away from inertia characteristics, It marks the universal application of CAD history of the development of the second revolution and the future development direction of the CAD. Parametric Technology is an unconstrained free than to create more innovative and better algorithm, which is characterized by : Based on the characteristics, full size constraints, all relevant data, dimension driven design changes. Parametric Technology Application dominated the history of the development of graphics third revolution. As the "full size constraint" mandatory interference and constrains the designer's creativity and imagination in the design of the key physical topology change, the loss of certain binding characteristicsdata, it will cause confusion, result in a Parametric Technology modeled on the more advanced solid modeling - variable technique.2, plastic mold CAECAE (Computer Aided Execution) (31 computer-aided engineering technology that It is the emergence of computer-aided design / computer-aided manufacturing (CAD/CAM) technology to the development of the depth direction requirements. M general it is a numerical calculation includes technical, databases, computer graphics , engineering analysis and simulation of an integrated software system Its core technology is the engineering model and numerical method. Mold on thecomputer-aided engineering technique , it is primarily the use of polymer rheology, heat transfer, numerical methods and computer graphics, and other basic theory the plastic molding process simulation of the mold before it can create the image , directly on the computer screen simulate the actual process of forming forecast mold design and molding conditions on the impact of products found potential defects for the judgment of mold design and molding conditions provide reasonable scientific basis. With computer technology is developing rapidly , the various plastic molding process simulation of plastic processing industry has become a hot spot. Now tell us about CAE technology in Note molding and gas-assisted injection molding applications.(1) Injection MoldingCAE in the injection mold, the mold filling for mobile, packing, cooling and warpage analyzed.Filling flow analysis can be made the following major tasks : optimization of casting system included in the balanced flow on the basis of rational flow channel size, distribution and the best runner number, location and shape; Optimal injection parameters, the analysis of flow front; Weld line gas analysis Point location; pressure field, temperature and velocity analysisPacking refers to the process in order to be satisfied with the products, Filling in the end still higher packing under pressure to the cavity to the feed to compensate for the temperature, pressure changes caused by volumetric shrinkage. Pressure is the essence of the process feed mainly used to predict melt in the cavity fill material and compaction process pressure, temperature, calculate volumetric shrinkage cavity and the shear stress and changes in the density.Cooling process melting plastics occurred curing, curing process by giving off heat from the coolant die away. The process of mold cavity temperature and the uniformity of a direct impact on the production of injection molded parts efficiency and quality. Chang main design parameters including : cooling pipe size, and the location of the cold tp pipeline linking relations geometric parameters and the coolant flow, inlet temperature and other physical parameters. A good cooling system should make Die achieve rapid and balanced cooling to reduce the cooling time, increase the productivity and and reduce or avoid warpage plastic parts, residual stress and surface quality defects, improving the quality of products. Warpage analysis is the application of the basic principles of mechanics and FEM / IsW difference numerical algorithm, Calculation of plastic parts molding size and deformation(2) gas-assisted injection moldingGas-assisted injection molding in the traditional injection molding developed on the basis of a new injection molding process. Its features is : filling stage to the cavity injection high-pressure gas; In the packing stage, and continue to inject high-pressure gas melt to compensate for the cooling caused by the contraction.Because gas, melt two entirely different dynamic interaction of material, makes molding process simulation is very complicated. The governing equations ofnon-Newtonian fluid under non-isothermal the Hele-Shaw flow, but assumed the gas cladding interface that the gas, melt two-phase media is not mixed. Thus, the flow field for the melt flow into the equation solving, Melting only in the gas interface of gas pressure at the top of the boundary conditions. Particle tracking normally used to control gas or gas fusion method to determine the interface.Gas Aided Design Process of CAE technology can resolve the following issues : gas. Cavitation potential -- through the quality issues; Melt determine the optimal size of the gas injected into the best time of process parameters; obtain more cavity system in the whole processing of materials and gas distribution; Optimization flow, the gating system size, Layout programs.3, CAE software briefMF software is a specialized in plastic computer aided engineering analysis (CAE) software and consulting firms (MOL DFLO~) development. MOLDFLO~ software can simulate the entire injection process and the process of injection molding products affected. They mainly consist of flow simulation module (MF~Flow), cooling simulation module (MF/Cool) Pressure Analysis Module (MFIPack), warping analysis module (MFl9Crrap) MF/SHRINK mold cavity to determine the size, structure MF/STRESS stress analysis MF/O~M injection molding machine parameter optimization, MF/GAS gas-assisted injection analysis, MF/FIBER plastic parts fiber orientation analysis module, simulation and analysis to determine the reasonable die structure and the technical parameters of appropriateness. The latest version can automatically establish and run a series of analysis, after the analysis of the structure will be automatically generated.4, CAD and CAE design of the general steps DiePlastic mold cavity and the core is plastic parts geometry of the design, therefore, first step in the design of plastic parts to describe the shape of three plastic parts of the computer model accurately established, Mold it into the entire design process and the basis for the foundation. Then proceed to the preliminary program of mold design, mold design of gating system and cooling systems. CAE software mode analysis, thereby optimizing the structure of mold design, based on an analysis of the results of pouring, Cooling system amendment. Plastic Parts will join three-dimensional model than can be contracted mold parts 3D model. Will tie in with the mold parts (such as Insert, pumping cores) 3D assembly model for interference checking. On the assembly projection models, such as cutting operation can be mold assembly map Molding of the same components for the operation is likely to be the two-dimensional engineering drawings. Mold parts of the 3D model can be used to generate CAM NC machining process.5, the end of LanguageCurrently, plastic mold design, generic manufacturers generally use the machinery CAD/CAM software and specialized CAE software combination. Development and promotion of specialized CAD software is the issue. CAE analysis only under the conditions set for plastic molding process simulation, it is not the result of analysis, the optimum nor can the design process for the guidance of experts. Integrated special plastic mold CAD and CAE software, enhancing its intelligence, CAD and CAE is a plastic mold technology development trends.CAD/CAE在塑料模具设计中的应用前言塑料作为高分子化学和材料科学发展的重要成果，早已为人们熟悉，塑料产品已经成为人类生产和生活中不可缺少的重要组成部分。

中英文文献翻译—UG NX简介

附录附录AUG NX Summarized AccountIntroduction to UG NX UG NX is American Unigraphics Solutions (UGS) company's PLM offering for the core components. UGS company is an American a global supplier of MCAD. PLM Solutions provide a powerful vitality of product lifecycle management (PLM) solutions, including product development, manufacturing, planning, product data management, e-commerce product solutions, but also offers a complete suite of services for product improvement. UG to the automotive and transportation, aerospace, consumer goods, General Engineering and electronic industries through its virtual product development (VPD) provide Multipole, integrated,enterprise-class products and services, including software, and a complete solution.CAD/CAM/CAE three systems tightly integrated. Users use the UG powerful solid modeling, surface modeling, virtual Assembly and create functions such as engineering drawings, you can use the CAE module for finite element analysis, kinematic analysis and simulation, to improve the design of reliability; according to the established 3D model, but also by CAM module directly generate CNC code, for product processing. Flexibility in the way of modeling. Composite modeling technology, is solid modeling, surface modeling, wireframe modeling, display the geometric modeling and parametric modeling. Parameter-driven, intuitive, easy to modify the image. Surface design for non-uniform rational B-spline curve-based, you can use several methods to build a complex surface, powerful. A good second development environment, users can use a variety of ways for the second development. Knowledge-driven automation (KDA), facilitate access to and reuse of knowledge.A brief, post processing both CAM software, its main purpose is to build components in the machine tool path (cutting). Generally speaking, you cannot directly transfer CAM software internally generated by the tool path to machine processing, because each type of machine in the physical structure and control system may be different, resulting in the NC program instructions and format requirements may differ. Therefore, the tool path data mustbe processed to fit each machine and control system for specific requirements. This processing, in most CAM software called "post-processing". Post-processing of results is to make the tool path data into machine recognizes tool path data, that is, the NC code. Visible, post-processing must have two elements: cutting-CAM homegrown tool path; after the processor —is a machine tool and its control system information. UG system provides general post processor — UG/Post, it uses the UG internal tool path data as input, output machine after via post-processing recognizes the NC code.UG/POST Organizational structureUG/Post has a strong user of capability, it can be adapted from the very simple to arbitrarily complex machine and control system for processing. Second, UG/Post composition structure mentioned UG/post processor, had to be simple to introduce MOM (Manufacturing Output Manager), i.e. processing output Manager. MOM is UG provides an event-driven tools, UG/CAM module output from it to manage, it is stored in the UG/CAM data to extract the data to generate output. UG/Post it is this tool a specific application. MOM is UG/post processor core, UG/post use MOM to start the interpreter, to explain the program provides the functionality and data, and loads the event processor (Event Handler) and the definition file (Definition File).In addition to the MOM, UG/post is primarily determined by the event generator, the event processor, definition file and output files, and so on of four elements. Once you start UG/post processor to handle UG internal tool path, its processes to the following: event Builder from beginning to end to scan the entire UG tool path data, extract each of the events and their associated parameters information, and they are passed to the MOM to MOM; then, delivered each event and its associated parameters to the user in advance to develop good event processor, and collected by the event processor based on their content to decide upon how each event for processing; then the event handler returns the data to MOM as their output, MOM read definition file content to determine how to format the output data; Finally, MOM to well-formatted output data to write to the specified output file. Figure 1 describes the concepts and content.附录BUG NX简介UG NX是美国Unigraphics Solutions（简称UGS）公司的PLM产品的核心组成部分。

solidworks产品设计流程

solidworks产品设计流程英文回答：SolidWorks is a popular computer-aided design (CAD) software used for product design and development. It provides a comprehensive set of tools and features that enable engineers and designers to create 3D models,simulate their performance, and generate detailed drawings for manufacturing.The product design process in SolidWorks typically involves several key steps. First, I start by gathering requirements and specifications from the client or end-user. This helps me understand their needs and expectations forthe product. For example, if I am designing a new smartphone, I would gather information about the desired screen size, camera specifications, and other features.Once I have a clear understanding of the requirements,I move on to the concept development phase. This is where Ibrainstorm and sketch out different ideas for the product.I can use SolidWorks' sketching tools to quickly create 2D sketches of the product's form and features. For instance,if I am designing a new chair, I would sketch out different shapes and configurations to explore various design possibilities.After the concept development phase, I select the most promising design and start creating a 3D model in SolidWorks. I can use a combination of sketching, extruding, and other modeling tools to create the geometry of the product. SolidWorks also allows me to apply materials and textures to the model, giving it a realistic appearance.For example, if I am designing a car, I can apply ametallic material to the body and rubber material to the tires.Once the 3D model is complete, I can simulate its performance using SolidWorks' analysis tools. This helps me evaluate factors such as structural integrity, fluid flow, and heat transfer. For instance, if I am designing a new bridge, I can simulate the load and stress on differentcomponents to ensure they can withstand the expected forces.After analyzing the design, I can make any necessary modifications to improve its performance or meet thedesired specifications. SolidWorks allows me to easily make changes to the 3D model and update associated drawings automatically. This flexibility and parametric modeling capability save time and effort in the design iteration process.Finally, I can generate detailed drawings and documentation for manufacturing. SolidWorks provides arange of tools for creating dimensioned drawings, assembly instructions, and bills of materials. These documents communicate the design intent and specifications to the manufacturing team, ensuring accurate production of the product.中文回答：SolidWorks是一款流行的计算机辅助设计（CAD）软件，用于产品设计和开发。

《微积分英文》课件 (2)

Methods for finding limits using algebra
Types of Limits
One-sided limits
Limits approached
from one direction
Limits at infinity
Behavior of functions at
infinity
● 02
第2章 Limits and Continuity
01 Definition of a limit
Explanation of what a limit is
02 Properties of limits
Key characteristics of limits
03 Calculating limits algebraically
Graphing functions by analyzing their derivatives and key points
Higher Order Derivatives
Second derivative
Rate of change of the rate of
change
nth derivative
● 03
第3章 Differentiation
Derivatives and Rates of
Change
A derivative is defined as the rate of change of a function at a given point. Notation for derivatives includes symbols such as f'(x) or dy/dx. Derivatives can be interpreted as rates of change in various realworld applications.

210988696_转子系统动力学仿真平台设计与实验验证

中图分类号: ＴＰ２０文献标志码: Ａ文章编号: １００５－３０２６(２０２３)０３－０３７５－０７
ＤｅｓｉｇｎａｎｄＥｘｐｅｒｉｍｅｎｔａｌＶｅｒｉｆｉｃａｔｉｏｎｏｆＲｏｔｏｒＳｙｓｔｅｍＤｙｎａｍｉｃｓ
ＳｉｍｕｌａｔｉｏｎＰｌａｔｆｏｒｍ
ＬＵＯＺｈｏｎｇ１ꎬ２ꎬ３ ꎬ ＷＵＤｏｎｇ￣ｚｅ１ꎬ２ ꎬＬＩＬｅｉ１ꎬ２ ꎬ ＧＥＣｈａｎｇ￣ｃｈｕａｎｇ４
Ｋｅｙｗｏｒｄｓ: ｒｏｔｏｒｓｙｓｔｅｍꎻ ｄｙｎａｍｉｃｓｓｉｍｕｌａｔｉｏｎｐｌａｔｆｏｒｍꎻ ｐａｒａｍｅｔｒｉｃｍｏｄｅｌｉｎｇꎻ ａｕｔｏｍａｔｅｄ
ｓｉｍｕｌａｔｉｏｎꎻ ＡＣＴ( ａｐｐｌｉｃａｔｉｏｎｃｕｓｔｏｍｉｚａｔｉｏｎｔｏｏｌｋｉｔ)
动力学仿真计算振动响应特性时ꎬ由于模型
动机燃烧室设计过程. 在动力学仿真平台方面ꎬ钟
应用层为平台的核心内容ꎬ主要包括仿真项
兴志 [１４] 通过Ｐｙｔｈｏｎ语言和ＡＣＴ进行ＡＮＳＹＳ
目、参数化建模、自动化仿真三个模块ꎬ模块与模
Ｗｏｒｋｂｅｎｃｈ的二次开发ꎬ集成了Ｗｏｒｋｂｅｎｃｈ软件
块之间相互配合. 本文所提设计主要是对这三个
将处理完成的数据提供给应用层ꎬ实现了应用程
序、软件、数据之间的交互.
数据层是作为研究人员的重要数据资料ꎬ主
要包括结果分析、模型、文档等数据ꎬ并且能在一
定程度上提供交互功能.
缺乏针对性的引导ꎬ容易造成操作上的遗漏或误
的建立费时费力ꎬ且模态分析、瞬态分析、谐响应
触. 同时材料赋予、支承设置、网格划分、分析设
分析、随机振动等众多仿真项目ꎬ仿真界面复杂ꎬ
置、约束设置等前处理操作过程繁琐ꎬ仿真效率

水文模型大全

Water Rights Analysis Package (WRAP)
GIS Application in Hydrology and Hydraulics
Automated Geospatial Watershed Assessment Tool eCoastal Program BASINS version 4.0 GIS Weasel HAZUS-MH HEC-GeoRAS HEC-GeoHMS MapWindow NHDPlus Append Tool NRCS Geo-Hydro_ArcGIS NRCS Geo-Hydro_ArcView StreamStats Soil and Water Assessment Tool (SWAT)
Environmental Models
Agricultural Non-Point Source Pollution Models (AGNPS 98) Areal Nonpoint Source Watershed Environmental Simulation (ANSWERS)
Continuous Annual Simulation Model (CALSIM) Erosion Productivity-Impact Calculator/ Environmental Policy Integrated Climate (EPIC) Hydrologic Simulation Program-Fortran (HSPF) LOAD ESTimator (LOADEST) One-dimensional Transport with EQuilibrium chemistry (OTEQ) Illinois Least-Cost Sewer System Design Model (ILSD) Illinois Urban Storm Runoff Model (IUSR) Water Quality/Solute Transport (OTIS) Soil Water Assessment Tool (SWAT) Large Scale Catchment Model, formerly CALSIM (WRIMS)

码垛机器人的机械结构设计毕业设计说明书

码垛机器人的机械结构设计毕业设计说明书毕业设计说明书（论文）论文题目：码垛机器人的机械结构设计系部：机械工程系专业：班级：学生姓名：学号：指导教师：2014年月日I摘要以码垛机器人本体为研究对象，通过分析其结构特点与性能参数，明确了设计的基本指标，为码垛机器人产品开发提供指标。

并依此为依据针对物流自动化行业中对箱包高速码垛的需求，并依据搬运机器人的性能要求，设计了一种四自由度的码垛机器人。

应用CERO2.0进行三维建模，并通过CERO2.0的机械设计分析模块，对其构建的三维模型的运动仿真。

结果表明，所设计机器人完全满足工业现场的需求。

关键词：码垛机器人机械设计CERO2.0运动分析AbstractIn this paper,the structure characteristics and performance indicators of palletizing robot body are an alyzed．It provides a reference for product development．In accordance with the requirement of robot palletizer in logistics automation technology，a universal robot palletizer was designed based on the functional requirement.Application CERO2.0 for 3 d modeling, and through the analysis of the mechanical design CERO2.0 module, the building of 3 d motion simulation of the model。

the experimental results show that the robot meets the objectives of the logistics automation requirements．Key Words：palletizing robot；machine design；cero2.0 motion analysisI目录绪论 (1)第1章码垛机器人现状研究 (2)1.1 引言 (2)1.2 结构分析 (2)1.3本体性能研究 (3)第2章码垛机器人的机械设计和电气控制 (6)2.1机械设计 (6)2.2电气控制系统 (13)第3章码垛机器人运动分析 (16)3.1 CERO2.0动态机构仿真简介 (16)3.2 CERO2.0机器人运动仿真 (20)致谢 (30)绪论随着21世纪工业及经济的蓬勃发展以及对产品精度的要求不断提高，机器人加工逐渐成为一种被普遍应用的加工方法，而码垛是物流自动化技术领域一门新兴技术，码垛按照一定模式，一件件堆成码垛，以便使单元化的码垛实现物料的搬运、存储、装卸运输等物流活动，随着工业化大生产规模的扩大，促使码垛自动化，以加快物流的速度，保护工人的安全和健康，同时也能获得整齐一致的物垛，减少物料的破损和浪费。

基于能源结构的IGT模型对CO2排放量的研究

摘要：能源结构指标Ｅ将引入到ＩＴ模型中，Ｇ并用模糊矩阵对Ｅ进行计算，。同时，用马尔可夫模型对湖南采省能源结构进行预测，果显示：２２结到００年，南省煤、油、然气、电在一次能源消费中的占比为：湖石天水
２ＫｙＬｂｒｔｙｏｎｉｎｅｔｉｌｇｎｏｕｉｎＣｎｒｌＨｕａｎｖｒｉ）．ｅａｏａｏｆｖｒｍｎａＢｏｙａｄＰｌｔｏｔ（ｎｎＵｉｅｓｙ，ｒＥｏｌｏｌｏｏｔ
Ｍｉｉｒｏｄｃｔｎｈｎｓａ４０８Ｃｉａｎｓｙｆｕａｉ，Ｃａｇｈ１０２，ｈｎ）ｔＥｏ
ＺｎｕｎｍｎＨａｇＨａｕＸａｇＱｕａ，ｅｇＧａｇｉｇ一，ｕｎｕｊｎ一，ｉｎｉｌｉ
（．ｏｅｅｏｎｉｎｎａＳｉｎｅａｄＥｇｅｒｇＨｕａｎｖｒｉ，ｈｎｓａ４０８，Ｃｉａ１Ｃｌｇｆｖｏｍｅｔｌｃｃｎｎｉｅｎ，ｎｎＵｉｓｙＣａｇｈ１０２ｈｎ；ｌＥｒｅｎｉｅｔ
Ｃ，放量仅为１１Ｏ排．９畦／万元，２１比００年下降４．％；００年因能源结构优化而减排的Ｃ２量为７０９７万１６２２Ｏ４．９
吨，当年Ｃ排放总量的１．５。此外，占Ｏ４４％当湖南省单位ＧＰ年均节能率达到７８左右时，Ｃ放量Ｄ．％其Ｏ排将维持２１００年的水平不变，而实现经济增长和Ｃ放量的脱钩。从Ｏ排关键词：源结构；Ｔ模型；Ｏ排放量；尔可夫模型能ＩＧＣ２马中图分类号：２Ｘ２Ｘ１２文献标识码：Ｂ

基于深度学习的多目标跟踪关联模型设计

第46卷第12期自动化学报Vol.46,No.12 2020年12月ACTA AUTOMATICA SINICA December,2020基于深度学习的多目标跟踪关联模型设计侯建华1张国帅1项俊1摘要近年来,深度学习在计算机视觉领域的应用取得了突破性进展,但基于深度学习的视频多目标跟踪(Multiple object tracking,MOT)研究却相对甚少,而鲁棒的关联模型设计是基于检测的多目标跟踪方法的核心.本文提出一种基于深度神经网络和度量学习的关联模型:采用行人再识别(Person re-identiﬁcation,Re-ID)领域中广泛使用的度量学习技术和卷积神经网络(Convolutional neural networks,CNNs)设计目标外观模型,即利用三元组损失函数设计一个三通道卷积神经网络,提取更具判别性的外观特征构建目标外观相似度;再结合运动模型计算轨迹片间的关联概率.在关联策略上,采用匈牙利算法,首先以逐帧关联方式得到短小可靠的轨迹片集合,再通过自适应时间滑动窗机制多级关联,输出各目标最终轨迹.在2DMOT2015、MOT16公开数据集上的实验结果证明了所提方法的有效性,与当前一些主流算法相比较,本文方法取得了相当或者领先的跟踪效果.关键词多目标跟踪,深度学习,度量学习,关联模型,多级关联引用格式侯建华,张国帅,项俊.基于深度学习的多目标跟踪关联模型设计.自动化学报,2020,46(12):2690−2700DOI10.16383/j.aas.c180528Designing Aﬃnity Model for Multiple Object Tracking Based on Deep LearningHOU Jian-Hua1ZHANG Guo-Shuai1XIANG Jun1Abstract While deep learning has made a breakthrough in many sub-ﬁelds of computer vision recently,there are only a few deep learning approaches to multiple object tracking(MOT).Since the key component in detection based multiple object tracking is to design a robust aﬃnity model,this paper proposes a novel aﬃnity model based on deep neural network and metric learning,that is,metric learning,a widely used technique in the task of person re-identiﬁcation(Re-ID),is exploited with convolutional neural networks(CNNs)to design the object s appearance model.Speciﬁcally,we adopt a three-channel CNNs that is learned by triplet loss function,to extract the discriminative appearance features and compute appearance similarity between objects.The appearance aﬃnity is then combined with motion model to estimate associating probability among trajectories.A hierarchical association strategy is employed by the Hungarian algorithm. At the low level,a set of short but reliable tracklets are generated in a frame by frame fashion.These tracklets are then further associated to form longer tracklets at the higher levels via an adaptive sliding-window mechanism.Experiment results in the challenging MOT benchmark demonstrate the validity of the proposed pared with several state-of-the-art approaches,our method has achieved competitive or superior performance.Key words Multiple object tracking(MOT),deep learning,metric learning,aﬃnity model,multi-level association Citation Hou Jian-Hua,Zhang Guo-Shuai,Xiang Jun.Designing aﬃnity model for multiple object tracking based on deep learning.Acta Automatica Sinica,2020,46(12):2690−2700多目标跟踪(Multi object tracking,MOT)是计算机视觉领域的一个研究热点,在视频监控、自动驾驶、机器人导航、行为分析等领域发挥着重要作收稿日期2018-08-02录用日期2019-01-09Manuscript received August2,2018;accepted January9,2019国家自然科学基金(61671484,61701548),湖北省自然科学基金(201 8CFB503),中南民族大学中央高校基本科研业务费专项资金项目(CZ Q17001,CZZ18001,CZY18046)资助Supported by National Natural Science Foundation of China (61671484,61701548),Hubei Provincial Natural Science Foun-dation of China(2018CFB503),and Fundamental Research Funds for the Central Universities,South-Central University for Nationalities(CZQ17001,CZZ18001,CZY18046)本文责任编委桑农Recommended by Associate Editor SANG Nong1.中南民族大学电子信息工程学院武汉4300741.College of Electronic Information Engineering,South-Central University for Nationalities,Wuhan430074用[1].近10年来,随着检测器性能的不断提升[2−4],基于检测的多目标跟踪算法[1,5−6]受到了广泛关注.这类方法基本流程如下:由离线检测器提供视频序列每一帧中各目标的位置(即检测响应),跟踪算法的任务是将这些检测响应与其对应的目标进行关联,最终得到每个目标完整的运动轨迹.基于检测的多目标跟踪包括两个主要模块:关联模型、关联状态推理(即优化策略),本文主要研究前者.关联模型即目标间亲密度模型,用以计算下一帧检测响应与当前帧目标之间的连接概率(或者关联代价),为关联状态推理提供有效的依据.特征表达是关联模型设计的核心,其中最常用的是目标外观特征和目标运动特征.近20年来,研究者们提出了多种特征用于构建目标的外观模型,例如颜色直12期侯建华等:基于深度学习的多目标跟踪关联模型设计2691方图[7]、方向梯度直方图(Histogram of oriented gradient,HOG)特征[8]、协方差特征[9]等;外观特征在多目标跟踪过程中发挥着重要的作用,但是在拥挤场景、以及目标(例如行人)具有相似外观的场景,仅靠外观特征易导致错误关联.因此很多研究工作通过建模目标动态特性,将运动特征与外观特征相融合,构建更鲁棒的目标特征表达[6,10].上述手工设计的特征有力推动了多目标跟踪研究的发展,但自2015年以来更具挑战性的MOTChallenge[11−12]数据集的公开,手工设计的特征已经难以取得令人满意的效果.例如,目标间的严重遮挡、剧烈的光照变化和目标形变等,将可能够导致相同目标之间距离远大于不同目标之间的距离,造成目标之间的错误关联.近几年,深度神经网络(Deep neural networks, DNNs)因其强大的特征学习与表达能力,在图像分类[13]、目标检测[3]等计算机视觉经典领域的应用取得了突破性进展;深度学习在视觉目标跟踪(通常是单目标跟踪)领域也得到了深入研究[14−16].但基于深度学习的多目标跟踪研究却相对甚少,其主要原因包括[17−18]:1)多目标跟踪算法训练样本集少,难以满足神经网络需要大量训练样本集的要求;2)现有的深度神经网络大多是在图像分类数据集上做离线训练基础上得到的,其缺陷是难以分辨目标间的细微差异、难以捕捉到视频目标中的运动特征.以下对近年来提出的一些代表性方法做一个简要回顾.在现有的基于深度学习的多目标跟踪方法中,一种常见思路是采用孪生卷积神经网络(Siamese convolutional neural networks,Siamese CNNs)提取外观特征,判断两个检测响应是否属于同一条轨迹.Leal-Taix´e等[19]以一对检测响应的原始图像和光流图像为输入,由Siamese CNNs提取局部时空域特征,再根据两个检测响应之间的几何及相对位置变化提取上下文信息;采用梯度增强分类算法,结合局部特征与上下文信息,利用对比损失训练整个网络,得到两个检测响应之间的关联概率;最后采用匈牙利算法进行数据关联.Sadeghian等[20]同时考虑目标外观、运动、以及目标间相互作用机制,提出了一种基于神经网络和多线索特征融合的关联模型,利用二元损失函数(验证损失)分别训练一个Siamese CNNs和一个长短时记忆网络(Long short-term memory,LSTM)作为外观模型和运动模型.Tang等[21]将行人多目标跟踪任务视为行人再识别问题,通过融合行人姿态信息,设计基于Siamese CNNs的外观模型.值得指出的是,为了改善关联模型的鲁棒性,近年来在多目标跟踪研究中,已有将度量学习嵌入到神经网络的报道.Wang 等[22]提出了一种联合学习卷积神经网络和时域约束度量的轨迹片关联方法,首先在辅助数据集上离线训练Siamese CNNs,用该网络提取两个检测响应的外观特征,同时对轨迹片施加一个分段时域约束,构建多任务损失函数,由一个Mahalanobis矩阵和时域约束矩阵组成,通过同时训练上述两个矩阵,得到鲁棒的目标外观相似度;在此基础上,再结合传统的方法得到运动相似度,将轨迹片关联问题转换为广义线性分配问题,采用软分配算法[23]求其优化解.此外,Milan等[24]将递归神经网络(Recurrent neural networks,RNNs)引入到多目标跟踪,依据贝叶斯滤波理论,采用RNNs对复杂的目标动态特性进行建模,实现目标状态的预测、更新、以及处理新目标的出现和旧目标消失;由于在状态更新时需要将下一时刻的观察值与对应的目标相匹配(即数据关联),文献[24]设计了一个LSTM网络来完成此功能;需要指出的是,该方法仅依据目标运动特性实现了端到端网络学习和多目标跟踪,虽然没有达到主流算法的跟踪结果,但为基于深度学习的视频多目标跟踪的深入研究提供了许多新的、有价值的思路[18,25−26].行人再识别(Person re-identiﬁcation,Re-ID)和行人多目标跟踪任务之间存在相似性[27],即两者都需要判断两个目标是否属于同一人;但与多目标跟踪不同,Re-ID问题主要依靠外观特征,未能有效利用目标的运动特性.文献[28]对Re-ID问题中损失函数的研究表明,三元组损失的性能要优于文献[20−21]中使用的二元损失(验证损失).受以上启发,本文将深度卷积神经网络与度量学习相结合,提出了一种新的多目标跟踪关联模型:1)采用Re-ID 领域中广泛使用的度量学习技术设计外观模型,即利用三元组损失函数设计一个三通道卷积神经网络,提取更具判别性的外观特征构建目标外观相似度,再结合运动模型计算轨迹片间的关联概率;2)采用匈牙利算法[29]通过多级关联生成各目标轨迹.在2DMOT2015[12]、MOT16[11]公开数据集上的实验结果表明,与当前一些主流算法相比较,本文方法取得了相当的跟踪效果,在部分指标上取得了领先.1整体算法框架图1给出了本文方法的整体框架.视频序列输入至目标检测器,检测器输出每一帧中的目标检测响应.1)初级关联:利用训练好的外观模型提取检测响应的外观特征,以此计算相邻两帧检测响应之间的外观关联代价矩阵L A;通过计算相邻帧行人检测框中心点坐标的距离获得运动关联代价矩阵L M;由关联代价矩阵L=L A+L M和匈牙利算法,采用逐2692自动化学报46卷图1多目标跟踪方法整体框架Fig.1The overall framework of multi-object tracking method帧关联的方式得到短小可靠的轨迹片集合.2)多级关联:在初级关联基础上,依据轨迹片时域约束关系,采用自适应时间滑动窗机制生成轨迹片集合A 和B ;利用关联模型计算两个集合轨迹之间的关联代价,再由匈牙利算法进行最优关联分配.2外观模型目标外观相似性度量是多目标跟踪关联模型的核心,也是本文研究的重点.外观模型的设计要求提取具有判别能力的视觉特征,且满足正样本(相同目标)之间外观关联代价小(或相似度大)、负样本(不同目标、或者目标与背景)之间具有较大的外观关联代价.上述要求与行人再识别(Re-ID)任务是一致的[27],本文基于深度神经网络、采用在Re-ID 领域应用广泛的度量学习技术设计外观模型.2.1三通道外观模型本文采用文献[30]的方法,构建三通道卷积神经网络模型,如图2所示,由三个结构相同、权值参数共享的子网络Model-A 组成三个通道,三元组样本分别输入至三个通道,提取外观特征后,再由三元损失函数训练整个网络.子网络Model-A 结构如下:在深度残差网络ResNet-50[31]的基础上去掉最后一层,再增加两个全连接层,其中第一个全连接层节点数为1024,采用批量归一化处理(Batch normalization)[32]和ReLU [33]激活函数;第二个全连接层包含128个节点,不使用激活函数,输出128维特征向量.输入图像尺寸为256×128像素.训练外观模型包括两部分:首先学习一个映射f (x ),将输入样本图像x 映射到嵌入特征空间(Em-bedding feature space)R r [34],r 为嵌入特征空间的维数;再设计一个度量函数,使得R r 中正样本对之间的距离远小于负样本对.在图2中,子网络Model-A 对应映射f (x ),即特征提取;三元损失函数对应度量学习.图2三通道外观模型训练框图Fig.2Three-channel appearance model trainingblock diagram设Z i = x a i ,x p i ,x ni 表示第i 个三元组样本,Z i由三张输入图片构成,其中,x a i 为锚点样本;x pi 来自与x a i 相同的目标,x p i 与x ai 构成一对正样本;x n i来自与x a i 不同的目标或者背景,x n i 与x ai 构成一对负样本.采用三元组损失训练图2所示网络,其中对三元组样本约束如下[30]:f (x a i )−f (x p i ) 2+α< f (x a i )−f (x ni ) 2(1)设N 为三元组训练样本总数,则三元组损失定义为L trip=1N Ni =1f (x a i )−f (x pi ) 2− f (x a i )−f (x ni ) 2+α+(2)其中,(·)+=max(0,·).三元组损失学习要求所有负样本对的距离要大12期侯建华等:基于深度学习的多目标跟踪关联模型设计2693于正样本对的距离一个正的最小间隔α,即在训练过程中损失不断下降,使得锚点与正样本越来越接近,而锚点与负样本距离则越远.α取值越小,则在特征空间中锚点与正样本集不需要拉的太近,锚点与负样本集也不需要拉的太远,容易满足收敛条件;但由于正负样本距离没有被拉开,存在的风险是不能够很好地区分有歧义的数据.反之,α取值越大,可以较有把握地区分较为相似的图像;但由于在学习过程中要拉近锚点与正样本之间的距离,同时拉远锚点与负样本之间的距离,将给网络训练带来一定的困难,即损失易处于一种很大的状态,参数更新震荡严重,模型训练困难.因此,设置一个合理的α对于基于三元损失的网络训练至关重要.本文参考文献[30]将α设置为1.0.三元组损失能够在特征空间拉近相同目标之间的距离,同时增加不同目标之间的距离.经过三元组损失训练的CNN,能够提取具有判别能力的特征,保证相同目标之间的外观差距远小于不同目标之间的外观差距.2.2训练三元组样本数据的生成是训练外观模型的关键之一.为保证训练样本的多样性,本文在多目标跟踪数据库2DMOT2015[12]和MOT16[11]训练集中的Ground truth的基础上,还使用了行人再识别数据库Market-1501[35]和CUHK03[36].需要说明的是, MOT16训练集中的视频MOT16-02作为实验部分的验证集,不用于本文训练.由于MOT训练数据集(Ground truth,GT)和测试数据集中的检测响应(Detection,DT)存在一定的差异性,因此直接用GT数据训练网络模型,在测试集上应用往往得不到理想的精度.针对此问题,本文采用在GT数据集上人为加入高斯噪声的方法来模拟DT的数据产生过程,即通过适量抖动GT行人检测框大小(即改变中心点位置及尺寸)的方式,产生与DT尽可能相似的分布数据训练网络模型.针对存在许多遮挡严重的视频,本文采用文献[10]提出的遮挡推理策略,通过估计每个运动目标特定时刻的可见度,将遮挡较大(可见度小于0.5)的目标剔除训练集,不参与模型训练.上述训练数据共包括3824个目标,以Batch 为单位生成三元组样本.首先从训练数据中随机选取5个不同目标作为锚点x a i,每一个锚点随机挑选对应的正样本x p i和负样本x n i,得到三元组 x a i,x p i, x ni.每一个目标对应20个三元组,一个Batch中包含100个三元组,用于外观网络模型的一次正向传播.使用TensorFlow深度学习框架进行网络训练,初始化权重来源于文献[37],使用AdamOpti-mizer[38]方法更新网络参数,最小化三元组损失.初始学习率设置为0.0001,每迭代10000次,学习率衰减10%.实验中迭代600000次时,网络已趋于收敛.在测试阶段,只使用一个子网络Model-A,依次输入各帧检测响应,Model-A输出其外观特征.3基于外观和运动特性的轨迹关联依据外观模型可以得到外观关联代价,但由于多目标跟踪场景的复杂性,不同目标也可能具有相似的外观,此时仅靠外观特征易导致错误关联.本文在外观模型的基础上,引入运动模型,利用目标空域信息提高关联的准确度.另一方面,为了降低计算复杂度,首先进行逐帧初级关联方式,得到可靠的短小轨迹片;再采用自适应时间滑动窗对短小轨迹片进行多级关联,得到完整的目标运动轨迹.以下分初级关联、以及其后的多级关联两种情况,分别介绍关联代价的计算.3.1初级关联3.1.1外观关联代价依次将各帧检测响应输入至外观网络Model-A,Model-A输出对应的外观特征.设当前帧目标集合为D t,下一帧检测响应集合为D t+1,计算两个集合中当前帧(第t帧)目标i的检测响应d t i和下一帧(第t+1帧)目标j的检测响应d t+1j之间的外观关联代价L A(d ti,d t+1j)=f(d ti)−f(d t+1j)2(3)需要说明,如果当前帧目标为轨迹片,则d t i为该轨迹片中置信度最高的检测响应.3.1.2运动关联代价依据检测响应框的中心点坐标,得到目标和检测响应的位置差,以此计算运动关联代价L M(d ti,d t+1j)=p ti−p t+1j2(4)其中,p t i,p t+1j分别为检测响应d t i,d t+1j的中心点二维坐标矢量.如果当前帧目标为轨迹片,则d t i为该轨迹片中的最后一帧检测响应.最终的初级关联代价为L(d ti,d t+1j)=L A(d ti,d t+1j)+L M(d ti,d t+1j)(5) 3.2多级关联3.2.1自适应时间滑动窗机制通过初级关联得到短小可靠的轨迹片后,本文使用自适应时间滑动窗机制对其做进一步关联,最2694自动化学报46卷终输出各目标完整的运动轨迹.其工作机理是采用自适应滑动时间窗的方式不断构建可关联轨迹片集.这里可关联轨迹片集是指两个轨迹片集合A与B,其中,A定义为当前时刻t s已跟踪目标轨迹集合,B 是滑动窗内待关联的轨迹片集合,集合B的构建满足:1)B中的轨迹集与A中任意轨迹不存在时间交集(也即可关联);2)B中任意两轨迹片间存在时间交叠.这里自适应窗口大小的含义是指从t s开始滑动时间窗口,窗口大小选取为满足集合B定义的第2个条件的最大值.设t s时刻A,B集如图3(a)所示,其中集合A为已跟踪轨迹,集合B中虚线框的宽表示滑动窗自适应时间跨度,其选取原则是恰好使得B集合里的所有轨迹片集满足“任意两两轨迹片彼此存在时间交叠”,如果继续增加滑动窗大小,也即B虚线宽增加,将11帧和12帧轨迹片包含到集合B,则B中轨迹片T1与第11帧轨迹片T2不满足集合B的第2个约束条件.在当前滑动窗机制下生成A,B集以后,采用匈牙利算法实现可关联轨迹片之间的最优关联状态.如图3(b)所示,将B中被关联的轨迹片与A中轨迹串联并加入到集合A(例如图中虚线所示),同时B中没有关联上的轨迹片也作为新目标轨迹加入A 集(例如图中轨迹片T3),至此完成了集合A的更新.下一时刻的滑动窗从t s=10时刻开始,重新按照自适应滑动方式构建新的集合B,直至滑动完整个视频集,算法结束.采用自适应滑动窗技术可以部分解决遮挡问题,假设因为遮挡导致某目标检测响应丢失,但自适应滑动窗可以考虑具有时间跨度的轨迹片间的相关性,即利用该目标在无遮挡阶段的检测数据进行关联.但基于滑动窗的数据处理模式存在一定的时间延迟,不能实现逐帧实时输出,无法满足对实时性要求较高的应用场景.3.2.2外观关联代价设T i,T j分别代表集合A,B的一条轨迹片,其外观关联代价为L A(T i,T j)= f(T i)−f(T j)2(6)其中,f(T i),f(T j)对应轨迹片中置信度最高的检测响应的外观特征.3.2.3运动关联代价采用文献[39]的方法构建多级关联的运动模型.首先定义轨迹片运动关联概率为g(T i,T j)=G(∆p1,Σ)G(∆p2,Σ)(7)其中,G(·,Σ)代表零均值高斯分布,Σ为高斯分布的方差.多级关联运动模型如图4所示,设p headj为轨迹片T j第一帧检测框的中心点位置(对应时刻记为t headj),v headj表示T j所对应目标j的运动速度;p taili 代表轨迹片T i最后一帧检测框的中心点位置(对应时刻记为t taili),v taili是轨迹片T i所对应目标i的运动速度.假设目标做匀速运动,∆t=t headj−t taili表示T j第一帧响应与T i最后一帧响应的时间间隔,则目标i在时刻t headj的预测位置为p taili+v taili·∆t,该预测值与待关联目标j实际位置的误差为∆p1=p taili+v taili×∆t−p headj(8)同理,∆p2=p headj−v headj×∆t−p taili(9)假定∆p1和∆p2服从高斯分布,则预测位置与待关联目标实际位置的误差越小,对应的轨迹片之间的运动相似度就越大.由式(7)计算运动关联概率,再由下式得到对应的运动关联代价为L M(T i,T j)=−ln(g(T i,T j))(10)最终的关联代价为L(T i,T j)=L A(T i,T j)+L M(T i,T j)(11)图3自适应时间滑动窗原理示意图Fig.3Diagram of adaptive time sliding window principle12期侯建华等:基于深度学习的多目标跟踪关联模型设计2695图4多级关联中的运动模型示意图Fig.4Diagram of motion model in multi-levelassociation4实验结果及分析本节首先介绍多目标跟踪性能评价指标;然后通过剥离对比实验(Ablation study)证明外观模型、运动模型及三元组损失的有效性;同时,在行人数据库MOTChallenge [11−12]测试集上与当前主流跟踪算法进行了比较分析,并给出了直观的可视化跟踪结果.最后在UA-DETRAC 车辆数据集上,进一步验证本文算法的有效性.本文实验的硬件环境:CPU 为Intel Xeon(R)************GHz,内存为128GB.4.1评价指标按照国际通行方法,采用文献[11,40]定义的标准来评估多目标跟踪算法的性能,包括:MT (↑):大部分被跟踪的目标.针对某一个目标,实际轨迹长度占该目标真实轨迹长度的百分比大于80%,则认为该目标被大部分跟踪.ML (↓):大部分未被跟踪的目标.针对某一个目标,实际轨迹长度占该目标真实轨迹长度的百分比小于20%,则认为该目标大部分未被跟踪.FP (↓):被预测为正的负样本,即虚检.FN (↓):被预测为负的正样本,即漏检.Frag (↓):轨迹断裂的次数.即还存在能关联,而没有被关联上的短小轨迹片.IDS (↓):针对某一条轨迹,其对应的目标身份发生变化的次数.MOTA (↑):多目标跟踪准确率.综合FP 、IDS 和FN 计算,是衡量多目标跟踪算法性能的主要指标.MOTP (↑):多目标跟踪精度.根据检测响应与真实数据的行人框的重合率计算,是衡量多目标跟踪算法性能的主要指标.在上述指标中,↑表示该指标越大性能越好,↓则表示该指标越小性能越好.4.2剥离对比实验及分析为了验证本文外观模型、运动模型在多目标跟踪中的作用,以MOT16-02作为验证集,设计了剥离实验(Ablation study).同时为了比较三元组损失与常规的二元损失的差异性,我们在不改变外观模型网络结构的前提下,设计并使用Siamese CNN,计算一对检测响应之间的关联概率,定义二元交叉熵损失来训练外观模型.实验结果如表1所示,其中A 表示外观模型,M 表示运动模型,T 表示使用三元组损失,V 表示使用二元交叉熵损失,加粗数据表示最好结果.从表1可以看出:1)外观特征在多目标跟踪过程中发挥主要作用.在使用三元组损失时,仅使用运动特征时,MOTA 为17.6%;而仅使用外观特征时,MOTA 为19.5%.上述结果与实际情况是相符的,因为在现实场景中,人眼主要通过目标衣着的颜色、形状等外观视觉特征来进行跟踪;同时卷积神经网络(CNN)具有很好的提取视觉结构化信息的能力.2)外观和运动特征相结合能够有效提升跟踪精度.同时使用外观和运动特征,得到的MOTA 最高,为21%.在复杂场景下,目标之间易出现遮挡;另一方面,距离较远的不同目标之间也有可能具有相同的外观.针对这些情况,仅使用外观特征不能得到准确的跟踪.本文结合运动模型,利用位置信息对关联进行约束,与外观特征协同发挥作用.3)与二元损失(验证损失)相比,使用度量学习技术中的三元组损失能够显著提升跟踪精度,表1剥离对比实验结果Table 1Results of ablation studyTrackers MOTA (↑)MOTP (↑)MT (↑)(%)ML (↓)(%)FP (↓)FN (↓)IDS (↓)A +T 19.574.67.4166.701091420243M +T 17.674.67.4064.803071432670A +M +T 21.074.39.2670.401751389316A +M +V14.775.11.8567.0060148043392696自动化学报46卷MTOA由14.7%上升到21.0%,说明了将行人再识别领域中的三元组损失应用于多目标跟踪的可行性.4.3与主流算法的对比表2给出了本文算法与当前主流多目标跟踪算法在MOT16测试集上的结果对比.从表2可以看出,本文算法在MOTA、MT、FP、FN指标上优于之前文献[41−44]中的跟踪算法.本文算法得到的MOTA为43.1%,低于文献[20]的AMIR方法.文献[20]同时考虑了目标外观、运动、以及目标间相互作用机制,并且采用LSTM网络建模目标在时域的非线性运动特性,而本文方法仅采用了传统的线性运动模型,且未能提取时域运动信息,这也是今后待改进之处.另外一点值得说明的是,本文算法的IDS优于文献[17,20,41−42,44−45]的方法,IDS 越小,说明跟踪轨迹的目标身份变化次数少,跟踪结果的可靠性高.表3列出了在2DMOT2015测试集上的结果对比.从表3可以看出,与基于孪生卷积神经网络(Siamese CNNs)计算关联概率的算法[19,22]相比,本文算法在MOTA上具有较大的优势.本文方法框架与文献[22]类似,都采用度量学习技术和神经网络提取外观特征,同时利用了传统的运动模型,最后用匈牙利算法进行轨迹关联;但不同的是本文采用三元组损失训练网络,用一个CNN提取外观特征.而文献[22]采用分段时域约束构建多任务损失函数训练Siamese CNNs,并用于外观特征提取.同时,本文三元组损失只对类间差距有约束,而对类内差距未做要求,因此文献[22]训练过程中的约束条件更强.但本文方法在实际中能获得更好的跟踪性能,与文献[22]相比,本文得到的MOTA提高了4.6个百分点(从29.6%到34.2%),说明了在这种遮挡严重的场景下,适当放松训练中的约束条件,可以使得模型更容易收敛,在测试时能够获得理想的跟踪效果.同时也说明了基于三元损失函数的三通道卷积表2MOT16测试集结果Table2Results of MOT16test setTrackers Mode MOTA(↑)MOTP(↑)MT(↑)(%)ML(↓)(%)FP(↓)FN(↓)IDS(↓)HZ(↑) AMIR[20]Online47.275.814.041.6268192856774 1.0 CDA[45]Online43.974.710.744.46450951756760.5本文Online43.174.212.447.74228990574950.7 EAMTT[41]Online38.875.17.949.1811410245296511.8 OVBT[42]Online38.475.47.547.3115179946313210.3 Quad-CNN[17]Batch44.176.414.644.9638894775745 1.8 LIN1[43]Batch41.074.811.651.3789699224430 4.2 CEM[44]Batch33.275.87.854.468371143226420.3表32DMOT2015测试集结果Table3Results of2DMOT2015test setTrackers Mode MOTA(↑)MOTP(↑)MT(↑)(%)ML(↓)(%)FP(↓)FN(↓)IDS(↓)HZ(↑) AMIR[20]Online37.671.715.826.87933293971026 1.9本文Online34.271.98.940.67965316657940.7 CDA[45]Online32.870.79.742.2498335690614 2.3 RNN LSTM[24]Online19.071.0 5.545.611578367061490165.2 Quad-CNN[17]Batch33.873.412.936.9787932061703 3.7 MHT DAM[46]Batch32.471.816.043.89064320604350.7 CNNTCM[22]Batch29.671.811.244.0778634733712 1.7 Siamese CNN[19]Batch29.071.28.548.451603779863952.8 LIN1[43]Batch24.571.3 5.564.65864402072987.5。

基于i-Tree_Eco模型的校园树木生态效益评估研究

Key words campus trees; morphological structure analysis; i-TreeEco model analysis校园树木在改善校园环境和增加生态效益方面有着重要作用，校园树木也是国内外众多研究者研究的方向，对校园树木的生态效益研究也较广泛。

国外学者Jayasooriya 等[1]利用i-Tree Eco 模型模拟建立绿色设施空间，分析出街道绿化与空间立体绿化结合，更有利于环境与人居和谐。

国内学者也相继关注i-Tree 模型在城市生态效益评价方面的研究，马宁等[2]分析了UFORE 和STRATUM 两个模块的研究及应用前景，从而为城市森林管理提供科学依据。

文章为探究上海应用技术大学徐汇校区校园树木生态景观效益，运用i-Tree Eco 模型来量化固碳释氧效益、雨水截留效益及净化空气质量改善效益，将量化值转换为货币值，评估徐汇校区生态效益情况，以期扩充国内生态景观效益数据[3]。

各类生态效益价值按i-Tree 软件自动设定价格汇率1美元=7.27元转化为人民币。

1研究方法1.1研究区域概况上海应用技术大学徐汇校区面积约11.3万 m 2，位于中国东部上海市，属于亚热带季风气候，全市年均日照时间约为1 886 h，年均降雨量为1 103 mm，蒸发量为1 257 mm，雨季多集中在5月至9月[4]。

1.2研究步骤研究主要有参数设置、可靠数据采集、数据结果分析三个过程[5]。

在模型中创建项目，由于研究地点在上海徐汇，且模型已存在的气象与污染数据最新为2014年的数据，摘要为评估上海应用技术大学徐汇校区校园树木生态效益，在对校区树木调研及形态测量基础上，文章运用i-Tree Eco 模型来量化固碳释氧、净化空气、雨水截留效益，为校区生态景观研究提供数据。

结果显示：共调查校区内323棵树，树高20＜h ≤25区间数量最多，为101棵，胸径多样性指数为1.30，指数值较高，有利于树木种群内部的更替生长。

参考文献——精选推荐

参考文献[1]何俊士，林洪孝.水资源概论[M].北京：中国农业大学出版社，2006，15．[2]张洪．蒸馏法脱盐技术的现状和发展趋势[J]．水处理技术，1991，17（16）：345-354．[3]解利昕，李凭力．海水淡化技术及各种淡化方法评述[J]．化工进展，2003，22（10）：1081-1084．[4]邹家庆．工业废水处理技术[M]．北京：化学工业出版社，2003，146-151．[5]唐受印，戴友芝，汪大翚，等．废水处理工程[M]．北京：化学工业出版社，1998，122-125．[6]陈秀芳．离子交换法在废水处理中的应用[J]．科技情报开发与经济，2004，14（7）：148-149．[7]Atungulu G，Koide S，Sasaki S，et al．Ion-exchange membrane mediated electrodialysis of scallopbroth：Ion，free amino acid and heavy metal profiles[J]．J．Food．Eng．，2007，78：1285-1290．[8]Gong Y an，Tang Y u，Wang Xiaolin，et al．The possibility of the desalination of actual 1，3-propanediol fermentation broth by electrodialysis[J]．Desalination，2004，161：169-l78．[9]黄万抚，罗凯，李新冬．电渗析技术研究现状与进展[J]．过滤与分离，2003，13（4）：20-24．[10]李广，梁艳玲，韦宏．电渗析技术的发展及应用[J]，化工技术与开发，2008，37（7）：28-30．[11]龚毅忠，张健春．反渗透除盐技术的应用及技术[J]．工业水处理，2002，22（7）：39-41．[12]L D Nghiem，A I Scher．Critical risk point of nanofiltration and reverse osmosis process in waterrecyling applications[J]．Desalination．2006，187：303-312．[13]侯运升，鹿笃实，范俊欣．纳滤法除盐研究[J]．膜科学与技术，1998，18（4）：30-33．[14]Schaep J，vander Bruggen B，V andecasteele C，et al．Influence of ion size and charge innanofiltration[J]．Sep．Purif．Technol．，1998，14（1-3）：155-162．[15]曾坚贤．膜技术制备饮用水的应用综述[J]．南京农专学报，2000，16（1）：23-27．[16]王晓林，张澄洪，赵杰．纳滤技术的应用现状[J]．南京化工大学学报，2000，22（3）：68-71．[17]Ban A，Schafer A，Wendt H.Journal of Applied[J].Electrochemistry，1998，28（3）：227-236.[18]陈福明.一种去除水中离子的新方法—充电富集法初探[J].化工学报，1999，50（1）：114-117.[19]尹广军，陈福明．电容去离子研究进展[J]．水处理技术，2003，29（2）：63~66．[20]RY oo Mirrwoo，SEO Gon．Improvement in capacitive deionization function of activated carboncloth by titania modification[J]．Water Research，2003，37：1527~1534．[21]Arnold B B，Murphy G W．Studies on the electrochemistry of carbon and chemically modifiedcarbon surfaces[J]．J．Phy．Chem．，1961，65（1）：135-138．[22]Caudle D D，Tucker J H，Cooper J L，et al．Electrochemical demineralization of water with carbonelectrodes[R]．Research and Development Progress，Report No．188，US Department of the Interior，May，1966：190．[23]Johnson A M，Newman J．Desalting by means of porous carbonelectrodes[J]．J．Electrochem．Soc．，1971，118（3）：510-517．[24]Oren Y，Soffer A．Electrochemical parametric pumping[J]．J．Electrochem．Soc．，1978，125（6）：869-875．[25]Farmer J C，Fix D V，Mack G V，et al．Capacitive deionization of NH4ClO4 solutions with carbonaerogel electrodes[J]．J．Appl．Electrochem．，1996，26（10）：1007-1018．[26]Farmer J C，Fix D V，Mack G V，et al．Capacitive deionization of NaCl and NaNO3 solutionswithcarbon aerogel electrodes[J]．J．Electrochem．Soc．，1996，143（1）：159-169．[27]Farmer J C．Method and apparatus for capacitive deionization,electrochemical purification，andregeneration of electrodes[P]．US：5425858，1995-06-20．[28]Andelman M D．Flow-through capacitor[P]．US：5192432，1993-03-09．[29]Andelman M D．Flow through capacitor and method[P]．US：6778378，2004-08-17．[30]Schultze J W，Rolle D．The electrosorption valency and charge distribution in the double layer．Theinfluence of surface structure on the adsorption of aromatic molecules[J]．Journal of Electroanalytical Chenistry，2003，552：163-169．[31]Ryoo M W，Seo G．Improvement in capacitive deionization function of activated carbon cloth bytitania modification[J]．Water Rsearch，2003，37：1527-1534．[32]Y ang C M，Choi W H，Na B K，et al．Capac itive deionization of NaCl solution with carbonaerogel-silica gel composite electrodes[J]．Desalination，2005，174：124-133．[33]Jeon Sang Hoon，Kim Chang-Koo，Shin Chee Burm，et al．Modeling of a capacitivedeionizationprocess[C]．205th Meeting of The Electrochemical Society，Mary，2004．[34]Welgemoed，Schutte T J，Frik C．Capacitive Deionization Technology[trademark]：Analternativedesalination solution[J]．Desalination，2005，183：327-340．[35]张登松，代凯，方建慧，等．多壁纳米碳管电极电吸附脱盐性能的研究[J]．功能材料，2005，36（2）：282-287．[36]秦仁喜，沈军，周斌，等．炭气凝胶电极用于NaCl溶液电容性除盐的研究[J]．水处理技术，2005，31（4）：53-55．[37]杨慧云，王岐东，李斌．活性炭纤维电极用于NaCl溶液电容性除盐[J]．北京工商大学学报（自然科学版），2006，24（2）：9-12．[38]刘海静，张鸿涛，孙晓慰．电吸附法去除地下水中离子的试验研究[J]．中国给排水，2003，19（11）：36-38．[39]孙晓慰，朱国富．电吸附水处理技术及设备[J]．工业水处理，2002，22（8）：1-4．[40]赵雪娜，倪文，林庚，等．电吸附技术在含盐水中的应用与研究进展[J]．工业水处理，2008，28（3）：5-8．[41]B．E．康维（陈艾等译）．电化学超级电容器—科学原理及技术应用[M]．北京：化学工业出版社，2005．[42]郭鹤桐，覃奇贤．电化学教程[M]．天津：天津大学出版社，2009：9．[43]Helmholtz H L．Ann[J]．Physik，1879，337（7）．[44]Kovalenko A．Molecular description of electrosorption in a nanoporous carbon electrode[J]．Journalof Computationnal and Theoretical Nanoscience，2004，1（4）：398-411．[45]Brockris J M．Comprehensive of electrochemistry[M]．New Y ork Plenum Press，1981：93．[46]刘海静．电吸附技术在水淡化/软化中的应用性实验研究[D]．北京：清华大学环境科学与工程系，2003．[47]查全性．电极过程动力学导论（第3版）[M]．北京：科学出版社，2002，31-38．[48]刘永辉．电化学测试技术[M]．北京：北京航空学院出版社，1978．[49]田昭武．电化学研究方法[M]．北京：科学出版社，1986．[50]Bladergroen B J，Y ang K，Ying T，et al．Electrosorption of ions from aqueous solutions by carbonaerogel an electrical double layer model[J]．Langmuir，2001，17（6）：1961-1969．[51]Huang Jingsong，Sumpter B G，Meunier V．A universal model for nanoporous carbonsupercapacitors applicable to diverse pore regimes carbon materials and electrolytes[J]．Chemistry，2008，14（22）：6614-6626．[52]Gabelich C J，Tran T D，Suffet I H．Electrosorption of nanostructured carbon-basedmaterials[J]．Environmental Science ang Technology，2002，36（13）：3010-3019．[53]陈兆林，宋存义，孙晓慰，等．电吸附除盐技术的研究与应用进展[J]．工业水处理，2011，31（4）：11-13．[54]查振林，余以雄．电吸附技术及其在水处理中的应用[J]．四川化工，2005，8（4）：52-54．[55]Zabasajja J，Savinell R F．Electrosorption of n alcohols on graphite particles[J]．Aiche，1989，35（5）：755-763．[56]Eisinger R S，Alkire R C，Separation by electrosoprtion of organic compounds in a fiow throughporous eleetrodeⅡ．Experimental validation of model[J]．Electrochem．Soc．，1983，130（1）：93-101．[57]Ahmad Alfarra，Elzbieta Frackowiak，Francois Beguin．Mechanism of lithium electrosorption byactivated carbons[J]．Electrochimica Acta，2002，47：1545–1553．[58]Abbas Afkhami，Brian E Conway．Investigation of Removal of Cr（VI），Mo（VI），W（VI），V（IV），and V（V）Oxy-ions from Industrial Waste-Waters by Adsorption and Electrosorption at High-Area Carbon Cloth[J]．Colloid and Interface Science，2002，251：248–255．[59]Zou Linda，Morris G，Qi Daoduo．Using activated carbon electrode in electrosorptive deion isationof brackish w ater[J]．Desalination，2008，225：329-340．[60]Zou Linda，Li Lixia，Song Huaihe，et al．Using mesoporous carbon electrodes for brackishwaterdesalination[J]．Water Res．，2008，42：2340-2348．[61]董晓宁，王亚丽．电吸附技术在水处理中的应用及其研究进展[J]．广东化工，2010，37（8）：26-27．[62]Y ue Xu，John W Zondlo，Harry O Finklea，et al．Electrosorption of uranium on carbon fibers as ameans of environmental remediation[J]．Fuel Processing Technology，2000，68：189-208．[63]Chen Rong，Xu Xien．Elecsorp tion of thiocyanate anions on activated carbon felt electrode in dilutesolution[J]．Colloid and Interface Science，2005，290：190-195．[64]Oh Hanjun，Lee Jongho，Ahn Hongjoo，etal．Nanoporous activated carbon cloth for capacitivedeionization of aqueous solution[J]．Thin Solid Fimls，2006，15：220-225．[65]Han Y anhe，Quan Xie，Chen Shuo，et al．Electrochemical enhancement of adsorption capacityofactivated carbon fibers and their surface physico chemical characterizations[J]．Electrochin Acta，2007，52：3075-3081．[66]Conway B E．电化学超级电容器—科学原理及技术应用[M]．北京：化学工业出版社，2005：577．[67]李定龙，申晶晶，姜晟，等．电吸附除盐技术进展及其应用[J]．水资源保护，2008（24）：63-66．[68]Famer J C，Pekala R W，Fix D V，et al．Capacitive deionization with carbon aerogel electrodes：carbonate，sulfateand phosphate[C]．International SAMPE Technical Conference，1995，27：294-304．[69]Pekala R W，Farmer J C，Alviso C T，et al．Carbon aeroges for electrochemicalapplications[J]．Non-Crystalline Solids，1998，225（1）：74-80．[70]Gabelich J，Tran T D，Suffet I H．Electrosorption ofinorganic salts from aqueous solution usingcarbon aerogels[J]．Environ Sci Tech，2002，36：3010-3019．[71]Ying Tungyu，Y ang Kunlin，Yiacoumis，etal．Electrorsorption of ions from aqueous so lutions bynanostructured carbon aerogel[J]．Colloid Interface Sci，2002，250：18-27．[72]Y ang Chun-Mo，Lee Gi Taek，Na Byung-Ki，et al．Capacitive deionization of NaCl solution usingfour-series cells with carbon aerogel composite electrodes[C]．2004 Joint International Meeting-206th Meeting of the Electrochemical Society，Japan，2004．[73]Y ang C，Choi W，Na B，et al．Capacition of NaCl solution with carbon aerogel-silicagel compositeelectrodes[J]．Desalination，2005，174（2）：125-133．[74]王万兵，张睿，刘希邈等．酚醛-糠醛基炭气凝胶对NaCl溶液的电吸附行为[J]．新型炭材料，2007，22（1）：65-69．[75]Iijima S．Helical microtubules of graphtic carbon[J]．Nature，1991，354：56-58．[76]Ozoemena K I，Pillay J，Nyokong T．Preferential electrosorption of cobalt（II）tetraram inophthalocyanineat singlewall carbon nanotubes immobilized on a basal planepy ro lytic graphiteelectrode[J]．Electrochem Commun，2006，8：1391-1396．[77]Zhang Dengsong，Shi Liyi，Fang Jianhui，etal．Preparation and desalination performance ofmultiwall carbon nanotube[J]．Mater Chem Phys，2006，97：415-419．[78]Dai Kai，Shi Liyi，Fang Jianhui，etal．NaCl adsorption in multiwalled carbon nanotubes[J]．MaterLett，2005，59：1989-1992．[79]Wang Shuo，Wang DaZhi，JI LiJun，et al．Equilibrium and kinetic studies on the removal of NaClfrom aqueous solutions by electrosorption on carbon nanotube electrodes[J]．Sep Purif Techno，2007，58：12-16．[80]施利毅．液流式电吸附脱盐装置的碳纳米管电极的制造方法[p]．ZL0410016261．9．[81]Xu Pei，Drewes J，Kim T U，et al．Effect of membrane fouling on tansport of organic contaminantsin NF/RO membrane applications[J]．Journal of membrane Science，2006，297（2）：165-175．[82]Ania C O，Beguin F O．Mechanism of adsorpti on and electrosorption of bentazone on activatedcarbon cloth in aqueous solutions[J]．Water Res．，2007，41：3372-3380．[83]Shi K，Shiu K．Adsorption of copper and lead species at electrochemically activated glassy carbonelectrodes[J]．Electrochimica Acta，2006，51（13）：2631-2638．[84]宋序彤，印明善．苦咸水淡化[M]．北京：科学出版社，1983．[85]陈兆林．操作参数对电吸附设备除盐的影响[D]．江苏工业学院，2006．[86]许可．电吸附法在污水会用工程中的应用研究[D]．清华大学，2008．[87]赵雪娜，倪文，陈兆林．结垢对电吸附除盐的影响极其控制[J]．过程工程学报，2008，8(5)：887-891．[88]陈兆林，孙晓慰，郭洪飞，等．电吸附技术处理首钢污水厂二级出水的中试研究[J]．中国给水排水，2010，26(9)：115-119．[89]赵雪娜，孙晓慰，郭洪飞，等．重碳酸盐碱度对电吸附设备除盐性能影响的研究[J]．环境工程学报，2008，2(5)：647-651．[90]石国乐，张凤英．给排水物理化学[M]．北京：中国建筑工业出版社，1997．81．[91]Hou C H，Liang C D.Electrosorption Capacitance of Nanostructured Carbon-based Material[J]．J.Colloid Interface Sci，2006，302(1)：54-61．[92]陈兆林，孙晓慰，朱广东，等．电吸附设备工作过程的研究[J]．环境工程学报，2009，3(7)：1307-1310．[93]赵雪娜，孙晓慰，郭洪飞，等．酸化吹脱去除重碳酸盐对电吸附设备性能的影响[J]．工业水处理，2009，29(1)：23-25．。

纳米催化二氧化碳制甲醇英文

纳米催化二氧化碳制甲醇英文Nanocatalysis for the Production of Methanol from Carbon Dioxide.Carbon dioxide (CO2) is a significant greenhouse gas that contributes to global warming. However, converting it into useful chemicals such as methanol offers a sustainable and environmentally friendly approach to mitigate its adverse effects. Nanocatalysis, a field that utilizes nanoscale materials to catalyze chemical reactions, has emerged as a promising technology for this purpose.Nanocatalysis Principles and Applications.Nanocatalysis leverages the unique properties of nanomaterials, including their large surface area and high reactivity, to enhance catalytic activity. These nanomaterials, often in the form of nanoparticles or nanostructures, can significantly improve the rate and selectivity of chemical reactions. In the context of CO2conversion, nanocatalysts can lower the activation energy required for the reaction, making it more energetically favorable.CO2 to Methanol Conversion.The conversion of CO2 into methanol involves a multi-step process known as the methanol synthesis. Typically, this process requires high temperatures and pressures, as well as a suitable catalyst. Nanocatalysts cansignificantly reduce these requirements, making the process more energy-efficient and cost-effective.The most common nanocatalysts used for CO2 hydrogenation to methanol are based on copper. Copper nanoparticles, due to their high activity and selectivity, are particularly effective in promoting this reaction. Other metals, such as palladium and platinum, have also been explored for this purpose.Nanocatalyst Design and Optimization.The design and optimization of nanocatalysts for CO2 conversion are crucial for achieving high catalytic performance. Factors such as particle size, shape, and composition can significantly influence the catalytic activity. For instance, smaller nanoparticles typically exhibit higher catalytic activity due to their increased surface area. Similarly, the choice of support material can also affect the stability and activity of the nanocatalyst.Challenges and Future Prospects.While nanocatalysis offers significant potential for CO2 conversion, several challenges need to be addressed. One of the main challenges is the scalability of nanocatalysts for industrial applications. Current methods for synthesizing nanomaterials are often not suitable for large-scale production. Additionally, the stability of nanocatalysts under reaction conditions is also a concern, as they can often deactivate or agglomerate over time.Future research efforts should focus on developing more stable and scalable nanocatalysts for CO2 conversion.Innovations in nanomaterials synthesis and characterization techniques can help address these challenges. Furthermore, integrating nanocatalysts with other renewable energy sources, such as solar or wind power, can further enhance the sustainability of the process.In conclusion, nanocatalysis holds promise for the efficient conversion of CO2 into methanol. By leveraging the unique properties of nanomaterials, we can develop more effective and sustainable catalysts for this important reaction. Future research in this area could lead to significant advancements in green chemistry and help mitigate the impact of climate change.。

基于CFD 数值仿真的飞行器气动系数计算方法

航空航天科学技术基于CFD数值仿真的飞行器气动系数计算方法刘岩松杜宇凡∗高婷（沈阳航空航天大学辽宁沈阳110136）摘要：为便捷地求取飞行器的气动力系数，本文以Bryan所提出的刚性对称飞行器运动方程为理论基础，该方程能直观方便地求出所需的力及力矩。

通过仿真得到各方向速度分量后，拟合系数，构建气动力模型方程。

在小扰动状态时，根据纵向及横侧向运动状态下得出的气动力系数构建方程，所得出的结果与CFD仿真模拟结果接近。

为验证模拟方法的准确性，将所求得的数据与风洞试验数据进行比对，其误差符合要求，说明在设计初期运用该方法求得设计机体的气动力系数是可行的。

关键词：CFD仿真平面拘束运动网格划分三维建模中图分类号：V221.3文献标识码：A文章编号：1674-098X(2021)09(c)-0009-03Aerodynamic Coefficient Calculation Method of Aircraft Based onCFD Numerical SimulationLIU Yansong DU Yufan∗GAO Ting(Shenyang Aerospace University,Shenyang,Liaoning Province,110136China) Abstract:In order to obtain the aerodynamic coefficients of the aircraft conveniently,this paper takes the rigid symmetrical aircraft motion equation proposed by Bryan as the theoretical basis.The equation can intuitively and conveniently obtain the required force and moment.After the velocity components in each direction are obtained by simulation,the fitting coefficient is used to construct the aerodynamic model equation.In the small disturbance state,the equations are constructed based on the aerodynamic coefficients obtained in the longitudinal and lateral motion states,and the obtained results are close to the CFD simulation results.In order to verify the accuracy of the simulation method,the obtained data is compared with the wind tunnel test data,and the error meets the require‐ments.It shows that it is feasible to use this method to obtain the aerodynamic coefficient of the designed engine block in the early stage of design.Key Words:CFD simulation;Planar constrained motion;Meshing;3D modeling飞机动力学建模是基于气动、动力、结构、总体等提供的飞行器数据库，构建飞机动力学模型，探寻其飞行力学规律，并指导制导控制系统设计。

parametric

Never enough建筑界如今使用Parametric的方式很肤浅和表面化, 缺少一个大的可做参的理论筐架, 表述, 历史和内部的动力. 并且鄙视和抵制那种后现代的符号化,那种将所有的视觉的东西转化成有隐喻内涵的设计方法(顺便提一句,上海世博会中国馆就完全属于被鄙视的这类). 最近的建筑界正在朝着后--后现代发展,朝着一种逃避符号化的象征发展. 如果考虑到parametric是一种重复的单元化的数量图形, 我会argue现在的parametric work可以看作是一种后现代的演进, 它的索引参考就是它自身和与它相似系统的东西. 如今建筑界对与parametric的应用,除了看多了让人麻木的复杂程度,还没有发掘到parametric真正的巨大的潜力, 比如说关联的多义的processing 方法, 复杂功能的需求, 场地的特殊环境,以及关联的network. 今天如果有什么东西想看起来很parametric的话, 通常是一种审美的production(被重复的quality和taste). 那些无比高科技的软件多是用来制造华丽的建筑decoration. 通过扭曲系统来作为下一个系统的参照.基于计算机技术本身的复杂特性,或者那种创造震撼视觉效果的渴望, 目前的parametric设计通常降低大form的多样性,但是在form本身内部最大化它的多样性(这句话是什么意思呢?我想是指比如说做一个盒子,他不会去做很多大小不同的盒子,但是会做同样大小但盒子自身复杂度很不同). 如今的parametric的审美很大程度上是基于American Formalism（新形式主义）, 是将功能主义教条当做积极的和天真的乌托邦（空想的不切实际的）融入到对它自己有利的位置.Parametric design是基于高度清晰的组织系统(diagram,parti等)和高度视觉复杂, 但是无论你parametric系统的单元有多复杂,系统多有逻辑,多么的parametric, 最终总是不可避免的是一个碎片,损坏的孤立的物体, 就算它身处于field中,或者它本身就是个field,都不能改变这点. Parametric系统需要differentiation来变成一个建筑. 是Social-Political允许parametric系统逃脱了作为物体的空洞. 建筑需要social engagement,需要culture/social relevance. 它不是parametric 或者form,或者自身的复杂性,but rather在设计建筑时的复杂的复杂关系.我们所感兴趣的parametric design是Pierre Bourdieau所说的”field of cultural production”, where form是在更大的culture field里的小field. 建筑的特殊性和建筑的最终目的只能当建筑置身于大的culture field中才能实现. 建筑已经不是马克思的”我们和他们”(二元对立)那种对立的辨证法模型. 在建筑中only 关键within the market culture, 建筑必须和它自身之外的事物联系才会变的重要.相似的是,我一直在考虑的是form和material的辨证法对立. 这是一种产生与后现代主义的对立法,where form和构成form的材料相比更关键. 但是这种distinction本身太简单和教条. 复杂parametric系统可以包含和调解这种虚伪的form VS material 二元. Parametric系统提供了我们一种不同的基于parametric的方法,but 它永远不会告诉我们那些parametric是建筑需要的.就想这本书里其他的作品一样,我们都是罪有应得的formalists,对于form异常的感兴趣.对于我们来说,百折不挠的formalism建筑叙事已经进化了,so that form的专门用途比form本身更重要. However, 这种专门用途决不应该是简单的功能主义, but rather 反对简单化和唯我论的建筑设计.”用途”是指建筑的performance:performance的两面,有益的和夸张的.今天, formal游戏(与80 和90年代不同)需要一个目的,而不是一直钻在它自己里面. 现在建筑缺少一种中心枢纽来被反对或者赞同.如果没有一个边界,那么form 的可能性就会无穷无尽,lost在自我表达里或者meaningless的建筑自己的field里.建筑决不是简单的form. Formal的最基本是需要和它自身之外联系的.有了这种联系,意义才可以被构建的locally and relationally.建筑的说教已经太过于和自己的建造绑在一起,而不是它的culture social situation. 当然建筑技术本身是文化的一部分,但是我们太过遗忘了剩下的别的部分. 建筑的目标变的如此的非政治化,于是建筑变成了一个个小范围理想. 而这种小范围的理想是不可以被说教的. 建筑的performance并不是建筑的基本问题.建筑更多的是culture social form,而不是技术结果.建筑是超级暧昧的, 关联的, parametric的, 但是它原比这些复杂,因为它超出了我们的掌握.并不是所有的事情都是简单数量,并不是所有的关系都是geometric,并不是所有的问题都能被关系所包含.传统上,先锋建筑的敌人是品位和质量,错误的文化表达方式.而我们在寻找一种specific与它本身的东西. 它不许要解释原本的culture,但是culture在起着重要的作用.我们感兴趣的是新的设计的方法.最终的,parametric系统吸引我们的恰恰是它所不包含的, 那就是Social political. Parametric的潜力在于创造parameters和relationships的network, the more 多样的目标the more 有意义和复杂.At the very least,Use的叙述可以给我们提供form的依据.但是这种功能的叙述又决不是重做一种后后功能主义,它反对功能和形式的二元对立或形式追随功能的教条.Use是一种叙述结构. 构建叙述的有利性给我们提供了一种逃脱反复的唯我论和空洞的formalism,提供form的指导,可以包括parametric的表现,教导或者意义.这些culture, social, formal的叙述是建筑有价值.Use使form不再是一个空洞的object,它提供了建筑物体与外界的联系,并且回应外界.通过叙述use,意义被表达,建筑被适合于它自己的场所.但是上面讲的这些如何来影响我们的设计方法呢?说实话我不确定.我们写script,我们用parametric 软件,我们构建系统,并且控制它们(这里讲了他们事物所里工作的方式,我会在后面的case study里具体讲)到目前,parametric design还没有任何的统治思想,我们工作的方式,使用的软件,思考的方法始终在不断变化,我们不断尝试接触新的媒介,新的方法,并且parametric modeling确实给我们提供了极大的帮助, 而且我们确实尝试着as parametric as we can.再一次,建筑的未来是属于parametric的,我丝毫不怀疑, 但是问题是科技不能解决我们自身的问题,而且我们自身的问题又much more deeper and much more human. 建筑只能在直面我们的文化或成为我们social culture network里面一部分的时候才能有意义. Parametric design可以变的更开放,更自由,少一些自我,允许新的model that is not build upon 那种形式和功能的二元法,而是更加紧密的联系social, 创造一个前瞻的utopia.(乌托邦)。

Generalizedadditivemixedmodels

Generalized Additive Mixed ModelsInitial data-exploratory analysis using scatter plots indicated a non linear dependence of the response on predictor variables. To overcome these difficulties, Hastie and Tibshirani (1990) proposed generalized additive models (GAMs). GAMs are extensions of generalized linear models (GLMs) in which a link function describing the total explained variance is modeled as a sum of the covariates. The terms of the model can in this case be local smoothers or simple transformations with fixed degrees of freedom (e.g. Maunder and Punt 2004). In general the model has a structure of:Where and has an exponential family distribution. is a response variable, isa row for the model matrix for any strictly parametric model component, is the correspondingparameter vector, and the are smooth functions of the covariates, .In regression studies, the coefficients tend to be considered fixed. However, there are cases in which it makes sense to assume some random coefficients. These cases typically occur in situations where the main interest is to make inferences on the entire population, from which some levels are randomly sampled. Consequently, a model with both fixed and random effects (so called mixed effects models) would be more appropriate. In the present study, observations were collected from the same individuals over time. It is reasonable to assume that correlations exist among the observations from the same individual, so we utilized generalized additive mixed models (GAMM) to investigate the effects of covariates on movement probabilities. All the models had the probability of inter-island movement obtained from the BBMM as the dependent term, various covariates (SST, Month, Chlorophyll concentration, maturity stage, and wave energy) as fixed effects, and individual tagged sharks as the random effect. The GAMM used in this study had Gaussian error, identity link function and is given as:Where k = 1, …q is an unknown centered smooth function of the k th covariate andis a vector of random effects following All models were implemented using the mgcv (GAM) and the nlme (GAMM) packages in R (Wood 2006, R Development Core Team 2011).Spatially dependent or environmental data may be auto-correlated and using models that ignore this dependence can lead to inaccurate parameter estimates and inadequate quantification of uncertainty (Latimer et al., 2006). In the present GAMM models, we examined spatial autocorrelation among the chosen predictors by regressing the consecutive residuals against each other and testing for a significant slope. If there was auto-correlation, then there should be a linear relationship between consecutive residuals. The results of these regressions showed no auto-correlation among the predictors.Predictor terms used in GAMMsPredictor Type Description Values Sea surface Continuous Monthly aver. SST on each of the grid cells 20.7° - 27.5°C Chlorophyll a Continuous Monthly aver. Chlo each of grid cells 0.01 – 0.18 mg m-3 Wave energy Continuous Monthly aver. W. energy on each of grid cells 0.01 – 1051.2 kW m-1Month Categorical Month the Utilization Distributionwas generated January to December (1-12)Maturity stage Categorical Maturity stage of shark Mature male TL> 290cmMature female TL > 330cmDistribution of residual and model diagnosticsThe process of statistical modeling involves three distinct stages: formulating a model, fitting the model to data, and checking the model. The relative effect of each x j variable over the dependent variable of interest was assessed using the distribution of partial residuals. The relative influence of each factor was then assessed based on the values normalized with respect to the standard deviation of the partial residuals. The partial residual plots also contain 95% confidence intervals. In the present study we used the distribution of residuals and the quantile-quantile (Q-Q) plots, to assess the model fits. The residual distributions from the GAMM analyses appeared normal for both males and females.MalesResiduals distribution ResidualsF r e q u e n c y-202402004006008001000120-4-2024-2024Q-Q plotTheorethical quantilesS a m p l e q u a n t i l e sFemalesHastie, T.J., and R.J. Tibshirani. 1990. Generalized Additive Models. CRC press, Boca Raton,FL. Latimer, A. M., Wu, S., Gelfand, A. E., and Silander, J. A. 2006. Building statistical models toanalyze species distributions. Ecological Applications, 16: 33–50. Maunder, M.N., and A.E. Punt. 2004. Standardizing catch and effort: a review of recentapproaches. Fisheries Research 70: 141-159. Wood, S.N. 2006. Generalized Additive Models: an introduction with R. Boca Raton, CRCPress.。