基于SiPESC通用试验设计和代理模型算法构架

Algorithmic framework for experimental design and surrogate model based on SiPESC

基于面向服务的开放软件平台SiPESC,针对试验设计和代理模型共性特征,采用面向服务插件编程的设计方法和软件设计模式,研发了试验设计和代理模型通用算法构架。构架的核心思想是将算法、数据模型和任务管理相分离,形成独立的服务,从而实现算法通用性。整个构架中,试验设计和代理模型分别由五个基本服务构成。重点阐述基本插件所代理功能的抽象过程和通用接口的设计思想,给出算法构架的使用步骤及扩展方式。利用该构架试验设计开发了均匀试验设计、正交试验设计、析因试验设计及中心复合试验设计;代理模型实现了响应面模型、径向基函数模型及Kriging模型,并对部分算法进行数值验证。研究工作表明,算法构架适用于通用试验设计和代理模型算法,可方便进行动态扩展,具备良好的开放性和重用性。

Based on the service-oriented open platform SiPESC,a new general algorithmic framework for experimental design and surrogate model is proposed by using service-oriented programming technology and software design patterns method,which considers the common features of experimental design and surrogate model algorithms.The core idea of the framework is that data model and algorithmic are designed with loose coupling in service-oriented architectures and encapsulated in different services.The whole framework includes five basic services in both experimental design and surrogate model.The abstraction of the basic services and the design of common interface are introduced,then,the aggregation and extending of services in this framework is given by instance.Furthermore,four experimental design methods and three surrogate model methods are implemented,two numerical verifications are employed to demonstrate the characteristics of the implemented framework.This investigation shows that the algorithmic framework has favorable extensibility and reusability which is applicable to the general analysis algorithm with settled procedural input and smart tasks.