Model-Driven Optimization in Software Engineering

各种软件工程问题都可以归结为软件模块化、软件测试、发布计划等优化问题。在基于搜索的软件工程(SBSE)中，元启发式方法被用于解决软件工程中的优化问题。迭代探索搜索空间的一种广泛使用的方法是进化算法。软件工程中的问题域通常是用向量或树编码的，因为进化操作符可以直接指定。当优化结果的质量不像预期的那样高时，这种影响的一个解释可能是在探索性搜索中没有充分捕获特定领域的知识。模型驱动工程(Model-Driven Engineering，MDE)提供了统一处理特定领域模型的概念、方法和技术。在SBSE中使用MDE被称为模型驱动优化(MDO)；它在文献中已在众所周知的优化问题中得到了演示。MDO很有前途，因为特定领域的知识可以系统地结合到SBSE中。为了加强MDO的愿景，本项目旨在巩固MDO，即为迄今所取得的结果发展科学基础，并加深对何时以及如何使用MDO来解决软件工程中的优化问题的理解。这个项目愿景可以分解为以下目标：(1)为MDO开发一个正式的框架，该框架定义了使用特定领域的知识来指定优化问题和进化算法的统一方法。该框架将用于澄清MDO中的概念和对进化算法的质量进行推理，以便开发人员能够做出明智的决策。(2)对MDO进行实证评价，考察其实践相关性。SBSE的两个主题领域已被确定用于此次评估，即突变测试和程序的遗传改进。作为这项评价的先决条件，将考虑到正式框架的所有实际相关概念和结果，为多学科发展目标开发一个综合工具环境。

A variety of software engineering problems can be considered as optimization problems such as software modularization, software testing, and release planning. In search-based software engineering (SBSE) meta-heuristic methods are applied to solve optimization problems in software engineering. One of the widely used approaches to iteratively explore a search space are evolutionary algorithms. The problem domains in software engineering are typically encoded with vectors or trees since evolutionary operators can be specified straightforwardly. When the quality of optimization results is not as high as expected, an explanation for this effect may be that domain-specific knowledge is not captured enough in the explorative search. Model-driven engineering (MDE) offers concepts, methods and techniques to process domain-specific models uniformly. The use of MDE in SBSE is called model-driven optimization (MDO); it has been demonstrated at well-known optimization problems in the literature. MDO is promising as domain-specific knowledge can be systematically incorporated into SBSE. To strengthen the MDO vision, this project aims to consolidate MDO, i.e., to develop a scientific basis for the results obtained so far and to obtain a deeper understanding when and how MDO shall be used to solve optimization problems in software engineering. This project vision can be broken down into the following objectives: (1) Develop a formal framework for MDO that defines a uniform approach for specifying optimization problems and evolutionary algorithms using domain-specific knowledge. The framework will be used for clarifying concepts and for reasoning about the quality of evolutionary algorithms in MDO such that developers can make informed decisions. (2) Perform an empirical evaluation of MDO to investigate its practical relevance. Two topical subject fields of SBSE have been identified for this evaluation, namely mutation testing and genetic improvement of programs. As a prerequisite for this evaluation, an integrated tool environment for MDO will be developed taking all concepts and results of the formal framework into account that are practically relevant.