Otimização com muitos objetivos por evolução diferencial aplicada ao escalonamento dinâmico de projeto de software

Abstract

Software Engineering problems often involve problems with many objectives and constraints, in most cases conflicting with each other. One trend toward solving these problems is the use of search and optimization algorithms to find solutions that automatically balance these objectives. In this work, we investigate a problem in the area of software planning, namely, the Software Project Scheduling Problem (SPSP), which aims to allocate people to tasks in a software project in order to optimize some objectives, such as project cost and duration. There are two main variations to this problem: static and dynamic. In static SPSP, the planning is done only at the beginning of the project, and the objectives to be optimized are project cost and duration. The dynamic model, called DSPSP, considers that the software project environment is susceptible to uncertainties, and the project may need to be rescheduled throughout the software development cycle. In dynamic approach, many objectives need to be optimized, such as cost, duration, stability and robustness of the schedule, to deal with the changes that may occur during the project development cycle. The dynamic model is still few explored in the literature. This work proposes an extension of the existing dynamic model in the literature, by considering two more dynamic events and the influence of team experience. The main focus of the work is the investigation of the suitability of the algorithm of optimization with many objectives by di erential evolution to the dynamic software project scheduling problem, considering the proposed model. Since the DSPSP involves dynamic optimization, six variants of the di erential evolution algorithm were investigated, each of them comprising one or more dynamic optimization techniques. The di erential evolution algorithm and its variants were compared to the evolutionary algorithm NSGA-III, also not yet explored for DSPSP. For the analysis of the algorithms investigated a battery of experiments was carried out. The results suggest that the di erential evolution algorithm with dynamic optimization techniques provides a better solutions for DSPSP.