Cooperatively Coevolving Particle Swarms for Large Scale Optimisation

在这个项目中，我们将开发新的协作式协同进化粒子群算法来解决大规模优化问题，特别是具有高维和不可分离性的问题。虽然已经有一些用于优化的协作式协同进化算法(CCEA)的工作，但这些协作式协同进化算法在处理不可分离的高维问题(例如，具有100个或更多实值变量)时很快就失效了。对于这类问题，迫切需要更有效的问题分解策略。我们将开发自适应分解策略，能够将大型问题分解为子组件，其中不同子组件之间的相互依赖保持在最低限度。这些更有效的分解策略将被结合到粒子群优化(PSO)算法中，以增强PSO处理高维不可分问题的能力，这是PSO目前非常薄弱的领域。经典的PSO算法在低维问题上表现得很好，但在高维问题上表现得很差。通过将协同进化框架与粒子群算法模型相结合，有望开发出更有效的解决大规模问题的粒子群算法。对不同的自适应分解策略和协作协进化粒子群算法(CCPSO)进行了深入的理论分析和计算研究。我们将使用维达1000(实值变量)的可分离和不可分离的基准函数，对所提出的CCPSO算法与其他现有CCEA算法进行全面比较。这将使我们能够确定我们提出的算法在处理这类特定问题时的优势和劣势。为了进一步评估所提出的CCPSO算法的性能，将使用形状优化中的实际应用程序。这项研究的预期结果不仅将使进化计算和群体智能社区的研究人员受益，也将使现实世界优化的实践者受益。

In this project, we will develop novel cooperative coevolutionary particle swarm algorithms for solving large scale optimisation problems, especially problems characterised by high dimensionality and non-separability. Although there have been some work on cooperative coevolutionary algorithms (CCEAs) for optimisation, these CCEAs break down quickly when dealing with non-separable high dimensional problems (e.g., with 100 or more real-valued variables). For this class of problems, more effective problem decomposition strategies are urgently needed. We will develop adaptive decomposition strategies capable of decomposing a large problem into subcomponents where the interdependencies among different subcomponents are kept at minimum. These more effective decomposition strategies will then be incorporated into a Particle Swarm Optimisation (PSO) algorithm to enhance PSO's ability in handling highdimensional non-separable problems, an area that PSO is currently very weak in.Classical PSO algorithms have been shown to perform well on low dimensionalproblems, but poorly on high dimensional problems. By combining a cooperativecoevolutionary framework with a PSO model, more effective PSO algorithms forlarge scale problems are expected to be developed. We will carry out in-depth theoretical analysis and computational studies of different adaptive decomposition strategies and cooperative coevolutionary PSO algorithms (CCPSO). Comprehensive comparisons between proposed CCPSO algorithms and other existing CCEAs will be carried out using both separable and non-separable benchmark functions with dimensions up to 1000 (real-valued variables). This will allow us to identify the strengths and weakness of our proposed algorithms in handling this particular class of problems. To evaluate further the performance of proposed CCPSO algorithms, a real-world application in shape optimisation will be used. The expected outcomes of this research will benefit not only researchers in the evolutionary computation and swarm intelligencecommunities, but also practitioners in real-world optimisation.