An Efficient Cooperative Co-Evolutionary Multitask Optimization Framework for Large-Scale Optimization
将协同进化框架与多任务优化结合，通过构建多任务问题和资源分配策略，提升大规模优化问题的求解效率，在基准测试和实际应用中表现优异。
Cooperative co-evolution (CC) framework is a classic decomposition-based method to solve large-scale optimization problems (LSOPs) by decomposing the original problem into several subproblems. CC is a single task paradigm that sequentially solves decomposed subproblems in a specific order, and it does not fully utilize similarities among these subproblems. Evolutionary multitask optimization (EMTO) employs the potential similarities and complementarities among distinct tasks to address multiple optimization tasks simultaneously through knowledge transfer mechanism. This study integrates the CC framework with the EMTO paradigm and proposes a cooperative co-evolutionary multitask optimization (CCMTO) framework for solving LSOPs. In the CCMTO framework, the original LSOP is redefined as a set of multitask optimization problems (MTOPs), and then the EMTO algorithm is used to solve them. To improve the optimization efficiency, this study proposes a construction strategy of multitask optimization problems and a contribution-based resource allocation strategy of MTOPs and subtasks. The construction strategy of multitask optimization problems can select the appropriate subproblems to construct MTOPs. The resource allocation strategy determines the optimization order of MTOPs based on their contribution to the improvement of the best fitness value, and reasonably allocates computational resources for each subtask. A multitask evolution strategy with dynamic distance threshold and adaptive elite sampling knowledge-guided external sampling (MTES-DAKG) is proposed and used to solve these MTOPs. Empirical results show that the proposed algorithm can significantly improve the optimization performance for solving LSOPs. Moreover, the proposed algorithm is superior to 14 state-of-the-art algorithms on 29 benchmark problems and performs well in real-world applications.