A bilevel coevolution framework with knowledge transfer for large-scale optimization and its application in multiperiod economic dispatch

Complex systems typically consist of multiple components and serve requirements across multiple periods. Their optimization involves large-scale parameters. If all parameters are considered at one time, the high dimensional searching space will present great challenge. Otherwise, if parameters are considered partially, the global fittest solutions can hardly be found. Keep these in mind, in this paper, a novel bilevel coevolution framework with knowledge transfer (BiKT) is introduced for large-scale multiobjective optimization. Specifically, this framework, the optimization problem is decomposed to several low-dimensional subproblems, establishing a bilevel structure. Then, the original problem and the subproblems are regarded as two tasks, their coevolution is realized by converting searching agents between the upper and lower optimization workflows, fulfilling exploration in global optimization and exploitation in local areas. Meanwhile, a knowledge transfer strategy is studied to adapt the search directions and accelerate convergence speeds. The superiority of the novel framework has been verified by experimental studies on large-scale benchmark problems, and the effectiveness of knowledge transfer has been discussed through ablation experiments. In the end, the proposed method employed to tackle a large-scale real-world challenge, the multiperiod economic dispatch problem in the power system. After problem formulation and analysis, the proposed method can perfectly solve the application.