A heterogeneous sparsity knowledge guided evolutionary algorithm for sparse large-scale multiobjective optimization

Research on sparse large-scale multiobjective optimization problems (LSMOPs) is rapidly growing due to their diverse applications in science and engineering. Existing studies typically employ static or dynamic knowledge to guide the search in evolutionary algorithms for solving sparse LSMOPs. However, relying solely on a single type of sparsity knowledge may result in ambiguous guidance and suboptimal optimization. To address this, we propose an evolutionary algorithm based on heterogeneous sparsity knowledge (HSKEA). In this approach, static knowledge is represented by a scoring vector to assess the importance of each decision variable, while dynamic knowledge is captured by indicator vectors that identify whether a decision variable is zero or non-zero and iteratively updates based on the population distribution. Two types of populations, including main and auxiliary populations, are initialized using both dynamic and static knowledge and evolved through a new genetic operator guided by heterogeneous sparsity knowledge and information sharing. Experimental results comparing HSKEA to four state-of-the-art algorithms across eight benchmark test problems and three real-world scenarios demonstrate its effectiveness.