A Random Projection Based Evolutionary Strategy for Solving Large-Scale Optimization Problems

Limited Memory Matrix Adaptation Evolution Strategy (LM-MA-ES) is an important branch of evolutionary strategy that utilizes a limited set of direction vectors to generate candidate solutions. However, it still faces challenges when applied to large-scale optimization problems. To address this issue, we propose a framework named Random Projection-Based Evolution Strategy (RP-ES). The core idea is to map the original population from high-dimensional space to a low-dimensional space using a random projection matrix. The offspring solutions are then generated in this low-dimensional space and projected back into the high-dimensional space. Additionally, the high-dimensional offspring are evaluated and selected based on the objective function. Experimental results demonstrate that the proposed framework outperforms existing algorithms in both benchmark tests and complex problem scenarios.