Random Matrix-Based Particle Swarm Optimization for Large-Scale Optimization

doi:10.1109/tbdata.2026.3679582

Return

Random Matrix-Based Particle Swarm Optimization for Large-Scale Optimization

2026-03-31

PRE

DOI：10.1109/tbdata.2026.3679582

Original

Original request for help

Save

Abstract

En 中文

Large-scale optimization problems (LSOPs) have attracted increasing attention in the Big Data era. Recently, matrix-based evolutionary computation (MEC) has been proposed as a new diagram for solving LSOPs within a short running time. However, compared with its corresponding classical non-matrix-based evolutionary computation (EC) algorithm, the MEC has not yet essentially improved its problem-solving ability or convergence speed. Therefore, how to enhance the problem-solving ability and convergence speed of MEC while maintaining its advantage of fast computational speed (i.e., short running time) has become a significant research topic. With this concern, this paper proposes a novel random matrix-based particle swarm optimization (RMPSO) algorithm, together with two novel designs. Firstly, a random matrix-based learning strategy is proposed, which enables each particle to learn from its superior particles by utilizing random matrices. Therefore, the RMPSO can effectively enhance the global search efficiency of the particles to improve problem-solving ability while maintaining fast computational speed. Secondly, a matrix-based knowledge and data-driven analysis strategy based on the dynamic system theory is proposed to analyze the algorithm convergence of RMPSO and help configure its parameters for fast convergence speed. To evaluate the proposed RMPSO, extensive experiments are conducted using 20 LSOPs, including 12 scalable problems with up to 10,000 dimensions and all the 8 multimodal LSOPs used in the latest IEEE CEC Large-Scale Global Optimization competition. Experimental results show that the RMPSO outperforms the compared state-of-the-art algorithms, including the champion algorithms from the LSGO competition, particularly on LSOPs with many local optima.

Keywords:

Large-scale optimization problem

matrix-based evolutionary computation

particle swarm optimization

Big Data