Information interaction among particles constitutes a fundamental mechanism in particle swarm optimization( PSO). To address limitations in the efficiency of information interaction and enhance the performance of PSO ...Information interaction among particles constitutes a fundamental mechanism in particle swarm optimization( PSO). To address limitations in the efficiency of information interaction and enhance the performance of PSO in complex optimization landscapes,an elite-sharing and rank-based learning( ESRBL) particle swarm optimization( ESRBL-PSO) framework was proposed in this paper. Departing from the classical PSO framework,where particles primarily interact with the global best information,ESRBL-PSO employs a hierarchical population architecture.Specifically,the original swarm is divided into multiple subpopulations of equal size,each yielding a locally optimal particle( designated as a local elite). These local elites are then aggregated into a shared elite pool,enabling shared information transfer across the entire population. During the updating phase,each particle not only interacts with information within its own subpopulation but also selects a local elite competitively from the shared elite pool for additional information interaction. This dual mechanism amplifies the diversity of information interaction during swarm evolution, facilitating the rapid identification of high-potential regions in expansive search spaces.Furthermore,to mitigate sensitivity to parameters,ESRBL-PSO eliminates all parameters in the particle velocity update process and proposes an adaptive population division strategy. Synergistically,these features enable ESRBLPSO to maintain a balance between exploration diversity and convergence precision,thereby achieving effective optimization across complex problem domains. Finally,extensive experiments executed on CEC2017 benchmark suites verify that ESRBL-PSO exhibits competitive or even superior performance compared to 10 state-of-the-art approaches and maintains robust capability and scalability in handling complex numerical optimization problems.展开更多
文摘Information interaction among particles constitutes a fundamental mechanism in particle swarm optimization( PSO). To address limitations in the efficiency of information interaction and enhance the performance of PSO in complex optimization landscapes,an elite-sharing and rank-based learning( ESRBL) particle swarm optimization( ESRBL-PSO) framework was proposed in this paper. Departing from the classical PSO framework,where particles primarily interact with the global best information,ESRBL-PSO employs a hierarchical population architecture.Specifically,the original swarm is divided into multiple subpopulations of equal size,each yielding a locally optimal particle( designated as a local elite). These local elites are then aggregated into a shared elite pool,enabling shared information transfer across the entire population. During the updating phase,each particle not only interacts with information within its own subpopulation but also selects a local elite competitively from the shared elite pool for additional information interaction. This dual mechanism amplifies the diversity of information interaction during swarm evolution, facilitating the rapid identification of high-potential regions in expansive search spaces.Furthermore,to mitigate sensitivity to parameters,ESRBL-PSO eliminates all parameters in the particle velocity update process and proposes an adaptive population division strategy. Synergistically,these features enable ESRBLPSO to maintain a balance between exploration diversity and convergence precision,thereby achieving effective optimization across complex problem domains. Finally,extensive experiments executed on CEC2017 benchmark suites verify that ESRBL-PSO exhibits competitive or even superior performance compared to 10 state-of-the-art approaches and maintains robust capability and scalability in handling complex numerical optimization problems.
