Scholarlink

Particle Swarm Optimization (PSO) is appreciated for its simplicity and ease of adaptation, yet its progress may stall when the swarm fails to identify promising regions early in the search. We present Neural-Network-Predicting Adaptive PSO (NNP-APSO), a variant that embeds an artificial neural network trained online to approximate the lower envelope of the objective landscape. At each iteration the network forecasts a candidate minimum, which is injected into the velocity update of under-performing particles; the swarm therefore shifts its exploration–exploitation balance automatically, without introducing additional control parameters. NNP-APSO is evaluated on the standard single-objective test suite (15 classical benchmarks) and compared, under identical computational budgets, with six established optimizers: Simulated Annealing, canonical PSO, Whale Optimization Algorithm, Grey Wolf Optimizer, Genetic Algorithm and Artificial Bee Colony. The experimental study indicates that NNP-APSO offers a promising alternative, delivering good and noteworthy results while highlighting the potential of real-time neural guidance within the PSO framework.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Neural-Network Guided Minima Forecasting for an Enhanced Particle Swarm Optimizer

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tri Ton That,
  • Binh Le-Van,
  • Thanh Cuong-Le

摘要

Particle Swarm Optimization (PSO) is appreciated for its simplicity and ease of adaptation, yet its progress may stall when the swarm fails to identify promising regions early in the search. We present Neural-Network-Predicting Adaptive PSO (NNP-APSO), a variant that embeds an artificial neural network trained online to approximate the lower envelope of the objective landscape. At each iteration the network forecasts a candidate minimum, which is injected into the velocity update of under-performing particles; the swarm therefore shifts its exploration–exploitation balance automatically, without introducing additional control parameters. NNP-APSO is evaluated on the standard single-objective test suite (15 classical benchmarks) and compared, under identical computational budgets, with six established optimizers: Simulated Annealing, canonical PSO, Whale Optimization Algorithm, Grey Wolf Optimizer, Genetic Algorithm and Artificial Bee Colony. The experimental study indicates that NNP-APSO offers a promising alternative, delivering good and noteworthy results while highlighting the potential of real-time neural guidance within the PSO framework.