Evaluation of ANN Training Methods: A Comparative Study of Back Propagation, Genetic Algorithm, and Particle Swarm Optimization for Predicting Electrical Energy Consumption

Abstract

This study compares the performance of ANN with three training methods: Backpropagation (BP), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO) in a simple classification case. The results show that ANN GA has the smallest average error (0.0308), followed by ANN BP (0.0569), while ANN PSO is much larger (0.7559). Thus, ANN GA proved to be the most stable and accurate, ANN BP still performed quite well, while ANN PSO had the weakest performance.