Particle Swarm Algorithm Setting using Deep Reinforcement Learning in the Artificial Neural Network Optimization Learning Process

Recent meta-heuristics for optimization issues include particle swarm optimization (PSO) techniques. They take their cues from the coordinated movements of swarms of fish and birds, which exhibit collective behaviors. A sophisticated learning phase, like back-propagation, is needed for artificial neural networks (ANNs), which are thought of as a source of artificial intelligence (AI). This phase allows for the calculation of the error gradient for each neuron, from the final layer to the first. However, some qualities of the objective function are necessary (cost). This inspired us to experiment with meta-heuristics to streamline the training of ANNs to manage complex nonlinear systems. The objective of this research is to apply deep reinforcement learning (DRL) to automatically calculate the PSO algorithm’s parameters while also optimizing the supervised learning process of ANN. After a number of case studies, our methodology consistently leads to the coefficients of the ideal ANN. © (2024), (International Association of Engineers). All Rights Reserved.