An Optimized-PSO Approach for Improving Elman Neural Network Performance in Time-Series Forecasting

The Elman Neural Network (ENN), a type of recurrent neural network, excels at modeling temporal dependencies and dynamic systems, making it well-suited for time-series prediction tasks. However, its performance heavily relies on the initialization of network parameters, which can result in slow convergence and suboptimal prediction accuracy. To address these limitations, we propose an Optimized-PSO-based Elman Neural Network (OP-ENN), where the traditional Particle Swarm Optimization (PSO) algorithm is enhanced to optimize the initial weights and thresholds of the Elman model. In our approach, several modifications are made to improve the convergence rate and avoid local optima: 1) Adaptive Inertia Weights and Learning Factors based on Nonlinear Variation, 2) Velocity Update Formula based on Second-order Oscillation Processing, and 3) Half-selection Logic. These improvements culminate in the Optimized-PSO algorithm, which is validated through two benchmark functions, showing significant improvements in convergence speed and accuracy over the conventional PSO. The OP-ENN model is then applied to a real-world problem of electrical load forecasting, demonstrating its superior performance in terms of both prediction accuracy and convergence speed. The results indicate that the proposed model is an effective solution for optimizing Elman networks, offering enhanced robustness and precision in time-series prediction tasks.