Short-Term Load Forecasting Using Chaos Theory and Generalized Regression Neural Network

This paper presents a method for short-term load forecasting that integrates chaos theory with generalized regression neural networks (GRNN), demonstrating satisfactory performance results. Chaos theory, as a novel mathematical approach, effectively characterizes complex dynamic behaviors in nonlinear power systems. The implementation typically involves phase space reconstruction using time-delay embedding techniques and calculating Lyapunov exponents to identify chaotic characteristics in load data. Meanwhile, GRNN serves as a machine learning algorithm specifically designed to handle nonlinear relationships through its probabilistic neural network architecture, featuring fast training convergence and inherent regularization properties. The integration methodology generally follows these steps: first applying chaos theory for system state identification and feature extraction, then utilizing GRNN's pattern recognition capabilities for prediction modeling. Key implementation aspects include using MATLAB's neural network toolbox for GRNN configuration with optimal spread parameter selection, and custom algorithms for chaos analysis including correlation dimension calculation. This combined approach enhances short-term load forecasting accuracy by adapting to varying load patterns and capturing complex nonlinear dynamics. The method's advantages lie in its adaptability to diverse load demands and significant improvement in prediction precision, making it particularly valuable for power industry applications where it facilitates better planning and management of electrical resources through reliable load forecasts.