Short-Term Net Load Forecasting Model Based on GRU Neural Network Optimized by Improved Arithmetic Optimization Algorithm

This paper develops a short-term net load forecasting model based on optimized GRU model with improved algorithm. The GRU model, a variant of recurrent neural networks, addresses the gradient vanishing problem in long-term time series prediction by introducing gating mechanisms. To further improve prediction accuracy, we employed an improved arithmetic optimization algorithm, capable of achieving global optimization, to train the weights of the GRU neural network and obtained the optimal solution through iterative updates. Simulation results based on real-world power system net load data demonstrate that the proposed model, which combines the GRU neural network with the improved arithmetic optimization algorithm, delivers superior forecasting accuracy compared to conventional methods.