Joint application of Crested Porcupine Optimizer and hybrid models in short-term wind power load forecasting

Abstract

With the large-scale grid integration of renewable energy sources such as wind power, their inherent uncertainty and prediction difficulty pose challenges to power grid dispatching. To address this, this paper proposes for the first time a combination of the Crested Porcupine Optimizer (CPO) with Temporal Convolutional Networks (TCN), Bidirectional Gated Recurrent Units (BiGRU), and self-attention mechanisms to construct a short-term wind power load forecasting model. Initially, TCN enhances the model’s ability to process time series data. Then, BiGRU captures bidirectional dependencies. Finally, the attention mechanism assigns weights to highly relevant information. Compared with traditional static parameter models, CPO dynamically adjusts the kernel size of TCN, the dimension of BiGRU hidden layers, attention weights, and regularization coefficients through the Cyclic Population Reduction (CPR) technique, significantly improving the model’s prediction accuracy. Compared with the unoptimized TCN-BiGRU-Attention model, our method reduces the Mean Squared Error (MSE) by 32.18%, and compared with the model optimized by the Grey Wolf Optimizer, the Root Mean Squared Error (RMSE) is reduced by 30.48%. The prediction error of CPO-TCN-BiGRU-Attention is significantly lower than that of other models, as confirmed by the Wilcoxon signed-rank test ($p$ $<$ 0.05), providing high-precision support for practical power grid dispatching.