An error correction ACKLSTM model combining data-driven and deep learning for wind power prediction

Abstract Accurate wind power forecasting is essential for improving renewable energy utilization efficiency and reducing uncertainties and fluctuations in power dispatch. To address this challenge, we propose a hybrid wind power prediction model that integrates a data-driven approach, an attention-based ConvLSTM (ACKLSTM) architecture, and an error correction mechanism. In this framework, the ACKLSTM model incorporates a self-attention (SA) mechanism built upon ConvLSTM and replaces the traditional multilayer perceptron (MLP) with a Kolmogorov-Arnold Network (KAN). Subsequently, a support vector machine (SVM) is employed to correct residual prediction errors, further enhancing forecasting accuracy. The proposed approach is validated through a comprehensive set of experiments. Results demonstrate that with an input sequence length of five time steps, the ACKLSTM model achieves the highest prediction accuracy, attaining an R 2 value of 95.70% ± 0.5. The comparative analysis demonstrates that the hybrid model, which integrates error correction and KAN, outperforms alternative methods, achieving the lowest mean squared error (MSE) of 0.091 ± 0.01 kW, representing a 70% improvement over the Transformer model. Furthermore, ablation experiments validate the effectiveness of both the error correction technique and the incorporation of KAN in improving model performance.