Efficient Load Prediction by Incorporating Kolmogorov–Arnold Network and Convolutional Neural Network-Based Transformer Encoder Available to Purchase

Load prediction technology offers significant opportunities for ensuring optimal energy distribution, enhancing grid stability, and supporting sustainable power system management. It remains challenging due to the high requirement of accurately capturing the complex and dynamic relationships between historical load data and various influencing factors. This study proposes a novel approach that integrates the Kolmogorov–Arnold network (KAN) with a convolutional neural network (CNN)-based transformer encoder to achieve highly accurate and efficient load prediction. Specifically, KAN is designed to capture the intricate relationships between diverse influencing factors, including geographical features, user behaviors, and environmental conditions. The CNN-based transformer encoder leverages temporal dependencies and spatial correlations inherent in historical load data, providing robust predictions even under varying conditions. Through comprehensive experiments on real-world datasets, the proposed hybrid model demonstrates superior performance compared to traditional standalone deep learning architectures. The findings suggest that this combined methodology offers a viable solution for enhancing the accuracy and reliability of load prediction.