Integrating Kolmogorov–Arnold Networks with Time Series Prediction Framework in Electricity Demand Forecasting
Yuyang Zhang, Lei Cui, Wenqiang Yan
Abstract
Electricity demand is driven by a diverse set of factors, including fluctuations in business cycles, interregional dynamics, and the effects of climate change. Accurately quantifying the impact of these factors remains challenging, as existing methods often fail to address the complexities inherent in these influences. This study introduces a time series forecasting model based on Kolmogorov–Arnold Networks (KANs), integrated with three advanced neural network architectures, Temporal Convolutional Network (TCN), Bidirectional Long Short-Term Memory (BiLSTM), and Transformer, to forecast UK electricity demand. The analysis utilizes real-world datasets from a leading utility company and publicly available sources. Experimental findings reveal that the integration of KANs significantly improves forecasting accuracy, robustness, and adaptability, particularly in modeling intricate sequential patterns in electricity demand time series. The proposed approach addresses the limitations of traditional time series models, underscoring the potential of KANs as a transformative tool for predictive analytics.