Multi-Load Power Forecasting Based on a Novel LSTM-KAN Network

Accurate power load forecasting is essential for the stable operation and long-term planning of power systems. This paper presents a long-term forecasting method for multi-load transformers based on a long short-term memory - Kolmogorov Arnold network (LSTM-KAN). The proposed model integrates the temporal feature extraction capabilities of LSTM networks with the adaptive attention mechanism of the KAN, enabling it to effectively capture the diverse and complex patterns of multi-load behaviors. Simulation results demonstrate that the model significantly outperforms traditional bidirectional LSTM and Transformer models, achieving superior forecasting accuracy and robustness, especially in scenarios with fluctuating load patterns.