Abstract:

To address such problems as the insufficient consideration of the interaction of meteorological factors, and the limitations of model's nonlinear expression ability, a power load forecasting method based on Kolmogorov-Arnold network (KAN) and bidirectional long short-term memory (BiLSTM) network utilizing composite factor construction is proposed. Firstly, the power load curves with similar characteristics are classified through the Gaussian mixture model. Secondly, a composite factor construction strategy is proposed: the linear correlation degree between meteorological factors and loads is quantified through Pearson correlation analysis; key meteorological variables are screened out and interaction terms are constructed to fully explore the potential interaction among meteorological factors, and nonlinear dependent features are further extracted with maximum information coefficient. Finally, aiming at the problem that the fully connected layer of the traditional BiLSTM model has limited ability to learn high-dimensional nonlinear features, the KAN is introduced to replace the fully connected layer, and a KAN-BILSTM hybrid prediction model is constructed using its nonlinear mapping ability. The experimental results demonstrate that the proposed method has high prediction accuracy and universality under four different load modes. It can provide a feasible solution for the precise prediction of power load in multi-meteorological coupling scenarios.

Key words: load forecasting, composite factor construction, bidirectional long short-term memory network, Kolmogorov-Arnold network, nonlinear characterization