Interpretable short-term electric load forecasting based on Bayesian Optimization for AM-CNN-KAN

With the continuous development of science and social progress, the difficulty of power system planning has gradually increased. Therefore, in order to reduce the difficulty of power system planning task, this paper uses the deep learning method to study the load forecasting problem, and proposes a hybrid forecasting model based on the combination of KAN (Kolmogorov Arnold networks) and CNN (convolutional neural networks), combined with Bayesian optimization algorithm and multi head attention mechanism. The model is named AM-CNN-KAN, and its evaluation indexes , MAE, MSE, RMSE, MAPE are improved by 0.072%, 7.52%, 2.44%, 1.23% and 7.10% respectively compared with the traditional model, the model also has hyperparameter adaptive optimization and good stability. In addition, in order to solve the black box problem in power load forecasting based on neural network, this study also added the SHAP model for interpretability analysis.