Comprehensive Energy Load Forecasting Method Based on Improved Deep Convolutional Fuzzy System

To address the short-term integrated energy load forecasting problem, this study proposes a prediction method based on an improved deep convolutional fuzzy system (IDCFS). DCFS is a hierarchical fuzzy system that uses the Wang–Mendel (WM) algorithm for model parameter tuning. To further enhance its prediction accuracy and enable real-time online training of model parameters, an improved DCFS is designed by incorporating the cubature Kalman filter (CKF) algorithm. The proposed IDCFS method is applied as a neutral prediction model for short-term integrated energy load forecasting. Validation using a public dataset of integrated energy systems demonstrates that, under identical conditions, the IDCFS method outperforms comparative methods such as LSTM, GRU, the interval Type-1 TSK fuzzy method (TSK-A1C1), and the original DCFS. The results indicate that the IDCFS method effectively combines the strengths of fuzzy systems and deep learning, achieving higher prediction accuracy and exhibiting promising application potential.