A robust deep fuzzy framework for load forecasting in integrated energy systems

To address the complex coupling among integrated energy loads and the limited forecasting accuracy of conventional single-interval type-2 fuzzy systems, on the basis of existing deep fuzzy system methods using type-1 fuzzy logic, this paper proposes a deep interval type-2 fuzzy system (DIT2FS) method using a nonlinear Kalman filter algorithm. First, in the data preprocessing stage of load forecasting, convergent cross mapping extracts key features from multi-source variables, effectively capturing nonlinear interactions between loads and meteorological factors while reducing redundancy and uncertainty. Second, interval type-2 fuzzy sets are introduced to enhance the ability of the model to capture uncertainty and complex nonlinear relationships. Within the framework of deep fuzzy systems, hierarchical modeling of multi-layer fuzzy subsystems improves the expressive power and interpretability of the model, while overcoming the shortcomings of traditional type-1 fuzzy systems in modeling uncertainty. Furthermore, the cubature Kalman filter is employed to perform recursive updates of model parameters, enabling dynamic parameter optimization and thereby enhancing robustness and online adaptability of the model. To validate the effectiveness of the proposed method, under identical experimental conditions, it was applied to a multi-load forecasting task in an integrated energy system and compared with existing deep convolutional fuzzy system and long short-term memory. The experimental results demonstrate that the proposed DIT2FS method has better prediction accuracy than the comparative methods in different seasons and indicates a reasonable model interpretability.