Recurrence Plot-Based Channel Shift Network for Hydrogen Production Load Forecasting in Integrated Energy System

Hydrogen load forecasting plays a crucial role in the optimal scheduling of integrated energy systems and the reduction of carbon emissions. To address the limitations of existing load forecasting methods in modeling temporal dependencies, feature extraction, and capturing multivariate feature dependencies, this paper proposes a recurrence plot-based channel shift network. First, multidimensional time series data related to load are transformed into recurrence plots to reveal the dynamic characteristics of time series in a high-dimensional space. Then, a parallel convolutional neural network with shared parameters is constructed to extract features from different variables, improving computational efficiency and enhancing the model's generalization capability. To achieve deep fusion of features across different variables, this paper introduces a feature channel shift method, which shifts a portion of feature channels among different variables' recurrence plots. This approach enhances the model's ability to capture temporal dependencies without increasing computational complexity. Extensive experiments on load forecasting dataset demonstrate that the proposed method achieves superior predictive accuracy compared to existing deep learning-based approaches, effectively improving the accuracy and stability of load forecasting.