Research on deep-learning-based load forecasting methods for integrated energy systems in chemical equipment

To enhance the modeling accuracy and robustness of integrated energy systems in chemical plants under multi-load forecasting, a Transformer-based prediction framework incorporating a multi-channel attention mechanism is constructed. Utilizing sliding window sequence construction, regularization optimization, and TPE hyperparameter search strategies, its generalization capability across various typical operating conditions is systematically analyzed. Results demonstrate that the model maintains high prediction accuracy and stability in non-steady-state scenarios, outperforming traditional recurrent structures.