Data-driven energy monitoring and prediction through an online visual platform for discrete manufacturing workshop

Efficient energy monitoring and prediction are vital for manufacturing sustainability, which provides technical support to optimize the energy structure and improve energy efficiency. A data-driven monitoring and prediction method embedded in an online visual platform is proposed for discrete manufacturing workshops under frequent disturbances. Firstly, a visual monitoring service layer is established to enable real-time energy monitoring. Subsequently, a residual LSTM with an improved adaptive width adjustment strategy is proposed. Compared to traditional point prediction, this approach provides superior uncertainty quantification for volatile energy data. Furthermore, an improved lifelong learning strategy is introduced, enabling the prediction model to adaptively evolve with continuous workshop changes while retaining historical knowledge, thereby maintaining high prediction accuracy over time. Finally, using a machining workshop as an application case, experimental results show that when the prediction interval nominal confidence is set to 0.9, the proposed method achieves the prediction interval coverage probability exceeding 0.91 while maintaining the narrow normalized average width below 0.14. Compared with baseline models, it reduces the interval root mean squared error by 5.22% to 7.64%. Practically, the platform enables real-time comparison of actual energy efficiency against theoretical benchmarks, empowering managers to detect anomalies and initiate dynamic decision-making.