Energy consumption data in current intelligent manufacturing processes is complex and exhibits strong real-time fluctuations, making it difficult for traditional prediction methods to maintain stable prediction accuracy in high-frequency data scenarios. This paper addresses this issue by proposing an energy consumption prediction model based on an improved attention mechanism using a bidirectional long short-term memory (Bi-LSTM-AM) network. This model acquires multi-source heterogeneous energy consumption data from a 5G industrial internet platform. It forms a high-dimensional input sequence through feature normalization and temporal reconstruction. The underlying bidirectional long short-term memory network achieves deep modeling of the bidirectional temporal dependencies of energy consumption. The improved attention mechanism dynamically allocates weights based on the importance of hidden states to strengthen the feature representation of key time segments. Finally, parameter back-optimization is performed with the goal of minimizing the prediction error function, achieving high-precision prediction of production energy consumption. Experimental results are pending. The research results demonstrate that this method effectively improves the accuracy of energy consumption prediction in intelligent manufacturing environments, providing solid technical support for industrial energy optimization decisions.
Xiong et al. (Thu,) studied this question.