A novel distributed fine-grained energy consumption monitoring for large-scale nonlinear dynamic industrial processes
针对大规模非线性动态工业过程的能耗监测难题，提出一种分布式框架，结合互信息、AE-LSTM-mRMR模型和支持向量数据描述，实现细粒度能耗监测，并在热轧带钢实际数据中验证了有效性。
To accelerate the transition of large-scale industrial processes toward green and low-carbon directions, reducing energy consumption is a top priority, large-scale industrial processes are characterised by nonlinearity and dynamics, and there are coupling relationships between the cascaded sub-blocks, which pose challenges for energy consumption monitoring. We propose a distributed energy consumption monitoring framework for large-scale nonlinear dynamic industrial processes in this paper. First, Mutual Information is employed to assess the relationship between process variables and energy consumption metrics, selecting relevant variables for energy consumption monitoring. Second, an AutoEncoder (AE)-Long and Short-Term Memory network (LSTM)-minimum Redundancy-Maximum Relevance analysis (mRMR) monitoring model is constructed. For series-coupled sub-blocks, the parallel AE-LSTM method can extract nonlinear and dynamic features and fuse them to comprehensively capture the mechanism of the energy consumption changes, while the mRMR method is capable of reducing the impact of redundant information. Then, based on the deep features, statistics are constructed for each sub-block using Support Vector Data Description to complete the energy consumption monitoring and detect whether it exceeds the limit. Finally, the method is validated using actual data from the real hot strip mill process, demonstrating its ability to accurately and effectively monitor excessive energy consumption.