Research On Non-Intrusive Load Monitoring Technology For Commercial And Industrial Loads In Different Scenarios

With the growing energy crisis and the gradual increase in the proportion of renewable energy sources,more and more attention has been paid to the application of demand-side management.In the current context,non-intrusive load monitoring has received extensive research and attention as a potential solution.However,existing studies mainly focus on residential energy use,while commercial and industrial loads,as an important demand-side resource,are characterized by large scale,variety and complexity of changes,leading to the problems of existing methods such as inapplicability to industry and commerce,poor model universality and single evaluation scenarios.Therefore,this paper proposes a variety of non-intrusive load monitoring methods for commercial and industrial loads in different scenarios through multi-perspective comparison and analysis.This paper first introduces the process of traditional non-intrusive load monitoring techniques and summarizes the advantages and disadvantages of existing methods at each step,which leads to the deep learning-based research method used in this paper.Subsequently,the differences between industrial and commercial loads and residential loads are compared from various perspectives,such as load characteristics,correlation,power consumption variation and number of events,to reveal the reasons for the problems of existing methods.Based on this,three non-intrusive load monitoring methods are proposed in this paper: first,a load decomposition method based on sequence-to-point Transformer is proposed to solve the limitation problem of relying on event detection methods and sequence-to-series models for commercial and industrial loads.The superiority of the method over traditional methods is demonstrated through experiments on loads in different scenarios and at different levels in the commercial load dataset COMBED.Second,various SGN-based network models are proposed to improve the shortcomings of a single regression network model for some industrial and commercial loads.Experimental results on the industrial load dataset IMDELD show that the proposed multiple models perform well,especially the TCN-Attention-SGN model,which has significant effect improvement in different scenarios.Finally,a multi-label classification model,TTRNet,is proposed,which autonomously learns the relevance of load through its unique network structure and solves the data imbalance problem using a multi-label focal loss function.The experimental results show that compared with existing multi-label classification methods,the method improves the average F1 scores in visible and invisible scenes by7.77% and 1.92%,respectively,with higher accuracy and practical value.In summary,this paper proposes a variety of effective solutions to the difficulties and challenges facing the application of non-intrusive load monitoring in industrial and commercial environments.These methods can help grid companies or business users to better manage the electricity load,which is important for improving energy utilization efficiency and ensuring safe and stable operation of the power system.Figure 56 Table 15 Reference 81...