Research On Non-intrusive Load Monitoring Based On Seq2point

In the process of realizing the goal of "carbon peaking" and "carbon neutrality",the energy saving efficiency of the electric power industry plays a vital role.As one of the key technologies in energy saving and efficiency improvement,the non-intrusive load monitoring is to monitor and decompose the total load of household electricity to achieve the purpose of monitoring each household appliance,and then provide users with a list of energy consumption of electrical equipment and scientific suggestions on electricity consumption,and guide users to actively save electricity.In recent years,the rapid development of deep learning not only accelerates the innovation of speech recognition,image recognition and other technologies,but also greatly promotes the development of non-intrusive load monitoring technology.This paper will carry out in-depth research on non-intrusive load monitoring technology based on the sequence-to-point deep learning architecture.The main research contents are as follows:1.In order to solve the problems of long model training time and poor decomposition performance in the application of sequence-to-point with deep learning method in the field of non-intrusive load monitoring,a non-intrusive load monitoring model based on Seq2 point was proposed.In this model,a forward gated recurrent unit network is used to efficiently mine the time-series relationship between input and output,and a standardized data preprocessing method is used to reduce the negative effects caused by unbalanced data distribution.The model was compared with the two mainstream models on the public dataset of AMPDS2,and the recall rate,precision rate,F1-score,mean absolute error of power estimation and relative error of total energy consumption were selected as metrics to evaluate the model.The results show that this model not only has superior decomposition performance,but also reduces the training time by 55%,which lays a good foundation for further research.2.A feasible idea is put forward to solve the problem that family training data is not easy to obtain,which leads to the problem that the technology cannot be popularized and applied.In this concept,the training sample data of the model will be provided by each electrical appliance manufacturer,the electrical appliance training set will be constructed from limited and unbalanced samples,and the training process of the model and the non-intrusive load decomposition process will be responsible by the power grid data center.In order to solve the problem that the training samples provided by manufacturers are limited and unbalanced,a balanced training set construction method is proposed.In this method,the samples in the training set are balanced and complete mainly through the steps of balanced expansion,random combination,cutting and arrangement of the limited samples.Finally,the dishwasher is selected as the main research object to verify the method.The results show that the model trained on the equilibrium training set has good generalization performance,and can be applied to obtain the energy consumption list of household electrical equipment by non-intrusive load decomposition.This fully demonstrates the effectiveness of the method and the feasibility of the proposed concept.