Residential Load Decomposition Based On Deep Learning

As an important part of the power system,the development of smart grid plays an important role in the economic operation and orderly power consumption.Power load monitoring is a focus technology of ubiquitous internet of things(iot)and smart grid combined with user demand-side response,and it is also a key method of energy saving and emission reduction.Among them,non-invasive load detection technology is the main research point of power load monitoring.Compared with traditional intrusive load monitoring,non-intrusive means that the energy consumption of a single power device in a certain area can be obtained only by analyzing the total meter in that area.In this paper,the non-intrusive load decomposition based on deep learning is studied by using the deep neural network model:Firstly,the household load is classified and the transient characteristics of the equipment are analyzed.In the present research,the load is decomposed after the load feature library is built by manual extraction of different load features,but there are some problems in this method,such as lack of information and large error,therefore,this paper will expand the load decomposition based on deep learning.Secondly,when total load is decomposed,active and reactive power are introduced as load characteristics,and a non-intrusive load decomposition method based on feature fusion and multi-task learning is proposed,which connects Bi-TCN and Bi-GRU in parallel,after learning the temporal features respectively,the features are fused into the multi-task learning module.The bottom layer of the multi-task module is composed of the codec structure based on full convolution,and the multi-convolution module is used to perform multi-task in the part near the output,make each task have its own unique layer to learn high-level features.The results show that,on the same data set,the decomposition effects of the six selected devices are better than those of seq2point,seq2seq and Dae,which verifies the validity of the model.Finally,the user’s power consumption behavior is analyzed by the level information of electrical apparatus obtained from non-intrusive load decomposition.The application of non-intrusive load decomposition can help to intuitively understand the electricity consumption behavior of residential users,promote the statistical business of electricity tariff,and then optimize the household energy consumption strategy,non-intrusive load decomposition also helps the grid to build user profiles,which in turn leads to personalized performance services and facilitates "Peak load reduction" among residential users.In general,non-intrusive load decomposition has great potential in smart power consumption and demand-side response.