Non-intrusive Load Identification Method Based On SAMME.R-DT And SRU

Increasing the demand-side response to more reasonably guide users to use electricity,with a view to achieving the goals of energy saving,emission reduction and sustainable development,has become the theme of the development of the times.To achieve this goal,non-intrusive load monitoring(NILM)based on user’s electricity usage behavior provides solid technical support.In NILM work,there are still problems that it is difficult to accurately identify load devices such as low power consumption and multi-state,and the problem of overlapping load characteristics such as power characteristics is still difficult to solve;On the other hand,with the development of artificial intelligence,deep learning has been widely used in NILM.However,most of the algorithms used in the field of deep learning in the field of NILM are too much to pursue the effect of load recognition and ignore the real-time.Such as RNN(recurrent neural network)and its variants LSTM((long short term memory)due to its own structure is limited,there is a problem that the parallel computing operation leads to too long calculation run time,the problem It will also directly affect the power department’s timely grasp of user-side data.In response to the above problems,this paper focuses on the application of new machine learning algorithms in NILM.The main work contents are as follows:1)The existing research methods of NILM are sorted out and classified from the two aspects of load characteristics and load recognition algorithms.The connotation of ensemble learning and deep learning are discussed respectively.The indicators of NILM algorithm evaluation are determined.Lay the foundation with the assessment.2)The use of integrated learning can effectively improve the characteristics of simple classifiers,and a combined integrated learning SAMME.R-DT method is proposed.The proposed method first uses SAMME.R for load identification,transforms the Adaboost algorithm from a two-class classification problem to a multi-class classification problem,avoiding the direct use of the classic Adaboost algorithm when performing load identification on multiple types of appliances.Sample weights need to be calculated by multiple sets of combinations.At the same time,the decision tree is selected as the base classifier.Compared with the four traditional supervised learning algorithms,it is found that the algorithm has improved the recognition accuracy of low power consumption and multi-state appliances,and has a better robustness when the number of appliances load gradually increases.3)In order to further improve the accuracy of load recognition,deep learning can effectively reveal the characteristics of sample data and the characteristics of internal laws.A new RNN variant,simple recurrent units(SRU),is used to build the NILM deep learning model.By comparing with LSTM,gated recurrent unit(GRU)and SAMME.R-DT algorithm,analyze the working performance of SRU in NILM.The analysis of calculation examples shows that: 1)Compared with traditional shallow learning,in terms of load recognition,after a large amount of data training,the recognition effect is obviously improved,but the model training also takes more time;2)SRU is relatively The operating efficiency of LSTM etc.in load identification has been significantly improved.The operating speed is almost 4 times that of LSTM and 2.5 times that of GRU,and does not affect the accuracy of load identification.This is of great significance for promoting the real-time performance of NILM in actual work.