Research On Non-Intrusive Load Monitoring Decomposition Algorithm Based On Deep Learning

Non-intrusive load monitoring is a technology that can obtain the load information of each sub-electric equipment from the total energy consumption information,which can help users understand their own electricity consumption behavior,guide users to form good electricity consumption habits,and achieve the purpose of saving electricity Reduce resource waste.At the same time,electricity consumption information can be fed back to the power sector to help them rationally allocate power resources,optimize power supply schemes,achieve energy conservation and emission reduction,and alleviate energy shortages.In this paper,based on deep learning related technologies,an optimization method is proposed for the shortcomings of the current non-intrusive load decomposition direction,which further improves the decomposition effect of the model.At the same time,the generalization ability and versatility of the model are improved,which is helpful for the promotion of non-intrusive load monitoring technology to practical applications.The features that can be extracted by non-intrusive load monitoring technology under low-frequency data sampling are limited,and it is difficult to mine effective information hidden in the timeline during the operation of electrical appliances,and the decomposition accuracy is low.Aiming at the above problems,a sequence-tosequence non-intrusive load decomposition algorithm based on the combination of convolutional neural network and long short-term memory network is proposed.The deep learning model takes the power time series as the input of the network,and performs feature extraction through the convolutional neural network.Considering the time series of power data,a long short-term memory network layer is added for electrical load decomposition,which reduces the number of network layers and simplifies the network structure compared to the sequence-to-sequence model in NILMTK(Toolkit for Non-intrusive Load Monitoring).The algorithm performance is evaluated on the REDD(Reference Energy Disaggregation Data Set)dataset.The proposed algorithm improves the performance of the entire network system,and the load decomposition accuracy is significantly improved compared with FHMM(Factorial Hidden Markov Model),CO(Combination optimization)and traditional sequence-to-sequence algorithms.Aiming at the problem that it is difficult to collect new data in the process of popularization and application of non-intrusive load monitoring technology and the model is easy to overfit when the amount of data is small,a load decomposition algorithm based on data enhancement and temporal convolutional neural network is proposed.Data augmentation allows efficient load decomposition with less data in new application scenarios,reduces labor costs for collecting new data and the cost of purchasing acquisition equipment and training models,and improves the applicability of non-intrusive load monitoring technology.On the basis of data augmentation,the decomposition accuracy is improved by mining the timeline information of the power data by using a temporal convolutional neural network.The proposed algorithm model is evaluated using REDD dataset and UK-DALE dataset.The load decomposition errors of the temporal convolutional network model before and after data enhancement and the ordinary convolutional neural network model are compared.The experimental results show that data augmentation effectively reduces the degree of overfitting under the condition of a small amount of new sample data,and improves the accuracy of load decomposition.