Research On Non-intrusive Load Decomposition Based On Deep Learning Algorithm

In the context of smart grids,non-intrusive load monitoring methods are becoming a research hotspot for many scholars in view of the high construction cost and difficulty of popularization of traditional intrusive load monitoring methods.Non-intrusive load disaggregation methods can help users understand the current status of electricity consumption,reduce energy consumption,and also help grid companies realize scientific power dispatch,so as to achieve the purpose of energy saving and consumption reduction.The non-intrusive load disaggregation is mainly realized by two means: event detection and load model.However,traditional feature extraction methods are relatively complex and difficult to meet high-precision load disaggregation,and cannot meet the needs of current smart grid development.At present,there are many non-intrusive load disaggregation methods based on deep learning models.But many of them have problems such as insufficient extraction of network model load features,low disaggregation accuracy,and large disaggregation errors for load equipment with low frequency of use.To solve the above problems,this thesis constructss two improved non-intrusive load disaggregation methods based on deep learning model.Firstly,a non-intrusive load disaggregation model based on long and short-term memory multi-output network is proposed.Load features are extracted by the improved multi-scale fusion residual block,the time series information is extracted by the LSTM structure,and the idea of cross-layer connection is used to combined data to achieve the effect of feature reuse.This effectively reduces the calculation cost and improves the disaggregation effect.Although the load disaggregation model based on long and short-term memory multi-output network has achieved certain experimental results.But for load equipment with low frequency of use,the disaggregation result is not very ideal.Therefore,an aggregation network based on dilated residual is proposed for non-intrusive load disaggregation.This network model enlarges the receptive field of the convolution kernel through dilated convolution and captures more data.The feature enhancement module is used to enhance the learning ability of subtle load features of the network and further improve the performance of load equipment.In summary,the construction of long and short-term memory multi-output network model and dilated residual aggregation convolutional network model in this thesis to carry out non-intrusive load disaggregation research has important reference significance for the development of smart grids in my country.