Community-oriented Non-intrusive Load Identification And Decomposition

Non-intrusive load monitoring(NILM)refers to the method of installing monitoring equipment on the home side of the electricity to obtain the status of each electrical equipment through the total electricity load information.Therefore,the NILM technology is to realize the intelligent power distribution and refined management at the user end.One of the key technologies.With the application of deep learning in NILM,the ability of load identification and power decomposition has been improved,but there is still room for optimization of the training rate and prediction accuracy of the model;now all online public data sets available for non-intrusive load monitoring are Based on the data collected by foreign households,there is a certain difference between the electricity consumption of foreign residents and domestic community households,so collecting domestic community residents’ electricity data is more meaningful for the study of non-intrusive load decomposition.This paper compares and analyzes public data sets,and builds a data collection device;uses existing public data sets to verify that the improved Attention mechanism algorithm has a good load decomposition effect in non-intrusive load decomposition;based on OS-CNN for the community User family types are classified.Firstly,the existing public data sets were compared and analyzed,and the different characteristics of each data set were found.According to the needs,a set of power data collection devices were built in the laboratory to simulate the electrical equipment used in the home,and the collection devices were installed to realize the data.Collection.On the basis of the successful data collection in the laboratory,the collection equipment platform is integrated into the gateway,and the mesh networking is used to gather the data of each floor gateway to the top gateway,and then upload it to the cloud through NB-Io T.Secondly,in view of the improvement in load identification and power decomposition capabilities,but the rate of training the model and the prediction accuracy of the model are still not high,the sequence-to-sequence algorithm is used to achieve non-intrusive load decomposition.After the Attention mechanism is added,the accuracy of the decomposition results is effectively improved,but the training time after the Attention mechanism is added is longer,so this article uses the non-invasive load decomposition model of the attention mechanism combining the global and the sliding window,and in the public data set REFIT The above verifies that this algorithm has a better effect on training rate and accuracy.Finally,analyze the data collected by the gateway in the community and classify user households according to the characteristics of electricity consumption;after analyzing the shortcomings of 1D-CNN in Fourier decomposition and kernel size selection,this article adopts the OS-CNN method.Find the appropriate convolution kernel size during the training process,effectively reducing the training time and improving the accuracy of the classification of family types.