Non-intrusive load disaggregation based on deep dilated residual network

Abstract A non-intrusive sequence to sequence load disaggregation method based on deep delated convolution residual network is proposed in this work. Firstly, the original power data is normalized, and then the sliding window used to create the input for the residual network. The dilated convolution residual network model reduces the difficulty of network optimization and solves the problem of vanishing gradient. The residual learning can improve the depth of the network and the ability of extracting data features. The difficulty of learning long time series data is solved by increasing the receptive field and capturing more data through dilated convolution, which can improve the ability of processing samples with low usage. The sequence to sequence disaggregation method can improve the disaggregation efficiency. Simulation results using two data sets of real house measurements show that proposed model can get better disaggregation results than existing studies, especially for the disaggregation of electrical appliances with low usage.