An Improved Temporal Convolutional Network for Non-intrusive Load Monitoring

Non-intrusive Load Monitoring(NILM), also known as load disaggregation or energy disaggregation, estimates the energy consumed by the individual appliance from aggregate power consumption of the entire house. Recently, NILM is used to provide users with reasonable energy saving solutions, optimize energy scheduling and fault diagnosis for the appliances. Deep learning is widely used for NILM because its remarkable achievements in neighbouring fields such as natural language processing. In this paper, an improved temporal convolutional network(TCN) in the form of sequence-to-sequence (seq2sqe) model is proposed for NILM to further improve the efficiency of load disaggregation. Specifically, the problem of gradient disappearance and gradient explosion in deep learning model of NILM is solved by the residual network. The dilated convolution reduces the number of the hidden layer of deep learning model and improves the training speed. In addition, our method retains the sequentiality and time dependence of the input, which is beneficial to improve the disaggregation accuracy. Experimental results show that the proposed model can achieve better disaggregation performance than the current state of the art.