Non-intrusive load monitoring through coupling sequence matrix reconstruction and cross stage partial network

Non-intrusive load monitoring can obtain the operation state of single device from the total power sequence. To achieve high-precision non-intrusive load monitoring, this study proposes a novel method by coupling sequence matrix reconstruction and cross stage partial network. The timing variation information hidden in the total power sequence is derived through sequence matrix transformation and matrix dimension rising, forming a three-dimensional matrix. Then, this matrix is sent into a cross stage partial network to estimate the device power consumption. Especially, the established cross stage partial network adopts a feature concatenation strategy, greatly reducing the network parameters. Unlike traditional neural network methods, the proposed method makes full utilization of the timing variation information of total power consumption. Experiments are carried out on the load monitoring platform to verify the method feasibility. Results show that the proposed method can non-intrusively decompose the device power consumption, with strong robustness and generalization ability.