Non-intrusive Load Decomposition Method Based on Deep Sequence Translation Model

Abstract

Power monitoring is very important for smart power, and non-intrusive load decomposition is its key technology. Deep sequence translation method is used for non-intrusive load decomposition to improve accuracy of residential load decomposition. Firstly, the actual powers of different appliance modes are calculated and the operating status of all appliances is encoded into status code. Secondly, the paper uses sequence translation model to learn the time correlation of electrical operation modes, and to train the decomposed signal and the status code of the appliance in the sequence translation model. At the same time, network parameters are optimized with dropout technology and sparse technology. The model learns the time-correlated information and signal amplitude characteristics of the electrical state. The decomposition of load energy is accomplished by translating the energy into status code. Finally, the paper uses public data set to confirm higher energy decomposition accuracy of the proposed method.