A non-intrusive load decomposition method of resident by multi-scale attention mechanism

Non-intrusive load monitoring (NILM) is one of the important technologies in home energy management and power demand response scenario. However, presence multi-mode appliances with close values have affected diminishing accuracy identification based NILM algorithms. To tackle these challenges, work proposes a resident decomposition method combining multi-scale attention mechanism convolutional neural network. At first stage, scores normal data at previous few moments model are smoothed dynamically against abnormal current moment. The optimized by constraint factors. Then, on this basis, convolution filters different sizes used to mixed electrical equipment, mine more abundant characteristic information. Finally, illustrate proposed processes validate its effectiveness, taking PLAID set as an example, article compared respect existing techniques. experimental results show that paper can greatly improve effect decomposition. Moreover, it reduces confusion problem appliance similar characteristics.