Multi-model aggregation method for non-intrusive load monitoring based on stacking

Abstract

Aiming at the problems that different non-intrusive load decomposition models are applicable to different loads and different states of the same load, a load decomposition integrated model based on stacking integrated denoising autoencoder model and gated neural network model is proposed. The model first uses the denoising autoencoder model and the gated neural network model to decompose the target load of the bus active power sequence, and obtains the active power sequence of the target load corresponding to the two models; Then the fully connected neural network is used to initially aggregate the two active power sequences into an active power sequence of the target load, and the denoising autoencoder model is used for optimization to obtain the final decomposition result of the target load. Experimental results show that the integration model based on stacking can combine the decomposition advantages of the base model, give full play to the respective advantages of DAE model and Gru model, and realize accurate load decomposition.