Non-Intrusive Load Decomposition Based on Instance-Batch Normalization Networks

At present, the non-intrusive load decomposition method for low-frequency sampling data is as yet insufficient within context of generalization performance, failing to meet accuracy requirements when applied novel scenarios. To address this issue, a based on instance-batch normalization network proposed. This uses an encoder-decoder structure with attention mechanism, in which skip connections are introduced at corresponding layers encoder and decoder. In way, decoder can reconstruct more accurate power sequence target. The proposed model was tested two public datasets, REDD UKDALE, performance compared mainstream algorithms. results show that F1 score higher by average 18.4 Additionally, mean absolute error reduced 25%, root square 22%.