Non-Intrusive Load Monitoring based on Cnn-Bigru and Mountain Multi-Layer Attention Mechanisms

In this paper, we propose a Non-Intrusive Load Monitoring method based on Convolutional Neural Network-Bidirectional Gated Recurrent Unit and Mountain Multi-Layer Attention Mechanisms. The method first utilizes Convolutional Neural Network to extract the local features of the short-time sequence of loads, and subsequently uses Bidirectional Gated Recurrent Unit to capture the global features of the long-time sequence of loads, on the basis of which, it incorporates the Mountain Multi-Layer Attention Mechanisms to enhance the learning capability of the active power features of the electrical equipment and dynamically adjusts the weight values, to excavate the correlation between the state of the electrical equipment and the load active power. Experiments on the real-world UK-DALE dataset show that the model significantly improves the load disaggregation performance of electrical equipment compared to existing typical disaggregation models.