A reliable deep learning-based algorithm design for IoT load identification in smart grid

In IoT load monitoring system of the smart grid, the non-intrusive load monitoring and identification (NILMI) has become the research focus. However, the existing researches focus on the accuracy of load identification, neglecting the effectiveness of data sampled, the distinction of load abstract feature representation, and the reliability of load identification model. This paper proposes a novel algorithm framework of spatial-temporal convolution neural network for NILMI, namely DST-CNN, to realize the fine-grained load identification. In the DST-CNN framework, to ensure the accuracy and reliability of data usage, an signal enhancement method, AM-PCA, is used. To enhance the distinction of load abstract feature representation, an extraction mechanism of the spatial-temporal features is developed, which utilizes deep convolution networks and time-series recurrent neural networks (RNN). To improve the accuracy and reliability of load identification model, a hierarchical load classification mechanism is constructed, and the deep long short-term memory (LSTM) structure as the classifier. A considerable amount of the high-frequent current signals are sampled to validation the performance of the proposed method. The experimental results demonstrate the good generalization performance and superiority for NILMI in IoT load monitoring system.

Automated summary

This paper proposes a novel algorithm framework of spatial-temporal convolution neural network, DST-CNN, for fine-grained load identification in IoT load monitoring system. AM-PCA signal enhancement method and spatial-temporal feature extraction mechanism are employed to improve the accuracy and reliability of data usage. The hierarchical load classification mechanism and deep long short-term memory structure are utilized for enhancing the accuracy and reliability of load identification model.