Non-invasive Load recognition of Electrical Signals based on Improved Conv-TasNet

With the rapid development of smart grids, electricity meters now collect aggregated data from various appliances. To obtain the individual power sequence of target appliances, algorithms are used. Traditional load recognition methods rely on machine learning algorithms, which struggle due to low-frequency data accuracy, complex feature extraction for high-frequency data, and poor network generalization. To address these issues, this paper introduces a non-invasive load decomposition method that manages computational complexity using improved Conv-TasNet and DAE-ResNet algorithms. First, a denoising model based on depthwise separable convolutional residual networks reduces the training network complexity. Second, the Conv-TasNet algorithm is enhanced to improve the load decomposition performance. Input data is encoded with deep convolution, and a separation network forms a mask for the target device. Finally, the decoder combines max pooling and deep convolution to decode the target electrical information. Experimental results show that the proposed method significantly outperforms existing techniques in load identification and decomposition. It improves both the accuracy and convergence speed.

Statistics

Show / Hide Statistics

Cumulative Counts from September 30th, 2019
Multiple requests among the same browser session are counted as one view. If you mouse over a chart, the values of data points will be shown.

Cite this article

[IEEE Style]

J. Xu, B. Hu, P. Zhang, Z. Xing, J. Cui, N. Han, "Non-invasive Load recognition of Electrical Signals based on Improved Conv-TasNet," Journal of Power Electronics, vol. 25, no. 7, pp. 1295-1308, 2025. DOI: 10.1007/s43236-024-00980-5.

[ACM Style]

Jian Xu, Bo Hu, Pengfei Zhang, Zuoxia Xing, Jia Cui, and Ni Han. 2025. Non-invasive Load recognition of Electrical Signals based on Improved Conv-TasNet. Journal of Power Electronics, 25, 7, (2025), 1295-1308. DOI: 10.1007/s43236-024-00980-5.