Distribution network reliability assessment and optimization method based on convolutional neural network

Abstract In light of the complex factors affecting the operation of power distribution networks and the limitations of traditional reliability assessment and optimization methods, this study proposes a novel approach based on convolutional neural networks. The model is trained and tested using data from the IEEE 123-node distribution system, and its performance is compared against several established methods, including fault tree analysis, support vector machine models, deep belief networks, and random forest algorithms. Experimental results demonstrate that, with respect to the System Average Interruption Duration Index, the proposed model achieves an average outage duration of 3.15 h per user, significantly lower than the 5.42 h per user recorded by the fault tree analysis model and the 4.96 h per user of the support vector machine model. In terms of prediction performance, the proposed model reaches an accuracy of 92.5 percent, a recall rate of 90.8 percent, and an F1 score of 91.6 percent—outperforming all comparison models. This approach provides a new and effective method for evaluating and optimizing the reliability of power distribution networks, thereby enhancing operational stability, cost efficiency, and the overall reliability of electricity supply.