Power Supply Reliability Evaluation Method of Distribution Network Based on Improved LSTM Neural Network

As the power supply load of the distribution network itself is volatile, it is necessary to consider the reserve demand brought by the load. At present, most studies set the reserve demand of the load as 6% of the load value at that time. As an important part of the modern distribution network, the power supply department can adjust the power supply of the flexible load according to the prior contract agreement. In this paper, the flexible load is taken to optimize the island range. Based on the improved LSTM neural network, the confidence level and the influence of different types of flexible loads on islanding are solved. It is concluded that the islanding reserve demand varies with the output and confidence level of intermittent power sources. The islanding stability and economic demand under different confidence levels can be balanced by adjusting the flexible load. The equivalent value of islanding restoration can be effectively increased. Compared with the post-fault power supply restoration without considering the islanding reserve demand, the proposed strategy improves the reliability of the power supply, which shows the effectiveness of the proposed strategy. The experimental evaluation results verify that the improved LSTM neural network has better classification performance, and its classification accuracy index has stronger adaptability and anti-interference in the experiment. Besides, it has a good classification effect in the task of fault type and fault feeder identification of the photovoltaic distribution network. By balancing the power supply stability and demand under different confidence levels for the regulation model, and effectively increasing the value of restoration equivalence, the reliability of power supply is improved, which is more practical.