Fault island partitioning algorithm for distributed distribution networks based on densely connected convolutional neural networks

If the time-varying characteristics and load priorities of distributed power sources are not fully considered, it may lead to power imbalance in isolated islands or the failure to recover important loads. To this end, a fault island partitioning algorithm for distributed distribution networks based on densely connected convolutional neural networks is proposed. The recovery priorities of different levels of loads were considered. The priority of secondary loads over tertiary loads was reflected by setting the weight size. The influence of isolated island operation on the load current of the power grid was taken into account. The limit range of bus voltage was set. Weights were assigned to various production loads according to the power value calculated by the input-output method. The densely connected convolutional network achieved feature reuse through dense skip connections. Solve the fault island partitioning model of the distributed distribution network. The experimental results show that the total load recovered by the proposed algorithm is 2463.57kW, which can effectively shorten the power-off time of the load, reduce the power-off loss of the load, and thereby improve the accuracy rate.