Research on Short Term Lightning Fault Warning in Distribution Network Based on Convolutional LSTM Network and Lightning Location System

Most lightning warning research focuses on the transmission network, as the density of distribution equipment per unit area is higher and the lightning protection ability is poor. Therefore, the research results cannot be simply applied to the distribution network. Firstly, divide the target area into equal area grids by $0.01^{\circ}\times 0.01^{\circ}$, and divide the lightning data from the first 15 minutes and the distribution network resource information of the selected area into three groups evenly over time; According to the fuzzy C-means clustering algorithm, the cluster center of each group of lightning data is generated through clustering analysis. The weight of the fault factor is obtained using the entropy weight method, and a short-term lightning strike fault warning model for the distribution network is trained based on the ConvLSTM (Convolutional long short-term memory) network; Based on a thunderstorm cloud trend prediction model, determine the location of lightning strikes in the target area within the next 15 minutes, match the factors of distribution network equipment, and predict the types and probabilities of lightning strikes in the distribution network for a period of time in the future; Finally, the method was validated using data from Jinshan Power Grid. The results indicate that on the basis of accurate prediction of the cluster center of thunderstorms, this algorithm can accurately predict the types and probabilities of distribution network faults caused by lightning strikes.