Objective Due to the wide distribution range and high height of transmission lines, they are highly susceptible to lightning strikes. In order to develop effective pre lightning protection measures and achieve full process protection against lightning strikes on transmission lines.

Method This study took the Guanzhong region as an example, selected thirty-three physical quantities highly correlated with lightning activity as features, and used the PSA algorithm to optimize the learning rate, number of hidden layer nodes and regularization coefficient. Based on a multivariate Long Short Term Memory (LSTM) network, a lightning potential prediction model for transmission line areas was constructed to predict lightning activity in the next twelve hours.

Result Research indicates that thirty-three atmospheric physical quantities, such as the modified K index and Charba modified index, are highly correlated with lightning activity and can serve as features for lightning potential prediction. In terms of model hyperparameters, the optimal combination determined through the PSA algorithm is: learning rate of 2.7e-4, number of hidden layer nodes of 29, and regularization coefficient of 4.1582e-4. Furthermore, when the batch size is set to 16 and the sequence length is set to 20, the model achieves the minimum error value. Under these conditions, the prediction accuracy is 80.94%, and the probability of detection (POD) is 71.3%.

Conclusion Taking into account the geographical distribution of transmission lines and their distance from meteorological sounding stations, the prediction results have a certain degree of validity.