<p>Lightning strike-induced transmission line tripping is an important reason affecting the safe operation of power grids. Timely and accurate prediction of transmission line tripping can provide a basis for power companies to adjust their operation modes beforehand and reduce economic losses. The traditional prediction model of line tripping due to lightning strikes predicts whether tripping will occur in the next time period based on the monitoring data of the lightning locating system (LLS) in one time period, which makes it difficult to learn the evolution pattern of lightning activity in space and time. Therefore, this paper proposes an improved 3-dimensional convolutional neural network (3DCNN)-based tripping prediction method for transmission lines. Firstly, based on the monitoring data from the LLS, a high-dimensional input matrix construction method considering the spatial–temporal characteristics of lightning activities is proposed. Secondly, based on the 3DCNN, self-attention mechanism and hierarchical classifier, the improved 3DCNN is proposed to learn the spatial–temporal law of lightning activities to realize the prediction of whether the transmission line is tripped or not, as well as the location of lightning strikes on lines. Subsequently, considering the imbalance of the samples, a focal loss function is introduced to guide the model training, which improves the accuracy of the model. Finally, the measured data of a local power grid in southern China is used as an example to verify the effectiveness and reliability of the method in this paper.</p>