Application of Machine Learning to Lightning Strike Probability Estimation

The risk of a lightning strike to a structure is influenced by geometry, the type of material of the structure, and proximity to other structures. The probability of lightning strike to different points on an object can be determined by applying the concept of probability modulated collection volume using the dynamic electro-geometrical model (DEGM). The numerical computation of the DEGM is computer resource-intensive and requires extensive programming and analysis to implement, which may be difficult to apply by engineers setting up lightning protection systems in the field. This study explores the feasibility of applying an artificial neural network (ANN) using a network with 18 hidden neurons and the Levenberg-Marquardt algorithm for training the model. The strike probability predictions by the neural network give close values to the numerical results. The ANN approach is further supported by predictive equation-based models developed using data pattern recognition and curve fitting techniques, as easy to use alternatives to the numerical DEGM simulations. The ANN model had an overall regression R
₂
of 1, and it performed better than the equation-based model.