A modular neural network approach for locating cloud-to-ground lightning strokes

A modular neuro-based approach is proposed for locating cloud-to-ground lightning strokes. The locating process is done in two stages. In the first stage, a multi-layer perceptron feed forward network with one hidden layer is used to estimate the location range of the return stroke channel (RSC). The estimated value of the RSC location range at the output of the first stage determines the appropriate ANN trained for three near (1-10 km), intermediate (10-20 km), and far (20-80 km) zones. The simulation results demonstrate that the RSC locations can be predicted with an absolute error not greater than 1 km and a relative error less than 5%.