The casualty prediction of earthquake disaster based on Extreme Learning Machine method

Abstract

In the prediction of casualties of earthquake disaster, the traditional prediction method requires strict sample data, and it is necessary to manually set a large number of parameters, resulting in poor prediction accuracy and slow learning speed. This paper introduces the Extreme Learning Machine (ELM) into the earthquake casualty prediction, aiming to improve the prediction accuracy. Through the data training, the ELM network structure of earthquake victims’ casualty prediction is established, and the number of hidden layer nodes and the excitation function are determined, which ensures the reliability of the ELM network prediction results. Based on the data of 84 groups of earthquake victims from China in 1970–2017, the ELM algorithm, BP neural network, SVM and modified partial Gaussian curve were compared and verified. The results show that the average relative error of ELM algorithm for earthquake disaster prediction is 3.37%, the coefficient of determination R-square is 0.96, the average relative error of injury prediction is 1.04%, and the coefficient of determination R-square is 0.97, which indicates that the ELM algorithm has good robustness and generalization ability.