Simulation of forest fire spread based on artificial intelligence

Abstract

This article aims to provide a more practical forest fire spread model for predicting and managing forest fires in Heilongjiang Province, China. Heilongjiang is dominated by spatially heterogeneous combustible forests with undulating terrain and steep slopes. In this article, an artificial neural network framework is used to generate an accurate flame propagation map. Considering inherent fire propagation uncertainties, a fire propagation model containing multidimensional physical and environmental variables is established. Based on fire propagation predictions, the physical fire propagation method is also effectively understood. Additionally, the artificial neural network model can analyse spatial time series patterns and is not a traditional fire spread model. Moreover, this study established a forest fire spread prediction model combining Heilongjiang's cellular automata and Wang Zhengfei's model for comparison with the artificial neural network model. After repeated training and testing, the forest fire prediction results based on the artificial neural network, were found to have average accuracy, sensitivity, and F-measure values of 85.02%, 95.26%, and 89.85%, respectively. The proposed model is suitable for prediction of fire spread beyond large forest fires (more than 1 ha and less than 100 ha of affected forest). Therefore, the model facilitates a better understanding of fire cover propagation behaviours and quickly generates fire peak profiles. The proposed model can enable forest managers and firefighting agencies to plan better firefighting operations and improved firefighting strategy effectiveness.