Fire frequency analysis and prediction based on back propagation neural network model and Long-Short Term Memory model

Abstract

It is reported that there have been many fires in many places this year. This article wants to explore the factors affecting the occurrence of fires and make predictions to prevent the occurrence of forest fires. The correlation between forest indexes and fire frequency in Fujian, Guangxi, Heilongjiang, Hunan, Sichuan, Yunnan and Zhejiang provinces was analyzed based on forest indexes and fire frequency prediction in seven provinces from 1998 to 2018. According to the six relevant indexes of seven provinces in 20 years, the Pearson correlation coefficient between them and the number of fires was analyzed to find the correlation between the variables. The Grey system theory Model (GM (1, 1)) was established to predict the six related index values. The BP neural network model was established based on the historical and the predicted values were substituted into the Back Propagation (BP) neural network model to predict the number of fires. Then, a prediction model based on Long-Short Term Memory (LSTM) was constructed through deep learning library Keras, and the predicted value was substituted into the model to predict the number of fires. Zhejiang Province and Hunan Province were taken as examples to compare the accuracy of the two models’ predictive values. It was found that the LSTM model had better fitting effect, and the LSTM model was finally used to predict the number of future fires. The forest ecological risk identification and prediction system based on Graphical User Interface system (GUI) was developed, the phase space reconstruction of one-dimensional data with time series characteristics was carried out, and the high-dimensional data were trained by using fuzzy neural network to realize data prediction and evaluation. This prediction can be applied to the forest prevention and control departments, and the annual fire situation can be predicted based on historical data, so as to prevent and reduce losses in advance.