Open AccessA prediction method of fire frequency: Based on the optimization of SARIMA model
Author(s) -
Shuqi Ma,
Qianyi Liu,
Yudong Zhang
Publication year - 2021
Publication title -
plos one
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.99
H-Index - 332
ISSN - 1932-6203
DOI - 10.1371/journal.pone.0255857
Subject(s) - outlier , autoregressive integrated moving average , statistics , quantile , mean squared error , time series , series (stratigraphy) , autoregressive model , interpolation (computer graphics) , computer science , environmental science , mathematics , geology , artificial intelligence , motion (physics) , paleontology
In the current study, based on the national fire statistics from 2003 to 2017, we analyzed the 24-hour occurrence regularity of fire in China to study the occurrence regularity and influencing factors of fire and provide a reference for scientific and effective fire prevention. The results show that the frequency of fire is low from 0 to 6 at night, accounting for about 13.48%, but the death toll due to fire is relatively high, accounting for about 39.90%. Considering the strong seasonal characteristics of the time series of monthly fire frequency, the SARIMA model predicts the fire frequency. According to the characteristics of time series data and prediction results, an optimized Seasonal Autoregressive Integrated Moving Average Model (SARIMA) model based on Quantile outlier detection method and similar mean interpolation method is proposed, and finally, the optimal model is constructed as SARIMA (1,1,1) (1,1,1) 12 for prediction. The results show that: according to the optimized SARIMA model to predict the number of fires in 2018 and 2019, the root mean square error of the fitting results is 2826.93, which is less than that of the SARIMA model, indicating that the improved SARIMA model has a better fitting effect. The accuracy of the results is increased by 11.5%. These findings verified that the optimized SARIMA model is an effective improvement for the series with quantile outliers, and it is more suitable for the data prediction with seasonal characteristics. The research results can better mine the law of fire aggregation and provide theoretical support for fire prevention and control work of the fire department.