欢迎访问林业资源管理

林业资源管理 ›› 2022, Vol. 0 ›› Issue (2): 109-116.doi: 10.13466/j.cnki.lyzygl.2022.02.015

• 技术应用 • 上一篇    下一篇

基于自适应模糊神经网络的滇中灌木林火灾发生预测研究

魏建珩1(), 赵恒1, 高仲亮1(), 王何晨阳1, 马泽南1, 王秋华1, 舒立福2, 杨红梅3   

  1. 1.西南林业大学 土木工程学院,昆明 650224
    2.中国林业科学研究院森林生态环境与保护研究所,北京 100091
    3.陕西省林业科学院,陕西 榆林 710082
  • 收稿日期:2022-01-09 修回日期:2022-03-22 出版日期:2022-04-28 发布日期:2022-06-13
  • 通讯作者: 高仲亮
  • 作者简介:魏建珩(1996-),女,重庆人,在读硕士,主要从事林火管理方面的研究。Email: jheng_w@163.com
  • 基金资助:
    国家自然科学基金项目(31860214);国家自然科学基金项目(301660210);国家自然科学基金项目(31960318);云南省教育厅科学研究基金项目(2021Y227);北京林业大学中央高校基本科研专项基金(BFUKF202107)

Prediction of Shrubbery Fire in Central Yunnan Province Based on the Adaptive NeuroFuzzy Inference System

WEI Jianheng1(), ZHAO Heng1, GAO Zhongliang1(), WANG Hechenyang1, MA Zenan1, WANG Qiuhua1, SHU lifu2, YANG Hongmei3   

  1. 1. Civil Engineering College,Southwest Forestry University,Kunming 650224,China
    2. Institute of Forest Ecological Environment and Protection,Chinese Academy of Forestry,Beijing 100091,China
    3. Shaanxi Academy of Forestry,Yulin,Shaanxi 710082,China
  • Received:2022-01-09 Revised:2022-03-22 Online:2022-04-28 Published:2022-06-13
  • Contact: GAO Zhongliang

摘要:

滇中地区原植被破坏严重,易燃灌木连片生长。全球气候变暖加剧,以灌木林为主的森林火灾频发,因此预测灌木林火对保护滇中地区森林资源有着重要作用。以云南省滇中地区1999—2019年灌木林火发生及其对应的气象数据为基础,选择自适应模糊神经网络推理系统(Adaptive Neuro Fuzzy Inference System,ANFIS)、逻辑斯蒂回归模型(Logistic Regression,LR),利用MATLAB、SPSS 25等软件,建立基于气象因子的滇中地区灌木林火发生预测模型,其中70%数据用于建立模型,30%用于模型检验。研究结果表明:通过主成分分析,将9个气象因子形成3个主成分作为ANFIS模型输入因子,3个主成分能解释9个气象因子77.663%的信息;LR模型经过多重共线性检验,依据VIF<10,得出24小时降水量、平均2分钟风速、日平均相对湿度、日最小相对湿度为LR模型的自变量输入。由2种模型的气象因子筛选结果可知,影响滇中地区灌木林火发生的主要影响因子为温度、风速、湿度。对比ANFIS,LR模型拟合结果,ANFIS模型训练集准确率大于LR模型12%,测试集准确率高于LR模型10%。ANFIS模型训练集、测试集AUC值分别为0.961,0.884;LR模型训练集、测试集AUC值分别为0.875,0.816。对比2种模型拟合结果,利用ANFIS模型建立滇中地区气象因子与灌木林火发生模型具有更好的适应性。研究结果能可为滇中地区灌木林火灾预测提供一定的科学依据。

关键词: 气象因子, 逻辑斯蒂回归模型, 自适应模糊神经网络算法, 灌木林火, 滇中地区

Abstract:

The original vegetation in central Yunnan has been seriously damaged,leading to the continuous growth of flammable shrubbery. With the intensification of global warming,forest fires occur frequently,mainly in shrubs. Therefore,the prediction of forest fires is of great importance to the protection of forest resources in central Yunnan. Based on the data of shrubbery fire and its corresponding meteorological data from 1999 to 2019 in central Yunnan,China.Adaptive Neuro Fuzzy Inference System (ANFIS) and Logistic Regression (LR) were selected,using MATLAB,SPSS 25 and other software to establish a prediction model of forest fire in central Yunnan based on meteorological factors(70% of the data),then the remaining 30% of the data was tested.The results showed that nine meteorological factors were formed into three principal components by principal component analysis. As the input factors of ANFIS model,3 principal components could explain 77.663% of the information of 9 meteorological factors. According to the multicollinearity test,the VIF of the LR model was less than 10,and the precipitation at 24 o'clock,the average 2-minute wind speed,the average daily relative humidity and the daily minimum relative humidity could be obtained,then they were inputted into the model as independent variables of the LR model. According to the screening results of meteorological factors of the two models,the main influencing factors of shrubbery fire in central Yunnan were temperature,wind speed and humidity. By comparing the results of ANFIS and LR model,the accuracy of ANFIS training set was 12% higher than that of LR model,and the accuracy of test set was 10% higher than that of LR model. The ANFIS training set value was 0.96,the test set AUC value was 0.88,and the LR model AUC values were 0.875 and 0.816,respectively. The results showed that the ANFIS model had better adaptability to the model of forest fire in central Yunnan. Results of the research can provide some scientific basis for shrubbery fire prediction in central Yunnan.

Key words: meteorological factors, logistic regression, Adaptive Neuro Fuzzy Inferences System, shrubbery fire, central Yunnan Province

中图分类号: