基于气象和时空因子的森林火险等级预测

Prediction of forest fire danger rating based on meteorological and space-time factors

传统森林火险预测方法在特征选择上主要依据气象因子,未考虑时间因子、空间地理位置、地形地貌等因素,可能在林火预测中产生较大误差,传统预测模型在面对不同林区更大的数据集和预测任务时,无法快速自动调参和自适应调整输入特征维度。笔者引入时间因子和空间因子,丰富火险预测特征,以降低森林火险等级预测中产生的误差;为解决传统火灾预测算法在大数据集和多分类问题上效率逐渐下降问题,提出核主成分分析算法(kernel principal component analysis,KPCA)和改进的极限学习机算法(extreme learning machine,ELM)相结合的森林火险等级预测模型。结果表明,该模型能有效提高森林火险等级预测的准确率和执行效率。相对传统预测模型,其准确率可达89%,在预测时间上也有一定优势。

The forest fire is a serious natural disaster, causing great damage to the economy and ecology. The meteoro- logical factors affect the forest fire potential. Global warming may result in more forest fires, and the geographic loca- tion and time are also important factors for forest fire danger rating. Therefore, the empirical analysis of the relation- ship between forest fires and these factors is of great significance. Generally, the areas where forest fires occur have spatial heterogeneity. In order to make forest fire prediction free from geographical restriction, it is necessary to put forward a method to provide technical support for forest fire management in most areas. In this study, the space-time factor was introduced to the forest fire danger rating model based on the meteorological factors to enrich the predicted features and reduce the predicted error. The predicted algorithm was optimized to improve the model efficiency and ac- curacy. In this study, a new predicted model based on kernel principal component analysis （KPCA） and extreme learning machine （ELM） were used. First, we analyzed the deficiency by using principal component analysis （PeA） to deal with the general nonlinear problems. Then, we expounded the PCA method based on kernel function and ap- plied it to reduce the dimension of forest fire danger rating index. Next, we used memetic algorithm to optimize the input weight and implicit layer deviation of the ELM, and classified the hidden layer nodes to reduce the training time of the ELM in order to obtain the best prediction model. Finally, we combined this model and cross-validation algo- rithm to rate forest fire danger. The experimental results showed that the model can effectively improve the accuracy of forest fire danger rating and has significant advantages in parameter selection and training speed.