基于支持向量机回归的NDVI组合预测模型

摘要

The MODIS NDVI data from 2004 to 2015 are used to make a time sequence.By using the temperature and precipitation data during this period as regression factor,support vector machine regression model is used to establish the NDVI short-term prediction model.First,grid search method,genetic algorithm,particle swarm optimization are used to optimize model parameters respectively.Then use the best parameters to train support vector machine respectively.The resuits show that the grid search method is the best parameter optimization algorithm.Build two single prediction model of NDVI from different angle using support vector machine regression model based on the grid search method.Do a linear combination with the two single prediction model and calculate the optimal weight coefficient.The results show that the combined model can predict NDVI effectively.%对2004年到2015年3～ 10月的MODIS NDVI数据建立时间序列,并利用同期的温度、降水数据做回归因子,采用支持向量机回归模型建立NDVI短期预测模型.首先用网格搜索法,遗传算法,粒子群算法对模型参数进行优化选择,然后用得到的最佳参数分别训练支持向量机,拟合结果显示网格搜索法是本实验数据的最佳优化算法.使用基于网格搜索法的支持向量机回归模型从2个角度建立了NDVI的单项预测模型,对2个单项模型做线性组合并计算最优权重系数,实验结果表明组合模型可以有效预测NDVI.