桂林地区稻飞虱发生等级气象预报模型

Meteorological forecast models for the occurrence grade of rice planthopper in Guilin of Guangxi,South China

利用广西桂林地区稻飞虱历史虫情以及相应站点同期地面气象观测资料,采用因子膨化技术对气象因子进行逐候膨化组合,通过相关分析筛选出与虫害发生相关性最为显著的关键因子及其组合时段,利用SPSS软件建立稻飞虱发生程度等级逐候气象预报多元线性回归模型,利用MATLAB软件建立稻飞虱发生程度等级逐候气象预报BP人工神经网络模型。结果表明:基于因子膨化技术的两种模型达到"一致"和"基本一致"的预报准确率在88%以上;其中人工神经网络在历史回代检验中对稻飞虱发生程度等级达到"一致"和"基本一致"的预报准确率比线性回归模型提高4%,在外推预报中对稻飞虱发生程度等级达到"一致"的预报准确率比线性回归模型提高14%;可见,利用因子膨化技术结合BP人工神经网络技术建立短期预报预警模型,不仅可实现稻飞虱发生程度等级的逐候动态预报,而且预报准确率和稳定性明显提高。

By using the historical data of rice planthopper in Guilin of Guangxi and related mete- orological data, the factors puffing technology was employed to assemble the meteorological data in pentad scale. Through the correlation analysis of the meteorological data and the occurrence of rice planthopper, the key factors and their assembling periods most significantly related with the occurrence of rice planthopper were screened. Two meteorological forecast models for the occur- rence grade of rice planthopper in pentad scale were built, of which, one applied multiple linear regression model was built by using SPSS software, and the another applied BP artificial neural network model was built by using MATLAB software. Based on the factors puffing technology, the prediction result as ＂correct＂ and ＂basically correct＂ of the two models were above 88%. The prediction results as ＂correct＂ and ＂basically correct＂ of the artificial neural network model was increased by 4% and the prediction result as carreet was increased by 14% , according to the historical samples and the independent samples, respectively. Therefore, a short-term forecasting model built with factors puffing technology and BP artificial neural network model could not only realize the dynamic updating and forecasting of the occurrence grade of rice planthopper, but also evidently improve the prediction accuracy and stability.