基于卡尔曼滤波算法的稻纵卷叶螟短期预测模型

摘要

利用1994-2014年中国南方四大稻区（华南、西南、江岭和江淮稻区）代表性病虫测报站的稻纵卷叶螟逐候田间赶蛾量资料，筛选出影响各站稻纵卷叶螟发生量的关键气象因子，应用卡尔曼滤波方法分别对各站建立稻纵卷叶螟迁入期候发生量的卡尔曼短期预测模型，并计算模型的准确率、误差大小和稳定性。结果表明：（1）稻纵卷叶螟发生量与前一候和前两候的田间蛾量呈极显著正相关（P<0.01），与前一候的近地面最低气温、平均气温和最高气温呈极显著正相关（P<0.01），与前一候的地面气压呈极显著负相关（P<0.01）。（2）经1994-2011年的回检拟合和2012-2014年试报检验，卡尔曼模型的发生量预测综合平均误差为-88.63，平均绝对误差为217.72，均方根误差为605.04。发生量预测综合准确率为84.33%，平均历史拟合率为83.33%，各站卡尔曼模型的预报结果与实测值基本吻合，表明模型可以应用于稻纵卷叶螟候发生量的预测。%In this paper, the pentad systematic investigation data ofC. Medinalis at the four representative plant protection stations of four main rice-growing regions (including the rice-growing region of the south China, the rice-growing region of the southwestern China, the rice-growing region between the Nanling mountains and the Yantze River valley and the rice-growing region between the Yantze River valley and the Huaihe River valley) in China was collected from 1994 to 2014, the key meteorological factors influencing onC. Medinalis’ occurrence amount were screened out and Kalman filter algorithm was used to establish the short-term forecasting models ofC. Medinalis’ pentad occurrence amount at the four plant protection stations, including Quanzhou in the Guangxi Zhuang Autonomous Region, Xiushan in Chongqing city, Xiangyin in Hunan province and Zhangjiagang in Jiangsu province in the immigration and damage period ofC. Medinalisrespectively. Based on the back substitution fittings and forecasting tests of the model, the errors and stability and accuracy rates of the Kalman model were calculated. The results showed as follows: (1) for four stations, the occurrence amount ofC. medinalis in the present pentad was significantly and positively correlated (P<0.01) with theC. medinalis’s moth amounts of the preceding pentad and the preceding two pentads in the field respectively. There were significantly positive correlations (P<0.01) between the occurrence amounts ofC. medinalis in the present pentad and the minimum air temperature, mean air temperature and maximum air temperature in the preceding pentad. But the pentad occurrence amount was significantly and negatively correlated with the surface pressure in the preceding pentad. (2) The back substitution fitting calculations from the Kalman model on the occurrence amount ofC. Medinalis from 1994 to 2011 and the trial forecast tests from 2012 to 2014 showed that the comprehensive mean error (ME) of the occurrence amounts by the Kalman model was-88.63, the mean absolute error (MAE) was 217.72, the comprehensive root mean square error (RMSE) was 605.04, the comprehensive mean accuracies (MA) was 84.33%, and the fitting rate was 83.33%. The Kalman model’s forecasting results were basically consistent with measured values, which indicated that the model could be applied to the prediction of occurrence amount of C. medinalis .