基于粒子群优化神经网络混合模型预测马尾松毛虫发生量.pdfVIP

36 11 Vo.l 36 No. 11 2008 11 JOURNAL OF NORTHEAST FORESTRY UN IVERSITY Nov. 2008 1) 陈绘画 崔相富 杨胜利 (, , 317300) 针对 BP网络具有局部精确搜索但易陷入局部极小粒子群算法具有全局寻优的特点, 将二 结合 起来形成一种训练神经网络的新算法) ) ) PSO BP算法根据相关系数法均生函数法及逐步回归法分别选取 与马尾松毛虫有虫面积虫口密度有虫株率相关关系密切的气象因子和延拓均生函数序列作为样本的输入特征, 分别建立马尾松毛虫有虫面积虫口密度有虫株率与气象因子和延拓均生函数序列的 PSO BP 混合模型结果 表明: 建立的各PSO BP 混合模型, 具有令人满意的拟合精度和预测精度当隐含层神经元 个数为 13 个, 预报因 子数为 6 个时, 3组预留有虫面积的平均预测误差为 4. 01% ; 虫口密度 PSO BP 混合模型的隐层神经元 个数为 9 个, 预报因子数为 4 个时, 预留样本的平均预测误差为 3. 32% ;有虫株率 PSO BP 混合模型的隐层神经元 个数为 9 个, 预报因子数为 4 个时, 预留样本的平均预测误差为 2. 83% 上述 3 个指标的预测准确率均为 100% 马尾松毛虫; 粒子群神经网络; 发生量; 预测预报; PSO BP混合模型 TP183 O ccurren ce F orecastin g ofD endrolim us p un cta tu s Based on the N eu ral N etw ork M ix ed M odel of P ar ticle Sw arm O p tim ization / Chen Huihua, CuiX iangfu, Yang Shengli( theForestry Bureau ofX ianju County, Zhejiang Province, X ian ju 317300, P. R. Ch ina) // Journal ofNortheast Forestry University. 2008, 36( 11). 88~ 91 PSO BP algorithm, a new neural network algorithm, was established by comb ining BP newt ork, which is characterized by local accuracy search but easily trapping in local m inmi um, w ith Particle Swarm Optmi ization ( PSO) w ith global optmi ization. Accord ing to related coefficient, m ean generating function and stepw ise regression methods, meteorological factors corresponding to area of infestedwood, population density and the attacking rate ofD endrolmi us punctatus Walker, and continuation mean gen erating function sequence w ere selected as the character of mi ported samples to establish the PSO BP m ixed mode.l Resu lts showed that the establishedPSO BPm ixed models had satisfactory fitting accuracy and forecasting precisi