基于相空间重构及GRNN的海防林害虫预测及效果检验

摘要

考虑到海防林环境气候、动植物组成的特殊性及林业害虫发生的非线性、复杂性、无规律性，通过重构非线性时间序列的相空间并将输出向量作为广义回归神经网络（GRNN, Generalized Regression Neural Network）的输入向量对海防林害虫的发生进行预测；并使用仙居县1983-2011年的马尾松毛虫（Dendrolimus punctatus）实测数据对该预测方法进行仿真检验：1983-2007年虫害数据用于重构相空间及训练神经网络，而2008-2011年的数据用于虫害发生的预测检验。结果表明，在将仙居县马尾松毛虫的危害程度划分为四个等级的前提下，预测方法预测精度较高，除了对2008年马尾松毛虫危害程度的预测出现一个数量级的偏差外，其余的预测准确率均为100%。%In this paper, reconstructed the phase space of the nonlinear time series and used the output vector as the input vector of Generalized Regression Neural Network (GRNN) to predict the occurrence of coastal protection forest pests.Thespecific method wasusingwavelet transform, noise reduction processingoftheraw data, then calculate thetime delay,embedding dimension and Lyapunov exponent, reusedcalculateddata reconstruction phase space. Finally,using theGRNNto predict thetest. Simulationtestdata was from theXianjuCounty from1983 to 2011. Specific ideas:pestdata of 1983-2007 forthereconstruction phase spaceandtrain the neural network, data of 2008-2011forpestprediction test. The result showed that theprediction has higherprediction accuracy, 2008(forecastdeviationisonlyan order of magnitude), therestoftheforecastaccuracies are all 100%. In short,thephase space reconstructionandGeneralizedRegression NeuralNetworksare allpopularmethodto researchnonlinear problems, this papercombined them asanew attempttopredicttheoccurrenceofcoastal protection forestpests.