Coastal Protection Forest Pest Forecasting Based On Wavelet Neural Network

Coastal shelter forest system has great practical and historical significance to improve the ecological environment in coastal areas against natural disasters and to protect people’s lives and property safety. But due to coastal protection forest tree species in a single, extensive management and humid climate, sufficient sunlight thus making pests easier outbreaks. Thus timely monitoring the occurrence of coastal protection forest pests,and making timely prevention is essential.The study applied Matlab programming wavelet neural network (Wavelet Neural Network WNN) to predict the occurrence of coastal protection forest pests. In the forecasting process, using SAS (Statistical Analysis System) Pearson correlation test and stepwise regression analysis to program screening forecast dominant factor. Finally, used the second layer coastal protection forest pests Dendrolimus punctatus as simulation tests in Xianju County, then compared the prediction effect with BP neural networkThe main conclusion of this study:1. SAS programming in a more convenient way to filter the dominant factor, even if the encounter a lot of basic data does not give any trouble forecasting work.2. Using SAS Pearson correlation test and stepwise regression analysis of simulation pests dominant meteorological factor screening; The results showed that the population density was significantly affected by the seven meteorological factors of22variables; Gondii strains rate by7meteorological factors of22variables significantly influence.3. The Wavelet Neural network prediction higher pest accuracy:2007-2011population density simulation accuracy rate more than90%; Insect area of simulation accuracy rate of more than95%in2007-2011; Worm strain rate simulation accuracy above96%in2007-2011.4. Predicted effect of wavelet neural network compared with BP neural network prediction results. The results showed that the average relative error of the wavelet neural network prediction of population density of only3.251%, and BP neural network to predict the relative error of the pest average of9.161%, and by analysis of variance P=0.0358<0.05; Wavelet neural network predicted the insect area average relative error was2.579%, BP neural network’s average relative error was6.570%, analysis of variance P=0.0334<0.05; Similarly, in the process to predict insect strain rate, the average relative error of wavelet neural network was1.9631%, the relative error of the BP neural network was8.0492%,analysis of variance P=0.0141<0.05. Wavelet neural network prediced experimentally simulation pests better than BP neural network.