基于神经网络的灯诱法预测马尾松毛虫发生量的研究

Using light traps to forecast the occurrence of Dendrolimus punctatus (Walker)

【目的】为了探索灯诱马尾松毛虫Dendrolimus punctatus(Walker)成虫预测下一代幼虫发生量的办法,建立预测预报模型。【方法】2013-2017年连续5年采用灯诱法对马尾松毛虫每年的越冬代、第1代灯诱成虫数、雌虫数、雄虫数、雌性比等数据进行收集,实地调查第1代、第2代(越冬代)幼虫林间的发生量(虫口密度),采用Excel2016进行相关性分析,筛选出与下一代幼虫发生量(虫口密度)关系密切的灯诱成虫数、雌虫数、雌性比等关联因子,并应用神经网络Matlab2016a建立预测模型。【结果】所建立的灯诱成虫预测模型,拟合度0.92以上,预测精度0.90以上。【结论】采用灯诱马尾松毛虫成虫预测下一代幼虫发生量的神经网络Matlab模型,十分适用于短期精细化预报,方法简单、实用,值得在马尾松毛虫生产性防控中大力推广。

[Objective] To develop a model to predict the abundance of larvae of the next generation of Dendrolimus punctatus from the number of adults caught in light traps. [Methods] For five consecutive years from 2013-2017 the number of adult females and males captured in light-traps, and the sex ratio of the overwintering and first generations of D. punctatus, were collected. During the same period the abundance（density） of the first and second（overwintering） larval generations were investigated in the field. Factors correlated with the abundance of the next larval generation, such as the number of adults, and sex ratio, were identified using Excel 2016. A predictive model using neural networks was established with Matlab 2016. [Results] The predictive model had a better than 0.92 fit to the actual data and was more than 0.90 accurate. [Conclusion] The Matlab model for predicting the abundance of the next larval generation of D. punctatus on the basis of adult light trapping data is very suitable for short-term, fine prediction. This method is simple and practical, and worth popularizing for the control of D. punctatus.