马尾松毛虫幼虫发生高峰期的三种预测模型
The three models for forecasting the peak larval period of Dendrolimus punctatus
余 燕；王振兴；李 尚；周夏芝；毕守东；方国飞；张国庆
点击：2019次 下载：5次
DOI：10.7679/j.issn.2095-1353.2018.091
作者单位：安徽农业大学理学院，合肥 230036；安徽农业大学林学与园林学院，合肥 230036； 国家林业局森林病虫害防治总站，沈阳 110034；安徽省潜山县林业局，潜山 246300
中文关键词：马尾松毛虫幼虫高峰期，平稳时间序列法，BP神经网络法，马尔科夫链法
英文关键词：the larvae peak period of Dendrolimus punctatus, periodical and regression method, stationary time series method and BP neural network, Markov chains and contingency table analysis method
中文摘要：
【目的】 为了科学确定马尾松毛虫Dendrolimus punctatus防治适期，提高防治效果。【方法】 采用平稳时间序列法、马尔科夫链法和BP神经网络法研究安徽省潜山县1983-2014年马尾松毛虫幼虫发生的高峰期预测模型，并用2015年和2016年发生资料进行验证。【结果】 显示：平稳时间序列法、马尔科夫链法、BP神经网络法模型预测2015年和2016年1代幼虫高峰期均为6月5日，2代高峰期均为9月6日，两年的1、2代实际发生期也是6月5日和9月6日，预测值与实际值完全一致。平稳时间序列法预测1代幼虫高峰期结果，若以大于2 d为误差标准，则1983-2014年的历史符合率为96.77%；若以小于和等于1 d为误差标准，历史符合率为74.19%，2代幼虫高峰的预测结果的历史符合率与1代相同。BP神经网络法预测结果若以1 d为误差标准，1983-2014年则1、2代预测结果的历史符合率均为100%。【结论】 在3种方法中，平稳时间序列法适用于害虫发生过程符合平稳时间序列的情况；马尔科夫链法预报量的分级标准科学与否直接影响预测结果的准确性；BP神经网络法可用于自变量与预报量非线性关系的研究，是一种比较理想的预报方法。
英文摘要：
[Objectives] To scientifically determine the optimum period for controlling Dendrolimus punctatus, and thereby improve the effectiveness of current control methods. [Methods] Stationary time series, BP neural network and Markov Chain analysis were used to develop models predicting the timing of peak larval abundance from 1983 to 2016 in Qianshan County Anhui Province. The predictions of these models were verified in 2015 and 2016. [Results] The peak of the first larval generation was predicted to occur on June 5 and that of the second generation on September 6. The predicted dates were exactly the same as the actual dates in 2015 and 2016. The stationary time sequence method predicted the timing of peak of larval abundance with 96.77% accuracy from 1983 to 2014 if the error criterion was > 2 days. If the error criterion is less ≤1, the accuracy was 74.19%. The accuracy of the prediction of the second generation larval peak was the same as that for the first generation. For the BP neural network, if the error criterion was one day, the accuracy of the predicted outcome was 100% from 1983 to 2014. [Conclusion] The stationary time sequence method was applicable to stable time series for predicting peak larval abundance whereas the Markov Chain was directly influenced by the accuracy of the prediction results. The BP Neural Network method can be used to determine nonlinear relationships between independent variables and predicted larval abundance, which is an ideal prediction method.