论文部分内容阅读
为了提高马尾松毛虫幼虫发生严重程度的预测精度,寻求简便准确的预测方法,采用时间平稳序列法、回归预测法、马尔科夫链法、BP神经网络法和列联表多因子多级相关分析法对安徽省潜山县1983—2014年的马尾松毛虫越冬代、一代和二代幼虫发生的严重程度进行预测,研究历史符合率,并用2015年和2016年的实际发生情况验证。结果表明,平稳时间序列法,列联表多因子多级相关分析法计算简便,预测结果准确;BP神经网络法和马尔科夫链法预测结果非常准确。回归模型中以当代卵盛期卵量预测当代幼虫发生严重程度的一元回归模型的预测结果准确性高,其余一元回归模型预测结果稍差,多元回归模型和逐步回归模型优于一元回归模型。BP神经网络模型是一种理想的预测模型。
In order to improve the prediction precision of Dendrolimus punctatus larvae, a simple and accurate prediction method was proposed. Time-series method, regression prediction method, Markov chain method, BP neural network method and Multivariate multistage correlation analysis The method was used to predict the severity of overwintering, first generation and second generation larvae of Dendrolimus punctatus from 1983 to 2014 in Qianshan County, Anhui Province. The historical coincidence rate was studied and verified with the actual occurrence in 2015 and 2016. The results show that the method of stationary time series and multi-factor multistage correlation analysis of contingency table is simple and the prediction result is accurate. The prediction results of BP neural network and Markov chain are very accurate. In the regression model, the prediction accuracy of one-dimensional regression model predicting the severity of contemporary larvae was high with the predictive value of the egg number in the current stage. The results of the other one-step regression models were slightly worse, and the multiple regression models and stepwise regression models were superior to the one-way regression models. BP neural network model is an ideal prediction model.