基于PCA建立蜡梅花初花期预测模型  

Establishment of Prediction Model of Early Flowering Period of Prunus Mume Based on PCA

作　　者：吉莉 刘晓冉 武强 李强 JI Li;LIU Xiaoran;WU Qiang;LI Qiang(Weather Bureau in Beibei District of Chongqing City,Chongqing 400700,China;Chongqing Institute of Meteorological Sciences,Chongqing 401147,China)

机构地区：[1]重庆市北碚区气象局,重庆400700 [2]重庆市气象科学研究所,重庆401147

出　　处：《西南师范大学学报（自然科学版）》2022年第10期59-66,共8页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：重庆市市场监管局第二批地方标准制修定计划项目(2022-71);北碚区科委项目(2022-32)。

摘　　要：以重庆市北碚区静观素心蜡梅早熟品种的初花期为研究对象,统计分析2007-2021年初花期变化特征,并基于主成分分析法(PCA),通过BP神经网络算法及逐步回归算法,构建了2007-2021年初花期预测模型,对2种预测模型的预报效果进行对比检验,筛选最优预测模型.结果表明:基于BP神经网络算法的预测模型在训练中的预报拟合率高达99%,与实测值的相关性超过了0.9,拟合度较高,在回代检验中拟合率低于训练时;基于逐步回归算法的预测模型在训练中与实测值误差大于基于BP神经网络算法,平均误差为1.7 d,在回代检验中效果明显优于基于BP神经网络算法,且线性相关性也较稳定;同时在回代模型中基于逐步回归算法的预测模型的独立样本值、标准差和平均绝对误差也同样优于基于BP神经网络算法的预测模型.总体来说,基于逐步回归算法的预测模型更优于基于BP神经网络算法的预测模型.Taking the early blooming period of Chongqing Beibei Jingguan Suxin Chimonanthus as the research object,the variation characteristics of the early blooming period from 2007 to 2021 were statistically analyzed.Based on principal component analysis(PCA),a prediction model for the early blooming period from 2007 to 2021 was constructed by BP neural network algorithm and stepwise regression algorithm.The prediction effects of the two prediction models were compared and tested to select the best prediction model.The results show that the prediction fitting rate of the prediction model based on BP neural network algorithm in training is as high as 99%,and the correlation with the measured value is more than 0.9.In the back substitution test,the fitting rate was lower than that in training;the error between the prediction model based on stepwise regression algorithm and the measured value in training is greater than that based on BP neural network algorithm,with an average error of 1.7 d.In the back substitution test,the effect is significantly better than what is based on BP neural network algorithm,and the linear correlation is also relatively stable.At the same time,the independent sample value,standard deviation and average absolute error of the prediction model based on stepwise regression algorithm are also better than the prediction model based on BP neural network algorithm.In general,the prediction model based on stepwise regression algorithm is better than the prediction model based on BP neural network algorithm.

关 键 词：气象因子 主成分分析法 BP神经网络 逐步回归 预测模型 

分 类 号：P49[天文地球—大气科学及气象学] S16[农业科学—农业气象学]