摘要
针对BP神经网络在参考作物腾发量(ET 0)计算方面存在稳定性差、易陷入局部最优的缺陷,通过平均影响值算法(MIV)和SPSS软件对影响ET 0的变量进行筛选,消除多重共线性,同时计算各变量对ET 0的相对贡献率,选择相对贡献率较高的3个变量作为输入;同时,提出改进的非线性随机递减蝙蝠算法(NRDBA)对BP神经网络的权值和阈值进行优化,该算法将非线性随机递减权重引入速度更新公式中以提高算法的全局收敛性。将改进的NRDBA-BP算法应用于商丘地区的ET 0预测中,并建立BP、BA-BP和PSO-BP三种预测模型与之进行对比。实验结果表明,4种模型中,NRDBA-BP模型的R 2值最大,MSE最小,这表明提出的NRDBA-BP模型与ET 0的真实值更加接近,其预测精度更高,可以有效提高ET 0的预测能力。
Aiming at the poor stability and inclination to fall into local optimal of BP neural network in ET 0 calculation,this paper uses the average influence value method and SPSS software to screen the variables that affect ET 0,eliminate multicollinearity,and calculate the relative contribution rate of each variable to ET 0,and the three variables with higher relative contribution rate are selected as inputs.At the same time,an improved nonlinear random decreasing bat algorithm(NRDBA)is proposed to optimize the weights and thresholds of BP neural network,which introduces non-linear random decreasing weights into the speed update formula to improve the global convergence of the algorithm.The improved NRDBA-BP algorithm is applied to the ET 0 prediction in Shangqiu area,and three prediction models of BP,BA-BP and PSO-BP are established for comparison.According to relevant experimental results,among the four models,the R 2 value of NRDBA-BP model is the largest,while the MSE value is the smallest,this indicates that the proposed NRDBA-BP model is closer to the true value of ET 0,and its prediction accuracy is higher,which can effectively improve the prediction ability of ET 0.
作者
任亚飞
田帅
邵馨叶
邵建龙
REN Ya-fei;TIAN Shuai;SHAO Xin-ye;SHAO Jian-long(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Information Center of Yunnan Power Grid Co.,Ltd.,Kunming 650200,China)
出处
《陕西理工大学学报（自然科学版）》
2021年第2期48-56,共9页
Journal of Shaanxi University of Technology:Natural Science Edition
基金
国家自然科学基金资助项目(61302042)
昆明理工大学教育技术研究项目(2506100219)。
关键词
参考作物腾发量
平均影响值
改进蝙蝠算法
非线性随机递减
预测模型
reference crop evapotranspiration
mean influence value
improved bat algorith
non-liner random decline
prediction model
