差分GWO优化RBFNN模型及粮食产量预测应用  

Differential evolution GWO optimized RBFNN and its grain yield prediction application

作　　者：张小庆 许荣杰 冯晓祥 叶亮 ZHANG Xiao-qing;XU Rong-jie;FENG Xiao-xiang;YE Liang(School of Mathematics and Computer Science,Wuhan Polytechnic University,Wuhan 430023,China)

机构地区：[1]武汉轻工大学数学与计算机学院,湖北武汉430023

出　　处：《计算机工程与设计》2024年第12期3802-3811,共10页Computer Engineering and Design

基　　金：武汉轻工大学校立科研基金项目(2023Y44);湖北省教育厅科技基金项目(B2020063)。

摘　　要：针对粮食产量预测方法预测精度的不足,提出一种融入差分进化自适应灰狼算法优化正则项径向基神经网络的粮食产量预测模型DEGWO-RBFNN。为提高灰狼算法的搜索精度,引入指数分布随机数初始化种群,提升初始种群质量;设计Sigmoid函数自适应缩放因子均衡算法搜索与开发;引入差分进化提高全局搜索能力。利用改进GWO搜索RBFNN超参数,解决网格调参易陷入局部最优及初值敏感的不足。实验结果表明,与GWO-RBFNN、RBFNN、DE-RBFNN、BPNN、GA-BPNN、支持向量机、随机森林相比,DEGWO-RBFNN预测精度达到96.06%,比对比模型可提高2.47%~14.79%。In response to the shortcomings of low prediction accuracy and large errors in current grain yield prediction methods,a grain yield prediction model DEGWO-RBFNN was proposed,in which the differential evolution adaptive grey wolf algorithm was integrated to optimize the regularization term radial basis function neural network.To improve the search accuracy of GWO,an exponential distribution random number was introduced to initialize the population and improve the initial population quality.An adaptive scaling factor was designed to balance global search and local development that combined sigmoid function.A differential evolution mechanism was introduced to improve the global search ability.The improved GWO was utilized to search for key parameters of radial basis function neural networks,solving the shortcomings that mesh tuning parameters is easy to fall into local optima and is sensitive to initial parameter values.The results show that compared with GWO-RBFNN,RBFNN,DE-RBFNN,BPNN,GA-BPNN,support vector machine and random forest,the prediction accuracy of DEGWO-RBFNN reaches 96.06%,which can be improved by 2.47%to 14.79%compared to that of the comparative models respectively.

关 键 词：径向基神经网络 粮食产量预测 灰狼优化算法 差分进化 指数分布 自适应缩放因子 正则项 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]