基于遗传算法优化神经网络的粮食产量组合预测研究

发布时间：2018-06-09 15:34

本文选题：遗传算法 + BP神经网络　；参考：《河南师范大学》2015年硕士论文

【摘要】：中共中央、国务院于2015年初印发了《关于加大改革创新力度加快农业现代化建设的若干意见》.意见指出:要不断增强粮食生产能力.由此可以看出,粮食产量是衡量一个国家经济实力的标准之一,是保障人民群众丰衣足食的不竭动力,是实现传统农业向现代农业过渡的重要保障.尽管我国的粮食产量年年稳定增长,但我们依然面临着很多的困难,例如:土地滥用、土地盐碱化、天气灾难等等,这些都是可能造成粮食减产的隐患因素.虽然,自古我国对于粮食危机均有相应丰富的应对经验,但相对于人口众多、粮食消费大、耕地资源匮乏的实际国情来说,保障农业安全和粮食的稳定可持续发展就成为农业科学研究中亟待解决的问题.因此,根据研究现阶段粮食生产发展的变动规律,并对其发展趋势进行预测,不仅可以为我国制定粮食政策与实施粮食生产系统控制提供决策依据,对保障国家粮食安全也具有重要的现实意义.本文首先分析了人工神经网络的概念、特性、网络结构和学习方式,在此基础上深入研究了BP神经网络、RBF神经网络和GRNN神经网络.在实际应用过程中,BP神经网络存在收敛速度较慢、甚至不能收敛的问题;初始连接权值、阈值和网络结构选择具有随机性,选取的初始点不一定具有全局性的问题,致使网络最后迭代出的结果也不一定是全局最优的问题.RBF神经网络存在隐含层单元通常是局部的,不能保证选择最优的隐层单元;隐层单元数量通常固定的,往往是通过经验选择,时间消耗大的问题.GRNN神经网络存在径向基函数的中心和宽度、隐含层到输出层的连接权选取对于神经网络的函数逼近能力具有很大的影响;且常规GRNN学习规则很容易使结果收敛到局部最小,甚至根本不收敛的问题.遗传算法是模仿自然界生物进化机制发展起来的随机全局搜索和优化方法,是一种高效、并行、全局搜索的方法,它能在搜索过程中自动获取和积累有关搜索空间的知识,并自适应地控制搜索过程以求得最优解.本文引入遗传算法对BP神经网络、RBF神经网络和GRNN神经网络进行优化,将优化后的GA-BP、GA-RBF和GA-GRNN网络用来预测粮食产量,以提高各单项模型的预测性能.传统的组合预测方法是按照单项预测方法的不同而赋予不同的加权平均系数,同一个单项预测方法在样本区间上各个时点的加权平均系数是不变的.然而实际上,就同一个单项预测方法而言,它在不同时刻的表现可能不相同,即在某个时点上预测精度较高,而在另一时点上预测精度较低.因此现有的组合预测方法存在与现实不符的缺陷.基于IOWA算子的组合预测模型,通过引进IOWA算子,对每个单项预测方法在样本区间上各个时点的拟合精度的高低按顺序赋权,以误差平方和为准则建立组合预测模型,因此,本文采用基于IOWA算子的组合预测模型将GA-BP、GA-RBF、GA-GRNN单项预测模型的结果融合,进一步提高预测精度.实验结果表明,采用本文方法可以有效提高粮食产量的预测精度.此外,本文在分析比较C#和MATLAB混合编程的几种方法的优缺点,利用C#作为前端开发环境,设计系统界面,并显示和输出结果;而采用MATLAB R2010a作为后端计算和图形绘制工具进行设计与开发,开发了粮食产量预测系统简化粮食预测流程,有效地减少了人工计算量,具有很广泛的应用前景.
[Abstract]:In the Central Committee of the Communist Party of China, in early 2015, the State Council issued a number of opinions on increasing the reform and innovation to speed up the construction of agricultural modernization. Although our country's grain output is steadily increasing year by year, we still face a lot of difficulties, such as land abuse, land salinization, weather disaster and so on. These are all potential risks of grain production reduction. Although in ancient times, China has a corresponding grain crisis. Rich in coping experience, but relative to the actual situation of large population, large food consumption and lack of arable land resources, it is an urgent problem to ensure the stability and sustainable development of agricultural safety and grain. Therefore, according to the study of the changing laws of the development of grain production and development at the present stage, the development trend is previewed. In this paper, the concept, characteristics, network structure and learning methods of artificial neural network are analyzed, and the BP neural network and RBF neural network are studied in this paper. And GRNN neural network. In the practical application process, the BP neural network has a slow convergence rate and even cannot converge. The initial connection weights, threshold and network structure selection are random, and the initial points selected are not necessarily global, and the results of the last iteration of the network are not necessarily the problem of global optimal. The RBF neural network has implicit layer units which are usually local and can not guarantee the optimal selection of hidden layer units. The number of hidden layer units is usually fixed. It is often selected through experience and the time consuming problem.GRNN neural network has the center and width of radial basis function, and the connection weight of the hidden layer to the output layer is selected for the neural network. The function approximation ability has a great influence, and the regular GRNN learning rules can easily make the result converge to the local minimum and even do not converge at all. The genetic algorithm is a stochastic global search and optimization method that imitates the evolution mechanism of natural organisms. It is a efficient, parallel, global search method. It can be used in the search process. It automatically obtains and accumulates knowledge about search space and adaptively controls the search process to get the optimal solution. In this paper, the genetic algorithm is introduced to optimize the BP neural network, RBF neural network and GRNN neural network, and the optimized GA-BP, GA-RBF and GA-GRNN networks are used to pretest grain output to improve the predictability of the single model. Yes. The traditional combination forecasting method is given the different weighted mean coefficients according to the difference of single prediction method. The weighted mean coefficient of the same single prediction method at each time point in the sample interval is constant. However, in fact, it may be different at different times in terms of the same single prediction method, that is, The prediction precision is high at some point, and the prediction accuracy is low at the other point. Therefore, the existing combination prediction method has the defects that are not consistent with the reality. The combined prediction model based on the IOWA operator, by introducing the IOWA operator, gives the right to the order of the fitting accuracy of each single prediction method at each time point in the sample interval. The square sum of error is a combination prediction model. Therefore, this paper uses a combination prediction model based on IOWA operator to combine the results of GA-BP, GA-RBF, GA-GRNN single item prediction model to further improve the prediction accuracy. The experimental results show that the prediction accuracy of high grain yield can be effectively raised by this method. In addition, the analysis ratio is also analyzed in this paper. Compared with the advantages and disadvantages of several methods of C# and MATLAB, C# is used as the front-end development environment, the system interface is designed, and the results are displayed and output, while MATLAB R2010a is used as the back end calculation and drawing tool to design and develop, and the grain yield prediction system is developed to simplify the grain prediction process and effectively reduce the artificial meter. It has a wide application prospect.
【学位授予单位】：河南师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F326.11;TP183

【参考文献】