研究粮食生产规律，做好粮食产量科学预测是制定农业政策的重要依据。本文以宁夏地区小麦的生产情况为研究对象，结合统计学习理论的相关知识和结构风险最小化原理，选择适合的科学原理和预测方法，通过针对性课题研究来建立预测模型，系统地对模型进行处理与分析，研制与实现小麦的产量预测系统，为粮食生产服务。 本文重点讨论了最小二乘支持向量机的算法，针对小麦生长特点，确定出影响小麦产量的主要因子，采用最小二乘支持向量机算法，建立了基于宁夏地区1981年至2010年的小麦产量预测模型。预测结果表明，该模型具有较高的预测精度；同时，研究了相应的预处理、图形界面交互等技术，实现了宁夏地区的小麦产量预测系统，该系统具有较好的扩展性，可供农业等相关部门使用，为粮食产量预测提供了一条新的途径。 首先，结合统计学习理论的相关内容，分析与学习支持向量机和最小二乘支持向量机的理论与应用机理，建立粮食产量预测模型。其次，以宁夏地区小麦产量数据为基础，研究和改进预测算法，进行相关预测实验分析，得出基于最小二乘支持向量机的小麦产量时间序列预测结果，分析其预测结果良好。然后，鉴于小麦生长过程的复杂性和信息不完全性，将气候数据引入作为小麦产量的主要影响因素，研究与改进相应的数据预处理技术，提出基于最小二乘支持向量机和气候数据的小麦产量预测方法，实验结果表明，该方法优于传统的时间序列预测方法。再次，针对小麦生产系统的多因素性，提出一种基于时间权重的最小二乘支持向量机的预测方法，旨在提高预测精度。最后，在基于最小二乘支持向量机和气候数据的小麦产量预测建模的基础上，研究了相应的图形界面交互技术，进行系统性开发，形成宁夏地区的小麦产量预测系统。 实验证明，最小二乘支持向量机不仅弥补了支持向量机所求解的一些不足，在保证预测精度的前题下，还有效的提高了大样本学习的求解速度。

Researching on the laws of grain production and carrying out scientific prediction on grain yield are the fundamentals for making agricultural policies. This paper takes the production situation of wheat in Ningxia area as the research object and aims to establish a prediction modeling line with relevant knowledge of statistical learning theory and principles of structural risk minimization by choosing proper scientific principles and prediction methods, and will systematically processing and analyzing the model to research and realize the wheat yield prediction system for grain production. This paper focuses on discussing the algorithm of Least Square Support Vector Machines (LS-SVM), determining the main factors affecting wheat yield in line with the growth features of wheat by means of LS-SVM algorithm, a wheat yield prediction model has been built for the wheat yield in Ningxia area from 1981 to 2010. The prediction results show that this model has high prediction accuracy. Meanwhile, the paper also researches such technologies as preprocessing and graphical interface interaction etc. and has realized the wheat yield prediction system in Ningxia area. With good expansibility, this system can be used by relevant departments such as agricultural department, etc., and it will provide a new approach for predicting grain yield. Firstly, this paper analyze and make study on the theoretical and application mechanisms of support vector machineand LS-SVM based on relevant contents of statistical learning theory, and builds a prediction model of grain yield. Secondly, it has researched and improved the prediction algorithm on the basis of the wheat yield data in Ningxia area; it also carries out relevant prediction and experimental analysis, and obtains the results of time series prediction of wheat yield based on LS-SVM. Analysis has showed that its prediction results are good. Then in view of the complicated growth process of wheat and the incomplete information, climate data is introduced as the main influencing factor affecting wheat yield to research and improve the corresponding data pre-processing technology. The paper proposes the method for wheat yield prediction based on LS-SVM and climate data. The experiment results show that this method is superior to traditional time series prediction method. In addition, considering the multi-factor property of wheat production system, it also proposes a prediction method of LS-SVM based on time weight, aiming at increasing the prediction accuracy. Finally, on the basis of the wheat yield prediction model based on LS-SVM and climate data, the paper researches corresponding graphical interface interaction technologies to conduct systematically development, thus forming the wheat yield prediction system in Ningxia area. These experiments prove that the LS-SVM have not only covered some shortages of support vector machine, but also have effectively increased the speed of large sample learning at the premise of ensuring the prediction accuracy.