【作者】 张磊；

【导师】 蒋华伟； 李战升；

【作者基本信息】 河南工业大学 ， 计算机技术（专业学位）， 2021， 硕士

【摘要】 小麦作为中国最重要的粮食作物之一,其储藏安全对于保障社会稳定发展和居民健康生活具有重要的意义,预测小麦品质有助于掌握劣变趋势情况,对保障小麦的储藏安全起着重要作用。小麦品质陈化变质会伴随着生理生化指标的数值变化,预测多指标数据变化趋势有利于及时发现小麦品质变化和减少储藏损失。针对现有小麦品质预测方法在准确性、稳定性、可信度等方面存在的不足,本文选择机器学习算法来探索多指标变化规律的相似性及差异性,实现了小麦多指标权重贡献作用的量化分析、单指标时序变化的准确预测、多指标整体品质的稳定预测、小麦品质状态类别的量化预测,为小麦储藏品质的精准预测分析提供一定的理论与技术支持。本文的主要创新之处如下:(1)小麦多指标权重计算方法研究针对现有研究工作中较少涉及多指标间复杂作用的问题,本文提出了基于信息熵模型的小麦多指标权重计算方法。由条件熵量化计算小麦单个指标从其它指标数据中获取的信息量,根据互信息量对小麦多指标的差异性作用进行权重分配,为准确预测小麦品质奠定了基础。(2)小麦品质单指标时序预测研究为提高不同储藏时期内小麦品质单指标时序预测的准确性,本文提出一种长短期记忆生成对抗网络模型。通过改进长短期记忆网络获得单指标的时序特征,基于对抗学习方法提取其它指标的数据变化特征,来进一步提升模型的预测性能,从而对小麦品质不同指标进行精准预测分析。(3)多指标小麦整体品质预测研究针对小麦品质劣变机理复杂导致的由单指标数据预测小麦品质信息量不足和稳定性不高的问题,本文提出并构建一种多指标宽度自适应提升算法模型。通过对多指标数据集进行宽度特征变换以获得特征学习器、增强学习器,并由AdaBoost算法加权组合为性能更优的强学习器,进而对小麦品质的整体变化规律进行预测分析。(4)小麦品质状态类别预测研究在单指标时序预测和多指标整体特性分析的基础上,为实现小麦品质状态的量化分类及准确预测,本文提出一种半监督小麦品质状态类别预测模型。分别采用改进孪生支持向量机、密度峰值聚类算法对小麦多指标数据集进行品质状态归类,并通过构建半监督协同分类框架以实现多指标小麦储藏品质状态类别的准确预测分析。更多还原

【Abstract】 As one of the most important food crops in China,the storage safety of wheat is of great significance to ensure the stable development of society and the healthy life of residents.Predicting wheat quality is helpful to understand the deterioration trend and plays an important role in ensuring the storage safety of wheat.The aging and deterioration of wheat quality is accompanied by the numerical changes of physiological and biochemical indexes.Predicting the variation trend of multi-index data is helpful to detect the change of wheat quality in time and reduce the storage loss.Aiming at the shortcomings of the existing wheat quality prediction methods in the aspects of accuracy,stability and reliability,the machine learning algorithm was selected in this paper to explore the similarity and difference of the change rules of multiple physiological and biochemical indexes,it realized the quantitative analysis of the contribution of wheat multi-index weight,the accurate prediction of single index time series change,the stable prediction of multi-index overall rule,and the quantitative prediction of wheat quality state category,which can provide some theoretical and technical support for the accurate prediction and analysis of wheat storage quality.The main innovations of this paper are as follows:(1)Study on the calculation method of wheat multi-index weightsIn view of the existing research work that rarely involves the complex effects among multiple indexes,a method for calculating the weight of multiple indexes of wheat based on the information entropy model was obtained in this paper.This paper was based on the conditional entropy quantitative calculates the information obtained by a single wheat index form other indexes,and weights were assigned to the differential effect of multiple indexes of wheat according to the amount of mutual information,which can laid a foundation for predicting wheat quality accurately.(2)Study on time series prediction of single index of wheat qualityIn order to improve the accuracy of single index prediction of wheat quality during different storage periods,a long and short memory generation adversarial network model was proposed in this paper.The model was based on the improved long short-term memory network to obtain the time series characteristics of a single index,and by extracting the data change characteristics of other indexes based on generative adversarial network to further improve the prediction performance of the model,so as to accurately predict and analyze different indexes of wheat quality.(3)Study on the prediction of multi-index wheat overall qualityFor the problem of insufficient information and low stability of single index data in predicting wheat quality due to the complex mechanism of wheat quality deterioration,a width-adaptive improvement algorithm model was constructed based on multi-index.In this model,a feature learner and an enhancement learner were obtained through width feature transformation of multi-index data sets,and the weighted combination of the Adaboost algorithm was used to make a stronger learner with better performance,and then the change law of the whole wheat quality was predicted and analyzed.(4)Study on classification prediction of wheat quality statusOn the basis of the time series prediction of wheat quality indexes and the analysis of the overall characteristics of multiple indexes,further research is carried out in this paper.In order to achieve the quantitative analysis and accurate prediction of different quality states of wheat during storage,a semi-supervised wheat quality category prediction model was proposed in this paper.In this model,the improved twin support vector machine and density peak clustering algorithm were used to classify the quality state of wheat storage,and a semi-supervised collaborative classification algorithm was used to realize the accurate prediction and analysis of the multi-index wheat storage quality states categories.更多还原