【作者】 张旭；

【导师】 奚小波； 丰有财；

【作者基本信息】 扬州大学 ， 农业硕士（专业学位）， 2025， 硕士

【摘要】 随着全球气候变化和人口增长的挑战,农业生产面临着日益严峻的压力。提高作物产量和光合作用效率成为应对这一挑战的关键所在。传统农业生产优化方法已难以满足现代农业的高效、精确需求。近年来,机器学习技术的快速发展为农业生产提供了全新的解决思路。本论文基于机器学习方法,结合作物光合作用效率的提升,探索了提高作物产量的有效途径。通过仿真和模型优化,本文提出了一种利用机器学习算法对作物产量进行预测的框架,并通过不同机器学习模型的比较与分析,优化了预测效果。论文主要研究工作如下:(1)结合本研究的背景与目的,明确了气候变化和人口增长对农业生产的影响,特别是如何通过提升光合作用效率来促进作物产量的提升。随着全球气候变化的加剧,传统的农业生产方式面临着不小的挑战,尤其是对气候变化的适应性要求越来越高。本文通过机器学习技术对农业生产过程进行优化,为农业生产提供科学的决策支持。(2)分析机器学习技术的基本概念与原理,包括人工神经网络(ANN)、支持向量机(SVM)、AdaBoost算法等,分析了这些算法的特点和适用范围,为后续实验提供了理论依据。(3)本研究采用系统化的机器学习方法实现农业产量预测模型的构建与应用。首先对原始农业数据进行预处理,包括缺失值填补、异常值检测与剔除,并通过标准化和归一化处理消除特征间的量纲差异。在特征工程阶段,采用主成分分析(PCA)进行特征降维,并构建了包含气象因子、土壤参数和遥感指标的综合特征集。基于Python平台,实现了三类机器学习算法:1)ANN模型采用双层感知器架构,包含16个神经元的隐藏层和Re LU激活函数,通过Adam优化器进行训练;2)SVM模型选用RBF核函数,采用网格搜索优化惩罚系数和核参数;3)AdaBoost模型集成50个决策树基学习器,通过迭代加权提升模型性能。(4)本研究对机器学习模型的训练结果进行了深入分析,比较了不同算法在预测地上总干重、单株籽粒数和籽粒与茎秆比方面的表现。结果表明,AdaBoost算法在提升预测准确性和模型稳定性方面表现最为出色,尤其在处理复杂非线性数据时,相较于ANN和SVM算法具有显著优势。同时,ANN算法在处理大规模数据及多变量非线性关系时拟合能力较强,能提供更优的预测效果;而SVM算法在数据量小或分布复杂时表现稍逊。通过均方误差和决定系数等指标对模型综合评价后,确定了最适合本研究的机器学习算法,并深入分析了自变量对因变量的影响权重,为后续农业数据分析提供了有力参考。本论文的研究创新点在于将机器学习算法应用于光合作用效率与作物产量预测的领域,尤其是在遥感数据与作物生长模型相结合的基础上,构建了一个高效的预测框架。通过对比不同算法的预测结果,提出了一种基于机器学习的作物产量预测模型,并优化了光合作用效率的提高策略,为提高农业生产力和应对气候变化提供了科学依据。更多还原

【Abstract】 (1)With the challenges posed by global climate change and population growth,agricultural production faces increasing pressure.Enhancing crop yield and improving photosynthetic efficiency have become key solutions to addressing these challenges.Traditional agricultural optimization methods can no longer meet the demands of modern precision and efficient farming.In recent years,the rapid advancement of machine learning technology has provided new approaches to agricultural production.This paper explores effective ways to improve crop yield by leveraging machine learning techniques in conjunction with photosynthetic efficiency enhancement.Through simulation and model optimization,we propose a framework for crop yield prediction using machine learning algorithms and optimize the prediction performance by comparing different machine learning models.The main contributions of this study are as follows.(2)This study presents the background and objectives,highlighting the impacts of climate change and population growth on agricultural production,and emphasizing enhancing photosynthetic efficiency to boost crop yields.With intensifying global climate change,traditional agricultural production faces significant challenges regarding climate adaptability.This research employs machine learning to optimize agricultural production processes and offers scientific decision-making support for agriculture.(3)This part introduces the fundamental concepts and principles of machine learning technologies,including Artificial Neural Networks(ANN),Support Vector Machines(SVM),and AdaBoost algorithms.It also analyzes their characteristics and application ranges,providing a theoretical foundation for subsequent experiments.(4)This research employs a systematic machine learning approach to build and apply agricultural yield prediction models.Firstly,it preprocesses raw agricultural data,including filling missing values,detecting and removing outliers,and standardizing and normalizing data to eliminate dimensional differences between features.During the feature engineering phase,it uses Principal Component Analysis(PCA)for feature dimensionality reduction and constructs a comprehensive feature set comprising meteorological factors,soil parameters,and remote sensing indicators.Based on the Python platform,it implements three machine learning algorithms:1)The ANN model adopts a two-layer perceptron architecture with a hidden layer of 16 neurons and a Re LU activation function,trained using the Adam optimizer;2)The SVM model selects an RBF kernel function and optimizes the penalty coefficient and kernel parameters through grid search;3)The AdaBoost model integrates 50 decision tree-based learners and enhances model performance through iterative weighting.(5)This study conducts an in-depth analysis of the machine learning model training results,comparing the performance of different algorithms in predicting above-ground total dry weight,number of grains per plant,and the ratio of grains to straw.The results indicate that the AdaBoost algorithm outperforms others in improving prediction accuracy and model stability,especially in handling complex nonlinear data,where it shows significant advantages over ANN and SVM algorithms.Meanwhile,the ANN algorithm demonstrates strong fitting ability in dealing with large-scale data and multivariable nonlinear relationships,providing better prediction results.In contrast,the SVM algorithm performs less effectively with small data volumes or complex data distributions.After a comprehensive evaluation of the models using metrics such as mean square error and the coefficient of determination,the most suitable machine learning algorithm for this study is identified.Furthermore,an in-depth analysis of the weight of independent variables on dependent variables is conducted,offering valuable insights for subsequent agricultural data analysis.(6)The innovation of this study lies in the application of machine learning algorithms to the field of photosynthetic efficiency and crop yield prediction.Specifically,by integrating remote sensing data with crop growth models,an efficient prediction framework is developed.Through comparative analysis of different algorithms,a machine learning-based crop yield prediction model is proposed,along with optimized strategies for improving photosynthetic efficiency.This research provides scientific evidence for enhancing agricultural productivity and addressing climate change challenges.更多还原