基于PCA-BPNN的温室番茄果实直径预测模型  

Prediction Model of Tomato Fruit Diameter in Greenhouses Based on PCA-BPNN

作　　者：韩坤林 王钊英[2] 杨会民[3] 陈毅飞 蒋永新[3] 张佳喜 HAN Kunlin;WANG Zhaoying;YANG Huimin;CHEN Yifei;JIANG Yongxin;ZHANG Jiaxi(College of Mechanical and Electrical Engineering,Xinjiang Agricultural University,Urumqi 830052, China;Xinjiang Academy of Agricultural Sciences, Urumqi 830091, China;Research Institute of Agricultural Mechanization,Xinjiang Academy of Agricultural Sciences, Urumqi 830091, China)

机构地区：[1]新疆农业大学机电工程学院,乌鲁木齐830052 [2]新疆农业科学院,乌鲁木齐8300091 [3]新疆农业科学院农业机械化研究所,乌鲁木齐830091

出　　处：《新疆农业科学》2022年第2期485-492,共8页Xinjiang Agricultural Sciences

基　　金：国家重点研发计划“蔬菜智能化精细生产技术与装备研发”(2017YFD0701301-03);新疆农业科学院重点项目前期预研专项“基于多源信息及云平台的设施智能化管控装备技术研究”(xjzdy-002)。

摘　　要：【目的】研究温室番茄果实直径变化量的动态预测模型,为番茄所需水肥规律提供数据支持。【方法】选择番茄果实横径为研究对象,以5株番茄果实膨大期的数据建立模型,采用主成分分析法对植物生理生态信息和环境信息进行分析,提取主要成分,以主成分为自变量,输出变量为因变量,建立一个包含空气温度、空气湿度、土壤含水率、叶片温度及果实横径的BP神经网络回归动态预测模型,并以3株番茄果实膨大期内所测的数据作为测试数据进行预测,比较预测值和实测值。【结果】第1株番茄预测值与实测值的决定系数为(R^(2))0.964,均方根误差(RMSE)为0.238,第2株番茄预测值与实测值的决定系数(R^(2))为0.960,均方根误差(RMSE)为0.051,第3株番茄预测值与实测值的决定系数(R^(2))为0.951,均方根误差(RMSE)为0.047。【结论】该模型可以预测温室短时内番茄果实直径变化量,可以用于新疆连栋温室内的秋季番茄果实直径变化预测,可根据预测量与实测量之间差值对水肥实行微调。【Objective】To study the dynamic prediction model of tomato fruit diameter variation in greenhouse,which can provide certain decision support for the law of water and fertilizer required by tomatoes.【Methods】Based on the fruit diameter as the research object,with five of tomato fruit enlargement period data as a model,using principal component analysis(PCA)to perform the plant physiological ecology and environment information analysis,extract the main ingredients,again with the principal components as independent variables,output variables as the dependent variable,to establish BP neural network regression dynamic prediction model including air temperature,air humidity,soil moisture content,leaf temperature and fruit diameter.In addition,the data measured in the fruit expansion period of 3 tomato plants were used as the test data to predict and compare the predicted and measured values.【Results】The first decision coefficient for tomato plant predicted and the measured values(R^(2))was 0.964,and the root mean square error(RMSE)was 0.238;The second tomato plant decision coefficient of the predicted values and the measured values(R^(2))was 0.960,and the root mean square error(RMSE)was 0.051;The first three decision coefficient of tomato predicted and the measured values(R^(2))was 0.951,and the root mean square error(RMSE)was 0.047.【Conclusion】The model can predict the change of tomato fruit diameter in the greenhouse in a short time,and it can be used to predict the change of tomato fruit diameter in the autumn in the greenhouse in Xinjiang.The water and fertilizer can be fine-tuned according to the difference between the predicted value and the actual measurement.

关 键 词：主成分分析 回归模型 神经网络 温室番茄 

分 类 号：S164.2[农业科学—农业气象学] S12[农业科学—农业基础科学]