摘要
针对温室番茄无法按需灌溉问题,提出了随机森林(Random forest,RF)结合门控循环单元(Gated recurrent unit,GRU)神经网络的温室番茄结果前期蒸腾量预测方法,并开发了一套基于番茄蒸腾量的智慧灌溉系统。基于物联网实时获取数据,采用RF算法对影响温室番茄蒸腾量的变量进行特征重要性排序,选取作物相对叶面积指数、温室内空气温度、相对湿度、光照强度、光合有效辐射、基质含水率和基质温度作为模型的输入变量,在此基础上,构建了基于GRU的番茄蒸腾量预测模型。试验结果表明:RF-GRU在番茄蒸腾量预测中具有准确的预测效果,决定系数(R^(2))、均方根误差(RMSE)、平均绝对误差(MAE)分别为0.9490、10.96 g和5.80 g。同时,基于此模型进行指导灌溉相比于定时灌溉,在番茄长势基本相同的情况下,灌溉量降低了20%,可为实际生产提供参考。
Taking greenhouse tomatoes as the research object,a forecasting method of transpiration of greenhouse tomatoes was proposed based on the real-time data of the Internet of things and random forest(RF)combined with gated recurrent unit(GRU)neural network.Firstly,the main factors affecting transpiration change collected by the sensor were preprocessed and RF was used to order the characteristic importance of the variables affecting the transpiration of tomato in greenhouse.Crop phenotypic parameters,including relative leaf area index,ecological parameters in greenhouse and cultivation environment parameters,including air temperature,relative humidity,light intensity,photosynthetically active radiation,substrate moisture content and substrate temperature were chosen as the input variables of the model.On this basis,a prediction model based on GRU was established to predict the transpiration of tomato.Finally,this model was compared with other models.At the same time,based on this model,a set of intelligent irrigation equipment was developed,which took the substrate water as the irrigation starting point and the predicted transpiration as the irrigation amount.The experimental results fully showed that the RFGRU model had accurate prediction effect in tomato transpiration prediction and showed good feature learning ability in agricultural big data mining.The determination coefficient(R^(2)),root mean square error(RMSE),mean absolute error(MAE)were 0.9490,10.96 g and 5.80 g,respectively.Compared with RF-LSTM and RF-RNN methods,the R^(2) was increased by 1.46%and 3.78%,the root mean square error was decreased by 1.38 g and 3.24 g,and the mean absolute error was decreased by 1.77 g and 0.14 g,respectively.At the same time,compared with regular irrigation,the intelligent irrigation system designed based on this model reduced the irrigation amount by 20%when the tomato growth was basically the same.This study could provide a reference for the research of greenhouse crop water requirements and it can be applied to water-saving greenhouse irrigation.
作者
李莉
李伟
耿磊
李文军
孙泉
SIGRIMIS N A
LI Li;LI Wei;GENG Lei;LI Wenjun;SUN Quan;SIGRIMIS N A(Key Laboratory of Agricultural Information Acquisition Technology,Ministry of Agriculture and Rural Affairs,China Agricultural University,Beijing 100083,China;Key Laboratory of Smart Agriculture System Integration,Ministry of Education,China Agricultural University,Beijing 100083,China;Department of Agricultural Engineering,Agricultural University of Athens,Athens 11855,Greece)
出处
《农业机械学报》
EI
CAS
CSCD
北大核心
2022年第3期368-376,共9页
Transactions of the Chinese Society for Agricultural Machinery
基金
国家重点研发计划项目(2019YFD1001903)
中央高校基本科研业务费专项资金项目(2021TC031)。
关键词
温室番茄
智慧灌溉
蒸腾量预测
随机森林
门控循环单元
greenhouse tomato
intelligent irrigation
transpiration prediction
random forest
gated recurrent unit
