【作者】 潘文杰； Simon X.Yang； 唐远驹； 姜超英；

【Author】 W.J.Pana,Simon X.Yangb,Y.J.Tanga,C.Y.Janga a:Guizhou Tobacco Research Institute.Guiyang,China 550003.b:Advanced Robotics & Intelligent Systems(ARIS)Lab,School of Engineering,University of Guelph,Canada,N1G 2W1,

【机构】 贵州省烟草科学研究所； 奎尔夫大学工程系机器人及人工智能实验室；

【摘要】 生态环境是影响烤烟烟叶无机元素含量最重要的因素之一。本次研究在统一烤烟栽培条件下(如施肥、田间管理及成熟采烤等),集中开发了一种神经网络模型来探索烟叶锌含量与它所生长的土壤、气候条件之间的关系。试验过程中,以Matlab软件为工具,7个土壤因子(有机质、全氮、全磷、全钾、速效氮、速效钾和有效锌),2个气象因子(日均温度、相对湿度)为输入,烟叶锌含量为输出。结果表明,一个基于正切(隐含层)、线性(输出层)激活函数,9-10-1的BP神经网络模型能较好地对烟叶锌含量进行预测,预测结果与测试数据之间相关系数达0.978,误差标准差为5.479%,平均误差为11.39%,明显优于作为对照的线性回归模型。更多还原

【Abstract】 All of the factors to control the contents of the inorganic elements of flue-cured tobacco leaves,the environment is one of the best important things.In this study,it focused on an artificial neural network(ANN)to know the relationship between Zn content of the flue-cured tobacco leaves and environment(including soil and weather)based on the uniform cultivation(such as fertilizer,field management,mature degree,curing condition and so on).In an ANN model,the input of model included 7 soil factors(organic matter,total nitrogen,total phosphorus,total potassium,available nitrogen,available potassium and available zinc)and 2 weather factors(daily average temperature and relative humidity).Based on the tangent activation function(hidden layer)and the linear activation function(output layer),it was concluded that an ANN including 9 input layer nodes,10 hidden layer nodes and 1 output layer node had great capabilities in estimating the Zn content of the flue-cured tobacco leaves.Between the predicted values based on the ANN and the observed ones,the relative coefficient was 0.978,the standard error of the relative error was only 5.479%,and the average error was 11.39%.It showed the ANN model can forecast the Zn content in the flue-cured tobacco leaves more than linear regression model.Above all,it is feasible to estimate Zn content in the Flue-cured tobacco leaves using the ANN based upon the environment factors.更多还原