Files in this item

FilesDescriptionFormat

application/pdf

application/pdf9971124.pdf (3MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Neural Networks for Setting Target Corn Yields
Author(s):Liu, Jing
Doctoral Committee Chair(s):Goering, Carroll E.
Department / Program:Agricultural Engineering
Discipline:Agricultural Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Artificial Intelligence
Abstract:In this research, artificial neural networks were employed to model corn yield. The overall objective was to test two hypotheses, i.e.: (1) That an artificial neural network (ANN) can be trained to approximate the nonlinear function relating corn yield to the factors that influence yield, and (2) That an ANN trained for one field can be retrained for a different field with a much sparser data set. The specific research objectives were: (1) To train an ANN using corn yield and input factor data from the Morrow Plots, (2) To evaluate the performance of the trained ANN, and (3) To retrain the ANN using data from the Dudley Smith Farm. First, the factors affected corn yields were analyzed. The corn yield was expressed as function of soil factors, weather factors, and management factors. 15 influencing factors were fed into input layer of the neural network. The neural network structure was designed with 15 input factors, one output, corn yield, and one hidden layer with 20 nodes. The network topology and parameters were set by trial and error. The Morrow Plots data was used to train the neural network. The data set was divided to training set and test set. After training with 5,000 epochs, the test set was used to verify this network. The RMS error was about 20%. Then, the network performance was evaluated in four aspects: for prediction yield trend with each influencing factor, the network gave realistic prediction trend with each factor; for interaction between nitrogen fertilizer and late July rainfall, the network captured the interaction between the two factors; for the influencing factors combination searching to get maximized yield using genetic algorithm, the network predicted a yield 75% larger than the maximum observed yield in the training set; for the influencing factors sensitivity analyzing, the calculated yields were most sensitive to the corn growing season rainfall, especially late July rain, then nitrogen fertilizer. The network was retrained with the Dudley Smith Farm data. The network topology was changed accordingly. The verification of the retraining model, the RMS error, was about 17%. Several training and retraining strategies were discussed.
Issue Date:2000
Type:Text
Language:English
Description:85 p.
Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2000.
URI:http://hdl.handle.net/2142/86103
Other Identifier(s):(MiAaPQ)AAI9971124
Date Available in IDEALS:2015-09-28
Date Deposited:2000


This item appears in the following Collection(s)

Item Statistics