标题：Artificial Neural Network Modelling for the Correlation of Nitrogen Level, Plant Density and Variety with Seed Yield and Six Yield Components in Soybean
期刊名称：Canadian Journal on Artificail Intelligence, Machin Learning and Pattern Recognition
出版年度：2011
卷号：2
期号：1
页码：17-27
出版社：AM Publishers Corporation Canada
摘要：The objective of this study was to develop an Artificial Neural Network (ANN) model to predict plant height, number of pod/plant, number of seed/pod, total dry weight, 1000-seed weight, yield, and harvest Index as a function of nitrogen level, plant density, variety, and block # in soybean. A total of 135 patterns, each had 11 components (x1, x2, x3, x4, y1, y2, y3,y4,y5,y6,y7), were used for training and testing the ANN. Four of the components were the input variables, whereas the last seven components were the output variables. The dataset was collected from the farm tests. The results show that feed-forward ANN trained by backpropagation algorithm is able to properly examine the relationship between the input and output parameters. Compared with 3-layer ANN models, 4-layer models provide better training and testing performance. Model in good performance is produced by the 4-28-9-7 structure with hyperbolic tangent transfer function. This model produces the smallest RMSE with 0.011 and 0.036 in training and testing, respectively. ANN model is able to predict the soybean seed yields and 6 yield components with mean RMSE, T value and R2 of 0.036, 0.91 and 0.93, respectively. The results confirm that a properly trained ANN model can be used to produce more than one output simultaneously, unlike traditional models where one regression is required for each output. Adjusting ANN parameters such as learning rate, momentum, and number of hidden nodes/layer affect the accuracy of crop yield predictions.
