Prediction of wheat stripe rust using neural network

By comprehensive analysis on wheat stripe rust disease and meteorological data from 1984 to 2008 in Pingliang, the main factors were selected, and based on it three modified BP prediction models were built to realize the prediction of wheat stripe rust bacteria amount during summer, the occurrence degree in autumn and in the next spring. The node numbers of three networks are 6-12-1, 3-10-1 and 6-10-1, and the training functions of hidden and output layers are both traingdx. The simulation results show that for training samples of three BP networks the prediction results are completely matching with real levels. The mean square deviations between the prediction results and real values of test samples are all within the range of 0.1~0.4 level. It proves that modified BP prediction model has better accuracy than stepwise regression method and can meet production needs.