Research On The Pest Prediction With BP Neural Network

Insect pest was frequently occurred in China, and caused severely damage to crops over years. The forecasting of pest occurrence was the key fundamental work in preventing disasters and reducing damages, and played an important role in crop safety production. Only forecast the pest occurrence accurately, can it provide effective reference to agriculture policy maker, early warning timely, and control the loss into economical permissible range.Insect pest data was influenced by external factors of global climate, weather, crop growth, the amount of chemical fertilizer, and internal remarkable dynamic time series, therefore it was difficult to forecast accurately. This thesis tried to solve this problem from the following aspects.(1)A new forecasting method of multi-dimensional time series was proposed based on IV (Impact Value) and REMCC-BPNN (Back Propagation Neural Network based on the Minimum Correlation Coefficient of the fitting Relative Error), namely IV-REMCC-BPNN. The steps were as follows. At first, normalized train set K was used to build the net by REMCC-BPNN, then for each independent variable in K, plus or minus ten percentage of original values to get two new train sets K1and K2. A pair of predicted values (YK1and YK2) was obtained by this net, and pairwise t test was conducted. Proper significance level was designed to screen independent variable automatically. At last, independent prediction was performed based on the selected variables and REMCC-BPNN.(2)Forecasting of pest occurrence based on meteorological factors. Combining IV-REMCC-BPNN with traditional meteorological factors such as temperature and rainfall, independent prediction was performed on the basis of historical statistical materials. The independent prediction results on rice gall midge and corn borer showed that, with robust performance and high accuracy, IV-REMCC-BPNN was better than commonly used comparative models including BPNN, REMCC-BPNN, SVR and SVR-CAR, which indicated this model held strong generalization ability.(3)Forecasting of pest occurrence based on macula. Sun was the supplier of light and heat on the earth, whose minor change would have an influence upon the earth. Dozens of studies at home and abroad showed that climate change, prevalence of disease and pest occurrence were objectively related to solar activities. Macula was the fiercest solar activity; therefore this paper selected the monthly historical data of macula as a factor, and applied it to the independent forecasting of pest occurrence and corn borer. The satisfactory forecasting result provided a new idea for the selection of influence factors about insect pests forecasting, and effective references for those who studied insect pests forecasting and macula data.