Research On Crop Pest Forecasting Based On Adaptive Probabilistic Neural Network

The agricultural economy suffers to giant loss due to serious damage of the main pests to crops perennial. According to the laws of pest occurrence and development, crop phenology and weather forecasting, comprehensive analysis is maked. While its future occurrence period and quantity and harmful levels are estimated. At the same time, the future trends of the pests are predicted. This work is called crop pest forecasting. The works to do the pests forecasting are a necessary premise to prevent it comprehensively. Only when the occurrence of pests is forecasted timely and accurately, comprehensive prevention control plan is drawn out, correctly. At the same time, necessary measures are timely adopted to reduce the number of the pest occurrence to ensure high yield of crops, economically and effectively. Hence, investigation on crop pest forecast technology has very important practical significance.The relative investigation on crop pest forecasting begins earlier. The traditional forecast methods include observation, statistics, and mathematic-ecologic-model method. The traditional methods have some limitation, for instance, more obvious human factors, low practice fitting rate, unstable prediction results, and model establishment more difficultly. In recent years, the new prediction methods were proposed such as BP neural network model, expert system (ES), and fuzzy neural network (FNN) model, and etc. Although these new prediction methods have their each advantage, but simultaneously, some defects are existed, say, BP neural network has slow training speed and low forecast accuracy, and falls into local minimum, easily; the knowledge acquisition in ES is also more difficult; FNN has more human interventions, and automatic generation and adjustment to the membership functions and fuzzy rules is more complicated. With the rapid development of agricultural technology, the existing methods can not meet the requirements of pest forecasting system very well, and so prediction technology of crop pest is required to be researched more deeply.Probabilistic neural network (PNN) possesses a good classification performance with the characteristics of global optimization. PNN calculates quickly, and guarantee the convergence to Bayesian Classifier, and it allows increasing or decreasing the training data not to train repeatedly. In view of the nonlinear, real-time and uncertain characteristics of the crop pest and the particular advantages on complex nonlinear problems solution of PNN, PNN is applied to forecast the crop pests. Moreover, the measuring space of all the dimensions of PNN has the same smoothing parameter values, and can not reflect the effect of input variables on the classification results really, and easily leads to inaccuracy and low computational efficiency usually. While adaptive probabilistic neural network(APNN) makes the improvements to traditional PNN, which optimizes different smoothing parameters of different measure space dimensions using repeat iterative method. APNN increases network training time, but generates higher classification accuracy. Hence, APNN is applied to crop pest forecasting in the thesis. The main works are as follows.By analyzing the variation rule of the crop pest, the occurrence of the crop pest is considered to be non-linear, complex and uncertain. Based on it, PNN is applied to implement crop pest forecasting, including PNN model establishment. Although the result shows that crop pest forecasting based on PNN is realistic, its prediction accuracy is low.In order to improve the prediction accuracy of pest forecasting further, APNN is applied to implement crop pest forecasting. To establish the prediction model of crop pests of APNN, and compare with prediction results of BP neural network model and PNN model, the result shows that APNN is feasible to crop pest forecasting, and higher prediction accuracy.