Research On The BP-Neural Network Based On Parameter-adjusted Automatically

Among the algorithms which being used to train the forward neural network, BP algorithm is the one, which get the most common used and achieve the best practical effect. Normal BP algorithm can be used in many fields and resolved many practical problems, however, normal BP algorithm has many limitations such as it's easy to fall into the local minimum in the course of convergence, its' convergent speed is very slow, the method which set the structural parameter and the operational parameter hasn't be widely accepted, and so on. Although it has been brought forward many improved algorithm, none hasn't being get wide recognition, so we have lots of work to do. This thesis brings forward a kind of improved BP algorithm. It's different from others algorithm that it's amplified at two aspects of normal BP algorithm. On one hand, it improved the node operational function to accelerate the speed of convergence, on the other hand, added the momentum item. According to the change of the error, let the parameters such as adjustion gene of the node operational function, study rate and inertial gene to be the function of error and adjust them automatically. It accelerates the speed of convergence and avoids falling into the local minimum. The simulated conclusion indicates that the improved algorithm achieved the anticipative purpose. Agriculture is a large complicated system. Among the factor which restrict the growth of the plants, the plant diseases and insect pests is a major one, it can make large economical losing if it can not be prevented and cured in time. Prevention and cure in time needs predict correctly. The artificial neural network can resolve the problems in the argriculral system such as connection complex, boundary faintness and strong uncertainty which cant not be strict described by the rule or mathematical model. Thus, in the last of the thesis, we construct a artificial neural network model based on the improved BP algorithm to predict the plant diseases and insect pests of cotton Heliothis armigera. The simulation used MATLAB tothe artifical neural network indicated that the prediction is correct and (completely.