Optimization Research Of BP Neural Network And Genetic Algorithm Based On Numerical Calculation Method

The artificial neural networks and genetic algorithm applied the biological principle to the bionics theory achievement of computer science. Because they have extremely strong ability to solve problem, it have drawn numerous scholars' interest and participation in recent years and has already become one of academia's interdisciplinary hot special topics.In the practical application of the artificial neural network, about 90% of the artificial neural network models adopt BP network or its change form , it is a key part of the feedforward network too. BP network applies to the approximation of function extensively, pattern-recognition / classification, the data compressed and so on.. It has already become one of the important fields which the artificial intelligence has studied now. However, because BP algorithm is a gradient dropping method of searching for, there are inherent deficiencies unavoidably, such as converging slowly, apt to fall into extremely some snack of the error function, as to the space of larger searching, many peak values and little function can search for reaching the overall situation snack very much effectively.The genetic algorithm as a kind of intelligent the overall searching for algorithms has optimized in number value since the eighties, system control, structural optimization design and applications in a great deal of fields show their characterized glamour, expose a lot of insufficient and defects at the same time .For example, totally rely on probability to operate at random, though extremely small can be avoided, It is sought the restriction of the excellent condition. The ones only can generally be received in the overall range approximatly and optimumly, it is very difficult to be solved optimumly; Adopting binary scale code to parameter, dispersing to take continuous space artificially result in the contradiction between calculating precision and string length and operation amount; So adopting and optimizing technology at random should spend a large amount of time; In branching and the mutant evolution course randomness is relatively strong, causing the efficiency of searching for to be low, embodied in evolving .changing and taking the place of course will appear subgeneration optimum individual lower than parent optimum individual. Though genetic algorithm have strongly overall situation searching for ability, its part searching for ability to be weaker and ( easier to appear the early-maturing phenomenon of disappearing). Groundwork of this paper:(1) Analyze and research the defect of BP neural network. With the shortcoming to the slow convergent speed of BP neural network, we carry on mathematics analysis to single polarity Sigmoid...