The Optimization Study On BP Algorithm Based On Psoand Ga

Artificial neural network is constructed inspired by biological network, owning thefollowing features: the distributed representation of information, the overall importanceof computation, adaptive learning and so on. Different network model corresponds todifferent learning algorithm in the artificial neural network. BP algorithm is the one ofthe learning method that used most infrequently in neural network. It updates theconnection weights and node threshold based on the Back Propagation transitivity ofnetwork error, so that it can make the error between the actual output and target outputof the network satisfy the precision stipulated in advance, achieving the effect ofnetwork training. Though the BP algorithm own the obvious advantages, it often showssome problems in practical application, such as easy to fall into local minimum,generalization being weak, convergence speed being slow. Specific to the problemsexisting in BP algorithm, corresponding improved method has been proposed. The maincontent studied are as follows:(1) This paper briefly introduces the research background and significance of theBP neural network algorithm and the research status.(2) This paper study on the math theory, learning rules, features belonging toartificial neural network. The network structure, fundamental theory, the faults andcorresponding improved measure of BP network also has been explained.(3) Improve based on the BP network optimized by traditional particle swarmalgorithm, proposing the improved BP algorithm based on artificial fish swarmalgorithm and particle swarm algorithm. The new optimization algorithm has introducedthe swarm behavior belonging to artificial fish swarm algorithm, as well as adaptiveweight and learning factor. The improved algorithm is used to predict the investmentrisk and the simulation shows that the improved algorithm can avoid the prematureconvergence well. It also can search the global optimal solution faster and raise theprediction accuracy and the rate of convergence.(4) Aiming at the problems existing in the process the standard genetic algorithmoptimize BP algorithm, and proposed a new algorithm. The new algorithm has beenused to predict the earthquake magnitude. The MATLAB simulation showed that the new algorithm can avoid falling into local extremum effectively and it can raise the rateof convergence of BP network and the characteristic of algorithm.The main work has been summarized in the end of this thesis, and it put forwardthe possible directions of the future study.