Analysis Of Genetic Algorithm And Neural Network For Forecasting Stock Prices

The stock market has characters of high risk and high return, its analysis and forecastingof research has been paid more attention. Whereas, as the stock market call be affected bysuch factors like politics, economy ere, and the internal running rules was complicated, thetraditional technological forecast was not ideal. In recent years, the rapid development ofartificial neural network provides a new technology and methods for modeling and predictionof the stock market.Artificial neural network is a non-linear science, it has the ability of tolerating mistake,self-adaptability,non-linear mapping and overcome the traditional artificial intelligencedefects in information processing, to make it has been successfully applied to the region ofneurologist system,pattern recognizing,intelligence control,combination optimization,prediction and other fields. This article aims to analyze stock market characteristics, grasp thechanges law, make better forecasts, then provide a reliable basis for stock control manager,investors to develop investment strategies and government to plan the development of thestock market.BP network is a feed-forward neural network, it has very strong non-linear handlingability. But there has some weakness such as the astringency slow, easily falling into localsolutions concerning bigger search space and complex function, and the neural networkbeginning construction and connection weights having no way in definition. These all affectANN's generalization ability. The paper introduces momentum method and learning rateself-adaptation adjusted strategy, the structure of network, the number of hidden nodes, thechoice and pretreatment of swatch datum and the determination of preliminary parameters. Inorder to avoid local extreme and promote convergence speed, GA-BP algorithm has beenadopted, which optimizes ANN by Genetic Algorithms. It defines connection weights ofANN by improved Genetic Algorithms and replaces defining connection weightsstochastically. So, this algorithm ensures a better search space to question. Then ANN carries on training, studying and searching the optimal solution or the approximate optimal solution.RBF neural network is a new and effective neural network, it has the best and universalapproximation property, simple structure and fast training speed. Therefore, it has particularadvantages in stock prediction. The choice of quantity and position of hidden layer radialbasis functions is very important and directly affects the goodness of overall networkapproximation capabilities.GRNN is a brace of RBF neural network, the node inputting layer only convey theinputting signature to the latent layer, and the latent layer is made up of the Gaussian functionthat play radiate role, the outputting layer is simple linear function. GRNN using the way togain the relations among dates are different from interpolation and imitating, which ratif iesthe network by the dates that come from computing or sampling, and needn't computeparameters again. Therefore, the character that GRNN has quickly computing ability, stableresult and few parameters selected by people, decides the forecast result avoiding the personalinfluence.This article describes different kinds of forecast methods in the stock market and theneural network structure algorithm, selecting the most representative of the Shanghai Indexand Pi Jiuhua to forecast stock price by using BP Network,GA-BP network,RBF Networkand GRNN network. The numerical experiment result of tests show that predicted value isalmost identical with the actual value, the method of stock prediction using neural network isfeasible and efficient. It has favorable foreground.