Research On Stock Price Prediction Model And System Based On Neural Network

The stock market plays an important role in financial investment. With the social and economic development, there is growing concern about financial management and investment, Stock market has gradually become an important investment channels that is equally with bank and insurance, is one of the leading personal and family finance. However, because of the inaccurate information, the different cognitive of investors, the complexity of a variety of analysis techniques and the randomness of stock price changes, the investment can not achieve the desired results, accordingly resulting in the loss of investor capital, that is so-called stock market risk. At home and abroad, Many scholars Dedicated to the research on trend of the stock market, the corresponding prediction models are established, have many efforts to avoid large fluctuations in the stock market, reduce investment risk and maintain economic prosperity and stability.Artificial neural network is a fast-growing emerging cross disciplinary, widely used in signal processing, adaptive control and financial forecasting and other fields. This thesis Studied how to better integrate the theory and technology of artificial neural network to predict stock prices, established the appropriate network model, analysis and evaluation of its efficiency through the forecast of specific examples of the stock, then the better network model is given. Specific work mainly are in the following areas:(1) The mainstream methods of stock analysis at the present stage are summed up, some applications of neural networks in forecasting stock price are discussed, and this study ideas are put forward.(2) The basic idea, standard methods and processes of Linear neural network, BP neural network, RBF neural network to forecast stock price are given, and taking two stocks Haihong Holding and Zhongguancun as examples, the advantages and disadvantages of the three networks in the stock price forecasting are analyzed and compared. The results show that, BP neural network is suitable to predict the stock price change case for the day and the day before, RBF neural network is suitable to predict future stock prices, and Linear neural network is not suitable for stock price prediction. A new idea that BP network and RBF network's good features are fused together is proposed.(3) A BP and RBF mixed neural network model BP-RBF is proposed. First, with support vector machines, the optimum initial RBF network structure and parameters are determined, Further, using BP algorithm, the final RBF network parameters are optimized. Based on the BP-RBF model to predict the two stocks Haihong Holding and Zhongguancun, the desired effect is achieved.(4) The stock price prediction system based on neural network model is developed. Using the COM Builder tool in the Matlab system, a neural network forecasting process is made into COM components, and called in VC++. In this way, Linear, BP, RBF and BP-RBF neural network, these 4 different stock price prediction are implemented.The results show that, With BP neural network and RBF neural network compared, The BP-RBF neural network model proposed in this thesis has higher accuracy rate, lower forecast errors for stock trend prediction, and can achieve better results. The stock price prediction system based on that can help the majority of stock market investors to avoid stock market risk effectively and get a better return on investment.