Research And Application Of Chaotic Time Series Based On Neural Network

With the continuous development of national income and people investment consciousness gradually strengthened, keeping healthy development for financial market has become a research focus in the financial sector. Stock as an important part of financial markets, whose prices reflects the health of the financial market to some extent, therefore timely predict stock price movements has become a top priority for the government and investors. The stock price movements has characteristics of nonlinear, especially for some complex system, whose internal information contains the chaos characteristics, that the traditional time series prediction model is unable to fully exert the intrinsic characteristics of the system, seriously affect the prediction effect of the model. Based on chaotic local prediction method, making system internal information from low-dimensional space mapped to high-dimensional space, fully explore and restore the intrinsic characteristics of the system. Since the RBF neural network has the advantages of fast convergence speed and simple structure, and PSO has the great global search ability, using the combination of RBF and PSO to fully explore the evolution regularity of phase points in the phase space, that chaotic local prediction model of RBF neural network based on particle swarm optimization was proposed. The main contents are as follow.Firstly,due to the effect of local prediction is subject to the selection of neighboring points, especially the existence of "pseudo neighboring points" will reduce the prediction accuracy, Therefore, to overcome the deficiency of near point selection method, an improved neighboring points selection method based on Lyapunov exponent was proposed which can lay the foundation to propose a chaotic local prediction model. Because the traditional Euclidean distance can not reflect influence on the prediction for each component of points, using the maximum Lyapunov index as weight coefficient to improve the Euclidean distance and then construct the distance correlation, combining the correlation distance and vector angle cosine as evaluation standard to determine the spatial relationships and the evolution trend between the phase points in phase space. Since there exist correlation between the phase points and their S-step preceding phase points, an improved neighboring point selection method is proposed by tracing the evolution trend of the S steps phase points ahead. Using the chaotic time series generated by x-component of Lorenz equations for test, the results showed that prediction effect of the improved method is improved obviously.Secondly, in view of some stock time series with chaos, since RBF neural network has the advantages of fast convergence speed and simple structure, chaotic global prediction model of RBF neural network was built to predict day’s closing price of Shanghai PuDong development bank stock, result showed that the prediction accuracy is poor, On the basis of chaotic local prediction, an improved stock price prediction model was proposed. At first, reasonably selected neighboring points of center point by using the improved method. Since PSO has the great global search ability, combining PSO and RBF to optimize weights, the basis function center and the standard deviation. Then, the chaotic local prediction model of RBF neural network based on particle swarm optimization was proposed which can fully explore the evolution regularity of phase points in the phase space. Finally using the improved model to predict the same data, Comparison results showed that the prediction accuracy of improved model is higher which illustrates the effectiveness of the improved method.