【摘要】：股票市場是現代金融市場中的重要組成部分。它對國家經濟的發(fā)展,股份制企業(yè)和股票投資者都具有無法替代的作用。同時,股票市場的波動對經濟建設也有不小的副作用。所以,對股票市場走勢的分析和預測具有重要的意義。由于股票價格的影響因素非常多而且非常復雜,研究者很難對股票市場進行精確的預測。時間序列方法對股價的預測是一個較好的選擇。然而股價具有非線性,高噪音和異方差的特點,這樣傳統的序列模型并不能很好地分析與預測股價。本文的主要工作是建立基于混沌優(yōu)化方法的多尺度小波核v-支持向量機的回歸預測模型,達到更準確預測股價的目的。與單尺度小波核v-支持向量機,小波神經網絡和徑向基函數ε-支持向量機等模型相比,這種模型可以更準確的預測股票價格的未來走勢。股票市場的參與者借助這種模型可以在降低投資風險的同時獲得更高的投資收益。 本文首先總結了股票價格時間序列預測方法的研究進展,特別是介紹了人工智能算法與小波理論的混合模型在股票價格時間序列預測中的應用；然后闡述了支持向量機的理論基礎和小波理論的基礎知識。根據支持向量機核函數構造的方法,本文著重分析了單尺度小波核和多尺度小波核作為v-支持向量機的核函數的合理性。接著,本文分析了以多尺度小波核作為核函數的v-支持向量機模型處理具有非線性,高噪音特點的時間序列的優(yōu)勢。多尺度小波核v-支持向量機模型的參數具有個數較多,每個參數的取值范圍不一的特點,使得支持向量機的常用參數選擇方法--交叉驗證方法不適合這一模型。針對這一問題,本文提出了混沌優(yōu)化方法作為多尺度小波核v-支持向量機模型的參數選擇方法。 本文具體的研究方法和成果是：首先利用上海證券交易所發(fā)布的建筑指數研究了混沌優(yōu)化方法對多尺度小波核v-支持向量機的優(yōu)化效果。本文通過多次試驗比較了粒子群優(yōu)化方法,混沌優(yōu)化方法和混沌粒子群優(yōu)化方法對模型的優(yōu)化效果,證明了混沌優(yōu)化方法對模型的優(yōu)化效果的有效性和穩(wěn)定性�；煦缌Ｗ尤簝�(yōu)化方法的優(yōu)化效果曲線顯示了粒子群優(yōu)化方法對從混沌優(yōu)化方法中得到的優(yōu)化粒子沒有進一步地優(yōu)化效果。接著,本文比較了多尺度小波核v-支持向量機,單尺度小波核v-支持向量機,小波神經網絡和徑向基核函數ε-支持向量機對上證指數的預測效果。在這次試驗中,本文將上證指數分為了牛市期,熊市期和震蕩期三個階段。在每一個階段中,基于混沌優(yōu)化方法的多尺度小波核v-支持向量機取得了比另外三個模型更好的預測效果。
[Abstract]:Stock market is an important part of modern financial market. It plays an irreplaceable role in the development of national economy, stock-holding enterprises and stock investors. At the same time, the volatility of the stock market has no small side effects on economic construction. Therefore, it is of great significance to analyze and forecast the trend of stock market. Because there are many and complicated factors affecting stock price, it is difficult for researchers to predict the stock market accurately. Time series method is a good choice for stock price prediction. However, the stock price has the characteristics of nonlinear, high noise and heteroscedasticity, so the traditional sequential model can not well analyze and predict the stock price. The main work of this paper is to establish the regression prediction model of multi-scale wavelet kernel v-support vector machine based on chaos optimization method, so as to predict the stock price more accurately. Compared with single scale wavelet kernel v- support vector machine, wavelet neural network and radial basis function 蔚-support vector machine, this model can predict the future trend of stock price more accurately. With the help of this model, participants in the stock market can achieve higher investment returns while reducing investment risk. This paper first summarizes the research progress of stock price time series prediction, especially introduces the application of hybrid model of artificial intelligence algorithm and wavelet theory in stock price time series prediction. Then, the theoretical basis of support vector machine and the basic knowledge of wavelet theory are expounded. According to the method of constructing kernel function of support vector machine, the rationality of single-scale wavelet kernel and multi-scale wavelet kernel as kernel function of v-support vector machine is analyzed in this paper. Then, this paper analyzes the advantage of the v-support vector machine model which uses multi-scale wavelet kernel as kernel function to deal with nonlinear and high-noise time series. The multi-scale wavelet kernel v-support vector machine model has a large number of parameters and a different range of values for each parameter, which makes the commonly used parameter selection method of support vector machine, the cross-validation method, not suitable for this model. To solve this problem, a chaotic optimization method is proposed as a parameter selection method for multi-scale wavelet kernel v-support vector machine model. The specific research methods and results are as follows: firstly, the optimization effect of chaos optimization method on multi-scale wavelet kernel v-support vector machine is studied by using the building index published by Shanghai Stock Exchange. In this paper, the effects of particle swarm optimization method, chaos optimization method and chaotic particle swarm optimization method on model optimization are compared, and the effectiveness and stability of chaotic optimization method for model optimization are proved. The optimization effect curve of chaotic particle swarm optimization method shows that the particle swarm optimization method has no further optimization effect on the optimization particles obtained from chaos optimization method. Then, this paper compares the prediction effect of multi-scale wavelet kernel v-support vector machine, single-scale wavelet kernel v-support vector machine, wavelet neural network and radial basis function 蔚 -support vector machine on Shanghai stock index. In this experiment, the index is divided into three stages: bull period, bear period and shock period. In each stage, the multi-scale wavelet kernel v-SVM based on chaotic optimization method achieves better prediction results than the other three models.
【學位授予單位】：蘭州大學
【學位級別】：碩士
【學位授予年份】：2012
【分類號】：O212.1;O211.61;F830.91

【參考文獻】

相關期刊論文 前5條

1 任世錦;吳鐵軍;;基于徑向基小波核的多尺度小波支持向量機[J];電路與系統學報;2008年04期

2 王彥峰;高風;;基于支持向量機的股市預測[J];計算機仿真;2006年11期

3 崔艷;程躍華;;小波支持向量機在交通流量預測中的應用[J];計算機仿真;2011年07期

4 李元誠;方廷健;;小波支持向量機[J];模式識別與人工智能;2004年02期

5 施燕杰;基于支持向量機(SVM)的股市預測方法[J];統計與決策;2005年04期

相關碩士學位論文 前4條

1 郝博乾;基于時間序列分析的股票預測模型研究[D];電子科技大學;2011年

2 張海珍;小波神經網絡在股價預測中的應用[D];西安科技大學;2008年

3 夏冬;小波神經網絡在我國股本權證市場中的應用[D];華東師范大學;2009年

4 張清華;小波神經網絡參數優(yōu)化及其應用[D];東北農業(yè)大學;2009年

，

本文編號：2324943