本文選題：支持向量機　切入點：粒子群　出處：《廣東工業(yè)大學》2012年碩士論文

【摘要】：股票在市場經(jīng)濟中占有重要地位,發(fā)行股票的公司可以通過從社會上融資來擴大公司規(guī)模,個體也可以通過控股的方式來影響公司的經(jīng)營,可以說股票對推動市場經(jīng)濟發(fā)展起到了重要的作用。因此,對股市進行分析和預測,從小的方面看,有利于市場中的個體獲益,從大的方面看,有利于決策者對宏觀經(jīng)濟進行調(diào)整,保持國民經(jīng)濟的平穩(wěn)發(fā)展。 隨著統(tǒng)計機器學習領域的發(fā)展,各種智能算法也不斷涌現(xiàn)。從股票的特點來看,它在短期投資上會產(chǎn)生很大的不確定性,但是在長期趨勢上則符合統(tǒng)計學規(guī)律。因此,在有限樣本的情況下,通過機器學習算法來預測股票,是股票預測研究中的一個重要發(fā)展方向。 本文在研究了各項股票預測技術的基礎上,將統(tǒng)計機器學習的思想作為算法的基礎,首先分析使用統(tǒng)計機器學習原理進行股票預測的可行性,然后提出了一種新的股票預測方法,這種方法以支持向量機分類為核心,首先使用K均值聚類,對股票的歷史數(shù)據(jù)從形態(tài)上進行分類,然后對每一類的歷史數(shù)據(jù),提取股票預測中的經(jīng)典時態(tài)指標作為特征,用支持向量機進行訓練,在訓練過程中,使用優(yōu)化的粒子群算法對支持向量機的關鍵參數(shù)進行調(diào)整,從而得到分類更準確的支持向量機模型。在預測時,首先使用最近鄰分類將待預測樣本歸到某一個聚類中,再使用該類相關的支持向量機進行預測。通過這種多級預測的算法,提高了分類的準確率。 為了真實全面地評估該算法的有效性,本文通過實際的上證大盤股票的歷史數(shù)據(jù)作為訓練集和預測集進行分析,在實驗的內(nèi)容上,也包括了本算法與其他預測算法的結果比較。實驗結果表明,本算法在預測精度和適應性上都比其他算法有了明顯進步,預測率得到了明顯提升。
[Abstract]:Stocks play an important role in the market economy. Companies that issue stocks can expand the size of the company through financing from the society, and individuals can also influence the operation of the company through holding shares.It can be said that the stock market economy to promote the development of an important role.Therefore, the analysis and prediction of the stock market is beneficial to the individual benefit in the market from the small aspect, and to the policy makers to adjust the macro economy and maintain the steady development of the national economy.With the development of statistical machine learning, a variety of intelligent algorithms are emerging.According to the characteristics of stock, it will produce a lot of uncertainty in short-term investment, but accord with the statistical law in long-term trend.Therefore, in the case of limited samples, it is an important development direction in stock forecasting research to predict stocks by machine learning algorithm.On the basis of studying the stock forecasting technology, this paper takes the idea of statistical machine learning as the basis of the algorithm, and analyzes the feasibility of using the statistical machine learning principle to predict the stock.Then, a new stock forecasting method is proposed, which takes support vector machine classification as the core. Firstly, K-means clustering is used to classify the historical data of stocks, and then the historical data of each class are classified.The classical temporal index of stock prediction is extracted as the feature, and the support vector machine is used to train. In the process of training, the key parameters of the support vector machine are adjusted by using the optimized particle swarm optimization algorithm.Thus, a more accurate support vector machine model is obtained.In prediction, the nearest neighbor classification is first used to classify the samples to a certain cluster, and then the support vector machine associated with this class is used to predict the prediction.The accuracy of classification is improved by using this multilevel prediction algorithm.In order to evaluate the validity of the algorithm, this paper analyzes it by using the actual historical data of Shanghai stock market as the training set and prediction set. In the content of the experiment, it also includes the comparison of the results of the algorithm and other prediction algorithms.The experimental results show that the prediction accuracy and adaptability of this algorithm are better than those of other algorithms, and the prediction rate is improved obviously.
【學位授予單位】：廣東工業(yè)大學
【學位級別】：碩士
【學位授予年份】：2012
【分類號】：TP18;F832.51

【參考文獻】

相關期刊論文 前10條

1 張玉川;張作泉;;支持向量機在股票價格預測中的應用[J];北京交通大學學報;2007年06期

2 莊新田;李冰;;對有效市場的挑戰(zhàn)——資本市場分形與混沌的研究綜述[J];東北大學學報(社會科學版);2008年02期

3 董曉莉;顧成奎;王正歐;;基于形態(tài)的時間序列相似性度量研究[J];電子與信息學報;2007年05期

4 竇全勝;周春光;徐中宇;潘冠宇;;動態(tài)優(yōu)化環(huán)境下的群核進化粒子群優(yōu)化方法[J];計算機研究與發(fā)展;2006年01期

5 陳成,王永縣;股市技術分析理論研究發(fā)展綜述[J];經(jīng)濟師;2005年05期

6 呂振亞;劉亞琴;嚴忠;;股票市場發(fā)展與經(jīng)濟增長關系研究文獻綜述[J];經(jīng)濟師;2006年05期

7 周廣旭;一種新的時間序列分析算法及其在股票預測中的應用[J];計算機應用;2005年09期

8 錢彭飛;章兢;謝燕江;;一種新的混合粒子群優(yōu)化算法[J];科技導報;2010年22期

9 楊穌;史耀媛;宋恒;;基于支持向量機的股市時間序列預測算法[J];科學技術與工程;2008年02期

10 肇瑩;劉紅星;王仲宇;王元慶;;最近鄰搜索用于分類問題的一種改進[J];南京大學學報(自然科學版);2009年04期

，

本文編號：1730331