【摘要】：隨著我國經(jīng)濟(jì)快速增長和股票市場的不斷擴(kuò)大，股票市場產(chǎn)生了大量有價(jià)值的數(shù)據(jù)信息，這些數(shù)據(jù)成為投資者進(jìn)行股票投資的重要分析主體。同時(shí)，股票價(jià)格的預(yù)測也成為投資者和相關(guān)學(xué)者的一個(gè)重要研究對象。日益增長的數(shù)據(jù)不僅難以處理，更給股票價(jià)格的預(yù)測者們帶來了無從選擇的難題。BP神經(jīng)網(wǎng)絡(luò)作為大數(shù)據(jù)預(yù)測方面的經(jīng)典算法備受投資者和研究人員的青睞，但是，BP算法本身固有的一些缺點(diǎn)也制約著其預(yù)測的效率和效果，在股票價(jià)格短期預(yù)測方面依然存在著預(yù)測精度方面的缺陷。 本文在深入分析股票價(jià)格短期預(yù)測面臨的問題和比較多種股票價(jià)格預(yù)測方法的基礎(chǔ)上，探討B(tài)P神經(jīng)網(wǎng)絡(luò)、主成分分析法和遺傳算法對股票價(jià)格進(jìn)行短期預(yù)測的可行性。BP神經(jīng)網(wǎng)絡(luò)能夠利用對過往股票市場數(shù)據(jù)的學(xué)習(xí)，找出股票市場發(fā)展變化的內(nèi)在規(guī)律，從而實(shí)現(xiàn)對未來一段時(shí)間內(nèi)股票價(jià)格數(shù)據(jù)變動(dòng)的預(yù)測。為此，本文所做的主要研究工作有： 針對股票價(jià)格數(shù)據(jù)影響因素多的問題，選用主成分分析法來解決了BP神經(jīng)網(wǎng)絡(luò)輸入向量的維數(shù)約減問題，同時(shí)，為了建立影響因素和預(yù)測向量之間的相關(guān)性關(guān)系，提高預(yù)測的精度，引入了計(jì)量經(jīng)濟(jì)學(xué)中擬合優(yōu)度的概念，結(jié)合傳統(tǒng)的主成分分析法，創(chuàng)新性的提出了相關(guān)主成分分析法。 針對BP算法容易陷入局部極小點(diǎn)而影響預(yù)測精度的缺點(diǎn)，利用遺傳算法對BP神經(jīng)網(wǎng)絡(luò)進(jìn)行優(yōu)化，構(gòu)建了遺傳算法優(yōu)化的BP神經(jīng)網(wǎng)絡(luò)預(yù)測模型。在前面算法的基礎(chǔ)上，建立了以相關(guān)主成分分析法和遺傳神經(jīng)網(wǎng)絡(luò)模型相結(jié)合的綜合預(yù)測模型，并在Matlab7.0中予以實(shí)現(xiàn)。 最后，，為了檢驗(yàn)所提出算法的有效性。文章最后進(jìn)行實(shí)驗(yàn)，利用上證指數(shù)數(shù)據(jù)，對文章提出的相關(guān)主成分分析法、遺傳神經(jīng)網(wǎng)絡(luò)模型和綜合預(yù)測模型分別進(jìn)行了仿真實(shí)驗(yàn)并進(jìn)行了誤差分析，誤差分析表明以相關(guān)主成分分析法為維數(shù)約減方法的遺傳神經(jīng)網(wǎng)絡(luò)模型在股票價(jià)格短期預(yù)測精度上有一定程度的改進(jìn)。
[Abstract]:With the rapid economic growth and the continuous expansion of the stock market, the stock market has produced a large number of valuable data information, which has become an important analysis of investors in stock investment. At the same time, stock price prediction has become an important research object for investors and related scholars. The growing data is not only difficult to deal with, but also a difficult problem for stock price forecasters. BP neural network is favored by investors and researchers as the classical algorithm of big data prediction. However, some inherent shortcomings of BP algorithm also restrict the efficiency and effect of its prediction, and there are still some defects in forecasting accuracy in the short-term forecasting of stock price. Based on the in-depth analysis of the problems faced by short-term forecasting of stock prices and the comparison of various methods of forecasting stock prices, this paper discusses BP neural network. Feasibility of short-term forecasting of stock price by principal component analysis and genetic algorithm. BP neural network can find out the inherent law of stock market development and change by learning from past stock market data. In order to achieve a period of future stock price data changes in the prediction. Therefore, the main research work in this paper is as follows: aiming at the problem that there are many factors affecting stock price data, principal component analysis (PCA) is used to solve the dimension reduction problem of BP neural network input vector, at the same time, In order to establish the correlation between influencing factors and prediction vectors and improve the accuracy of prediction, the concept of goodness of fit in econometrics is introduced. Combining with the traditional principal component analysis, the correlation principal component analysis is innovatively proposed. In view of the disadvantage that BP algorithm is prone to fall into local minima and affect the prediction accuracy, BP neural network is optimized by genetic algorithm, and a BP neural network prediction model optimized by genetic algorithm is constructed. Based on the previous algorithms, a comprehensive prediction model based on correlation principal component analysis (PCA) and genetic neural network (GNN) is established and implemented in Matlab7.0. Finally, in order to verify the effectiveness of the proposed algorithm. At the end of the paper, the experiment is carried out, and the related principal component analysis method, genetic neural network model and comprehensive prediction model are simulated and the error is analyzed by using the index data of Shanghai Stock Exchange. The error analysis shows that the genetic neural network model with correlation principal component analysis as dimension reduction method has a certain degree of improvement in the short-term forecasting accuracy of stock price.
【學(xué)位授予單位】：重慶交通大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：F832.51;TP183

【參考文獻(xiàn)】

相關(guān)期刊論文 前10條

1 夏莉,黃正洪;馬爾可夫鏈在股票價(jià)格預(yù)測中的應(yīng)用[J];商業(yè)研究;2003年10期

2 張U

本文編號(hào)：2177348