基于KPCA-GWO-SVR组合模型的股票价格预测

Stock Price Prediction Based on KPCA-GWO-SVR Model

股票市场是反映经济动向的晴雨表，准确预测股票价格能够帮助市场稳定运行，指导投资者做出正确的投资策略。本文运用灰狼优化（GWO）算法对支持向量回归（SVR）中的参数进行优化，用优化后的SVR模型对股票价格进行建模预测，并通过核主成分（KPCA）技术对变量进行降维处理。从可推广性的角度考虑，本文在主板、中小板、创业板中各选取若干只股票，利用该组合模型进行中长期预测，并与其它模型进行对比。实验结果表明：该组合模型在股票价格预测中显著地减小了真实值与预测值之间的误差。在预测精度上，BP神经网络和SVR的预测精度均低于KPCA—GWO—SVR的组合模型，该模型针对不同市值规模的股票价格都可以进行有效预测。

The stock market is a barometer which can reflect economic activity. Predicting the stock price accurately can stabilize the market and guide investors to make the right investment strategy. This paper optimizes the parameters of Support Vector Regression （SVR） by applying the Grey Wolf Optimizer （GWO） algorithm,then uses the optimized SVR to predict the stock price, and finally process variables by using kernel principal component analysis （KPCA） technology. Based on the perspective of extensibility, this paper selects a number of stocks in the main board, medium - small board and gem board, uses the model for medium - term and long - term to predict and compares it with other models. The experimental results show that the composite model further reduces the error between the real and the predicted values in the stock price forecast. The prediction accuracy of BP neural network and SVR is not high in the KP- CA - GWO - SVR model. Meanwhile, the model can be effectively predicted for stock prices of different market value.