基于混合量子-经典神经网络模型的股价预测  

Stock Price Prediction Based on a Hybrid Quantum-Classical Neural Network Model

作　　者：张晓旭 高振涛 吴磊 李鑫 卢明静 ZHANG Xiaoxu;GAO Zhentao;WU Lei;LI Xin;LU Mingjing(CCB Fintech Co.Ltd.Pudong New Area,Shanghai 200120;Sichuan Yuanjiang Technology Co.Ltd.,Chengdu 611730)

机构地区：[1]建信金融科技有限责任公司,上海浦东新区200120 [2]四川元匠科技有限公司,成都611730

出　　处：《电子科技大学学报》2022年第1期16-23,共8页Journal of University of Electronic Science and Technology of China

基　　金：建信金融科技有限责任公司研究性课题(KT2000050);成都市重点研发支撑计划重大科技应用示范项目(2021-YF09-00114-GX)的资助。

摘　　要：股票预测本质上是数据挖掘的问题,大盘走势是一个很好的股票买卖时机抉择信号。在量化分析中,常用深度学习技术对大盘历史数据进行拟合与特征提取,为股票投资提供决策参考。该文首先训练了一个经典深度神经网络对沪深300的日K量价数据进行监督学习,实现了一个输出"涨跌"概率的二分类预测器,并以此制定策略进行模拟交易,利用测试集数据计算累积收益率,从而评估投资策略的优劣。此外,还构造了一种混合量子-经典神经网络模型,充分利用量子计算的线路模型特点,构造参数化变分量子线路,实现了量子前馈神经网络。在量子线路学习框架中,将股票的特征因子编码到量子态的振幅上,通过训练量子神经网络U的参数θ,迭代得到一个最优的分类器。量子算法的运行时间比经典算法少了7.7%,预测准确率更高,回报率高出3%,因此证明了量子算法的表达力强、鲁棒性高的特点。Stock price prediction is essentially a problem of data mining.The market trend is a good signal for stock trading timing.In quantitative analysis,deep learning technology is often used to fit and extract the characteristics of the market historical data,so as to provide decision-making references for stock investment.In this paper,a classical deep neural network is presented and trained to implement supervised learning based on the daily candlestick chart’s volume and price data of CSI 300;a binary predictor is realized to output the’’rise’’or’’fall’’labels and simulate the transaction with the strategy of the predictor;and then the test data set is utilized to calculate the cumulative rate of return,so as to estimate the advantages and disadvantages of the investment strategy.In addition,another hybrid quantum-classical neural network model is constructed,it makes full use of the characteristics of the circuit model of quantum computing to form the parameterized variational quantum circuit and realize the quantum implementation of feedforward neural network.In the frame of quantum circuit learning,the stock indicators are encoded into the amplitude of quantum states and the parametersθof the quantum neural network U are trained,and finally an optimal classifier is obtained after iterations.The results show the characteristics of strong expression and high robustness:the run-time of the quantum algorithm is 7.7%shorter than that of the classical algorithm and the prediction accuracy is higher,resulting in a return rate advantage of 3%.

关 键 词：深度学习 混合量子-经典神经网络 量子金融 股价预测 技术面因子 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]