论文总字数：26304字

摘 要

长期以来，深度强化学习在人工智能领域始终是研究的热门，更是争相讨论的焦点，而股票市场在金融领域也一直是重要的不可或缺的存在。在股票交易时，怎样获取有价值有效的交易信息一直是相当重要的内容；获取的信息，进行的投资交易能否获利，如何获利，获利多少同样是核心探讨领域，对于股票投资策略，我们一直期望有更优化的方式出现，从而尽可能的降低风险，节省运营成本，充分进行资源整合，实现收益最大化，合理准确的预测投资策略。

本文主要是以深度强化学习理论为理论基础，进而将深度强化学习模型与国内沪深300的A股股票交易结合起来，做基于深度强化学习算法的A股投资分析与实证研究。文中综述了近些年来国内外发展起来的深度强化学习的基本理论及其模型概况，其中有DQN算法，Double DQN算法，Deuling DQN算法，Policy Gradient算法，DDPG算法等。实验部分使用了bigquant平台策略构造了基于DQN算法和Policy Gradient算法的股市深度强化学习模型，并且进行了模型之间的策略优化程度对比以及回测分析和泛化分析；也深入了解了Double DQN算法和DDPG算法；在实验分析中总结了基于此模型自动化投资的优缺点，便于后续深度研究和探讨。

通过本文深度强化学习的部分算法与股票投资交易结合的实验分析，验证了算法在股票投资领域的可操作性，可实施性，可塑造性，更加验证了自动化精准投资的可能性很大，实践性很强。这项研究对于投资者搭建自动化投资模型来提高投资策略收益率非常有利，既有利于股市投资策略的发展和深入研究，为未来智能化的股票投资市场奠定基础，加强实证证明，也为人工智能技术在金融投资领域的应用做出了不错的尝试，提出了更多的技术要求和调整需求。

关键词：深度强化学习 AI股票投资 实证分析 DQN算 Policy Gradient算法

ABSTRACT

For a long time, deep reinforcement learning has been a hot research topic in the field of artificial intelligence, and the stock market has been an important and indispensable existence in the field of finance. How to acquire valuable and effective trading information is very important in stock trading. Access to information, to make the investment trading profits, how to profit, how much is the core areas, same for stock investment strategy, we always expect to have more optimization approach, as much as possible to reduce risk, save operating costs, fully integrating resources, maximizing benefits, reasonable and accurate prediction of investment strategy.

The paper is based on the deep reinforcement learning theory, and then combines the deep reinforcement learning model with the a-share stock trading of csi 300 in China to do the investment analysis and empirical research of a-share based on the deep reinforcement learning algorithm. In this paper, the basic theories and models of depth reinforcement learning developed at home and abroad in recent years are summarized, including DQN theory, Double DQN theory, Deuling DQN theory ,Policy Gradient algorithm, DDPG theory, etc. In the experiment part, bigquant platform strategy was used to construct the stock market depth reinforcement learning model based on DQN algorithm and Policy Gradient algorithm, and the comparison of strategy optimization degree between the models as well as back-test analysis and generalization analysis were conducted. Also deeply understand the Double DQN algorithm and DDPG algorithm; The advantages and disadvantages of automatic investment based on this model are summarized in the experimental analysis, which is convenient for further research and discussion.

Through the experimental analysis of the combination of some algorithms of deep reinforcement learning and stock investment trading in this paper, the operability, enforceability and molding ability of the algorithm in the field of stock investment are verified, which further verifies the great possibility of automatic precise investment and strong practicality. The study for the investors to build automation model to improve the investment strategy yields very good, not only promotes the development of stock market investment strategy and in-depth study, laying the groundwork for future intelligent stock investment market, strengthen the empirical proof, also for the application of artificial intelligence technology in the field of financial investment has made a good attempt, the demand for more technical requirements and adjustment.

Keywords: deep reinforcement learning;AI automated investment;empirical analysis ;DQN algorithm;Policy Gradient algorithm

目录

基于深度强化学习算法的A股投资分析与实证 2

摘要 2

第一章 绪论 5

1.1 研究背景和研究意义 5

1.1.1 研究背景 5

1.1.2 研究意义 5

1.2 研究理论和研究内容 5

1.2.1 研究理论 5

1.2.2研究内容 6

第二章 国内外文献综述 7

2.1 深度学习文献综述 7

2.2 强化学习文献综述 7

2.3 深度强化学习文献综述 7

第三章 深度强化学习算法概述 9

3.1 DQN算法 9

3.2 Double DQN算法 10

3.3 Dueling-DQN算法 11

3.4 Policy Gradient算法 11

3.5 DDPG算法 12

第四章 基于深度强化学习建模实证分析 14

4.1 模型构建 14

4.1.1 数据选取 14

4.1.2 模型选取 14

4.2 实证分析 14

4.2.1 模型实验图解（2018年沪深300A股数据） 14

4.2.2 模型实验图解（2017年沪深300A股数据） 16

4.2.3 回测分析 17

4.2.4 泛化分析 18

4.2.5 模型对比 18

4.3 实验结论 18

第五章 总结与展望 19

5.1 本文总结 19

5.1.1 全文概述 19

5.1.2 创新之处 19

5.1.3 不足之处 19

5.2 未来展望 19

参考文献 21

致谢 23

请支付后下载全文，论文总字数：26304字