二维基于旋转不变技术信号参数估计(2D-estimating signal parameter via rotational invariance techniques,2D-ESPRIT)算法是估计几何绕射理论(geometric theory of diffraction,GTD)模型参数的一种经典算法,但在信噪比较低的条件下,2...二维基于旋转不变技术信号参数估计(2D-estimating signal parameter via rotational invariance techniques,2D-ESPRIT)算法是估计几何绕射理论(geometric theory of diffraction,GTD)模型参数的一种经典算法,但在信噪比较低的条件下,2D-ESPRIT算法的参数估计精度明显下降,噪声鲁棒性较差。针对这一问题,提出一种极化平方前后向平滑2D-ESPRIT(polarized-quadratic-forward-backward 2D-ESPRIT,PQ-FB-2D-ESPRIT)算法,有效地提高了算法的噪声鲁棒性与参数估计性能。改进算法利用目标散射回波数据的极化信息,并通过对协方差矩阵平方处理和前后向空间平滑处理,提高了算法的参数估计性能与数据利用率,同时达到了去相关的效果。仿真结果表明,提出的PQ-FB-2D-ESPRIT算法的参数估计性能及噪声鲁棒性要优于经典2D-ESPRIT算法、前后向平滑2D-ESPRIT(forward-backward 2D-ESPRIT,FB-2D-ESPRIT)算法及平方FB-2D-ESPRIT(quadraticFB-2D-ESPRIT,Q-FB-2D-ESPRIT)算法。基于不同算法估计得到的GTD模型参数对散射中心的定位精度进行比较,进一步验证了改进算法的优越性与有效性。展开更多
2024年9月中国A股市场大涨,再次点燃了全民的“炒股热”。然而,牵动股民心弦的股价涨跌——却跟许多因素息息相关。对于散户来说,除了筛选信息进行股票的买进卖出以外,通过算法模型预测也能够起到事半功倍的效果。上世纪六十年代初便有...2024年9月中国A股市场大涨,再次点燃了全民的“炒股热”。然而,牵动股民心弦的股价涨跌——却跟许多因素息息相关。对于散户来说,除了筛选信息进行股票的买进卖出以外,通过算法模型预测也能够起到事半功倍的效果。上世纪六十年代初便有了通过计算机技术进行量化交易的雏形,随着技术的迭代,通过统计学和模型构建成为量化交易的主流选择。而本论文构建了一个使用A2C (优势行动–评论家)强化学习算法的股票交易模型。利用“gym-anytrading”库创建一个股票交易环境,并使用Stable-Baselines库训练一个策略网络来学习如何在该环境中进行交易以最大化收益。该模型的数据来源于Yahoo-Finance的阿里巴巴股票信息(2022年12月至2024年9月),通过pandas-datareader库的接口获取。In September 2024, a significant surge in China’s A-share market reignited the public’s “stock trading frenzy”. However, the fluctuating stock prices that excited stock investors were closely related to many factors. For individual investors, in addition to screening information for buying and selling stocks, using an algorithm model to predict can also have a twice-as-effective effect. In the early 1960s, the embryo of quantitative trading using computer technology had appeared, and with the advancement of technology, quantitative trading based on statistics and model building became the mainstream choice. This paper constructs a stock trading model using the A2C (Advantage Actor-Critic) reinforcement learning algorithm. By using the “gym-anytrading” library to create a stock trading environment and training a policy network using the Stable-Baselines library to learn how to trade in this environment to maximize profits. The data source for the model comes from the stock information of Alibaba (2022 December to 2024 September) obtained through the interface of the pandas-datareader library.展开更多
文摘二维基于旋转不变技术信号参数估计(2D-estimating signal parameter via rotational invariance techniques,2D-ESPRIT)算法是估计几何绕射理论(geometric theory of diffraction,GTD)模型参数的一种经典算法,但在信噪比较低的条件下,2D-ESPRIT算法的参数估计精度明显下降,噪声鲁棒性较差。针对这一问题,提出一种极化平方前后向平滑2D-ESPRIT(polarized-quadratic-forward-backward 2D-ESPRIT,PQ-FB-2D-ESPRIT)算法,有效地提高了算法的噪声鲁棒性与参数估计性能。改进算法利用目标散射回波数据的极化信息,并通过对协方差矩阵平方处理和前后向空间平滑处理,提高了算法的参数估计性能与数据利用率,同时达到了去相关的效果。仿真结果表明,提出的PQ-FB-2D-ESPRIT算法的参数估计性能及噪声鲁棒性要优于经典2D-ESPRIT算法、前后向平滑2D-ESPRIT(forward-backward 2D-ESPRIT,FB-2D-ESPRIT)算法及平方FB-2D-ESPRIT(quadraticFB-2D-ESPRIT,Q-FB-2D-ESPRIT)算法。基于不同算法估计得到的GTD模型参数对散射中心的定位精度进行比较,进一步验证了改进算法的优越性与有效性。
文摘2024年9月中国A股市场大涨,再次点燃了全民的“炒股热”。然而,牵动股民心弦的股价涨跌——却跟许多因素息息相关。对于散户来说,除了筛选信息进行股票的买进卖出以外,通过算法模型预测也能够起到事半功倍的效果。上世纪六十年代初便有了通过计算机技术进行量化交易的雏形,随着技术的迭代,通过统计学和模型构建成为量化交易的主流选择。而本论文构建了一个使用A2C (优势行动–评论家)强化学习算法的股票交易模型。利用“gym-anytrading”库创建一个股票交易环境,并使用Stable-Baselines库训练一个策略网络来学习如何在该环境中进行交易以最大化收益。该模型的数据来源于Yahoo-Finance的阿里巴巴股票信息(2022年12月至2024年9月),通过pandas-datareader库的接口获取。In September 2024, a significant surge in China’s A-share market reignited the public’s “stock trading frenzy”. However, the fluctuating stock prices that excited stock investors were closely related to many factors. For individual investors, in addition to screening information for buying and selling stocks, using an algorithm model to predict can also have a twice-as-effective effect. In the early 1960s, the embryo of quantitative trading using computer technology had appeared, and with the advancement of technology, quantitative trading based on statistics and model building became the mainstream choice. This paper constructs a stock trading model using the A2C (Advantage Actor-Critic) reinforcement learning algorithm. By using the “gym-anytrading” library to create a stock trading environment and training a policy network using the Stable-Baselines library to learn how to trade in this environment to maximize profits. The data source for the model comes from the stock information of Alibaba (2022 December to 2024 September) obtained through the interface of the pandas-datareader library.
