همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

A Spatio-Temporal Deep Reinforcement Learning Framework for Algorithmic Trading Using Genetically Evolved Dynamic Knowledge Graphs

نویسندگان :

Fatemeh Yousefloie¹ Ali Tavassolian² Mehrnoosh Shamsfard³ Monireh Abdoos⁴

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی 3- دانشگاه شهید بهشتی 4- دانشگاه شهید بهشتی

کلمات کلیدی :

Algorithmic Trading،Reinforcement Learning،Dynamic Knowledge Graphs،Genetic Algorithms،Explainable AI (XAI)

چکیده :

Financial markets dynamic and non-stationary behavior challenges traditional, static algorithmic trading systems. While modern deep learning models offer high predictive power, their black-box nature creates a critical gap in trust and explainability. To address these challenges, this paper introduces a novel spatio-temporal deep reinforcement learning framework for algorithmic trading. At each time step, our system employs a genetic algorithm to construct a dynamic knowledge graph. This graph serves as a real-time, optimized map of effective technical analysis strategies for the current market state. A sequence of these evolving graphs is then fed into a hybrid neural architecture. This architecture leverages a Graph Attention Network (GAT) to capture the complex relationships within each strategic map and an LSTM to model their temporal dynamics. The resulting rich spatio-temporal representation serves as the state representation for a deep reinforcement learning agent. This agent is trained using the PPO algorithm to learn an optimal policy for executing trades and selecting risk management levels (take-profit and stop-loss). This approach represents a paradigm shift from static prediction models to an adaptive agent that learns to make decisions based on the evolution of the market's strategic structure. Experimental results demonstrate that our framework not only achieves superior performance in generating profitable signals but also offers a novel form of hybrid explainability, enabling the analysis of the strategic knowledge graphs underlying the agent's decisions.