Deep Reinforcement Learning Stock Trading Strategy Optimization Framework Based on TimesNet and Self-Attention Mechanism

In the financial market, the design and optimization of stock trading strategies have become a key focus for investors. With the globalization and digitization of markets, traditional trading strategies often struggle to cope with complex market dynamics, especially in environments characterized by high-frequency volatility and multiple influencing factors. Deep Reinforcement Learning (DRL), as an emerging intelligent algorithm, has shown potential in complex nonlinear markets by learning and optimizing strategies through interactions with the market environment. However, existing DRL models still face challenges in handling long-term dependencies in time series data and market noise. Therefore, this study proposes a deep reinforcement learning framework based on TimesNet and self-attention mechanisms, aiming to overcome the limitations of traditional methods in time series modeling, complex data feature capturing, and strategy optimization. By integrating the multi-scale feature extraction capability of TimesNet with the global dependency capturing advantages of self-attention mechanisms, this research seeks to enhance the intelligence level and trading effectiveness of stock trading strategies, thereby providing investors with more adaptive decision support.