Artificial Intelligence has become a key technology in the financial sector, improving the analysis of market data and supporting investment decisions. Reinforcement learning provides a framework for developing trading agents that learn through interaction with dynamic market environments.

This paper investigates the application of reinforcement learning to automated intraday stock trading using a Dueling Deep Q-Network (Dueling DQN) agent. The trading environment is based on historical price and volume data, while the reward function is defined using changes in portfolio value (Profit and Loss – PnL). The study also examines the use of the Volume-Weighted Average Price (VWAP) as an additional feature in the agent's state representation.

The proposed approach is evaluated by comparing a baseline model with a VWAP-enhanced model. The results show that the Dueling DQN agent can learn effective intraday trading strategies. Although the baseline model achieved higher cumulative returns, the VWAP-enhanced agent adopted a more conservative trading strategy, suggesting potential benefits for risk management during volatile market conditions.