Maximum Power Point Tracking is essential for maximizing the efficiency of photovoltaic systems under varying irradiance and temperature con-ditions. Conventional MPPT techniques are simple to implement but often suffer from slow dynamic response, steady-state oscillations, and re-duced performance under rapidly changing environmental conditions. Advanced approaches improve robustness but may introduce chattering and remain suboptimal for the nonlinear characteristics of PV systems, especially under partial shading. To address these challenges, this paper proposes a hybrid intelligent MPPT strategy that integrates a Long Short-Term Memory (LSTM) neural network, Terminal Sliding Mode Control (TSMC), and Proximal Policy Optimization (PPO). The LSTM model predicts the optimal maximum power point voltage in real time, while the TSMC ensures fast and robust tracking. A PPO-based reinforcement learning agent adaptively optimizes the TSMC control parameters online, enhancing dynamic performance and reducing chattering. Simulation results under diverse operating conditions demonstrate that the proposed hybrid controller significantly outperforms conventional methods. The proposed ap-proach achieves a high tracking efficiency of 97.4%, a low average tracking error of 0.25, and a fast response time of 65 ms. These results confirm the effectiveness of combining deep learning and reinforcement learning for robust and adaptive MPPT in PV systems.