Transformer-based reinforcement learning for optical cavity temperature control system

Abstract

The accuracy of laser gas detection technology is influenced by the temperature of the optical cavity. Traditional control methods suffer from inadequacies in fully considering the coupling effects between features and the time delay in heat transfer. To address these issues, a method combining Transformer and reinforcement learning (RL) has been proposed. By using Transformer, this method generates enhanced features that are then used by the RL algorithm for iterative learning, aiming to optimize the control strategy. Additionally, a dual attention mechanism is introduced to enhance the model’s comprehension of the complex dynamics within the optical cavity. This study represents the first application of Transformer in the field of temperature control, paving the way for the utilization of advanced machine-learning techniques in optical cavity temperature regulation. Experimental results confirm the proposed method’s efficiency and long-term effectiveness in ensuring precise temperature control, demonstrating its potential in managing the complex cross-coupling effects within temperature control systems.