Autonomous collaborative optimization control of earth pressure balance shield machine based on hierarchical control architecture

Due to manual operations are unable to adjust parameters promptly to adapt to continuously changing geological conditions, which can easily lead to safety accidents. Therefore, this study aims to solve the real-time dynamic optimization problem of shield machines and achieve autonomous optimal control. Based on cyber-physical system (CPS), this paper proposes a Twin Double Deep Deterministic Policy Gradient-Model Predictive Control (TD3-MPC) hierarchical autonomous control scheme, which is divided into coordination level and execution level. Based on Twin Double Deep Deterministic Policy Gradient (TD3) algorithm, TD3 agent is designed. A virtual tunnel environment has been designed based on sealed cabin pressure mechanism model. TD3 agent and virtual environment are used as coordination level. A dynamic state-space model for sealed cabin pressure is established, incorporating shield machine dynamics model and mechanism of sealed cabin pressure. On this basis, model predictive controller is designed as execution level. TD3 agent interacts with virtual tunneling environment based on geological information parameters of real-time sampling. An optimal sealed cabin pressure curve under current sampling interval is found and transmitted to executive level as the control target. Execution level solves the quadratic programming through real-time feedback of sealed cabin pressure and continuous rolling optimizes screw conveyor speed and propulsion speed to achieve accurate tracking of pressure controlling target. Simulated results demonstrated that this control approach has a good pressure control effect and strong soil adaptive ability. From multiple perspectives, viability and efficiency are further verified for the proposed controlling scheme. This paper further promotes the application of AI technology in underground construction.