A study on the development of an automated control system for slipform construction based on deep reinforcement learning

Due to the relatively fixed structure and parameters of the linear sliding mode control, it lacks the ability to adjust the control strategy in real time according to the system state. When the template position deviation is large, the fixed control gain may lead to too strong control action, which makes the system response too drastic, and may even exceed the physical limitations of the system, resulting in instability or damage. For this reason, a deep reinforcement learning-based automated control method for slipform construction is proposed. In the hardware design part of the system, the slide mold construction automation control system is constructed based on the information acquisition module, human–computer interaction module, control module and analog output module; In the software design section, the sensor layout is optimized using particle swarm optimization algorithm to achieve accurate measurement of template positions. Design an iterative sliding mode controller to compensate for the position error of the template, and achieve precise control of the template position through error compensation. And further adjust the controller parameters through deep reinforcement learning algorithms. The experimental results show that this method has good template position control effect and strong anti-interference ability.