基于启发式动态规划的盾构土压平衡优化控制

Optimization control for soil pressure balance of shield based on heuristic dynamic programming

为了避免盾构施工造成地表变形而引发安全事故,建立有效的密封舱土压平衡自动控制方法是至关重要的.为此,提出了一种新的基于启发式动态规划(heuristic dynamic programming,HDP)的盾构土压平衡优化控制方法,该方法能够实现盾构机多参数实时协调优化控制.基于神经网络,构建组成密封舱土压HDP控制器的执行网络、模型网络和评价网络,采用梯度下降算法对网络权值进行在线训练;利用评价网络迭代逼近密封舱土压优化控制代价函数,将多级优化问题简化为单级优化,从而显著降低实时计算负荷;执行网络以最小化代价函数为目标,对多参数同步协调优化,获得推进速度、刀盘扭矩、总推力、螺旋输送机转速的最优值.仿真结果表明,该方法能够有效控制密封舱土压平衡,HDP控制器具有较强的抗干扰能力,并且控制过程更加快速、稳定.

In order to avoid the safety accidents caused by surface deformation in shield construction,it is very important to establish an effective automatic control method for the soil pressure balance of sealed cabin.For that,a new optimization control method for soil pressure balance of shield is proposed based on heuristic dynamic programming(HDP),which can realize coordinated optimization control of shield machine′s multi-parameter in real time.Firstly,based on neural network,the action network,model network and critic network of the HDP controller for soil pressure in sealed cabin are established,and the gradient descent algorithm is used to train the network weights online.Then,the critic network is used to iteratively approximate the cost function of soil pressure control in sealed cabin,so that the multi-stage optimization problem is simplified to single-stage optimization,which can greatly reduce the calculation load.Finally,with the goal of minimizing the cost function of soil pressure control,the action network realizes the synchronous coordinated optimization of multi-variable such as advancing speed,cutterhead torque,total thrust and screw conveyor speed.The simulation results show that the method can effectively control the soil pressure balance of sealed cabin,and the HDP controller has strong anti-interference ability.What′s more,the control process is quicker and steadier.