基于Q学习的超超临界机组协调系统模型预测控制研究

Research on Model Predictive Control of Ultra-supercritical Unit Coordination System Based on Q Learning

超超临界机组具有多输入多输出、非线性严重、耦合强、时滞大的特点,其控制系统的设计要求是在保证系统稳定运行的同时,最大限度地提高机组的热效率和功率输出、快速准确地调节机组负荷、确保机组主蒸汽压力和温度在允许范围。通过调整系统模型以及阶段成本,进而获取最优策略。选取强化学习中的Q学习方法。将参数化模型预测控制作为Q学习中的一种新型函数逼近器,通过调整模型预测控制的参数来获取近似最优闭环控制策略以及最优的值函数。将该方法应用到1 GW超超临界锅炉汽轮机协调系统的控制中。在升负荷以及小范围升负荷变化情况下进行仿真。仿真结果表明,与传统模型预测控制方法对比,该方法可以在满足约束的情况下使系统输出值更加准确地到达设定值。

Ultra-supercritical unit is a complex system with multiple inputs and multiple outputs,serious nonlinearity,strong coupling and large time delay,which is difficult to model.The control system designed for it should ensure the stable operation of the system and maximize the thermal efficiency and power output of the unit,adjust the load of the unit quickly and accurately,and ensure that the main steam pressure and temperature of the unit are within the allowable range.Based on the principle that the optimal strategy can be obtained by adjusting the model of the system and the stage cost,this paper selects the Q learning method in reinforcement learning,takes parametric model predictive control as a new function approximator in Q learning,and obtains the approximate optimal closed-loop control strategy and optimal value function by adjusting the parameters of model predictive control.This method is applied to the control of the coordinated system of 1000 MW ultra-supercritical boiler and steam turbine.The simulation is conducted under conditions of load increase and small-range load increase,and results show that this method can reach the set value accurately and effectively under the condition of satisfying the constraints compared with traditional model predictive control methods.