采用双速率框架的快速预测控制算法
Double-speed-frame-based fast predictive control algorithm

约束模型预测控制(model predictive control, MPC)在实际应用中优化计算复杂度高, 无法在采样周期内完 成优化以保证系统实时性. 本文针对这一问题, 提出采用双速率框架的快速预测控制算法(DSF–MPC). 该算法将实 时控制量的求解分解到两个时间尺度上进行, 即双速率框架: 每隔数个采样周期, 慢速率层负责完成一次对完 整MPC优化问题的求解; 而在每个采样周期, 快速率层负责根据系统反馈信息和慢速率层算法预测信息的差值, 朝 着使目标函数值下降的负梯度方向, 修正慢速率层的优化结果来获取实际控制量, 以满足控制的实时性要求. 该算 法不要求在每个采样周期内都完成MPC中的在线优化, 故能在继承MPC优点的同时, 满足快速系统的控制实时性 要求. 针对直流电动机和倒立摆系统的仿真结果, 验证了该算法的有效性, 反映了其在快速系统中的应用潜力.
The constrained model predictive control (MPC) is often with high computational complexity, there may not be enough time to solve the optimization problem in a sampling period. To solve this problem, we propose a double-speedfame- based fast predictive control algorithm (DSF–MPC), in which the solving process of real-time control is decomposed into two parts with high and low speeds. In every finite number of sampling periods, the low-speed part solves once the MPC optimization problem to obtain the control value. In each sampling period, the fast part determines the real control value by modifying the control value obtained from the low-speed part in the negative gradient direction of the objective function, based on the difference between the feed-back information from the system and the predictive information from the low speed part. Since this algorithm doesn’t solve the optimization problem in each sampling period, it not only meets the real-time requirement of fast-speed system, but also inherits advantages of MPC. Simulation results of the system model of DC motor with inverted pendulum show the effectiveness of this proposed algorithm and reflect its application potentiality for fast systems.