基于多输出最小二乘支持向量回归建模的自适应非线性预测控制及应用

Multi-output least squares support vector regression modeling based adaptive nonlinear predictive control and its application

提出一种可有效提高常规预测控制方法控制性能与计算效率的数据驱动自适应非线性模型预测控制方法.首先,为了提高多输出非线性系统最小二乘支持向量回归(least squares support vector regression, LS–SVR)建模的精度,考虑各维输出间的耦合关系,采用在目标函数中加入样本整体拟合误差项,实现多输出LS–SVR(multi-output LS–SVR,M–LS–SVR)预测建模,同时采用粒子群算法优化模型参数;其次,针对动态过程建模的模型失配问题以及由于M–LS–SVR模型复杂导致传统智能算法求解预测控制律缓慢的问题,提出自适应非线性模型预测控制策略,包括两个非线性优化层:第1层采用梯度下降算法实时优化模型和实际过程输出的偏差,以自适应调节模型参数;第2层采用具有全局收敛性和超线性收敛速度序列二次规划(sequential quadratic programming, SQP)算法设计非线性预测控制器,以加速预测控制律的求解速度. Benchmark仿真实例及在高炉炼铁过程的数据试验表明:所提基于M–LS–SVR预测建模的自适应非线性模型预测控制具有较快的求解速度、较好的设定值跟踪和干扰抑制性能以及较强的鲁棒性.

This paper proposes a novel data-driven adaptive nonlinear predictive control method,which can effectively improve the control performance and computing efficiency of the conventional predictive control methods.First,in order to improve the accuracy of least squares support vector regression(LS–SVR)modeling for multi-output nonlinear systems,and considering the coupling relationship among multiple outputs,multi-output LS–SVR(M–LS–SVR)prediction modeling is proposed in this paper by adding the whole sample fitting error term to the optimization objective function.And the particle swarm optimization(PSO)algorithm is used to optimize the model parameters.Then,in view of the model mismatch of dynamic process modeling,and considering that the complexity of the M–LS–SVR model may lead to the slow optimization speed of predictive control with traditional intelligent algorithms,a novel adaptive nonlinear predictive control scheme is proposed,which includes two phases of nonlinear optimization.The first phase is to adopt the gradient descent algorithm to optimize the error between the model outputs and the actual ones in real time,so as to adjust the model parameters.And the second one is to use sequential quadratic programming(SQP)algorithm with global convergence and superlinear convergence speed to design the nonlinear predictive controller,so as to accelerate the speed of predictive control solving.Benchmark simulation and data experiment in a blast furnace ironmaking process show that the proposed method has fast computing speed,and good performances of setpoint tracking,disturbance rejection and robustness.