Fuel cell starvation control using model predictive technique with Laguerre and exponential weight functions
Fuel cell starvation control using model predictive technique with Laguerre and exponential weight functions
复制标题
DOI：
10.1007/s12206-014-0348-3
复制
发表时间：
2014-05
影响因子：
1.6
通讯作者：
M. Abdullah;M. Idres
中科院分区：
文献类型：
--
作者：
M. Abdullah;M. Idres
燃料电池系统是一个复杂的系统，需要一个有效的控制器。模型预测控制由于其优化和约束处理的特点而成为一个重要的控制器。在这项工作中，一个改进的模型预测控制（MPC）与Laguerre和指数权重函数提出了控制燃料电池的氧饥饿问题。为了获得最佳性能的MPC，控制和预测的范围内选择尽可能大的计算限制。在优化过程中，应用指数权重函数来更加强调当前时间，而不太强调未来时间。这导致稳定的数值解大预测视野。Laguerre函数用于捕获大部分控制轨迹，同时减少控制器计算时间和大预测范围的存储器。所提出的控制器的鲁棒性和稳定性进行了评估，使用蒙特-卡罗仿真。结果表明，改进的MPC能够模仿无限时域控制器，离散线性二次型调节器（DLQR）的性能。与传统的MPC方案相比，控制器的计算时间减少了约一个数量级。蒙特-卡罗仿真结果表明，所设计的控制器对40%的系统参数不确定性具有鲁棒性和稳定性。
Fuel cell system is a complicated system that requires an efficient controller. Model predictive control is a prime candidate for its optimization and constraint handling features. In this work, an improved model predictive control (MPC) with Laguerre and exponential weight functions is proposed to control fuel cell oxygen starvation problem. To get the best performance of MPC, the control and prediction horizons are selected as large as possible within the computation limit. An exponential weight function is applied to place more emphasis on the current time and less emphasis on the future time in the optimization process. This leads to stable numerical solution for large prediction horizons. Laguerre functions are used to capture most of the control trajectory, while reducing the controller computation time and memory for large prediction horizons. Robustness and stability of the proposed controller are assessed using Monte-Carlo simulations. Results verify that the modified MPC is able to mimic the performance of the infinite horizon controller, discrete linear quadratic regulator (DLQR). The controller computation time is reduced approximately by one order of magnitude compared to traditional MPC scheme. Results from Monte-Carlo simulations prove that the proposed controller is robust and stable up to system parameters uncertainty of 40%.