Handling Constraints and Uncertainty in Chemical Process Operation Using Nonlinear Model Predictive Control

长期以来，化工过程的控制一直面临着诸如输入限制(阀门和泵的能力有限)、非线性和不确定性(扰动或对象模型不匹配)等挑战。虽然在过去的几十年里在控制理论方面有了重要的研究，但现代传感和计算能力的可获得性已经开启了过去难以计算的控制设计的可能性。*使用控制策略来实现过程所需操作的概念是一个公认的概念。控制算法的实施所依据的三个“支柱”是传感器(如温度传感器)、模型和控制执行器或输入(如阀门或泵)。根据测量的变量来量化各种系统的期望操作。连接这些部分的是过程的模型，或者本质上是对控制执行器如何影响测量的过程变量的理解(通常是定量的)。然后，控制器需要确定如何最好地移动控制执行器以实现所需的行为。*早期的控制方法往往忽略了非线性，即其他操纵变量对所讨论的过程变量的影响，通常倾向于采取不那么激进的控制行动，以避免这些未测量的影响或干扰的负面影响。*因此，整个过程自动化领域将受益于更新的控制方法，这些方法充分利用现有的计算资源来解决非线性、约束、故障处理等重要问题，并辅之以改进的模型。拟议的研究通过解决尚未解决的问题，即确定“最佳方向”，以推动系统在执行机构受限的情况下实现预期的操作，将影响涉及自动化的每一个过程(从炼油厂到建筑物温度控制，再到车辆控制)。保守估计，效率提高1%，这很容易转化为加拿大数十亿美元的节省，以及全球范围内的更多节省。改进模型的另一个重点是将统计建模工具与基于状态空间的控制工具相结合，用于批处理过程控制，从而节省从生物制药到特种化学品等生产过程的成本。最后，将解决对加拿大工业中的环境和安全问题产生积极影响的大型化学过程的故障检测和处理问题。直接技术转让将通过与麦克马斯特先进控制联盟的工业合作伙伴合作进行，PI是该联盟的成员。考虑到受过区域过程控制培训的学生的快速安置，这笔赠款产生的25个HQP将满足加拿大工业目前和未来的需求。*

Control of chemical processes has long grappled with challenges such as the presence of constraints on the inputs (valves and pumps limited in capacity), nonlinearity and uncertainty (disturbances or plant model mismatch). While there has been significant research in control theory over the last several decades, the availability of modern sensing and computational capabilities have opened up the possibilities of control designs that were in the past computationally intractable. ***The notion of use of control strategies to achieve the desired operation of a process is a well established concept. The three `pillars' on which the implementation of control algorithm stands are sensors (such as temperature sensors), a model, and a control actuator or input (such as valves or pumps). The desired operation of various systems is quantified in terms of the measured variables. Connecting these pieces is a model of the process, or essentially an understanding (typically quantitative) of how the control actuators effect the measured process variables. The controller then needs to determine how best to move the control actuators around to achieve the desired behavior. ***Early control approaches often neglected nonlinearity, the effect of other manipulated variables on the process variable in question, and generally favored making less aggressive control actions to avoid the negative impact of these unmeasured effects or disturbances.*The entire field of process automation thus stands to benefit from newer control approaches that fully harness the available computational resources to address the important problems of nonlinearity, constraints, handling of faults, and aided by improved models. The proposed research, by addressing the unsolved problem of determining the `best direction' to push the system to achieve desired operation subject to actuator limitations, will impact every process where automation is involved (from refineries to building temperature control to vehicle control). With a conservative estimate of one percent improvement in efficiency, this readily translates into billions of dollars in savings in Canada, and much more worldwide. The other thrust on improved models will unite statistical modeling tools with state-space based control tools for batch process control saving costs for production processes ranging from bio-pharmaceuticals to specialty chemicals. Finally, the problem of fault-detection and handling will be addressed for large scale chemical processes positively impacting environmental as well as safety issues in Canadian industry. Direct technology transfer will occur through collaboration with the industrial partners of the McMaster Advanced Control Consortium, of which the PI is a member. Given the rapid placement of students trained in the area process control, the 25 HQP resulting from the grant will fulfill current and future need in Canadian industry.*** **