从模型预测控制技术 50 年看成功的技术突破

模型预测控制(model predictive control, MPC)是一种使用数学模型在有限时间内实时优化控制系统的技术。MPC 自 20 世纪 70 年代问世以来，已广泛应用于化学工程、炼油、先进制造、机器人和航空航天等各个领域。机器学习(machine learning, ML)是人工智能的一个分支，研究如何使计算机从数据中学习并执行需要人类智能的任务。随着神经网络、遗传算法和专家系统的发展，ML 在 20 世纪 80 年代中期成为一个独特的领域。统计监控 (statistical process monitoring, SPM) 是收集和分析数据以检测系统或过程中的异常、故障或变化的过程，SPM 一直用于质量控制、故障诊断和可靠性评估。首先回顾我作为这些领域的研究者和从业者的旅程，并介绍我的一些工作。然后，我将回顾过去几十年 MPC、ML 和 SPM 的主要发展和挑战，以及它们如何落地和影响工程实践。最后，讨论目前 ML和 AI 等新技术落地工业应用的必要条件和应对策略。

Model predictive control (MPC) is a technique that uses mathematical models to optimize a control system in real time in a finite time. Since its inception in the 1970s, MPC has found extensive applications in various fields, including chemical engineering, oil refining, advanced manufacturing, robotics, and aerospace. Machine learning (ML), a branch of artificial intelligence, focuses on enabling computers to learn from data and perform tasks requiring human intelligence. With the development of neural networks, genetic algorithms, and expert systems, ML emerged as a distinct field in the mid-1980s. Statistical process monitoring (SPM) involves collecting and analyzing data to detect anomalies, faults, or changes in a system or process. SPM has been widely used for quality control, fault diagnosis, and reliability assessment. In this paper, I first review my journey as a researcher and practitioner in these fields and introduce some of my work. Then, I will review the main developments and challenges in MPC, ML, and SPM over the past few decades, along with their practical implementations and impact on engineering practices. Finally, the essential requirements and strategies for the industrial application of new technologies such as ML and AI will be discussed.