Model-free framework for dynamic characterization and optimization of plant-wide operation

在当今动态的市场和复杂的化工过程制造环境中，装置的所有单元的运行状态必须仔细协调，才能保持在经济最优附近。实时优化技术(RTO)解决了机组间的协调问题，引起了流程工业的广泛关注。传统的RTO系统是基于工厂的稳态流程图模型，对于大多数具有不同时间尺度的机组相互作用的综合工厂来说，其公式可能非常有限。其他利用整个工厂的动态行为的方法在任何可预见的未来都不太可能扩展到真正的工厂范围的问题，因为过高的计算要求。与实际系统缺乏可扩展的全厂动态和动态优化方案相反，分布式控制系统和数据采集工具的快速发展导致了海量和高维数据的可用。这些数据记录了各种运行模式和事件的历史。本研究的目的是开发基础理论和计算框架，通过从工厂历史运行数据中提取有价值的信息，而不是构建工厂范围的动态模型，来推导出改进的工厂范围运行策略。预计将所建议的框架应用于流程工业的决策层次将显著改善流程和经济绩效。

In today's dynamic market and complex manufacturing environment for chemical process industries, operating conditions of all units of a plant must be carefully coordinated in order to remain near an economic optimum. Real time optimization (RTO) solves the problem of coordinating various plant units and has attracted the attention of process industry. The conventional RTO system is based on a steady-state flow-sheet model of a plant, and its formulation can be extremely limited for most integrated plants with their units interacting across different time scales. Other approaches utilizing dynamical behaviour of an entire plant are unlikely to scale to real plant-wide problems in any foreseeable future because of exorbitant computational requirements. Contrary to the paucity of scalable plant-wide dynamics and dynamic optimization scheme for practical systems, the rapid growth of distributed control systems and data acquisition tools has resulted in the availability of massive and high dimensional data. Such data capture the history of various operation modes and events. The objective of my research is to develop fundamental theories and computational frameworks for deriving an improved plant-wide operation strategy by extracting valuable information from historical plant operation data without constructing a plant-wide dynamic model. It would be expected that applications of the suggested framework to the decision hierarchy in process industry will result in significantly improved process and economic performance.