Management Method for Commissioning AI-Based Real-Time Closed-Loop Operation Systems in Refining and Petrochemical Units

Abstract

AI-based real-time closed-loop operation systems for refining and petrochemical units integrate model computation, strategy generation, and control execution into the production control chain. Consequently, management evaluation of their closed-loop commissioning cannot be limited to confirming whether the software system has completed engineering delivery. It must also determine separately whether the models can continuously and stably produce reliable computational results, whether the control chain can execute strategies under constraints, and whether the system achieves the expected operational outcomes after being placed in sustained operation. To meet the management needs associated with the application of this innovative technology, this paper proposes a category-based evaluation method for closed-loop commissioning. The method divides the evaluation objects into four categories—engineering delivery, model capability, closed-loop execution, and sustained operation and outcomes—establishes indicators, evidence, a statistical basis, and evaluation conclusions for each category, and defines rules for matching evidence to conclusions, making itemized determinations, and updating conclusions. Using the commissioning of an AI-based real-time closed-loop operation system in Yunnan Petrochemical's 13-million-metric-ton-per-year atmospheric and vacuum distillation unit (AVDU) as a case study, the category-based evaluation method produces separate, evidence-based conclusions on construction results, model performance, closed-loop capability, and operational outcomes, thereby providing refining and petrochemical units with an approach to managing and evaluating system commissioning that is suited to technological innovation.