【摘要】 流程工业智能工厂在精益化运行方面已取得显著成效,但在数据透明化程度、核心工业软件之间信息互通以及分析结果深度洞察等方面仍面临瓶颈.为了解决这些挑战并建设从自主运行到自主优化的新型流程工业智能工厂,本文探讨了将流程工业智能工厂的核心工业软件与新一代人工智能的大模型技术结合的新路径.基于此,本文提出了工业大模型赋能的新型流程工业智能工厂核心工业软件体系.该体系构建了基于大语言模型的工业大模型,其架构分为模型底座层、公共能力层和业务应用层.其中,公共能力层提供时序数据、图像数据、文本数据等多模态处理能力,业务应用层则结合具体业务场景开发了多种类型的智能体,包括图表智能体、低代码智能体、感知智能体、分析智能体、诊断智能体、决策优化智能体和控制智能体.这些智能体能够准确执行人类通过自然语言发出的各项任务,并通过组合调用公共能力层的能力来处理数据和知识.这种新型体系旨在提高核心工业软件的数据透明化程度、实现更完善的信息互通和利用,并提供更具价值的营运分析结果.本文以比例–积分–微分控制(proportional-integral-derivative control, PID)性能评估与整定系统为例,展示了工业大模型赋能核心工业软件的应用效果.在此示例中,分析智能体负责整体调度,协调感知智能体获取数据、图表智能体进行绘制、诊断智能体分析原因以及控制智能体提供参数优化方案,各智能体通过事件总线实现松耦合交互.这证明了通过智能体之间的协同作用,能够提升工业数据的透明化和工业场景的智能化水平.更多还原

【Abstract】 Smart factories in the process industries have made notable strides in lean operations, yet challenges remain in achieving data transparency, seamless interoperability among core industrial software systems, and the generation of deep, actionable insights. To overcome these limitations and advance toward a new paradigm-shifting from autonomous operation to autonomous optimization, this paper proposes an innovative architecture that integrates core industrial software with next-generation industrial foundation models. Built upon large language models, the architecture is structured into three layers: the model foundation layer, the public capabilities layer,and the business application layer. The public capabilities layer enables multimodal data processing across time series, image, and text data, while the business application layer incorporates specialized intelligent agents,such as charting, low-code, perception, analytics, diagnostic, decision optimization, and control agents. These agents can interpret natural language instructions, invoke shared capabilities, and manage domain knowledge flexibly. The proposed architecture enhances data transparency, improves software interoperability, and delivers deeper operational insights. A PID performance evaluation and tuning case study demonstrates the framework’s effectiveness, where multiple agents collaborate via an event bus to coordinate data acquisition, visualization,diagnosis, and parameter optimization. This example highlights how agent-based collaborative intelligence can significantly elevate the transparency and intelligence of industrial process operations.更多还原