- 學位論文
藉工廠資產架構裡定義之控制迴路導入智慧化模型預測控制
Implementation of intelligent model predictive control using the control loop defined in plant asset structure
摘要
因應工業4.0的興起，以及因而發展出來的眾多國際標準，例如ISA-95與OPC UA，整合與融合工廠內部資訊與操作控制成為製造業首要努力目標。針對目前多數工廠的問題，包括缺乏定期調校控制迴路中的控制器參數，以及未納入預測控制概念，有必要探討如何依據ISA-95定義出控制迴路之資產架構，進而導入自製智慧化模型預測控制。 本報告首先說明ISA-95資產架構下之控制迴路概念，包括該標準如何定義工廠PID控制迴路名稱、類別與能力，以及如何建立系統化資訊。其次，本報告討論工廠分散式控制系統、PID控制器與設備之間的關係，說明控制器參數定期調校的重要性，並介紹控制器參數理論與IMC理論。關於操作預測控制的概念，本報告分別說明先進預測控制(APC)、模型預測控制(MPC)，以及結合類神經網路技術所延伸出來的智慧化模型預測控制(NN MPC)。 本報告以一個實廠案例，完成了資產管理架構的建立、控制器參數調校、並利用實廠數據建立NNMPC。透過NNMPC來預測未來10分鐘的液位趨勢走向，經由測試集測試的模型性能達到0.75，其他數據集測試模型性能達到0.73。
並列摘要
In response to the rise of Industry 4.0 and the development of many international standards, such as ISA-95 and OPC UA, the integration and integration of factory internal information and operation control has become the primary goal of the manufacturing industry. In view of the current problems of most factories, including the lack of regular adjustment of the controller parameters in the control loop, and the failure to include the concept of predictive control, it is necessary to explore how to define the asset structure of the control loop based on ISA-95, and then introduce a self-made intelligent model prediction control. This report first explains the concept of control loops under the ISA-95 asset structure, including how the standard defines the names, types and capabilities of plant PID control loops, and how to create systematic information. Secondly, this report discusses the relationship between factory distributed control systems, PID controllers and equipment, explains the importance of regular adjustment of controller parameters, and introduces controller parameter theory and IMC theory. Regarding the concept of operational predictive control, this report separately describes advanced predictive control (APC), model predictive control (MPC), and intelligent model predictive control (NN MPC) extended by combining neural network technology. This report uses a real factory case to complete the establishment of asset management framework, controller parameter adjustment, and use real factory data to establish NNMPC. Through NNMPC to predict the trend of the liquid level in the next 10 minutes, the performance of the model tested by the test set reached 0.75, and the performance of the other data set tested models reached 0.73.