Cascaded MPC Constraint Abstraction for Real-Time Plantwide Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional multi-variable model predictive control (MPC) systems in industrial processes face challenges in optimizing overall plant operations due to the complexity and size of upper tier models, which become unwieldy and difficult to solve in real-time, lacking a true multi-scale approach that effectively integrates constraints and variables across different scales.
Innovation Solution
A cascaded MPC approach that constrains conjoint measured variables (CMVs) by abstracting fundamental limitations onto a small set of CMVs, allowing the upper tier controller to pass target values to lower tier controllers, which optimize and update operating points while communicating selected numerical constraints back to the upper tier for overall business optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a super-model approach is used to capture all variables and constraints from lower tier controllers, then the upper tier controller can achieve global optimization, but the model size grows very quickly and becomes too unwieldy to solve in real-time
Solution Approach 1:
The patent segments the control system into multiple tiers (upper tier and lower tier controllers) with distinct responsibilities. The upper tier controller handles global optimization with a simplified model, while lower tier controllers manage local control with detailed models. This segmentation allows the upper tier model to remain computationally tractable while still achieving global optimization through coordinated control of key variables.
Solution Approach 2:
The patent extracts only the essential constraints and variables from lower tier controllers that are relevant to global optimization, rather than including all variables and constraints. This extraction process creates a simplified upper tier model that captures the critical relationships needed for plantwide optimization without the computational burden of a complete super-model.
2Reliability
If all variables from lower tier controllers are included in the upper tier model, then complete system optimization is achieved, but the computation time exceeds real-time requirements
Solution Approach 1:
The patent divides the optimization problem into two segments: global optimization at the upper tier focusing on key performance variables, and local optimization at the lower tier handling detailed process variables. This time segmentation allows the upper tier to compute quickly with reduced variables while the lower tier handles the computationally intensive local optimizations, achieving both completeness and real-time performance.
Solution Approach 2:
The patent applies partial action by including only the necessary subset of variables and constraints in the upper tier model that are sufficient for global optimization. Rather than modeling every variable, the approach uses a simplified representation that captures the essential system behavior and interactions, achieving adequate optimization without excessive computation time.
3Productivity
If the upper tier controller uses a simplified model with fewer variables, then real-time computation is achieved, but the coordination between process units is insufficient
Solution Approach 1:
The patent applies local quality by assigning different levels of model detail to different tiers. The upper tier uses a simplified model with aggregated variables appropriate for global coordination, while lower tier controllers use detailed local models for precise unit-level control. This differentiated approach ensures real-time computation at the upper tier while maintaining accurate inter-unit coordination through the exchange of key variables and constraints.
Data Source
AI summary
A cascaded MPC system includes an upper tier controller and lower tier controller having stored constraints for controlling a process having manipulated variables (MVs), controlled variables (CVs), and conjoint manipulated variable (CMV). The upper tier passes a target value for the CMV to the lower tier which optimizes for determining a local optimal operating point for the MVs, CVs, and CMV, moves towards the target value starting at the CMVs local operating point, and optimizes for identifying of the constraints as selected constraint(s) when the moving is truncated, passes the selected constraint(s) to the upper tier which performs an overall optimization for the process using the selected constraint to generate an optimal value for the CMV that lower tier uses as a new CM V target value for redetermining updated local optimal operating points for the MVs, CVs, and CMV, and for controlling the process utilizing the updated operating points.