Hybrid Model Control Optimization for Physically Consistent Set Points
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current techniques for generating optimization solutions for physical systems using machine learning models may not be as accurate as desired, as they rely solely on empirical relationships without considering fundamental principles, leading to suboptimal set points for controlling complex production processes.
Innovation Solution
A computer-implemented method that combines a machine learning model with a first principle model to form an optimization model, using an objective function and constraints to determine set points for input variables where both models agree, thereby optimizing the target value for a physical system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a data-driven regression optimization using only machine learning models is used, then the system can handle complex relationships and provide optimization, but the accuracy and precision of the optimization solutions are insufficient
Solution Approach 1:
The patent combines machine learning models with first principle models into a hybrid optimization framework. The machine learning model captures complex empirical relationships from data, while the first principle model ensures physical consistency and accuracy. By merging these two approaches, the system achieves both adaptability to complex relationships and precision in optimization solutions through the complementary strengths of data-driven flexibility and physics-based reliability.
2Reliability
If a data-driven regression optimization is used, then the system can learn from historical data, but the solutions may not align with fundamental physical principles
Solution Approach 1:
The first principle model serves as an intermediary that mediates between the data-driven machine learning model and the physical system. It acts as a constraint and guide to ensure that the optimization solutions learned from historical data remain consistent with fundamental physical laws. The hybrid framework uses the first principle model to validate and correct predictions, ensuring physical plausibility while maintaining the learning capability from historical data.
3Ease of manufacture
If purely empirical relationships are used for optimization, then the system can be simpler to implement, but the set points generated are suboptimal for controlling production processes
Solution Approach 1:
The patent transforms the optimization approach by changing the parameters and structure of the modeling framework from purely empirical to a hybrid structure. This parameter change involves integrating first principle constraints into the optimization objective function and constraints, thereby improving the effectiveness of control actions and productivity while maintaining implementability through systematic mathematical formulation.
Data Source
AI summary
A computer implemented method for controlling determining an optimization solution for controlling a physical system. An optimization model is formed using an objective function and a set of constraints, a machine learning model that predicts a target value for a target variable for a physical system in response to receiving inputs for input variables for the physical system, and a first principle model that predicts the target value for the target variable for the physical system in response to receiving the inputs for the input variables for the physical system. Set points are determined for the input variables for an extremum for the target value for the target variable in the optimization model using regions in which the set points result in agreement between the target value predicted by the machine learning model and the target value predicted by the first principle model. The set points form the optimization solution.