Hybrid Control Objective Integration for Interpretable Expert Behavior
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control systems either rely solely on explicit performance criteria or expert models, leading to limitations such as insufficient insight into control actions and failure to capture deep performance measures or natural human behavior.
Innovation Solution
A control objective integration system that combines an expert model unit, a transformer, and a combiner to generate predicted expert control actions and compute an optimal set of weights for an aggregated cost function, balancing explicit performance optimization and expert behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If only expert models are used to generate control actions, then the system can capture natural human behavior, but the system lacks interpretability and may not provide sufficient insight into why control actions were generated
Solution Approach 1:
The patent combines expert models with explicit performance criteria into a unified control system. The controller integrates both approaches by using the expert model to generate predicted control actions and simultaneously using explicit performance criteria to evaluate and interpret these actions, thereby maintaining interpretability while capturing expert behavior.
Solution Approach 2:
The patent introduces explicit performance criteria as an intermediary between the expert model and the control output. This intermediary layer provides interpretability by translating the black-box expert model predictions into understandable performance metrics and reasons for control actions.
2Loss of information
If only explicit performance criteria are used for control optimization, then the system provides interpretability and clear performance measures, but the system fails to fully capture natural expert behavior
Solution Approach 1:
The patent merges explicit performance criteria with expert models to create a hybrid control system. The explicit performance criteria provide interpretability and structured optimization, while the expert models contribute natural human behavior patterns, achieving both goals simultaneously.
Solution Approach 2:
The control system uses a composite approach by combining two different control paradigms (explicit performance-based control and expert-model-based control) into a unified framework, similar to how composite materials combine different materials to achieve properties that neither material alone could provide.
3Adaptability or versatility
If the system integrates both expert models and explicit performance criteria, then the system can balance performance optimization and expert behavior, but the system complexity increases
Solution Approach 1:
The patent segments the control system into distinct functional modules: an expert model unit for generating predicted control actions, a performance evaluation unit for assessing these actions against explicit criteria, and an integration unit for combining both approaches. This modular segmentation manages complexity by organizing the integrated system into manageable, independent components.
Data Source
AI summary
An expert model unit 81 generates predicted expert control actions based on an expert model which is a machine learning model trained using data collected when an expert operated a plant which is a control target or a plant of the same or similar characteristics. A transformer 82 constructs metrics or error measures involving the predicted expert control actions from the expert model unit 81 as an objective term. A combiner 83 collects different objective terms from the transformer 82 and a learner which outputs machine-learning models as objective terms and computes an optimal set of weights or combinations of the objective terms to construct an aggregated cost function for use in an optimizer.