Model Predictive Control Using Deep Causal Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current model predictive control (MPC) methods lack precision in capturing system uncertainties and interactions between subsystems, leading to sub-optimal process control decisions due to linear approximations and inability to quantify variance in control moves.
Innovation Solution
Deep Causal Learning (DCL) introduces randomized controlled signals to subsystems, computes confidence intervals for causal relationships, and uses time-varying Jacobian and Hessian matrices to predict optimal control moves, accounting for non-linearities and temporal dynamics, thereby refining the internal model for MPC.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional linear approximations are used in MPC, then computational complexity is reduced, but measurement precision and manufacturing precision of system predictions deteriorate
Solution Approach 1:
The patent transforms the linear approximation parameters into time-varying parameters that adapt to system dynamics. By using time-varying Jacobian and Hessian matrices instead of constant linear approximations, the system maintains computational tractability while significantly improving prediction precision for non-linear and dynamic systems.
Solution Approach 2:
The patent introduces dynamic elements by using time-varying matrices that capture changing system characteristics over time. The internal model transitions from static linear approximations to dynamic representations that adapt to temporal variations in system behavior, improving precision without excessive computational burden through efficient temporal modeling.
2Ease of operation
If traditional MPC methods are used, then implementation simplicity is maintained, but the ability to quantify variance and uncertainties in control moves deteriorates
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously monitors performance and uses computed confidence intervals to adjust control moves. The internal model incorporates uncertainty information from randomized controlled signals, creating a feedback loop that quantifies and responds to variances while maintaining operational simplicity through automated confidence-based decision making.
3Measurement precision
If the internal model complexity is increased to capture non-linearities, then prediction accuracy improves, but computational burden and device complexity increase
Solution Approach 1:
The patent applies partial linearization by using time-varying Jacobian and Hessian matrices that capture essential non-linear characteristics without requiring full non-linear model complexity. This partial action approach provides sufficient prediction accuracy for control purposes while avoiding the excessive computational burden of complete non-linear modeling.
Data Source
AI summary
Method for predictive control of a system having subsystems. The method includes providing signal injections relating to performance of the system. The signal injections include various operational controls for the system or its subsystems. Response signals corresponding with the signal injections are received, and a utility of those signals is measured. Based upon the utility of the response signals, data relating to operational controls is modified to optimize performance of the system via its subsystems.