Hybrid State Inference for Physical System Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining the state of a physical system from sensor data, such as Kalman filtering, face limitations in accuracy due to noise and complexity in modeling dynamics, especially when dealing with non-linear and complex systems.
Innovation Solution
A hybrid approach combining a prior knowledge-based mathematical model with a learned model, where the learned model corrects the initial inference from the mathematical model using ground truth data, allowing for improved state inference accuracy by refining the state estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Kalman filtering is used to estimate state from sensor measurements, then the state can be recursively estimated based on time-series data, but the accuracy is limited due to noise and uncertainty in measurements
Solution Approach 1:
The patent combines a mathematical model representing prior knowledge with a learned model (neural network) to create a hybrid state estimation system. The mathematical model provides physics-based constraints and relationships, while the learned model captures complex patterns from data, together improving estimation accuracy beyond what either model could achieve alone
Solution Approach 2:
The learned model acts as an intermediary between the mathematical model and sensor measurements. It processes the relationship between observed measurements and inferred states, learning complex mappings that bridge the gap between simplified mathematical models and real-world sensor data with noise and uncertainties
2Measurement precision
If a learned model is used to correct initial inference from mathematical model, then state inference accuracy is improved, but the system requires training data and computational resources
Solution Approach 1:
Instead of training a complete learned model from scratch to handle all aspects of state estimation, the patent uses the learned model only for correction of initial mathematical model inferences. This partial approach requires less training data and computational resources while still achieving improved accuracy by focusing the learning process on correcting specific errors
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system (100) and computer-implemented method are provided for enabling control of a physical system (60) based on a state of the physical system which is inferred from sensor data (122). The system and method may iteratively infer the state by, in an iteration, obtaining an initial inference of the state using a mathematical model representing a prior knowledge-based modelling of the state, and by applying a learned model to the initial inference of the state and the sensor measurement, wherein the learned model has been learned to minimize an error between initial inferences provided by the mathematical model and a ground truth and to provide a correction value as output for correcting the initial inference of the state of the mathematical model. Output data (142) may be provided to an output device (40) to enable control of the physical system based on the inferred state.