Safe Dynamic Model Training With Gaussian Process Exploration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Active learning for time series models representing physical systems faces challenges in safely exploring dynamic processes without damaging the systems, as existing methods lack effective criteria for balancing information gain and safety during dynamic stimulation.
Innovation Solution
The method combines dynamic, active, and safe exploration by using Gaussian processes with a non-linear exogenous structure, parameterizing input curves as successive sections of ramp or step functions, and solving an optimization problem with secondary safety conditions to maximize information gain while ensuring system safety.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If dynamic stimulation is applied to explore the input area of a physical system, then information gain about dynamic processes is improved, but the risk of damaging the physical system increases
Solution Approach 1:
A safety model is trained beforehand to predict unsafe input areas before actual exploration occurs. This preliminary safety assessment allows the active learning process to avoid dangerous regions proactively, enabling aggressive exploration in safe regions without risking system damage.
Solution Approach 2:
The safety model acts as an intermediary between the active learning process and the physical system. It receives exploration requests, evaluates them for safety, and only permits safe inputs to reach the physical system, thus mediating the conflict between information gathering and system protection.
2Measurement precision
If selective data identification is performed in a static state, then measurement precision is improved, but the ability to detect dynamic processes deteriorates
Solution Approach 1:
The system transitions from static to dynamic exploration by sequentially determining curve sections based on previous observations. The input curve is adapted dynamically through an explorative approach that responds to system behavior, allowing both precise measurement and dynamic process detection.
Solution Approach 2:
The exploration process continues continuously by appending successive curve sections to form a complete input curve. This continuous explorative action maintains both measurement precision through selective data identification and dynamic process detection through ongoing adaptation to system responses.
3Measurement precision
If more pieces of information are collected from the physical system, then model accuracy is improved, but the cost and risk of measurements increase
Solution Approach 1:
The safety model extracts and isolates the safety assessment function from the main active learning process. By separating safety evaluation, the system can efficiently identify and avoid unsafe measurement points without compromising model accuracy, reducing unnecessary measurement costs and risks.
Solution Approach 2:
The system changes the parameterization of input curves from fixed pre-defined curves to dynamically determined successive sections. This parameter change allows the exploration to adapt to system responses, collecting maximum information with minimum measurements by focusing on informative regions identified through the safety model.
Data Source
AI summary
A computer-implemented method for the safe, active training of a computer-aided model for modeling time series of a physical system using Gaussian processes, including the steps of establishing a safety threshold value α; initializing by implementing safe initial curves as input values on the system, creating an initial regression model and an initial safety model; repeatedly carrying out the steps of updating the regression model; updating the safety model; determining a new curve section; implementing the determined new curve section on the physical system and measuring output variables; incorporating the new output values in the regression model and the safety model until N passes have been carried out; and outputting the regression model and the safety model.