Continuous-Time Machine Learning Models for Technical System Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for simulating technical systems using digital twins with machine learning fail to accurately represent continuous state transitions due to the use of discrete measurement points, which do not align with the underlying continuous state transitions in real-world systems.
Innovation Solution
A computer-implemented method for generating a trained machine learning model that incorporates time invariance during training, using extended training data tuples that include time parameters, allowing for the simulation of continuous state changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If discrete measurement points are used for training the machine learning model, then the training data is simplified and easier to process, but the simulation accuracy of continuous state transitions deteriorates
Solution Approach 1:
The patent transforms the discrete state parameters into continuous parameters by introducing time as a continuous variable. The state is no longer represented as discrete jumps between measurement points, but as a continuous function of time that passes through all intermediate values. This allows the model to simulate continuous state transitions while training on discrete measurement data.
Solution Approach 2:
The patent adds the time dimension to the state representation. Instead of modeling state transitions as discrete jumps between measurement points, the patent models the state as a continuous function evolving over time. This dimensional addition allows the system to capture continuous state changes while maintaining compatibility with discrete training data.
2Device complexity
If discrete state jumps are used to model system behavior, then the model complexity is reduced, but the reliability of simulating real-world continuous physical processes deteriorates
Solution Approach 1:
The patent changes the fundamental parameter representation from discrete state jumps to continuous state evolution. By treating the state as a continuous function of time rather than discrete transitions, the model maintains physical realism for continuous processes while using the same underlying machine learning architecture, thus not significantly increasing complexity.
Solution Approach 2:
The patent creates a continuous-time copy of the discrete measurement data by interpolating between measurement points. The neural network learns to generate continuous state trajectories that are consistent with the discrete measurement data, effectively creating a continuous version of the discrete data without requiring continuous physical sensors.
3Measurement precision
If time invariance is incorporated into the training process, then the simulation accuracy of continuous state changes is improved, but the training computational requirements increase
Solution Approach 1:
The patent performs preliminary data preparation by organizing the discrete measurement data into sequences with time stamps before training begins. The neural network architecture is pre-configured to handle continuous time inputs, and the training data is pre-processed to include temporal information. This preliminary setup enables efficient training by avoiding the need for complex runtime calculations to handle time invariance.
Data Source
Figure 1~3
AI summary
The invention relates to a computer-implemented method for generating a trained machine learning model for simulating the behavior of a technical system, comprising the steps: a. providing a plurality of training data tuples (S1); wherein each training data tuple of the plurality of training data tuples comprises a first data element of a first state of the technical system and an action; wherein the first state changes to a second state as a result of the action; b. training a machine learning model based on the plurality of training data tuples and taking time invariance into account (S2); and c. providing the trained machine learning model (S3). The invention further relates to a technical system and a corresponding computer program product.