Hybrid Dynamic Model for Technical System Load Change Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Dynamic processes in complex technical systems, such as start-up and shutdown processes and load changes, are difficult to model with computer support due to insufficient understanding of the systems and complex mathematical equations, leading to high manual effort and economic unviability, which limits the use of advantages like virtual commissioning and virtual sensors.
Innovation Solution
A hybrid computer-aided dynamic model combining a physical simulation model and a data-driven model, trained using time-dependent process variables, to reproduce second process variables and control start-up, shutdown, and load changes, incorporating expert knowledge and reducing the need for extensive measurement data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a purely physical simulation model is used to model dynamic processes, then modeling accuracy can be achieved, but the modeling effort and complexity increase significantly
Solution Approach 1:
The patent combines a physical simulation model with a data-driven model (neural network) into a hybrid model. The physical model provides the structural framework while the neural network learns from measurement data to capture complex dynamic behaviors, achieving high accuracy without requiring complete physical understanding of all system aspects.
Solution Approach 2:
The neural network acts as an intermediary that bridges the gap between the simplified physical model and the actual complex system behavior. It processes measurement data and supplements the physical model's predictions, allowing the system to achieve high accuracy without requiring a complete and complex physical model.
2Ease of manufacture
If a purely data-driven model is used to predict process variables, then less expert knowledge is required, but the amount of measurement data needed increases significantly
Solution Approach 1:
The hybrid model merges the advantages of both physical modeling (requiring less data) and data-driven modeling (requiring less expert knowledge). The neural network component learns patterns from relatively small amounts of measurement data while the physical model provides the structural constraints and reduces the dimensionality of the learning problem.
Solution Approach 2:
The model applies different approaches to different aspects of the system: the physical model handles aspects where theoretical understanding exists and data requirements would be high, while the neural network handles aspects where data is available but physical understanding is incomplete, optimizing the data-to-knowledge ratio.
3Measurement precision
If complex mathematical equations are used to model technical systems, then modeling precision can be improved, but the manual effort and economic viability deteriorate
Solution Approach 1:
The patent replaces complex analytical mathematical modeling with a computational approach using neural networks that learn system behavior directly from data. This substitution of traditional mechanical/mathematical modeling methods with data-driven computational methods achieves high prediction accuracy while dramatically reducing the manual effort and expert knowledge required.
Solution Approach 2:
The neural network automatically learns the system's dynamic behavior patterns from measurement data without requiring manual derivation of complex mathematical equations. The model trains itself on the available data, eliminating the need for extensive manual modeling work while achieving high prediction accuracy.
Data Source
Figure 1~4
AI summary
The invention relates to a device (100) and a computer-implemented method for controlling a start-up and/or shutdown process and/or load change of a technical system (TS). The device comprises: - an interface (101) configured to provide a computer-based dynamic model (CM), wherein the computer-based dynamic model comprises a physical dynamic simulation model (SIM) and a data-driven model (NN), wherein the physical dynamic simulation model models only one aspect of the dynamic processes during the start-up and/or shutdown process and/or load change of the technical system, and the data-driven model models the remaining aspects of the dynamic processes, and wherein the computer-based dynamic model is trained using time-dependent process variables of the technical system toto reproduce second process variables of the technical system depending on a first process variable, - a sensor (102) configured to detect a process variable of the technical system, - a forecasting module (103) configured to predict further process variables of the technical system depending on the detected process variable by executing the computer-aided, dynamic model, and - an output module (104) configured to output the predicted process variables for controlling a start and/or shutdown process and/or load change of the technical system.