Data-Based Controller Configuration for Real-Time MPC Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Efficient operational control of complex technical systems, such as process-engineering plants and power plants, is challenging due to high computational demands in existing control methods, particularly for systems requiring model-predictive control.
Innovation Solution
A method that configures a data-based control model using a model-predictive control model to set configuration parameters, allowing the data-based control model to reproduce its output, thereby reducing computational effort and covering a larger state space, including rare but critical scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If model-predictive control is used for complex technical systems, then optimized control parameters can be determined, but computational effort becomes too high for operational control
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network controller offline using model-predictive control calculations. The neural network learns the optimal control strategy in advance during system commissioning, so that during operational control, no complex real-time calculations are needed. This transfers the computational burden from runtime to setup phase, enabling fast operational control while maintaining optimized control parameters.
2Adaptability or versatility
If traditional control methods are used, then control can be maintained, but the system cannot cover rare but critical scenarios effectively
Solution Approach 1:
The patent implements feedback by using the neural network's ability to learn from training data that includes rare and critical scenarios. During the offline training phase, the system processes diverse state space data including edge cases, and the neural network adjusts its weights to capture these patterns. During operation, the trained network provides reliable control even for rare scenarios because it has learned from them during training, effectively incorporating feedback from comprehensive scenario analysis.
Data Source
AI summary
A method and a device for configuring a controller and to a method and a controller for controlling a technical system by means of a data-based control model is provided, in particular a model based on reinforcement learning. This data-based control model is configured using a model-predictive control model. Configuration parameters of the data-based control model are set by mapping the model-predictive control model onto the data-based control model in such a way that the data-based control model reproduces the output data of the model predictive control model depending on state data of the technical system read in, and determines optimized control parameters configured in this way. A computationally intensive training procedure for configuring the data-based control model can thus be avoided.