Diese LVA soll die Teilnehmer befähigen, geeignete fortschrittliche Modellierungsansätze auszuwählen und einzusetzen, um genaue und effiziente Modelle für den späteren Einsatz modellbasierter Regelungsmethoden bereitzustellen.
Diese LVA wird in englischer Sprache abgehalten.
Beachten Sie das als Download verfügbare Info-Sheet!
In dieser LVA werden fortgeschrittene Methoden zur Modellbildung und Systemidentifikation für darauf aufbauende modellbasierte Regelung erarbeitet, wobei auf Anwendungen in der Ingenieurspraxis eingegangen wird.
The course will cover advanced model design and system identification techniques for modelling of dynamic systems in engineering practice. The paradigms presented are tailored in such a way to obtain models which can then be employed in model based control strategies. Different model based control methodologies such as model predictive control, internal model control, adaptive control, etc. embody different classes of models with various types of structures and parametrisations. Since the efficiency and quality of a model based control is strongly related to the accuracy of the internal model, it is therefore important to select a modelling technique which will generate the best possible model for particular application. Modelling approaches such as local model networks, Takagi-Sugeno fuzzy models, neural networks and support vector machines, as well as probabilistic models like Gaussian processes will be presented. Pros and cons of each of them will be highlighted. Machine learning and adaptive algorithms for structure identification and parametrisation will be presented in detail. Difficulties related to curse of dimensionality, rank deficiency, overfitting, bias-variance tradeoff and off-equilibrium dynamic will be pointed out and practical guidelines how to overcome them will be given. Model properties such as transparency and inevitability will also be discussed. Utilisation of modelling techniques presented, for model based control will be demonstrated on industrial examples taken from automotive applications and process control. Scientific computing, simulation and animation software will extensively be used in order to provide efficient and attractive presentation of course material.
Gutes Verständnis grundlegender regelungstechnischer Methoden ist erforderlich.
Vorerfahrungen in der Modellbildung, Identifikation, Simulation und Optimierung sind hilfreich; die Vorlesung wird aber an die spezifischen Vorkenntnisse der Teilnehmer anknüpfen.
