First edition of the Spring Doctoral School on
Data-driven Model Identification of Dynamical Systems
Nancy, France April, 3-6, 2017
The 4-days Spring Doctoral School consisted of a series of lectures followed by computer exercises.
April 3, 2017 - 9h00-17h00
Lecturer: Prof. Lennart Ljung (Linköping University, Sweden)
2-day course on System Identification: A Prediction Error Perspective
Overview
- Theme 1: Overview of the problem area. Overall principles. Basic model structures. Bias and Variance. Model quality and uncertainty. Model validation.
- Computer exercise 1. Getting hands on experience with real data. Simple models and model validation.
- Theme 2: More on Linear models. Frequency domain data. Subspace Techniques. New approaches with regularization. High order, regularized simple models, advantages and disadvantages.
- Computer exercise 2: Tests of subspace methods (N4SID, MOESP, CVA) and regularization. Experience of choice of model orders and auxiliary variables.
April 4, 2017 - 8h30-16h30
Lecturer: Prof. Lennart Ljung (Linköping University, Sweden)
- Theme 3: Non-linear model structures. Wiener and Hammerstein models, Neural Networks. Grey box models with physical structure.
- Computer exercise 3: Hands on identification of non-linear models.
- Theme 4: Special questions. Experiment design. Concluding remarks.
April 5, 2017 - 8h30-17h00
Lecturer: Prof. Hugues Garnier (University of Lorraine, France)
Data-driven approaches for direct continuous-time model identification
Overview
- Theme 1: Overview of continuous-time linear model identification. Basic SVF-based and optimal instrumental variable (IV)-based estimators. Practical and software aspects.
- Computer exercise 1. Getting hands on experience with the CONTSID toolbox methods with real data.
- Theme 2: Benefits for practical applications. IV extensions to advanced situations (MISO systems, time-delay systems, simple process models, LTV systems,...).
- Computer exercise 2: Getting hands on experience on advanced situations.
April 6, 2017 - 8h30-12h15
Lecturer: Prof. Marion Gilson (University of Lorraine, France)
Frequency-domain, closed-loop, Hammerstein and LPV model identification
Overview
- Theme 1: Frequency-domain identification. Closed-loop system identification. Instrumental variable-based techniques.
- Computer exercise 1: Hands on frequency-domain and closed-loop model identification.
- Theme 2: Hammerstein and Linear Parameter Varying (LPV) model identification.
- Computer exercise 2: Hands on identification of Hammerstein and LPV models.