System Identification
Graduate system identification: building models from data with least squares, prediction error methods, recursive estimation and model validation.
Building a mathematical model from observations of a system’s behavior is the subject of this graduate class. It reviews linear systems and stochastic processes, then treats nonparametric identification, input design and least squares. It goes on to prediction error methods, their convergence and asymptotic distribution, instrumental variables, closed loop data, recursive estimation and practical model validation.
What You Will Achieve
- Estimate models with least squares and analyze their statistics
- Design inputs with persistence of excitation
- Apply prediction error methods and maximum likelihood
- Estimate model order and validate a model structure
Program Overview
Foundations and Inputs
The first lectures set the framework.
- Review of linear systems
- Nonparametric identification
- Input design
- Least squares
Parameter Estimation
The central lectures develop the main theory.
- Parametrized model structures
- Minimum prediction error paradigm
- Convergence and consistency
- Asymptotic distribution of PEM
Methods in Practice
The last lectures turn to use.
- Instrumental variable methods
- Identification in closed loop
- Recursive estimation
- Model structure validation
Course Completion
Job Outlook
These methods are used in control engineering, signal processing and other fields that build models from measured data.
Good for: Graduate engineering students who have completed MIT’s 6.241 and 6.432 or similar classes.
Keep in mind: You need Ljung’s textbook. The syllabus outline mentions adaptive control and worst case identification, which are not on the lecture schedule. The material is highly mathematical.
Bottom line: A mathematical companion to Ljung’s book, with notes and problem sets that have solutions.
Frequently Asked Questions
Which textbook goes with it?
System Identification: A Theory for the User by Lennart Ljung, 2nd edition, published by Prentice Hall in 1998.
How can I practice?
Five problem sets are published together with solutions, spread across the 24 lectures, and lecture notes are available.
Does it cover Kalman filters?
The description and outline relate recursive estimation to Kalman filters, under the heading of a Kalman filter interpretation.



There are no reviews yet.