System Identification: Theory for the User (Prentice Hall Information and System Sciences Series)
Appropriate for courses in System Identification. This book is a comprehensive and coherent description of the theory, methodology and practice of System Identification-the science of building mathematical models of dynamic systems by observing input/output data. It puts the user in focus, giving the necessary background to understand theoretical foundation and emphasizing the practical aspects of the options and choices that face the user. The Second Edition has been updated to include material on subspace methods, non-linear black box models-such as neural networks-and methods that use frequency domain data.
Why Read This Book
You should read this book if you need a rigorous but practical guide to building mathematical models of dynamic systems from input/output data; it explains when to choose different model structures and estimation methods and shows how to validate and interpret results. You will get working knowledge of prediction-error methods, subspace identification, frequency-domain techniques, and nonlinear black-box approaches (including neural networks) with an emphasis on how to apply them in real situations.
Who Will Benefit
Graduate students, signal-processing/control engineers, and researchers who need to identify linear and nonlinear dynamic models from data for estimation, prediction, or controller design.
Level: Advanced — Prerequisites: Undergraduate-level signals & systems, linear algebra, probability & stochastic processes; familiarity with MATLAB is strongly recommended.
Key Takeaways
- Formulate system identification problems and choose appropriate model structures (ARX/ARMAX/OE/BJ, state-space).
- Apply prediction-error methods to estimate parameters and understand their statistical properties.
- Use subspace identification techniques to obtain state-space models directly from time-series data.
- Validate models and design experiments to ensure identifiability and adequate model fidelity.
- Implement and evaluate nonlinear black-box models (e.g., neural networks) and frequency-domain identification methods.
- Assess estimation uncertainty and apply asymptotic theory for hypothesis testing and confidence intervals.
Topics Covered
- Introduction and basic concepts of system identification
- Problem formulation, data collection, and model structures
- Prediction-error methods and parameter estimation
- Classical linear model structures: ARX, ARMAX, OE, Box-Jenkins
- State-space models and identification
- Subspace identification methods
- Frequency-domain identification methods
- Nonlinear black-box models (neural networks, NARX, etc.)
- Model validation, residual analysis, and experiment design
- Statistical properties, asymptotic theory and model selection
- Practical implementation topics and algorithms (numerics)
- Appendices and MATLAB examples / System Identification Toolbox notes
Languages, Platforms & Tools
How It Compares
Ljung is more user-focused and balanced between theory and practice than Söderström & Stoica's mathematically heavy treatment, and it covers a broader set of topics (including prediction-error and nonlinear models) than the more narrowly focused Van Overschee & De Moor subspace text.
