Nonlinear System Identification: From Classical Approaches to Neural Networks and Fuzzy Models
Written from an engineering point of view, this book covers the most common and important approaches for the identification of nonlinear static and dynamic systems. The book also provides the reader with the necessary background on optimization techniques, making it fully self-contained. The new edition includes exercises.
Why Read This Book
You should read this book if you need a practical, engineering-focused guide to modeling and estimating nonlinear systems: it walks you from classical series and block-structured approaches through modern black‑box methods such as neural networks and fuzzy models. You will get applied guidance on optimization, model selection and validation so you can build reliable nonlinear models for DSP, control, or communications problems.
Who Will Benefit
Engineers and researchers with some DSP or systems-background who must build or validate nonlinear models for signals, control systems, or communications receivers.
Level: Advanced — Prerequisites: Undergraduate-level signals & systems, linear system identification basics, linear algebra and probability; familiarity with optimization concepts is helpful (the book provides a review).
Key Takeaways
- Formulate nonlinear identification problems and choose appropriate model structures (Volterra, Hammerstein–Wiener, block-structured, NARX/NARMAX).
- Apply parametric estimation and optimization techniques (least squares, prediction error methods, regularization) to fit nonlinear models.
- Design and validate identification experiments and model validation tests to assess model fitness and generalization.
- Implement black-box modeling using neural networks and fuzzy models and understand their strengths/tradeoffs for system identification.
- Analyze model structure selection, complexity control and methods to avoid overfitting in nonlinear contexts.
Topics Covered
- Introduction and problem statement
- Mathematical and optimization preliminaries
- Linear identification review and extensions
- Static nonlinear modeling and regression methods
- Volterra series and polynomial-based models
- Block-structured models: Hammerstein, Wiener and Wiener–Hammerstein
- Dynamic black-box models: NARX/NARMAX approaches
- Neural network methods for system identification
- Fuzzy modelling and neuro-fuzzy approaches
- Parameter estimation and numerical optimization techniques
- Model validation, experiment design and noise handling
- Regularization, model selection and practical considerations
- Examples, exercises and case studies
Languages, Platforms & Tools
How It Compares
Covers nonlinear model families and practical estimation in more application-focused detail than Ljung's System Identification (which is broader and more theory-heavy for linear models), and is complementary to Billings' NARMAX-focused texts which dive deeper into that specific family.
