Computational intelligence in time series forecasting. Theory and engineering applications (Q2576561)

This is a monograph whose aim is of special and singular interest: to present systematic and comprehensive methods and techniques of computational intelligence and soft computing for solving forecasting and prediction problems of various types (e.g., nonlinear, multivariable, seasonal, and chaotic) of time series. The book is designed to be largely self-contained and is devoted to offer researchers, practicing engineers, and applications-oriented professionals a reference volume and a valuable guide for the design, building and execution of forecasting and prediction experiments, from the collection and structuring of time-series data up to the evaluation of experimental results. The entire monograph is sensibly structured in four parts, containing ten chapters. Part I (Introduction) of the book consists in the first two chapters of the objective of introducing the reader to the evolution and computational intelligence methods, and to the traditional formulations and approaches of time series forecasting problems. Chapter 1 (Computational Intelligence: An Introduction) shows how the field of computational intelligence has grown up from fuzzy logic, neuro-computing and probabilistic reasoning (as postulated by L. Zadeh since 1994) with the addition of genetic algorithms (GAs), genetic programming, evolutionary strategies, and evolutionary programming. Particular attention is paid to hybrid computational intelligence, which deals with parameter tuning of fuzzy systems using neural networks, performance optimization of neural networks through monitoring, and parameter adaptation by fuzzy logic systems. The application fields of computational intelligence today are surveyed. Chapter 2 (Traditional Problem Definition) is devoted to presenting the classical definition and solving of the time-series forecasting problem. The main characteristic features of time series, their classification, and the problem of time series modeling are discussed, including linear regression-based time series models such as ARMA, ARIMA, CARIMA, as well as multivariate, nonlinear, and chaotic time series models. The core part of the chapter deals with the forecasting approaches of time series based on Box-Jenkins methods and the approaches using exponential smoothing, adaptive smoothing, and nonlinear combination of forecasts, illustrated with an example of control engineering in industry. Part II (Basic Intelligent Computational Technologies) encloses Chapters 3--5. In Chapter 3 (Neural Networks Approach) the neuro-technology and forecasting technology is described that comprise data preparation, determination of network architecture, training strategy, training stopping, validation, etc. Advanced use of neural networks in combination with traditional approaches and nonlinear combination of forecasts are performed. Chapter 4 (Fuzzy Logic Approach) provides the foundations of fuzzy logic methodology and its application to fuzzy modelling on examples of building the Mamdani, relational, singleton, and Takagi-Sugeno models that are suitable for time series modelling and forecasting. Issues such as optimal shaping of membership functions, automatic rule generation and building a non-redundant and conflict-free rule database are pointed out, and examples involving chaotic time series forecasting, modeling and prediction of second-order nonlinear plant outputs using fuzzy logic systems, and temperature prediction in a chemical reactor are discussed. Chapter 5 (Evolutionary Computation) presents the main approaches of evolutionary computations and intelligent optimal solution search algorithms: genetic algorithms (GAs), genetic programming, evolutionary strategies, evolutionary programming, and differential evolution. Part III (Hybrid Computational Technologies) deals with various combinations of basic computational technologies that can work in a cooperative way. Chapter 6 (Neuro-fuzzy Approach) describes the combination of neuro and fuzzy logic technology, with two major issues: the training of typical neuro-fuzzy networks, and their application to modelling nonlinear dynamic systems. Forecasting examples are given from industrial practice: short-term forecasting of electrical load, prediction of material properties, correction of pyrometer readings, as well as prediction of Wang data and chaotic time series. Chapter 7 (Transparent Fuzzy / Neuro-fuzzy Modelling) investigates mainly the topics of model transparency and the interpretability of data-driven automated fuzzy models. Strong emphasis is placed on compact and transparent modelling schemes that include model structure solutions, data clustering, similarity-based simplification, and model visualization. The formal technique of generating the ``white-box''-like model is proposed, in contrast to the black-box model generated by a neural network. Chapter 8 (Evolving Neural and Fuzzy Systems) covers the application of GAs and evolutionary programming in the evolution design of neural networks and fuzzy systems. The chapter focuses on the optimal selection of fuzzy rules and the optimal shaping of membership function parameters when evolving fuzzy logic systems. Chapter 9 (Adaptive Genetic Algorithms) deals with the problem of adaptation of GAs using fuzzy logic systems for optimal selection and tuning of genetic operators, parameters, and fitness functions. An example of dynamically controlled GAs using a rule-based expert system with a fuzzy government module for tuning the GA parameters is given. Part IV (Recent Developments) and its (single) Chapter 10 (State of the Art and Development Trends) introduces the reader to more recently developed computationally intelligent technologies, such as support vector machines, wavelet and fractal networks, the entropy and Kohonen networks-based fuzzy clustering approaches, the design of Takagi-Sugeno fuzzy models, etc. Development trends of computational intelligence technologies are outlined: advanced bioinformatics, swarm engineering, multi-agent systems, and fuzzy-logic-based data understanding.