303 research outputs found
Evolving Takagi-Sugeno fuzzy systems from data streams (eTS+).
No full textIt is a well known fact that nowadays we are faced with not only large data sets that we need to process quickly, but with huge data streams (Domingos and Hulten, 2001). Special requirements are also placed by the fast growing sector of autonomous systems where systems that can re-train and adapt ‘on-fly’ are required (Patchett and Sastri, 2007). Similar requirements are enforced by the advanced process industries for self-developing and self-maintaining sensors (Qin et al., 1997). Now they even talk about self-learning industries (EC, 2007). All of these requirements cannot be met by using off-line methods and systems that can only adjust their parameters and/or are linear (Astroem and Wittenmark, 1989). These requirements call for a new type of systems that assumes the structure of non-linear, non-stationary systems to be adaptive and flexible. The author of this chapter started research work in this direction around the turn of the century (Angelov and Buswell, 2001; Angelov, 2002) and this research culminated in proposing with Dr. D. Filev the so called evolving Takagi-Sugeno (eTS) fuzzy system (Angelov and Filev, 2003). Since then a number of improvements of the original algorithm has been done, which require a systematic description in one publication. In this chapter an enhanced version of the eTS algorithm will be described which is called eTS+. It has been tested on a data stream from real engine test bench (data provided courtesy of Dr. E. Lughofer, Linz, Austria). The results demonstrate the superiority of the proposed enhanced approach for modeling real data stream in precision, simplicity and interpretability, and computational resources used. (c) IEEE Press and John Wiley and Son
Evolving intelligent systems : methodology and applications.
No full textThe newly established concept of evolving intelligent systems is a result of the synergy between conventional systems, neural networks and fuzzy systems as structures for information representation and the real time methods for machine learning. It targets non-stationary processes by developing novel on-line learning methods and computationally efficient algorithms for real-time applications. Fuzzy systems have demonstrated the ability to formalize in a computationally efficient manner the approximate reasoning typical of humans. Neural Networks present a convenient framework for synthesis and analysis of complex non-linear systems. Neuro-fuzzy systems combine the advantages of both areas and often use established machine learning methods for design. One of the important research challenges today is to further develop the intelligent systems theory towards the design of truly intelligent systems with a higher level of flexibility and autonomy that can develop their understanding of the environment and ultimately their intelligence. To address the problems of modelling, control, prediction, classification and data processing in such environments a system must be able to fully adapt its structure rather than adjust its parameters based on a pre-trained and fixed structure. That is, the system must be able to evolve, to self-develop, to self-organize. Evolving intelligent systems are inspired by the idea of system model evolution. They focus on the evolution of an open family of system models representing the system in different situations and operating conditions. In this sense they differ from the traditional evolutionary algorithms. They use inheritance and gradual change with the aim of life-long learning and adaptation, self-organization (including system structure evolution) in order to adapt to the (unknown and unpredictable) environment. Embedded soft computing diagnostics and prognostics algorithms, intelligent agents and controllers are the natural implementation area of evolving systems as a realistic and practical tool for design of real time intelligent systems
Evolving Fuzzy Systems, Proc. of the 2006 International Symposium on Evolving Fuzzy Systems EFS'06.
No full textA key characteristic of intelligent systems is their ability to deduce new knowledge, to predict and make decisions. The theoretical underpinning of these capabilities mainly resides in approximate reasoning, which itself is based on fuzzy logic and fuzzy sets. Fuzzy systems have demonstrated the ability to formalize in a computationally efficient manner the approximate reasoning typical of humans. The main challenge today is to design the next generation of intelligent systems able to have a higher level of flexibility and autonomy that can develop their understanding of the environment and ultimately their intelligence. It is to be noticed that the environments in which such systems are required to successfully operate are very often challenging – they are non-stationary, (often unpredictably) changing, and partially or completely unknown. To address the problems of modeling, control, prediction, classification and data processing in such environments a system model must be able to fully adapt, not simply to adjust parameters of a pre-trained and fixed structure. That is, the system model must be able to evolve, to self-develop, to self-organize. A new area that is addressing these challenges is now emerging on the cross-roads of reasoning-based fuzzy systems and evolution-inspired principles of adaptation, life-long learning, self-development and self-organization. It aims to develop systems that are more flexible than conventional adaptive systems that usually assume linearity and fixed structures of the underlying models. The emerging area of evolving fuzzy systems targets non-stationary processes by developing novel on-line learning methods and computationally efficient algorithms for real-time applications. Evolving fuzzy systems are evolution-inspired. They focus on the evolution of individual fuzzy systems. They use inheritance and gradual change with the aim of life-long learning and adaptation, as well as self-organization (including system structure evolution) in order to adapt to the (unknown and unpredictable) environment. This complements well with the established area of genetic fuzzy systems (GFS), that uses techniques based on population-based evolutionary and bio-inspired algorithms (genetic algorithms, genetic programming, evolution strategies, particle swarm optimization methods, bio-memetic approaches, ant colony optimization, etc.) to design fuzzy systems. These two topics make the methodological body of the present Symposium, which is the second in the series of events organized by the GFS Task Force from the Fuzzy Systems Technical Committee, IEEE Computational Intelligence Society. The present Symposium is organized by the Department of Communication Systems, Infolab21, Faculty of Science and Technology, Lancaster University, and technically co-sponsored by the Computational Intelligence Society and Systems, Man, and Cybernetics Society, IEEE, by the International Fuzzy Systems Association (IFSA), and by the European Society on Fuzzy Logic and Technology (EUSFLAT). The Symposium was also generously sponsored by EPSRC-UK arranging a number of student and young researchers’ grants that made possible attendance for a number of talented young people, and co-sponsored by Nokia-UK, BAE Systems, Retail Analytics, and J&S Marine offering a range of ‘best paper’ awards. This volume collects 55 full papers by 122 authors from 23 countries from all continents that have been accepted after being evaluated by at least three independent referees from the International Program Committee. All together, 72 papers authored by 153 authors from 26 countries were submitted and all of them were subjected to anonymous peer review process. The rejection rate of 24% (due to quality or irrelevance) is one of the illustrations of the high quality. The overall number of authors who have contributed to the symposium is another illustration of the International status of the event. The presence of several IEEE Fellows and a number of Senior Members and IEEE Technical Committees members is another fact in support of this qualification. Bearing in mind that this is only the second event of this series, the significant industrial participation (General Electric, USA, Ford Motor Co., USA, Nokia-UK, CEPSA, Spain etc.) illustrates the applicability of the methodologies that are discussed at the Symposium. The structure of the Symposium is composed of two main parts and eleven sessions: i) methodology-based sessions (covering evolving Takagi-Sugeno fuzzy models, evolving neuro-fuzzy systems, evolving fuzzy clustering, evolutionary fuzzy systems, genetic fuzzy systems); and ii) applications-oriented sessions (including industrial applications of real-time evolving fuzzy systems, evolving neuro-fuzzy systems, genetic fuzzy systems, developments in fuzzy systems and a session comprising papers that will discuss and explore the frontiers of computational intelligence and will pose some challenging questions. The discussion will take place in addition to scheduled sessions and networking events during the Round Table. (c) IEEE Pres
Architectures of evolving fuzzy rule-based classifiers
Get PDFPrediction of the properties of the crude oil distillation side streams based on statistical methods and laboratory-based analysis has been around for decades. However, there are still many problems with the existing estimators that require a development of new techniques especially for an on-line analysis of the quality of the distillation process. The nature of non-linear characteristics of the refinery process, the variety of properties to measure and control and the narrow window that normally refinery processes operates in are only some of the problems that a prediction technique should deal with in order to be useful for a practical application. There are many successful application cases that refinery units use real plant data to calibrate models. They can be used to predict quality properties of the gas oil, naphtha, kerosene and other products of a crude oil distillation tower. Some of these are distillation end points and cold properties (freeze, cloud). However, it is difficult to identify, control or compensate the dynamic process behaviour and the errors from instrumentation for an online model prediction
Evolving intelligent sensors in chemical industry.
No full textThis chapter presents a new promising technique for design of inferential sensors in chemical process industry which has a broad range of applicability. It is based on the concept of evolving fuzzy rule-based systems (EFS). Mathematical modeling was used to infer difficult to measure otherwise variables such as product quality since 1980s, including in on-line mode. The challenge today is to develop such adaptive, flexible, self-calibrating on-line inferential sensors that reduce the maintenance costs while keeping high precision and interpretability/transparency. The methodology of fuzzy rule-based models of Takagi-Sugeno type (Takagi-Sugeno, 1985) which have flexible, open structure and are therefore called ‘evolving’ (see also chapter 2) is particularly suitable for addressing this challenge. The basic concept is described from the point of view of the implementation of this technique to self-maintaining and self-calibrating inferential sensors for several chemical industry processes. The sensitivity analysis (input variables selection) was performed on-line and this was compared to the off-line input variables selection using genetic programming (GP). A case study based on four different inferential sensors for estimating chemical properties is presented in more detail, while the methodology and conclusions are valid for the broader area of chemical and process industry in general. The results demonstrate that well interpretable and with simple structure inferential sensors can be designed automatically from the data stream in real-time that provide estimation of the real values of process variables of interest. The proposed approach can be used as a basis for development of a new generation of inferential sensors that can address the challenges of the modern advanced process industry. (c) IEEE Press and John Wiley and Son
An approach to online identification of Takagi-Sugeno fuzzy models
Get PDFAn approach to the online learning of Takagi-Sugeno (TS) type models is proposed in the paper. It is based on a novel learning algorithm that recursively updates TS model structure and parameters by combining supervised and unsupervised learning. The rule-base and parameters of the TS model continually evolve by adding new rules with more summarization power and by modifying existing rules and parameters. In this way, the rule-base structure is inherited and up-dated when new data become available. By applying this learning concept to the TS model we arrive at a new type adaptive model called the Evolving Takagi-Sugeno model (ETS). The adaptive nature of these evolving TS models in combination with the highly transparent and compact form of fuzzy rules makes them a promising candidate for online modeling and control of complex processes, competitive to neural networks. The approach has been tested on data from an air-conditioning installation serving a real building. The results illustrate the viability and efficiency of the approach. The proposed concept, however, has significantly wider implications in a number of fields, including adaptive nonlinear control, fault detection and diagnostics, performance analysis, forecasting, knowledge extraction, robotics, behavior modeling
- …
