Intelligent Control Systems using Computational Intelligence Techniques
Intelligent Control techniques are becoming important tools in both academia and industry. Methodologies developed in the field of soft-computing, such as neural networks, fuzzy systems and evolutionary computation, can lead to accommodation of more complex processes, improved performance and considerable time savings and cost reductions.
Inspec keywords: Gaussian processes; identification; optimal control; neural nets; nonlinear control systems; neurocontrollers; multi-agent systems; learning (artificial intelligence); fault diagnosis; intelligent control; nonlinear dynamical systems; adaptive control; evolutionary computation; fuzzy systems
Other keywords: multiagent control; neuro-fuzzy model construction; neural networks; computational intelligence techniques; nonlinear dynamical systems; reinforcement learning; multiobjective evolutionary computing solutions; nearly optimal control; intelligent control systems; nonlinear control; adaptive local linear modelling; fault diagnosis; nonlinear system identification; Gaussian process; nonlinear modelling; fuzzy systems
Subjects: Optimisation techniques; Control engineering computing; Simulation, modelling and identification; Expert systems and other AI software and techniques; Specific control systems; Optimal control; Other topics in statistics; Artificial intelligence (theory)
- Book DOI: 10.1049/PBCE070E
- Chapter DOI: 10.1049/PBCE070E
- ISBN: 9780863414893
- e-ISBN: 9781849190527
- Page count: 476
- Format: PDF
-
Front Matter
- + Show details - Hide details
-
p.
(1)
-
1 An overview of nonlinear identification and control with fuzzy systems
- + Show details - Hide details
-
p.
1
–36
(36)
This chapter gives an overview of system identification techniques for fuzzy models and some selected techniques for model-based fuzzy control. It starts with a brief discussion of the position of fuzzy modelling within the general nonlinear identification setting. The two most commonly used fuzzy models are the Mamdani model and the Takagi-Sugeno model. An overview of techniques for the data-driven construction of the latter model is given. We discuss both structure selection (input variables, representation of dynamics, number and type of membership functions) and parameter estimation (local and global estimation techniques, weighted least squares, multi-objective optimisation). Further, we discuss control design based on a fuzzy model of the process. As the model is assumed to be obtained through identification from sampled data, we focus on discrete-time methods, including: gain-scheduling and state-feedback design, model-inverse control and predictive control. A real-world application example is given.
-
2 An overview of nonlinear identification and control with neural networks
- + Show details - Hide details
-
p.
37
–88
(52)
This chapter has introduced the basic concepts related with the use of neural net works for nonlinear systems identification, and has briefly reviewed neuro-control approaches. Several important issues for the design of data-driven models, as neural networks are, such as data acquisition and the design of excitation signals could not be covered here, but the will be discussed in Chapter 5. On purpose, important topics such as neuro-fuzzy and local linear models were not discussed, as they will be treated in other chapters of this book. Neural network modelling is an iterative process, requiring, at the present stage, substantial skills and know ledge from the designer. It is our view that the methodology presented in the example, employing multi-objective evolutionary algorithms for the design of neural models, is a suitable tool to aid the designer in this task. It incorporates input model order and structure selection, as well as parameter estimation, providing the designer with a good number of well performing models with varying degrees of complexity. It also allows the incorporation of objectives which are specific for the ultimate use of the model. Through the analysis of the results obtained in one iteration, the search space can be reduced for future iterations, therefore allowing a more refined search in promising model regions.
-
3 Multi-objective evolutionary computing solutions for control and system identification
- + Show details - Hide details
-
p.
89
–118
(30)
Systems engineers are increasingly faced with problems for which traditional tools and techniques are ill suited. Such problems are posed by systems that are complex, uncertain and not conducive to deterministic analysis. Modern tools and techniques are being applied in these subject areas in order to address such problems. This chapter introduces the multi-objective genetic algorithm (MOGA) of Fonseca and Fleming [1] as one of the tools for addressing these problems. Background to genetic algorithms (GAs) is given in terms of their operators and abstraction of the problem domain. Multi-objective optimisation is introduced using the concept of Pareto dominance and trade-off between competing objectives. Visualisation techniques are illustrated in which 'many-objective' problems mabe handled and preference articulation maybe implemented. The motivation for using GAs for multi-objective problems such as control systems design and systems identification is given. The chapter concludes with case studies illustrating the use of MOGA for control system design and multi objective genetic programming (MOGP) for system identification.
-
4 Adaptive local linear modelling and control of nonlinear dynamical systems
- + Show details - Hide details
-
p.
119
–152
(34)
In this chapter, the problem of nonlinear system identification and control system design was addressed under the divide-and-conquer principle. This principle motivated the use of multiple local models for system identification in order to simplify the modelling task. Especially in the case of unknown dynamics, where only input output data from the plant is available, the proposed method is able to approximate the nonlinear dynamics of the plant using a piecewise linear dynamical model that is optimised solely from the available data. Especially when local linear models are used as described, it also became possible to design a piecewise linear controller for the plant, whose design is based on the identified model. The questions of the existence and validity of input-output models as described and utilised was addressed theoretically using the implicit function inversion theorem that points out the observability conditions under which such models are possible to build from input-output data alone. The performance of the proposed local linear modelling scheme and the associated local linear controllers was tested on a variety of nonlinear dynamical systems including chaotic systems, a NASA aircraft and the NASA Langley transonic wind tunnel.
-
5 Nonlinear system identification with local linear neuro-fuzzy models
- + Show details - Hide details
-
p.
153
–176
(24)
The local linear modelling approach is capable of modelling complex nonlinear processes. The transparency of the model architecture permits the incorporation of prior knowledge in different ways and the integration of process knowledge and experimental data has been demonstrated in detail for the modelling of the heat exchanger. Utilising an arbitrary local modelling strategy does not automatically guarantee a transparent model. The employed LOLIMOT algorithm has proven to yield transparent and well interpretable models. The appropriate choice of an excitation signal is vital for a good model. Here, the transparency of the local structure again aids the engineer in the design and the assessment of the excitation signal. The interpretability of the local structure offers simple ways for the validation of the identified model based on local characteristic values as process gains or time constants. The obtained local linear models can further be utilised in model-based control approaches as well as process monitoring or fault detection.
-
6 Gaussian process approaches to nonlinear modelling for control
- + Show details - Hide details
-
p.
177
–218
(42)
In the past years many approaches to modelling of nonlinear systems using neural networks and fuzzy modelshave been proposed. The difficulties associated with these black-box modelling techniques are mainly related to the curse of dimension ality and lack of transparency of the model. The local modelling approach has been proposed to increase transparency as well as reduce the curse of dimensionality. Difficulties related to partitioning of the operating space, structure determination, local model identification and off-equilibrium dynamics are the main drawbacks of such local modelling techniques. To improve the off-equilibrium behaviour, the use of non-parametric probabilistic models, such as Gaussian process priors was proposed. The Gaussian process prior approach was first introduced in Reference 6 and revised in References 7-9. The ability to make a robust estimation in the transient region, where only a limited number of data points is available, is one of the advantages of the Gaussian process in comparison to the local model network.
-
7 Neuro-fuzzy model construction, design and estimation
- + Show details - Hide details
-
p.
219
–252
(34)
Neuro-fuzzy model uses a set of fuzzy rules to model unknown dynamical systems with a B-spline functional. T-S inference mechanism is useful to produce an operating point dependent structure facilitating the use of state-space models for data fusion or data-based controller design. This chapter aims to introduce some neuro-fuzzy model construction, design and estimation algorithms that are developed to deal with fundamental problems in data modelling, such as model sparsity, robustness, transparency and rule-based learning process. Some work on ASMOD that has been derived based on ANOVA are reviewed initially and then a locally regularised orthogonal least-squares algorithm, based on T-S and ANOVA and combined with a D-optimality used for subspace-based rule selection, has been proposed for fuzzy rule regularisation and subspace-based information extraction.
-
8 A neural network approach for nearly optimal control of constrained nonlinear systems
- + Show details - Hide details
-
p.
253
–292
(40)
A rigorous computationally effective algorithm to find nearly optimal controllers in state feedback form for general nonlinear systems with constraints is presented that approaches the problem of constrained optimisation from a practical engineering tractable point. The control is given as the output of a neural network. Conditions under which the theory of successive approximation applies were shown. Several numerical examples were discussed and simulated. This algorithm requires further study into the problem of increasing the region of asymptotic stability. Moreover, adaptive control techniques can be blended to formulate adaptive optimal controllers for general nonlinear systems with constraints and unknown system dynamics. In this chapter, we study the constrained optimal control problem through the framework of the HJB equation. We show how to systematically obtain approximate solutions of the HJB equation when nonquadratic performance functions are used. The solution of the HJB equation is a challenging problem due to its inherently nonlinear nature. For linear systems, this equation results in the well-known Riccati equation used to derive a linear state feedback control. But even when the system is linear, the saturated control requirement makes the required control nonlinear, and makes the solution of the HJB equation more challenging.
-
9 Reinforcement learning for online control and optimisation
- + Show details - Hide details
-
p.
293
–326
(34)
Current control methodologies can generally be divided into model based and model free. The first contains conventional controllers, the second so-called intelligent controllers. Conventional control designs involve constructing dynamic models of the target system and the use of mathematical techniques to derive the required control law. Therefore, when a mathematical model is difficult to obtain, either due to complexity or the numerous uncertainties inherent in the system, conventional techniques are less useful. Intelligent control may offer a useful alternative in this situation.
-
10 Reinforcement learning and multi-agent control within an internet environment
- + Show details - Hide details
-
p.
327
–348
(22)
A multi-agent system consists of many individual computational agents, distributed throughout an environment, capable of learning environmental management strategies, environmental interaction and inter-agent communication. Multi-agent controllers offer attractive features for the optimisation of many real world systems. One such feature is that agents can operate in isolation, or in parallel with other agents and traditional control systems. This allows multi-agent controllers to be implemented incrementally as funds allow The distrusted nature of multi-agent control allows local control to be integrated into a global system. Each of the individual agent controllers can be comparatively simple and optimised for local problems. Through agent communication, the agents can take into consideration the global goals of a system when they are developing their local behaviour strategies. Multi-agent controllers have adaptive behaviour, developed by the artificial intelligence community, to learn and optimise local behaviour and at the same tune contribute to the global system performance.
-
11 Combined computational intelligence and analytical methods in fault diagnosis
- + Show details - Hide details
-
p.
349
–392
(44)
In this chapter, selected aspects of the fault diagnosis problem for control systems have been considered with special attention paid to fault detection and, to a lesser extent, fault isolation. In particular, the robustness problem of FDI with respect to the requirement to maximise sensitivity to faults while minimising (or decoupling) the effect of uncertain effects through unknown inputs has been described. This severe robustness challenge accompanying analytical methods of FDI has led to the requirement of developing new methodologies for FDI in which analytical and CI techniques are combined to achieve good global and nonlinear modelling and robustness. The chapter has outlined some of the recent research on FDI and fault diagnosis for dynamic systems, using this integrated approach. The examples have shown that by using evolutionary computing, neural networks and NF modelling structures realistic solutions are achievable. It is hoped that this direction of research will stimulate an increased adoption of real industrial application to make mode-based FDI more usual and effective in real process systems.
-
12 Application of intelligent control to autonomous search of parking place and parking of vehicles
- + Show details - Hide details
-
p.
393
–414
(22)
This chapter has presented an intelligent system for autonomous parking of vehicles, including intelligent navigation and searching of parking place. Thus, an intelligent control method has been implemented. This method uses soft-computing techniques to: (1) take advantage of knowledge of expert drivers, (2) manage imprecise information from sensors, (3) navigate searching for a parking place and (4) select and perform a parking manoeuvre. Soft-computing techniques facilitate combining different strategies, providing the adaptation of the control motion to the environment and improving manoeuvre performance. The efficiency of the proposed method is demonstrated using the non holonomic mobile robots ROMEO-3R and ROMEO-4R, developed at the University of Seville.
-
13 Applications of intelligent control in medicine
- + Show details - Hide details
-
p.
415
–446
(32)
The first part of the chapter describes the results of two literature surveys on intelligent systems allied to medicine which were commissioned by the European Networks of Excellence ERUDIT and EUNITE. The first survey covered the use of fuzzy technology across the whole of medicine divided into ten sub-disciplines, ranking from surgery to many branches of internal medicine. The second survey covered the field of intelligent adaptive systems in medicine, again applied to the same ten subdivisions as above, but reduced to five sub-areas in the analysis of the results. In this case a wide range of intelligent techniques was considered including fuzzy inference, neural networks and model-based reasoning. Both surveys attempted to show the merits and limitation of the current work and to give signposts for future developments in the field. The second part of the chapter contains two relevant case studies. The first concerns intelligent systems in anaesthesia monitoring and control. The theme is unconsciousness management in operating theatres, which is a particularly challenging application for AI techniques. The second case study concerns the developments of nonlinear models based on ANN and neuro-fuzzy techniques for both classification and prediction in cancer survival studies. The particular application is that of bladder tumour prognosis using gene expression markers together with demographic data and social conditions such as smoking.
-
Back Matter
- + Show details - Hide details
-
p.
447
(1)