Anytime information processing based on fuzzy and neural network models

In modern measurement and control systems, the available time and resources are often not only limited, but could change during the operation of the system. In these cases, the so-called anytime algorithms could be used advantageously. While diflerent soft computing methods are wide-spreadly used in system modeling, their usability in these cases are limited.

Estimating fuzzy membership functions parameters by the levenberg-marquardt algorithm

In previous papers from the authors fuzzy model identification methods were discussed. The bacterial algorithm for extracting fuzzy rule base from a training set was presented. The Levenberg-Marquardt algorithm was also proposed for determining membership functions in fuzzy systems. In this paper the Levenberg-Marquardt technique is improved to optimise the membership functions in the fuzzy rules without Ruspini-partition. The class of membership functions investigated is the trapezoidal one as it is general enough and widely used. The method can be easily extended to arbitrary piecewise linear functions as well.

Training neuro-fuzzy models using evolution based algorithms

The normal design process for neural networks or fuzzy systems involve two different phases: the determination of the best topology, which can be seen as a system identification problem, and the determination of its parameters, which can be envisaged as a parameter estimation problem. This latter issue, the determination of the model parameters (linear weights and interior knots) is the simplest task and is usually solved using gradient or hybrid schemes. The former issue, the topology determination, is an extremely complex task, especially if dealing with real-world problems.

Design of neuro-fuzzy models by evolutionary and gradient-based algorithms

All systems found in nature exhibit, with different degrees, a nonlinear behavior. To emulate this behavior, classical systems identification techniques use, typically, linear models, for mathematical simplicity. Models inspired by biological principles (artificial neural networks) and linguistically motivated (fuzzy systems), due to their universal approximation property, are becoming alternatives to classical mathematical models. In systems identification, the design of this type of models is an iterative process, requiring, among other steps, the need to identify the model structure, as well as the estimation of the model parameters. This thesis addresses the applicability of gradient-basis algorithms for the parameter estimation phase, and the use of evolutionary algorithms for model structure selection, for the design of neuro-fuzzy systems, i.e., models that offer the transparency property found in fuzzy systems, but use, for their design, algorithms introduced in the context of neural networks. A new methodology, based on the minimization of the integral of the error, and exploiting the parameter separability property typically found in neuro-fuzzy systems, is proposed for parameter estimation. A recent evolutionary technique (bacterial algorithms), based on the natural phenomenon of microbial evolution, is combined with genetic programming, and the resulting algorithm, bacterial programming, advocated for structure determination. Different versions of this evolutionary technique are combined with gradient-based algorithms, solving problems found in fuzzy and neuro-fuzzy design, namely incorporation of a-priori knowledge, gradient algorithms initialization and model complexity reduction.

A neural network seismic detector

This experimental study focuses on a detection system at the seismic station level that should have a similar role to the detection algorithms based on the ratio STA/LTA. We tested two types of neural network: Multi-Layer Perceptrons and Support Vector Machines, trained in supervised mode. The universe of data consisted of 2903 patterns extracted from records of the PVAQ station, of the seismography network of the Institute of Meteorology of Portugal. The spectral characteristics of the records and its variation in time were reflected in the input patterns, consisting in a set of values of power spectral density in selected frequencies, extracted from a spectro gram calculated over a segment of record of pre-determined duration. The universe of data was divided, with about 60% for the training and the remainder reserved for testing and validation. To ensure that all patterns in the universe of data were within the range of variation of the training set, we used an algorithm to separate the universe of data by hyper-convex polyhedrons, determining in this manner a set of patterns that have a mandatory part of the training set. Additionally, an active learning strategy was conducted, by iteratively incorporating poorly classified cases in the training set. The best results, in terms of sensitivity and selectivity in the whole data ranged between 98% and 100%. These results compare very favorably with the ones obtained by the existing detection system, 50%.

Identification of aircraft gas-turbines dynamics: comparison of classical and neural techniques

A gas turbine is made up of three basic components: a compressor, a combustion chamber and a turbine. Air is drawn into the engine by the compressor, which compresses it and delivers it to the combustion chamber. There, the air is mixed with the fuel and the mixture ignited, producing a rise of temperature and therefore an expansion of the gases. These are expelled through the engine nozzle, but first pass through the turbine, designed to extract energy to keep the compressor rotating [1]. The work described here uses data recorded from a Rolls Royce Spey MK 202 turbine, whose simplified diagram can be seen in Fig. 1. Both the compressor and the turbine are split into low pressure (LP) and high pressure (HP) stages. The HP turbine drives the HP compressor and the LP turbine drives the LP compressor. They are connected by concentric shafts that rotate at different speeds, denoted as NH and NL.

Neural network classification of cerebral embolic signals

The presence of circulating cerebral emboli represents an increased risk of stroke. The detection of such emboli is possible with the use of a transcranial Doppler ultrasound (TCD) system.

B-splines neural network assisted PID autotuning

This papers describes an extantion of previous works on the subject of neural network proportional, integral and derivative (PID) autotuning. Basically, neural networks are employed to supply the three PID parameters, according to the integral of time multiplied by the absolute error (ITAE) criterion, to a standard PID controller.

On-line adaptation of neural network

The Proportional Integral and Devirative (PID) controller autotuning is an important problem, both in practical and theoretical terms. The autotuning procedure must take place in real-time, and therefore the corresponding optimisation procedure must also be executed in real-time and without disturbing on-line control.

Recent developments in neural network PID autotuning

PID controllers are widely used in industrial applications. Because the plant can be time variant, methods of autotuning of this type of controllers, are of great economical importance, see (Astrom, 1996). Since 1942, with the work of Ziegler and Nichols (Ziegler and Nichols, 1942), several methods have been proposed in the literature. Recently, a new technique using neural networks was proposed (Ruano et al., 1992). This technique has been shown to produce good tunings as long as certain limitations are met.

Neural network assisted autotuning of industrial PID controllers

A recent servey (1) has reported that the majority of industrial loops are controlled by PID-type controllers and many of the PID controllers in operation are poorly tuned. poor PID tuning is due to the lack of a simple and practical tuning method for avarage users, and due to the tedious procedurs involved in the tuning and retuning of PID controllers.

Comparison of alternative approaches to neural network PID autotuning

In this paper, a scheme for the automatic tuning of PID controllers on-line, with the assistance of trained neural networks, is proposed. The alternative approaches are presented and compared.

A neural network PID autotuner

Proportional, Integral and Derivative (PID) regulators are standard building blocks for industrial automation. Their popularity comes from their rebust performance and also from their functional simplicity. Whether because the plant is time-varying, or because of components ageing, these controllers need to be regularly retuned.

A neural network controller

Proportional, Integral and Derivative (PID) regulators are standard building blocks for industrial automation. The popularity of these regulators comes from their rebust performance in a wide range of operating conditions, and also from their functional simplicity, which makes them suitable for manual tuning.