Repository logo
 
Loading...
Profile Picture

Search Results

Now showing 1 - 10 of 187
  • An efficient parallel implementation of a least squares problem
    Publication . Ruano, Antonio; Fleming, P. J.; Jones, D. I.
    Least squares solutions are a very important problem, which appear in a broad range of disciplines (for instance, control systems, statistics, signal processing). Our interest in this kind of problems lies in their use of training neural network controllers.
  • Single and multi-objective genetic programming design for B-spline neural networks and neuro-fuzzy systems
    Publication . Cabrita, Cristiano Lourenço; Ruano, Antonio; Fonseca, C. M.
    The design phase of B-spline neural networks and neuro-fuzzy systems is a highly computationally complex task. Existent heuristics, namely the ASMOD algorithm, have been found to be highly dependent on the initial conditions employed. A Genetic Programming approach is proposed, which produces an efficient topology search, obtaining additionally more consistent solutions. The facility to incorporate a multi-objective approach to the GP algorithm is exploited, enabling the designer to obtain better conditioned models, and more adequate for their intended use.
  • Applications of neural networks to control systems
    Publication . Ruano, Antonio
    This work investigates the applicability of artificial neural networks to control systems. The following properties of neural networks are identified as of major interest to this field: their ability to implement nonlinear mappings, their massively parallel structure and their capacity to adapt. Exploiting the first feature, a new method is proposed for PID autotuning. Based on integral measures of the open or closed loop step response, multilayer perceptrons (MLPs) are used to supply PID parameter values to a standard PID controller. Before being used on-line, the MLPs are trained offline, to provide PID parameter values based on integral performance criteria. Off-line simulations, where a plant with time-varying parameters and time varying transfer function is considered, show that well damped responses are obtained. The neural PID autotuner is subsequently implemented in real-time. Extensive experimentation confirms the good results obtained in the off-line simulations. To reduce the training time incurred when using the error back-propagation algorithm, three possibilities are investigated. A comparative study of higherorder methods of optimization identifies the Levenberg-Marquardt (LM)algorithm as the best method. When used for function approximation purposes, the neurons in the output layer of the MLPs have a linear activation function. Exploiting this linearity, the standard training criterion can be replaced by a new, yet equivalent, criterion. Using the LM algorithm to minimize this new criterion, together with an alternative form of Jacobian matrix, a new learning algorithm is obtained. This algorithm is subsequently parallelized. Its main blocks of computation are identified, separately parallelized, and finally connected together. The training time of MLPs is reduced by a factor greater than 70 executing the new learning algorithm on 7 Inmos transputers.
  • Seismic event detection with artificial neural networks
    Publication . Madureira, G.; Ruano, Antonio
    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 spectrogram 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%.
  • A neural network PID autotuner
    Publication . Ruano, Antonio; Lima, João; Mamat, R.; Fleming, P. J.
    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.
  • Comparison of alternative approaches to neural network PID autotuning
    Publication . Ruano, Antonio; Lima, João; Mamat, R.; Fleming, P. J.
    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.
  • Fuzzy model identification by evolutionary, gradient based and memtic algorithms
    Publication . Botzheim, J.; Kóczy, László T.; Ruano, Antonio
    One of the crucial problems of fuzzy rule modeling is how to find an optimal or at least a quasi-optimal rule base fro a certain system. In most applications there is no human expert available, or, the result of a human expert's decision is too much subjective and is not reproducible, thus some automatic method to determine the fuzzy rule base must be deployed.
  • Accelerating multi-objective control system design using a neuro-genetic approach
    Publication . Duarte, N. M.; Ruano, Antonio; Fonseca, C. M.; Fleming, P. J.
    Designing control systems using multiobjective genetic algorithms can lead to a substantial computational load as a result of the repeated evaluation of the multiple objectives and the population-based nature of the search. Here, a neural network approach, based on radial basis functions, is introduced to alleviate this problem by providing computationally inexpensive estimates of objective values during the search. A straightforward example demonstrates the utility of the approach.
  • Recent developments in neural network PID autotuning
    Publication . Ruano, Antonio
    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.
  • Configuration space synthesis for robotic manipulators using neural networks
    Publication . Pashkevich, A.; Ruano, Antonio; Kazheunikau, M.
    The paper deals with configuration space syntheses for industrial robotic manipulators. A new efficient method is proposed that is based on a neural network collision model. To generate the collision model, a modification of the Radial Basis Function Network (RBFN) is used, which is trained applying the developed algorithm. An obstacle transformation algorithm that is based on conjugate vector model of a robotic cell is proposed. The method has been successfully applied to the design of a robotic manufacturing cell for the automotive industry.