Multi-valued Neuron Research Articles

A complex Neural Classifier for the Fault Prognosis and Diagnosis of Overhead Electrical Lines

The technique proposed in this work is finalized to the non-intrusive monitoring of high voltage electrical networks. In order to develop a prognostic method capable of avoiding failures on overhead transmission grids, the connection joints between two sections of the line are considered. The method is based on the use of Frequency Response Analysis (FRA) and machine learning, represented by a neural classifier based on a Multi-Valued Neuron (MVN) neural network. The procedure can be considered as a smart measurement block, where a single measure is used by a neural classifier to extract information able to diagnose an electrical system. This means that the method shown in this paper can be developed and adapted to solve many different problems in the world of industry, such as the management of the most worn electrical devices. In this sense, the maintenance organization plays a fundamental role and the prognostic approach allows the reduction of the recovery times by locating critical components. This monitoring system increases the global availability of the electrical grid in which it is used and, from a practical point of view, it can be used by network operators to obtain online control of operating conditions.

MLMVNNN for Parameter Fault Detection in PWM DC–DC Converters and Its Applications for Buck and Boost DC–DC Converters

This paper presents an effective approach to the fault diagnosis of pulsewidth modulated (PWM) DC–DC power converters. It is based on a multilayer multivalued neuron neural network with a complex QR decomposition. This network is used to identify the parameter values running out of tolerance range in two topologies of PWM DC–DC converters, namely, the buck and boost circuits. The proposed technique is based on measurements of steady-state voltages and currents waveforms.

A MLMVN with Arbitrary Complex-Valued Inputs and a Hybrid Testability Approach for the Extraction of Lumped Models Using FRA

Abstract A procedure for the identification of lumped models of distributed parameter electromagnetic systems is presented in this paper. A Frequency Response Analysis (FRA) of the device to be modeled is performed, executing repeated measurements or intensive simulations. The method can be used to extract the values of the components. The fundamental brick of this architecture is a multi-valued neuron (MVN), used in a multilayer neural network (MLMVN); the neuron is modified in order to use arbitrary complex-valued inputs, which represent the frequency response of the device. It is shown that this modification requires just a slight change in the MLMVN learning algorithm. The method is tested over three completely different examples to clearly explain its generality.

A Multi-Valued Neuron Based Complex ELM Neural Network

In this paper, a new efficient model of neural network is proposed, which is realized by the combination of two recent and successful neurocomputing paradigms. The idea behind the work is to realize a neural network constituted by multi-valued complex neurons, which are trained with the principles of extreme learning machine (ELM). The specific kind of used neuron allows a very straightforward derivation of the ELM, with no substantial modification in the weight adjustment procedure. The main advantages that clearly emerge by this model are represented by the further increasing of generalization performances in presence of noise, together with a generalized reduction of needed nodes (neurons) in the hidden layer. The effectiveness of the proposed model will be shown by some benchmark results, also compared with the original techniques.

Modified Multivalued Neuron With Periodic Tolerant Activation Function

The multivalued neuron with periodic activation function (MVN-P) was proposed by Aizenberg for solving classification problems. The boundaries between two distinct categories are crisply specified in MVN-P, which may result in slow convergence or being unable to converge at all in the learning process. In this paper, we propose a revised model of MVN-P based on the idea of unsharp boundaries. In this revised model, a fuzzy buffer is provided around a boundary between two distinct categories, allowing incorrect assignments with membership degree less than a threshold to be tolerated in the training phase. Genetic algorithms are applied to derive optimal values for the parameters involved in this model, alleviating the burden of setting them manually by the user. Besides, MVN-P has difficulties solving the classification problems having a large number of categories. A tree structure is developed to overcome these difficulties. Simulation results demonstrate the effectiveness of our proposed ideas.

MLMVN With Soft Margins Learning

In this paper, we consider a modified error-correction learning rule for the multilayer neural network with multivalued neurons (MLMVN). This modification is based on the soft margins technique, which leads to the minimization of the distance between a cluster center and the learning samples belonging to this cluster. MLMVN has a derivative-free learning algorithm based on the error-correction learning rule and demonstrate a higher functionality and better generalization capability than a number of other machine learning techniques. The discrete k-valued multivalued neuron activation function divides a complex plane into k equal sectors. For more efficient and reliable solving of classification problems it is possible to modify the MLMVN learning algorithm in such a way that learning samples belonging to different classes (clusters) will be located as close as possible to the bisector of a desired sector (the cluster center) and as far as possible from each other, respectively. Such a modification based on the soft margins learning technique is considered in this paper. This modified learning algorithm improves the generalization capability of MLMVN when solving classification problems.

Multi-Valued Neuron with Sigmoid Activation Function for Pattern Classification

Multi-Valued Neuron (MVN) was proposed for pattern classification. It operates with complex-valued inputs, outputs, and weights, and its learning algorithm is based on error-correcting rule. The activation function of MVN is not differentiable. Therefore, we can not apply backpropagation when constructing multilayer structures. In this paper, we propose a new neuron model, MVN-sig, to simulate the mechanism of MVN with differentiable activation function. We expect MVN-sig to achieve higher performance than MVN. We run several classification benchmark datasets to compare the performance of MVN-sig with that of MVN. The experimental results show a good potential to develop a multilayer networks based on MVN-sig.

System identification and modelling based on a double modified multi-valued neural network

A novel identification technique for the extraction of lumped circuit models of general distributed or stray devices is presented. The approach is based on two multi-valued neuron neural networks used in a joined architecture able to extract hidden parameters. The convergence allows the validation of the approximated lumped model and the extraction of the correct values. The inputs of the neural network are the geometrical parameters of a given structure, while the outputs represent the estimation of the lumped circuit parameters. The method uses a frequency response analysis approach in order to elaborate the data to present to the net.

Hebbian and error-correction learning for complex-valued neurons

In this paper, we observe some important aspects of Hebbian and error-correction learning rules for complex-valued neurons. These learning rules, which were previously considered for the multi-valued neuron (MVN) whose inputs and output are located on the unit circle, are generalized for a complex-valued neuron whose inputs and output are arbitrary complex numbers. The Hebbian learning rule is also considered for the MVN with a periodic activation function. It is experimentally shown that Hebbian weights, even if they still cannot implement an input/output mapping to be learned, are better starting weights for the error-correction learning, which converges faster starting from the Hebbian weights rather than from the random ones.

Introduction to the special issue on SM2ACD 2010

Welcome to the Special Issue of SM2ACD 2010 workshop (initially SMACD). This workshop is the XIth International Workshop on Symbolic and Numerical Methods, Modeling and Applications to Circuit Design held on 4–6 October 2010 in Gammarth—Tunisia. The history of SM(2)ACD workshops goes back to the early nineties. The first SMACD was held in 1991 in France. Throughout the past 20 years, SMACD has been organized at various European locations: Italy (1992), Spain (1994), Belgium (1996), Germany (1998), Portugal (2000), Romania (2002), Poland (2004), Italy (2006), and Germany (2008). This Special Issue of the Analog Integrated Circuits and Signal Processing consists of the expanded and carefully revised version of papers presented at SM2ACD 2010. Based on workshop review scores and feedback from the sessions chairs on the presentations, nine best papers were selected for publication in this issue from all the contributions at the workshop, and were subject to the normal journal review and refereeing process. These papers cover a large variety of the SM2ACD topics. (Eleven invitations for contribution to this special issue were sent. Ten were received. After the assessment of the reviewers, nine of them have been accepted for publication). The first paper, by Zhigang Hao, Sheldon X.-D. Tan, Esteban Tlelo-Cuautle, Jacob Relles, Chao Hu, Wenjian Yu, Yici Cai and Guoyong Shi, presents a novel method for statistical inductance extraction and modeling for interconnects considering process variations. The proposed method uses the collocation-based spectral stochastic method where orthogonal polynomials are used to represent the statistical processes. The second paper, by Francesco Grasso, Antonio Luchetta, Stefano Manetti and Maria Cristina Piccirilli, presents a novel identification technique for lumped models of general distributed circuits (microwave transmission lines, monolithic integrated circuits and filters) that is based on a hybrid multi-valued neuron neural network with a modified layer and learning process. The third paper, by David Guilherme, Jorge Guilherme and Nuno Horta, presents an optimization methodology for continuous time loop-filters design applied to class-D amplifiers. The proposed methodology is based on an evolutionary optimization approach which integrates both the topology selection and circuit sizing by automatically generating optimal sized topologies and performance tradeoffs. The fourth paper, by Antonietta De Nardo, Nicola Femia and Walter Zamboni, presents a new symbolic method for the design of filter capacitors in DC–DC switching converters, using a combination of a ceramic and an electrolytic capacitor; an effective method for the design of mixed electrolytic/ceramic capacitor-based passive filters. The fifth paper, by Pedro Pereira, Maria Helena Fino, Fernando Coito and Mario Ventim-Neves, addresses a technology-aware modified genetic optimization-based approach for the design of RF integrated inductors where the double p-model is used for the characterization of the inductor’s behavior. The sixth paper, by Florin Constantinescu, Miruna Nitescu, Alexandru Gabriel Gheorghe, Aurelian Florea and Olivier Llopis, presents a nonlinear circuit model of power BAW resonators and filters, which takes into account the parasitic of the resonator connection to the measurement bench. New models for the anti-series and the anti-parallel connections of two resonators are also proposed. The seventh paper, by Elisenda Roca, Manuel VelascoJimenez, Rafael Castro-Lopez and Francisco V. Fernandez, M. Fakhfakh (&) University of Sfax, B.P. 1069, 3018 Sfax, Tunisia e-mail: mourad.fakhfakh@ieee.org

A new multi-valued neural network for the extraction of lumped models of analog circuits

A novel identification technique for lumped models of general distributed circuits (i.e. microwave transmission lines, monolithic integrated circuits and filters) is presented. The approach is based on a hybrid multi-valued neuron neural network with a modified layer and learning process, whose convergence allows the validation of the approximated lumped model. The modified layer is generated by symbolic analysis of the model under exam. The inputs of the neural network are geometrical parameters, while the outputs represent the estimation of the lumped circuit parameters.

Periodic Activation Function and a Modified Learning Algorithm for the Multivalued Neuron

In this paper, we consider a new periodic activation function for the multivalued neuron (MVN). The MVN is a neuron with complex-valued weights and inputs/output, which are located on the unit circle. Although the MVN outperforms many other neurons and MVN-based neural networks have shown their high potential, the MVN still has a limited capability of learning highly nonlinear functions. A periodic activation function, which is introduced in this paper, makes it possible to learn nonlinearly separable problems and non-threshold multiple-valued functions using a single multivalued neuron. We call this neuron a multivalued neuron with a periodic activation function (MVN-P). The MVN-Ps functionality is much higher than that of the regular MVN. The MVN-P is more efficient in solving various classification problems. A learning algorithm based on the error-correction rule for the MVN-P is also presented. It is shown that a single MVN-P can easily learn and solve those benchmark classification problems that were considered unsolvable using a single neuron. It is also shown that a universal binary neuron, which can learn nonlinearly separable Boolean functions, and a regular MVN are particular cases of the MVN-P.

Multilayer Feedforward Neural Network Based on Multi-valued Neurons (MLMVN) and a Backpropagation Learning Algorithm

A multilayer neural network based on multi-valued neurons (MLMVN) is considered in the paper. A multi-valued neuron (MVN) is based on the principles of multiple-valued threshold logic over the field of the complex numbers. The most important properties of MVN are: the complex-valued weights, inputs and output coded by the kth roots of unity and the activation function, which maps the complex plane into the unit circle. MVN learning is reduced to the movement along the unit circle, it is based on a simple linear error correction rule and it does not require a derivative. It is shown that using a traditional architecture of multilayer feedforward neural network (MLF) and the high functionality of the MVN, it is possible to obtain a new powerful neural network. Its training does not require a derivative of the activation function and its functionality is higher than the functionality of MLF containing the same number of layers and neurons. These advantages of MLMVN are confirmed by testing using parity n, two spirals and "sonar" benchmarks and the Mackey---Glass time series prediction.

Image coding and decoding by discrete‐time cellular neural networks

AbstractWhen the original signal to be encoded has a strong correlation in the neighborhood, as in the case of the natural image, it can be expected that the original signal is represented by a smaller number of bits.This paper proposes the following low‐bit coding system. the analog image is converted into the digital image by A‐D conversion in the transmitter based on the dynamics of the locally interconnected discrete‐time cellular neural network (DTCNN). the digital image is passed through a lowpass filter in the receiver so that the original analog image is restored. In this method, the coding of the static image is reduced to the optimization problem where the error between the original image and the decoded image is minimized. the distortion function is defined so that the noise generated by encoding is suppressed from the viewpoint of the whole image.A high‐quality image is reconstructed by the dynamics of the multivalued neuron, which is an extension of the output of the neuron to multiple values. In this method, a parallel processing dynamics is employed, where a complex function can be realized from simple devices, and it is expected to realize the digital conversion in real‐time of the analog information which is input in parallel, as in the case of the light input to the retina.