F. Vacca

Descriptive and Inferential Statistics for Supervising and Monitoring the Operation of PV Plants

This paper deals with the problem of supervising and monitoring a photovoltaic (PV) plant. First, an offline descriptive and inferential statistical procedure for evaluating the goodness of system performance is presented. Then, an online inferential algorithm for real-time monitoring and fault detection is introduced. The two methodologies utilize the energy output of inverters as input data and are valid for both Gaussian and non-normal distribution of data. The procedures have been tested on a real PV installation, and results are reported for the case of a grid-connected PV plant in Italy for which one PV module over 132 resulted in being badly connected.

Centroid estimation by means of uncompetitive unsupervised neural element

In this paper the drawbacks of classical unsupervised learning laws are discussed and the paradigms of an alternative clustering algorithm are carried out. Then a new model of neuron element able to search the centroid of clusters without competition with other neurons, as in an unsupervised competitive learning law, is singled out.<<ETX>>

Time domain analysis of switching circuits by using the simulink-based co-simulation

The paper deals with the transient analysis in the simulation techniques of switching circuits typically employed in power electronics. It is not easy to simulate power electronic circuits, because they are variant topology circuits. Most simulators consider an approximation model for the switch in order to obtain an invariant topology, but this approach introduces new problems. A good solution to simulate power switching circuits could be the co-simulation. In the paper mathematical basics of a co-simulation are recalled and a set of four simulations is considered. A first comparison reports the simulation results utilizing a general purpose simulation software (PSpice) and those obtained by means of the co-simulation PSpice-Simulink. In the second comparison, the simulation results utilizing specific simulation software (PSIM) and those obtained by means of the co-simulation PSIM-Simulink are examined.

Multivariate data projection techniques based on a network of enhanced neural elements

In this paper two techniques to project high dimensional data into a bidimensional space are introduced. These techniques are based on an unsupervised neural network of enhanced processing elements. The proposed approaches are compared with some widely known projection techniques based on unsupervised neural networks. These comparisons show that the new projection techniques perform comparably or slightly better than the traditional techniques and are promising in term of computational burden.

Local competitive signals for an unsupervised competitive neural network

Unsupervised Competitive Neural Networks (UCN) have been recognized as a powerful tool for pattern analysis, feature extraction and clustering analysis. Nevertheless, the inhibitory interactions among the units of the network, required by the winner-take-all paradigm, constitute a crucial step for the implementation of competitive networks in analog VLSI. The aim of this paper is to present an unsupervised competitive neural network characterized by local inhibitory interactions among its cells. The kernel of this network is a neural unit based on a modified competitive learning law in which the threshold changes in the learning stage. It is shown that the proposed neuron unit is able, during the learning stage, to perform an automatic selection of patterns that belong to a cluster, moving towards its centroid. The properties of this network, related to the robustness of the final results and to the choice of the number of the elements, are examined in a set of numerical simulations adopting a data set composed of Gaussian mixtures and uniform noise.

Performance evaluation of a "parallel collision control" unsupervised neural network

In this paper a new neural approach to clustering tasks in handwritten numeral recognition problems is compared to classical unsupervised neural networks techniques. The kernel of the proposed network is a neural unit able to perform clustering acting alone. The network is able to find directly dense zones of the input space without requiring competition and thus overcoming the major drain backs of classical unsupervised architectures.

Adjoint network theory to analyse the power converters with respect to their line-side behaviour

This chapter presents a straightforward approach for the sensitivity analysis of the line-side connected converters with respect to the variation of their parameters. The sensitivity analysis is carried out based on the adjoint network theory. This allows studying the variation of an electrical quantity with respect to the perturbation of any circuit parameter, considering only one circuit besides the one assigned. As the control of a converter plays an important role in the propagation of the harmonics in the circuit, it is necessary to implement a detailed model of the controller too. Starting from the requirements of the digital controller, an analogue circuit that is able to represent the behavior of the controller is synthesized. Substituting the controller with the virtual circuit, an analogue network, representing the power converter and its control, is obtained. This homogenous model allows a rigorous and straightforward sensitivity analysis. This straightforward approach does not neglect anything of both the physical system and the digital controller. Also, with this approach it is easy to evaluate the problems of the use of the line-connected converter related to the propagation of harmonics in an industrial plant.

Clustering noisy data by a principal feature extraction unsupervised neural network

Principal feature classification is based on a sequential procedure for finding the principal features from an assigned data set. This paper presents an unsupervised neural network which is able to find principal features, based on neural units sensitive to density of the data space. These units adopt a modified competitive learning law, which utilizes only local information to specialize toward a single cluster. It is shown that the network presented is able to automatically select the number of units as in the rival penalized competitive network, and also to correctly detect features when the number of clusters exceed the number of units. Simulations on IRIS data set are provided and it is shown that the proposed network presents property of robust noise rejection and is suitable for features extraction in noise data sets.

A bit-oriented parallel A/D converter

In this paper, the electrical circuit scheme for the Bit-Oriented A/D converter compatible with most CMOS process technologies is proposed. In order to overcome the problems of CMOS resistances implementation, further current is used as a processing signal. The most important factors affecting conversion resolution and speed are discussed. It is pointed out that a 3-/spl mu/m CMOS implementation of bit oriented A/D converter can achieve 8-bit resolution to maintain an absolute current error of half LSB. The bit-oriented converter can operate at a sampling rate of 8 MHz. The simple structure of the converter can give rise to other digital codes through variation of the reference voltages and current source connections.<<ETX>>

An alternative approach to parallel A/D conversion

A functional diagram of a novel type of A/D converter, described as bit oriented, for high-speed application is discussed. Although the system can be regarded as the natural development of the well-known flash ADC, a different approach to the conversion process has led to a reduction in the number of comparators required, with no loss in flash performances. There is also no need for sample and hold (S/H) and digital circuits for the coding. A comparison with the most up-to-date high conversion techniques (subranging, folding and interpolation) points out the advantages of the technique.<<ETX>>