Fast Learning by Bounding Likelihoods in Sigmoid Type Belief Networks

Sigmoid type belief networks, a class of probabilistic neural networks, provide a natural framework for compactly representing probabilistic information in a variety of unsupervised and supervised learning problems. Often the parameters used in these networks need to be learned from examples. Unfortunately, estimating the parameters via exact probabilistic calculations (i.e, the EM-algorithm) is intractable even for networks with fairly small numbers of hidden units. We propose to avoid the infeasibility of the E step by bounding likelihoods instead of computing them exactly. We introduce extended and complementary representations for these networks and show that the estimation of the network parameters can be made fast (reduced to quadratic optimization) by performing the estimation in either of the alternative domains. The complementary networks can be used for continuous density estimation as well.

Monitorització de la qualitat de l'electricitat d'un edifici universitari mitjançant l'analitzador CM4000

Es va instal.lar un analitzador CM4000 a la Facultat de Ciències per tal de poder enregistrar en temps real les incidències en la xarxa elèctrica que s'estaven produint (caigudes de tensió i problemes amb els harmònics). El projecte vol monotoritzar la qualitat dels registres d'aquest analitzador que queden enregistrats a la base de dades Power Server. A partir d'aquestes dades es realitza un estudi sobre la freqüència de les incidències, buscant el patró dels dies i hores en què les incidències són màximes

Integrating simulated annealing and delta technique for constructing optimal prediction intervals

This paper aims at developing a new criterion for quantitative assessment of prediction intervals. The proposed criterion is developed based on both key measures related to quality of prediction intervals: length and coverage probability. This criterion is applied as a cost function for optimizing prediction intervals constructed using delta technique for neural network model. Optimization seeks out to minimize length of prediction intervals without compromising their coverage probability. Simulated Annealing method is employed for readjusting neural network parameters for minimization of the new cost function. To further ameliorate search efficiency of the optimization method, parameters of the network trained using weight decay method are considered as the initial set in Simulated Annealing algorithm. Implementation of the proposed method for a real world case study shows length and coverage probability of constructed prediction intervals are better than those constructed using traditional techniques.

Construction of optimal prediction intervals for load forecasting problems

Short-term load forecasting is fundamental for the reliable and efficient operation of power systems. Despite its importance, accurate prediction of loads is problematic and far remote. Often uncertainties significantly degrade performance of load forecasting models. Besides, there is no index available indicating reliability of predicted values. The objective of this study is to construct prediction intervals for future loads instead of forecasting their exact values. The delta technique is applied for constructing prediction intervals for outcomes of neural network models. Some statistical measures are developed for quantitative and comprehensive evaluation of prediction intervals. According to these measures, a new cost function is designed for shortening length of prediction intervals without compromising their coverage probability. Simulated annealing is used for minimization of this cost function and adjustment of neural network parameters. Demonstrated results clearly show that the proposed methods for constructing prediction interval outperforms the traditional delta technique. Besides, it yields prediction intervals that are practically more reliable and useful than exact point predictions.

Optimizing the quality of bootstrap-based prediction intervals

The bootstrap method is one of the most widely used methods in literature for construction of confidence and prediction intervals. This paper proposes a new method for improving the quality of bootstrap-based prediction intervals. The core of the proposed method is a prediction interval-based cost function, which is used for training neural networks. A simulated annealing method is applied for minimization of the cost function and neural network parameter adjustment. The developed neural networks are then used for estimation of the target variance. Through experiments and simulations it is shown that the proposed method can be used to construct better quality bootstrap-based prediction intervals. The optimized prediction intervals have narrower widths with a greater coverage probability compared to traditional bootstrap-based prediction intervals.

Comparação entre os intumescimentos de borracha natural e polibutadieno com alto teor 1,4-cis em solventes (n-octano, ciclo-hexano e esqualeno) e em soluções poliméricas

Amostras de borracha natural foram reticuladas por meio de radiação gama (doses de 2, 4, 6 e 8 MRad) e intumescidas: a) em n-octano, ciclo-hexano e esqualeno; b) em polibutadieno líquido (Mn=1830 g/mol; alto teor 1,4- cis) a 45°C; C) em solução de polibutadieno de baixo peso molecular (M,= 1830 g/mol; alto teor 1,4-cis) em n-octano e ciclo-hexano; d) em solução de polibutadieno de alto peso molecular (Mn= 175.000 g/rnol; alto teor 1,4-cis) emn-octano e ciclo-hexano ; e) em solução de borracha natural (Mn=779.000 g/mol) em ciclo- hexano, a 25°C. Amostras de polibutadieno (alto teor 1,4-cis) foram reticuladas por meio de radiação gama (doses de 1,s; 18 e 25 MRad) e intumescidas: a) em n-octano, ciclo-hexano e esquaieno. Salvo quando mencionado em contrário, as amostras de borracha natural e de polibutadieno foram intumescidas nas temperaturas de 25, 35 e 45°C. A partir dos ensaios de intumescirnento acima mencionados, foram determinadas as solubilidades de borracha natural e de polibutadieno em n-octano, ciclo-hexano e esqualeno através dos parâmetros de Flory-Huggins, calculados com o emprego da equação de Flory-Rehner. Verificou-se a influência da temperatura e da concentração de polimero no gel sobre o valor do parâmetro de Flory-Huggins. Também através desse parâmetro procurou-se determinar a compatibilidade entre borracha natural e polibutadieno. Foi demonstrado que cadeias lineares de baixo peso molecular de polibutadieno conseguem penetrar em amostras reticuladas de borracha natural, quando o peso molecular das cadeias lineares for bem inferior ao do arco de rede Mc das amostras reticuladas. Cadeias lineares de polibutadieno e de borracha natural com peso molecular superior ao do arco de rede da borracha natural reticulada, entretanto, não conseguem penetrar no retículo. Quranto mais alta a concentração da solução externa em moléculas de alto peso molecular, tanto menor é o grau de intumescimento da amostra reticulada.

A Learning Function for Parameter Reduction in Spiking Neural Networks with Radial Basis Function

Spiking neural networks - networks that encode information in the timing of spikes - are arising as a new approach in the artificial neural networks paradigm, emergent from cognitive science. One of these new models is the pulsed neural network with radial basis function, a network able to store information in the axonal propagation delay of neurons. Learning algorithms have been proposed to this model looking for mapping input pulses into output pulses. Recently, a new method was proposed to encode constant data into a temporal sequence of spikes, stimulating deeper studies in order to establish abilities and frontiers of this new approach. However, a well known problem of this kind of network is the high number of free parameters - more that 15 - to be properly configured or tuned in order to allow network convergence. This work presents for the first time a new learning function for this network training that allow the automatic configuration of one of the key network parameters: the synaptic weight decreasing factor.

Static converter: didactic support software

Teaching a course of special electric loads in a continuing education program to power engineers is a difficult task because they are not familiarized with switching topology circuits. Normally, in a typical program, many hours are dedicated to explain the thyristors switching sequence and to draw the converter currents and terminal voltages waveforms for different operative conditions. This work presents teaching support software in order to optimize the time spent in this task and, mainly to benefit the assimilation of the proposed subjects, studying the static converter under different non-ideal operative conditions.

The ZVS-PWM commutation cell applied to the DC-AC converter

This paper presents the analysis of a dc-ac converter using a zero-voltage-switching (ZVS) commutation cell. First, we show the cell applied to the buck converter. The stages of operation are presented along with the main current and voltage equations. Next, we adapt the converter to the regenerative-operation mode. Hence, the full-bridge converter at low-frequency operation is connected in the dc-dc output stage (at high frequency). The main switches commute at zero voltage. The converter operated at constant frequency with pulse-width modulation (PWM), and neither overvoltage nor additional current stress was observed by digital simulation. A design example and experimental results obtained by prototype, rated at 275 V and 1 kW, are also presented. © 1997 IEEE.

On optimizing micropower MOS regulated cascode circuits on switched current techniques

Trade-off between settling time and micropower consumption in MOS regulated cascode current sources as building parts in high-accuracy, current-switching D/A converters is analyzed. The regulation-loop frequency characteristic is obtained and difficulties to impose a dominant-pole condition to the resulting 2nd-order system are discussed. Raising pole frequencies while meeting consumption requirements is basically limited by parasitic capacitances. An alternative is found by imposing a twin-pole system in which design constraints are somewhat relaxed and settling slightly faster. Relationships between pole frequencies, transistor geometry and bias are established. Simulated waveforms obtained with PSpice of designed circuits following a voltage perturbation suggest a good agreement with theory. The proposed approach applied to the design of a micropower current-mode D/A converter improves its simulated settling performance.

Design of a LNA and a Gilbert Cell Mixer MMICs with a GaAs PHEMT technology

The design of a Gilbert Cell Mixer and a low noise amplifier (LNA), using GaAs PHEMT technology is presented. The compatibility is shown for co-integration of both block on the same chip, to form a high performance 1.9 GHz receiver front-end. The designed LNA shows 9.23 dB gain and 2.01 dB noise figure (NF). The mixer is designed to operate at RF=1.9 GHz, LO=2.0 GHz and IF=100 MHz with a gain of 14.3 dB and single sideband noise figure (SSB NF) of 9.6 dB. The mixer presents a bandwith of 8 GHz.

New 12kW three-phase 18-pulse high power factor AC-DC converter with regulated output voltage for rectifier units

This work presents a new high power factor three-phase rectifier based on a Y-connected differential autotransformer with reduced kVA and 18-pulse input current followed by three DC-DC boost converters. The topology provides a regulated output voltage and natural three-phase input power factor correction. The lowest input current harmonic components are the 17th and the 19th. Three boost converters, with constant input currents and regulated parallel connected output voltages are used to process 4kW each one. Analytical results from Fourier analyses of winding currents and the vector diagram of winding voltages are presented. Simulation results to verify the proposed concept and experimental results are shown in the paper.

Reconfiguracao de sistemas de distribuicao radiais utilizando o criterio de queda de tensao

Network reconfiguration in distribution systems can be carried out by changing the status of sectionalizing switches and it is usually done for loss minimization and load balancing. In this paper it is presented an heuristic algorithm that accomplishes network reconfiguration for operation planning in order to obtain a configuration set whose configurations have the smallest active losses on its feeders. To obtain the configurations, it is used an approached radial load flow method and an heuristic proceeding based on maximum limit of voltage drop on feeders. Results are presented for three hypothetical systems largely known whose data are available in literature and a real system with 135 busses. In addition, it is used a fast and robust load flow which decreases the computational effort.

A high speed 3.3V current mode CMOS comparators with 10-b resolution

This paper presents a high speed current mode CMOS comparator. The comparator was optimized for allows wide range input current 1mA, ±0.5uA resolution and has fast response. This circuit was implemented with 0.8μm CMOS n-well process with area of 120μm × 105μm and operates with 3.3V(±1.65V).

A new three-phase low THD power supply with high-frequency isolation and 60V/200A regulated DC output

This work proposes a new isolated high power factor 12kW power supply based on an 18-pulse transformer arrangement. Three full-bridge converters are used for isolation and to balance the DC-link currents, without current sensing or a current controller. The topology provides a regulated DC output with a very simple control strategy. Simulation and experimental results are presented in this paper.