Francisco Javier de Cos Juez

Bankruptcy forecasting: A hybrid approach using Fuzzy c-means clustering and Multivariate Adaptive Regression Splines (MARS)

During the last years, hybrid models have proven to be a promising approach for the design of classification systems for the forecasting of bankruptcy. In the present research we propose a hybrid system which combines fuzzy clustering and MARS. Both models are especially suitable for the bankruptcy prediction problem, due to their theoretical advantages when the information used for the forecasting is drawn from company financial statements. We test the accuracy of our approach in a real setting consisting of a database made up of 59,336 non-bankrupt Spanish companies and 138 distressed firms which went bankrupt during 2007. As benchmarking techniques we used discriminant analysis, MARS and a feed-forward neural network. Our results show that the hybrid model outperforms the other systems, both in terms of the percentage of correct classifications and in terms of the profit generated by the lending decisions.

Battery State-of-Charge Estimator Using the MARS Technique

State of charge (SOC) is the equivalent of a fuel gauge for a battery pack in an electric vehicle. Determining the state of charge is thus particularly important for electric vehicles (EVs), hybrid EVs, or portable devices. The aim of this innovative study is to estimate the SOC of a high-capacity lithium iron phosphate (LiFePO4) battery cell from an experimental dataset obtained in the University of Oviedo Battery Laboratory using the multivariate adaptive regression splines (MARS) technique. An accurate predictive model able to forecast the SOC in the short term is obtained and it is a first step using the MARS technique to estimate the SOC of batteries. The agreement of the MARS model with the experimental dataset confirmed the goodness of fit for a limited range of SOC (25-90% SOC) and for a simple dynamic data profile [constant-current (CC) constant-voltage charge-CC discharge].

Modeling a MEMS deformable mirror using non-parametric estimation techniques

Using non-parametric estimation techniques, we have modeled an area of 126 actuators of a micro-electro-mechanical deformable mirror with 1024 actuators. These techniques produce models applicable to open-loop adaptive optics, where the turbulent wavefront is measured before it hits the deformable mirror. The models input is the wavefront correction to apply to the mirror and its output is the set of voltages to shape the mirror. Our experiments have achieved positioning errors of 3.1% rms of the peak-to-peak wavefront excursion.

Using artificial neural networks for open-loop tomography

Modern adaptive optics (AO) systems for large telescopes require tomographic techniques to reconstruct the phase aberrations induced by the turbulent atmosphere along a line of sight to a target which is angularly separated from the guide sources that are used to sample the atmosphere. Multi-object adaptive optics (MOAO) is one such technique. Here, we present a method which uses an artificial neural network (ANN) to reconstruct the target phase given off-axis references sources. We compare our ANN method with a standard least squares type matrix multiplication method and to the learn and apply method developed for the CANARY MOAO instrument. The ANN is trained with a large range of possible turbulent layer positions and therefore does not require any input of the optical turbulence profile. It is therefore less susceptible to changing conditions than some existing methods. We also exploit the non-linear response of the ANN to make it more robust to noisy centroid measurements than other linear techniques.

Missing data imputation of solar radiation data under different atmospheric conditions

Global solar broadband irradiance on a planar surface is measured at weather stations by pyranometers. In the case of the present research, solar radiation values from nine meteorological stations of the MeteoGalicia real-time observational network, captured and stored every ten minutes, are considered. In this kind of record, the lack of data and/or the presence of wrong values adversely affects any time series study. Consequently, when this occurs, a data imputation process must be performed in order to replace missing data with estimated values. This paper aims to evaluate the multivariate imputation of ten-minute scale data by means of the chained equations method (MICE). This method allows the network itself to impute the missing or wrong data of a solar radiation sensor, by using either all or just a group of the measurements of the remaining sensors. Very good results have been obtained with the MICE method in comparison with other methods employed in this field such as Inverse Distance Weighting (IDW) and Multiple Linear Regression (MLR). The average RMSE value of the predictions for the MICE algorithm was 13.37% while that for the MLR it was 28.19%, and 31.68% for the IDW.

Deformable mirror model for open-loop adaptive optics using multivariate adaptive regression splines

Open-loop adaptive optics is a technique in which the turbulent wavefront is measured before it hits the deformable mirror for correction. We present a technique to model a deformable mirror working in open-loop based on multivariate adaptive regression splines (MARS), a non-parametric regression technique. The models input is the wavefront correction to apply to the mirror and its output is the set of voltages to shape the mirror. We performed experiments with an electrostrictive deformable mirror, achieving positioning errors of the order of 1.2% RMS of the peak-to-peak wavefront excursion. The technique does not depend on the physical parameters of the device; therefore it may be included in the control scheme of any type of deformable mirror.

Hybrid Intelligent System to Perform Fault Detection on BIS Sensor During Surgeries.

This paper presents a new fault detection system in hypnotic sensors used for general anesthesia during surgery. Drug infusion during surgery is based on information received from patient monitoring devices; accordingly, faults in sensor devices can put patient safety at risk. Our research offers a solution to cope with these undesirable scenarios. We focus on the anesthesia process using intravenous propofol as the hypnotic drug and employing a Bispectral Index (BISTM) monitor to estimate the patient’s unconsciousness level. The method developed identifies BIS episodes affected by disturbances during surgery with null clinical value. Thus, the clinician—or the automatic controller—will not take those measures into account to calculate the drug dose. Our method compares the measured BIS signal with expected behavior predicted by the propofol dose provider and the electromyogram (EMG) signal. For the prediction of the BIS signal, a model based on a hybrid intelligent system architecture has been created. The model uses clustering combined with regression techniques. To validate its accuracy, a dataset taken during surgeries with general anesthesia was used. The proposed fault detection method for BIS sensor measures has also been verified using data from real cases. The obtained results prove the method’s effectiveness.

Missing data imputation of questionnaires by means of genetic algorithms with different fitness functions

This article proposes a new missing data imputation method based on genetic algorithms. The algorithm presented in this paper is a useful tool for the completion of missing data in knowledge and skills tests. This algorithm uses both Bayesian and Akaikes information criterions as fitness functions and applies them to the classical item response theory models of one, two and three parameters. The results obtained by this new algorithm have been compared with those achieved by means of the Multivariate Imputation by Chained Equations (MICE) algorithm. For all the missing data ratios checked, the average incorrect imputation percentages obtained with the GA algorithm were, statistically, significantly lower than the results obtained with the MICE method. The most favorable frameworks for the use of the algorithm developed in the present research are those questionnaires in which missing answers would be considered as missing completely at random (MCAR). In other words, those questionnaires in which the same questions are present for all the examinees, but not necessarily in the same order. A genetic algorithm for missing data imputation is proposed.The algorithm is tested in the context of the item response theory.Optimum parameters of the algorithm are analyzed.The proposed algorithm performs better than MICE algorithm.

An ANN-Based Smart Tomographic Reconstructor in a Dynamic Environment

In astronomy, the light emitted by an object travels through the vacuum of space and then the turbulent atmosphere before arriving at a ground based telescope. By passing through the atmosphere a series of turbulent layers modify the lights wave-front in such a way that Adaptive Optics reconstruction techniques are needed to improve the image quality. A novel reconstruction technique based in Artificial Neural Networks (ANN) is proposed. The network is designed to use the local tilts of the wave-front measured by a Shack Hartmann Wave-front Sensor (SHWFS) as inputs and estimate the turbulence in terms of Zernike coefficients. The ANN used is a Multi-Layer Perceptron (MLP) trained with simulated data with one turbulent layer changing in altitude. The reconstructor was tested using three different atmospheric profiles and compared with two existing reconstruction techniques: Least Squares type Matrix Vector Multiplication (LS) and Learn and Apply (L + A).

Forecasting SO2 Pollution Incidents by means of Elman Artificial Neural Networks and ARIMA Models

An SO2 emission episode at coal-fired power station occurs when the series of bihourly average of SO2 concentration, taken at 5-minute intervals, is greater than a specific value. Advance prediction of these episodes of pollution is very important for companies generating electricity by burning coal since it allows them to take appropriate preventive measures. In order to forecast SO2 pollution episodes, three different methods were tested: Elman neural networks, autoregressive integrated moving average (ARIMA) models, and a hybrid method combining both. The three methods were applied to a time series of SO2 concentrations registered in a control station in the vicinity of a coal-fired power station. The results obtained showed a better performance of the hybrid method over the Elman networks and the ARIMA models. The best prediction was obtained 115 minutes in advance by the hybrid model.