ENBIS: European Network for Business and Industrial Statistics

My abstracts

The following abstracts have been accepted for this event:

• Short-Term Forecasting of National Imbalance Volume
  
  Authors: Shen Huang (EDF R&D)
  Primary area of focus / application: Modelling
  Secondary area of focus / application: Business
  Keywords: Energy markets, Time series, Forecasting, Model selection
  

  Submitted at 28-Mar-2018 16:55 by Shen HUANG
  Accepted
  

  3-Sep-2018 14:40 Short-Term Forecasting of National Imbalance Volume
  The balance between the production and the consumption of electricity at the national scale must be ensured at every moment. Each imbalance volume is physically compensated by the Transmission System Operator, but it is financially charged a posteriori via the electricity Balancing Mechanism. The design of imbalance volume prices aims to penalise parties who contribute to an overall imbalance and to be beneficial for parties who help the system equilibrium. So an accurate forecast of Net Imbalance Volume (NIV) at national level is critical to make trading decisions and to avoid being out of balance in the same direction to the overall system.
  
  In this talk, we are interested in the forecasting of UK-national net imbalance volume for the next few hours. Different statistical or machine learning models are tested such as linear regression, generalized additive models and random forest. The performance of tested models is evaluated on classical statistical criteria and also earnings/loss-based criteria, which is more meaningful for traders; according to our earnings/loss-based criteria, the annual financial gain made by our best forecast model is estimated to several million British Pounds. A comparison of those tested models will be provided and some further improvements of net imbalance volume forecasts will also be discussed.
• Variance-Sensitive Cost-Optimal Control Charts for Healthcare Data
  
  Authors: András Zempléni (Eötvös Loránd University, Budapest), Balázs Dobi (Eötvös Loránd University)
  Primary area of focus / application: Process
  Keywords: Control chart, Cost-effectiveness, Healthcare, Markov-chain
  

  Submitted at 28-Mar-2018 19:14 by András Zempléni
  Accepted (view paper)
  

  3-Sep-2018 14:40 Variance-Sensitive Cost-Optimal Control Charts for Healthcare Data
  Recently [1], we developed a Markov chain-based method for the economically optimal design of Shewhart-type control charts, which is suitable for real-life medical applications. In this model not only the shift size (i.e. the degradation of the patient's health) can be random, but the sampling interval (due to possible noncompliance) and the effect of the repair (i.e. treatment) too.
  Further developing the method, we introduce a target function to be minimised, which also incorporates the variance of the cost, which is often very important from a process control viewpoint. The resulting model requires several parameters to be estimated, and the accuracy of these estimations may have a significant effect on the results. Because of this, we investigate the sensitivity of the optimal parameters (the critical value and the sampling interval), and the resulting average cost and cost variance on different parameter values. We demonstrate the usefulness of the approach for real-life data of patients treated in Hungary – e.g. monitoring the cholesterol level of patients with cardiovascular event risk.
  Reference
  [1] B. Dobi, A. Zempléni, Cost-optimal Control Charts for Healthcare Data. 17th Annual Conference of the European Network for Business and Industrial Statistics, Naples, 2017.
• Global Sensitivity Analysis and Bayesian Calibration of a Clogging Numerical Model
  
  Authors: Bertrand Iooss (EDF R&D), Loïc Le Gratiet (EDF R&D), Guillaume Damblin (CEA), Sandrine Gyuran (CEA), Laurent Lefebvre (Framatome), Mathieu Segond (Framatome), Roberto Spaggiari (Framatome)
  Primary area of focus / application: Design and analysis of experiments
  Secondary area of focus / application: Modelling
  Keywords: Calibration, Clogging, Metamodel, Sensitivity analysis, Steam generators, Uncertainty
  

  Submitted at 29-Mar-2018 08:57 by Bertrand Iooss
  Accepted
  

  3-Sep-2018 15:30 Global Sensitivity Analysis and Bayesian Calibration of a Clogging Numerical Model
  Steam generators are massive heat exchangers transferring the heat from the primary to the secondary fluid to produce the steam needed by the turbines of nuclear power plants. After several years of operation, they can be subject to clogging that limits their heat exchange capacity and causes important economic and safety issues. In order to understand and predict this phenomenon, several non-destructive examinations (televisual data and eddy-current signals) have been gathered at various times of the heat exchanger operation, and a numerical mechanistic model has been recently developed by EDF.
  
  The objective of this work is to improve the modeling of clogging phenomenon to increase the predictive capability of the computer code. A global sensitivity analysis, based on Sobol’ indices, is first performed by the use of a neural network metamodel that has learnt on several runs of the computer code. By discussing the results with the clogging specialist engineers, this step helps improving the understanding of the clogging phenomenon. A Bayesian calibration of an epistemic model parameter is then applied in order to fit simulations to data. The resulting model allows compensating for physical phenomena not taken into account by the initial clogging numerical model.
• Performance of Bayesian Weibull Credible Intervals for Weibull Modulus in Small Samples
  
  Authors: Meryem Yalçinkaya (Kirikkale University), Burak Birgören (Kirikkale University)
  Primary area of focus / application: Reliability
  Keywords: Weibull modulus, Bayesian estimation, Classical estimation, Credible interval, Confidence interval, Prior elicitation, Monte Carlo simulation
  

  Submitted at 29-Mar-2018 09:23 by MERYEM YALÇINKAYA
  Accepted
  

  3-Sep-2018 15:10 Performance of Bayesian Weibull Credible Intervals for Weibull Modulus in Small Samples
  The Weibull modulus has been widely used in recent years to describe the statistical behaviour of mechanical properties of advanced materials. In the literature, various methods have been proposed for estimating confidence intervals for the Weibull modulus. This study proposes the Bayesian Weibull credible intervals as an alternative using the information that ceramic and composite materials have increasing failure rates, which requires the Weibull modulus to be more than 1. It provides a comprehensive comparison with existing methods for small samples. In the first step of the study, credible and confidence interval estimation algorithms have been developed for several methods such as variations of maximum likelihood, linear and weighted linear regression with interchanged axes, Bayesian Weibull methods with different prior elicitations and by using the prior information that the modulus is more than 1. In the second step, Monte Carlo simulations have been designed and run in the C++ language for the comparisons. The simulation studies showed that Bayesian methods have significantly better performance as compared to various classical methods.
• qgam: Quantile Additive Regression Models in R
  
  Authors: Matteo Fasiolo (University of Bristol), Yannig Goude (EDF R&D), Raphaël Nedellec (EDF R&D), Simon N. Wood (University of Bristol)
  Primary area of focus / application: Other: Electricity Data Analysis with R
  Secondary area of focus / application: Modelling
  Keywords: Quantile regression, Electricity load forecasting, Statistical software, Generalized additive models, R package
  

  Submitted at 29-Mar-2018 23:03 by Matteo Fasiolo
  Accepted (view paper)
  

  4-Sep-2018 15:50 qgam: Quantile Additive Regression Models in R
  The qgam R package offers tools for creating, fitting, checking and visualising non-parametric quantile additive models. Such models are an extension of traditional Generalized Additive Models (GAMs), and are particularly useful in the context of electricity demand forecasting. In fact, quantile GAMs inherit the flexibility and interpretability of GAMs (the latter property is essential in an operational setting), and have the additional advantages of not making any parametric assumption on the distribution of the response and of allowing modellers to focus directly on the quantile(s) of interest. qgam is an extension of the popular mgcv R package, which means that it can be used to fit quantile GAMs containing a wide variety of fixed, random or smooth effects (such as multivariate tensor product smooths, smooths on the sphere and adaptive smooths).
• Implementation of Standardised Multivariate Capability Indices in R
  
  Authors: Emilio L. Cano (University of Castilla-La Mancha), Matías Gámez Martínez (University of Castilla-La Mancha), Noelia García Rubio (University of Castilla-La Mancha)
  Primary area of focus / application: Process
  Keywords: Capability analysis, Multivariate process capability indices, Multivariate analysis, Statistical software, ISO standards, Statistical Process Control
  

  Submitted at 31-Mar-2018 16:18 by Emilio L. Cano
  Accepted (view paper)
  

  4-Sep-2018 09:00 Implementation of Standardised Multivariate Capability Indices in R
  Univariate capability indices are well known by industry as they are usually implemented in widely spread commercial software. However, in many industrial situations, Critical to Quality (CTQ) characteristics are multivariate. A plethora of multivariate capability indices can be computed, but the most common software packages lack of such computations. Nevertheless, such complex indices are defined both in scientific publications and in international standards. But such computations are not trivial, and it might be cumbersome, or even technically impossible, to be performed by busy industry teams. Thus, there is a gap,not only between Academia and Industry, but also between International Standards and Industry, which is a paradox. Open source software is an innovative way of filling that gap.
  
  In this work, the process capability indices for characteristics following a multivariate normal distribution defined in the ISO 22514-6 International Standard are implemented in the R Statistical Software and programming language. The methods are illustrated with the numerical examples included within the own Standard.Therefore, the Open Source developed software can be easily checked against the Standard, both via the source code and the results on example data. At the end, compliance can be assured to third parties, overcoming the main barrier for free-software-skeptical SPC practitioners.