Probability Estimation and Data Analysis in Engineering

Modelling of Supermarket Energy Control Systems and its Application in Design Integration and Optimisation Dr Yunting Ge Conventional supermarket refrigeration systems are responsible for considerable CO2 emissions due to high energy consumption and large quantities of refrigerant leakage. In the effort to conserve energy and reduce environmental impacts, an efficient design tool for the analysis, evaluation and comparison of the performance of alternative system designs and controls is required. In this presentation, a detailed development procedure of supermarket simulation model ‘SuperSim’ is described. The model which has been validated against measurement data from an operational supermarket has been used for a number of significant applications. These include performance comparisons of refrigeration systems with different refrigerants, integration of a trigeneration system and employment of heat recovery etc. High Performance Data Analysis for Smart Grids Dr Maozhen Li Smart grids are a term that refers to modernisation of current electric power grids through the integration of information technology, digital communication, sensor devices and Internet. Smart grids use two way digital communication with advanced sensors to collect and analyze data for transforming the existing electric grid into intelligent, dynamic self-healing, selfoptimizing transmission and distribution grid. Smart grids are characterised by the exponential growth of data to be collected from many high frequency devices such as phasor measurement units (PMUs) and smart meters. In this talk, we will introduce the recently EPSRC funded project on "Developing Scalable Smart Grid Infrastructure to Enable Secure Transmission System Control" which is in collaboration with China involving a number of industrial partners such as the UK National Grid, Alstom Grid Ltd, Intel, China Southern Power Grid. We will look at how high performance computing technologies such as Cloud computing facilitate massive and scalable data analysis for smart grid systems. The Art of PD Curve Calibration Dr Dirk Tasche PD curve calibration refers to the task of transforming a set of conditional probabilities of default (PDs) to another average PD level that is determined by a change of the underlying unconditional PD. This paper presents a framework that allows to explore a variety of calibration techniques and the conditions under which they are fit for purpose. We test the techniques discussed by applying them to a publicly available dataset of agency rating and default statistics that can be considered typical for the scope of application of the techniques. We show that the popular technique of 'scaling the PD curve' is theoretically questionable and does not perform well on the test datasets. We identify two calibration techniques that are both theoretically sound and perform much better on the test datasets. Nonparametric predictive inference for ordered three-class ROC analysis with continuous measurements Dr Tahani Coolen-Maturi Receiver operating characteristic (ROC) curves are widely used to assess the performance of a binary classifier. ROC curves have been used in many fields such as signal detection, medicine, radiology, biometrics, machine learning, data mining and credit scoring. ROC surfaces (3D surfaces) are currently used to assess the performance of three-class classifiers. Classification of a given (future) observation to one of three classes is an important task in many decision making problems. We present the nonparametric predictive inference (NPI) approach to threeordered classes ROC analysis, including results on the volumes under the ROC surfaces and consideration of the choice of decision thresholds for the classification. Parameter Inference for discrete Weibull distribution Dr Veronica Vinciotti Discrete Weibull distribution is useful to model discrete variables in a number of applications, including discrete failure data. Very little work has been done for estimating the parameters of this distribution and, in particular, for building confidence intervals. In this talk, we present a method to derive exact confidence intervals for the parameters of this distribution. The method has potential to be applied to other discrete-type distributions, where bootstrap or asymptotic confidence intervals are often employed. Probabilistic Failure Analysis of Underground Pipes