A postdoctoral research position is available for a three-year, EPSRC funded, collaborative project in the Department of Mathematical Sciences, University of Durham, England, entitled:
"Bayes Linear Forecasting and Decision-Making for Large-Scale Physical Systems in the Petroleum Industry"
The objectives of this three-year research project, funded by the UK's Engineering and Physical Sciences Research Council, are (i) to develop a Bayes linear approach to forecasting and decision making using complex, high dimensional models for physical phenomena such as hydrocarbon reservoirs; (ii) to devise useful prior descriptions for the input and output uncertainties in such models, synthesising graphical elicitation tools for expert beliefs, theoretical analysis and the study of simplified forms for the models; and (iii) to combine (i) and (ii) to derive improved methods for asset management for hydrocarbon reservoirs and for decision problems such as leak detection for pipeline networks, and test these methods in a series of case studies. The project is in collaboration with Scientific Software-Intercomp (UK) Ltd, a leading software developer for the oil industry.
This project follows on from a recent EPSRC funded project held by the investigators which concerned strategies for matching hydrocarbon reservoir simulator output to actual reservoir production history.
The project involves a combination of statistical, numerical and computational skills and an ability to develop innovative solutions to challenging practical problems. Desirable qualities include experience of Bayesian methodology, modelling of complex stochastic systems and fluent computing skills. The appointee will be expected to collaborate closely with the investigators in all areas.
The position is for three years from 1st October 1996, or later by negotiation, and initial salary will be up to 15986 UK pounds on the RA(1A) scale. We welcome informal enquiries, which should be addressed either to Professor Michael Goldstein (e-mail: email@example.com, tel.: +44-191-374-2365) or to Dr Allan Seheult (e-mail: firstname.lastname@example.org, tel.: +44-191-374-2371). Application forms and further particulars are available from the Director of Personnel, Old Shire Hall, Durham DH1 3HP, England (e-mail: email@example.com, fax: +44-191-374-7253). Please quote reference A582. The closing date for applications is 19th July 1996.
For more information about the statistics group, the Department of Mathematical Sciences and about Durham, see the WWW page (URL http://fourier.dur.ac.uk:8000/stats/announce.html), which also has information about the project and a paper outlining a case study for this problem.
---------- Further details about Durham and the project follow ----------
Durham is a beautiful city occupying a magnificent site in a meander of the river Wear. Its cathedral (the finest example of Romanesque architecture in Britain) and castle (which became the first college of the university) are recognised as a supremely important part of our national heritage. Durham lies on the main North-South rail network (less than three hours by train from London), is just two miles from the A1(M) motorway and thirty minutes drive from Newcastle International Airport.
The University was founded in 1832 as the third university in England. Durham is a collegiate university. Its twelve colleges and two societies offer first class accommodation close to the academic departments, and to facilities for sport, music and drama. They help promote the University's very friendly atmosphere. The University of Durham is one of the UK's leading centres of scientific research. Mathematical Sciences (Applied Section), Physics, Chemistry and Engineering were all awarded the highest grade (5) research rating by the HEFCE in the last national research assessment.
Durham University Library contains an excellent collection of books, monographs and conference proceedings in statistics and other subject areas, and a good selection of current periodicals. In addition, it provides access to and training opportunities for a range of electronic information resources. These include catalogues of other libraries, bibliographic data sources such as BIDS and OCLC FirstSearch, and access to the rapidly expanding range of information (bulletin boards, online journals and others) available through the Internet.
The Department of Mathematical Sciences also has its own research library, the Collingwood Library, which contains a smaller collection of books and collected works primarily for research, and an undergraduate library.
The University and the Department of Mathematical Sciences together offer excellent computer facilities for calculation, document preparation and communication via the Internet and with external computing services. The facilities are provided on both UNIX workstations and PCs.
THE STATISTICS AND PROBABILITY RESEARCH GROUP
Members of the research group in statistics and applied probability are:
FPA Coolen (Bayesian reliability, software testing, foundations of statistics) PS Craig (spatial statistics, dynamic models, applied Bayes linear methods) M Goldstein (Bayes (linear) methods, decision theory, foundations) IM MacPhee (stochastic decision processes, applied probability) MD Penrose (geometric probability, percolation, interacting particle systems) AH Seheult (robust analysis of designed experiments, applied Bayesian methods) DA Wooff (Bayes linear methods, graphical models)
The interests of the group range over a wide range of topics associated with statistics and probability. These include:
Bayes linear methodology offers a systematic way of analysing uncertainty, based on the combination of statistical data and a linear analysis of limited aspects of expert judgements. Similar in spirit to other Bayesian approaches, it is often more straightforward to apply to complex problems. The approach addresses fundamental practical and philosophical issues about learning-based on partial knowledge. To implement the rich mathematical theory underlying this methodology, we have developed a general purpose programming language, which handles large practical applications by using graphical models to analyse and display information flow. You can find out about more about our development on the World Wide Web by accessing the Durham statistics home page (URL http://fourier.dur.ac.uk:8000/stats/home.html).
We are particularly interested in substantial research problems, arising from our contacts with other academic departments and with industry. Current projects include: computer experiments for history matching for hydrocarbon reservoirs; applications of Bayes linear methods to problems in oil and gas pipeline technology; sales forecasting in large competitive markets; industrial experimentation for quality control; collaborations with engineers developing hip replacements; with archaeologists using spectrometry to determine object composition; with medical physicists on measurement problems in dermatology.
Robust analysis of designed experiments:
Anomalous behaviour in data from designed experiments is common and can seriously affect their classical analysis. Robust methods are tailored to detect, highlight and accommodate any such unusual behaviour.
Other areas of methodology:
These include: modelling and inference for spatial phenomena; time series analysis and forecasting; statistics in the earth sciences; Bayesian reliability theory; software testing; expert judgements and uncertainty; and statistical selection.
Stochastic decision processes are used to model and guide decision making for systems with incomplete information. The use of semi-Markov decision processes together with modern optimisation methods has led to many successful applications to fields as diverse as admission to, routing within and collision resolution for various telecommunication networks; machine replacement and maintenance problems; various search and resource allocation questions.
Percolation and geometric probability:
Some rather easily described spatial random systems have deep properties, the analysis of which is relevant to the modelling of biological systems such as epidemics, and in statistical physics, and also to certain statistical tests. There is scope both for analysis and for computer simulation in this field of research.
We have regular seminars in statistics and probability, in conjunction with the University of Newcastle upon Tyne. Additionally, we have a more informal, internal seminar series. We have an expanding role in statistical consultancy, and recently established the University of Durham Statistics and Mathematics Consultancy Unit, which offers statistical and mathematical advice and analysis for all kinds of industrial, commercial, academic and institutional clients.
The investigators are statisticians Michael Goldstein, Allan Seheult, Peter Craig, David Wooff and numerical analyst Alan Craig. The statistics group in Durham University are leaders in the theory and applications of the Bayes linear methodology underpinning the proposed research. It is also a leading statistics group working on history matching and related problems in the oil industry. Our industrial collaborator, Scientific Software-Intercomp (UK) Ltd (SSI), is a leading software developer for the oil industry.
This project is concerned with problems of forecasting and decision-making based on information derived from large physical models, with particular application to the petroleum industry. These models are implemented as pieces of computer software known as simulators. A simulator takes as input a complete physical description of the system and instructions on system operation. However, some aspects of the physical description are not known. Even for the best possible choices, there will be differences between computed model behaviour and real system behaviour. Predictions and consequent decisions should take into account such uncertainties, but in many applications a single choice of physical description is made, with predictions and decisions based solely on that choice, perhaps supported by a simple sensitivity analysis. The choice is made by `history matching', where it is attempted to make simulated production match the historical record, by running the simulator a number of times with different inputs. More careful treatment, for example using a full Bayes analysis, is usually infeasible as the models are very high dimensional and slow to run.
Bayes linear strategies for history matching of hydrocarbon reservoirs were the subject of a previous EPSRC grant held by four of the applicants. A successful approach to history matching was developed, incorporating the various sources of uncertainty about the physical system by using computer based elicitation tools and the analysis of simplified versions of the simulators. This approach suggests a practicable real time methodology for forecasting and decision-making, as the simplified models are sufficiently fast that we may account for model uncertainties using repeat runs of these models to produce prior judgements as to forecast outcomes and good decisions, which are updated by selected evaluations on the full model. To develop this methodology, we must (i) construct detailed spatio-temporal prior descriptions relating the uncertainties in the models and the physical system, (ii) derive Bayes linear methodology for forecasting and decision-making using such high dimensional descriptions, and (iii) review the various elements of our history matching strategy, given our requirement to seek matches which give reliable forecasts or identify good decisions.
We shall apply the methodology to reservoir simulators and pipeline network simulators: two important types of model in the petroleum industry. The methodologies will be tested in case studies concerning asset management for oil reservoirs and real time decision problems such as leak detection for pipeline networks, as supplied by our industrial collaborator Scientific Software-Intercomp (UK) Ltd (SSI).
The petroleum industry will benefit from improved forecasting and decision-making methods for reservoirs and pipelines. Methodologies developed should generalise to a wide class of physical phenomena for which computer simulators are used to assist prediction and decision-making. Finally, there is general statistical and scientific value in developing Bayes linear methodology for such high dimensional decision problems, as the ideas will be transferable to many other areas where good decision-making requires careful use of expert judgements but the problem is too complex readily to allow a full Bayes solution.
PROGRAMME OF WORK
The work involves two related themes: (1) developing informative prior specifications, and (2) developing the Bayes linear approach to forecasting and decision-making for such complex models. These strands will proceed in parallel for the first eighteen months. As our approach builds strongly upon our previous work on history matching, we will also use this opportunity to review, critically, the various elements of that strategy, in the light of our requirement to seek matches which give reliable forecasts or identify good decisions. While our intention is to produce methodology which is widely applicable to general problems using physical models, the theory and models will be tested during this development upon trial problems in reservoir and pipeline management, suggested by SSI, for which it will be possible to compare our results with objective solution criteria.
The second eighteen month period will be devoted largely to two case studies in which we apply the methodology to selected problems for real physical systems, and to critical feedback to allow the development of the theory to react to case study results. The first case study will involve applying the methodology to a number of genuine oil reservoirs. The second will concern gas and liquid pipeline networks, including the problems of batch arrival forecasting and fast automatic leak detection and location systems, based on our experience of using dynamic linear models to monitor pipeline networks. The case study will be based on actual oil and gas networks owned by international petroleum companies collaborating with SSI.
#--------------------------------------------------------------------# | E-mail: P.S.Craig@durham.ac.uk Telephone: +44-91-3742376 (Work) | | Fax: +44-91-3747388 +44-91-3846041 (Home) | | | | WWW: http://fourier.dur.ac.uk:8000/stats/psc.html | | | | Snail: Peter Craig, Dept. of Math. Sciences, Univ. of Durham, | | South Road, Durham DH1 3LE, England | #--------------------------------------------------------------------#