![]() |
|||
![]()
|
![]() |
![]() |
![]() |
3.1.4. Forecasting Gas Demand British Gas plc. buys gas from 40 different suppliers, with distribution controlled by various control centers around the different regions. Gas is acquired at a steady rate in order to avoid large fluctuations at the wellheads. However, customers do not use gas at a steady rate and a varying consumption can mean a fivefold increase in demand between the morning and the evening. This variation is controlled by the use of regional gas storage. The gas company has to estimate how much gas is needed for the next day of service, and this estimation is used to order the gas before the next day (see Figure 2). One principal factor affecting the state of the demand for gas is the weather, as demand is high during the winter, especially if the winter is severe. In order to produce a good estimate, a good forecast for the future weather is needed. To aid in this task, the local Meteorological Offices for the regions send forecast data to the control centers in a standard format, which are used in the predictions. Inaccurate predictions will either produce a lower amount of gas than the demand, resulting in a shortage, or a higher amount than the demand, which will result in a waste of fuel. Current forecasting systems vary from region to region. At the time of the study (Leonard et al., 1996), the northern region used a Box-Jenkins approach employing PASCAL software and Lotus 1-2-3 spreadsheets, whereas other areas used databases to predict the next day's gas demand from historical data. Regression analysis methods were available in stand-alone software packages and macros within spreadsheets. However, no method was proven to be significantly better than the prediction obtained from an experienced Shift Control Officer (the domain expert).
At the time of the expert system development project, there were 72 expert forecasters spread geographically around the country in various regions. The experts worked different shift patterns and so could never be gathered in one place, even within a single region. Knowledge acquisition was used in two stages: first to get an overall picture of the national problem of forecasting gas demand, and second, to improve the forecasting within the northern region. Knowledge acquisition for the national picture consisted of the use of structured interviews and questionnaires. Four regional grid control centers were visited, and the knowledge acquisition sessions used a questionnaire as the basis of the interview. Questions were asked in order but interviewees were allowed to digress from the question. Questionnaires were also sent to all of the regions, and the replies were used to establish what methods and data each region used in their forecasting procedures. Knowledge acquisition for the northern region involved structured interviews, the card-sorting method, and the use of retrospective case descriptions. Structured interviews had two purposes; first, to explain the objectives of the expert system and, second, to discover each expert's individual technique in estimating the demand for gas. From the interviews it was established that all of the experts followed the same basic method in arriving at a forecast estimate. From the knowledge acquisition sessions, a model was derived to estimate the gas delivery for the next day's demand. The actual expert system was built using the Crystal expert system shell with the aim of making it generic for use by each region and also giving it the ability to interface with other prediction tools. The finished system was evaluated by a Shift Control Officer; after he had manually predicted the next day's gas forecast in his usual manner, he would then use the expert system to see what forecast it gave. The results showed that the expert system was more accurate in forecasting than the Shift Control Officer and also more accurate than any of the other forecasting tools (Leonard et al., 1996). 3.2. GAS TURBINE DIAGNOSTICSAdvanced gas turbines have offered substantial gains in firing temperatures, thermal efficiency, and electrical output. However, problems resulting in the shutdown of a gas turbine can be very serious, with disastrous safety, energy, and production cost implications. Gas turbines also have the feature of being able to start quickly, which makes them important for peak duty, i.e., to provide short-term power when there is a high demand. Failure to start on these turbines must be diagnosed quickly and efficiently (Milne, 1992; Armor et al., 1993). Expert systems have been developed to monitor gas turbine performance and condition, and to determine when a fault is occurring in its early stages before an alarm or trip (an emergency shutdown of the turbine) is triggered. Alarms themselves do not provide any in-depth information about a problem, which means that engineers must diagnose the problem manually in detail to determine what caused the alarm.
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |