In 1985, Northwestern Utilities Limited (NUL), Edmonton, Alberta, Canada, initiated the development of the Northwestern Utilities Gas Operations Support System (NUGOSS). This real-time gas pipeline modeling system provides for monitoring of real-time operations, predictive network behavior analysis, and short-term demand forecasting. This paper describes the development and implementation of the real-time and predictive gas load models required for NUGOSS. The development of mathematical models for generation of real-time loads at unmonitored system delivery points and 24 hour forecasting of future gas loads is presented. These models are based on statistical analysis of the hourly temperature and gas load data for seven City of Edmonton area gas load sites from October, 1986 to March, 1987. Linear regression models were developed to estimate real-time gas load demand based on time and temperature for non-telemetered stations. These models are used to calculate temperature-related differences In gas demand between stations that display similar load patterns. This "Load Generator" model predicts real-time values by scalar association of the non-telemetered station to a telemetered station. The development of a "Load Forecaste P linear regression model that is used to predict energy load profiles inadvance is then discussed. The model explains 80 X to 89 X of the load values. The model uses time, previous 24 hour load values and temperature values, and temperature forecasts. All equations for both the Load Generator and Load Forecaster models werestatistically significant. Comnents on model performance and suggestions for future development are noted. Recornendations to address a variety of improvements are suggested including treatment of expected non-linear gas load changes at cold temperatures and additional environmental parameters.
The incorporation of real-time modeling represents a new development in gas pipeline monitoring and control systems. Successful application of these new techniques requires knowledge of the current and expected future state of the pipeline network and impacting factors. While direct measurement of all important modeling factors would be desirable, this is not economically feasible for some gas delivery networks. The alternative is to estimate current or future gas loads usingpredictive toolsthat can then be used by a real-time model.