PIPELINE SIMULATION-CONCLUSION MODELS BEAT DATA PROBLEMS, ACCURATELY TRACK AUSTRALIAN GAS SYSTEM
William J. Turner
William J. Turner Pty. Ltd.
Willetton, Australia
Prue A. Maguire
Csiro Division of Mineral & Process Engineering
Menai, Australia
Chris Smith
Leeds and Northrup (Australia)
Alexandria, Australia
Patrick McConnell
AGL Sydney Ltd.
Concord, Australia
Mike Severs
Pipeline Authority
Canberra
Despite difficulties caused by measurement data that were delivered less frequently than desired and by often inaccurate data, pilot testing of a computer simulation program on a high-pressure Australian network has shown the program able to model the pipeline well enough to track its current state and predict its next 16 days of operation.
This conclusion of a two-part series on SIROGAS covers details and results of its pilot testing in late 1991 that compared measured, tracking, and prediction results. The tests indicated that when the model receives valid data, good agreement is obtained with actual operation.
Development and use of SIROGAS are described in Part 1 (OGJ, June 21, p. 80).
The program simulates the steady state and transient behavior of single-phase fluid (principally gas) in pipeline networks. It is used by most operators of high pressure, long distance natural-gas pipelines in Australia.
In 1991, the program was installed on a new Leeds & Northrup Australia supervisory control and data acquisition (scada) computer at the Australian Gas Light Co.'s control center in Sydney.
SIROGAS' tracking model uses information provided by the scada system (pressures and flows, for example) to simulate each previous minute.
These data are measured at Moomba, Newcastle, and other places on the New South Wales (NSW) gas-pipeline network. (See map in Fig. 1 of Part 1.)
The tracking model also provides a starting state to simulate the operation of the network over the next 16 days. This is done routinely by the program's prediction model once each day and at other times as requested by the operator.
The tracking model performed well over the trial period under adverse conditions in which data on compressor operations were obtained only about twice a day and, during several periods, grossly incorrect measurement data were passed to the model.
In spite of these difficulties, the average pressure difference between the model and measured pressure at the line's terminus at Eastern Nitrogen Ltd. (ENL) north of Newcastle was less than 1%, with all excursions above 1% accounted for by known data failures.
PREINSTALLATION TESTING
The tracking model's point-identification (PID) file contains descriptions of the flow, pressure, and temperature measurement points in the network. Data for each measurement point include PID, type, location, and limits to be placed on the measurement value.
The PID file and pseudo-scada data required for testing each model at Csiro were generated with special SIROGAS programs developed for this purpose. Some of these data were passed to Leeds & Northrup for use in factory tests.
A special version of SIROGAS called RTMGEN produced the PID file and the scada data required by the tracking model.
This was done by either or both of the following procedures:
- Simulating the network and multiplying selected SIROGAS outputs by a random number to simulate the effect of measurement errors
- Changing some outputs to 9 x 109 to simulate the effect of extreme data failures.
The net effect of the simulated measurement failure was that about 10% of measurements were extreme at any one time and remained so for a few hours.
So that the model would be robust, Csiro versions of the tracking model's scada-interface routine were written to read the pseudo-scada data and PID lists from files and to control the calculation.
PID files and pseudo-scada data were also generated for other networks to test the capability of the tracking model to adapt to changes in the network and available measurements.
PID lists and scada data were generated for several networks with a similar generating program PREDGEN. A Csiro version of the prediction model's scada interface was written to read the pseudo-scada data and PID lists from files, thus impersonating the scada.
In the prediction model, the initial state of the network is determined from an historical restart file written by the tracking model for the same network. This file was generated during test runs of the tracking model.
Thus, in addition to generating the PID file and pseudo data required, testing the prediction model involved first generating data for and running the tracking model to produce this restart file.
INSTALLATION, TESTING
Fortran source and all data files including the pseudo-scada data used in preinstallation testing were transferred to the LN2068 computer that eventually became an additional module in the Leeds & Northrup distributed processing scada at Australian Gas Light Co. (AGL).
The pseudo scada real time data were used to test the tracking model. This testing in turn generated restart files that enabled the prediction model to be tested.
Acceptance tests were conducted by AGL staff from Monday, Nov. 18, to Wednesday, Nov. 27, 1991.
During this time the compressor station at Bulla Park, 362 miles downstream from Moomba, was running with pressure-discharge setpoint varying between 870 and 910 psia (Fig. 1). This variation was obtained by phone from the Pipeline Authority (TPA) and entered into the scada data base about three times a day.
Most of the New South Wales network is operated by TPA, an agency of the Australian government.
Except for 1 hr, the compressor station at Young, 646 miles from Moomba, was not used.
Thus, minute-by-minute information on the operation of the compressor stations was unavailable to the tracking model. For example, no change at Bulla Park was reported during the 2-hr shutdown of the Bulla Park station.
During the test period several events occurred that had a significant effect on the simulation (Table 1).
MOOMBA FLOW
At the beginning of the trial, the Moomba tracking model (Fig. 2a) was about 6% high because the ground temperatures specified to the model were recently changed from 77 F. to better estimates ranging from 71.2 F. at Moomba to 55.4 F. at Lithgow.
Cooling of the gas then occurred as heat transfer from the gas to the ground was simulated, causing pressures to fall and the flow from the constant pressure inlet at Moomba to rise.
The two cessations of flow from Moomba (Table 1) are clearly evident in both the tracking model and measured flows in Fig. 2a. The boundary condition is the measured pressure at Moomba (Fig. 1).
The tracking model finds that a flow of about 25 MMscfd is needed to follow the Moomba pressure dip. This difference from 0 is partly the continuation of the difference of about 13 MMscfd at the start of the trial and partly an effect of the limiting and filtering which stops the tracking model's boundary condition following the rapid pressure transient.
A drop in flow is also evident at Wilton in both the tracking model and the measurements (Fig. 2b) about 5 hr later.
Following the second Moomba shutdown, the model Moomba flow oscillates around the measured flow until it reaches close agreement late on Nov. 21. Then follows a period of good agreement until 10 a.m. on Nov. 25.
It is unclear why the agreement gradually degrades from then until 3 p.m. when the Bulla Park compressor tripped, but it seems likely that the lack of real time data from the station may be the main factor.
The compressor station trip at 3 p.m. caused a decrease in measured flows at Moomba and Wilton, which of course do not appear in the tracking model results because no information on the trip was passed to the model.
By about 4 a.m. on Nov. 26, the tracking model was again in good agreement with the measurements which continued until disconnection of the flow signal at Moomba during witness testing at 1:40 p.m. on Nov. 26. Shortly after this signal ceased, good agreement again prevailed until the end of the test.
WILTON FLOW;
ENL PRESSURE
At 10:40 a.m. on Nov. 20, the value for the Horsley Park flow used by the tracking model changed to 0, when in fact about 80 MMscfd continued to flow out of the network at this point.
Thus, the tracking model pressures began to rise relative to the measured pressures (Figs. 2c and 2d) and the tracking model flow at Wilton (Fig. 2b) 32 miles upstream drops by the same amount as the data error at Horsley Park.
This had no effect at Moomba because of the isolating effect of the compressor at Bulla Park.
The presence of a large load with a minimum pressure right at the end of the network-Eastern Nitrogen Ltd. (ENL) north of Newcastle-dominates operation of the network. Thus, prediction of this pressure and the confidence of the operations staff in that prediction are considerably important.
This confidence will only grow if the agreement between measured and tracking model's ENL pressures is seen to be reasonable.
The tracking model begins about 1.6% low because of changes made the previous Friday already discussed.
By the start of Nov. 19, however, the agreement is within 1% and remains so for the duration, excluding the effect of the Horsley Park measurement failure on Nov. 20.
PREDICTION MODEL;
PROBLEMS
The prediction model is automatically run at 6:15 a.m. each day to predict the network state over the next 16 days. Its most important use is to confirm that the forecast loads input to the model can be met at acceptable pressure.
Fig. 2 shows the results for the first 24 hr of each of the nine current-day prediction runs made each day during the test period.
Since the starting state is taken from the tracking model, each prediction model run agrees with the tracking model at 6:15 a.m. and tends to diverge from it over the following 24 hr shown in the graphs.
The level of agreement is of course mainly a matter of load forecasting and is not discussed here.
Among some minor problems noted in the acceptance report were the following:
- The lack of measured data on flow from laterals upstream of Wilton makes the modeling of these laterals too inaccurate.
Steps have since been taken to obtain hourly data on many of these flows, and negotiations are in progress to obtain data automatically from TPA.
- Time delays and omission of information on changes to compressors.
This information is now much better. Automatic supply of these data is also being negotiated.
- It is too difficult to enter predicted loads and events into the prediction model. A much improved method of data entry is being developed.
- The longer time required to run the prediction and hypothetical model on the scada computer (a shared 68030 microprocessor) compared to the previous prediction method LOOPY1 running on a large IBM mainframe.
REFERENCES
- Smith, R.H., "Practical Application of Transient Analysis to an Australian Pipeline System," Pipeline Simulation Interest Group, Oct. 30, 1986, New Orleans.
Copyright 1993 Oil & Gas Journal. All Rights Reserved.