Simulation forecasts complex flow streams from Ekofisk

Fatos Çeçen Arnes, Henning Lillejord
Phillips Petroleum Co. Norway A/S
Tananger, Norway

Andrew Vieler
Simulation Engineering
Krimpen, The Netherlands

A commercial steady-state process flowsheet simulation program serves as the basis for a rigorous calculation model for predicting produced flow rates from the Ekofisk complex in the Norwegian sector of the North Sea.

The complex is the center of an extensive gathering system that collects oil and gas streams from several producing fields.

Prior to running a production forecast, the simulation model is initiated by matching several years of production. Once the simulation model matches historical production data within acceptable limits, it then is driven by production forecasts from reservoir simulations to develop long-term forecasts of gas, NGL, and oil production.

Ekofisk

Phillips Petroleum Co. Norway A/S (PPCoN) operates the Ekofisk field, the first field developed and one of the larger oil fields in the Norwegian sector of the North Sea. Ekofisk started producing in 1971.

Gas from Ekofisk is piped to Emden, Germany. The oil and NGL are sent to Teesside, U.K.

Since the original development, many nearby satellite fields and platforms have been linked into the Ekofisk complex. The complex itself has been extended several times with additional platforms and facilities.

Many operators and companies share the ownership of the various fields. The destinations of the production streams from the various satellite fields into the Ekofisk complex are also different, with some streams being processed in the main oil and gas separation stages, other streams entering the gas compression and condensate recycle sections, and still other streams feeding directly into the pipelines.

The Statpipe gas pipeline terminates at Ekofisk, and gas from it enters the Emden gas pipeline. A number of third-party gas lines also enter the Emden gas pipeline downstream from the Ekofisk production and treatment facilities.

The oil pipeline also serves a number of third parties whose oil production enters the Teesside oil pipeline downstream of the Ekofisk processing facilities.

Fig. 1 [53491 bytes] shows the current and future Ekofisk complex with satellite fields.

The Ekofisk oil and gas production facility includes several gas and oil separation stages with oil stabilization, condensate knock-out, gas compression, and an NGL stabilizer column. The NGLs are normally spiked into the oil stream to Teesside.

The current process includes several recycles of condensates back to separation stages, and several alternative routes for the feeds and the NGL stream produced by the NGL stabilizer.

The Teesside plant includes six parallel stabilization columns, which remove the NGLs from the oil stream from Ekofisk and produce a stabilized oil product. The NGLs then enter a separation plant, which produces a range of almost pure ethane, propane, and i and n-butane streams.

With the number of producing fields, owners, and operators, one of the more difficult problems faced by the Ekofisk complex operators is the allocation of production back to the various owners and forecasting of future production and reserves.

Over the years PPCoN has developed an in-house steady-state process simulation model for the surface facilities. Links from the reservoir simulations, which predicted production from each field over its lifetime, drove the simulation model and predicted the products from the surface facilities.

But model maintenance costs, and the rather complex links, which involved large Ascii batch input and output files, led PPCoN to change to a Hysim-based model. Hysim is a steady-state flowsheet simulator from Hyprotech Ltd.

A second reason for changing to Hysim was that PPCoN's process department uses it in general process simulation for process design and operations analysis.

Stream characteristics

In the original in-house model, thirty components-including nitrogen, carbon dioxide, standard saturated hydrocarbons, and some hypothetical hydrocarbons with characteristics between saturated straight-chain and unsaturated hydrocarbons-defined the fluid composition from each producing field.

It was decided to keep these definitions, and the components in the flowsheet models were simply standard library components or hypothetical components with some given properties. Hysim's standard methods calculated the other properties.

The historical data from each producing field include the oil, gas, and fuel flow rates. The reservoir forecasts predict flow rates and compositions of the total streams from the various fields. Fuel consumption for each producing field is also forecast.

These flow rates and compositions vary with time over a field's life. For the purposes of transferring to and using these data in Hysim, the gas and oil streams from the historical data were combined by the data preparation program for producing fields with a production separator. The total streams from the reservoir forecasts were used directly.

Process simulation

For a thermodynamic model, Phillips Petroleum Co. developed its own component and interaction parameter data bases for Hysim. These utilize the Soave-Redlich-Kwong (SRK) EOS (equation of state) model in Hysim. The original intention was to use the Hysim SRK model with the Phillips components and interaction parameters.

Some problems in matching the simulation case results with the in-house simulation program were found.

Eventually, it was found that the best match was achieved by using the standard SRK property package with thirty components and using the standard Hysim components and SRK interaction parameters except for a number of hypothetical components for heavier oil fractions.

Flowsheet models

The forecast period covers two phases of Ekofisk production: the current phase 1995-96, termed Ekofisk I, and the revised production scheme, Ekofisk II, which starts in 1998.

Ekofisk I has more producing fields feeding into the system but has a simpler process with two main recycle streams under normal operating conditions. Ekofisk II has a revised process scheme with a third condensate recycle stream and a turboexpander for cooling the feed to the NGL stabilizer.

Both flowsheets include a rigorous simulation model of the NGL stabilizer. The original in-house simulation program contained an equivalent of the Hysim fractionator model with predefined component-by-component cut-factors (or separation or recovery factors) that are derived from historical material balance data.

A rigorous column model has the advantage that it will account for changes in the feed and produced products that then meet the product specifications. A fractionation model will only split the feed components into two product streams with the user-specified component recovery factors.

It should be noted that the Teesside plant was initially modeled with simple fractionator models without representation of parallel trains and without removal of sour gas components from the off-gas streams. Several simulation model/system versions were developed ranging from the simple cut-factor models for all columns to a rigorous representation of the crude stabilizers with the other columns modeled as fractionator blocks and finally full rigorous simulation of the storage tanks from which off-gas is drawn as fuel, the crude oil stabilizers, and the fractionation trains.

The more rigorous models improved accuracy but the computational costs did not warrant the increased accuracy.

The complexity and number of unit operations in the simulation model were minimized to reduce calculation times. Some "tricks" were:

• Avoid wherever possible flash calculations where Hysim would be required to search for a temperature or a pressure. Examples are calculations of compressor outlet conditions, valve outlet conditions, and adiabatic mixing of multiple streams.

• Lump unit operation models where the outlet conditions (temperature, pressure) of a block of unit operations are known. An example is a compressor unit with a specified discharge pressure followed by an aftercooler with a specified outlet temperature and pressure drop.

• Reduce recycles by judicious placement of tears in the flowsheet. Tears placed in main process streams where stream conditions (temperature and pressure) are fixed is an example of a "good" tear location.

• Reduce calculations outside recycle loops by ordering the flowsheet or forcing the calculation order by ignoring unit operations immediately outside the recycle loops.

Thus, the surface facilities flowsheet model could be reduced almost entirely to a set of flash calculations and separations at known temperatures and pressures.

Fig. 2 [74250 bytes] shows an overview of the two simulation models.

Historical data

A centralized data base system contains averaged monthly operating data. The information required to drive the simulation model is extracted from this databank and written to files.

The historical production data include dates, production flow rates, oil and gas stream compositions, fuel flow rates, and separator conditions. Monthly data from the 2 years preceding the forecast data are included.

These data are then converted to Hysim command files. Hysim processes the command file, runs the simulations, and writes output in specific formats to files for each time-step.

After Hysim has run the simulations, comparisons between actual and calculated flow rates for product and significant intermediate streams can be made.

Reservoir data

The reservoir forecasts predict average flow rates for more than 30 years of future production. The first 4 years include monthly average flow rate forecasts and the remaining years include quarterly average flow rate forecasts.

Various sources provide the reservoir forecasts. Some are run by PPCoN and others are from other field operators or third parties. The forecast data are all made available on one computer.

The links to and from Hysim were developed using Hysim command files and Hysim calculator programs. Forecast data are collected and translated by a computer program into a Hysim command file. The input command file contains all data required to drive the simulation models for about 150 case studies per forecast.

The simulation models were designed to be initialized by a set of feed stream and unit operation parameters before being run in time-steps with exception specifications. Exception specifications mean that a simulation case will use data from a previous time-step unless this is altered by a command (user-input). This minimizes the volume of data written for and then processed by Hysim.

The main Hysim input command file contains the following information:

• Initialization data for streams and unit operations

• Year and month for a time-step followed by separator operating conditions if applicable, the production flow rates, and compositions

• Other data for each time step that may have altered a time-step

• Calls to the calculator programs to produce output for each time-step.

Once the input command file is complete, Hysim is called and processes the command file. The output from each time-step is written in the specified formats to a file with filenames being the month and year of the time-step. These files are then eventually concatenated into one large file for post-simulation processing. Users have the options of saving the individual data files for examination.

Stream identification

The entire system is data driven. The flowsheet models are data-driven as the standard operation models in Hysim are data-driven.

The flowsheet model streams, for which output information is required, are specified by the user in data files, which are read by the routines which produce the customized output. Thus, if a flowsheet model is altered by addition or removal of a stream, the user simply changes the data file rather than having to re-code routines or data statements in Fortran programs.

History match

The data required for history matching is relatively simple and includes stream names, mass, and standard volume flow rates.

Allocation data

PPCoN's product allocation and analysis program requires a complex set of information about many of the streams in the flowsheet models in a specified Ascii format. The information is generally not in standard flowsheet simulator output and uses calculation methods fixed in the field operation contracts.

Simple examples are:

• Stream component flow rates reported in an analysis up to pentanes and a lumped C₅+ fraction

• Heating values calculated using specific data

• Wobbe numbers reported for the gas product stream.

Hysim calculator programs analyze the data specification files, retrieve the information from the calculated flowsheet, perform the relevant calculations, and then write out the data to a file in the correct formats.

Implementation

Hysim models and calculator programs were developed using standard 486 personal computers operating under MS DOS. Initial model testing was also done on personal computers until the UNIX version of Hysim had been installed at PPCoN.

The initial machine was an IBM RS/6000 Model 570 which was replaced by an IBM RS/6000 Model 590. The entire system now runs on the latter machine with some post-processing possible on personal computers by exporting data files into spreadsheet or presentation graphics software.

All development work of the Hysim models and calculator programs were completed in The Netherlands while testing and implementation were done in Norway. Files were transferred back and forth between the two countries by simple modem transfer.

Testing

The initial Hysim flowsheet model tests involved direct comparison of results from the in-house simulation program and flowsheet models for single time steps. After resolving the problems with the thermodynamic models (this took a considerable amount of time and effort) the results were generally very close and therefore acceptable.

Once the Hysim models were running, the next step in the testing phase was to drive Hysim using the automatically generated command files for the history-matching period and then to compare results from simulations run with the in-house simulation program. The period covered by the initial history-matching included the years 1993 and 1994.

A data base program stored results and then created summary tables and plots of mass and volume flow rates of various intermediate and product streams for the history-matching period for the plant data, the in-house program, and the Hysim results.

For forecasting the models were driven by the command files generated, and comparisons were made between results from the in-house simulation program and Hysim. For these cases, no plant data were of course available for comparison for the initial forecast testing.

However, later during the development work, the first 6 months of production data from 1995 became available. Comparing both historical and forecast data for this period was then possible. Some discrepancies were found between historical and forecast data. This was clearly the result of the historical operations not following the forecast for 1 or 2 months. Nevertheless, the two surface facility models tracked each other; therefore, the test results were accepted.

Because the forecasts predict the total production over a long period, it was essential to show that the results generated by Hysim models did not differ from those produced by the previous system that was established and accepted by the field owners.

The initial testing and comparison included historical data from 1993-94 and forecasts for the period from January 1996 to the end of 2028. The tests established that the Hysim-based model results were within acceptable tolerances and, in fact, the Hysim-based simulation models with a rigorous simulation of the offshore NGL stabilizer produce better history-match than the models with fractionator blocks.

With acceptable accuracy from the simpler models and a need to minimize computation times required for the forecasts, it was decided not to add the rigorous distillation simulation of the Teesside plant to the forecasting models but to keep these models for off-line testing engineering and operation forecasting.

Final implementation

Scripts control the entire system. The user, before submitting the simulation run as a batch job, needs to specify whether a history-match or a forecast is to be run and to give the dates for the simulation runs as well as enter all the input parameters.

The four main steps in the batch process are:

1. Convert data from the in-house files. (Fortran program). The process creates a file called the Horizon file, which holds pressure, temperature, fuel, and third-party production feed data.

2. Convert all input data to Hysim commands.

3. Start Hysim as a separate process on the UNIX machine. The user can monitor the progress.

4. Read Hysim output, with the allocation program and allocate products back to owners. The allocated products are used in the long-term forecast-reporting system where all official reports are produced.

Fig. 3 [49664 bytes] shows an overview of the system.

Problems

Problems were experienced in getting the entire system to run properly. These problems were documented to serve as lessons that may benefit other Hysim users. The main difficulties involved the following:

• Interaction parameters-It was not possible to determine whether Hysim reads and/or correctly uses user component or interaction parameter data. It was found that the user interaction parameters were not being used after property packages had been switched for comparison purposes.

• Feed compositions and flows-As feed streams dropped to zero during the forecast, Hysim was more comfortable with feed streams to separators with small positive flow rates (1.0E-6) than with zero flow rates. Also, it was necessary to specify these streams to have non-zero compositions because zero compositions result in unknown compositions that cascade unknowns through the flowsheet.

• Reverse flows-On occasion, errors in the programs creating the command files allocated incorrectly high-flow rates to streams leaving the flowsheet. This usually resulted in negative flows in a flowsheet which then created problems in the post-processing. The positive aspect of this was an immediate flagging of an error.

• Matching cut factors with rigorous columns-Generally, it was reasonably easy to implement quality specifications in the columns, which would then produce fractionations similar to those calculated by empirical cut-factors. One general problem remains in modeling crude stabilization towers where the rigorous column models force more propane out of the top of the towers than is ever produced.

  Interestingly, use of the completely independent Hysim-based simulation models of the Teesside plant by the plant operating staff confirmed this anomaly. This is probably being caused by the thermodynamic models and interaction parameters between the propane and the heavier components, which were generated automatically by Hysim.

• Calculator function limitations-Certain data such as the time and date and version number of Hysim are not directly available to the calculator. A dummy file needed to be written and then re-read by the calculator to retrieve this information.

  This would not have been a hindrance if the re-entry to the calculator for further work could be achieved but it was found that, after processing a set of commands created by the calculator using the playback feature, Hysim required the calculator program to be terminated and a second started. The problem was eventually avoided by using script files on the UNIX computer and removal of the requirement of having the Hysim version number in the output.

• Unit operations outside recycles-Unit operations outside recycles are not always ignored by Hysim during recycle calculations. This can be a result of the order in which a flowsheet model is developed. A calculation sequence in Hysim can sometimes be altered by changing the order of the unit operations, or the order in which specifications are given.

  It is not always possible to find a way to force the calculation order to converge recycles first. These models unfortunately have this characteristic and it was found necessary to issue an ignore command to Hysim to ignore downstream unit operations, complete all recycle calculations, and then restore the ignored unit operations.

• Speed of calculator operations-Optimization of the calculator programs has not been done and some looping is relatively slow. Insufficient work has been done to establish exactly where the losses in speed occur.

• Quality assurance and integrity of Hysim cases-During the development and testing phases, it was found to be all too easy to make incorrect changes to Hysim models and then run simulations which then gave unexpected results. Occasionally, considerable detective work was required to determine where and why two Hysim cases with the same names in two different directories or on two separate computers had slightly different specifications.

The Hysim cases have been "protected" from these changes by the users of the overall system not needing to access Hysim directly at all. Any changes to the flowsheet models and associated data files are made by authorized personnel only.

History-matching results

The production flow rates calculated by the Hysim simulations being driven by historical data were compared with the measured production flow rates and with the production flow rates predicted by the in-house simulation program.

Comparisons of the average deviations of the main product flow rates for a number of simulations are shown in Fig. 4 [38943 bytes]. The bars in the graphs without crosshatching are from the simulation models selected, that is, the simulation models with rigorous representation of the Ekofisk NGL stabilizer and fractionation block models for the Teesside distillation columns.

Forecasting results

To check the forecast simulations, Hysim results were compared to the in-house program results. During the post-processing analysis of the simulation results, the production rates from each of the time steps are integrated over time to predict total reserves for all the fields in the entire system.

Table 1 [34090 bytes] summarize the differences in the results. It should be noted that the flow rates of the NGL product streams (ethane, propane, butanes) are small when compared to the main gas product and the stabilized oil stream so that very minor differences in the main product streams were the most important comparison criteria used.

The primary conclusion drawn was that the forecasting system based on Hysim simulation models could replace the previous models based on the in-house program. The Hysim models include a rigorous simulation model of the main NGL stabilizer, and this was found to improve history-matches.

In addition to the immediate use for product allocation and production forecasting, with the possibility of automatic generation of Hysim input for past and future operating scenarios, the system has application in the following areas:

• Examining production facilities by process and operations departments with a view to process improvements and debottlenecking

• Determining future process designs over many time steps, as opposed to the more usual designs based on average compositions and flow rates

• Examining past plant operations to investigate problems and performance

• Predicting future plant performance to predict possible capacity constraints. It is planned to develop calculator routines to allow automatic checking of equipment capacities and generation of warning messages and reports

• Obtaining "what-if" studies of future operations for situations such as a temporary shutdown or perhaps replacement of equipment

• Enabling off-line studies of the NGL production plant using the rigorous column simulation models combined with rigorous predictions of the plant feed stream.

Acknowledgment

The authors acknowledge permission to publish the above article from Phillips Petroleum Co. Norway and co-venturers, including Fina Exploration Norway S.C.A., Norsk Agip A/S, Elf Petroleum Norge AS, Norsk Hydro Produksjon a.s., Total Norge A.S., Den norsk stats oljeselskap a.s., Elf Rex Norge A/S, and Saga Petroleum a.s.

Bibliography

1. Hysim Version 2.50 Users Guides, Hyprotech Ltd., 1994.

2. Phillips Petroleum Co. Norway internal documentation.

3. Private correspondence with Phillips Petroleum Co. U.K. Ltd.

The Authors