Real-Time Production Surveillance and Optimization in a Mature Subsea Asset

ExxonMobil

You have access to this full article to experience the outstanding content available to SPE members and JPT subscribers.

To ensure continued access to JPT's content, please Sign In, JOIN SPE, or Subscribe to JPT

A real-time production-surveillance and -optimization system has been developed to integrate available surveillance data with the objective of driving routine production optimization. The system aims to streamline data capture, automate data quality assurance, integrate high- and low-frequency data to extract maximum value, optimize the design and analysis of commingled well tests, and provide real-time multiphase well-rate estimates for continuous well‑performance evaluation.

Introduction

The technology was piloted in an offshore field consisting of stacked deepwater channel deposits developed with five individual subsea drill centers, 15 active oil producers, one gas injector, and five water injectors. Equipment is controlled remotely, and produced ­fluids are routed to surface by two 8-in. risers from each drill center. Produced gas is treated for use as fuel gas on the surface and for gas lift by reinjecting into each production well. All wells were originally deployed with a full suite of instrumentation, as well as valve- status sensors for each well flowline and riser. The three-phase separators on the surface use orifice plate meters for rate measurement.

The approach taken was to integrate real-time data and physical models for real-time production surveillance and optimization. The deployed system uses software that incorporates in-house-developed techniques for continuous model tuning. First, a complete integrated production model (IPM) of the production system was built, spanning reservoir inflow and wells on the one hand and flowlines, risers, and topsides on the other. The IPM is embedded in a field management software platform that houses several standard work flows as well as proprietary algorithms. Calculation results and real-time data are visualized with a visualization application.

As with other digital-oilfield applications, the key is continuous calibration of the physical models when field producing conditions change. The standard solution within the chosen field-­management software relies on calibrating the well models with single-producing-well tests. However, most of the subsea wells in the asset have lost the ability to flow alone, and only commingled well tests are available. Operators thus resort to a test-by-­difference protocol to obtain periodic well rates for allocation, which introduces significant errors when test conditions deviate largely from production conditions. To overcome this issue, a customized commingled well-test-analysis (CWTA) work flow was developed to analyze multisegment commingled well tests simultaneously.

Calibrated models coupled with real-time field sensor measurements allow for virtual flowmetering (VFM). Well rates can be calculated in multiple ways using the same sensor data but with different well-performance models. However, conventional VFM implementations treat each well separately and only use a limited number of sensors for any given well. In addition, many current VFM implementations require inputs such as water cut and gas/oil ratio (GOR) of the well from the last well test. The flow-rate estimates therefore will be biased as these values change over time. To overcome these issues, the chose software provides a multiwell allocation (MWA) work flow. Although MWA can close the overall material balance of the field, it relies strongly on the accuracy of the wellhead pressure (WHP) for each well.

In cases where many of the WHP sensors are either broken or have drifted, and where some wells have no working sensors at all, the VFM work flows provided by the chosen software cannot be used directly. To address this issue, a customized daily-well-rate-estimation (DWRE) work flow was developed, which, like MWA, also expresses the well-rate-­estimation problem as a nonlinear least-squares problem in which the productivity index (PI), water cut, and GOR are unknowns for each well. However, instead of using individual-well performance models to predict sensor measurements as in MWA, a network model is used and the reservoir pressures and separator pressures are set as boundary conditions. This reduces the strong dependence on individual-sensor data such that the work flow continues functioning even if sensors fail.

Implementation Overview

The deployed real-time production-surveillance and -optimization system consists of the standalone CWTA module and a software platform with DWRE. After commingled well tests are performed, the engineer uses CWTA to analyze the captured data, calibrate the model, and calculate ­individual-well rates. DWRE, anchored by the calibrated model, is embedded in the software as a scheduled work flow. The visualization software collects sensor data in real time and feeds them to DWRE, which calculates the individual-well rates with the model on a daily basis. The model is automatically recalibrated by DWRE whenever the calculated total flow rates deviate significantly from measured values.

CWTA

One crucial assumption that underlies the test-by-difference protocol is that the production rate of a well is constant across all segments. However, a historical look at commingled tests in this specific asset reveals that the downhole pressure of any given well can fluctuate by as much as 150 psi from segment to segment.

CWTA was developed to overcome this difficulty and to improve the estimates for the production rates of subsea wells. CWTA attempts to obtain estimates for unknown quantities at an asset by matching an integrated production model to data obtained over many segments of a commingled well test, the quantities in this deployment being the water cut, GOR, and PI of each well.

A graphical interface exists for CWTA to allow an engineer to configure the optimizer to analyze commingled well tests. Once a test is configured, it is fed into the optimizer to obtain an estimate for each unknown parameter. CWTA generates a report with summary statistics indicating the quality of the match that allows the engineer to go back and adjust the settings if the results are deemed unsatisfactory. Once a satisfactory result is obtained, the engineer chooses one final segment in the CWTA interface that is representative of normal operating conditions (as opposed to test conditions).

DWRE

CWTA is used to analyze well-test results and establish a calibrated model for DWRE. Between well tests, DWRE is then used to estimate multiphase-flow rates for each well and automatically recalibrate the model if the well parameters change from the preceding-well-test results. The DWRE work flow consists of three steps:

  1. Data import and segmentation
  2. Data-quality analysis/control
  3. DWRE calculation

DWRE is scheduled to run daily at the end of each allocation period to calculate rates for each well. The first step is to import real-time sensor data from the last 24 hours from PI through the IVM database. Together with well parameters, these inputs uniquely define the network model and field-operating conditions.

The second step is to post-process the sensor data for each segment. Extreme and unphysical values are removed, first on the basis of some simple rules. Then, a moving median filter is used to smooth all the data. The median value of the filtered data is chosen as the one representative value for each segment and is used as input to the model.

DWRE Historical Production Allocation: Synthetic and Real Data

Because only limited sensor data were available in the pilot field, synthetic data were used initially to verify the DWRE algorithm. Synthetic well parameters PI, warter cut, and GOR were generated from reservoir simulation; then, the model was used to generate 10 years of production data with the synthetic well parameters and actual field-operation conditions. With the true well parameters at well startup as the initial conditions, DWRE was used to calculate the well flow rates for the next 10 years.

After validating DWRE with synthetic data, real sensor data from the field were used. The results comparing the raw metered data with a traditional daily allocation method and a conventional VFM implementation where the model is anchored to well tests with no model recalibrations in between are plotted with DWRE results in Fig. 1.

Fig. 1—Historical total production allocation results from DWRE for the pilot field from the last 10 years.

 

Because of frequent single-well testing performed to 2008, results from all methodologies appear fairly consistent and well parameters were not changing dramatically. After 2008, how­ever, wells gradually lost their ability to flow individually up the risers and test frequency decreased.

As shown in the figure, both conventional allocation and VFM methods begin to deviate as conditions change more frequently between well tests. On the other hand, if DWRE is used to re­calibrate the model between well tests, the model prediction is almost the same as the metered data, showing the benefits of the DWRE work flow.

Post-Implementation Benefits

A study of the historical production allocation for each well was performed with CWTA by reanalyzing past commingled well tests.

Total field allocated volumes remained constant, but production-rate discrepancies on a well-by-well basis were identified and corrected. Leveraging the new, more-representative well rates, the reservoir team was able to improve the quality of their simulation models. Additionally, given the increased frequency of (virtual) well tests, the engineering team was able to justify reducing the number of actual well tests by 25%.

Automatic data streams were created to take advantage of the newly available high-frequency well rates and operating conditions being calculated in the real-time system by the network solver. Today, daily virtual well tests are submitted to the well-rate database to be used as historical production data for reservoir management.

More importantly, tuned production models enable real-time production optimization.

The engineering team leveraged the built-in production-optimization work flows in the digital-oilfield platform to identify candidate wells. During a 2-week field trial, a 2% production uplift was realized after performing a single well-routing change and reallocating available lift gas between four wells.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 181103, “Real-Time Production Surveillance and Optimization at a Mature Subsea Asset,” by Xiang Ma, Zachary Borden, Paul Porto, Damian Burch, Nancy Huang, Paul Benkendorfer, Lynne Bouquet, Peng Xu, Cassandra Swanberg, Lynne Hoefer, Daniel F. Barber, and Tom C. Ryan, ExxonMobil, prepared for the 2016 SPE Intelligent Energy Conference and Exhibition, Aberdeen, 6–8 September. The paper has not been peer reviewed.

Real-Time Production Surveillance and Optimization in a Mature Subsea Asset

01 March 2017

Volume: 69 | Issue: 3

STAY CONNECTED

Don't miss the latest content delivered to your email box weekly. Sign up for the JPT newsletter.