SPE-176751-MS

Real-Time Monitoring of Asphaltene Deposition for Solvent Job Cycle

Optimization
J. P. de Boer, R. J. P. van der Linden, and W. A. Renes, TNO

Copyright 2015, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Middle East Intelligent Oil & Gas Conference & Exhibition held in Abu Dhabi, UAE, 15–16 September 2015.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
Asphaltene agglomeration and deposition is a complex process, leading to progressive production losses
without mitigative actions. Injection of asphaltenes inhibitors (solvents) in the tubing is a common
practice to avoid deposition in production tubing. This does however not avoid deposition in the
near-wellbore. Wells suffering from asphaltenes depositions are regularly shut in for ‘solvent jobs’,
injection of chemicals to dissolve asphaltenes in the (near-)wellbore. After a solvent job the well
performance slowly degrades again due to new deposition. The goal of this study was to optimize the
production-solvent cycle in terms of maximizing production profit, by means of a data-driven model
capturing both well and reservoir performance.
TNO and ENI have developed a grey-box model to capture the impact of asphaltenes deposition in the
near-wellbore on oil production. This effective PI-model is calibrated to actual field data and has shown
to predict the effect of deposition and the resulting production decline well. This paper describes the
background and modelling approach for both the grey-box reservoir deposition as well as the wellbore and
choke pressure drop – predicting near-future oil prediction based on current performance. Examples of
calibrating the grey-box deposition parameters are given and for the used well data the prediction window
shows to be consistent for at least 20 days ahead. The outcome of the developed functionalities is an
estimated ‘optimal cycle length’ and therefore a suggested next date of performing a solvent job at a
specific well.
The methodology was developed using production data from a well equipped with multiphase flow
meter (MPFM), but is applicable to wells with intermittent flow readings as well. The methodology can
handle any realistic trend in production fluid (GOR, WC) as well as operator input (choke, FLP back
pressure) and economic constraint (OPEX, solvent job cost relation).
The developed functionalities are applicable for real-time implementation, even without optimization
of the code in terms of computational speed. Current production data is available on an hourly basis,
wheras the calculation of the optimal solvent job cycle (including predicting oil production rates one
month ahead) takes a few minutes maximum.
A demonstrator implementation in OVS software can be used to optimize solvent job cycles on
individual well basis. The operator can through choke variations manipulate the predicted date of the next
optimal solvent job when scheduling solvent jobs at multiple wells (under equipment constraints).
2 SPE-176751-MS

Extension of this work can include optimization of an asset of wells, maximizing total production under
practical solvent job constraints.

Introduction and Problem Description

Traditionally, ENI and other oil producers faced with asphaltenes deposition issues have approached the
problem by studying in detail the asphaltenes molecules and its micro and macroscopic properties, such
as onset of precipitation, aggregation and deposition rates [1, 2]. The main objective of this project was
to attempt a different approach, where asphaltenes control strategies are evaluated based on models for
asphaltenes deposition which are based on actual production data.
The current procedure at ENI to mitigate production decline due to asphaltenes deposition is to perform
a solvent job when tubinghead pressure is below a specified value. During this solvent job the well is
shut-in and asphaltenes solving chemicals are injected to remove deposited asphaltenes. The wells are
completed with continuous solvent injection (tubing) so these intermittent solvent jobs focus on cleaning
the near-wellbore and well below the injection point. After the asphaltenes are (partially) removed the well
is reopened and production is continued until the tubinghead pressure is too low again. The cycle is then
repeated.
The objective of this study is to evaluate the feasibility to optimize the production/solvent job cycle
based on real-time production data. The resulting operational benefit is in reducing production loss due
to shut-in time and minize the amount (and related cost) of solvent chemicals used.
An overall (field scale) optimization is not included in this study. This might be needed in case of
limited equipment, when performing multiple solvent jobs simultaneously is not possible. This could be
solved by including a ‘delay-cost’ in the economic objective function of each individual well, as will be
described below.

Method

A grey-box modelling approach is used to capture the effects of asphaltenes deposition in the
near-wellbore on oil production. This means that very limited physics of the complex agglomeration/
deposition mechanisms is captured by the model. The model is however fitted to recent production data
and therefore expected to represent the near-wellbore behavior well.
The modelling and optimization is performed using MathWork’s MATLAB software. The non-linear
optimizer is available within the Optimisation Toolbox of MATLAB. Lift curves of the well were made
using OLGA steady state simulations, covering a wide range of WC, GOR, oil flow rates and topside
backpressure. The ultimate lift curve (e.g. at slightly differing WC, GOR, . . .) is interpolated from the
OLGA results to accurately represent field conditions.
For the presented well data continuous flow measurements were available from a wellhead multiphase
flow meter (MPFM). MPFMs are by no means standard well equipment so the ultimate workflow aims
on intermittent well test data only, to verify the flow prediction model.
The boundary conditions (inputs) to the model are:
– Reservoir pressure, Pres (t) – estimated
– Gas-oil-ratio, possibly varying over time, GOR (t) – estimated
– Watercut, possibly varying over time, WC (t) – estimated
– Choke setting, X(t) - measured
– Flowline pressure, Pfl (t) - measured
A schematic representation of the reservoir-well model is shown in Figure 1. This illustrates the model
boundary conditions of reservoir and flowline pressure (Pres and Pfl) and the pressure drops in between:
SPE-176751-MS 3

reservoir (dPres), wellbore (dPwell) and choke (dPchoke). The models for the three pressure drops are further
described below.

Figure 1—Schematic layout of reservoir-well model, indicating the chain-wise representation of production performance by the
reservoir pressure drop (red), wellbore pressure drop (blue) and choke pressure drop (purple).

Reservoir pressure may be taken constant or be approximated by a slow linear decline in time estimated
from pressure buildup tests. Future GOR and WC behavior may be extrapolated using recent historic data
and validated with regular test separation results. Flowline pressure and choke setting are both measured
and are free to be changed by the operator – as long as the specific scenario is provided to the prediction
algorithm to be taken into account. TNO and IPCOS have developed an OVS-based interface that allows
for monitoring and input of such scenarios, see further below.
The model has three parameters that are estimated from production data of historic solvent job cycles.
These parameters are:
– The starting value K0 immediately after a solvent job has been performed
– The production decline factor ␣
– The choke valve discharge coefficient cd
The inflow from the reservoir into the well is described as
(1)

The factor capturing the effect of asphaltenes deposition on inflow is effectively a time-dependent
reservoir productivity index (PI), since total deposition is a function of total production – here simplified
as a function of time since the previous solvent job:
(2)

All effects of asphaltenes deposition are concentrated in the term K(t) and incorporated in the inflow
model. The parameters K0 and ␣ of the function are estimated using a purely data-driven approach on the
observed production decline in previous cycles, see further below.
The lift curve is interpolated from a large set of OLGA steady state simulations capturing the pressure
drop as function of a wide range of wellhead pressure, oil flow rate, WC and GOR variations:
(3)

From equation (3) the bottomhole pressure is estimated. Therefore, Pbh measurements are not required
4 SPE-176751-MS

for this modelling approach, however they can serve as additional verification of model-based estimation.
The pressure drop and mass flow rate over the production choke are related as [3]:
(4)

The above equations are now able to describe both the pressure and flow throughout the wellbore
model, from the upstream boundary condition (reservoir pressure) down to the flowline pressure
downstream boundary condition, by calculating or using dPres, dPwell and dPchoke as illustrated in Figure
1. This enables prediction of future oil production rates, which is the base of estimating the optimal solvent
job cycle length. The objective function describes the economic balance of production and all involved
cost for one solvent job cycle. The optimization yields the optimal solvent job cycle length, based on oil
revenue (W1-terms), OPEX (W2-term), solvent job cost (W3) and cleaning shut in time Ts, see equation (5)
and (6). Additional terms (and stronger dependency on e.g. time) can be included, for example to optimize
cyclic production of multiple wells. The objective function for one well can be described as
(5)

Which can be rewritten as

(6)

The time index t refers to the current moment in time, t⫽0 refers to the restart of production after the
previous performed solvent job and T is the moment in time of the next solvent job. For a list of all
symbols, see Nomenclature. Furthermore, note that equation (6) in principle does not require flow rate
measurements: once a model has been benchmarked and accurately predict flow rates, one only needs the
predicted oil flow rate (Qoil, pred) which is based on real-time pressure measurements (Pfl, Pth, Pwh), using
equations (1) – (4).
The procedure to compute the optimal solvent job timing thus consist of the following steps:
1. Calculate the amount of total oil produced since last solvent job, IQoil(t)
2. Predict the future behavior of GOR
3. Predict the future behavior of WC
4. Predict the future behavior of Pres
5. Define future scenario for flowline pressure and choke setting (operator input)
6. Compute future oil production using the prediction model
7. Solve the optimization problem, resulting in optimal value of production cycle time, T
Note that the decline parameters (K0, ␣) are only re-fitted to production data when necessary. The
predicted oil flow rate (and Pbh) then differ significantly from actual field data. This can for example be
after a sub-optimally performed solvent job – at which K0 cannot be assumed to be ‘reset’ to its initial
value. This re-fitting is therefore not a default step, but included in the final effective workflow, see below
in Figure 2.
SPE-176751-MS 5

Figure 2—Daily workflow of solvent job optimization steps.

The consecutive calculation steps as described above can be formed into a workflow for optimal
production under cyclic solvent jobs. An example of a daily workflow is shown in Figure 2. The typical
cycle length between solvent jobs is 30-45 days for the example well, hence executing the workflow daily
is sufficient – but can also be performed at e.g. weekly well test intervals. The steps performed by the
workflow are:
A. Verification of prediction-model accuracy against most recent production data. This is done based
on recent pressure drop measurements over the choke and wellbore;
B. If this matches the expected decline curve, a new forecasting simulation can be performed. The
operator can indicate if over the next prediction window (e.g. 2 months ahead) choke or flowline
pressure changes are anticipated, so that the model can take this into account;
C. Actual calculation of predicted oil flow rate;
D. Based on the expected future oil flow rate, find the optimum time for the next solvent job in order
to minimize lost production during shut-ins;
E. If a defined accuracy threshold is exceeded, the operator is explicitly required to check of the
deviation from the expected decline is relevant or not;
F. If the deviation is acceptable, the most recent production decline curve (since the last solvent job)
is used to re-fit the decline parameters (K0 and ␣). With the new parameters, a new calculation of
predicted oil flow rate can be performed (B-D);
G. If the performance of the cleaned well is unacceptable, the physical well deposition cycle can be
‘reset’ by an additional solvent job, making sure all asphaltenes are dissolved.
As Figure 1. describes, the first step in the workflow is to compare the newest production data against
expected decline trends due to deposition. The production data consists of flowline pressure, choke setting
and (estimated) reservoir pressure, as well as reconstructed bottomhole pressure. The expected decline
trend is an estimate of bottomhole pressure and tubinghead pressure, based on predicted flow rate decline.
The flow rate decline is described by equation (2), a time-dependent Productivity Index (PI), here
labelled K. After performing a successful wash of the near-wellbore, the original PI is restored (K0). The
effective PI is a function of produced oil volume (here simplified as production time). Figure 3 shows
6 SPE-176751-MS

historic production data over 6 solvent jobs cycles. The first cycle (fit shown in red) is used to calibrate
the parameters (K0, ␣) that describe the effective PI, see equation (2). The top figure shows that the
estimated flow rate (green) matches the measured flow rate (blue) very well. This shows that the
assumption of restoring the effective near-wellbore PI after each performed solvent job is correct. Note
that it also shows that the operator performs the solvent jobs consistently and repeatable – this is key in
a model-based prediction method that relies heavily on field data.

Figure 3—Top: measured (blue) and modelled oil flow rate (green) based on productivity index (PI, or K) fitted to production interval
(red). Bottom: derived well productivity based on pressures (Pres, Pth, Pfl) and flow rates.

After relating the reservoir pressure drop and production flow accurately, the wellbore pressure drop
model was constructed. In this well multiphase flow is present and therefore OLGA steady state
calculations were preferred over simpler static correlations. To be able to predict the wellbore pressure
drop accurately for a wide range of operating conditions (WC, GOR, Qoil, p) an extensive OLGA
Parameter Study was performed, calculating the pressure drop under widely varying conditions. The
ultimate pressure drop is interpolated using MATLAB based on instantaneous WC, GOR, etc.
A (mass) flow rate estimate is obtained from the choke pressure drop, see equation (4). The choke
discharge parameter can be adjusted during well test intervals (or MPFM measurements) to calibrate this
flow estimation. Figure 4 shows once more the model-estimated oil flow rate (green), the MPFM-
measured flow rate (blue) and model-predicted near-future trend (red).
SPE-176751-MS 7

Figure 4 —Measured (blue) and model-estimated (green) oil flow rate. Based on historic data, a prediction of future oil flow rate is made
(red). Date of optimal next solvent job indicated by red circle.

To better illustrate the accuracy of the prediction, Figure 5 shows a “magic eye” prediction on date
03/10. The production data that later came available (“actual future”, blue) is compared against the
prediction of that date (red). The top figure shows the predicted oil flow rate, the middle figure shows the
predicted (and actual) tubinghead pressure and the bottomhole figure shows the prediction for the
(unmeasured) bottomhole pressure.
8 SPE-176751-MS

Figure 5—Top: Model-predicted oil flow rate (green) against actual measured flow rate (blue) and predicted (red). Middle: measured
tubinghead pressure (blue) against predicted (red). Bottom: measured bottomhole pressure (green) against predicted (red).

As soon as one is able to predict (near-) future production rates well, one can start evaluating until when
production can be economically sustained. Shutting in a well for solvent jobs too often results in high cost
of solvent chemicals and many days of down time. However producing for too long results in very low
production rates due to asphaltene deposition. The economic balance over one solvent job cycle
(consisting of production time and related revenue as well as solvent job cost and shut-in losses) is
described by equation (6) and solved for a global maximum using a constrained NLP solver. The outcome
of such optimization is a suggested date of performing the next solvent job, given current operating
conditions and a near-future scenario of the model boundary conditions (reservoir pressure decline, WC
and GOR-trends, choke variations etc.).
Figure 6 shows the result of such consequtive estimates. Starting on the 22nd of February (02/22) an
estimated optimal solvent job cycle length is 36 days. Knowing that the previous solvent job was at that
time 15 days ago (blue), this results in an estimated optimal solvent job date 21 days ahead (red). In the
daily updates since 02/22, the cycle length has been fairly constant, indicating that the deposition and
wellbore model perform consistently in predicting future production. Sensitivity analysis shows that
variations in the computed optimal cycle time are mainly caused by a changing trend in water cut.
SPE-176751-MS 9

Figure 6 —Estimated cycle length and predicted time until optimal solvent job time (top) and estimated oil flow rate at optimal moment
of solvent job time, against instantaneous oil flow rate (bottom)

TNO and IPCOS have developed a demonstrator interface in OVS software for the functionalities
described in this paper. The interface (see Figure 7 for an example view) allows for scheduling solvent
jobs for multiple wells based on their actual production data, as well as monitoring well performance in
real-time. For example, deviation in effective PI (Figure 4, bottom) can give early indications of a
sub-optimally performed solvent job, requiring a re-evaluation of model performance (Figure 3, step E).
Similar to Figure 6. It illustrates if today’s estimate are in line with earlier estimated optimal solvent job
dates, to visualize any ‘drift’ from earlier predictions.

Figure 7—Example of described solvent job optimization functionality as developed in OVS interface.

With a typical solvent job cycle of 30-45 days and hourly production data, the optimization calculations
can be implemented directly for real-time application. The typical calculation time – including model
10 SPE-176751-MS

retrieval from the model manager, production data preprocessing and overall optimization – is 1-2 minutes
per well, without any efforts to optimize calculation time thus far.

Conclusions
Asphaltene agglomeration and deposition is a complex process, leading to progressive production losses
without mitigative actions. Injection of asphaltenes inhibitors (solvents) in the tubing is a common
practice to avoid deposition in production tubing. This does however not avoid deposition in the
near-wellbore. TNO and ENI have developed a grey-box model to capture the impact of asphaltenes
deposition in the near-wellbore on oil production. This effective PI-model is calibrated to actual field data
and has shown to predict the effect of deposition and the resulting production decline well. The
assumption that the effective PI is restored by a solvent job has shown to be justified.
The methodology was developed using production data from a well equipped with MPFM, but is
applicable to wells with intermittent flow readings as well. The methodology can handle any realistic trend
in production fluid (GOR, WC) as well as operator input (choke, FLP back pressure) and economic
constraint (OPEX, solvent job cost relation).
The developed functionalities are applicable for real-time implementation, even without optimization
of the code in terms of computational speed. Currently data is available on an hourly basis, wheras the
calculation of the optimal solvent job cycle (including predicting oil production rates one month ahead)
takes a few minutes maximum.
The MATLAB-based modelling and optimization functionalities are embedded in an OVS (One
Virtual Source, [4]) environment to facilitate implementation in field operations. Production data retrieval
and access to well/reservoir model managers are included in a demo-interface that allows for scheduling
of solvent jobs for multiple wells – while updating forecasts as soon as new production data is available.
The implemented functionalities can be used to optimize solvent job cycles on individual well basis.
The operator can through choke variations manipulate the predicted date of the next optimal solvent job
when scheduling solvent jobs at multiple wells (under equipment constraints). Extension of this work can
include optimization of an asset of wells, maximizing total production under practical solvent job
constraints.
Field trials are currently being considered to further benchmark the performance of the optimizer and
production models.

Acknowledgements
We would like to thank ENI as sponsor for this project and for the permission to use representative
production data.

Nomenclature
␣ ⫽ Production decline parameter in K [1/day]
␳ ⫽ Density [kg/m3]
cd ⫽ Choke discharge coefficient [-]
t ⫽ Time [days]
tjob ⫽ Time of previous performed solvent job [days]
GOR ⫽ Gas-oil-ratio [Nm3/Nm3]
IQoil ⫽ Integrated oil flow rate since previous preformed solvent job [m3]
J ⫽ Objective function [€/day]
K ⫽ Production decline factor [-]
K0 ⫽ Initial value of K immediately after solvent job [-]
Pbh ⫽ Bottomhole pressure [bar]
SPE-176751-MS 11

Pfl ⫽ Flowline pressure [bar]

Pres ⫽ Reservoir pressure [bar]
Pth ⫽ Tubinghead pressure [bar]
Pwh ⫽ Wellhead pressure [bar]
T ⫽ Solvent job cycle production time [days]
Ts ⫽ Solvent job cycle shut-in time [days]
Qoil ⫽ Oil flow rate [Nm3/d]
W1 ⫽ Value of oil [€/Nm3]
W2 ⫽ Production operation cost [€/day]
W3 ⫽ Solvent job cost [€]
WC ⫽ Watercut [-]
X ⫽ Choke setting (between 0-1) [-]

References
1. Schutte et al (2015), Hydrodynamic Perspective on Asphaltene Agglomeration and Deposition,
Energy Fuels, 2015, 29 (5), pp 2754 –2767. DOI: 10.1021/ef50193h
2. Ramirez-Jaramillo et al, Modelling Asphaltene Deposition in Production Pipelines, Energy Fuels,
2006, 20 (3), pp 1184 –1196. DOI: 10.1021./ef050262s
3. SPE 109243, Al-safran, Predictions of Two-Phase Critical Flow Boundary, 2007
4. http://ovsgroup.com/workflows