Oil Reservoir Production Optimization using Single Shooting and ESDIRK Methods

Transcription

1 Downloaded from orbit.dtu.dk on: Jun 7, 8 Oil Reservoir Production Optimization using Single Shooting and ESDIRK Methods Capolei, Andrea; Völcker, Carsten; Frdendall, Jan; Jørgensen, John Bagterp Published in: in Offshore Oil and Gas Production Link to article, DOI:.38/53--NO-4.3 Publication date: Document Version Publisher's PDF, also known as Version of record Link back to DTU Orbit Citation (APA): Capolei, A., Völcker, C., Frdendall, J., & Jørgensen, J. B. (). Oil Reservoir Production Optimization using Single Shooting and ESDIRK Methods. In in Offshore Oil and Gas Production (Vol., pp. 86-9). International Federation of. (IFAC Proceedings Volumes (IFAC-PapersOnline) ). DOI:.38/53--NO-4.3 General rights Copright and moral rights for the publications made accessible in the public portal are retained b the authors and/or other copright owners and it is a condition of accessing publications that users recognise and abide b the legal requirements associated with these rights. Users ma download and print one cop of an publication from the public portal for the purpose of private stud or research. You ma not further distribute the material or use it for an profit-making activit or commercial gain You ma freel distribute the URL identifing the publication in the public portal If ou believe that this document breaches copright please contact us providing details, and we will remove access to the work immediatel and investigate our claim.

2 Proceedings of the IFAC Workshop on Automatic Control in Offshore Oil and Gas Production, Norwegian Universit of Science and Technolog, Trondheim, Norwa, Ma 3 - June, FrBT.3 Oil Reservoir Production Optimization using Single Shooting and ESDIRK Methods Andrea Capolei Carsten Völcker Jan Frdendall John Bagterp Jørgensen Department of Informatics and Mathematical Modeling & Center for Energ Resources Engineering, Technical Universit of Denmark, DK-8 Kgs. Lngb, Denmark. ( {acap,cv,jf,jbj}@imm.dtu.dk). Abstract: Conventional recover techniques enable recover of 5% of the oil in an oil field. Advances in smart well technolog and enhanced oil recover techniques enable significant larger recover. To realize this potential, feedback model-based optimal control technologies are needed to manipulate the injections and oil production such that flow is uniform in a given geological structure. Even in the case of conventional water flooding, feedback based optimal control technologies ma enable higher oil recover than with conventional operational strategies. The optimal control problems that must be solved are large-scale problems and require specialized numerical algorithms. In this paper, we combine a single shooting optimization algorithm based on sequential quadratic programming (SQP) with eplicit singl diagonall implicit Runge-Kutta (ESDIRK) integration methods and a continuous adjoint method for sensitivit computation. We demonstrate the procedure on a water flooding eample with conventional injectors and producers. Kewords: Optimal Control, Optimization, Numerical Methods, Oil Reservoir. INTRODUCTION The growing demand for oil and the decreasing number of newl discovered significant oil fields require more efficient management of the eisting oil fields. Oil fields are developed in two or three phases. In the primar phase, the reservoir pressure is large enough to make the oil flow to the production wells. In the secondar phase, water must be injected to maintain pressure and move the oil towards the producers. In some cases, a tertiar phase known as enhanced oil recover is considered. Enhanced oil recover includes technologies such as in situ combustion, surfactant flooding, polmer flooding, and steam flooding (Thomas, 8). After the secondar phase, tpicall the oil recover is somewhere between % and 5% (Chen, 7; Jansen, ). Optimal control technolog and Nonlinear Model Predictive Control have been suggested for improving the oil recover of the secondar phase (Jansen et al., 8). In such applications, the controller adjusts the water injection rates and the bottom hole well pressures to maimize oil recover or a financial measure such as net present value. In the oil industr, this control concept is also known as closed-loop reservoir management (Jansen et al., 9). The controller in closed-loop reservoir management consists of a state estimator for histor matching and an optimizer that solves a constrained optimal control This research project is financiall supported b the Danish Research Council for Technolog and Production Sciences. FTP Grant no problem for the production optimization. The main difference of the closed-loop reservoir management sstem from a traditional Nonlinear Model Predictive Controller (Binder et al., ) is the large state dimension ( 6 is not unusual) of an oil reservoir model. The size of the problem dictates that the ensemble Kalman filter is used for state estimation (histor matching) and that single shooting optimization algorithms compute gradient based on adjoints (Jansen, ; Jørgensen, 7; Sarma et al., 5; Suwartadi et al., ; Völcker et al., ). In this paper, we propose a high order temporal integration method (Eplicit Singl Diagonall Implicit Runge- Kutta, ESDIRK) for forward computation of the initial value problem and for backward solution of the associated continuous-time adjoint. Conventional practice b commercial reservoir simulators is limited to the use of first order temporal implicit or semi-implicit integrators for the initial value problem and the adjoints. Völcker et al. (a,b, 9) introduce high order ESDIRK methods in two phase reservoir simulation. The high order scheme allows larger steps and therefore faster solution of the reservoir model equations. To compute the gradient of the objective function in a single shooting optimization method, Völcker et al. () propose a method based on adjoints for the discretized equations. Cao et al. () and Jansen () provide an overview of gradient computation using the adjoint. Brouwer and Jansen (4) and Sarma et al. (5) eplain and demonstrate gradient computation b the adjoint equations based on the implicit Euler discretization. Kourounis et al. () suggest the Copright held b the International Federation of 86

3 continuous-time high order adjoint equations for gradient computation in production optimization. Nadarajah and Jameson (7) compare gradients computed b discrete and continuous adjoints for problems arising in aerodnamics. The conclude that the gradients computed from continuous adjoints is accurate enough to be used in optimization algorithms. Since computation of gradients based on continuous time adjoints is faster than gradients based on discrete adjoints, this conclusion implies that the gradient computations can be accelerated b using the continuous time adjoint equations. The novel contribution in this paper is an etension of the adjoint based optimization method suggested b Völcker et al. () to include gradient computation based on the continuous-time adjoint equation. Using a conventional oil field as case stud, we demonstrate the new single-shooting optimization algorithm based on ESDIRK integration of the initial value problem and ESDIRK integration of the continuous-time adjoint equation. The case stud illustrates the potential of optimal control for production optimization of water flooded oil reservoirs b maimizing the net present value. We do a parameter stud to illustrate the sensitivit of the optimal solution to the discount factor. The paper is organized as follows. Section states the general constrained optimal control problem using a novel representation of the sstem dnamics. The ESDIRK algorithm for solution of the differential equation sstems is described in Section 3, while Section 4 presents the continuous adjoint method. Section 5 describes the numerical case stud and discusses the sensitivit of the optimal solution to the discount factor in the net present value. Conclusions are presented in Section 6.. OPTIMAL CONTROL PROBLEM In this section, we present the continuous-time constrained optimal control problem and its transcription b the single shooting method to a finite dimensional constrained optimization problem. First we present the continuoustime optimal control problem. Then we parameterize the control function using piecewise constant basis functions, and finall we convert the problem into a constrained optimization problem using the single shooting method. Consider the continuous-time constrained optimal control problem in the Bolza form tb min b )) + (t),u(t) t a Φ((t), u(t))dt (a) subject to (t a ) = (b) d dt g( (t) ) = f((t), u(t)), t [t a, t b ], (c) u(t) U(t) (d) (t) R n is the state vector and u(t) R nu is the control vector. The time interval I = [t a, t b ] as well as the initial state,, are assumed to be fied. (c) represents the dnamic model and includes sstems described b inde- differential algebraic equations (DAE). (d) represents constraints on the input values, e.g. u min u(t) u ma, c(u(t)), and some constraints related to rate of movement that are dependent on the input parametrization. Copright held b the International Federation of Path constraints η((t), u(t)) () ma render the optimization problem infeasible. For this reason and due to computational efficienc considerations when computing the sensitivities b the adjoint method (Capolei and Jørgensen, ; Jørgensen, 7), we include these constraints as soft constraints using the following smooth approimation χ i ((t), u(t)) = ( ) η i ((t), u(t)) + β i ) η i ((t), u(t) (3) to the eact penalt function ma(, η i ((t))) for i {,..., n η }.With this approimation of the path constraints, the resulting stage cost, Φ((t), u(t)), used in (a) consist of the inherent stage cost, Φ((t), u(t)), and terms penalizing violation of the path constraints () Φ(, u) = Φ(, u) + χ(, u),q + χ(, u),q (4). Discretization Control Parametrization Let T s denote the sample time such that an equidistant mesh can be defined as t a = t <... < t S <... < t N = t b (5) with t j = t a + jt s for j =,,..., N. We use a piecewise constant representation of the control function on this equidistant mesh, i.e. we approimate the control vector on ever subinterval [t j, t j+ ] b the zero-order-hold parametrization u(t) = u j, u j R nu, t j t < t j+, j,..., N (6) Input Constraints The input constraints (d) include bound constraints u min u k u ma. In the discrete problem using the zero-order-hold parametrization, we also include rate of movement constraints in the form u min u k u ma with u k = u k u k.. Single Shooting Optimization For the single shooting approach (control vector parametrization), we introduce the function ψ({u k } N k=, ) = { tb J = Φ((t), u(t))dt + ˆΦ((t b )) : t a (t ) =, d dt g((t)) = f((t), u(t)), t a t t b, } u(t) = u k, t k t < t k+, k =,,..., N (7) such that () can be approimated with the finite dimensional constrained optimization problem min {u k } N k= ψ = ψ({u k } N k=, ) (8a) s.t. u min u k u ma k N (8b) u min u k u ma k N (8c) c k (u k ) k N (8d) with N = {,,..., N }. 87

4 3. ESDIRK METHODS In this section, we describe our implementation of the ESDIRK method for the computation of ψ({u k } N k=, ) in (7). Computation of ψ({u k } N k=, ) consists of two major operations: ) For each integration step we first compute the model states (t) solving the initial value problem (c), ) and then we compute, using the same quadrature points, the value of the Lagrange term t ψ(t) := Φ((t), u(t))dt t a t a t t b. (9) in the cost function (a). Let t n denote the integration times chosen b the step size controller in the integrator. Each integration step size, h n, is chosen such that it is smaller than or equal to the sample time, T s. Therefore, one sample interval contains man integration steps. The numerical solution of the IVP (c) b an s-stage, stiffl accurate, Runge-Kutta ESDIRK method with an embedded error estimator, ma in each integration step [ t n, t n+ ] be denoted (Capolei and Jørgensen, ; Völcker et al., a) T = t n, T i = t n + c i h n (a) X = n (b) i φ i ({X j } i j=, u) = g(x ) + h n a ij f(x j, u) (c) j= g(x i ) = φ i ({X j } i j=, u) + h nγf(x i, u) (d) n+ = X s (e) s e n+ = h n d i f(x j, u) (f) j= with i =,..., s. X i denotes the numerical solution at time T i for i {,..., s}. n+ is the numerical solution at time t n+ = t n + h n. e n+ is the estimated error of the numerical solution, i.e. e n+ g( n+ ) g(( t n+ )). Subsequent to solution of (), we compute the numerical solution of the cost function (9) ψ( t n+ ) = ψ( t s n ) + h n b i Φ(X i, u) () i= When t n+ = t b, we add the Maer term of (a) such that ψ({u k } N k=, ) = ψ(t b ) = ψ(t b ) + ˆΦ((t b )) () The main computational effort in the ESDIRK method is solution of the implicit equations (d) using a Newton based method. (d) is solved b sequential solution of R i (X i ) := [g(x i ) h n γf(x i, u)] φ i ({X j } i j=, u) = (3) for i =,..., s. (3) is solved using an ineact Newton method. Each iteration in the ineact Newton method for solution of (3) ma be denoted M X [l] i = R i (X [l] i ) (4a) X [l+] i = X [l] i + X [l] i (4b) The iteration matri, M, is an approimation M J(X [l] i ) (5) to the Jacobian of the residual function J i (X i ) = R i (X i ) = g X i (X i) h n γ f (X i, u) (6) Copright held b the International Federation of The iteration matri, M, and its LU factorization is updated adaptivel b monitoring the convergence rate of the ineact Newton iterations. Convergence of the ineact Newton iteration is measured b R i (X [l] i ) = ma j,...,n (R i (X [l] i ) j < τ (7) ma{atol j, rtol j g j (X [l] i )} where atol is the absolute tolerance and rtol is the relative tolerance. Steps are accepted if this measure of the residual is smaller than τ. In case of divergence or slow convergence, the iterations are terminated, the step size, h n, is decreased and the Jacobian of the iteration matri is re-evaluated and factorized. As eplained in e.g. Völcker et al. (b) and Capolei and Jørgensen (), the step size controller adjust the temporal step sizes such that the error estimate satisfies a norm similar to the norm used in (7). 4. CONTINUOUS ADJOINT METHOD Gradient based methods such as sequential quadratic programming (SQP) methods for solution of (8) require the gradient of the objective function (7) with respect to the control vector parameters, i.e. ψ/ u k for k =,,..., N. In this section, we describe a continuoustime adjoint based method for computation of these gradients. Proposition. (Gradients based on Continuous Adjoint). Consider the function ψ = ψ({u k } N k= ; ) defined b (7). The gradients, ψ/ u k, ma be computed as ψ tk+ ( Φ f = λt u k u u t k ) dt k =,,..., N (8) in which (t) is computed b solution of (b)-(c) and λ(t) is computed b solution of the adjoint equations dλ T g f + λt dt Φ = (9a) ˆΦ ((t b)) + λ T (t b ) g ((t b)) = (9b) Remark. (Computation using ESDIRK). (t) is computed using the ESDIRK method applied to (b)-(c) and integration forwards. This solution is stored. The same ESDIRK method is applied for computation of λ(t) b solving (9) integrating backwards in time. Remark 3. (Gradients Computed b Continuous Adjoint). The gradients computed using the continuous adjoints are not the eact gradients, ψ/ u k, when the involved differential equations and integrals are computed b discretization using the ESDIRK method. However, the can be made sufficientl precise for the optimizer such that the do not affect the convergence (Nadarajah and Jameson, 7). The advantage of the continuous adjoint equations (9) is that the can be solved faster than the adjoint equations for the discretized sstem ()-(). 5. PRODUCTION OPTIMIZATION FOR A CONVENTIONAL OIL FIELD In this section, we appl our algorithm for constrained optimal control problems to production optimization in a 88

5 k min :.4 Darc, k ma :.3 Darc 3 4 Fig.. The permeabilit field and the location of wells. A circle indicates the location of an injector and a cross indicates the location of a producer Table. Parameters for the two phase model and the discounted state cost function (). Smbol Description Value Unit φ Porosit. - c r Rock compressibilit Pa ρ o Oil densit (4 atm) 8 kg/m 3 ρ w Water densit (4 atm) kg/m 3 c o Oil compressibilit 5 /atm c w Water compressibilit 5 /atm µ o Dnamic oil viscosit 3 Pa s µ w Dnamic water viscosit 3 Pa s S or Residual oil saturation. - S ow Connate water saturation. - n o Core eponent for oil.5 - n w Core eponent for water.4 - P init Initial reservoir pressure 4 atm S init Initial water saturation. - r o Oil price USD/m 3 r w Water production cost USD/m 3 conventional horizontal oil field that can be modeled as two phase flow in a porous medium (Chen, 7; Völcker et al., 9). The reservoir size is 45 m 45 m m. B spatial discretization this reservoir is divided into 5 5 grid blocks. The configuration of injection wells and producers as well as the permeabilit field is illustrated in Fig.. As indicated in Fig., the four injectors are located in the corners of the field, while the single producer is located in the center of the field. The specification of the two phase oil model consists of the injector (i I) and the producer (i P) location, the permeabilit parameters indicated in Fig., and the parameters listed in Table. The initial reservoir pressure is 4 atm everwhere in the reservoir. The initial water saturation is. everwhere in the reservoir. This implies that initiall the reservoir has a uniform oil saturation of.9. The inherent discounted stage cost function (see (4)) Φ(t) = Φ((t), u(t)) = (r ( + b) t/365 o ( f w ) f w r w ) q j (t) () j P accounts for the value of the oil produced minus the processing cost of the produced water. In this cost function, we have neglected the processing cost of injected water as well as the effect of pressure on injecting water. b is the discount factor. The fractional flow of water, f w = λ w /(λ w + λ o ), indicates the relative flow of water. λ w = ρ w kk rw /µ w log (K) (Darc) and λ o = ρ o kk ro /µ o are the water and oil mobilities, respectivel. In the problems considered, we do not have an cost-to-go terms, i.e. ˆΦ(t b ) =. Neither do we have an path constraints (). Therefore, maimizing the net present value of the oil field corresponds to minimization of tb J(t b ) = NPV(t b ) = Φ((t), u(t))dt () t a with Φ((t), u(t)) = Φ((t), u(t)). The optimizer maimizes the net present value b manipulating the injection of water at the injectors and b manipulation of the total fluid production (oil and water) at the producers. Hence, the manipulated variable at time period k N is u k = {{q w,i,k } i I, {q i,k } i P } with I being the set of injectors and P being the set of producers. For i I, q w,i,k is the injection rate (m 3 /da) of water in time period k N at injector i. For i P, q i,k is the total flow rate (m 3 /da) at producer i in time period k N. Therefore, at producer i P, the water flow rate is q w,i,k = f w q i,k and the oil flow rate is q o,i,k = ( f w )q i,k. The bound constraints (8b) appear in the production optimization problem because the water injected at injectors and the production at the producers must both be positive and because each production facilit has a maimum flow capacit. In the considered problem we have q i,k q i,k 5 i I P, k N (a) q i,k q ma i P, k N (b) The maimum flow capacit, q ma, is the same for all injectors and producers in this case stud. The rate of change for all injectors and producers are q i,k q i,k 5 for i I P and k N. Since the injection of oil is zero, q o,i,k = for i I, we get q w,i,k q w,i,k 5 for i I and k N. This leads to the rate of movement constraints (8c). In addition we use a voidage replacement constraint (Brouwer and Jansen, 4; Jansen, ) q i,k = q w,i,k = q i,k k N (3) i I i I i P and enforce a constant total injection, i I q w,i,k = q ma for k N. This translates into constraints of the tpe (8d). B the total injection constraint, the optimization problem reduces to a problem of redistributing the flows among the injectors. The prediction and control horizon is t b = 47 das and the sampling period is T s = 35. Hence the prediction and control horizon corresponds to N = periods. With a total injection at each time period of q ma = m 3 /da, these specifications corresponds to injection of.5 pore volume during operation of the reservoir. The prediction horizon is optimal in the reference case for a total injection of m 3 /da. The optimal water injection rates computed b solution of the constrained optimal control problem () for different discount factors, b, are illustrated in Fig.. In addition, a base case with constant and equal water injection rates is illustrated. It is evident that the optimal injection rates are ver sensitive to the discount factor, b. The corresponding cumulative oil and water production are plotted in Fig. 3. Independent of the discount factor value, the optimized strategies produce more oil than the base case. For the high discount factor case, b =., less oil is recovered than in Copright held b the International Federation of 89

6 CUMULATIVE PRODUCTION (m 3 ) 5 3 opt oil ref oil opt water.5 ref water.5.5 CUMULATIVE PRODUCTION (m 3 ) 5 3 opt oil ref oil opt water.5 ref water.5.5 CUMULATIVE PRODUCTION (m 3 ) 5 3 opt oil ref oil opt water.5 ref water TIME (DAY) (a) Discount factor b = TIME (DAY) (b) Discount factor b = TIME (DAY) (c) Discount factor b =.. Fig. 3. Cumulative oil and water productions for different discount factors, b. Million dollars opt. b= ref. b= opt. b=.6 ref. b=.6 opt. b=. ref. b= opt. b= opt. b=.6 opt. b=. reference opt. b= opt. b=.6 opt. b=. reference das (a) NPV das (b) Water cut das (c) Water fraction (kg water/kg fluid). Fig. 4. The net present value (NPV), water cut (accumulated water production per produced fluid), and the water fraction as function of time for the scenarios considered. Table. Ke indicators for the optimized cases. Improvements are compared to the base case. b NPV NPV Cum. Oil Oil Cum. water Water Oil Rec. factor Oil Rec. factor 6 USD % 5 m 3 % 5 m 3 % % %-point injector 3 (m 3 /da) injector (m 3 /da) time t (Das) 5 b= b=.6 b=. reference 3 4 time t (Das) injector 4 (m 3 /da) injector (m 3 /da) time t (Das) time t (Das) Fig.. Optimal water injection rates for different discount factors, b. the low discount factor case, b =. However, the produced oil is alwas above the reference case when b =.. This is not the case for b = and b =.6. Fig. 4 illustrates the net present value, the water cut and the water fraction for the base case scenarios as well as the optimized scenarios. The plot of NPV demonstrates that when b =, the NPV is lower than the base case NPV at some time during the production. At the end of the production the optimized NPV is largest. In order to recover the maimum amount of oil less oil must be produced at some times. This is also confirmed b the water fraction curves. The results are summarized in Table. Table shows that most oil is recovered in the case without discount (b = ), while least oil is recovered when the discount factor is high (b =.). Fig. 5 illustrates the evolution of the oil saturation for the optimized case (b = ) and the base case. The figures show that initiall, less oil is produced from the upper left corner in the optimal case compared to the base case. This gives a better sweep of the oil field and results ultimatel in higher oil recover. 6. CONCLUSIONS In this paper, we solve constrained optimal control problems using a single shooting method based on a quasi- Newton implementation of Powell s sequential quadratic programming (SQP) algorithm. The sstem of differential equations are formulated in a novel wa to ensure mass conservation and the resulting initial value problem (c) is solved with tailored ESDIRK integration methods. We also introduce a high order continuous adjoint sstem for efficient computation of the gradients. The algorithm is implemented in Matlab. Copright held b the International Federation of 9

7 TIME: 5 (Da) TIME: 35 (Da) TIME: 5 (Da) TIME: 35 (Da) TIME: 3 (Da) TIME: 47 (Da) TIME: 3 (Da) TIME: 47 (Da) (a) Optimal solution (b = ). (b) Reference solution. Fig. 5. Oil saturations at different times for the optimal solution and the reference solution. The resulting algorithm is tested on a production optimization problem for an oil reservoir with two phase flow. For all cases considered, the dnamic optimization increase the net present value of the oil field and give increased oil production. However, the optimal injection rates are ver sensitive to the discount factor. REFERENCES Binder, T., Blank, L., Bock, H.G., Burlisch, R., Dahmen, W., Diehl, M., Kronseder, T., Marquardt, W., Schlöder, J.P., and von Strk, O. (). Introduction to model based optimization of chemical processes on moving horizons. In M. Grötschel, S. Krumke, and J. Rambau (eds.), Online Optimization of Large Scale Sstems. Springer. Brouwer, D.R. and Jansen, J.D. (4). Dnamic optimization of waterflooding with smart wells using optimal control theor. SPE Journal, 9(4), Cao, Y., Li, S., Petzold, L., and Serban, R. (). Adjoint sensitivit analsis for differential-algebraic equations: The adjoint DAE sstem and its numerical solution. SIAM Journal on Scientific Computing, 4(3), Capolei, A. and Jørgensen, J.B. (). Solution of constrained optimal control problems using multiple shooting and ESDIRK methods. In American Control Conference. Accepted. Chen, Z. (7). Reservoir Simulation. Mathematical Techniques in Oil Recover. SIAM, Philadelphia, USA. Jansen, J.D., Douma, S.D., Brouwer, D.R., Van den Hof, P.M.J., Bosgra, O.H., and Heemink, A.W. (9). Closed-loop reservoir management. In 9 SPE Reservoir Simulation Smposium, SPE 998. The Woodlands, Teas, USA. Jansen, J.D., Bosgra, O.H., and Van den Hof, P.M.J. (8). Model-based control of multiphase flow in subsurface oil reservoirs. Journal of Process Control, 8, Jansen, J. (). Adjoint-based optimization of multiphase flow through porous media - A review. Computers & Fluids, 46, 4 5. Jørgensen, J.B. (7). Adjoint sensitivit results for predictive control, state- and parameter-estimation with Copright held b the International Federation of nonlinear models. In Proceedings of the European Control Conference 7, Kos, Greece. Kourounis, D., Voskov, D., and Aziz, K. (). Adjoint methods for multicomponent flow simulation. In th European Conference on the Mathematics of Oil Recover. Oford, UK. Nadarajah, S.K. and Jameson, A. (7). Optimum shape design for unstead flows with time-accurate continuous and discrete adjoint methods. AIAA Journal, 45(7), Sarma, P., Aziz, K., and Durlofsk, L.J. (5). Implementation of adjoint solution for optimal control of smart wells. In SPE Reservoir Simulation Smposium, 3 Januar- Feburar 5, The Woodlands, Teas. Suwartadi, E., Krogstad, S., and Foss, B. (). Nonlinear output constraints handling for production optimization of oil reservoirs. Computational Geosciences. doi:.7/s Accepted. Thomas, S. (8). Enhanced oil recover - an overview. Oil & Gas Science and Technolog, 63, 9 9. Völcker, C., Jørgensen, J.B., Thomsen, P.G., and Stenb, E.H. (a). Eplicit singl diagonall implicit Runge- Kutta methods and adaptive stepsize control for reservoir simulation. In ECMOR XII - th European Conference on the Mathematics of Oil Recover. Oford, UK. Völcker, C., Jørgensen, J.B., Thomsen, P.G., and Stenb, E.H. (b). Adaptive stepsize control in implicit Runge-Kutta methods for reservoir simulation. In M. Kothare, M. Tade, A.V. Wouwer, and I. Smets (eds.), Proceedings of the 9th International Smposium on Dnamics and Control of Process Sstems (DYCOPS ), Leuven, Belgium. Völcker, C., Jørgensen, J.B., and Stenb, E.H. (). Oil reservoir production optimization using optimal control. In 5th IEEE Conference on Decision and Control and European Control Conference. Orlando, Florida. Accepted. Völcker, C., Jørgensen, J.B., Thomsen, P.G., and Stenb, E.H. (9). Simulation of subsurface two-phase flow in an oil reservoir. In Proceedings of the European Control Conference 9, 6. Budapest, Hungar. 9