Well Location Optimisation Methodology

Well location optimisation (WLO) typically requires 100’s of simulated forecasts for different well locations, despite using an optimisation algorithm (particle swarm optimisation in this case). For the field under study, a single full field model forecast simulation takes approximately 12 hours to complete and hence it would not be feasible to perform the optimisation directly on the full field model. Instead, the model was split into sectors of more manageable size with associated forecast taking only 10-25 minutes.

 The complete workflow consists of the following steps:

1. Create sectors in the full field model taking the well locations into account.

2. Run the full field model history and forecast with original well locations. 

a. The original well locations are created using the parametrized well trajectory with the custom workflow developed specifically for this study. This will create the base line for oil recovery.

b. It will also calculate the fluxes across sector boundaries at every timestep. Those fluxes are then used in each sector to simulate the in and out fluxes at the edges of the sector.

c. This step took approximately 72 hours of simulation time. 

3. Run the well optimisation for each sector.

a. The objective function to be maximised is the total oil recovery of the sector.

b. If there are more than one well to be optimised within a sector, the optimisation is done on one of the wells first, with the other wells at their original locations.

c. Then the next well is optimised, with the previously optimised wells at their optimised locations.

d. On average, it requires 100 iterations per well or about 24-48 hours of computation time per well.

4. Run the full field model with all the optimised well locations found in the previous steps.

The well optimisation part of the workflow (step 3 above) is shown in Figure 1.

The Particle Swarm Optimisation (PSO) algorithm was coupled with dynamic simulation to optimise well locations and trajectories. The objective function drives the optimisation algorithm forward. In this study, the objective function was the total cumulative oil of the given sector over a period of 10 years, i.e. from the first 2021/2022 well start-up. 

To optimise well locations and well trajectories, the following parameters were considered as variables (Figure 2):

1. Well pad surface location X and Y coordinates (X_SURF, Y_SURF, fixed for each well);

2. The maximum step-out from the well pad location to the reservoir target (STEPOUT, to be optimised);

3. The azimuth from the well pad to the subsurface target (AZ_SURF, to be optimised);

4. The top and bottom simulation layers of the target zone, as k1 to k2 layer number (K_TOP, K_BOT, fixed for each well);

5. The well length in the target layers (LW, to be optimised);

6. The azimuth at subsurface (AZ_SS, fixed).

A custom WLO script was developed using Python coding to generate the well trajectories based on the parameters outlined above. The script creates well trajectories and penetrations within specified target layers. For example, a completion can be constrained to reservoirs “layer 2” by defining “K_TOP” and “K_BOT” variables to correspond to the grid layers of “layer 2” as defined in the simulation model. In order to stay within the target layers along the entire well length, the well trajectory follows the layers topography, as illustrated in Figure 3 (left plot). For very long wells that target only a few simulation layers, it is challenging to stay exactly within the target layers and it was observed in some cases that layers above and below were penetrated in a few places (right plot, Figure 3). This was resolved by manual modification of the trajectory as a post-processing QAQC step.

The benefit of scripting is that the WLO script has been integrated into an automated workflow that allows to place the created well location and trajectory into the reservoir sector and launches the optimisation algorithm. This optimisation algorithm will search the variables space specified while trying to maximise the total oil recovery of the sector. 

Application to the field case and results:

Reservoir Context

The reservoir of interest for this study is a significantly sized; multi-layered and geologically complex carbonate field located in the North of Kuwait. The oil exhibits an API gravity varying between 20 – 28 degrees. The field started production in early 1958 with initial dry oil production rates from the discovery well. Sustained production began in the second half of 1960, extending until the present date. The current reservoir production is in the order of 130 Mbopd with an average water cut of ~60% and a GOR of 400 scf/stb from more than 300 active producing wells assisted by Electro-Submersible Pumps (ESPs). The total oil produced until end 2020 gives a current recovery factor of approximately 7%. The reservoir remained under primary recovery for close to 40 years of production; however, the general pressure behaviour of all the reservoir units showed poor to non-existent aquifer support. Therefore, waterflooding was started in 1997 through commingled vertical injectors located in the crest of the field. Presently there are around 100 active and mostly horizontal, injection wells; with a total average field injection rate of +360 Mbwpd.

Full Field Dynamic Model

The dynamic model used for the WLO application comprises 27 mln total grid cells; with 12 million active cells. Given the highly faulted nature of the reservoir, more than 800 faults were modelled; however, the big majority of them are open to cross-fault flow. The reservoir fluid description is complex; depicting API gravity increase with depth and 4 model regions really demarcating different oil types. Seven rock types were modelled; comprising different levels of permeability; from very permeable cemented cycle tops (high-perm, > 1 Darcy) to very fined grain and low permeability mudstones (< 10 mD). The saturation input to the model was done through the assignment of average J function and relative permeability tables allocated for the different rock types. The free water level (FWL) also varies considerably across the field; hence 8 equilibration regions exist in the model. 

Results

The WLO integrated workflow successfully identified more optimal locations for the producers in the SAMA simulation model using the PSO algorithms coupled with dynamic simulation. The WLO optimised well locations yield an expected recovery of 76% additional oil recovery with respect to the originally planned locations (Figure 4). 

Using a powerful multi-core CPU/GPU workstation partnered with the scalability of tNavigator’s simulation engine allowed an efficient evaluation of ranges of possible scenarios to be carried out. The project was completed within 2 weeks of equivalent computational time which is a significant acceleration compared to an approach of running optimisation on a full field model. Additionally, this is significantly more straight forward than the conventional manual location selection process, which could take months.