Systems and methods for reducing reservoir simulator model run time

1. A method comprising:

determining, by a processor, a plurality of decision variables for a reservoir simulator model configured to simulate a reservoir, wherein each decision variable among the plurality of decision variables is a manipulatable parameter that effects how the reservoir simulator model simulates the reservoir;

determining, by the processor, a base case by executing the reservoir simulator model using a set of predefined values for the plurality of decision variables as input, the base case representing production rates or injection rates over a predetermined time period;

determining, by the processor, a total number of time steps based on the production or injection rates from the base case;

determining, by the processor, a plurality of triangular distributions by, for each time step over the total number of time steps:

calculating a respective triangular distribution for each respective decision variable among the plurality of decision variables, wherein the respective triangular distribution includes (i) a minimum value for the respective decision variable, (ii) a maximum value for the respective decision variable, and (iii) a most likely value for the respective decision variable;

generating, by the processor, reservoir simulator model input data by executing a stochastic simulation algorithm that is constrained by (i) a predetermined minimum number of cases, (ii) the total number of time steps, and (iii) the plurality of triangular distributions;

executing, by the processor, the reservoir simulator model using the reservoir simulator model input data as input to produce corresponding output data as output from the reservoir simulator model;

generating, by the processor, training data that includes relationships between the reservoir simulator model input data and the corresponding output data from the reservoir simulator model;

creating, by the processor, a neural network that is separate from the reservoir simulator model;

training, by the processor, the neural network using the relationships expressed in the training data to enable the neural network to serve as a proxy model for the reservoir simulator model, wherein the neural network is more quickly executable than the reservoir simulator model; and

executing, by the processor, the proxy model to generate a forecast indicating a characteristic of the reservoir at a future point in time.

2. The method of claim 1 , wherein the minimum value is selected and represents a value for a choke in a completely closed position at a respective time step.

3. The method of claim 1 , wherein the maximum value is selected and represents a value for a choke in a wide open position at a respective time step.

4. The method of claim 1 , wherein the most likely value is selected based on each time step in the total number of time steps being identical to each time step in the predetermined time period.

5. The method of claim 1 , wherein the most likely value is calculated using the production rates or injection rates from the base case.

6. The method of claim 1 , wherein the predetermined minimum number of cases is at least ten times the number of decision variables in the plurality of decision variables.

7. The method of claim 1 , wherein the stochastic simulation algorithm includes a Monte Carlo algorithm, an orthogonal array algorithm, or a Latin hypercube algorithm.

8. The method of claim 1 , further comprising creating the neural network at least in part by determining a total number of nodes for the neural network based on a total number of proxy model input variables.

9. The method of claim 1 , further comprising adjusting a setting associated with an asset-production tool based on the forecast.

10. A non-transitory program carrier device tangibly carrying instructions executable by a processor for causing the processor to perform operations including:

determining a plurality of decision variables for a reservoir simulator model configured to simulate a reservoir, wherein each decision variable among the plurality of decision variables is a manipulatable parameter that effects how the reservoir simulator model simulates the reservoir;

determining a base case by executing the reservoir simulator model using a set of predefined values for the plurality of decision variables as input, the base case representing production rates or injection rates over a predetermined time period;

determining a total number of time steps based on the production or injection rates from the base case;

determining a plurality of triangular distributions by, for each time step over the total number of time steps:

calculating a respective triangular distribution for each respective decision variable among the plurality of decision variables, wherein the respective triangular distribution includes (i) a minimum value for the respective decision variable, (ii) a maximum value for the respective decision variable, and (iii) a most likely value for the respective decision variable;

generating reservoir simulator model input data by executing a stochastic simulation algorithm that is constrained by (i) a predetermined minimum number of cases, (ii) the total number of time steps, and (iii) the plurality of triangular distributions;

executing the reservoir simulator model using the reservoir simulator model input data as input to produce corresponding output data as output from the reservoir simulator model;

generating training data that includes relationships between the reservoir simulator model input data and the corresponding output data from the reservoir simulator model;

creating a neural network that is separate from the reservoir simulator model;

training the neural network using the relationships expressed in the training data to enable the neural network to serve as a proxy model for the reservoir simulator model, wherein the neural network is more quickly executable than the reservoir simulator model; and

executing the proxy model to generate a forecast indicating a characteristic of the reservoir at a future point in time.

11. The non-transitory program carrier device of claim 10 , wherein the minimum value is selected and represents a value for a choke in a completely closed position at a respective time step.

12. The non-transitory program carrier device of claim 10 , wherein the maximum value is selected and represents a value for a choke in a wide open position at a respective time step.

13. The non-transitory program carrier device of claim 10 , wherein the most likely value is selected based on each time step in the total number of time steps being identical to each time step in the predetermined time period.

14. The non-transitory program carrier device of claim 10 , wherein the most likely value is calculated using the production rates or injection rates from the base case.

15. The non-transitory program carrier device of claim 10 , wherein the predetermined minimum number of cases is at least ten times the number of decision variables in the plurality of decision variables.

16. The non-transitory program carrier device of claim 10 , further comprising instructions executable by the processor to create the neural network at least in part by determining a total number of nodes for the neural network based on a total number of proxy model input variables.

17. The non-transitory program carrier device of claim 10 , further comprising instructions executable by the processor to adjust a setting associated with an asset-production tool based on the forecast.