Method and equipment for searching non-unique solution of petrophysical property combination in history matching
View Patent ↗The present disclosure provides a method and equipment for searching a non-unique solution of a petrophysical property combination in history matching. The method includes: receiving field historical data, and dividing the field historical data into a tuning historical data portion, and a validation historical data portion; inputting initial data for the petrophysical property combination into an proxy model to obtain field response data; optimizing the petrophysical property combination to minimize an error between the field response data corresponding to the optimized petrophysical property combination and the field response data of the tuning historical data portion; inputting the optimized petrophysical property combination into the proxy model to obtain field response data corresponding to the optimized petrophysical property combination; matching the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion; and if the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion match each other, determining the optimized petrophysical property combination as the non-unique solution of the petrophysical property combination.
1. A method for searching a non-unique solution of a petrophysical property combination in history matching, comprising:
receiving field historical data, and dividing, according to a timeline, the field historical data into a tuning historical data portion for tuning the petrophysical property combination, and a validation historical data portion for validating the petrophysical property combination;
inputting initial data for the petrophysical property combination into a machine-learning-based proxy model to obtain field response data;
optimizing the petrophysical property combination based on the tuning historical data portion to minimize an error between the field response data corresponding to the optimized petrophysical property combination and the field response data of the tuning historical data portion;
inputting the optimized petrophysical property combination into the machine-learning-based proxy model to obtain field response data corresponding to the optimized petrophysical property combination;
matching the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion; and
if the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion match each other, determining the optimized petrophysical property combination as the non-unique solution of the petrophysical property combination.
2. The method according to claim 1 , wherein the optimization operation is performed through a heuristic optimizer, and the heuristic optimizer defines a search interval of uncertain petrophysical property parameters.
3. The method according to claim 1 , wherein the machine-learning-based proxy model is obtained by training the machine-learning-based proxy model using the field historical data through a machine-learning-based algorithm.
4. The method according to claim 1 , wherein the computational speed of the machine-learning-based proxy model is many orders of magnitudes less than a high-fidelity numerical simulation model.
5. The method according to claim 1 , wherein if the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion do not match each other, the petrophysical property combination is continued to be optimized.
6. The method according to claim 1 , wherein after the non-unique solution of the petrophysical property combination is obtained, the field historical data is re-divided, and the above operations is re-executed according to the re-divided field historical data, so as to obtain another non-unique solution of the petrophysical property combination.
7. The method according to claim 6 , wherein the above operations are iteratively performed until the number of non-unique solutions of the petrophysical property combination reaches a preset number.
8. Equipment for searching a non-unique solution of a petrophysical property combination in history matching, comprising:
a data division module, configured to receive field historical data, and divide, according to a timeline, the field historical data into a tuning historical data portion for tuning the petrophysical property combination, and a validation historical data portion for validating the petrophysical property combination;
a machine-learning-based proxy model, configured to receive an initial data input for the petrophysical property combination, and output field response data;
an optimizer, configured to optimize the petrophysical property combination based on the tuning historical data portion to minimize an error between the field response data corresponding to the optimized petrophysical property combination and the field response data of the tuning historical data portion; and
a matching module, configured to input the optimized petrophysical property combination into the machine-learning-based proxy model to obtain field response data corresponding to the optimized petrophysical property combination, and match the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion, wherein if the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion match each other, the optimized petrophysical property combination is determined as the non-unique solution of the petrophysical property combination.
9. The equipment according to claim 8 , wherein the optimizer is a heuristic optimizer, and defines a search interval of uncertain petrophysical property parameters.
10. The equipment according to claim 8 , wherein the machine-learning-based proxy model is obtained by training the machine-learning-based proxy model using the field historical data through a machine-learning-based algorithm.
11. The equipment according to claim 8 , wherein the computational speed of the machine-learning-based proxy model is many orders of magnitudes less than a high-fidelity numerical simulation model.
12. The equipment according to claim 8 , wherein if the field response data corresponding to the optimized petrophysical property combination and the field response data of the validation historical data portion do not match each other, the optimizer continues to optimize the petrophysical property combination.
13. The equipment according to claim 8 , wherein after the non-unique solution of the petrophysical property combination is obtained, the data division module re-divides the field historical data, and the machine-learning-based proxy model, the optimizer and the matching module re-execute the above operations according to the re-divided field historical data, so as to obtain another non-unique solution of the petrophysical property combination.
14. The equipment according to claim 13 , wherein the data division module, the machine-learning-based proxy model, the optimizer and the matching module iteratively perform the above operations until the number of non-unique solutions of the petrophysical property combination reaches a preset number.