Progressive modeling of optical sensor data transformation neural networks for downhole fluid analysis

1. A method, comprising:

applying a subset of synthetic optical measurements to an optical data transformation of an adaptive model to identify predicted optical sensor responses;

applying the predicted optical sensor responses to identify predicted synthetic optical measurements;

generating calibration data for calibrating the optical data transformation of the adaptive model based on a comparison of the predicted synthetic optical measurements and the subset of synthetic optical measurement responses;

receiving optical sensor measurements generated by one or more downhole optical sensors in a wellbore;

applying the adaptive model and the optical sensor measurements to determine synthetic data for fluid characterization; and

applying the synthetic data to determine one or more physical properties of a fluid in the wellbore for which the optical sensor measurements are received.

2. The method of claim 1 , wherein determining the synthetic data comprises:

selecting, as the adaptive model, a candidate reverse optical data transformation neural network and a candidate forward optical data transformation neural network from a model base of optical data transformation neural networks;

applying the optical sensor measurements to the candidate reverse optical data transformation neural network to yield simulated synthetic data;

applying the simulated synthetic data to the candidate forward optical data transformation neural network to yield simulated optical measurement data; and

selecting the simulated synthetic data as the synthetic data based on a prediction error between the simulated optical measurement data and the optical sensor measurements.

3. The method of claim 2 , wherein if the prediction error is equal to or less than a threshold, the simulated synthetic data is selected as the synthetic data for determining the one or more physical properties of the fluid.

4. The method of claim 2 , wherein if the prediction error is more than a threshold, the method further comprises:

iteratively selecting further candidate reverse optical data transformation neural networks and candidate forward optical data transformation neural networks and repeating steps of applying the optical sensor measurements, applying the simulated synthetic data and selecting the simulated synthetic data until the prediction error is equal to or less than the threshold.

5. The method of claim 1 , wherein the adaptive model comprises a fluid characterization neural network and an optical data transformation neural network; and the synthetic data is applied as input to a pre-calibrated fluid characterization neural network to receive, as output the one or more physical properties of the fluid.

6. The method of claim 5 , wherein the adaptive model is pre-calibrated in a laboratory setting.

7. The method of claim 6 , wherein pre-calibrating the adaptive model comprises calibrating the optical data transformation neural network.

8. The method of claim 7 , wherein the pre-calibrating comprises applying a progressive modeling to calibrate forward and reverse optical data transformation neural networks.

9. The method of claim 1 , wherein the optical data transformation is a forward optical data transformation.

10. The method of claim 5 , wherein the fluid characterization neural network had two hidden layers and the optical data transformation neural network has a single hidden layer.

11. A device, comprising:

memory having computer-readable instructions stored therein; and

one or more processors configured to execute computer—readable instructions to:

apply a subset of synthetic optical measurements to an optical data transformation of an adaptive model to identify predicted optical sensor responses;

apply the predicted optical sensor responses to identify predicted synthetic optical measurements;

generate calibration data for calibrating the optical data transformation of the adaptive model based on a comparison of the predicted synthetic optical measurements and the subset of synthetic optical measurement responses;

receive optical sensor measurements generated by one or more downhole optical sensors in a wellbore;

determine synthetic data for fluid characterization using an adaptive model and the optical sensor measurements; and apply the synthetic data to determine one or more physical properties of a fluid in the wellbore for which the optical sensor measurements are received.

12. The device of claim 11 , wherein the one or more processors are configured to further execute the computer-readable instructions to:

select, as the adaptive model, a candidate reverse optical data transformation neural network and a candidate forward optical data transformation neural network from a model base of optical data transformation neural networks;

apply the optical sensor measurements to the candidate reverse optical data transformation neural network to yield simulated synthetic data;

apply the simulated synthetic data to the candidate forward optical data transformation neural network to yield simulated optical measurement data; and

select the simulated synthetic data as the synthetic data based on a prediction error between the simulated optical measurement data and the optical sensor measurements.

13. The device of claim 12 , wherein if the prediction error is equal to or less than a threshold, the simulated synthetic data is selected as the synthetic data for determining the one or more physical properties of the fluid.

14. The device of claim 12 , wherein if the prediction error is more than a threshold, the one or more processors are configured to further execute the computer-readable instructions to:

iteratively select further candidate reverse optical data transformation neural networks and candidate forward optical data transformation neural networks and repeat steps of applying the optical sensor measurements, applying the simulated synthetic data and selecting the simulated synthetic data until the prediction error is equal to or less than the threshold.

15. The device of claim 11 , wherein

the adaptive model comprises a fluid characterization neural network and an optical data transformation neural network; and

the synthetic data is applied as input to a pre-calibrated fluid characterization neural network to receive, as output the one or more physical properties of the fluid.

16. The device of claim 15 , wherein the adaptive model is pre-calibrated in a laboratory setting.

17. The device of claim 16 , wherein the one or more processors are configured to further execute the computer-readable instructions to pre-calibrate the adaptive model by calibrating the optical data transformation neural network.

18. The device of claim 17 , wherein the one or more processors are configured to further execute the computer—readable instructions to apply a progressive modeling to calibrate forward and reverse optical data transformation neural networks.

19. The device of claim 11 , wherein

the optical data transformation is a forward optical data transformation.

20. The device of claim 11 , wherein the one or more optical sensors are communicatively coupled to the device.

21. The device of claim 11 , wherein the adaptive model can be at least one of a concatenated neural network, a genetic algorithm, an evolutionary method, a reinforcement learning, a bootstrap aggregating, a temporal difference learning, a w-net, or a growing self-organizing map.