Machine learning server-side flow measurement
Systems and methods enable predicting flow rate in a pipeline by using machine learning. Data is collected from a pipeline with an acoustic device and transmitted from the acoustic device to a server. The data is processed and, based on the processing, a reference model is selected that is most appropriate for the processed data. Features extracted from the data are input to the reference model. The reference model outputs a predicted flow rate.
1. A computer-implemented method for measuring the flow rate of a fluid or gas or multi-phase flow flowing through a pipeline, the method comprising:
transmitting, by an acoustic device, signals into an outer shell of a pipeline;
receiving, by the acoustic device, received signals reflected from the contents of the pipeline;
processing, by the acoustic device, the received signals to generate a data stream;
transmitting, by the acoustic device, the data stream to a server; computing a flow rate based on the data stream by analyzing a Doppler frequency of the received signals;
performing dimensionality reduction, by the server, on the data stream and one or more acoustic device attributes to generate a device signature;
identifying, by the server, one of a plurality of reference models having a device signature that is most similar to the device signature of the acoustic device to select a most similar reference model;
applying, by the server, an attribute identification machine learning model to identify a plurality of important features from the data stream and one or more device attributes; and
applying a multivariate machine learning model associated with the most similar reference model, using the identified important features in the multivariate machine learning model, to predict a flow rate.
2. The method of claim 1 , wherein each of the plurality of reference models comprises an associated multivariate machine learning model.
3. The method of claim 1 , wherein the attribute identification machine learning model is a nearest neighbors algorithm.
4. The method of claim 1 , further comprising selecting the most similar reference model by using a distance metric.
5. The method of claim 1 , further comprising applying a feature extraction module to the data stream.
6. The method of claim 1 , further comprising modeling the flow rate by using a physics model.
7. The method of claim 1 , further comprising training the attribute identification machine learning model using parameters from one or more of the reference models.
8. The method of claim 1 , further comprising:
performing dimensionality reduction, by the server, on the data stream and one or more indicators of a firmware update installed on the acoustic device to generate a firmware update signature;
identifying, by the server, one of a plurality of firmware models having a firmware update signature that is most similar to the firmware update signature of the acoustic device to select a most similar firmware model; and
transmitting a firmware update associated with the firmware model to the acoustic device.
9. The method of claim 8 , further comprising selecting the most similar firmware model by using a distance metric.
10. The method of claim 8 , further comprising measuring, by the acoustic device, the flow rate in the pipeline using updated firmware settings.
11. A non-transitory computer-readable medium comprising instructions for:
transmitting, by an acoustic device, signals into an outer shell of a pipeline;
receiving, by the acoustic device, received signals reflected from the contents of the pipeline; processing, by the acoustic device, the received signals to generate a data stream;
transmitting, by the acoustic device, the data stream to a server;
computing a flow rate based on the data stream by analyzing a Doppler frequency of the received signals;
performing dimensionality reduction, by the server, on the data stream and one or more acoustic device attributes to generate a device signature;
identifying, by the server, one of a plurality of reference models having a device signature that is most similar to the device signature of the acoustic device to select a most similar reference model;
applying, by the server, an attribute identification machine learning model to identify a plurality of important features from the data stream and one or more device attributes; and
applying a multivariate machine learning model associated with the most similar reference model, using the identified important features in the multivariate machine learning model, to predict a flow rate.
12. The non-transitory computer-readable medium of claim 11 , wherein each of the plurality of reference models comprises an associated multivariate machine learning model.
13. The non-transitory computer-readable medium of claim 11 , wherein the attribute identification machine learning model is a nearest neighbors algorithm.
14. The non-transitory computer-readable medium of claim 11 , further comprising instructions for selecting the most similar reference model by using a distance metric.
15. The non-transitory computer-readable medium of claim 11 , further comprising instructions for applying a feature extraction module to the data stream.
16. The non-transitory computer-readable medium of claim 11 , further comprising instructions for modeling the flow rate by using a physics model.
17. The non-transitory computer-readable medium of claim 11 , further comprising instructions for training the machine learning model using parameters from one or more of the reference models.
18. The non-transitory computer-readable medium of claim 11 , further comprising instructions for:
performing dimensionality reduction, by the server, on the data stream and one or more indicators of a firmware update installed on the acoustic device to generate a firmware update signature;
identifying, by the server, one of a plurality of firmware models having a firmware update signature that is most similar to the firmware update signature of the acoustic device to select a most similar firmware model; and
transmitting a firmware update associated with the firmware model to the acoustic device.
19. The non-transitory computer-readable medium of claim 18 , further comprising instructions for selecting the most similar firmware model by using a distance metric.
20. The non-transitory computer-readable medium of claim 18 , further comprising instructions for measuring, by the acoustic device, the flow rate in the pipeline using updated firmware settings.