Adaptive power generation management

• US 10,309,372 B2
• Filed: 05/25/2017
• Issued: 06/04/2019
• Est. Priority Date: 05/25/2017
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Huyn, Nam, Venkatraman, Chandrasekar
• Primary Examiner(s): Ortiz Rodriguez, Carlos R
• Application Number:
  US15/605,184
• Publication Number:
  US 20180340515A1
• Time in Patent Office: 740 Days
• Field of Search: None
• CPC Class Codes: F03D 17/00 Monitoring or testing of wind motors, e.g. diagnostics testing during commissioning of wind motors F03D13/30
• CPC Class Codes: F03D 7/028 controlling wind motor output power
• CPC Class Codes: F03D 7/045 with model-based controls
• CPC Class Codes: F03D 7/046 with learning or adaptive control, e.g. self-tuning, fuzzy logic or neural network
• CPC Class Codes: F03D 7/048 controlling wind farms
• CPC Class Codes: F03D 9/11 storing electrical energy
• CPC Class Codes: F03D 9/257 the wind motor being part of a wind farm
• CPC Class Codes: F05B 2260/821 Parameter estimation or prediction
• CPC Class Codes: F05B 2260/8211 of the weather
• CPC Class Codes: F05B 2260/84 Modelling or simulation
• CPC Class Codes: F05B 2270/32 Wind speeds
• CPC Class Codes: F05B 2270/335 Output power or torque
• CPC Class Codes: F05B 2270/404 active, predictive, or anticipative
• CPC Class Codes: H02K 7/183 wherein the turbine is a wind turbine adaptation of a wind turbine to an electric generator F03D9/25
• CPC Class Codes: Y02E 10/72 Wind turbines with rotation axis in wind direction
• CPC Class Codes: Y02E 70/30 Systems combining energy storage with energy generation of non-fossil origin

Abstract

In some examples, a system receives first sensor data from respective wind turbines of a plurality of wind turbines. For instance, the first sensor data may include at least a power output and a wind speed per time interval. The system trains at least one respective model for each respective wind turbine based on the first sensor data received from that respective wind turbine. Further, the system receives, for a second time period, respective second sensor data from the respective wind turbines. The system executes, using the respective second sensor data, the respective model trained using the first sensor data received from that respective wind turbine to determine, for each respective wind turbine, a predicted power output for an upcoming period. The predicted power outputs may be aggregated to determine a total predicted power output and at least one action is performed based on the total predicted power output.

Claims

1. A system comprising:

one or more processors; and

one or more non-transitory computer-readable media storing executable instructions, which, when executed by the one or more processors, configure the one or more processors to perform operations comprising;

receiving, for a first time period, first sensor data from respective wind turbines of a plurality of wind turbines, the first sensor data including at least a power output and a wind speed per time interval, wherein the first time period includes a plurality of the time intervals;

training at least one respective model for each respective wind turbine based at least partially on the first sensor data received from the respective wind turbine;

receiving, for a second time period, respective second sensor data from the respective wind turbines;

executing, using the respective second sensor data received from the respective wind turbines, the at least one respective model that was trained using the first sensor data received from that respective wind turbine to determine, for each respective wind turbine, a predicted power output for an upcoming time period;

aggregating the predicted power outputs to determine a total predicted power output for the plurality of wind turbines; and

based at least in part on the total predicted power output, at least one of;

sending a communication to cause a shut down or a start up of at least one wind turbine of the plurality of wind turbines;

sending a communication to cause activation of a switch to divert at least a portion of produced power to a power storage;

sending a communication to cause activation of a switch to send stored power from the power storage to a grid;

or

sending a communication to a utility computing device indicating an expected power output for a least a portion of the upcoming time period.

2. The system as recited in claim 1, further comprising a plurality of model computing devices, each model computing device including at least one of the processors, wherein respective ones of the model computing devices are configured to train and execute the at least one respective model for a respective one of the wind turbines.

3. The system as recited in claim 1, further comprising a data collection computing device located proximate to a geographic region of the plurality of wind turbines for receiving the sensor data from the plurality of wind turbines, wherein the plurality of model computing devices are located proximate to the geographic region of the plurality of wind turbines and are able to communicate over a local area network with the data collection computing device.

4. The system as recited in claim 1, the operations further comprising:

receiving, from an external source, weather forecast data for a geographic region in which the plurality of wind turbines are located; and

training the at least one respective model for each of the respective wind turbines based at least partially on the weather forecast data, wherein executing, using the second sensor data received from each respective wind turbine, the at least one respective model that was trained using the first sensor data received from that respective wind turbine includes using received weather forecast data when executing the at least one respective model.

5. The system as recited in claim 1, the operations further comprising:

when executing the at least one respective model for a first one of the respective wind turbines, determining one or more second respective wind turbines within a threshold distance proximate to first respective wind turbine; and

wherein executing, using the second sensor data received from the first respective wind turbine, the at least one respective model that was trained using the first sensor data received from the first respective wind turbine includes using sensor data received from the one or more second respective wind turbines.

6. The system as recited in claim 1, wherein:

the at least one respective models are support vector machine regression models, and the sensor data used as inputs to execute the at least one respective models includes a time series of a plurality of the time intervals received from the respective wind turbines, where a measured power output and measured wind speed are averaged over each time interval in the time series; and

at least one of the time intervals in the time series at least partially overlaps another one of the time intervals in the time series.

7. The system as recited in claim 1, wherein, for a first one of the respective wind turbines, the at least one respective model is a first model, the operations further comprising:

training a second model using additional sensor data received from the first wind turbine prior to receiving the second sensor data;

executing the second model using the second sensor data to determine a second predicted power output for the first wind turbine;

based on subsequently received sensor data from the first respective wind turbine indicating an actual power output, determining whether predicted power output from the first model or the second model is closer to the actual power output; and

based at least partially on the predicted power output from the second model being closer to the actual power output, replacing the first model with the second model for the first respective wind turbine.

8. A method comprising:

training, by one or more processors, a respective model for each respective wind turbine of a plurality of wind turbines based at least partially on first sensor data received from the respective wind turbines for a first time period;

receiving, for a second time period, respective second sensor data from the respective wind turbines;

executing, using the respective second sensor data received from the respective wind turbines, the respective model trained for the respective wind turbine to determine, for each respective wind turbine, a predicted power output for an upcoming time period;

aggregating the predicted power outputs for the respective wind turbines to determine a total predicted power output for the plurality of wind turbines; and

based at least in part on the total predicted power output, performing at least one action.

9. The method as recited in claim 8, wherein performing at least one action comprises at least one of:

sending a communication to cause a shut down or a start up at least one wind turbine of the plurality of wind turbines;

sending a communication to cause activation of a switch to divert at least a portion of produced power to a power storage;

or

sending a communication to cause activation of a switch to send stored power from the power storage to a grid.

10. The method as recited in claim 8, wherein performing at least one action comprises sending a communication to a utility computing device indicating an expected power output for a least a portion of the upcoming time period.

11. The method as recited in claim 8, further comprising configuring a plurality of model computing devices at a geographic region at which the wind turbines are located, wherein respective ones of the model computing devices are configured to train and execute the respective model for a respective one of the wind turbines.

12. The method as recited in claim 8, further comprising:

when executing the respective model for a first one of the respective wind turbines, determining one or more second respective wind turbines within a threshold distance proximate to first respective wind turbine; and

wherein executing, using the respective second sensor data received from the respective wind turbines, the respective model trained for the respective wind turbine includes using sensor data received from the one or more second respective wind turbines.

13. The method as recited in claim 8, wherein:

training a respective model for each respective wind turbine comprises training at least one respective support vector machine regression model for each respective wind turbine, wherein the sensor data used as inputs to execute the respective models includes a time series of a plurality of the time intervals received from the respective wind turbines, wherein a measured power output and measured wind speed are averaged over each time interval in the time series; and

at least one of the time intervals in the time series at least partially overlaps another one of the time intervals in the time series.

14. The method as recited in claim 8, wherein, for a first one of the respective wind turbines, the respective model is a first model, the method further comprising:

training a second model using additional sensor data received from the first wind turbine prior to receiving the second sensor data;

executing the second model using the second sensor data to determine a second predicted power output for the first wind turbine;

based on subsequently received sensor data from the first respective wind turbine indicating an actual power output, determining whether predicted power output from the first model or the second model is closer to the actual power output; and

based at least partially on the predicted power output from the second model being closer to the actual power output, replacing the first model with the second model for the first respective wind turbine.

15. A system comprising:

one or more processors; and

one or more non-transitory computer-readable media maintaining executable instructions, which, when executed by the one or more processors, program the one or more processors to perform operations comprising;

training a respective model for each respective wind turbine of a plurality of wind turbines based at least partially on first sensor data received from the respective wind turbines for a first time period;

receiving, for a second time period, respective second sensor data from the respective wind turbines;

executing, using the respective second sensor data received from the respective wind turbines, the respective model trained for the respective wind turbine to determine, for each respective wind turbine, a predicted power output for an upcoming time period;

aggregating the predicted power outputs for the respective wind turbines to determine a total predicted power output for the plurality of wind turbines; and

based at least in part on the total predicted power output, performing at least one action.

16. The system as recited in claim 15, wherein, for a first one of the respective wind turbines, the respective model is a first model, the operations further comprising:

training a second model using additional sensor data received from the first wind turbine prior to receiving the second sensor data;

executing the second model using the second sensor data to determine a second predicted power output for the first wind turbine for the upcoming time period;

based on subsequently received sensor data from the first respective wind turbine indicating an actual power output, determining whether predicted power output from the first model or the second model is closer to the actual power output; and

based at least partially on the predicted power output from the second model being closer to the actual power output, replacing the first model with the second model for the first respective wind turbine.

17. The system as recited in claim 15, wherein, for a first one of the respective wind turbines, the respective model is a first model, the operations further comprising:

training a second model using additional sensor data received from the first wind turbine prior to receiving the second sensor data;

executing the second model using the second sensor data to determine a second predicted power output for the first wind turbine for the upcoming time period; and

determining a predicted power output for the first wind turbine for the upcoming time period based on the predicted power output determined from the first model and the predicted power output determined from the second model.

18. The system as recited in claim 15, further comprising a plurality of model computing devices at a geographic region at which the wind turbines are located, each model computing device including at least one of the one or more processors, wherein respective ones of the model computing devices are configured to train and execute the respective model for a respective one of the wind turbines.

19. The system as recited in claim 15, wherein the operation of performing at least one action comprises performing at least one of:

sending a communication to cause a shut down or a start up at least one wind turbine of the plurality of wind turbines;

sending a communication to cause activation of a switch to divert at least a portion of produced power to a power storage;

sending a communication to cause activation of a switch to send stored power from the power storage;

or

sending a communication to a utility computing device indicating an expected power output for a least a portion of the upcoming time period.

20. The system as recited in claim 15, the operations further comprising:

when executing the respective model for a first one of the respective wind turbines, determining one or more second respective wind turbines within a threshold distance proximate to the first respective wind turbine; and

wherein executing, using the respective second sensor data received from the respective wind turbines, the respective model trained for the respective wind turbine includes using sensor data received from the one or more second respective wind turbines.