APPLICATION OF ARTIFICIAL INTELLIGENCE TECHNIQUES AND STATISTICAL ENSEMBLING TO FORECAST POWER OUTPUT OF A WIND ENERGY FACILITY

• US 20140195159A1
• Filed: 01/09/2014
• Published: 07/10/2014
• Est. Priority Date: 01/09/2013
• Status: Abandoned Application

• Current Assignee: ClearAg, Inc. (TBG AG)
• Original Assignee: Iteris Incorporated
• Inventors: MEWES, JOHN J., OSBORNE, LEON F.
• Application Number:
  US14/150,784
• Publication Number:
  US 20140195159A1
• Time in Patent Office: Days
• US Class Current: 702/3
• CPC Class Codes: G01R 21/00 Arrangements for measuring electric power or power factor G01R7/12 takes precedence
• CPC Class Codes: G01W 1/02 Instruments for indicating weather conditions by measuring two or more variables, e.g. humidity, pressure, temperature, cloud cover or wind speed G01W1/10 takes precedence
• CPC Class Codes: G01W 1/10 Devices for predicting weather conditions computers per se G06; display devices G09

Abstract

A wind energy forecasting system processes data from one or more numerical weather prediction models with power output data from a wind energy facility using artificial intelligence. This artificial intelligence is applied in one or more neural networks that produce specific power output forecasts for each numerical weather prediction model. A statistical ensembling approach is then applied to the resulting numerical weather prediction model forecasts and integrated with a persistence power output forecast to arrive at a consensus, overall forecasted power output for the wind energy facility.

Claims

1. A method of forecasting power output of a wind energy facility, comprising:

ingesting one or more data sets representative of meteorological forecasts for an area in which a wind energy facility is located from at least one of multiple numerical predictive weather models;

extracting weather variables from the one or more data sets having an expected relationship to a power output production generated by the wind energy facility;

ingesting an actual power output data that is representative of historical power output of the wind energy facility for a specified period of time;

applying, within a computing environment comprised of at least one computer processor configured to model a specific power output forecast for a wind energy facility within a plurality of data processing modules, the weather variables and the actual power output data to heuristically build one or more neural networks to infer non-linear relationships between the weather variables and the actual power output data of the wind energy facility to produce a specific power output forecast for each numerical weather prediction model;

projecting a current time-series representation of power output of the wind energy facility to create a persistence power output forecast; and

creating an ensemble average consensus power output forecast for the specified period of time from numerical weather prediction models and persistence power output forecast comprising ensemble members, each ensemble member having a weight determined by recent and real-time statistical assessments of an accuracy of each specific power output forecast for the wind energy facility.

2. The method of claim 1, wherein the applying the weather variables and the actual power output data to one or more neural networks further comprises heuristically building multiple neural networks for a given numerical weather prediction model, each neural network using a different network structure and a different set of predictors drawn out of the weather variables.

3. The method of claim 1, further comprising repetitively training the one or more neural networks to continuously update the specific power output forecasts from the numerical weather prediction models to account for seasonal changes in weather and for the effects of unusually severe and unusually mild weather conditions at the location of the wind energy facility.

4. The method of claim 1, wherein the creating an ensemble average consensus power output forecast further comprises applying a minimum variance estimation to improve overall forecast accuracy.

5. The method of claim 1, wherein the creating an ensemble average consensus power output forecast further comprises weighting each power output forecast, for an individual numerical weather prediction model and/or a persistence power output forecast, to produce a minimum variance estimate.

6. The method of claim 5, wherein the weighting each power output forecast for an individual numerical weather prediction model and/or persistence power output forecast to arrive at a minimum variance estimate further comprises assigning a weighted estimate based on degrees of similarity of present conditions to past conditions where the actual power output data of the wind energy facility is already known.

7. The method of claim 1, further comprising determining a specified period of time over which the ensemble average consensus power output forecast for the wind energy facility is generated, wherein the specified period of time determines a time-dependent weighting applied to each specific numerical weather prediction model and/or persistence power output forecast.

8. The method of claim 1, wherein the weather variables define an atmospheric profile that at least includes at least one of a vertical profile of expected wind speed and wind direction characteristics, temperature, humidity, stability, turbulent transfer, and precipitation, at the location of the wind energy facility.

9. The method of claim 1, wherein the ensemble average consensus power output forecast comprises at least one of:

a set of one or more different numerical weather prediction models, a set of one or more different physical or dynamical schemes within each numerical weather prediction model, and a set of one or more different lead times for a specific numerical weather prediction model.

10. The method of claim 1, further comprising applying, to the ensemble of numerical weather prediction models and persistence power output forecast from which the specific power output forecasts are produced, historical climatology data representing average power output production for a time interval, and further comprising applying a Fast Fourier analysis to the average power output production to identify long-term cycles in the historical climatology data that explain variability in power output over the specified period of time.

11. The method of claim 1, wherein the persistence power output forecast are generated by comparing a power output of the wind energy facility for a particular time interval and with power output over preceding, similar time intervals.

12. The method of claim 1, further comprising enabling graphical displays of power output forecast data for operators of wind energy facilities on a graphical user interface, the graphical displays providing one or more of visualizations and animations of the consensus power output forecast.

13. A wind energy forecasting system, comprising:

at least one computer processor operably coupled to at least one computer-readable storage medium having program instructions stored therein, the at least one computer processor configured to execute the program instructions to model weather variables representative of one or more meteorological conditions and generate specific power output forecasts for a wind energy facility in a plurality of data processing modules, the plurality of data processing modules including;

an input data collection module configured to continually ingest weather data comprising the weather variables from one or more numerical weather prediction model runs, an actual power output data collection module configured to ingest data relative to a real-time and historical power output of the wind energy facility over a specified period of time;

a plurality of neural networks heuristically built from the weather variables from the input data collection module and the historical power output from the actual power output data collection module, and trained to infer non-linear relationships between the weather variables and the historical power output of the wind energy facility to produce a specific power output forecast for each numerical weather prediction model for the meteorological conditions represented in the weather variables;

an ensembling module configured to aggregate the specific power output forecasts from each numerical weather prediction model run, and a persistence power output forecast based on real-time power output, into an ensemble of members by continuously tracking statistical properties of each specific power output forecast to assign a weight for each specific power output forecast, the weight determined based on a minimum variance estimation; and

a power production module configured to generate output data representative of a consensus power output forecast for the wind energy facility.

14. The system of claim 13, wherein the plurality of neural networks further comprises multiple neural networks for a given numerical weather prediction model, and each neural network for a given numerical weather prediction model uses a different network structure and a different set of predictors drawn out of the weather variables.

15. The system of claim 13, wherein the plurality of neural networks are repetitively training to continuously update the specific power output forecasts from the numerical weather prediction models to account for seasonal changes in weather and for the effects of unusually severe and unusually mild weather conditions at the location of the wind energy facility.

16. The system of claim 13, wherein the weather variables define an atmospheric profile that at least includes a vertical profile of expected wind speed and wind direction characteristics, temperature, humidity, stability, turbulent transfer, and precipitation, at the location of the wind energy facility.

17. The system of claim 13, wherein the ensemble of members comprises at least one of:

a set of different numerical weather prediction models, a set of different physical or dynamical schemes within each numerical weather prediction model, and different lead times for a specific numerical weather prediction model.

18. The system of claim 13, wherein the persistence power output forecasts are generated by comparing a power output of the wind energy facility for a particular time interval and with power output over preceding, similar time intervals.

19. The system of claim 13, further comprising a graphics module configured to generate one or more of visualizations and animations of the output data display on a graphical user interface for operators of wind energy facilities.

20. A method of forecasting power output of a wind energy facility, comprising:

modeling weather variables from a plurality of numerical weather prediction models and actual power output data of wind energy facility, by;

heuristically building one or more neural networks by training each neural network to infer non-linear relationships between the weather variables and the actual power output data of wind energy facility to produce a specific power output forecast for each numerical weather prediction model for the meteorological conditions represented in the weather variables;

aggregating the numerical weather prediction models from which the specific power output forecasts are produced into an ensemble of numerical weather prediction model members; and

statistically combining the specific power output forecast for each numerical weather predictive model with a real-time persistence power output forecast of the wind energy facility by assigning a weight to each specific power output forecast to produce a minimum variance estimate.

21. The method of claim 20, wherein the heuristically building one or more neural networks further comprises heuristically building multiple neural networks for a given numerical weather prediction model, each neural network using a different network structure and a different set of predictors drawn out of the weather variables.

22. The method of claim 20, further comprising repetitively training the one or more neural networks to continuously update the specific power output forecasts from the numerical weather prediction models to account for seasonal changes in weather and for the effects of unusually severe and unusually mild weather conditions at the location of the wind energy facility.

23. The method of claim 20, further comprising determining a specified period of time over which a consensus power output forecast for the wind energy facility is generated, wherein the specified period of time determines a time-dependent weighting applied to each specific numerical weather prediction model and persistence power output forecast.

24. The method of claim 20, further comprising generating one or more of visualizations and animations of the output data display on a graphical user interface for operators of wind energy facilities.

25. The method of claim 20, wherein the weather variables define an atmospheric profile that at least includes a vertical profile of expected wind speed and wind direction characteristics, temperature, humidity, stability, turbulent transfer, and precipitation, at the location of the wind energy facility.