Machine Learning Model-Based Analysis for Monitoring the Power Performance of Wind Farms

Abstract

The present invention relates to machine learning model-based analysis for monitoring the power performance of a wind farm. Specifically, a method for controlling a wind turbine includes detecting a plurality of analysis outputs related to the power performance of the wind turbine from a plurality of different analyses. The method also includes analyzing the plurality of analysis outputs related to the power performance of the wind turbine. Additionally, the method includes using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of the wind turbine. Further, the method includes training the computer-based model(s) of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine. Moreover, the method includes using the computer-based model(s) of machine learning to estimate the power magnitude of the wind turbine. Accordingly, the method includes implementing a control action when the power magnitude of the wind turbine is outside a selected range.

Description

Technical Field

[0001] The present disclosure generally relates to wind farms and, more particularly, to machine learning model-based analysis for monitoring wind farm performance. Background Art

[0002] Wind power is considered one of the cleanest and most environmentally friendly energy sources currently available, and wind turbines have received increasing attention in this regard. Modern wind turbines typically include a tower, a generator, a gearbox, a nacelle, and one or more rotor blades. The rotor blades capture the kinetic energy of the wind using known airfoil principles. For example, the rotor blades typically have an airfoil cross-sectional profile such that during operation, air flows over the blade creating a pressure difference between the sides. Thus, a lift force directed from the pressure side towards the suction side acts on the blade. The lift force creates a torque on the main rotor shaft, which is connected via a gear to a generator for electricity generation.

[0003] Multiple wind turbines are typically used in combination with each other to generate electricity and are commonly referred to as a "wind farm". During operation, it is advantageous to utilize various analysis outputs to evaluate wind turbine and / or wind farm performance to ensure that the wind turbine(s) and / or wind farm are operating properly. However, it is difficult to classify situations where wind turbine performance is insufficient using such individual analysis outputs. As a result, the prior art provides an excessive number of false alarms, causing operators to ignore the degraded analysis outputs when performance is insufficient. In addition, not all analysis outputs are calculated and become available simultaneously. However, a decision still needs to be made on the queried cases.

[0004] Accordingly, the present disclosure relates to systems and methods for combining low-performing subsets of available analysis streams to create machine learning model-based analysis with high precision and accuracy to better estimate wind turbine performance. In addition, the systems and methods of the present disclosure also use power ensembles and / or other analysis to rank underperforming wind turbines, which quantify the extent of energy loss, enabling field engineers to focus on the key underperforming wind turbines in the wind farm. Summary of the Invention

[0005] Aspects and advantages of the invention will be set forth in part in the description which follows, or may be obvious from the description, or may be learned by practice of the invention.

[0006] In one aspect, the present disclosure relates to a method for controlling a wind turbine. The method includes detecting, via a controller, a plurality of analysis outputs related to the power performance of the wind turbine from a plurality of different analyses. The method also includes analyzing, via the controller, the plurality of analysis outputs related to the power performance of the wind turbine. Additionally, the method includes generating, via the controller, at least one computer-based model of the power performance of the wind turbine using at least a portion of the analyzed plurality of analysis outputs. Further, the method includes training, via the controller, at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine. Moreover, the method includes estimating the power magnitude of the wind turbine using at least one machine learning-based computer model. Thus, the method includes implementing a control action when the power magnitude of the wind turbine is outside a selected range.

[0007] In another aspect, the present disclosure relates to a system for controlling a wind turbine. The system includes a plurality of analyses for generating a plurality of analysis outputs related to the power performance of the wind turbine. Additionally, the system includes a controller communicatively coupled to the plurality of analyses. The controller is configured to perform a plurality of operations including, but not limited to, receiving the plurality of analysis outputs from the plurality of analyses, analyzing the plurality of analysis outputs related to the power performance of the wind turbine, generating at least one computer-based model of the power performance of the wind turbine using at least a portion of the analyzed plurality of analysis outputs, training the computer-based model(s) of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine via the controller, estimating the power magnitude of the wind turbine using the machine learning-based computer model(s), and implementing a control action when the power magnitude of the wind turbine is outside a selected range. It will be understood that the system may include any one or more of the additional features described herein.

[0008] In yet another aspect, the present disclosure relates to a wind farm. The wind farm includes a plurality of wind turbines, each of which includes a turbine controller and a field-level controller communicatively coupled to each of the turbine controllers. The field-level controller is configured to perform a plurality of operations, including but not limited to receiving a plurality of analysis outputs related to the power performance of each of the wind turbines from a plurality of different analyses, analyzing the plurality of analysis outputs related to the power performance of each of the wind turbines, using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of each of the wind turbines, using the annotated analysis outputs related to the power performance of each of the wind turbines to train at least one computer-based model of the power performance of each of the wind turbines, using at least one machine learning-based computer model to estimate the power magnitude of each of the wind turbines, and implementing a control action when the power magnitude of any one of the wind turbines is outside a selected range. It will be understood that the wind farm may include any one or more of the additional features described herein.

[0009] Specifically, the present disclosure also discloses the following technical solutions.

[0010] Technical solution 1. A method for controlling a wind turbine, the method comprising:

[0011] Detecting, via a controller, a plurality of analysis outputs related to the power performance of the wind turbine from a plurality of different analyses;

[0012] Analyzing, via the controller, the plurality of analysis outputs related to the power performance of the wind turbine;

[0013] Generating, via the controller, at least one computer-based model of the power performance of the wind turbine using at least a portion of the analyzed plurality of analysis outputs;

[0014] Training, via the controller, the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine;

[0015] Estimating the power magnitude of the wind turbine using the at least one machine learning-based computer model; and

[0016] Implementing a control action when the power magnitude of the wind turbine is outside a selected range.

[0017] Solution 2. The method according to Solution 1, wherein the plurality of analysis outputs related to the power performance of the wind turbine include at least two of the following: low production rate of the power curve, power curve history, power curve residuals, or power set.

[0018] Solution 3. The method according to Solution 1, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0019] Filtering the plurality of analysis outputs related to the power performance.

[0020] Solution 4. The method according to Solution 1, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0021] Using at least one of principal component analysis or factorization to reduce the dimensionality of the plurality of analysis outputs.

[0022] Solution 5. The method according to Solution 1, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0023] Organizing the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set via the controller.

[0024] Solution 6. The method according to Solution 5, wherein the first data set in the plurality of analysis data sets includes data from a first time length, and the second data set includes data from a second time length, and the first time length is longer than the second time length.

[0025] Solution 7. The method according to Solution 1, wherein using the annotated analysis outputs to train the at least one computer-based model of the power performance of the wind turbine further includes:

[0026] Continuously receiving the power measurements of the wind turbine;

[0027] Classifying each of the received power measurements of the wind turbine as underperformance, overperformance, or standard performance;

[0028] Annotating the received power measurements of the wind turbine; and

[0029] Using the annotated power measurements of the wind turbine to machine-learn the at least one computer-based model of the power performance.

[0030] Technical solution 8. The method according to technical solution 7, characterized in that training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis output further includes:

[0031] Performing a root cause analysis on the annotated power values of the wind turbine.

[0032] Technical solution 9. The method according to technical solution 8, characterized in that the method further includes storing the root cause analysis of the annotated power values for future use.

[0033] Technical solution 10. The method according to technical solution 1, characterized in that the method further includes: determining an uncertainty level associated with the power values of the wind turbine, and displaying the uncertainty level via a user interface of the controller.

[0034] Technical solution 11. The method according to technical solution 1, characterized in that the at least one computer-based model includes a support vector machine.

[0035] Technical solution 12. The method according to technical solution 1, characterized in that the at least one computer-based model is a microservice.

[0036] Technical solution 13. A system for controlling a wind turbine, the system includes:

[0037] Multiple analyses for generating multiple analysis outputs related to the power performance of the wind turbine;

[0038] A controller communicatively coupled to the multiple analyses, the controller configured to perform multiple operations, the multiple operations including:

[0039] Receiving the multiple analysis outputs from the multiple analyses;

[0040] Analyzing the multiple analysis outputs related to the power performance of the wind turbine;

[0041] Using at least a part of the analyzed multiple analysis outputs to generate at least one computer-based model of the power performance of the wind turbine;

[0042] Via the controller, training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine;

[0043] Using the at least one machine learning-based computer model to estimate the power values of the wind turbine; and

[0044] Implement a control action when the power quantity value of the wind turbine is outside a selected range.

[0045] Technical solution 14. The system according to technical solution 13, characterized in that the plurality of analysis outputs related to the power performance of the wind turbine include at least two of the following: low production rate of the power curve, power curve history, power curve residuals, or power set.

[0046] Technical solution 15. The system according to technical solution 13, characterized in that analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0047] Filter the plurality of analysis outputs related to the operation of the wind turbine.

[0048] Technical solution 16. The system according to technical solution 13, characterized in that analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0049] Use at least one of principal component analysis or factorization to reduce the dimensionality of the plurality of analysis outputs.

[0050] Technical solution 17. The system according to technical solution 13, characterized in that analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0051] Via the controller, organize the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set, wherein the first data set of the plurality of analysis data sets includes data from a first time length and the second data set includes data from a second time length, and the first time length is longer than the second time length.

[0052] Technical solution 18. The system according to technical solution 17, characterized in that using the annotated analysis outputs related to the power performance of the wind turbine to train the at least one computer-based model of the power performance of the wind turbine further includes:

[0053] Continuously receive the power quantity value of the wind turbine;

[0054] Classify each power quantity value of the received power quantity value of the wind turbine as insufficient performance, excessive performance, or standard performance;

[0055] Annotate the received power quantity value of the wind turbine; and

[0056] Use the annotated power quantity value of the wind turbine to machine-learn the at least one computer-based model of the power performance.

[0057] Technical solution 19. The system according to technical solution 18, wherein training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis output related to the power performance of the wind turbine further includes:

[0058] Performing a root cause analysis on the annotated power magnitude of the wind turbine; and

[0059] Storing the root cause analysis of the power magnitude of the wind turbine for future use.

[0060] Technical solution 20. A wind farm, comprising:

[0061] A plurality of wind turbines, each of the wind turbines including a turbine controller;

[0062] A field-level controller communicatively coupled to each of the turbine controllers, the field-level controller configured to perform a plurality of operations, the plurality of operations including:

[0063] Receiving a plurality of analysis outputs related to the power performance of each of the wind turbines from a plurality of different analyses;

[0064] Analyzing the plurality of analysis outputs related to the power performance of each of the wind turbines;

[0065] Using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of each of the wind turbines;

[0066] Training the at least one computer-based model of the power performance of each of the wind turbines using the annotated analysis output related to the power performance of each of the wind turbines;

[0067] Using the at least one machine learning-based computer model to estimate the power magnitude of each of the wind turbines; and

[0068] Implementing a control action when the power magnitude of any one of the wind turbines is outside a selected range.

[0069] Referring to the following description and the appended claims, these and other features, aspects and advantages of the present invention will become better understood. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the present invention and, together with the description, serve to explain the principles of the present invention. Description of the Drawings

[0070] The disclosure of the present invention, which is complete and enabling for a person of ordinary skill in the art (including its best mode), is set forth in the specification with reference to the accompanying drawings, in which:

[0071] Figure 1 Illustrates a perspective view of an embodiment of a wind farm in accordance with the present disclosure;

[0072] Figure 2 Illustrates a perspective view of an embodiment of a wind turbine in accordance with the present disclosure;

[0073] Figure 3 Illustrates a block diagram of an embodiment of a controller for a wind turbine and / or a wind farm in accordance with the present disclosure;

[0074] Figure 4 Illustrates a flowchart of an embodiment of a method for controlling a wind turbine in accordance with the present disclosure;

[0075] Figure 5 Illustrates a schematic diagram of a system for controlling a wind turbine in accordance with the present disclosure; and

[0076] Figure 6 Illustrates a schematic diagram of an embodiment of an analysis microservices architecture in accordance with the present disclosure. DETAILED DESCRIPTION

[0077] Reference will now be made in detail to embodiments of the invention, one or more examples of which are illustrated in the accompanying drawings. Each example is provided by way of explanation of the invention and not as a limitation of the invention. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the scope or spirit of the invention. For example, features illustrated or described as part of one embodiment can be used in combination with another embodiment to yield yet another embodiment. Accordingly, the present invention is intended to cover such modifications and variations as fall within the scope of the appended claims and their equivalents.

[0078] Generally, the present disclosure relates to machine learning model-based analysis for monitoring wind farm performance such that early detection of performance issues can be achieved. More specifically, the machine learning model-based analysis of the present disclosure combines several wind performance analyses having lower precision and accuracy to achieve a single analysis having high precision and accuracy. For example, the model-based analysis uses supervised machine learning and continuous learning of the identified data along with various preprocessing steps to create an analysis and system capable of detecting wind turbine performance deficiencies with minimal misses and fewest false alarms. Accordingly, the present disclosure may also provide a methodology for automatically determining the correct dimensions to include in the model using factorization and / or principal component analysis. Additionally, the model may include power ensemble analysis as one of the set of features used in the model. Thus, the models of the present disclosure may be continuously improved over time and new analyses may be continuously added as they become available.

[0079] Referring now to the drawings, Figure 1 illustrating an exemplary embodiment of a wind farm 100 including a plurality of wind turbines 102 in accordance with aspects of the present disclosure. The wind turbines 102 may be arranged in any suitable manner. By way of example, the wind turbines 102 may be arranged in an array of rows and columns, a single row, or randomly. Additionally, Figure 1 illustrating an example layout of one embodiment of the wind farm 100. Typically, the arrangement of wind turbines in a wind farm is determined based on numerous optimization algorithms such that the AEP is maximized for the corresponding site wind climate. It should be understood that any wind turbine arrangement may be implemented on, for example, uneven land without departing from the scope of the present disclosure.

[0080] Furthermore, it should be understood that the wind turbines 102 of the wind farm 100 may have any suitable configuration, for example, as Figure 2 shown. As shown, the wind turbine 102 includes a tower 114 extending from a support surface, a nacelle 116 mounted on top of the tower 114, and a rotor 118 coupled to the nacelle 116. The rotor includes a rotatable hub 120 on which a plurality of rotor blades 112 are mounted, which in turn is connected to a main rotor shaft that is coupled to a generator (not shown) housed within the nacelle 116. Thus, the generator generates electricity (or electrical power) from the rotational energy generated by the rotor 118. It should be recognized that Figure 2 the wind turbine 102 of

[0081] As generally shown in the figure, each wind turbine 102 of the wind farm 100 may also include a turbine controller 104 communicatively coupled to a field controller 108. Additionally, in one embodiment, the field controller 108 may be coupled to the turbine controller 104 via a network 110 to facilitate communication between various wind farm components. The wind turbine 102 may also include one or more sensors 105, 106, 107 configured to monitor various operating, wind, and / or loading conditions of the wind turbine 102. For example, the one or more sensors may include blade sensors for monitoring rotor blades 112; generator sensors for monitoring the load, torque, speed, acceleration, and / or power output of the generator; wind sensors 106 for monitoring one or more wind conditions; and / or shaft sensors for measuring the load and / or rotational speed of the rotor shaft. Additionally, the wind turbine 102 may include one or more tower sensors for measuring the load transmitted through the tower 114 and / or the acceleration of the tower 114. In various embodiments, the sensors may be any one or combination of the following: accelerometers, pressure sensors, angle-of-attack sensors, vibration sensors, micro inertial measurement units (MIMUs), camera systems, fiber optic systems, anemometers, wind vanes, acoustic detection and ranging (SODAR) sensors, infrared lasers, light detection and ranging (LIDAR) sensors, radiometers, pitot tubes, radiosondes, other optical sensors, and / or any other suitable sensors.

[0082] Now referring to Figure 3, which illustrates a block diagram of one embodiment according to the present disclosure that may include suitable components within the site controller 108, turbine controller(s) 104, and / or other suitable controllers. As shown, the controller(s) 104, 108 may include one or more processors 150 and associated memory device(s) 152, which are configured to perform various computer-implemented functions (e.g., execute methods, steps, calculations, etc. and store relevant data as disclosed herein). Additionally, the controller(s) 104, 108 may further include a communication module 154 to facilitate communication between the controller(s) 104, 108 and various components of the wind turbine 102. Additionally, the communication module 154 may include a sensor interface 156 (e.g., one or more analog-to-digital converters) to allow signals transmitted from one or more sensors 105, 106, 107 (e.g., the sensors described herein) to be converted into signals that can be understood and processed by the processor 150. It should be appreciated that the sensors 105, 106, 107 may be communicatively coupled to the communication module 154 using any suitable means. For example, as shown, the sensors 105, 106, 107 are coupled to the sensor interface 156 via a wired connection. However, in other embodiments, the sensors 105, 106, 107 may be coupled to the sensor interface 156 via a wireless connection (e.g., by using any suitable wireless communication protocol known in the art).

[0083] As used herein, the term "processor" refers not only to integrated circuits known in the art as being included in a computer, but also to controllers, microcontrollers, microcomputers, programmable logic controllers (PLCs), application specific integrated circuits, and other programmable circuits. Additionally, the memory device(s) 152 may generally include memory element(s), which include but are not limited to computer-readable media (e.g., random access memory (RAM)), computer-readable non-volatile media (e.g., flash memory), floppy disks, compact disc read-only memory (CD-ROM), magneto-optical discs (MOD), digital versatile discs (DVD), and / or other suitable memory elements. Such memory device(s) 152 may generally be configured to store suitable computer-readable instructions that, when implemented by the processor(s) 150, configure the controller(s) 104, 108 to perform the various functions described herein.

[0084] In addition, the network 110 coupling the field controller 108, the turbine controller 104, and / or the wind sensors 106 in the wind farm 100 may include any known communication network, such as a wired or wireless network, an optical network, etc. In addition, the network 110 may adopt any known topology, such as ring, bus, or hub, and may have any known contention resolution protocol not departing from the art. Thus, the network 110 is configured to provide data communication between the (multiple) turbine controllers 104 and the field controller 108 in near real-time.

[0085] Now referring to Figure 4 and Figure 5 , illustrated are a method 200 and a system 300 for controlling a wind turbine (e.g., one of the wind turbines 102 in the wind farm 100). More specifically, Figure 4 FIG. illustrates a flowchart of a method 200 for controlling a wind turbine in accordance with the present disclosure, while Figure 5 FIG. illustrates a schematic diagram of a system 300 for controlling a wind turbine in accordance with the present disclosure. Generally speaking, as Figure 4 shown in, the method 200 is described herein as being implemented for controlling the wind turbine 102 and / or the wind farm 100 described above. However, it should be recognized that the disclosed method 200 can be used to operate any other wind turbine and / or wind farm having any suitable configuration. In addition, although Figure 4 the steps depicted are performed in a particular order for purposes of illustration and discussion, the methods described herein are not limited to any particular order or arrangement. Using the disclosure provided herein, one of ordinary skill in the art will recognize that the various steps of the method can be omitted, rearranged, combined, and / or adjusted in a variety of ways.

[0086] As shown in (202), the method 200 includes detecting, via a controller, a plurality of analysis outputs related to the power performance of the wind turbine 102 from a plurality of different analyses. It will be understood that the controller configured to implement the method can be one or more of the field controller 108, the turbine controller 104, and / or any other suitable controller located within or remote from the wind farm 200. In addition, as is commonly understood, a wind turbine typically includes a plurality of performance analyses, which generally refer to the collected and analyzed data associated with the performance of the wind turbine, and the data is or can be classified, stored, and / or analyzed to study various trends or patterns in the data.

[0087] Thus, in one embodiment, as Figure 5As shown, system 300 may include a controller 302 (e.g., one of turbine controller 104 or farm-level controller 108), which receives various analysis outputs related to the power performance of one or more wind turbines in wind turbine 102, as shown at 304. Such analysis outputs may be calculated, for example, via various performance analyses. Additionally, as shown at 308, controller 302 is configured to detect various performance analysis outputs, which may be related to power curve production rate (e.g., low or high; contractual power curve), power curve threshold (e.g., farm-level learning curve), power curve history, power curve residuals (e.g., farm average comparison), and / or power ensemble (e.g., turbine-level learning model; anemometer agnostic).

[0088] More specifically, as described herein, "power ensemble" wind turbines generally refer to wind turbines identified as important elements in determining the power of a turbine of interest. Thus, power ensemble verification uses the average power from key reference wind turbines to determine power expectations. The power ensemble for a given wind turbine is determined by the wind turbines most relevant to the wind turbine of interest, which together provide the lowest uncertainty in determining the performance of the wind turbine of interest. The advantage of the power ensemble is that uncertainty is reduced by using only the power from multiple sensors.

[0089] Referring back to Figure 4 , as shown at (204), method 200 includes analyzing, via controller 302, multiple analysis outputs related to the power performance of wind turbine 102. For example, in one embodiment, controller 302 may filter the multiple analysis outputs related to power performance via a low-pass filter, a high-pass filter, a band-pass filter, or a combination thereof. More specifically, as shown at 306 in Figure 5 , controller 302 may filter the analysis outputs before detecting the type of performance analysis. Additionally, in other embodiments, controller 302 may also analyze the analysis outputs, for example, using principal component analysis or factorization to reduce the dimensionality in the analysis outputs.

[0090] Still referring to Figure 5 , controller 302 may also be configured to analyze the analysis outputs by organizing the analysis outputs into at least a first data set 310 and a second data set 312. In such embodiments, the first data set 310 of the multiple analysis data sets may include data from a first time length, and the second data set 312 may include data from a second time length. Thus, the first time length may be longer than the second time length. For example, as shown, the first data set 310 may include long-term data (e.g., two months), while the second data set 312 may include short-term data (e.g., one week).

[0091] Thus, referring back to Figure 4 , as shown at (206), method 200 includes generating or establishing, via controller 302, at least one computer-based model 314 of the power performance of wind turbine 102 using at least a portion of the analyzed plurality of analysis outputs. It will be understood that any number of models may be generated such that individual models may be created for subsets of the set of elements such that the lack of one or more element analyses will not prevent the algorithm from operating properly.

[0092] For example, in one particular embodiment, stepwise linear regression may be utilized to establish model(s) 314. Generally speaking, stepwise linear regression adds or deletes elements one at a time in an attempt to obtain the best regression model without overfitting. Additionally, stepwise regression typically has two variants, including forward and backward regression, both of which are within the scope and spirit of the present invention. For example, forward stepwise regression is a stepwise process of building a model by successively adding predictor variables. At each step, the models with and without the potential predictor variable are compared and the larger model is only accepted if it results in a significantly better fit of the data. Alternatively, backward stepwise regression starts with a model with all the predictors and removes terms that do not have statistical significance in modeling the response variable.

[0093] Another statistical method that may be used to generate model 314 may be the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm. Generally speaking, the LASSO algorithm minimizes the sum of squared residuals subject to the constraint that the sum of the absolute values of the coefficients is less than a constant. Still another statistical algorithm that may be used to generate model 314 is the M5 Prime (M5P) algorithm, which is a tree-based regression algorithm that is effective in many domains. For example, stepwise linear regression produces a single global linear model for the data, while tree-based regression algorithms perform logical tests on the elements to form a tree structure. Generally speaking, the M5P algorithm uses a linear regression model at each node of the tree, thus providing a more specialized model. Machine learning models that necessarily include direction may also be used along with the average of the power ensemble to determine equity (i.e., power expectation). This may be considered an improvement over previous methods of filtering data into specific direction sectors (which then form separate models for each sector). Other machine learning methods that may be used to generate model 314 may also include Gaussian process models, random forest models, support vector machines, and / or microservices, which are discussed in more detail herein.

[0094] Referring back to Figure 4 , as shown at (208), method 200 further includes training (e.g., via machine learning), via controller 302, the computer-based model(s) 314 of the power performance of wind turbine 102 using the annotated analysis output 316 related to the power performance of wind turbine 102. Thus, referring back toFigure 4 , as shown at (210), method 200 includes using a computer-based model 314 of at least one machine learning to estimate the power magnitude of wind turbine 210.

[0095] For example, in one embodiment, as shown at 318 in Figure 5 , controller 302 is configured to continuously train the computer-based model(s) by continuously determining the power magnitude of wind turbine 102 via model 314. Thus, as shown at 320, a human annotator can then classify each of the received power magnitudes from model 314 as underperforming, overperforming, or standard performance, and can also annotate the received power magnitude of wind turbine 102, i.e., by correcting the received power magnitude. As used herein, annotation in machine learning (e.g., annotated analysis) generally refers to the process of identifying data in a way that can be recognized by a machine or computer. Additionally, since human annotators are generally better at interpreting subjectivity, intent, and ambiguity within data, such annotation can be done manually by a human. Thus, the machine can learn from the annotated data by recognizing the human annotation over time. In some cases, the annotation can be learned through artificial intelligence and / or other algorithms (such as semi-supervised learning or clustering) and any other suitable accurate identification process.

[0096] The annotated power magnitudes can then be fed into model(s) 314 for training and / or correction. In some cases, as shown at 322, the human annotator can also determine a root cause analysis of the annotated power magnitude of wind turbine 102. As shown at 316 and as previously mentioned, the annotated power magnitude (and / or the root cause analysis of the annotated power magnitude) can also be stored in a dataset, which can be used to further update model 314 and / or for future use.

[0097] In other words, controller 302 can include a supervised machine learning algorithm that can apply past learned knowledge to new data using identified data to predict future performance (as shown at 324). From model building, the learning algorithm generates an inference function to make predictions about output values. Thus, controller 302 is capable of providing a target for any new input after sufficient training. The learning algorithm can also compare its output with the corrected expected output and find an error in order to modify the model accordingly.

[0098] In a particular embodiment, as Figure 6As shown, an example is a schematic diagram of an embodiment of an analysis microservices architecture 400 in accordance with the present disclosure. As shown, an analysis application programming interface (API) 402 is configured to send a power performance model output 404 to a controller 302, which provides the performance model output to a model 314. The model 314 then uses the new identification data to train the data. A model storage device 406 can be used to store the trained model, and a model file 408 can be read from the model storage device 406 and loaded to generate predictions. Feedback 410 from field engineers allows the model 314 to be improved over time.

[0099] Thus, as shown at Figure 4 (212) in, method 200 includes implementing a control action when the power magnitude of wind turbine 102 is outside a selected range (e.g., below a predetermined threshold or above a predetermined threshold). In one embodiment, for example, the control action can include generating an alert. It will be understood that the control actions as described herein can also include any suitable commands or constraints by controller 302. For example, in some embodiments, the control action can include derating (or reducing the rating) or ramping up (or increasing the rating) of wind turbine 102 temporarily.

[0100] Derating or ramping up wind turbine 102 can include derating or ramping up speed, derating or ramping up torque, or a combination of both. Additionally, as mentioned, wind turbine 102 can be derated or ramped up by pitching one or more rotor blades 22 of the rotor blades about its pitch axis 28. Wind turbine 10 can also be temporarily derated or ramped up by yawing nacelle 106 to change the angle of nacelle 106 relative to the wind direction. In additional embodiments, controller 302 can be configured to actuate one or more mechanical brakes to reduce the rotational speed of rotor blade 112. In yet another embodiment, controller 302 can be configured to perform any suitable control actions known in the art. Additionally, controller 302 can implement a combination of two or more control actions.

[0101] Furthermore, in some embodiments, method 200 can include determining a level of uncertainty associated with the power magnitude of wind turbine 102 and displaying the level of uncertainty via a user interface 322 of system 302. Uncertainty information can be useful because less analysis can lead to more decision / recommendation uncertainty.

[0102] Aspects and embodiments of the present invention are defined by the following numbered clauses:

[0103] Clause 1. A method for controlling a wind turbine, the method comprising:

[0104] Via a controller, detect a plurality of analysis outputs related to the power performance of the wind turbine from a plurality of different analyses;

[0105] Via the controller, analyze the plurality of analysis outputs related to the power performance of the wind turbine;

[0106] Via the controller, use at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of the wind turbine;

[0107] Via the controller, use the annotated analysis outputs related to the power performance of the wind turbine to train the at least one computer-based model of the power performance of the wind turbine;

[0108] Use the at least one machine learning-based computer model to estimate the power magnitude of the wind turbine; and

[0109] Implement a control action when the power magnitude of the wind turbine is outside a selected range.

[0110] Clause 2. The method according to clause 1, wherein the plurality of analysis outputs related to the power performance of the wind turbine include at least two of the following: low production rate of the power curve, power curve history, power curve residuals, or power set.

[0111] Clause 3. The method according to any one of the preceding clauses, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0112] Filter the plurality of analysis outputs related to the power performance.

[0113] Clause 4. The method according to any one of the preceding clauses, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0114] Use at least one of principal component analysis or factorization to reduce the dimensionality in the plurality of analysis outputs.

[0115] Clause 5. The method according to any one of the preceding clauses, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0116] Via the controller, organize the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set.

[0117] Clause 6. The method according to any one of the preceding clauses, characterized in that the first data set among the plurality of analysis data sets includes data from a first time length, and the second data set includes data from a second time length, and the first time length is longer than the second time length.

[0118] Clause 7. The method according to any one of the preceding clauses, characterized in that training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis output further includes:

[0119] Continuously receiving the power measurements of the wind turbine;

[0120] Classifying each of the received power measurements of the wind turbine as underperformance, overperformance, or standard performance;

[0121] Annotating the received power measurements of the wind turbine; and

[0122] Using the annotated power measurements of the wind turbine to machine-learn the at least one computer-based model of the power performance.

[0123] Clause 8. The method according to any one of the preceding clauses, characterized in that training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis output further includes:

[0124] Performing a root cause analysis on the annotated power measurements of the wind turbine.

[0125] Clause 9. The method according to any one of the preceding clauses, characterized in that the method further includes storing the root cause analysis of the annotated power measurements for future use.

[0126] Clause 10. The method according to any one of the preceding clauses, characterized in that the method further includes: determining a level of uncertainty associated with the power measurements of the wind turbine, and displaying the level of uncertainty via a user interface of the controller.

[0127] Clause 11. The method according to any one of the preceding clauses, characterized in that the at least one computer-based model includes a support vector machine.

[0128] Clause 12. The method according to any one of the preceding clauses, characterized in that the at least one computer-based model is a microservice.

[0129] Clause 13. A system for controlling a wind turbine, the system comprising:

[0130] A plurality of analyses for generating a plurality of analysis outputs related to the power performance of the wind turbine;

[0131] A controller communicatively coupled to the plurality of analyses, the controller configured to perform a plurality of operations, the plurality of operations including:

[0132] Receiving the plurality of analysis outputs from the plurality of analyses;

[0133] Analyzing the plurality of analysis outputs related to the power performance of the wind turbine;

[0134] Using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of the wind turbine;

[0135] Via the controller, training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine;

[0136] Using the at least one machine learning-based computer model to estimate the power magnitude of the wind turbine; and

[0137] Implementing a control action when the power magnitude of the wind turbine is outside a selected range.

[0138] Clause 14. The system according to clause 13, wherein the plurality of analysis outputs related to the power performance of the wind turbine include at least two of the following: low power curve production rate, power curve history, power curve residuals, or power set.

[0139] Clause 15. The system according to clauses 13-14, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0140] Filtering the plurality of analysis outputs related to the operation of the wind turbine.

[0141] Clause 16. The system according to clauses 13-15, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0142] Using at least one of principal component analysis or factorization to reduce the dimensionality of the plurality of analysis outputs.

[0143] Clause 17. The system according to clauses 13-16, wherein analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes:

[0144] Via the controller, organize the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set, wherein a first data set of the plurality of analysis data sets includes data from a first time length and the second data set includes data from a second time length, and the first time length is longer than the second time length.

[0145] Clause 18. The system according to clauses 13-17, wherein training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine further includes:

[0146] Continuously receive the power measurements of the wind turbine;

[0147] Classify each power measurement in the received power measurements of the wind turbine as underperformance, overperformance, or standard performance;

[0148] Annotate the received power measurements of the wind turbine; and

[0149] Use the annotated power measurements of the wind turbine to machine-learn the at least one computer-based model of the power performance.

[0150] Clause 19. The system according to clauses 13-18, wherein training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine further includes:

[0151] Perform a root cause analysis on the annotated power measurements of the wind turbine; and

[0152] Store the root cause analysis of the power measurements of the wind turbine for future use.

[0153] Clause 20. A wind farm, comprising:

[0154] A plurality of wind turbines, each of the wind turbines including a turbine controller;

[0155] A field-level controller communicatively coupled to each of the turbine controllers, the field-level controller configured to perform a plurality of operations, the plurality of operations including:

[0156] Receive a plurality of analysis outputs related to the power performance of each of the wind turbines from a plurality of different analyses;

[0157] Analyze the plurality of analysis outputs related to the power performance of each of the wind turbines;

[0158] Use at least a portion of the plurality of analyzed analysis outputs to generate at least one computer-based model of the power performance of each wind turbine in the wind turbines;

[0159] Use the annotated analysis outputs related to the power performance of each wind turbine in the wind turbines to train the at least one computer-based model of the power performance of each wind turbine in the wind turbines;

[0160] Use the at least one machine learning-based computer model to estimate the power magnitude of each wind turbine in the wind turbines; and

[0161] Implement a control action when the power magnitude of any one of the wind turbines is outside a selected range.

[0162] This written description uses examples to disclose the invention, including the best mode, and also enables one of ordinary skill in the art to practice the invention, including making and using any device or system and performing any incorporated method. The patentable scope of the invention is defined by the claims and may include other examples that occur to one of ordinary skill in the art. If such other examples include structural elements that are not different from the literal language of the claims or if such other examples include equivalent structural elements that are not materially different from the literal language of the claims, they are considered to be within the scope of the claims.

Claims

1. A method for controlling a wind turbine in a wind farm, the method comprising: Via a controller, detecting, from a plurality of different analyses, a plurality of analysis outputs related to the power performance of the wind turbine, the plurality of analysis outputs related to the power performance of the wind turbine including at least a power set analysis, the power set analysis using the average power from key reference wind turbines in the wind farm to determine a power expectation value of the wind turbine; Via the controller, analyzing the plurality of analysis outputs related to the power performance of the wind turbine; Via the controller, using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of the wind turbine, wherein the at least a portion of the analyzed plurality of analysis outputs includes the power set analysis; Via the controller, using the annotated analysis outputs related to the power performance of the wind turbine to train the at least one computer-based model of the power performance of the wind turbine; Using the at least one machine learning-based computer model to estimate the power magnitude of the wind turbine; and Implementing a control action when the power magnitude of the wind turbine is outside a selected range.

2. The method according to claim 1, characterized in that The plurality of analysis outputs related to the power performance of the wind turbine further includes at least one of the following: low power curve production rate, power curve history, or power curve residual.

3. The method according to claim 1, characterized in that, Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: Filtering the plurality of analysis outputs related to the power performance.

4. The method according to claim 1, characterized in that, Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: Using at least one of principal component analysis or factorization to reduce the dimensionality in the plurality of analysis outputs.

5. The method according to claim 1, characterized in that, Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: Via the controller, organizing the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set.

6. The method according to claim 5, characterized in that The first data set in the plurality of analysis data sets includes data from a first time length, and the second data set includes data from a second time length, the first time length being longer than the second time length.

7. The method according to claim 1, characterized in that, Using the annotated analysis outputs to train the at least one computer-based model of the power performance of the wind turbine further includes: Continuously receiving the power magnitude of the wind turbine; Classifying each of the received power magnitudes of the wind turbine as underperformance, overperformance, or standard performance; Annotating the received power magnitudes of the wind turbine; and Using the annotated power magnitudes of the wind turbine to machine learn the at least one computer-based model of the power performance.

8. The method according to claim 7, wherein Using the annotated analysis outputs to train the at least one computer-based model of the power performance of the wind turbine further includes: Performing a root cause analysis on the annotated power magnitudes of the wind turbine.

9. The method according to claim 8, wherein The method further includes storing the root cause analysis of the annotated power magnitudes for future use.

10. The method according to claim 1, wherein The method further includes: determining an uncertainty level associated with a power magnitude of the wind turbine, and displaying the uncertainty level via a user interface of the controller.

11. The method according to claim 1, wherein The at least one computer-based model includes a support vector machine.

12. The method according to claim 1, characterized in that, The at least one computer-based model is a microservice.

13. A system for controlling a wind turbine in a wind farm, the system comprising: a plurality of analyses for generating a plurality of analysis outputs related to a power performance of the wind turbine, the plurality of analysis outputs including at least a power ensemble analysis, the power ensemble analysis using an average power from a key reference wind turbine in the wind farm to determine a power expectation value of the wind turbine; a controller communicatively coupled to the plurality of analyses, the controller configured to perform a plurality of operations, the plurality of operations including: receiving the plurality of analysis outputs from the plurality of analyses; analyzing the plurality of analysis outputs related to the power performance of the wind turbine; using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of the wind turbine, wherein the at least a portion of the analyzed plurality of analysis outputs includes the power ensemble analysis; training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine via the controller; estimating a power magnitude of the wind turbine using the at least one machine learning-based computer-based model; and implementing a control action when the power magnitude of the wind turbine is outside a selected range.

14. The system according to claim 13, wherein The plurality of analysis outputs related to the power performance of the wind turbine further includes at least one of the following: a low production rate of a power curve, a power curve history, or a power curve residual.

15. The system according to claim 13, wherein Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: filtering the plurality of analysis outputs related to the operation of the wind turbine.

16. The system according to claim 13, wherein Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: using at least one of principal component analysis or factorization to reduce a dimension in the plurality of analysis outputs.

17. The system according to claim 13, wherein, Analyzing the plurality of analysis outputs related to the power performance of the wind turbine further includes: organizing, via the controller, the plurality of analysis outputs related to the power performance of the wind turbine into at least a first data set and a second data set, wherein a first data set of the plurality of analysis data sets includes data from a first time length and the second data set includes data from a second time length, the first time length being longer than the second time length.

18. The system according to claim 17, wherein, Training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine further includes: continuously receiving a power magnitude of the wind turbine; classifying each power magnitude of the received power magnitudes of the wind turbine as underperformance, overperformance, or standard performance; annotating the received power magnitudes of the wind turbine; and Machine learning the at least one computer-based model of the power performance using the annotated power quantity values of the wind turbine.

19. The system according to claim 18, wherein Training the at least one computer-based model of the power performance of the wind turbine using the annotated analysis outputs related to the power performance of the wind turbine further includes: Performing a root cause analysis on the annotated power quantity values of the wind turbine; and Storing the root cause analysis of the power quantity values of the wind turbine for future use.

20. A wind farm, comprising: A plurality of wind turbines, each of the wind turbines including a turbine controller; A field-level controller communicatively coupled to each of the turbine controllers, the field-level controller configured to perform a plurality of operations, the plurality of operations including: Receiving a plurality of analysis outputs related to the power performance of each of the wind turbines from a plurality of different analyses, the plurality of analysis outputs related to the power performance of the wind turbine at least including a power set analysis that uses the average power of key reference wind turbines in the wind farm to determine the power expectation value of the wind turbine; Analyzing the plurality of analysis outputs related to the power performance of each of the wind turbines; Using at least a portion of the analyzed plurality of analysis outputs to generate at least one computer-based model of the power performance of each of the wind turbines, wherein the at least a portion of the analyzed plurality of analysis outputs includes the power set analysis; Training the at least one computer-based model of the power performance of each of the wind turbines using the annotated analysis outputs related to the power performance of each of the wind turbines; Using the at least one machine-learned computer-based model to estimate the power quantity values of each of the wind turbines; and Implementing a control action when the power quantity value of any one of the wind turbines is outside a selected range.

Images