System and method for evaluating and verifying wind turbine and wind farm performance

Abstract

A method for evaluating or validating wind turbine or wind farm performance resulting from one or more upgrades is provided. Measurements of operational data from wind turbines in the wind farm are obtained. A baseline model of performance is generated, and each of the baseline models is developed from a different portion of the operational data. The generating step filters the wind turbines so that they are in a random state of equilibrium. An optimal baseline model of performance is selected from the baseline models, and the optimal baseline model includes a direction. The optimal baseline model of performance is compared with actual performance of the wind farm or wind turbines. The comparing step determines a difference between the optimal baseline model of power output and the actual power output of the wind farm / turbine. This difference reflects the change in power output resulting from the upgrade.

Description

Technical Field

[0001] The present disclosure relates generally to wind turbines and wind farms, and more particularly to a system and method for evaluating the benefits of wind farm performance and validating wind farm performance to determine improvements and / or underperformance (eg, power output). Background Art

[0002] Wind power is considered to be one of the cleanest and most environmentally friendly energy sources currently available, and wind turbines have received increased 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 use the known airfoil principle to obtain the kinetic energy of the wind. For example, the rotor blades typically have a cross-sectional profile of an airfoil so that during operation, air flows over the blades to create a pressure difference between the sides. Therefore, a lift force directed from the pressure side toward the suction side acts on the blades. The lift force generates a torque on the main rotor shaft, which is engaged to a generator to generate electricity.

[0003] Multiple wind turbines are often used in conjunction with each other to generate electricity, and are often referred to as a "wind farm". Wind turbines on a wind farm often include their own meteorological monitors that perform, for example, temperature, wind speed, wind direction, atmospheric pressure, and / or air density measurements. In addition, a separate meteorological mast or tower ("met mast") with higher quality meteorological instrumentation may also be provided that can provide more accurate measurements at one point in the farm. The correlation of meteorological data to power output provides for the empirical determination of a "power curve" for an individual wind turbine.

[0004] Typically, in a wind farm, each wind turbine attempts to maximize its own power output while keeping its fatigue loads within desired limits. To this end, each turbine includes a control module that attempts to maximize the turbine's power output in the face of varying wind and grid conditions while satisfying constraints such as subsystem ratings and component loads. Based on the determined maximum power output, the control module controls the operation of various turbine components (such as generators / power converters, pitch systems, brakes, and yaw mechanisms) to achieve maximum power efficiency.

[0005] Typically, while maximizing the power output of a single wind turbine, adjacent turbines may be negatively affected. For example, a downwind turbine may experience a large wake effect caused by an upwind turbine. The wake effect includes a reduction in wind speed downwind from the wind turbine and increased wind turbulence, which is usually caused by the conventional operation of the upwind turbine (i.e., for maximum power output). Due to these wake effects, the downwind turbine receives wind at a lower speed, which greatly affects its power output (because the power output is proportional to the wind speed). In addition, wind turbulence can have a negative impact on the fatigue load placed on the downwind turbine, and thereby affect its life (because the life is proportional to the fatigue load). Therefore, the maximum efficiency of some wind turbines can lead to suboptimal power output, performance or life of other wind turbines in the wind farm. Therefore, modern control technology attempts to optimize the wind farm power output rather than the power output of each individual wind turbine.

[0006] In addition, there are many products, features and / or upgrades available for wind turbines and / or wind farms to increase the power output or annual energy production (AEP) of the wind farm. Once the upgrade is installed, it is advantageous to effectively determine various wind turbine performance improvement measurements to verify the benefits of the upgrade. For example, a typical method for evaluating wind turbine performance measurements is a baseline power for wind speeds such as those evaluated by a turbine nacelle anemometer. However, the nacelle anemometer approach is sometimes hindered by the inaccuracy of the nacelle anemometer measurements and the input of these measurements into the AEP estimate. In addition, this approach may not be as preferred as using an external meteorological mast in front of the wind turbine, but this approach is widely used due to the generally excessive cost of the meteorological mast approach. In addition, even when the nacelle anemometer is correctly calibrated, the wind curve method alone cannot discern the benefits of the upgrade, such as wake minimization technology, which can create more wind for the field to use. Other types of upgrades are related to software / controls and involve optimizing the pitch, speed, or yaw angle of a given turbine to maximize turbine and field power. These upgrades can also potentially confound anemometer-based performance assessments because controlling these parameters can alter the air flow to the anemometer.

[0007] In addition to evaluating the performance of upgrades to turbines and farms, performance benefit (power expectancy) measurement techniques and methods are needed to discern whether a particular wind turbine is operating properly or is underperforming due to material or software issues. Likewise, anemometer-based assessment techniques can have difficulties if there are problems with the anemometer sensor itself. Summary of the invention

[0008] Aspects and advantages of the invention will be set forth in part in the following description, or may be obvious from the description, or may be learned through practice of the invention.

[0009] In one aspect, the present disclosure relates to a method for evaluating or verifying the performance of a wind turbine or wind farm resulting from one or more upgrades. The method includes a measuring step for measuring operational data of one or more wind turbines from a wind farm via one or more sensors. A generating step generates multiple baseline models of the performance of the wind farm or multiple baseline models of the performance of wind turbines in the wind farm from at least a portion of the operational data via a processor. Each of the baseline models of performance is developed from a different portion of the operational data of one or more wind turbines from the wind farm. The generating step includes filtering one or more wind turbines so that the one or more wind turbines are in a random state of equilibrium. The selecting step selects an optimal baseline model of performance from the multiple baseline models via a processor, wherein the optimal baseline model includes a direction. A comparing step compares the optimal baseline model of performance with an actual performance of the wind farm or wind turbine via a processor. The actual performance of the wind farm or wind turbine is determined after one or more wind turbines of the wind farm are modified by one or more upgrades that have been installed for the one or more wind turbines. The upgrade includes at least one of a tip speed ratio, a pitch setting, or a yaw offset setting. Comparing the optimal baseline model of performance with the actual performance of the wind farm or wind turbines includes determining a difference between the optimal baseline model of power output and the actual power output of the wind farm or wind turbines. The difference reflects a change in power output resulting from one or more upgrades.

[0010] On the other hand, the present disclosure relates to a method for evaluating or verifying the performance of a wind turbine. The measuring step measures operational data from the wind turbine via one or more sensors. The generating step generates multiple baseline models of the performance of the wind turbine from at least a portion of the operational data via a processor. Each of the baseline models of performance is developed from a different portion of the operational data. The generating step includes filtering the data so that the data is in a balanced random state. The selecting step selects an optimal baseline model of performance from a plurality of baseline models via a processor, and the optimal baseline model includes a direction. The comparing step compares the optimal baseline model of performance with the actual performance of the wind turbine via a processor. The baseline model includes at least one of a tip speed ratio, a pitch setting, or a yaw offset setting. Comparing the optimal baseline model of performance with the actual performance of the wind turbine includes determining a difference between the optimal baseline model of power output and the actual power output of the wind turbine. This difference reflects a change in power output resulting from the absence or presence of one or more upgrades to the wind turbine.

[0011] In yet another aspect, the present disclosure relates to a system for verifying one or more wind farm performance measurements generated by one or more upgrades. The system includes a plurality of sensors configured to measure operational data from one or more wind turbines in a wind farm. A controller is configured to perform a plurality of operations or steps. A generating step generates a plurality of baseline models of performance of the wind farm from at least a portion of the operational data. Each of the baseline models of performance is developed from a different portion of the operational data of one or more wind turbines in the wind farm. The generating step includes filtering one or more turbines so that the one or more turbines are in a random state of equilibrium. The selecting step selects an optimal baseline model of performance from the plurality of baseline models, and the optimal baseline model includes a direction. The comparing step compares the optimal baseline model of performance with an actual performance of the wind farm. The actual performance of the wind farm is determined after one or more wind turbines in the wind farm are modified by one or more upgrades that have been installed for the one or more wind turbines in the wind farm. The upgrades include at least one of a tip speed ratio and a pitch setting. Comparing the optimal baseline model of performance with the actual performance of the wind farm includes determining a difference between the optimal baseline model of power output and the actual power output of the wind farm. This difference reflects a change in power output generated by the one or more upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] A full and complete disclosure of the present invention including the best mode thereof to one of ordinary skill in the art is set forth in the specification with reference to the accompanying drawings, in which:

[0013] Figure 1 A perspective view of one embodiment of a wind farm according to the present disclosure is shown.

[0014] Figure 2 A perspective view of one embodiment of a wind turbine according to the present disclosure is shown.

[0015] Figure 3 A block diagram of one embodiment of a controller for a wind turbine and / or wind park according to the present disclosure is shown.

[0016] Figure 4 A graph illustrating one embodiment of the percentage (%) uncertainty of AEP change in response to one or more upgrades as a function of the number of wind turbines according to the present disclosure.

[0017] Figure 5 A flow chart of one embodiment of a method for validating wind farm performance measurements according to the present disclosure is shown.

[0018] Figure 6 A graph illustrating one embodiment of average power output of multiple statistical algorithms according to the present disclosure.

[0019] Figure 7A graph illustrating another embodiment of average power output using multiple statistical algorithms of different characteristics according to the present disclosure.

[0020] Figure 8 A graph of one embodiment of power output versus time for multiple baseline models according to the present disclosure is shown.

[0021] Fig. 9 A graph of one embodiment of power output versus time for an optimal baseline model according to the present disclosure is shown.

[0022] Fig.10 A layout of one embodiment of a wind farm according to the present disclosure is shown, particularly showing selected wind turbines of one of the baseline models.

[0023] Fig.11 A graph illustrating one embodiment of the average residual between actual power and predicted power according to the present disclosure.

[0024] Fig.12 A graph showing one embodiment of actual and predicted power output as a function of fan zone in accordance with the present disclosure.

[0025] Fig.13 Several graphs of long term wind speed distribution and one embodiment of a reference power curve for integrating power changes relative to the long term power distribution are shown in accordance with the present disclosure.

[0026] Fig.14 A flow chart showing another embodiment of a method for validating wind farm performance measurements according to the present disclosure is shown.

[0027] Fig.15 Results of a Gaussian process model according to the present disclosure for a turbine of interest are shown.

[0028] Fig.16 is a flow chart of a method for determining a power set for a selected wind turbine according to the present disclosure. DETAILED DESCRIPTION

[0029] Reference will now be made in detail to embodiments of the present invention, one or more examples of which are shown in the accompanying drawings. Each example is provided by way of illustration of the present invention without limiting the present invention. In fact, it will be clear 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 present invention. For example, a feature shown or described as a part of an embodiment can be used in conjunction with another embodiment to produce yet another embodiment. Therefore, it is intended that the present invention covers such modifications and variations as fall within the scope of the appended claims and their equivalents.

[0030] In general, the present disclosure relates to improved systems and methods for verifying or identifying wind farm performance measurements in a wind farm. For example, in one embodiment, the system is configured to verify wind farm performance measurements, i.e., power output, in response to one or more upgrades installed within a wind farm, in order to optimize wind farm performance. Alternatively, the system may be configured to identify wind farm performance measurements (i.e., problems or errors) in a wind farm, such as, for example, misconfigurations, material or software failures, and / or other problematic conditions that result in sub-par performance of the wind farm.

[0031] In various embodiments, a park controller is configured to estimate a change in annual energy production (AEP) after installing an upgrade for one or more wind turbines (e.g., a turbine of interest) of a wind farm. More specifically, the park controller is configured to receive and store operational data collected from a plurality of wind turbines (including the turbine of interest) of the wind farm measured by one or more sensors. The park controller then generates a plurality of baseline models of performance (e.g., power output) of the wind farm from the operational data. For example, in one embodiment, the park controller selects a plurality of wind turbine subsets from the wind farm to form a baseline model that does not include the turbine of interest, and selects a model that achieves optimal accuracy, i.e., an optimal baseline model. Then, after one or more upgrades have been installed, the park controller compares the optimal baseline model of performance with the actual performance of the wind farm. For example, in one embodiment, the controller compares the optimal baseline model of power output with the actual power output, and determines the associated uncertainty of the change in power output. It should be understood that the upgrade may include any suitable upgrade now known in the art or later developed, including but not limited to rotor blade chord extension, software upgrade, control upgrade, hardware upgrade, wake control, aerodynamic upgrade, blade tip extension, vortex generator, winglet, etc. Thus, the park controller is configured to verify an improvement in wind park performance (such as an improvement in wind park power) that occurs in response to installing the at least one upgrade.

[0032] The present disclosure has many advantages not present in the prior art. For example, the present disclosure uses machine learning algorithms to utilize and fuse accurate available sensor data. That is, the more relevant and better quality sensors are used and the more data collected under similar conditions, the lower the prediction error of the optimal baseline model will be. Therefore, the accuracy of wind farm performance, improved measurements and the associated costs and time can be reduced.

[0033] Now referring to the accompanying drawings, Figure 1 An exemplary embodiment of a wind farm 100 including a plurality of wind turbines 102 according to aspects of the present disclosure is shown. The wind turbines 102 may be arranged in any suitable manner. For example, the wind turbines 102 may be arranged in an array of rows and columns, in a single row, in a random arrangement, or in an environmentally advantageous position. In addition, Fig.10An example layout of one embodiment of a wind farm 100 is shown. Typically, the wind turbine arrangement in a wind farm is determined based on a number of optimization algorithms so as to maximize the AEP for the corresponding on-site wind climate. It should be understood that any wind turbine arrangement may be implemented, such as on uneven land, without departing from the scope of the present disclosure.

[0034] Furthermore, it should be appreciated that wind turbines 102 of wind farm 100 may have any suitable configuration, such as, for example, Figure 2 As shown, wind turbine 102 includes a tower 114 extending from a supporting surface, a nacelle 116 mounted atop tower 114, and a rotor 118 coupled to nacelle 116. The rotor includes a rotatable hub 120 having a plurality of rotor blades 112 mounted thereon, which in turn is connected to a main rotor shaft that is coupled to a generator housed within nacelle 116 (not shown). Thus, the generator generates electricity from the rotational energy generated by rotor 118. It should be appreciated that Figure 2 The wind turbine 102 is provided for illustration purposes only. Therefore, one of ordinary skill in the art should understand that the present invention is not limited to any particular type of wind turbine configuration.

[0035] As generally shown, each wind turbine 102 of the wind farm 100 may also include a turbine controller 104 communicatively coupled to a farm controller 108. In addition, in one embodiment, the farm controller 108 may be coupled to the turbine controller 104 via a network 110 to facilitate communication between the various wind farm components. The wind turbine 102 may also include one or more sensors 105 configured to monitor various operational, wind, and / or load conditions of the wind turbine 102. For example, the one or more sensors may include a blade sensor for monitoring the rotor blades 112 or the pitch setting of the rotor blades; a generator sensor for monitoring the generator load, torque, speed, acceleration, and / or generator power output; a wind sensor for monitoring one or more wind conditions; and / or a shaft sensor for measuring the load of the rotor shaft and / or the rotational speed of the rotor shaft. In addition, the wind turbine 102 may include one or more tower sensors for measuring loads transmitted through the tower 114 and / or accelerations of the tower 114. In various embodiments, the sensor may be any one or combination of the following: an accelerometer, a pressure sensor, an angle of attack sensor, a vibration sensor, a micro inertial measurement unit (MIMU), a camera system, a fiber optic system, an anemometer, a wind vane, a sound detection and ranging (SODAR) sensor, an infrared laser, a light detection and ranging (LIDAR) sensor, a radiometer, a pitot tube, a radiosonde, other optical sensors and / or any other suitable sensor.

[0036] Reference now Figure 3, shows a block diagram of one embodiment of suitable components that may be included within a field controller 108 and / or a turbine controller 104 according to aspects of the present disclosure. As shown, the controller 108 may include one or more processors 150 and associated memory devices 152, which are configured to perform various computer-implemented functions (e.g., perform methods, steps, calculations, etc., and store relevant data as disclosed herein). In addition, the controller 108 may also include a communication module 154 to facilitate communication between the controller 108 and various components of the wind turbine 102. In addition, 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 103, 105, 107, 109 (such as 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 103, 105, 107, 109 can be communicatively coupled to the communication module 154 using any suitable means. For example, as shown, the sensors 103, 105, 107, 109 are coupled to the sensor interface 156 via a wired connection. However, in other embodiments, the sensors 103 , 105 , 107 , 109 may be coupled to the sensor interface 156 via a wireless connection, such as by using any suitable wireless communication protocol known in the art.

[0037] As used herein, the term "processor" refers not only to what is known in the art as an integrated circuit included in a computer, but also to controllers, microcontrollers, microcomputers, programmable logic controllers (PLCs), application-specific integrated circuits, and other programmable circuits. In addition, the memory device 152 may generally include memory elements, including but not limited to computer-readable media (e.g., random access memory (RAM)), computer-readable non-volatile media (e.g., flash memory), floppy disks, compact disk read-only memory (CD-ROM), magneto-optical disks (MOD), digital versatile disks (DVD), and / or other suitable memory elements. This memory device 152 may generally be configured to store suitable computer-readable instructions, which, when implemented by the processor 150, configure the controller 108 to perform various functions as described herein. In addition, the network 110 that connects the field controller 108, the turbine controller 104, and the wind sensor 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 be connected in any known topology, such as a ring, a bus, or a hub, and may have any known contention resolution protocol without departing from the art. Thus, network 110 is configured to provide data communication between turbine controller 104 and park controller 108 in near real-time.

[0038] In addition, the farm controller 108 is configured to obtain or collect data from the sensors 103, 105, 107, 109 and / or other data sources such as turbine empirical models. Using these inputs, the farm controller 108 is configured to determine one or more baseline models of the performance of the wind farm 100, which can be used to verify the farm-level performance (e.g., farm-level power output) of the wind farm 100. More specifically, the farm controller 108 uses multi-feature estimation to establish a wind turbine performance baseline, which normalizes the AEP uncertainty estimate and does not rely solely on nacelle anemometer estimates or expensive additional sensors. For example, in one embodiment, the present disclosure provides a unified paradigm that uses a machine learning algorithm to utilize and fuse the best available sensors. Therefore, regardless of the machine learning model used, the farm controller 108 is configured to utilize big data and many sensors to improve the accuracy of the estimates. For example, as Figure 4 As shown in , as the number of wind turbines 102 used to collect operational data increases, the percentage (%) uncertainty decreases. In addition, as the time period for collecting data increases (e.g., from a few minutes to a few years), as shown by the different line types, the % uncertainty decreases. Therefore, the more relevant and better quality sensors are used and the more data collected under similar conditions, the lower the prediction error of the model will be.

[0039] Reference now Figure 5 , a simplified flow chart is shown illustrating one embodiment of a method 200 for validating performance improvement measurements of a wind farm 100 according to the present disclosure. As shown at 202, the farm controller 108 may include a global repository configured to collect and store operational data collected from sensors 103, 105, 107, 109. It should be understood that the wind turbine operational data may include any relevant operational data about the wind turbine 102, including but not limited to pitch angle, generator speed, power output, torque output, air density, temperature, pressure, wind speed, wind peak, wind turbulence, wind shear, wind direction, tip speed ratio, etc.

[0040] At 204, the operational data may be filtered to provide filtered data 206. For example, in one embodiment, the data may be filtered for a variety of reasons, including but not limited to: when one of the wind turbines 102 is not fully operational during a specific time interval (e.g., a 10 minute interval), when a curtailment is detected, and / or when a data point has a very low wind speed (e.g., less than 2 meters / second). It should be understood that any other suitable filtering parameters may be used to filter the data in addition to those specifically described herein. The operational data is filtered so that the turbines are in a state of equilibrium in a random manner. This can be accomplished by utilizing a threshold of the standard deviation of the average state turbine state through some historical window. For example, the turbine state in this context refers to the minimum tip speed ratio and / or pitch offset setting of the control system, but can be expanded to include any turbine configuration parameters that result in performance optimization. This ensures that there is a switch in turbine state within the period of interest. This metric may be combined with a metric related to the entropy of the turbine state to ensure that all turbines do not change state together. Any method of ensuring randomization of turbine states will be sufficient. Other filtering may include filtering time periods when turbines are curtailed, or time periods when a certain minimum number of turbines are not operating.

[0041] At 208, data may optionally be synchronized or coordinated to establish consistency between data collected from multiple sources. In addition, at 210, the field controller may be configured to estimate data when there are missing values ​​in the historical data. For example, the above initial filtering may produce gaps in the operational data. Since various embodiments of the present disclosure require estimating the power from one or more wind turbines of interest and using information from neighboring turbines, the absence of measured values ​​at a given time interval requires the corresponding data records to be completely removed, resulting in severe data loss. Therefore, the field controller 108 is configured to estimate missing data, that is, replace missing values ​​with estimated values ​​to obtain a more complete data set for modeling and analysis. It should be understood that any estimation method known in the art may be used, including choosing not to estimate but using the mean of non-missing values ​​as the optimal value. For example, in one embodiment, a k-nearest neighbor algorithm may be used to estimate missing data. This algorithm replaces the missing data of a given turbine with the weighted mean of the measured values ​​of its k nearest neighbors. In some embodiments, the weight is inversely proportional to its Euclidean distance from the neighboring turbine.

[0042] In other embodiments (e.g., when using wind speed data from adjacent turbines), linearization of the operational data may be useful prior to incorporation into the power model. For example, as shown at 212, a Bayesian power curve approach may be utilized to linearize the operational data to produce a power estimate for input into the modeling algorithm. At 214, the linear and nonlinear data are integrated to provide integrated data 216.

[0043] At 218, one or more statistical models are established to estimate the performance measurement improvement (e.g., power output) from the individual wind turbines 102. For example, in certain embodiments, stepwise linear regression may be used to estimate the power output from the individual wind turbines 102. In general, stepwise linear regression adds or removes one feature at a time to try to obtain the optimal regression model without overfitting. In addition, stepwise regression generally has two variants, including forward regression 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, models with and without potential predictor variables are compared, and larger models are accepted only when the larger model results in a significantly better fit to the data. Alternatively, backward stepwise regression starts with a model with all predictors and removes items that are not statistically significant in modeling the response variable.

[0044] Another statistical method that can be used to verify the performance measurement value improvement (such as power output) of the wind farm 100 is the least absolute convergence and selection operator (LASSO) algorithm. In general, the LASSO algorithm minimizes the residual sum of squares under the constraint that the absolute value sum of the coefficients is less than a constant. Another statistical algorithm that can be used to verify the performance measurement value improvement from the wind farm 100 is the M5 Prime (M5P) algorithm, which is a tree-based regression algorithm that is effective in many fields. For example, given that stepwise linear regression produces a single global linear model for the data, the tree-based regression algorithm performs logical tests on the features to form a tree structure. In general, the M5P algorithm uses a linear regression model at each node of the tree, providing a more specialized model. The machine learning model that must include the direction is used together with the mean of the power set group to determine the benefit (i.e., power expectation). This can be considered as an improvement to the previous method of filtering the data to a specific direction sector (which then forms a separate model for each sector).

[0045] Other machine learning approaches that have shown success in determining performance benefit models include Gaussian process models, random forest models, and support vector machines. In one embodiment, a Gaussian process model is used with three features: the sine and cosine of the field median direction, and the NaN mean of the power of the "power aggregate wind turbines." The "power aggregate wind turbines" are wind turbines that are identified as important features in determining the power of the turbines of interest. Fig.15 Results for the Gaussian process model for the turbine of interest are shown. Power aggregation validation utilizes the average power from key reference wind turbines to determine power expectations. The power aggregation for a given wind turbine is determined by the wind turbines that are 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 power aggregation is that uncertainty is reduced by using only power from multiple sensors.

[0046] Fig.16 1 is a flow chart of a method 1600 for determining a power set for a selected wind turbine. The method 1600 reduces uncertainty using a turbine group with equilibrium and random states, and the method can be applied to a single wind turbine or multiple wind turbines of interest. The optimal power set is determined by using the NaN mean of the control wind turbine group. The null case can be used to ensure that there is no bias. The method incorporates the direction by inputting the field-level direction into a Gaussian process model together with the tip speed ratio and the pitch setting. The annual energy production gain is determined based on executing the Gaussian process model. In step 1610, a power set is determined from a correlation matrix to reduce or minimize uncertainty. For example, a synchronized wind farm-level power matrix is ​​created from input data (which is obtained from multiple sensors). The input data is filtered to obtain an experimental state design of a balanced set and a predetermined number of operating wind turbines. The power set wind turbines are then sorted by pairwise correlation and operating probability. The power set is then determined by starting with the most relevant wind turbine and allowing missing values ​​that reduce or minimize uncertainty.

[0047] In step 1620, a machine learning model is created. For example, the model may be a Gaussian process model, support vector regression, random forest, neural network model or other machine learning techniques. The machine learning model is created from the power set NaN mean input, direction and experimental state design to provide the estimated power output and estimated uncertainty of the wind turbine of interest. In step 1630, the annual power generation gain is determined from the estimated wind turbine power and the uncertainty of the value. Machine learning or Gaussian process models allow the creation of models that interpolate well between points and provide uncertainty measures. Support vector machines and random forests will also work in this application. The sine and cosine of the direction are included in a single model as explicit features. Additional advantages are provided by using the NaN mean (i.e., the mean of all non-missing wind turbine sensor values ​​is equally weighted). This method is particularly suitable for a single wind turbine, but can also be used for multiple wind turbines or even an entire wind farm.

[0048] Return to reference Figure 6, compare sample results from statistical algorithms for verifying power output, i.e., stepwise linear regression (line 224), LASSO (line 226), and M5P (line 228). As shown, stepwise algorithm 224 performs as well as the more complex LASSO (line 226) and M5P (line 228) algorithms for a particular turbine, but with some variations. In addition, as shown, LASSO (line 226) and M5P (line 228) algorithms provide similar power estimates for wind farm 100. However, in some months, M5P (line 228) performs best, while in other months, LASSO 226 performs better. It should be understood that the various statistical models described herein are provided only as examples, and any other statistical models are within the spirit and scope of the present invention.

[0049] Reference now Figure 7 , shows the root mean square error (RMSE) of the predicted power of various models divided by the average power produced in the data for a given month, which can be interpreted as a measure of the prediction error normalized by the output power. In certain embodiments, using training data equivalent to about one month provides an effective means of predicting turbine power for a field of 100 turbines, however it should be understood that any amount of data can be used to develop various models. For example, as shown in the illustrated embodiment, when all features including Bayesian power curve power estimation are included in the feature set (line 230), the wind turbine 102 trained using January data can produce an error of about 5% training. However, when only wind speed and power are used (line 232), the error of the training data is about 8%, and when only power from other turbines is used (line 234), the error rises to 15%. If only wind speed is used in the model (line 236), the error of predicting 10-minute intervals from the model for the training data exceeds 23%. When these models trained in January are applied to test data for other months of the year, the error characteristics are shown. Thus, as shown in the illustrated embodiment, optimal results are obtained when an onboard nacelle anemometer and a Bayesian power curve estimate are included as features (line 230 ).

[0050] Reference now Figure 8 and Fig. 9 , the present disclosure includes comparing and analyzing multiple baseline models and selecting the model that provides the best or most accurate performance (i.e., the optimal baseline model) to verify the field power output improvement in response to the upgrade. For example, Figure 8As shown in , the field controller 108 measures multiple operational data from one or more subsets of wind turbines 102 within the wind farm 100. The controller 108 may then generate multiple baseline models for each subset of data. In certain embodiments, the controller 108 is configured to verify each of the baseline models by applying new data to each model (which was not used to create the baseline model) to determine how well each model predicts the total field power output. For example, in one embodiment, the first operational data set may include a first subset and a second subset. Therefore, one of the baseline models may be generated using the first subset and then verified using the second subset. For example, in a wake control embodiment, the baseline model (i.e., when wake control is not in effect) may be executed for new data (when wake control is also not in effect) so that the wind farm power output can be predicted with a small error margin. In an alternative embodiment, the controller 108 may verify the baseline model based only on the training data, so that it is not necessary to verify using new data (e.g., the second subset).

[0051] In another embodiment, the park controller 108 may also eliminate the baseline model that includes one or more wind turbines that have been modified by one or more upgrades. In still other embodiments, the park controller 108 may strategically provide or reject (e.g., turn on and off) certain upgrades or modifications (e.g., wake control) from one or more wind turbines in the wind park in order to provide a basis for evaluating the performance of the park with the desired accuracy. In addition, the degree to which modifications are retained or removed provides the ability to trade off verification accuracy for improvements in park performance.

[0052] Return to reference Figure 5 Once the baseline model has been established and the optimal baseline model has been selected, an AEP uncertainty estimate and static optimal settings are determined in step 220, which results in an AEP uncertainty value 222. The trained or optimal model is applied to the test data, and if no upgrades are applied, the controller 108 can determine a baseline predicted power that represents the power estimate. In addition, the residual power is defined as the difference between the observed or actual power and the baseline predicted power, which represents the power change due to the upgrade and is used to evaluate the AEP change. For example, if Fig. 9As shown in , the actual power 238 is compared with the predicted power 242 determined by the optimal baseline model. In the illustrated embodiment, the optimal baseline model is based on the predicted power 242 as determined by sensors from two wind turbines 102. In addition, line 240 represents an estimate of the total field power when half of the wind turbines 102 are used in the model to predict power. As shown, the predicted power 242 as determined by the optimal model is similar to the actual power 238 as determined by one or more sensors. By inputting the tip speed ratio and pitch settings into the machine learning model and querying the model using all options to find which setting performs best on average, a static optimal setting for turbine performance optimization is determined. These "optimized" settings can then be applied to the desired wind turbine.

[0053] Reference now Fig.10 , and then applies the optimal baseline model to the wind farm 100. More specifically, as shown, the farm controller 108 determines the control sensor used to control a given sample within the turbine set (as shown by reference numeral 252), and then applies the optimal baseline model thereto. Fig.11 As shown in , the farm controller 108 can then model the error 244 or residual between the actual power and the predicted power to determine the performance variation / uncertainty. It should be understood that although any amount of data can be used to build the model, the accuracy of the baseline model generally improves with more data. In some embodiments, the controller 108 can also determine the average error 246 between the actual power and the predicted power to determine the average uncertainty value.

[0054] In another embodiment, Fig.12 and Fig.13 As shown in , the park controller 108 can calculate the AEP change before and after the upgrade by integrating the power change relative to the long-term power profile 304. For example, in one embodiment, the park controller 108 converts the pre-specified long-term wind speed profile 300 based on the reference power curve 302 as shown. In addition, as shown in Fig.12 As shown in , the controller 108 may determine the AEP as a function of the fan zone. More specifically, as shown, line 248 represents the AEP with wake control, while line 250 represents the AEP without wake control. Therefore, the controller 108 may be configured to calculate a baseline model for each possible wind direction. For example, in one embodiment, the controller 108 may determine a model set for fan zones with a median wind direction of 15 degrees. The field controller 108 may then train the model on data with a median wind direction of 15 degrees, and use the model to predict the field output when the wind direction is 15 degrees. In such an embodiment, there are multiple different model sets, each having a field output model using every permutation and combination of sensor inputs. Other wind directions or sectors may also be used, such as Fig.12(i.e., a 20 degree sector), in this case there would be 18 different model sets, and so on.

[0055] Reference now Fig.14 , a flow chart of a method 1400 for validating wind farm performance improvement measurements resulting from one or more upgrades is shown. At 1402, the method 1400 includes measuring operational data from one or more wind turbines of the wind farm via one or more sensors. At 1404, the method includes generating a plurality of baseline models of performance of the wind farm from at least a portion of the operational data, wherein each of the baseline models of performance is developed from a different set of operational data from one or more wind turbines of the wind farm. The generating step 1404 includes filtering the data so that the turbine is in a random state of equilibrium. The control turbine does not need to be in a fixed state, but a method of forcing a random state of equilibrium is introduced so that any turbine in the entire experimental farm design is considered to be part of the control turbine group. At 1406, the method includes selecting an optimal baseline model of performance from the plurality of baseline models via a processor. A machine learning model that must include a direction is used together with a mean of a power set group to determine the benefit. This is an improvement over previous methods that filter to a specific directional sector (forming a separate model for each sector). At 1408, the method includes comparing the optimal baseline model of performance with the actual performance of the wind farm, wherein the actual performance of the wind farm is determined after installing one or more upgrades. Additionally, in step 1408, static optimal settings for turbine performance optimization are determined by inputting tip speed ratio and blade pitch settings into a machine learning model and querying the model using all options to find which setting(s) perform best on average. These improved settings can then be applied to the desired wind turbine.

[0056] As described above, the systems and methods described herein may also be configured to identify one or more performance measurements (e.g., errors) of a wind farm and performance improvements. As described herein, the term "error" is intended to encompass its ordinary meaning as known in the art, as well as misconfigurations, material or software failures, or other problematic situations that may result in subpar performance of a wind farm. In yet other embodiments, errors may occur in situations where a wind farm performs below baseline expectations for a variety of reasons. For example, in one embodiment, information from a field verification method may be used to determine when a wind turbine in a wind farm or the entire wind farm is not properly configured or has suffered an incident.

[0057] This written description uses examples to disclose the invention, including the best mode, and also enables any person skilled in the art to implement 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 a person skilled in the art. If such other examples include structural elements that are not different from the written language of the claims, or if they include equivalent structural elements that are not substantially different from the written language of the claims, such other examples are intended to be within the scope of the claims.

Claims

1. A method for evaluating or verifying wind turbine performance resulting from one or more upgrades, the method comprising: measuring operational data from one or more wind turbines of the wind farm via one or more sensors; generating, via a processor, a plurality of baseline models of performance of wind turbines in the wind farm from at least a portion of the operational data, wherein each of the baseline models of performance is developed from a different portion of the operational data from one or more wind turbines in the wind farm, the generating step comprising filtering the one or more wind turbines such that the one or more wind turbines are in a random state of equilibrium; selecting, via the processor, an optimal baseline model of performance from the plurality of baseline models, the optimal baseline model including direction, wherein a machine learning model including direction is used with a mean of a power aggregate group to determine revenue, wherein a Gaussian process model is used using sine and cosine of field median direction and a NaN mean of power aggregate wind turbine power, wherein the method incorporates direction by inputting the field level direction into the Gaussian process model along with a tip speed ratio and a pitch setting; comparing, via the processor, the optimal baseline model of performance with an actual performance of the wind turbine, wherein the actual performance of the wind turbine is determined after one or more wind turbines of the wind farm are modified by the one or more upgrades that have been installed for the one or more wind turbines, the upgrades comprising at least one of a tip speed ratio, a pitch setting, or a yaw offset setting, and wherein comparing the optimal baseline model of performance with the actual performance of the wind turbine comprises determining a difference between the optimal baseline model of power output and an actual power output of the wind turbine, wherein the difference reflects a change in power output resulting from the one or more upgrades.

2. The method according to claim 1, wherein: The optimal baseline model of power output includes the annual power output of a single wind turbine. 3 . The method of claim 1 , further comprising offering or denying the one or more upgrades to wind turbines of the wind farm based on an expected verified accuracy of the performance improvement measurements of the wind farm.

4. The method according to any one of claims 1 to 3, wherein: The one or more upgrades include any one or combination of the following: rotor blade chord extension, software upgrade, control upgrade, hardware upgrade, wake control, aerodynamic upgrade, blade tip extension, vortex generators, winglets, turbine performance optimization control.

5. The method of any one of claims 1-3, further comprising generating a baseline model of power output for multiple wind directions, or utilizing the sine and cosine of the median wind direction as features for a single model in a machine learning algorithm.

6. The method according to any one of claims 1-3, further comprising developing a long term power profile by converting a pre-specified long term wind speed profile based on a reference power curve.

7. The method of claim 6, further comprising determining a change in annual energy production (AEP) of the wind farm in response to the one or more upgrades by integrating a change in power output resulting from the one or more upgrades relative to the long-term power profile.

8. The method according to any one of claims 1 to 3, wherein: The one or more sensors include any one or combination of the following: an accelerometer, a pressure sensor, an angle of attack sensor, a vibration sensor, a micro inertial measurement unit (MIMU), a camera system, a fiber optic system, an anemometer, a wind vane, a sound wave detection and ranging (SODAR) sensor, an infrared laser, a light detection and ranging (LIDAR) sensor, a radiometer, a pitot tube or a radiosonde anemometer.

9. A method for evaluating or verifying wind turbine performance, the method comprising: measuring operational data from the wind turbine via one or more sensors; generating, via a processor, a plurality of baseline models of performance of the wind turbine from at least a portion of the operational data, wherein each of the baseline models of performance is developed from a different portion of the operational data, the generating step comprising filtering the data such that the data is in a balanced random state; selecting, via the processor, an optimal baseline model of performance from the plurality of baseline models, the optimal baseline model including direction, wherein a machine learning model including direction is used with a mean of a power aggregate group to determine revenue, wherein a Gaussian process model is used using sine and cosine of field median direction and a NaN mean of power aggregate wind turbine power, wherein the method incorporates direction by inputting the field level direction into the Gaussian process model along with a tip speed ratio and a pitch setting; comparing, via the processor, the optimal baseline model of performance with actual performance of the wind turbine, the baseline model comprising at least one of a tip speed ratio, a pitch setting, or a yaw offset setting, and wherein comparing the optimal baseline model of performance with the actual performance of the wind turbine comprises determining a difference between the optimal baseline model of power output and an actual power output of the wind turbine, wherein the difference reflects a change in power output resulting from the absence or presence of one or more upgrades.

10. The method according to claim 9, wherein: The optimal baseline model of power output includes an annual power output of the wind turbine.

11. The method according to claim 9, wherein: The one or more upgrades include any one or combination of the following: rotor blade chord extension, software upgrade, control upgrade, hardware upgrade, wake control, aerodynamic upgrade, blade tip extension, vortex generators, winglets, turbine performance optimization control.

12. The method of any one of claims 9-11, further comprising generating a baseline model of power output for multiple wind directions, or utilizing the sine and cosine of the median wind direction as features for a single model in a machine learning algorithm.

13. The method according to any one of claims 9-11, further comprising developing a long term power profile by converting a pre-specified long term wind speed profile based on a reference power curve.

14. The method according to any one of claims 9 to 11, wherein: The one or more sensors include any one or combination of the following: an accelerometer, a pressure sensor, an angle of attack sensor, a vibration sensor, a micro inertial measurement unit (MIMU), a camera system, a fiber optic system, an anemometer, a wind vane, a sound wave detection and ranging (SODAR) sensor, an infrared laser, a light detection and ranging (LIDAR) sensor, a radiometer, a pitot tube or a radiosonde anemometer.

15. A system for evaluating or verifying wind turbine performance, the system comprising: a plurality of sensors configured to measure operational data from one or more wind turbines in a wind farm; as well as A controller configured to perform the method according to any one of claims 1-14.

16. The system of claim 15, further comprising evaluating performance of an individual wind turbine or group of wind turbines in the absence of an upgrade to determine underperformance.

Images