WO2018121668A1

WO2018121668A1 - A method and system for evaluating wind turbine generator performance

Info

Publication number: WO2018121668A1
Application number: PCT/CN2017/119358
Authority: WO
Inventors: Matthew SUMMERS
Original assignee: Envision Energy Jiangsu Co Ltd
Current assignee: Envision Energy Jiangsu Co Ltd
Priority date: 2016-12-30
Filing date: 2017-12-28
Publication date: 2018-07-05
Anticipated expiration: 2019-06-30
Also published as: CN110168220B; CN110168220A

Abstract

Disclosed is a method of evaluating (100) performance of a specific wind turbine generator (WTG) (11) amongst a set of wind turbine generators (WTGs) (23). The method may comprise the following acts: measuring (110) actual power production (31) of the specific WTG; receiving (120) a set of data (40, 41) comprising control settings (50) from the specific WTG and at least one set of data (40, 42) comprising control settings from at least one of the other WTGs (12) in the set of WTGs; estimating (130) estimated power production (33) of the specific WTG as a function of the received set of data from the specific WTGs and the received at least one set of data from at least one of the other WTGs; comparing (140) the actual power production to the estimated power production. The act of estimating estimated power production is based on a computer implemented evaluation (200). Also disclosed is a wind turbine evaluating system.

Description

A Method and System for Evaluating Wind Turbine Generator Performance

Field of the Invention

The present invention relates to a method of evaluating performance of a specific wind turbine generator (WTG) amongst a set of wind turbine generators (WTGs) . The method may comprise one or more of the following acts. There may be an act of measuring actual power production of the specific WTG. There may be an act of receiving a set of data from the specific WTG and at least one set of data from at least one of the other WTGs in the set of WTGs. There may be an act of estimating power production of the specific WTG as a function of the received set of data from the specific WTGs and the received at least one set of data from at least one of the other WTGs. There may be an act of comparing the actual power production to the estimated power production. The act of estimating power production may be based on a computer implemented evaluation.

The present invention also relates to a system for evaluating performance of a WTG amongst a set of WTGs.

Background of the Invention

Traditional methods to predict power performance of wind turbine generators (WTGs) rely on a “power curve” describing the expected power generation as a function of wind speed measured by an anemometer mounted on a nacelle or a met mast, along with other atmospheric properties like density and possibly even humidity and temperature among others. The actual power is then compared to values looked up from the power curve to evaluate the WTG performance. Such methods are subject to a variety of problems associated with anemometer measurements.

Improvement has been made by comparing the actual power production of pairs of WTGs. The general idea is to make a performance metric from the pair of powers, typically a ratio or difference. Pairs may be pre-selected to account for spatial differences to some extent.

There are a few issues with this approach. First, when one of the WTGs is waked, the metric will be skewed. Second, in complex terrains, the metric is very wind direction and shear dependent, and the standard deviation of such metrics (e.g. a power ratio) is typically large.

US 2011/0270450 describes a method and system for evaluation of a wind turbine performance for a specific wind turbine in a wind farm. The measurement of actual power production (wind turbine performance) of a specific wind turbine and similar wind turbines within the pair is performed. However, the suggested pairing and handling of paired wind turbines have shown to be problematic.

In general, there is a need for alternative methods and methods that are more robust. Specifically issues with skewness are to be mitigated or overcome.

Patent application US20160084233 discloses the application of segmented regression to model the power produced by an entire farm, using data from a subset of WTGs as input variables.

Object of the Invention

It is an objective of the present disclosure to overcome or improve one or more of the above-cited problems or shortcomings.

It is an objective to provide a simple and robust way of assessing power performance amongst wind turbines. Even without direct access to the control or operation of all the wind turbines.

Description of the Invention

An objective is achieved by continuously evaluating performance of a specific wind turbine generator (WTG) amongst a set of wind turbine generators (WTGs) . The method may comprise one or more of the following acts.

There may be an act of measuring actual power production of the specific WTG.

There may be an act of receiving a set of data from the specific WTG and at least one set of data from at least one of the other WTGs in the set of WTGs.

There may be an act of estimating power production of the specific WTG as a function of the received set of data from the specific WTGs and the received at least one set of data from at least one of the other WTGs.

There may be an act of comparing the actual power production to the estimated power production.

The act of estimating power production may be based on a computer implemented evaluation.

Such method allows WTGs to be continuously compared relatively to each other and thereby giving the operator or manufacturer valuable information about the performance of one or more wind turbines by using data empirically available. The method gives an easy and reliable indication of the performance of a specific WTG.

In an aspect of the invention, the computer implemented evaluation may be implemented as computer instructions representing one or more of the following types of models.

There may be an implementation of a regression model. There may be an implementation of a support vector regression model. There may be an implementation of machine learning such as a neural network. Machine learning may be under a category of models known as ” supervised learning” .

Pre-processing, normalisation and general preparation of data may be required and depend on the actual model implementation.

A person skilled in the art will appreciate a variety of models chosen amongst the mentioned types of models. A person skilled in the art will also appreciate varieties of naming of models. At a high level, a model may be found amongst models known as supervised learning or regression. Implementing such models will give the power for the specific WTG. A person skilled in the art and familiar with such classes of algorithms will readily know where to find literature or software libraries to implement the methods.

One starting point of supervised learning method may be the so-called “Random Forest Regression” [Ho, Tin Kam (1998) . "The Random Subspace Method for Constructing Decision Forests" .

Another starting point may be “IEEE Transactions on Pattern Analysis and Machine Intelligence. 20 (8) : 832–844] ” .

A starting point for “k-Nearest Neighbors” may be Altman, N.S. (1992) . "An introduction to kernel and nearest-neighbor nonparametric regression" . The American Statistician. 46 (3) : 175–185.

A starting point for “Support Vector Regression” may be Drucker, Harris; Burges, Christopher J.C.; Kaufman, Linda; Smola, Alexander J.; and Vapnik, Vladimir N. (1997) ; "Support Vector Regression Machines" , in Advances in Neural Information Processing Systems 9, NIPS 1996, 155–161, MIT Press.

Implementing such models will require some adjustments. For example, when implementing a decision tree model. For Random Forest regression there may be used 20 trees with a minimum of 2 samples per leaf node and a maximum tree depth of 20. Depending on the actual data 1-50 trees may be used.

For k-nearest neighbors a distance metric is required and it will be a matter of choice for the person skilled in the art, but for the purposes here a weighted Euclidean distance metric may be used.

The model inputs are the data from the specific WTG and other WTGs, excluding the specific WTG’s power. The model output is the specific WTG’s power. The models are trained with historical data including both the inputs and outputs and then, after training, make predictions of the output using the input.

In an aspect of the invention, there may further be an act of triggering an alarm if the actual power production is below the estimated power production.

An alarm may be based on a more generalized criteria, e.g. an alarm may be activated if the average power for the last N-hours is a certain percentage less than the average predicted power for the last N hours. Alternatively, an alarm may be triggered using confidence intervals.

In an aspect of the invention, there may further be an act of identifying the reason why the actual power production is below the estimated power production.

The act of identifying may be based on characteristics of how the underperformance in power production happens in time. Slow gradual reduction in performance can be caused by blade contamination.

If the underperformance happens on a faster time scale under a condition with temperature below zero, the reason could be icing. If an abrupt decrease in actual power production is observed, the reason could be a software (controller) update, blade change or breakage.

This can be used in conjunction with other algorithms like the pitch and torque schedule change detection and may be used to evaluate e.g. the annual energy production impact of other changes.

In an aspect of the invention, the received data from the specific WTG includes at least the actual pitch setting of the specific WTG.

The selected model may then use pitch settings of the specific WTG as a parameter and produce an estimated power output based on the pitch settings of other WTGs with actual power outputs. Hence, the method allows for the actual power production to be compared with an estimated power production based on the power production of other WTGs by fitting, modelling or predicting.

In an implementation, the model input data includes pitch from other WTGs, but not the specific WTG.

The selected model may thus optionally include the pitch setting of the specific WTG and/or other WTGs.

Other settings of the specific WTG may be included in the received data. For example, torque settings. In this case, the selected model may then use torque settings of other WTGs.

In an implementation, the model input data includes torque from other WTGs, but not the specific WTG.

The selected model may thus optionally include the torque setting of the specific WTG and/or other WTGs.

In an aspect of the invention, received data from the other WTGs include at least pitch settings and corresponding actual power production of one or more of the other WTGs.

In an aspect of the invention, the act of receiving data may further comprise receiving a set of meteorological conditions pertinent to each WTG and wherein the act of estimating furthermore is a function of the set of meteorological conditions.

The collected parameters or data from the various WTGs and other devices, such as e.g. meteorological station providing meteorological conditions or data in the farm (excluding the actual WTG power) , may be delivered as independent variables. The values of these parameters at each point in time may be used in the function or modelling such as used to create the regression function, which targets the power on the specific WTG.

This has shown to be much more reliable than traditional methods using power ratios and power curves; since it has been observed that the method relying on power curves is not useful for assessing most types of changes. Power ratios may be good in some cases, but are sensitive to wind shear, wind direction, curtailment of compared WTG. Using meteorological data or conditions in the methods outlined here overcomes most of those problems.

In an aspect of the invention, there may further be an act of updating or adjusting the computer implemented evaluation after performing the act of comparing.

The evaluation or model may be adjusted or refined. Weights or free parameters may be adjusted according to the output or other factual information available that will increase the reliability of the evaluation.

For instance a regression model may show that one or more particular WTGs are outside a certain statistical threshold and the particular WTGs are given less weight or eliminated in following evaluation or modelling.

In another instance, more models or evaluations are used, and it turns out that amongst those specific models or evaluations some are more precise than others. The use of models may then be adjusted to rely more on those more precise models. One model may show indications of being “under-fitted” and then the number of free parameters, layers or complexity may be increased. Likewise, a model may show indications of being “over-fitted” and then the number of free parameters, layers or complexity may be decreased. In an aspect of the invention, the set of wind turbine generators (WTGs) comprises at least one reference WTG.

One particular WTG amongst the other WTGs may be maintained or empirically observed to be performing more reliable than the remaining WTGs. Such particular WTG may be considered as a reference WTG and used with a higher weight than other WTGs.

A model or algorithm, like the random forest, may also automatically learn if the power output is more related with some WTGs than with others. E. g. trying to predict the power of a WTG-A having two neighboring WTGs –WTG-B and WTG-C; and if WTG-B’s power perfectly matches with WTG-A’s power while WTG-C’s power is completely unrelated, then the algorithm will automatically learn this and use WTG-B’s power while completely or mostly ignoring WTG-C’s power.

In an aspect of the invention, evaluating performance may be performed with more or each one of the WTGs in a set of wind turbine generators (WTGs) as the specific wind turbine generator (WTG) .

Such acts may be used to calibrate a model or evaluation. Furthermore, such acts may be used to classify or order the WTGs relatively to each other.

In a particular instance where a subset of, e.g. a pair of, WTGs are more related to each other, then this subset, or the pair, may be identified and form the basis for a better and more precise evaluation. Likewise, the acts may be used to reduce the complexity by identifying the most suitable set of other WTGs and then reduce required data or computational efforts.

In an aspect of the invention, the specific WTG comprises one or more additional WTG features such as vortex generators, Gurney devices, and serrations, when compared to the other wind turbine generators (WTGs) .

This allows for performance evaluation and operational effects of WTG additions to be assessed.

In an aspect of the invention, the specific wind turbine (WTG) has one or more control settings that vary compared to the other wind turbine generators (WTGs) .

This allows for performance evaluation and operational effects of changes in operational or control settings for a WTG to be assessed.

An object may be achieved by a computer program product comprising instruction to perform one or more actions of evaluating performance according to outlined methods or actions.

An objective may be achieved by a wind turbine evaluating system comprising one or more of the following features.

There may be a power production measuring system configured to measure actual power production of a specific WTG.

There may be a data receiving system configured to receive a set of data from the specific WTG and a at least one set of data from at least one of the other WTGs in the set of WTGs.

There may be a computer configured to estimate power production of the specific WTG, as a function of the received set of data from the specific WTGs and the received at least one set of data from at least one of the other WTGs, and to compare the actual power production with the estimated power production.

There may be a computer program product comprising instruction to cause the computer to execute the methods or actions disclosed.

Definitions:

By “continuously” is understood more than one and preferably periodically or regularly. In principle, there may be a first (former) evaluation and later on a second (later) evaluation. In practice, a person skilled in the art will know how to adjust the regularity, but an evaluation may be performed periodically with 10 minutes intervals.

By “data” is understood parameters or values relating to the operation of the WTG.

By “set” is understood a collection of items or objects. The set may be a non-empty set. A “set” may be a group of WTGs that share some grouping features or similarities, but a “set” may also be one or more WTGs collected to form a set of other WTGs to be used for the analysis. The set may change over time or the WTGs in the “set of WTGs” may change or be changed.

Hence, a set of wind turbines may be a wind turbine farm located in the same area. A set of wind turbines may also be similar models of wind turbines located at different locations. A set of wind turbines may also be wind turbines of different models.

Herein, a set of “WTG” include at least one specific WTG i.e. a WTG that is to be compared or evaluated against the other WTGs. Hence, a set may comprise of WTG “A, B, C” . “A” may be the specific WTG and “B and C” the other “WTGs” . “B” may also be the specific WTG and “A and C” the other “WTGs” . Likewise, a set may be “A, B” and “A” , or “B” may either be the specific WTG that is evaluated against the other WTG, i.e. “B” or “A” or vice versa.

By “model” is understood a principle expressed as either a formula, a set of actions or steps performed in one or more algorithms. The model may also be considered a neural network.

Examples

The methods, actions and system disclosed may be used as a method for evaluating the performance of a wind turbine generator as generally illustrated in figure 1. The method may employ a regression model to calculate the power output of the wind turbine generator based on some parameters of the specific, or target, wind turbine generator, and one or more other wind turbine generators located relative to the target wind turbine generator. The specific wind turbine and the other wind turbines may be as generally illustrated in figure 2.

The regression model may include any of the following models: machine learning or statistical methods including neural networks, decision trees, support vector regression, nearest neighbors, etc.

The parameters include power output from the other wind turbine generators, blade pitch from the other wind turbine generators, torque from the other wind turbine generators, air properties, wind direction, temperature, wind vane, time of day, precipitation, or density measurements, or wind turbine generators status codes, etc. The data and evaluation may be as illustrated in figure 2.

The data collected may be processed, evaluated and compared as generally illustrated in figure 3, where calculated power of the target wind turbine generator is compared to the actual (measured) power of the target wind turbine generator to evaluate the performance of the target wind turbine generator. Performance metrics is calculated on the basis of the comparison of the calculated power and the actual power in form, a difference between the powers, or a ratio of the powers. An underperforming wind turbine can hence be identified on the basis of the comparison. Further, the method can be used to detect any errors on the wind turbine, including the yaw error or wind turbine blade cracking or breaking, etc. Also, this method can be used to evaluate the performance impact of any upgrades or add-ons added to the wind turbine generator or any change in the control settings of the wind turbine generator.

The following pointers have been identified based on the elements of the invention. All the relevant and/or related results identified in the search are mapped on these key features to properly depict their significance.

A method for evaluating performance of a wind turbine generator may comprise evaluating the performance of a wind turbine generator using a regression model wherein a regression model fits the wind turbine generator power as a function of the one or more other parameters measured in the wind farm.

The method may be comparing the measured wind turbine generator power (s) to (a) value (s) computed from parameters measured on this wind turbine generator and one or more other wind turbine generator located relative to this wind turbine generator.

The data may include power output from the other wind turbine generators, blade pitch from all wind turbine generators, or torque output from other wind turbine generators.

The data may include air properties, wind direction, temperature, wind vane, time of day, precipitation, or density measurements.

There may be an indicator of wind turbine generator performance which includes the actual power relative to the regressed power expressed in the form of performance metrics which includes:

a. the ratio of the actual power to the regressed value, or

b. difference between the actual power and the regressed value,

c. wind turbine generator status codes may be used as the parameters.

Periodic aggregates (e.g. daily or hourly means or medians, potentially with confidence intervals or standard deviations) may be used to provide additional statistical certainty of the conclusions.

For some controller types, a rotor RPM may be substituted for power to reduce statistical uncertainty.

This method may be applied in a software product to generate alarms when performance is too low.

This method may be used to evaluate the performance impact of wind turbine generator add-ons like vortex generators, Gurney devices, and serrations.

The method may be used to detect yaw error or suboptimal control settings if the associated things are varied while monitoring performance i.e. different operating wind vane, pitch, or torque schedule.

This method may be applied to determine when a blades geometry has been unintentionally compromised (e.g. blade soiling, leading edge erosion, cracked/broken, etc. ) .

The disclosed actions or systems describe an improved methodology for evaluating the performance of a WTG. In an example, the method or principles are based on comparing the measured WTG power (s) to a value/values computed from parameters measured on this WTG and one or more other WTGs. The other WTG may be located relative to this WTG.

The data or parameters typically include power output from the other WTGs, blade pitch from all the WTGs, and air properties such as wind direction, temperature, TI, among others.

Other data or parameters may include, but not limited to torque, wind vane, time of day, precipitation, or density measurements.

Additionally, data may include WTG status codes.

Some of these parameters like blade pitch and torque can be included from several WTGs.

In an example, the method includes a regression model to fit WTG power as a function of the one or more other parameters measured in the wind farm. The actual power relative to the regressed power is an indicator of WTG performance. Performance metrics may include e.g. the ratio of the actual power to the regressed value or the difference between the actual power and the regressed value.

Periodic aggregates, e.g. daily or hourly means or medians, potentially with confidence intervals or standard deviations, may be used to provide additional statistical certainty of the conclusions.

This method may be used to evaluate the performance impact of WTG add-ons like vortex generators, Gurney devices, serrations among others.

The method may be used to detect yaw error or suboptimal control settings if the associated things are varied while monitoring performance (i.e. different operating wind vane, pitch, or torque schedule) .

The outlined methods or actions may be applied to determine when a blade’s geometry has been unintentionally compromised (e.g. blade soiling, leading edge erosion, cracked/broken, etc. )

The method may be applied to real data and to detect when a software update changed the WTG torque schedules, significantly affecting the production.

The method may also be applied to detect yaw error on several WTGs. It may be applied to detect when a blade cracks or breaks;. in general, anything that would cause a WTG to produce less power.

The method may also be applied to study the effect of changes. Such changes may include, but are not limited to, control changes of pitch, torque, or yaw or even blade geometry changes such as VGs, gurney devices, tip extensions, boundary layer fences, or other geometry changes.

The collected data or parameters from the various WTGs and other devices (e.g. met mast) in the farm (excluding the target WTG power) are the independent variables. The values of these parameters at each point in time are used to create the regression function which targets the power on the target WTG.

The regression may be performed using any machine learning or statistical methods including neural networks, decision trees, or nearest neighbors. Such regression model may optionally be preloaded with historical data. Regression model may or may not update (i.e. continuous learning) as new data is received. Regression may update periodically, e.g. daily or monthly. Regression model may update as a function of user input, e.g. in some conditions a user may command the module to stop updating for a period of time to prevent learning known bad data.

The choice of which independent variables to use depends on the application. One system may have several regression functions, e.g. one for detecting blade geometry problems, one for detecting torque problems, and one for detecting yaw error problems. E. g. if torque investigation is a goal, then the torque on the target machine may not be used. In this case, the performance as a function of torque can subsequently be investigated to detect potential issues.

Description of the Drawing

Embodiments of the invention will be described in the figures, whereon:

Fig. 1 illustrates a general wind turbine generator (WTG) ;

Fig. 2 illustrates an exemplary arrangement of WTGs with a specific WTG and other WTGs;

Fig. 3 illustrates a result of comparing actual performance with estimated performance over time;

Fig. 4 illustrates a set of WTGs having a specific WTG and other WTGs;

Fig. 5 illustrates a method of evaluating WTG performance;

Fig. 6 illustrates a computer implemented evaluation based on one or more models; Fig. 7 illustrates further acts when evaluating WTG performance;

Fig. 8 illustrates further inclusion of metrological data into evaluating WTG performance; and

Fig. 9 illustrates an act of updating a model.

Detailed Description of the Invention

Item	No
Wind turbine generator, WTG	1
Tower	2
Nacelle	3
Rotor	4
Blade	5
Controller	6
Wind turbine generator (WTG)	10
Specific WTG	11
Other WTG	12
Set of WTGs	23
Reference WTG	25
Power production	30
Actual power production	31
Estimated power production	33
Data/set of data	40
Data specific WTG	41
Data other WTG	42
Set of data	43
Control setting	50
Pitch setting	52
Meteorological conditions	70
Method of evaluating	100
Measuring	110
Receiving	120
Estimating	130
Comparing	140
Identifying	150
Triggering an alarm	160
Computer implemented evaluation	200

Regression model	210
Support vector regression model	215
Machine learning Model	220
Neural Network model	225
Decision Tree model	230
Nearest Neighbor’s model	240
Updating computer implemented evaluation	400
Evaluating system	1000
Power production measuring system	1100
Data receiving system	1200
Computer	2000
Computer program product	2200

Figure 1 illustrates a generic wind turbine generator, a WTG, 10 having a tower 2 supporting a nacelle 3 with a rotor 4 having blades 5. The WTG 10 represent a typical wind turbine generator referred to in the following.

The WTG 10 may have a controller 6 for controlling the WTG 10. The controller 6 will generally be configured to communicate externally. Various operational data will generally be available based on monitoring systems. There may be a Supervisory Control and Data Acquisition (SCADA) implementation.

Figure 2 illustrates a wind turbine evaluating system 1000 comprising a power production measuring system 1100 configured to measure 110 actual power production of a specific WTG 11 and other WTGs 12.

There is a data receiving system 1200 configured to receive a set of data 40, 41 from the specific WTG 11 and a at least one set of data 40, 42 from at least one of the other WTGs 12 in the set of WTGs 23. The data may comprise control settings 50 such as pitch settings 52. There is a computer 2000 (with a computer program product 2200) configured to estimate 130 power production of the specific WTG 11 as a computer implemented evaluation 200 function or model of the received 120 set of data 40, 41 from the specific WTG 11 and the received at least one set of data 40, 42 from at least one of the other WTGs 12 and for comparing 140 the actual power production 31 to the estimated power production 33.

If the actual power production 31 of the specific WTG 11 is “normal” or within a level of the estimated power production 33, then no action is required. If the actual power production is not normal, or e.g. below the estimated power production 33, then further action may be performed. There may be an act of identifying 150 the reason or/and there may be an act of triggering an alarm 160, which may also be an act of informing the operator.

Figure 3 illustrates an example of evaluating a set of WTGs as outlined in figure 2 and the result estimating 130 and comparing 140 the actual power production 31 to the estimated power production 33 over time. The dots represent actual power production 31 of a specific WTG 11 compared to the estimated 130 WTG power obtained by a model such as a regression model. As indicated, there is a period in which the actual power production of the specific WTG 11 is underperforming with respect to the estimated power production 33. Underperforming may be defined as when the act of comparing 140 results in a measure where the actual power 31 is statistical or significantly below a certain value. In the case (s) of underperforming, there may be several further actions. An alarm may be triggered. There may be an investigation or evaluation to identify a potential problem. There may be produced a notification, which may include a list of potential problems, and the produced notification may be sent to the operator or manufacturer.

Figure 4 illustrates a specific wind turbine generator WTG 11 amongst a set of wind turbine generators WTGs 23. The set of WTGs 23 comprises the specific WTG 11 and other WTGs 12.

The selection or choice of specific WTG may be changed or cycled. In any combination, the disclosed evaluation of performance may be performed with more WTGs or with each one of the WTGs in the set of wind turbine generators WTGs as the specific wind turbine generator WTG 11.

Figure 5 illustrates, with reference to the previous figures, a method of continuously evaluating 100 performance of a specific wind turbine generator WTG 11 amongst a set of wind turbine generators WTGs 23. The method comprises an act of measuring 110 actual power production 31 of the specific WTG 11.

There is an act of receiving 120 a set of data 40, 41 from the specific WTG 11 and a at least one set of data 40, 42 from at least one of the other WTGs 12 in the set of WTGs 23.

There is an act of estimating 130 estimated power production 33 of the specific WTG 11 as a function of the received set of data 40, 41 from the specific WTGs 11 and the received at least one set of

data

40, 12 from at least one of the other WTGs 12.

There is an act of comparing 140 the actual power production 31 to the estimated power production 33.

The act of estimating 130 the estimated power production 33 is based on a computer implemented evaluation 200.

With reference to figure 2 and 4, the received data 40 from the specific WTG 11 includes at least actual pitch settings 51 of the specific WTG 11. As an example, the received data 40 from the other WTGs 12 include at least actual pitch settings 51 and corresponding actual power production 31 of one or more of the other WTGs 12.

The outlined methods may also be performed where the specific WTG 11 comprises one or more additional WTG features, such as vortex generators, Gurney devices, and serrations, as compared to the other wind turbine generators WTGs 12. Also, the specific wind turbine WTG 11 may have one or more control settings 50 varied or altered as compared to control settings 50 of the other wind turbine generators WTGs 12.

Figure 6 illustrates that the computer implemented evaluation 200 is implemented as computer instructions representing one or more of a regression model 210, a support vector regression model 215, a machine learning 220 such as a neural network 225, a decision tree model 230, and/or a nearest neighbor’s model 240.

Figure 7 illustrates the method outlined in figure 5 and further comprises an act of identifying 150 the reason why the actual power production 31 is below the estimated power production 33.

Further illustrated is an act of triggering an alarm 160 if the actual power production 31 is below the estimated power production 33.

Figure 8 illustrates, with reference to figure 2, the method wherein the act of receiving 120 further comprises receiving 122 a set of meteorological conditions 60 pertinent to each WTG 10 and wherein the act of estimating 130 furthermore is a function of the set of meteorological conditions 60.

Figure 9 illustrates an act of updating or adjusting 400 the computer implemented evaluation 200 after performing the act of comparing 140.

Claims

A method of continuously evaluating (100) performance of a specific wind turbine generator (WTG) (11) amongst a set of wind turbine generators (WTGs) (23) , the method comprising acts of:

- measuring (110) actual power production (31) of the specific WTG (11) ;

- receiving (120) a set of data (40, 41) comprising control settings (50) from the specific WTG (11) and a at least one set of data (40, 42) comprising control settings (50) from at least one of the other WTGs (12) in the set of WTGs (23) ;

- estimating (130) estimated power production (33) of the specific WTG (11) as a function of the received set of data (40, 41) from the specific WTGs (11) and the received least one set of data (40, 12) from at least one of the other WTGs (12) ;

- comparing (140) the actual power production (31) to the estimated power production (33) ; and

- wherein the act of estimating (130) estimated power production (33) is based on a computer implemented evaluation (200) .
The method according to claim 1, wherein the computer implemented evaluation (100) is implemented as computer instructions representing:

- a regression model (210) ;

- a support vector regression model (215) ;

- a machine learning (220) such as a neural network (225) ;

- a decision tree model (230) ; and/or

- a nearest neighbor’s model (240) .
The method of claim 1 or 2 further comprising an act of triggering an alarm if the actual power production (31) is below the estimated power production (33) .
The method of any one or more of claim 1 to 3, further comprising an act of identifying (150) the reason why the actual power production (31) is below the estimated power production (33) .
The method of any one or more of claim 1 to 4, wherein the received data (40) from the specific WTG (11) includes at least actual pitch settings (51) of the specific WTG (11) .
The method of any one or more of claim 1 to 5, wherein the received data (40) from the other WTGs (12) include at least actual pitch settings (51) and corresponding actual power production (31) of one or more of the other WTGs (12) .
The method of any one or more of claim 1 to 6, wherein the act of receiving (120) further comprises receiving (122) a set of meteorological conditions (60) pertinent to each WTG (10) and wherein the act of estimating (130) furthermore is a function of the set of meteorological conditions (60) .
The method of any one or more of claim 1 to 7, further comprising an act of updating or adjusting (400) the computer implemented evaluation (200) after performing the act of comparing (140) .
The method of any one or more of claim 1 to 8, wherein the set of wind turbine generators (WTGs) (23) comprises at least one reference WTG (25) .
The method of any one or more of claim 1 to 9, wherein evaluating performance is performed with more or each one of the WTGs in set of wind turbine generators (WTGs) as the specific wind turbine generator (WTG) (11) .
The method of any one or more of claim 1 to 10, wherein the specific WTG (11) comprises one or more additional WTG features such as vortex generators, Gurney devices, and serrations, as compared to the other wind turbine generators (WTGs) (12) .
The method of any one or more of claim 1 to 11, wherein the specific wind turbine (WTG) (11) has one or more control settings (50) that vary or is varied as compared to control settings (50) of the other wind turbine generators (WTGs) (12) .
A computer program product (200) comprising instruction to perform one or more actions of evaluating performance according to the methods of one or more of claim 1 to 12.
A wind turbine evaluating system (1000) comprising:

- a power production measuring system (1100) configured to measure actual power production of a specific WTG (11) ;

- a data receiving system (1200) configured to receive a set of data (40, 41) comprising control settings (50) from the specific WTG (11) and a at least one set of data (40, 42) comprising control settings (50) from at least one of the other WTGs (12) in the set of WTGs (23) ;

- a computer (2000) configured to estimate power production (30) of the specific WTG (11) as a function of the received set of data (40, 41) from the specific WTG (11) and the received at least one set of data (40, 42) from at least one of the other WTGs (12) and for comparing (140) the actual power production (31) to the estimated power production (33) .
A computer program product (2200) comprising instruction to cause the computer (2000) of claim 14 to execute the methods or actions of one or more of claim 1 to 13.