Systems and methods for reducing risk of avian collision with wind turbines

Abstract

A system and a method for reducing the risk of collision between flying avians (3) and rotor blades (2) of a wind turbine (1) by combining data on target tracks (7) obtained by at least one avian radar (5) and environmental risk factors (9) affecting avian behavior in an avian detection area (4) to determine a risk level (11) of a possible collision, and using this risk level (11) and technological parameters (13) of the wind turbine (1) for determining a control value (12) defining a recommended change of blade velocity; or to trigger an avian dispersal system (42) that disperses avians (3) depending on whether the determined risk level (11) exceeds an avian dispersal threshold.

Description

[0001] SYSTEMS AND METHODS FOR REDUCING RISK OF AVIAN COLLISION WITH WIND TURBINES

[0002] TECHNICAL FIELD

[0003] The present invention relates to a system and a method for reducing risks of collision between flying avians and objects that may pose a collision risk with the flying avians, and in particular for reducing risks of collision between flying avians and wind turbine blades. The invention further relates to a system and a method for controlling wind turbines to reduce the risk of such collisions.

[0004] BACKGROUND

[0005] Wind turbines represent a hazard to flying avians such as migratory birds, raptors and other avians of interest. A bird hit by a wind turbine rotor blade may be killed, and the collision may also damage the rotor blade, which can result in stopping of the turbine and costly repairs of the blade. Wind farms can further impact birds through direct habitat loss due to wind farm construction; and displacement through indirect loss of habitat due to birds avoiding a new wind farm area.

[0006] There exist a number of solutions aimed for preventing avians from hitting wind turbine blades. US 8,742,977 B1 describes a solution that detects birds in the vicinity of wind turbines and engages a deterrent, like intense lights or sounds, to scare the birds away. Similar patents on detecting and repelling birds exist in the field of aviation. Some of these systems employ bird radars, such as disclosed in US 8456349, that have a static 3D coverage zone. Within that zone specific birds (e.g. large birds and flocks) generate an audio or visual alarm. This alarm can be used to e.g. scare away birds or to shut down wind turbines. The alarms are based on size and location of the bird only.

[0007] In practice this leads to situations where there may be too many alarms, too many false alarms or alarms that come too late. In addition, employed frequently and on a broad scale, such deterrents could have negative ecological impacts, driving away not only birds, but also other non-targeted animals living in the vicinity of wind turbines. There is also the danger of desensitization, where over repeated exposures the birds become accustomated to the deterrents, thereby reducing the efficiency of such solutions.

[0008] WO 2010 / 076500 A1 describes a different prior art approach where flying objects in the vicinity of a single wind turbine are detected using one or more radar. Safety zones are defined based on the area swept by the rotor blades and, if an object is detected within the safety zones, the wind turbine is automatically slowed or stopped. When the object leaves the safety zones, the wind turbine is returned to operation. DE10 2005 046 860.8 describes a similar method where a region around a wind turbine is monitored for birds or bats and, if a threshold number are detected, the wind turbine rotor is automatically braked or stopped, to reduce the danger of collision.

[0009] These existing methods involve automatically changing the operating state of the wind turbine from a normal operating state to one in which the rotor speed is reduced or completely stopped, in order to reduce the danger of collision with detected flying objects. Apart from the fact that slowing down and stopping wind turbines results in curtailment of power production, due to their design wind turbines cannot just be stopped or started instantaneously. In fact, stopping the blades results in wear and tear that can significantly reduce the lifetime of wind turbines over time. Therefore, shutdown events should not happen automatically and intermittently, and without taking into account the types and technological limitations of the wind turbines themselves.

[0010] Furthermore, while reducing bird mortality is an important purpose, it often has to be weighed against economic and legal factors, such as wind farms with limited budgets for shutdown events, and different types of protective measures applying for different avian species.

[0011] In addition, weather, time-of-day and sun elevation, as well as seasonal patterns related to migration or foraging can have a big impact on avian behaviour and therefore risk of avian mortality in the vicinity of wind turbines.

[0012] Because of the increasing volume wind turbine installations, and increasingly strict regulations on protection of avians of interest, there is thus a growing need for a reliable, cost-effective and robust system for reducing the risk of collision with certain avian species, while also taking into account the environmental, legal, and economic context of wind turbines.

[0013] SUMMARY

[0014] On the above background it is an object of the present invention to provide a method and system that reduces the risk of collision between flying avians and wind turbine rotor blades while also overcoming some or all of the disadvantages of the prior art solutions.

[0015] The above and other objects are obtained by the provision of a system that combines target track data obtained by radar devices configured for detection of avian objects with environmental risk factors affecting flying avian behaviour and technological parameters of any wind turbine in question.

[0016] The combined data is used as input for a risk model configured for determining risk levels of possible collisions between a wind turbine and avian objects. The determined risk level can be used to determine a control value for a recommended change of blade velocity for one or more wind turbine, taking into account the corresponding technological parameters; or to trigger an avian dispersal system that disperses avian objects depending on the determined risk level and other parameters.

[0017] According to a first aspect there is provided a method for controlling at least one wind turbine to reduce risk of collision between at least one flying avian and at least one wind turbine rotor blade. The method comprising the steps of obtaining a number of radar plots corresponding to at least one avian detected in an avian detection area by an avian detection system comprising at least one avian radar; creating and storing a number of target tracks based on the obtained radar plots, each target track comprising target data determined from radar plots of corresponding avians; obtaining a number of environmental risk factors affecting flying avian behavior in the avian detection area; determining a risk level with respect to a possible collision between the at least one wind turbine and at least one avian using a risk model based on target data of the number of target tracks and the number of environmental risk factors; and determining a control value for at least one wind turbine based on the risk level and technological parameters of the wind turbine, wherein the control value defines a recommended change of blade velocity for the at least one wind turbine.

[0018] In an embodiment each target track is determined based on at least one obtained radar plot, and further based on sensor data on the corresponding avian detected in the avian detection area, wherein the sensor data is received from an additional sensor, such as a second radar, a camera, or a transponder.

[0019] In a possible implementation form of the first aspect, each target track is based on a match between corresponding target data of at least two, three or four radar plots. In a possible implementation form of the first aspect, the match between corresponding target data of radar plots comprises a match between range data. In a possible implementation form of the first aspect, the match between corresponding target data of radar plots comprises a match between radial velocity data. In a possible implementation form of the first aspect, the match between corresponding target data of radar plots comprises a match between return energy data.

[0020] Examples of matching conditions are known in the art of radar scanning and track generation, but a match condition may have to be fulfilled for a set of target data, which are part of the radar plots being compared, when a new target track is generated based on a number radar plots, while the data of a new radar plot has to match corresponding data of a stored target track, when a new radar plot is used for updating an existing target track. A match condition may be fulfilled for corresponding radar plot or target track, when the difference between the target data being matched is below a predetermined threshold difference.

[0021] In an embodiment each target track is created based on a plurality of obtained radar plots, preferably at least 4 consecutive radar plots obtained along a path corresponding to an avian flight path.

[0022] In an embodiment the environmental risk factors comprise information about at least one of: time of year, time of day, weather, visibility, and sun elevation in the avian detection area. In an embodiment the technological parameters of the wind turbine comprise information about at least one of: shutdown delay, wear-and-tear, and expected turbine lifetime.

[0023] In an embodiment the collision risk level is a value between 0 and 100.

[0024] In a possible implementation form of the first aspect, the control value is a value between 0 and 100, wherein 0 corresponds to unimpeded turbine operation and 100 corresponds to a strong recommendation to completely stop the blades of the wind turbine.

[0025] In a possible implementation form of the first aspect, the control value is determined using an economic model, wherein the risk level and technological parameters of the wind turbine are inputs for the economic model.

[0026] In a possible implementation form of the first aspect, the economic model further takes as input economic factors for determining the control value, the economic factors comprising at least one of a minimum shutdown duration based on predictable power production and shutdown-related turbine wear-and- tear, and a predetermined budget corresponding to a remaining amount of allowable reduction of blade velocity based on time windows of different power demands and type of avian.

[0027] In an embodiment the economic model further takes as input legal factors for determining the control value, such as expected reduction in collision casualties between flying avians and the wind turbine rotor blades, expected reduction in loss of habitat of flying avians, or a minimum barrier performance.

[0028] In a possible implementation form of the first aspect, the risk model comprises a decision tree model based on at least one ruleset, each ruleset comprising at least one rule.

[0029] In a possible implementation form of the first aspect, the target data determined from radar plots comprises target classification; target track altitude; and / or speed, heading direction and sinuosity with respect to the at least one wind turbine; wherein the decision tree model comprises a track filtering ruleset for determining whether a target track is a track of interest based on at least one of the target data fulfilling the conditions of a respective rule in the track filtering ruleset.

[0030] In an embodiment applying the decision tree model comprises filtering out all target tracks which do not correspond to a track of interest such as aircraft, birds flying sufficiently high, tracks that are not straight that exhibit foraging behavior, tracks which are not headed towards the at least one wind turbine, or tracks coming from a non-migrating direction.

[0031] In an embodiment the method comprises obtaining transponder data, and the target tracks which do not correspond to a track of interest are determined at least partially based on the transponder data.

[0032] In a possible implementation form of the first aspect, the decision tree model comprises an alarm ruleset for determining whether and to what extent a wind turbine presents a risk for collision with an avian based on whether and how long the corresponding target track is located within a predefined alarm area and / or whether the corresponding target track crosses a predefined alarm line.

[0033] In a possible implementation form of the first aspect, the decision tree model comprises a risk ruleset for determining the level of risk a wind turbine presents for a target track based on at least one environmental risk factor, such as time of year, time of day, weather, visibility, and sun elevation in the avian detection area.

[0034] In an embodiment the at least one rule is manually adjustable by a user via a GUI, by values for parameters defining the rules.

[0035] In an embodiment multiple rules may produce different risk levels for a single wind turbine, such as one rule to detect bird activity around sunset and another rule during daytime; and wherein either the highest risk level is applied for determining the control value, or wherein the different risk levels are converted to probabilities and are combined using probability theory for determining the control value.

[0036] In an embodiment, multiple wind turbines may be controlled by a single rule, such as detecting bird migration based on track density. In a possible implementation form of the first aspect, the method comprises calculating statistical data from target data on heading direction and sinuosity of the number of target tracks for at least one predefined alarm area or alarm line, the statistical data comprising at least one of track density in the alarm area, migration traffic rate in the at least one alarm area, and crossing rate of the at least one alarm line; and wherein the risk model further takes into account the statistical data for determining the risk level.

[0037] In a possible implementation form of the first aspect, the risk model comprises a decision tree model with an avian activity ruleset for determining whether a target track corresponds to local avian activity or migrating avian activity, based on the target data, and further based on the statistical data on at least one of track density in the at least one alarm area, migration traffic rate in the at least one alarm area, or crossing rate of the at least one alarm line.

[0038] In a possible implementation form of the first aspect, the risk model comprises a decision tree model with a migration activity ruleset for determining whether migration activity has started within an alarm area comprising grid cells, based on whether the number of grid cells classified as migration activity within the alarm area is over a predefined migration activity threshold.

[0039] In a possible implementation form of the first aspect, the method comprises the steps of: defining a regular grid over the alarm area, the regular grid comprising grid cells, recording target track activity over time in each grid cell, determining a track vector for each target track in each grid cell at each point in time, calculating a cell vector for each grid cell by adding up the track vector of each target track in the grid cell over time, normalizing the resulting cell vector for the number of target tracks to determine a normalized cell vector for each grid cell, and classifying each grid cell as local activity or migration activity based on the length of its respective normalized cell vector, wherein grid cells with a normalized cell vector below a predefined cell vector threshold are classified as local activity, and grid cells with a normalized cell vector over the predefined cell vector threshold are classified as migration activity.

[0040] In a possible implementation form of the first aspect, classifying each grid cell further comprises the steps of: locating busy cells within the regular grid, wherein a grid cell is defined as a busy cell if its recorded target track count is larger than a predefined cell activity threshold, clustering adjacent busy cells into busy clusters, classifying each busy cluster as local activity or migration activity by determining a track vector for each target track in each busy cluster at each point in time, calculating a cell vector for each busy cluster by adding up the track vector of each target track in the busy cluster over time, normalizing the resulting cell vector for the number of target tracks to determine a normalized cell vector for each busy cluster, and classifying each busy cluster as local activity or migration activity based on the length of its respective normalized cell vector, wherein busy clusters with a normalized cell vector below a predefined cell vector threshold are classified as local activity, and busy clusters with a normalized cell vector over the predefined cell vector threshold are classified as migration activity, and applying the classification of each busy cluster to all cells in the convex hull of the respective busy cluster.

[0041] In a possible implementation form of the first aspect, classifying each grid cell comprises assigning a time-dependent track vector weight to each track vector based on respective age, and normalizing the cell vector takes into account each track vector weight to result in a smoothed cell vector. In an embodiment classifying each grid cell comprises calculating a running average for the normalized cell vector of each grid cell, and classifying each grid cell based on the running average.

[0042] In a possible implementation form of the first aspect, determining the control value is further based on directional migration activity towards the at least one wind turbine, wherein determining the directional migration activity comprises using a combination of temporal and spatial information by defining at least one probe line in proximity of a wind turbine, the probe lines being arranged substantially perpendicular to an expected migration direction towards the turbine, recording line crossing rate of target tracks crossing each probe line over time, calculating a migration profile corresponding to expected traffic over time by combining line crossing rates from the probe lines, and estimating directional migration activity towards the wind turbine based on the migration profile.

[0043] In an embodiment the economic model further takes into as input the directional migration activity towards the wind turbine for determining control values distributed over time, based on a remaining available shutdown budget and optionally based on time windows of different power demands.

[0044] In an embodiment determining the risk level comprises applying a machine learning model based on historical data on bird migration patterns and corresponding labelling of ground truth data.

[0045] According to a second aspect, there is provided a system for controlling at least one wind turbine to reduce risk of collision between at least one flying avian and at least one wind turbine rotor blade, the system comprising: at least one wind turbine; at least one avian radar configured to detect avians in an avian detection area and return radar plots corresponding to the avians; and one or more processors configured to obtain the radar plots; create and store a number of target tracks based on the obtained radar plots, each target track comprising target data determined from radar plots of corresponding avians; obtain a number of environmental risk factors affecting flying avian behavior in the avian detection area; determine a risk level with respect to a possible collision between the at least one wind turbine and at least one avian using a risk model based on target data of the number of target tracks and the number of environmental risk factors; and determine a control value for at least one wind turbine based on the risk level and technological parameters of the wind turbine, wherein the control value defines a recommended change of blade velocity for the at least one wind turbine.

[0046] In possible implementation forms and embodiments of the second aspect, the one or more processors are configured to execute a method according to possible implementation forms and embodiments of the first aspect.

[0047] According to a third aspect, there is provided a method for avian dispersal, the method comprising obtaining radar plots corresponding to at least one avian detected in an avian detection area by an avian detection system comprising at least one avian radar; creating and storing a number of target tracks based on the obtained radar plots, each target track comprising target data determined from radar plots of corresponding avians; obtaining a number of environmental risk factors affecting flying avian behavior in the avian detection area; determining a risk level with respect to a possible collision between the at least one wind turbine and at least one avian using a risk model based on target data of the number of target tracks and the number of environmental risk factors; and triggering an avian dispersal event of an avian dispersal system based on whether the risk level calculated for the avian detection area exceeds an avian dispersal threshold.

[0048] In an embodiment the method comprises defining at least one of an alarm area or at least one alarm line within the avian detection area, and wherein determining the risk level is based on whether and how long the corresponding target track is located within the alarm area and / or whether the corresponding target track crosses the at least one alarm line.

[0049] In an embodiment each target track is determined based on at least one obtained radar plot, and further based on sensor data on the corresponding avian detected in the avian detection area, wherein the sensor data is received from an additional sensor, such as a second radar, a camera, or a transponder.

[0050] In a possible implementation form of the third aspect, each target track is based on a match between corresponding target data of at least two, three or four radar plots. In a possible implementation form of the third aspect, the match between corresponding target data of radar plots comprises a match between range data. In a possible implementation form of the third aspect, the match between corresponding target data of radar plots comprises a match between radial velocity data. In a possible implementation form of the third aspect, the match between corresponding target data of radar plots comprises a match between return energy data.

[0051] Examples of matching conditions are known in the art of radar scanning and track generation, but a match condition may have to be fulfilled for a set of target data, which are part of the radar plots being compared, when a new target track is generated based on a number radar plots, while the data of a new radar plot has to match corresponding data of a stored target track, when a new radar plot is used for updating an existing target track. A match condition may be fulfilled for corresponding radar plot or target track, when the difference between the target data being matched is below a predetermined threshold difference. In an embodiment each target track is created based on a plurality of obtained radar plots, preferably at least 4 consecutive radar plots obtained along a path corresponding to an avian flight path.

[0052] In an embodiment the environmental risk factors comprise information about at least one of: time of year, time of day, weather, visibility, and sun elevation in the avian detection area.

[0053] In an embodiment the collision risk level is a value between 0 and 100.

[0054] In a possible implementation form of the third aspect, the risk model comprises a decision tree model based on at least one ruleset, each ruleset comprising at least one rule.

[0055] In a possible implementation form of the third aspect, the target data determined from radar plots comprises target classification; target track altitude; and / or speed, heading direction and sinuosity with respect to the at least one wind turbine; wherein the risk model comprises a decision tree model with a track filtering ruleset for determining whether a target track is a track of interest based on at least one of the target data fulfilling the conditions of a respective rule in the track filtering ruleset.

[0056] In an embodiment applying the decision tree model comprises filtering out all target tracks which do not correspond to a track of interest such as aircraft, birds flying sufficiently high, tracks that are not straight that exhibit foraging behavior, tracks which are not headed towards the at least one wind turbine, or tracks coming from a non-migrating direction.

[0057] In an embodiment the method comprises obtaining transponder data, and the target tracks which do not correspond to a track of interest are determined at least partially based on the transponder data.

[0058] In a possible implementation form of the third aspect, the risk model comprises a decision tree model with a risk ruleset for determining the level of risk a wind turbine presents for a target track based on at least one environmental risk factor, such as time of year, time of day, weather, visibility, and sun elevation in the avian detection area. In an embodiment the at least one rule is manually adjustable by a user via a GUI, by values for parameters defining the rules.

[0059] In an embodiment multiple rules may produce different risk levels for a single wind turbine, such as one rule to detect bird activity around sunset and another rule during daytime.

[0060] In a possible implementation form of the third aspect, the method comprises calculating statistical data from target data of the number of target tracks for at least one predefined alarm area or alarm line, the statistical data comprising at least one of track density in the at least one alarm area, migration traffic rate in the at least one alarm area, and crossing rate of the at least one alarm line; and wherein the decision tree model further takes into account the statistical data for determining the risk level.

[0061] In a possible implementation form of the third aspect, the risk model comprises a decision tree model with an avian activity ruleset for determining whether a target track corresponds to local avian activity or migrating avian activity, based on the target data on heading direction and sinuosity, and further based on the statistical data on track density in the at least one alarm area, migration traffic rate in the at least one alarm area, and / or crossing rate of the at least one alarm line.

[0062] In a possible implementation form of the third aspect, the risk model comprises a decision tree model with a migration activity ruleset for determining whether migration activity has started within an alarm area comprising grid cells, based on whether the number of grid cells classified as migration activity within the alarm area is over a predefined migration activity threshold.

[0063] In a possible implementation form of the third aspect, the method comprises the steps of: defining a regular grid over the alarm area, the regular grid comprising grid cells, recording target track activity over time in each grid cell, determining a track vector for each target track in each grid cell at each point in time, calculating a cell vector for each grid cell by adding up the track vector of each target track in the grid cell over time, normalizing the resulting cell vector for the number of target tracks to determine a normalized cell vector for each grid cell, and classifying each grid cell as local activity or migration activity based on the length of its respective normalized cell vector, wherein grid cells with a normalized cell vector below a predefined cell vector threshold are classified as local activity, and grid cells with a normalized cell vector over the predefined cell vector threshold are classified as migration activity.

[0064] In a possible implementation form of the third aspect, classifying each grid cell further comprises the steps of: locating busy cells within the regular grid, wherein a grid cell is defined as a busy cell if its recorded target track count is larger than a predefined cell activity threshold, clustering adjacent busy cells into busy clusters, classifying each busy cluster as local activity or migration activity by determining a track vector for each target track in each busy cluster at each point in time, calculating a cell vector for each busy cluster by adding up the track vector of each target track in the busy cluster over time, normalizing the resulting cell vector for the number of target tracks to determine a normalized cell vector for each busy cluster, and classifying each busy cluster as local activity or migration activity based on the length of its respective normalized cell vector, wherein busy clusters with a normalized cell vector below a predefined cell vector threshold are classified as local activity, and busy clusters with a normalized cell vector over the predefined cell vector threshold are classified as migration activity, and applying the classification of each busy cluster to all cells in the convex hull of the respective busy cluster. In a possible implementation form of the third aspect, classifying each grid cell comprises assigning a time-dependent track vector weight to each track vector based on respective age, and normalizing the cell vector takes into account each track vector weight to result in a smoothed cell vector.

[0065] In an embodiment classifying each grid cell comprises calculating a running average for the normalized cell vector of each grid cell, and classifying each grid cell based on the running average.

[0066] In a possible implementation form of the first aspect, triggering an avian dispersal event is further based on directional migration activity towards the at least one wind turbine, wherein determining the directional migration activity comprises using a combination of temporal and spatial information by defining at least one probe line in proximity of a wind turbine, the probe lines being arranged substantially perpendicular to an expected migration direction towards the turbine, recording line crossing rate of target tracks crossing each probe line over time, calculating a migration profile corresponding to expected traffic over time by combining line crossing rates from the probe lines, and estimating directional migration activity towards the wind turbine based on the migration profile.

[0067] In an embodiment determining the risk level comprises applying a machine learning model based on historical data on bird migration patterns and corresponding labelling of ground truth data.

[0068] According to a fourth aspect, there is provided a system for avian dispersal, the system comprising: at least one wind turbine; at least one avian radar configured to detect avians in an avian detection area and return radar plots corresponding to the avians; an avian dispersal system; and one or more processors configured to obtain the radar plots; create and store a number of target tracks based on the obtained radar plots, each target track comprising target data determined from radar plots of corresponding avians; obtain a number of environmental risk factors affecting flying avian behavior in the avian detection area; determine a risk level with respect to a possible collision between the at least one wind turbine and at least one avian using a risk model based on target data of the number of target tracks and the number of environmental risk factors; and trigger an avian dispersal event of the avian dispersal system based on whether the risk level calculated for the avian detection area exceeds an avian dispersal threshold.

[0069] In possible implementation forms and embodiments of the fourth aspect, the one or more processors are configured to execute a method according to possible implementation forms and embodiments of the third aspect.

[0070] These and other aspects will be apparent from and the embodiment(s) described below.

[0071] BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In the following detailed portion of the present disclosure, the aspects, embodiments and implementations will be explained in more detail with reference to the example embodiments shown in the drawings, in which:

[0073] Fig. 1 is a schematic illustration of a system for reducing risk of collision between flying avians and wind turbine rotor blades according to an example of the disclosure,

[0074] Fig. 2 shows a schematic top view illustrating an alarm area around a wind turbine with detected target tracks according to an example of the disclosure,

[0075] Fig. 3 shows an overview flow diagram of a method for determining a control value for a wind turbine according to an example of the disclosure, Fig. 4 shows a flow diagram of a method determining a risk level of a possible collision between a wind turbine and an avian according to an example of the disclosure,

[0076] Fig. 5 shows a flow diagram of a method for determining whether a target track corresponds to local avian activity or migrating avian activity according to an example of the disclosure,

[0077] Fig. 6 shows a flow diagram of a method for classifying grid cells as local or migrating avian activity according to an example of the disclosure,

[0078] Fig. 7 shows a flow diagram of a method for determining start of a migration activity within an alarm area according to an example of the disclosure,

[0079] Fig. 8 shows a schematic top view illustrating how a track vector is calculated for a target track in a grid cell according to an example of the disclosure,

[0080] Fig. 9 shows a schematic top view illustrating how a normalized cell vector is calculated for grid cells in a grid according to an example of the disclosure,

[0081] Fig. 10 shows a schematic top view illustrating how a smoothed cell vector is calculated for a grid cell according to an example of the disclosure,

[0082] Fig. 11 shows a schematic top view illustrating a method of locating busy cells within the regular grid according to an example of the disclosure,

[0083] Fig. 12 shows a schematic top view illustrating a method of clustering adjacent busy cells into busy clusters according to an example of the disclosure,

[0084] Fig. 13 shows a schematic top view illustrating a method of determining directional migration activity according to an example of the disclosure,

[0085] Fig. 14 illustrates calculating a migration profile over time according to an example of the disclosure,

[0086] Fig. 15 is a schematic illustration of a system for avian dispersal at a wind turbine according to an example of the disclosure, and

[0087] Fig. 16 shows a flow diagram of a method for avian dispersal at a wind turbine according to an example of the disclosure. DETAILED DESCRIPTION

[0088] The following detailed description relates to specific, but non-limiting, exemplary embodiments of the invention.

[0089] Referring to Fig. 1 there is shown a schematic block diagram illustrating the basic structure of an example embodiment of the present disclosure.

[0090] The system comprises basically at least one wind turbine 1 comprising rotor blades 2 and an avian detection system with at least one avian radar 5 emitting radar signals and receiving reflections from avians 3 in an avian detection area 4, for instance a flock of birds 3 heading towards the wind turbine 1 (the terms ‘birds’ and ‘avians’ are being used interchangeably throughout the description).

[0091] In Fig. 1 only a single avian radar 5 is shown, but it is understood that more than one avian radar 5 may be used at the same time. The radars can be local radars but also long-range radars may be applied. According to an embodiment, two radars, one vertical and one horizontal are used. The horizontal radar may provide latitude / longitude and size information of avians 3, while the vertical radar may provide height information and detailed information like wing beats.

[0092] The avian detection system may have a static coverage range corresponding to the avian detection area 4. It would however also be possible to use non-static coverage ranges that might for instance adapt to certain specific dynamic conditions at the site of application of the detection system and embodiments applying such non-static coverage ranges will also fall within the scope of the present invention.

[0093] In an embodiment the avian detection system is a 3D radar system that includes an avian radar 5 with an antenna provided with means for varying its effective pointing direction in elevation. A radar transmitter is operatively connected to the antenna for generating a radar signal for emission via the antenna. In an exemplary embodiment the radar transmitter transmits pulses of constant width at a constant pulse repetition frequency (PRF) at X-band or S-Band (or other bands). In an embodiment the avian detection system is a high range resolution FMCW radar system designed for detecting and characterizing avian objects 3 by using micro-Doppler analysis of obtained range-Doppler data. The relative motion is characteristic for different classes of targets, e.g. the flapping motion of a bird’s wing vs. the spinning of propeller blades. In a range-Doppler map or range-radial velocity map, the moving parts of a body causes a characteristic Doppler signature, where the main contribution comes from the torso of the body, which causes the Doppler frequency of the target, while the flapping motion of bird wings or propeller blades induces modulation on the returned radar signal and generates sidebands around the central Doppler frequency of a Doppler signature, which may be referred to as micro-Doppler signatures. The width of the sidebands of a micro-Doppler signature within a range-Doppler map / matrix may therefore be indicative of the type of target being hit by the transmitted radar waves. When using radial velocity for a range-Doppler map, the width of the sidebands of a micro-Doppler signature will then be given by the width of the radial velocity span of the micro-Doppler signature.

[0094] A radar receiver is operatively connected to the antenna. One or more processors may be operatively connected to the receiver. A sampling system may digitize the radar return video signal. The processor(s) may be configured for detecting and localizing objects, such as birds 3, in azimuth and range.

[0095] In an embodiment the processor(s) may further be configured for estimating a height of a detected object or bird 3 based on relative amplitudes of echo returns as a function of elevation pointing direction of the antenna.

[0096] In another possible embodiment, an azimuth scanner is operatively coupled to the antenna for rotating the radar 5 about an axis. The radar 5 antenna may rotate continuously 360 degrees in azimuth while transmitting and receiving. The inclination of the radar antenna may be fixed or variable. Signals (raw data) from the radar 5 may be provided to a first processing unit that comprises image processors and filters known from the prior art and used for instance to remove signals from rain, ground clutter and moving objects that are not birds 3. It is important that the limitations of the used avian radar 5 are taken into account in the risk model 10. For example, it is usually not possible to detect birds 3 that are flying between the waves due to clutter.

[0097] As illustrated in Fig. 2, the processed data is analysed in order to detect and define radar plots 6, and these plots 6 are searched within the avian detection area 4, e.g. in relation to an alarm line 21 or an alarm area 20 defined around a wind turbine 1 .

[0098] A radar plot 6 may be defined as a reflection (location and strength of the reflection) that is most probably a bird 3 and could be part of a target track 7. Data from a transponder receiver may also be analysed to define transponder plots 50.

[0099] The radar plots 6 (of multiple radars) are used to generate (3D) target tracks 7. The tracker that can be used is a tracker developed by the applicant as a tool that uses multiple tracking algorithms like Kalman filtering. Furthermore, a target track 7 may be further determined using additional sensor data on the corresponding avian 3 detected in the avian detection area 4, such as a camera used to compare radar reflection with captured images (species recognition).

[0100] Besides data from radars 5 and additional sensors, the system also relies on a number of environmental risk factors 9 affecting flying avian behavior in the avian detection area 4. These environmental risk factors 9 comprise information about time of year, time of day, weather, visibility, and sun elevation in the defined avian detection area 4, among others, and can be provided by a local weather station and / or derived from historical radar data, field work, or biology.

[0101] The data from the radars 5, transponder receivers, and other sensors is provided to a processing unit, which in practice consists of a number of processors (which may depend on number of servers). This input data may have multiple formats, but for the radars 5 it may consist of raw images with radar reflections. The information from other sensors may be provided in the form of raw data for further processing to the processing unit.

[0102] The information provided by the avian detection system may be stored in a database (e.g. PostgreSQL or MySQL). This database is preferably located on a separate server, with powerful data management / back-up facilities. Different processing steps performed in accordance with embodiments of the system and method of the present disclosure are discussed in the following and illustrated in connection with Figs. 3-14.

[0103] Fig. 3 shows an overview flow diagram of a method for reducing risk of collision between flying avians 3 and wind turbine rotor blades 2. An avian radar system comprising at least one avian radar 5 is provided as shown in Fig. 1 , and radar plots 6 are obtained relating to avian objects 3 detected by the radar system. New data are regularly provided by the avian radar system and the optional transponder receiver and further sensors, and corresponding new or updated radar plots 6 and transponder plots 50 are obtained. In case of separate horizontal and vertical radars, new horizontal radar data may be provided at regular intervals, such as intervals of or about 1 .3 seconds, and new vertical radar data may be provided at regular intervals, such as intervals of or about 3 seconds, while new transponder data may be provided at intervals which may be a bit random, with intervals in the range of 2-4 seconds.

[0104] For horizontal radar plots, raw image data is obtained when the horizontal radar detects an object. The raw data is processed and filtered to remove clutter, where static clutter, such as clutter provided by buildings, can be removed based on raw data, which has been obtained within the last couple of hours. Semi dynamic clutter, such as clutter provided by falling leaves, can be removed based on raw data, which has been obtained within the last minutes. Dynamic clutter, such as clutter provided by rain, can be removed based on raw data, which has been obtained from the last received images. The final obtained horizontal plot holds size data and location data of the detected object, where the location data comprises latitude and longitude data of the object. The obtained plots may be used to create or update target tracks 7, as will be explained below.

[0105] For vertical radar plots, raw image data is obtained when the vertical radar detects an object. The raw data is processed and filtered to remove clutter in processes similar to the processes for the horizontal radar date. The final obtained vertical plot holds size data, location data in the form of range data, and height data of the detected object. The obtained plots may be used to create or update target tracks 7, as will be explained below.

[0106] For the transponder plots 50, transponder data is generally ADS-b transponder data associated with certain aircraft (manned or unmanned). The relevant parameters of the received transponder data are selected to obtain a transponder plot 50 holding data which may represent all or part of the following: location, height, direction of movement, speed, and type (classification). The obtained transponder plots 50 may be used to create transponder tracks, as will be explained below.

[0107] Optionally, transponder data can be in the form of a Passive Integrated Transponder (PIT) tag, is received by a transponder receiver. The obtained (optionally combined horizontal and vertical) radar plots 6 are compared in order to create and store a number of target tracks 7. If applicable, received transponder plots 50 are used to generate transponder tracks, and these transponder tracks are associated with target tracks 7 based on radar plots 6 to classify target tracks 7 e.g. as an aircraft or UAV.

[0108] When a new plot 6 is provided, then the first step is to try to match the new plot 6 with an already existing target track 7. If there is a match to an existing target track 7, then the data of the matched track is updated with data of the new track to obtain an updated or extended target track 7. If there are no existing tracks, or there is no match to any of the existing tracks, then the new plot 6 is stored in order to be matched with other plots 6 to create a new target track 7. If there is no match to other plots 6 within a predetermined time frame T, such as a time frame of about 10 seconds, then the plot 6 is dropped or discarded. If there is a match to other plots 6, a new track or target track 7 is created based on the matching plots 6 as follows.

[0109] If there is a match of position data, then the next step is a match of direction of movements of object of the plots. Received radar plots 6 may not hold direction data, and if there is no match between direction data due to lack of such data, the matching proceeds to a match of speed data. Received radar plots 6 however may not hold speed data, and if there is no match between speed data due to lack of such data, the matching proceeds to a match of size data. All received plots hold size data, where for transponder plots the size data may be represented by classification data. If two matched plots 6 have matching size data, then the matching proceeds to the next step, where a new target track 7 is created holding object data 8 of the matched plots 6. It is preferred that a new target track 7 is based on at least four matching radar plots 6.

[0110] When a new target track 7 has been created, the target track 7 is stored and will be matched with new incoming or received plots 6. Here, the first data to be matched is position or location data, including range and latitude and longitude data for vertical and horizontal radar plot 6 data. If there is no match between position / location data, the plot data of the received plot is stored. If there is a match of position data, then the next step is a match of data for direction of movements of the avian objects 3. Received radar plots 6 may not hold direction data, and if there is no match between direction data due to lack of such data, the matching proceeds to a match of speed data es described above, such as if there is no match between speed data due to lack of such data, the matching proceeds to a match of size data, and if there is no match between the size data, then the plot is stored.

[0111] It is preferred that the determination of when a match condition is fulfilled for a set of corresponding plot data or a set of corresponding plot and track data is based on a difference measure of values representing the data being matched, where the difference measure may be smaller than a predetermined threshold difference.

[0112] Received radar plots 6 hold data representing position or location of the detected avian objects 3, but data for direction of movement and speed may not be directly available from the radar plots 6. When a number of matching radar plots 6 has been received, the matching plots 6 may be part of a track 7, and based on position / location data of a number of received or successively received plots 6 of the same track 7, data for direction of movement and / or speed can be determined for the object of the track 7. The track 7 can now be updated to also include data for direction of movement and / or speed. It is preferred that calculation or determination of data for direction of movement and / or speed is based on at least 4 or 5 received or successively received plots.

[0113] It is noted that a target track 7 can represent a single bird 3 or several birds 3, such as a flock of birds 3, and the size data of a target track 7 can represent a single bird 3, several birds 3 or the size of a flock of birds 3.

[0114] After creating and storing target tracks 7 based on the obtained radar plots 6, the subsequent method steps include, as illustrated, obtaining environmental risk factors 9 affecting flying avian behavior in the avian detection area 4; and determining a collision risk level 11 using a risk model 10 that relies on the obtained environmental risk factors 9 and target data 8 of the target tracks 7, as will be explained below in more detail with respect to Fig. 4.

[0115] In a final step of the method, a recommended change of blade velocity is determined in the form of a control value 12 for the wind turbine 1 using an economic model 14, wherein the calculated collision risk level 11 and technological parameters 13 of the wind turbine 1 (such as shutdown delay, wear- and-tear, and expected turbine lifetime) are inputs for the economic model 14.

[0116] As also illustrated in Fig. 3, the economic model 14 can further take as input economic factors 15 for determining the control value 12, these economic factors 15 comprising, for example, a minimum shutdown duration based on predictable power production and shutdown-related turbine wear-and-tear, and / or a predetermined budget corresponding to a remaining amount of allowable reduction of blade velocity.

[0117] In an embodiment the economic model 14 can further take as input legal factors for determining the control value 12, such as expected reduction in collision casualties between flying avians 3 and the wind turbine rotor blades 2, expected reduction in loss of habitat of flying avians 3, and / or a minimum barrier performance (e.g. at least an average of 30%).

[0118] In simplified terms, the more emergency stops are triggered by the system, the more the technical or economic lifetime of a turbine 1 is reduced. Therefore, it is advantageous to develop a system that can give an early warning with a recommended control value 12 corresponding to a pause or slowdown recommendation, so the turbine can be operated in a more “gentle” fashion. Such an output control value 12 can for example be a value between 0 and 100, wherein 0 corresponds to a suggested unimpeded turbine operation and 100 corresponds to a strong recommendation to completely stop the blades 2 of the wind turbine 1. This allows wind turbine operators to already slow down the turbine blades 2 during low pause recommendation in order to prevent an emergency stop at 100, in contrast to the current practice of automatic shutdown events triggered based on a simple threshold.

[0119] Fig. 4 shows a more detailed flow diagram of a method for determining a collision risk level 11 between wind turbine(s) 1 and avian(s) 3 using a risk model 10 according to an embodiment of the present disclosure.

[0120] The risk model 10 in this and related embodiments comprises a decision tree model 16 based on at least one ruleset, each ruleset comprising at least one rule.

[0121] As explained before with respect to Fig. 3, the target data 8 determined from radar plots 6 can comprise any or all of the following data regarding the target track 7 :

[0122] - target classification;

[0123] - target track altitude;

[0124] - speed, and / or

[0125] - heading direction with respect to the at least one wind turbine 1 .

[0126] For tracks holding data for direction of movement, it is also possible to determine the sinuosity of a target track 7, where sinuosity can be determined by measuring the change in direction of movement per timeframe.

[0127] The classification of the tracks may further comprise a classification based on stored track data representing size of the object. The classification classes of a bird track may comprise: large bird, medium bird and small bird. The classification classes of a collision object track may comprise: plane and vehicle. If received, the target classification can also be determined based on transponder data. As shown in Fig. 4, the decision tree model 16 comprises a track filtering ruleset 17 for determining whether a target track 7 is a track of interest 18 based on the target data 8 fulfilling the conditions of a respective rule in the track filtering ruleset 17, such as the altitude, speed, and / or heading direction of the detected avian 3 being within certain predefined ranges or above / below a certain threshold. Applying the decision tree model 16 this way helps filtering out all target tracks 7 which do not correspond to a track of interest 18 such as aircraft, birds flying sufficiently high, tracks that are not straight that exhibit foraging behavior, tracks which are not headed towards the at least one wind turbine 1 , or tracks coming from a non-migrating direction.

[0128] It is also within embodiments of the disclosure, that tracks of interest 18 are selected according to any or all of the following requirements:

[0129] - the reflection and or the radar cross section (RCS) of a detected target track 7is above a given threshold, such as above 0.025 m2 (in this case the RCS can represent one or more avian 3, and an RCS can be the sum of individual RCSs);

[0130] - the height above ground of the target track 7is within a predefined altitude window, such as being less than 150 meters above ground;

[0131] - the ground or airspeed of the detected target track 7is above a given threshold, such as above 20 m / s;

[0132] - the track length of the detected target track 7(expressed in meters, seconds, or even number of plots) is above a given threshold, such as above 15 seconds when measured as a length of time, or the path of the target track 7 meets certain predefined criteria (such as the path having a predefined shape).

[0133] If the target track 7 is not a track of interest 18, then the track 7 is filtered out, no risk level will be determined, and no resulting control value 12 needs to be calculated. If the target track 7 is a track of interest 18, then at least moving direction and speed of the avian 3 is determined for further risk assessment.

[0134] As further shown in Fig. 4, the decision tree model 16 also comprises an alarm ruleset 19 for taking into account whether and how long the corresponding target track 7 that has been identified as a track of interest 18 is located within a predefined alarm area 20 and / or whether the corresponding target track 7 crosses a predefined alarm line 21 , for determining whether and to what extent a wind turbine 1 presents a risk for collision with an avian 3.

[0135] As illustrated also in Fig. 1 and 2, while the avian detection area 4 is determined by the detection range of the avian detection system, the alarm area 20 and the alarm line(s) 21 are determined based on the current or planned location of the wind turbine(s) 1 in question, in particular based on the circular area where rotor blades 2 can be spinning, i.e. the rotor swept area.

[0136] A possible requirement is that the flying height of the avian 3 shall be less than a predetermined height, which can be set to 150 meters. This corresponds to defining the height of the alarm area 20. The alarm area 20 therefore can be considered to extend in three dimensions, with the third dimension being the height, which can be defined by a maximum flying height for a bird 3 to be classified based on its target track 7 as a track of interest 18.

[0137] In the examples illustrated in Fig. 2, the target track 7 on the left is detected as a track of interest 18, and although it is not in the alarm area 20, it has crossed a predefined alarm line 21 arranged outside the alarm area to indicate a direction towards the rotor swept area. Therefore, this target track 7 is taken into account for determining a higher risk level 11 resulting in a control value 12 triggering or strongly suggesting a pause or slowdown for the rotor blades 2 of the wind turbine 1.

[0138] In the other example illustrated in Fig. 2, the target track 7 on the right is also detected as a track of interest 18, but it leaves the alarm area 20 before a determined time threshold, and heads in a direction away from the rotor swept area. Therefore, this target track 7 is not taken into account for calculating a control value 12.

[0139] In another possible scenario, if the target track 7 on the right is detected earlier, it may still result in a higher risk level 11 and a suggested slowdown of the rotor blades 2 of the wind turbine 1 , based on sensitivity settings, turbine parameters 13 and further economic factors 15. In a possible implementation of the system, if a detected bird 3 is flying towards the wind turbine 1 , but it is so close to the rotor swept area that it may collide with the rotor blades 2 within a time period being shorter than a determined reaction time, such as 100 seconds, then there is not enough time to react to the bird 3, and hence there is no reason to determine a risk level 11 or calculate a control value 12.

[0140] As further shown in Fig. 4, the decision tree model 16 also comprises a risk ruleset 22 for taking into account at least one environmental risk factor 9, such as time of year, time of day, weather, visibility, temperature, and sun elevation in the avian detection area 4. These all have a big impact on bird behavior and therefore risk of bird mortality.

[0141] Taking into account these environmental risk factor 9 is at least partially data- driven, i.e. based on detecting patterns in the available data and associating these patterns with biological explanations. The following is a few examples of how environmental risk factor 9 can be taken into account in the model.

[0142] Time of year: Migration is clearly seasonal, so both population and behavior will be different. If birds have offspring, they may fly more or between different areas.

[0143] Time of day: Some species are not active at night or vice versa. This can be especially relevant for the risk ruleset 22 if the goal is to protect a specific species. For example, in cases where wind turbines are located between a nesting and feeding area of a specific species, these species can be expected to exhibit very regular patterns moving from the nesting area to the feeding area. This can be used to enable / disable rules intended to protect this bird species based on time of day.

[0144] Sun elevation: Biologically, night and day are more related to sun elevation than exact time of day, so the same considerations apply as above, as an alternative.

[0145] Visibility: One obvious consideration is an elevated risk in case of low visibility, due to the fact that if birds can't see the turbine blades, it becomes more difficult to avoid a collision. Another consideration is that birds of prey are less active in low visibility as they can't see prey, resulting in lower risk if the goal is protecting birds of prey. Temperature is also a relevant factor: on very hot days bird activity is lower.

[0146] Another possible use of these environmental risk factor 9 originates from the fact that the radar 5 can also miss certain targets. The above factors 9 can be used to estimate what is missed, i.e. what target tracks 7 have a higher possibility of going undetected. Accordingly, the model output can result in a higher control value to slow down or turn off a turbine, even based on the same radar measurements, when more bird activity is expected due to e.g. time of day.

[0147] As further shown in Fig. 4, and will be explained in more detail in relation to Figs. 5-14, the risk model 10 takes into account further statistical data 24 derived from target data 8 on heading direction and sinuosity of a number of target tracks 7, in relation to at least one predefined alarm area 20 or alarm line 21 .

[0148] For example, during a pre-construction evaluation phase of a wind turbine 1 installation, avian radars 5 can be used as a reliable and long-term measurement tool, gathering scientific data on bird movements in the area, as well as migration activity. They can automatically detect and log thousands of birds 3 simultaneously, including RCS (as a proxy for physical size), speed, direction and flight path. Historically, this type of information would be gathered by human observers over one or more limited observation periods. The information on bird activity, based on avian radar 5 data, can not only be used as an input for a risk model 10 for determining a collision risk level 11 , but also compared both pre- and post-construction of wind turbines 1 to measure the true impact on the local and migratory bird population.

[0149] The calculated statistical data 24 may comprise:

[0150] - track density in the alarm area 20,

[0151] - migration traffic rate in the alarm area 20, and / or

[0152] - crossing rate of the alarm line 21 .

[0153] Some wind farm operators have different shutdown budgets for local activity and migrating birds. Different rules can be written for these two cases. Rules to protect local birds typically use a small alarm area defined around the turbine they control. Only a few tracks in the area may trigger a shutdown event. Users typically add constraints such as time-of-day or sun elevation in order to capture known-behavior. Some species have very regular behavior in feeding or flying patterns, which may be captured in rules implemented in a decision tree model 16. Bird migration is an event that involves large numbers of birds over large areas. Since there is a larger risk of high bird mortality during migration events, operators often want to shut down the wind farm completely when migration begins. A goal of the methods and systems described herein is to avoid the need for such shutdown events and still reduce risk of collision and thus bird mortality during migration events.

[0154] In order to detect migration, an alarm area 20 at some distance from the wind turbine 1 can be used. Its location depends on the expected direction of migration (e.g. North-South) and local features in the landscape such as rivers. Tracks 7 are filtered by their sinuosity (straightness) and heading. Then, track density or estimated migration traffic rate (MTR) is used to assess the likelihood of migration. MTR is a standard metric that can be estimated using track length or track speed. Alternatively, an alarm line 21 is used instead of an alarm area 20, and the number of line crossings is estimated per hour per kilometer.

[0155] Fig. 5 illustrates an example of the method of the disclosure, where the risk model 10 takes into account the statistical data 24 via a decision tree model 16, either implemented separately or as part of the decision tree model 16 described before in relation to Fig. 4 for determining whether a target track 7 corresponds to local avian activity or migrating avian activity. The decision tree model 16 includes an avian activity ruleset 23 based partly on the target data 8 obtained from the avian detection system such as sinuosity and heading direction, and further based on the statistical data 24 on track density in an alarm area 20, migration traffic rate in an alarm area 20, and / or crossing rate of an alarm line 21 .

[0156] Figures 6-12 illustrate an embodiment where the risk model 10 comprises a decision tree model 16 with a migration activity ruleset 37. First, as shown best in in Figs. 11 -12, a regular grid 27 is defined over the alarm area 20 with regular grid cells 28. The risk model 10 relies on this regular grid 27 for classifying each grid cell 28 as local activity 25 or migration activity 26 as illustrated in the flowchart of Fig. 6.

[0157] The model 10 can then determine whether migration activity 26 has started within an alarm area 20, based on a rule on whether the number of grid cells 28 classified as migration activity 26 is over a predefined migration activity threshold, as illustrated in the flowchart of Fig. 7.

[0158] For classifying grid cells 28 as local activity 25 or migration activity 26, the target track activity 44 is recorded over time in each grid cell 28, as illustrated in Fig. 8. Based on the recorded track activity 44, a track vector 29 is calculated for each target track 7 in each grid cell 28 at each point in time.

[0159] As shown in Fig. 8, at plot 0 (p=0) there is a single plot and hence no velocity I direction. At p=1 ,2,3 an estimated velocity is used (tracker, black dashed arrows) for determining a track vector 29 for the target track 7. These track vectors 29 are each associated with a grid cell 28. Each track vector’s timestamp (darkness of gray dashed arrow) is also stored for weighing by age, as will be explained in relation with Fig. 10.

[0160] Next, a cell vector 30 is calculated for each grid cell 28 by adding up the track vector 29 of each target track 7 in the grid cell 28 over time, as illustrated in Fig.

[0161] 9 (left arrows). The resulting cell vector 30 is then normalized for the number of target tracks 7 to determine a normalized cell vector 31 for each grid cell 28 (right arrows).

[0162] For local activity, this normalized cell vector 31 tends to be small, while during migration this normalized cell vector 31 tends to be large. The reason is simply because during migration almost all birds are traveling in the same direction. If there are sufficiently many migration cells it can be determined that migration has started.

[0163] Each grid cell 28 is thus classified as local activity 25 (top right arrow) or migration activity 26 (bottom right arrow) based on the length of its respective normalized cell vector 31 , wherein grid cells 28 with a normalized cell vector 31 below a predefined cell vector threshold are classified as local activity 25, and grid cells 28 with a normalized cell vector 31 over the predefined cell vector threshold are classified as migration activity 26.

[0164] In an embodiment illustrated in Fig. 10, classifying each grid cell 28 further comprises a smoothing step. A time-dependent track vector weight 32 is assigned to each track vector 29 based on its respective age as described before, and the step of normalizing the cell vector 30 takes into account each track vector weight 32 to result in a smoothed cell vector 33.

[0165] Due to limitations of avian radars 5, a grid with relatively small grid cells 28 can exhibit sampling effects. To mitigate this effect, in another possible embodiment, a running average is calculated for the normalized cell vector 31 of each grid cell 28 and classifying each grid cell 28 is based on this running average. This way sampling effects can be effectively smoothed in time, which in turn gives a more robust cell classification.

[0166] Figs. 11 and 12 illustrate an embodiment of the method where the classification of grid cells 28 is based on clustering. This allows the model to make decisions based on a limited number of clusters rather than many tracks.

[0167] In this embodiment, busy cells 34 are located within the regular grid 27 as shown in Fig. 11 , wherein a grid cell 28 is defined as a busy cell 34 if its recorded target track 7 count is larger than a predefined cell activity threshold.

[0168] Once located, adjacent busy cells 34 are clustered into busy clusters 35, as shown in Fig. 12, and each busy cluster 35 is classified as local activity 25 or migration activity 26 according to one of the embodiments described before. Finally, the classification of each busy cluster 35 is extended to all cells in the convex hull 36 of the respective busy cluster 35.

[0169] The rules set for the decision tree model 16 and described before in relation to Figs. 4-10 may be manually adjustable by a user via a GUI, by values for parameters defining the rules.

[0170] The collision risk level 11 calculated may be given as a percentage number ranging from 0 -100, and a collision risk level above 80 is considered critical. In an embodiment, determining the risk level 11 comprises applying a machine learning model based on historical data on bird migration patterns and corresponding labelling of ground truth data.

[0171] In some embodiments, multiple rules may produce different risk levels 11 for a single wind turbine 1 , such as one rule to detect bird activity around sunset and another rule during daytime. In such embodiments, either the highest risk level 11 is applied for determining the control value 12, or the different risk levels 11 are converted to probabilities and are combined using probability theory for determining a control value 12.

[0172] In other embodiments, multiple wind turbines 1 may be controlled by a single rule, such as detecting bird migration based on track density.

[0173] A migration event typically starts with only a few birds. This by itself is hard to distinguish from local activity (“just a few birds”). The birds may also come in waves. This makes it hard to effectively plan turbine shutdown, especially given minimum reaction time and shutdown duration, and limited shutdown budgets. In order to facilitate shutdown planning a migration profile 41 can be used, based on both temporal and spatial information.

[0174] Using probe lines 39 it is possible to monitor traffic over time. Based on the location of the probe line 39 and average bird speed it is possible to estimate when the target tracks 7 will arrive at a turbine 1 (or complete wind farm). Nearby probe lines 39 can give relatively accurate estimates of core decision variables such as line crossings or MTR, but limited reaction time, while probe lines 39 at a greater distance give less accurate estimates but more reaction time. Combining the data from several probe lines 39 is therefore the best approach to estimate the expected traffic over time. This migration profile 41 can then be used to distribute the shutdown budget, given all constraints.

[0175] Figs. 13 and 14 illustrate such an embodiment of the method wherein determining the control value 12 is at least partially based on directional migration activity 38 towards the at least one wind turbine 1. Herein, determining the directional migration activity 38 comprises using a combination of temporal and spatial information. As illustrated in Fig. 13, in a first step probe lines 39 are defined in proximity of a wind turbine 1 . These probe lines 39 are arranged substantially perpendicular to an expected migration direction towards the turbine.

[0176] In the setup shown in Fig. 13, it is possible to detect the vanguard of migrating birds in the nearest probe, while the bulk of the migrating birds is visible in other probes. This allows us a shutdown already when we the vanguard is detected, because it can be predicted that the bulk will arrive later. Advantageously, the probe lines 39 can be placed along known migration routes (a river in this case).

[0177] Once probe lines 39 are defined, line crossing rate 40 of target tracks 7 crossing each probe line 39 can be recorded over time, and a migration profile 41 corresponding to expected traffic over time can be calculated by combining line crossing rates 40 from the probe lines 39, as shown in Fig. 14. Based on the migration profile 41 , directional migration activity 38 towards the wind turbine 1 can be estimated. In the shown example, a migration profile 41 is created based on 3 measurements for each probe line 39. Two shutdown events are then planned for the two migration waves estimated.

[0178] The directional migration activity 38 towards a wind turbine 1 can also be taken as input for the economic model 14 for determining control values 12 distributed over time, based on a remaining available shutdown budget and optionally based on time windows of different power demands.

[0179] Clearly, wind turbine operators want to avoid shutdown as much as possible. It is this advantageous to provide a solution that offers bird dispersal, i.e. scaring away birds e.g. using lights, gas cannons or distress calls. If effective, dispersal is clearly preferable over shutdown. Instead of triggering a shutdown event it is easier and cheaper to trigger a dispersal event.

[0180] Currently, dispersal is typically employed by human operators, e.g. the bird control unit at airports. Unfortunately, dispersal methods are only effective in a relatively small area. Thus, multiple dispersal devices are necessary to cover a significant area. Figs. 15 and 16 illustrate another aspect of the present disclosure, wherein the collision risk between flying avians 3 and wind turbine rotor blades 2 is reduced by an avian dispersal system 42.

[0181] In this illustrated aspect, structures and features that are the same or similar to corresponding structures and features previously described or shown herein are denoted by the same reference numeral as previously used for simplicity.

[0182] As shown in Fig. 15, the system is essentially the same as described before with respect to Fig. 1 , with the difference of an additional avian dispersal system 42 arranged in the vicinity of the wind turbine(s) 1. In a proposed system, multiple dispersal devices are placed at carefully selected locations, e.g. at the center of alarm areas 20.

[0183] The avian dispersal system 42 can be any system known from the prior art that is suitable for locating around wind turbines 1 and can generate e.g. an audio or visual alarm that can deter flying avians 3.

[0184] As illustrated in Fig. 16, the system relies on detected target tracks 7 and environmental risk factors 9 affecting flying avian behavior in the avian detection area 4. These and further optional input, as described before in relation to Figs. 3-7, are used as input for a risk model 10 to determine a risk level 11 with respect to a possible collision between the wind turbine 1 and at least one avian 3.

[0185] In an embodiment, an additional difference is that at least one alarm area 20 and / or alarm line 21 is defined within the avian detection area 4, and determining the risk level 11 is based on whether and how long the corresponding target track 7 is located within the alarm area 20 and and / or whether the corresponding target track 7 crosses the at least one alarm line 21 .

[0186] In a simple embodiment, as shown in Fig. 16, this risk level 11 is compared to an avian dispersal threshold, and an avian dispersal event 43 is triggered if the risk level 11 calculated for the avian detection area 4 exceeds this avian dispersal threshold. If the risk level 11 obtained is below the predetermined avian dispersal threshold value, no avian dispersal event 43 is triggered, and the target track 7 is stored. In an embodiment that the predetermined avian dispersal threshold value of the collision risk level 11 is set to a value of 60.

[0187] Clearly, the use of dispersal techniques does not rule out shutdown. It is entirely possible to have a completely separate set of rules for the aspect of wind turbine control using a recommended control value 12, next to a set of rules for dispersal as described above. Ideally, a combined system would be configured in such a way that wind turbine control is only triggered when a dispersal effort fails.

[0188] However, given the effective range of dispersal devices and the time required for shutdown this may not be practically feasible. Instead, the choice may be solely based on risk levels and / or the availability of shutdown budget.

[0189] In the present specification, the term "comprising" as used in the claims does not exclude other elements or steps. The term "a" or "an" as used in the claims does not exclude a plurality.

[0190] Although the present invention has been described in detail for purpose of illustration, it is understood that such detail is solely for that purpose, and variations can be made therein by those skilled in the art without departing from the scope of the invention.

Claims

CLAIMS1 . A method for controlling at least one wind turbine (1 ) to reduce risk of collision between at least one flying avian (3) and at least one wind turbine rotor blade (2), the method comprising obtaining a number of radar plots (6) corresponding to at least one avian (3) detected in an avian detection area (4) by an avian detection system comprising at least one avian radar (5); creating and storing a number of target tracks (7) based on the obtained radar plots (6), each target track (7) comprising target data (8) determined from radar plots (6) of corresponding avians (3); obtaining a number of environmental risk factors (9) affecting flying avian behavior in the avian detection area (4); determining a risk level (11 ) with respect to a possible collision between the at least one wind turbine (1 ) and at least one avian (3) using a risk model (10) based on target data (8) of the number of target tracks (7) and the number of environmental risk factors (9); and determining a control value (12) for at least one wind turbine (1 ) based on the risk level (11 ) and technological parameters (13) of the wind turbine (1 ), wherein the control value (12) defines a recommended change of blade velocity for the at least one wind turbine (1 ).

2. A method according to any preceding claim, wherein the control value (12) is a value between 0 and 100, wherein 0 corresponds to unimpeded turbine operation and 100 corresponds to a strong recommendation to completely stop the blades (2) of the wind turbine (1 ).

3. A method according to any preceding claim, wherein the control value (12) is determined using an economic model (14), wherein the risk level (11 ) andtechnological parameters (13) of the wind turbine (1 ) are inputs for the economic model (14).

4. A method according to claim 3, wherein the economic model (14) further takes as input economic factors (15) for determining the control value (12), the economic factors (15) comprising at least one of a minimum shutdown duration based on predictable power production and shutdown-related turbine wear-and- tear, and a predetermined budget corresponding to a remaining amount of allowable reduction of blade velocity.

5. A method according to any preceding claim, wherein the risk model (10) comprises a decision tree model (16) based on at least one ruleset, each ruleset comprising at least one rule.

6. A method according to claim 5, wherein the target data (8) determined from radar plots (6) comprises at least one of:- target classification;- target track altitude; and- speed, heading direction and sinuosity with respect to the at least one wind turbine (1 ); wherein the decision tree model (16) comprises a track filtering ruleset (17) for determining whether a target track (7) is a track of interest (18) based on at least one of the target data (8) fulfilling the conditions of a respective rule in the track filtering ruleset (17).

7. A method according to any one of claims 5 or 6, wherein the decision tree model (16) comprises an alarm ruleset (19) for determining whether and to what extent a wind turbine (1 ) presents a risk for collision with an avian (3) based on whether and how long the corresponding target track (7) is located within apredefined alarm area (20) and and / or whether the corresponding target track (7) crosses a predefined alarm line (21 ).

8. A method according to any one of claims 5 to 7, wherein the decision tree model (16) comprises a risk ruleset (22) for determining the level of risk a wind turbine (1 ) presents for a target track (7) based on at least one environmental risk factor (9), such as time of year, time of day, weather, visibility, and sun elevation is the avian detection area (4).

9. A method according to any preceding claim, wherein the method comprises calculating statistical data (24) from target data (8) on heading direction and sinuosity of the number of target tracks (7) for at least one predefined alarm area (20) or alarm line (21 ), the statistical data (24) comprising at least one of track density in the alarm area (20), migration traffic rate in the alarm area (20), and crossing rate of the alarm line (21 ); and wherein the risk model (10) further takes into account the statistical data (24) for determining the risk level (11 ).

10. A method according to claim 9, wherein the risk model (10) comprises a decision tree model (16) with an avian activity ruleset (23) for determining whether a target track (7) corresponds to local avian activity or migrating avian activity, based on the target data (8), and further based on the statistical data (24) on at least one of track density in the alarm area (20), migration traffic rate in the alarm area (20), or crossing rate of the alarm line (21 ).

11. A method according to any preceding claim, wherein the risk model (10) comprises a decision tree model (16) with a migration activity ruleset (37) for determining whether migration activity (26) has started within an alarm area (20) comprising grid cells (28), based on whether a number of grid cells (28) classified as migration activity (26) within the alarm area (20) is over a predefined migration activity threshold.

12. A method according to claim 11 , wherein the method comprises the steps of:1 ) defining a regular grid (27) over the alarm area (20), the regular grid (27) comprising grid cells (28),2) recording target track activity (44) over time in each grid cell (28),3) determining a track vector (29) for each target track (7) in each grid cell (28) at each point in time,4) calculating a cell vector (30) for each grid cell (28) by adding up the track vector (29) of each target track (7) in the grid cell (28) over time,5) normalizing the resulting cell vector (30) for the number of target tracks (7) to determine a normalized cell vector (31 ) for each grid cell (28), and6) classifying each grid cell (28) as local activity (25) or migration activity (26) based on the length of its respective normalized cell vector (31 ), wherein grid cells (28) with a normalized cell vector (31 ) below a predefined cell vector threshold are classified as local activity (25), and grid cells (28) with a normalized cell vector (31 ) over the predefined cell vector threshold are classified as migration activity (26).

13. A method according to claim 12, wherein classifying each grid cell (28) further comprises the steps of:7) locating busy cells (34) within the regular grid (27), wherein a grid cell (28) is defined as a busy cell (34) if its recorded target track (7) count is larger than a predefined cell activity threshold,8) clustering adjacent busy cells (34) into busy clusters (35),9) classifying each busy cluster (35) as local activity (25) or migration activity (26) by applying the steps 3 to 6 of claim 12, and10) applying the classification of each busy cluster (35) to all cells in the convex hull (36) of the respective busy cluster (35).

14. A method according to any one of claims 12 or 13, wherein classifying each grid cell (28) comprises assigning a time-dependent track vector weight (32) to each track vector (29) based on respective age, and wherein normalizing the cell vector (30) takes into account each track vector weight (32) to result in a smoothed cell vector (33).

15. A method according to any preceding claim, wherein determining the control value (12) is further based on directional migration activity (38) towards the at least one wind turbine (1 ), wherein determining the directional migration activity (38) comprises using a combination of temporal and spatial information by defining at least one probe line (39) in proximity of a wind turbine (1 ), the probe lines (39) being arranged substantially perpendicular to an expected migration direction towards the turbine; recording line crossing rate (40) of target tracks (7) crossing each probe line (39) over time; calculating a migration profile (41 ) corresponding to expected traffic over time by combining line crossing rates (40) from the probe lines (39); and estimating directional migration activity (38) towards the wind turbine (1 ) based on the migration profile (41 ).

16. A system for controlling at least one wind turbine (1 ) to reduce risk of collision between at least one flying avian (3) and at least one wind turbine rotor blade (2), the system comprising: at least one wind turbine (1 ); at least one avian radar (5) configured to detect avians (3) in an avian detection area (4) and return radar plots (6) corresponding to the avians (3); and one or more processors configured to obtain the radar plots (6);create and store a number of target tracks (7) based on the obtained radar plots (6), each target track (7) comprising target data (8) determined from radar plots (6) of corresponding avians (3); obtain a number of environmental risk factors (9) affecting flying avian behavior in the avian detection area (4); determine a risk level (11 ) with respect to a possible collision between the at least one wind turbine (1 ) and at least one avian (3) using a risk model (10) based on target data (8) of the number of target tracks (7) and the number of environmental risk factors (9); and determine a control value (12) for at least one wind turbine (1 ) based on the risk level (11 ) and technological parameters (13) of the wind turbine (1 ), wherein the control value (12) defines a recommended change of blade velocity for the at least one wind turbine (1 ).

17. A method for avian dispersal, the method comprising obtaining radar plots (6) corresponding to at least one avian (3) detected in an avian detection area (4) by an avian detection system comprising at least one avian radar (5); creating and storing a number of target tracks (7) based on the obtained radar plots (6), each target track (7) comprising target data (8) determined from radar plots (6) of corresponding avians (3); obtaining a number of environmental risk factors (9) affecting flying avian behavior in the avian detection area (4); determining a risk level (11 ) with respect to a possible collision between the at least one wind turbine (1 ) and at least one avian (3) using a risk model (10) based on target data (8) of the number of target tracks (7) and the number of environmental risk factors (9); andtriggering an avian dispersal event (43) of an avian dispersal system (42) based on whether the risk level (11 ) calculated for the avian detection area (4) exceeds an avian dispersal threshold.

18. A method according to claim 17, wherein the target data (8) determined from radar plots (6) comprises at least one of:- target classification;- target track altitude; and- speed, heading direction and sinuosity with respect to the at least one wind turbine (1 ); and wherein the risk model (10) comprises a decision tree model (16) with a track filtering ruleset (17) for determining whether a target track (7) is a track of interest (18), based on at least one of the target data (8) fulfilling the conditions of a respective rule in the track filtering ruleset (17).

19. A method according to any one of claims 17 or 18, wherein the risk model (10) comprises a decision tree model (16) with a risk ruleset (22) for determining the level of risk a wind turbine (1 ) presents for a target track (7) based on at least one environmental risk factor (9), such as time of year, time of day, weather, visibility, and sun elevation is the avian detection area (4).

20. A method according to any one of claims 17 to 19, wherein the method comprises calculating statistical data (24) from target data (8) on heading direction and sinuosity of the number of target tracks (7) for at least one predefined alarm area (20) or alarm line (21 ), the statistical data (24) comprising at least one of track density in the alarm area (20), migration traffic rate in the alarm area (20), and crossing rate of the alarm line (21 ); and wherein the risk model (10) comprises a decision tree model (16) with an avian activity ruleset (23) for determining whether a target track (7) corresponds to localavian activity or migrating avian activity, based on the target data (8), and further based on the statistical data (24) on at least one of track density in the alarm area(20), migration traffic rate in the alarm area (20), or crossing rate of the alarm line(21 ).21 . A method according to any one of claims 17 to 19, wherein the risk model (10) comprises a decision tree model (16) with a migration activity ruleset (37) for determining whether migration activity (26) has started within an alarm area (20) comprising grid cells (28), based on whether a number of grid cells (28) classified as migration activity (26) within the alarm area (20) is over a predefined migration activity threshold.

22. A method according to claim 21 , wherein the method comprises the steps of:1 ) defining a regular grid (27) over the alarm area (20), the regular grid (27) comprising grid cells (28),2) recording target track activity (44) over time in each grid cell (28),3) determining a track vector (29) for each target track (7) in each grid cell (28) at each point in time,4) calculating a cell vector (30) for each grid cell (28) by adding up the track vector (29) of each target track (7) in the grid cell (28) over time,5) normalizing the resulting cell vector (30) for the number of target tracks (7) to determine a normalized cell vector (31 ) for each grid cell (28), and6) classifying each grid cell (28) as local activity (25) or migration activity (26) based on the length of its respective normalized cell vector (31 ), wherein grid cells (28) with a normalized cell vector (31 ) below a predefined cell vector threshold are classified as local activity (25), and grid cells (28) with a normalized cell vector (31 ) over the predefined cell vector threshold are classified as migration activity (26).

23. A method according to claim 22, wherein classifying each grid cell (28) further comprises the steps of:7) locating busy cells (34) within the regular grid (27), wherein a grid cell (28) is defined as a busy cell (34) if its recorded target track (7) count is larger than a predefined cell activity threshold,8) clustering adjacent busy cells (34) into busy clusters (35),9) classifying each busy cluster (35) as local activity (25) or migration activity (26) by applying the steps 3 to 6 of claim 22, and10) applying the classification of each busy cluster (35) to all cells in the convex hull (36) of the respective busy cluster (35).

24. A method according to any one of claims 22 or 23, wherein classifying each grid cell (28) comprises assigning a time-dependent track vector weight (32) to each track vector (29) based on respective age, and wherein normalizing the cell vector (30) takes into account each track vector weight (32) to result in a smoothed cell vector (33).

25. A method according to any one of claims 17 to 24, wherein triggering an avian dispersal event (43) is further based on directional migration activity (38) towards the at least one wind turbine (1 ), wherein determining the directional migration activity (38) comprises using a combination of temporal and spatial information by defining at least one probe line (39) in proximity of a wind turbine (1 ), the probe lines (39) being arranged substantially perpendicular to an expected migration direction towards the turbine; recording line crossing rate (40) of target tracks (7) crossing each probe line (39) over time; calculating a migration profile (41 ) corresponding to expected traffic over time by combining line crossing rates (40) from the probe lines (39); andestimating directional migration activity (38) towards the wind turbine (1 ) based on the migration profile (41 ).

26. A system for avian dispersal, the system comprising: at least one wind turbine (1 ); at least one avian radar (5) configured to detect avians (3) in an avian detection area (4) and return radar plots (6) corresponding to the avians (3); an avian dispersal system (42); and one or more processors configured to obtain the radar plots (6); create and store a number of target tracks (7) based on the obtained radar plots (6), each target track (7) comprising target data (8) determined from radar plots (6) of corresponding avians (3); obtain a number of environmental risk factors (9) affecting flying avian behavior in the avian detection area (4); determine a risk level (11 ) with respect to a possible collision between the at least one wind turbine (1 ) and at least one avian (3) using a risk model (10) based on target data (8) of the number of target tracks (7) and the number of environmental risk factors (9); and trigger an avian dispersal event (43) of the avian dispersal system (42) based on whether the risk level (11 ) calculated for the avian detection area (4) exceeds an avian dispersal threshold.

Images