A method and apparatus for predicting wind turbine icing and assessing power output loss.

By acquiring meteorological, operational, and topographical variables of wind turbines, calculating topographic correction and component exposure coefficients, correcting icing prediction data, and combining the output data of adjacent wind turbines, the problem of icing prediction errors and misjudgments in complex mountainous wind farms was solved, improving the accuracy of icing probability and output loss assessment.

CN123115608APending Publication Date: 2026-11-17ELECTRIC POWER RES INST CHINA SOUTHERN POWER GRID CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611211927.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing wind turbine icing prediction schemes are unable to reflect the differences in icing risk and component icing exposure at different turbine locations within complex mountainous wind farms, leading to icing prediction errors. Furthermore, the operational data inversion process is susceptible to misjudgments due to factors such as power rationing and malfunctions.

Method used

By acquiring meteorological, operational, and topographic variables of the target wind turbine, the topographic correction coefficient and component exposure coefficient are calculated to correct the prior physical icing data. Combined with the output data of adjacent wind turbines, dual output deviation observation is performed, and physical and inverted icing thickness is integrated to reduce the impact of misjudgment.

Benefits of technology

It improves the accuracy of icing probability and output loss assessment, expresses the differences in icing risk and component icing exposure between complex mountainous sites, and reduces misjudgments caused by factors such as power outages and malfunctions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115608A_ABST
    Figure CN123115608A_ABST
Patent Text Reader

Abstract

This application relates to the field of wind turbine icing prediction technology, and provides a method and apparatus for wind turbine icing prediction and power loss assessment. The method includes: acquiring meteorological variables, operational variables, and topographic variables of a target wind turbine; calculating physical icing prior data based on the operational and meteorological variables; calculating the topographic correction coefficient of the target wind turbine based on the topographic variables; calculating the component exposure coefficient of the target wind turbine based on the operational variables; correcting the physical icing prior data based on the topographic correction coefficient and the component exposure coefficient to obtain the physical icing thickness; obtaining a dual power output deviation observation based on the power output data of multiple adjacent wind turbines; obtaining the inverted icing thickness based on the dual power output deviation observation and the operational variables; fusing the physical icing thickness and the inverted icing thickness to obtain an icing thickness fusion result; and obtaining the power loss rate and icing probability based on the icing thickness fusion result. This application can improve the accuracy of the final output icing probability and power loss assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of wind turbine icing prediction, and more particularly to a method and apparatus for predicting wind turbine icing and assessing power loss. Background Technology

[0002] Wind farms located in mountainous and high-altitude areas in mid-to-low latitude regions are often affected by a combination of factors during winter, including southward cold waves, warm and humid vapor transport, freezing rain, freezing fog, and wet snow. This makes wind turbine blades, nacelles, towers, and sensors prone to icing. Icing alters the aerodynamic shape of the blades, reduces wind energy capture efficiency, increases structural load and vibration risk, and in severe cases, leads to turbine shutdown, power limitation, or blade damage. Current wind turbine icing prediction methods typically include three categories: the first is the meteorological threshold method, which assesses icing risk based on thresholds such as temperature, humidity, wind speed, precipitation, or freezing rain forecasts; the second is the physical model method, which uses numerical weather prediction data to drive physical models such as the Makkonen icing model and freezing rain icing model to calculate icing growth or thickness; and the third is the data-driven method, which inputs meteorological variables and SCADA operational variables into a machine learning model to directly output icing probability, icing level, or power loss.

[0003] Inventors in this field have discovered that mountain wind farms in mid-to-low latitude regions have various micro-topographical features such as ridges, passes, windward slopes, leeward slopes, wind tunnel convergence zones, and high-altitude turning points. These features result in significant differences in temperature, humidity, wind conditions, and cloud / water supply capacity at different turbine locations within the same wind farm. Existing three types of icing prediction schemes, which use site-level meteorological station data or coarse-resolution numerical weather forecast data, are insufficient to reflect the local icing environment at the hub height and blade swept layer. They also fail to adequately express the differences in icing risk between complex mountain turbine locations and the differences in icing exposure of components such as turbine blades, nacelles, and towers, leading to icing prediction errors. At the same time, wind turbine operating data contains information on the response of icing to actual output. This information can be used to invert the current icing state and splice it with the icing prediction results for output. However, if non-icing factors such as power curtailment, faults, maintenance, abnormal yaw, and communication anomalies are not excluded, misjudgments can easily occur during the inversion process, ultimately leading to repeated propagation of errors after splicing. Summary of the Invention

[0004] This application provides a method and apparatus for predicting wind turbine icing and assessing power loss. It can express the differences in icing risk between complex mountain turbine locations and the differences in icing exposure of components such as turbine blades, nacelles, and towers in the icing prediction results. At the same time, it reduces the misjudgment of operation data inversion caused by power curtailment, faults, and local wind condition differences, and improves the accuracy of the final output icing probability and power loss assessment.

[0005] In a first aspect, embodiments of this application provide a method for predicting wind turbine icing and assessing power output loss, including:

[0006] Obtain the meteorological variables, operational variables, and terrain variables of the target wind turbine;

[0007] Calculate prior data on physical icing based on operational and meteorological variables;

[0008] Calculate the terrain correction factor for the target wind turbine based on terrain variables;

[0009] Calculate the component exposure coefficient of the target wind turbine based on the operating variables;

[0010] The physical icing thickness of the target wind turbine is obtained by correcting the prior data of physical icing based on the terrain correction coefficient and component exposure coefficient.

[0011] A dual output deviation observation was obtained based on the output data of multiple adjacent wind turbines;

[0012] The inverted ice thickness was obtained based on the dual output deviation observation and operating variables.

[0013] By fusing the physical icing thickness and the inverted icing thickness, the icing thickness fusion result is obtained;

[0014] The power loss rate and icing probability are obtained based on the fusion results of icing thickness.

[0015] Optionally, the method further includes:

[0016] After obtaining the power loss rate and icing probability, the risk level is obtained by querying the risk level table based on the icing thickness fusion result, power loss rate, and icing probability.

[0017] Optionally, the method further includes:

[0018] Quality control was performed on the acquired meteorological variables, operational variables, and topographic variables;

[0019] Meteorological variables are interpolated based on topographic variables, and the interpolated meteorological variables are spatiotemporally registered according to the location of the target wind turbine and the preset time step.

[0020] Optionally, the dual output deviation observations include the relative air density corrected power curve deviation of the target fan and the deviation relative to the nearest reference fan;

[0021] The above-mentioned dual output deviation observation, obtained based on the output data of multiple adjacent wind turbines, includes:

[0022] Multiple adjacent wind turbines are filtered according to preset filtering rules to obtain at least one reference wind turbine;

[0023] The weighted output of the neighboring reference wind turbines is obtained based on the output data of the reference wind turbines;

[0024] The relative air density correction power curve deviation and the relative adjacent reference fan deviation are calculated based on the target fan's theoretical reference power, actual active power, and weighted output of the adjacent reference fan.

[0025] Optionally, the above-mentioned fusion of physical icing thickness and inverted icing thickness to obtain icing thickness fusion result includes: fusing physical icing thickness and inverted icing thickness based on a preset fusion rule based on data quality weights to obtain icing thickness fusion result. The data quality weights are obtained based on numerical weather forecast confidence, real-time meteorological data quality, operational data quality, and operational observation reliability.

[0026] Optionally, the above-mentioned output loss rate and icing probability obtained based on the fusion results of icing thickness include:

[0027] By inputting the icing thickness fusion result, the target wind turbine's operating variables, and the current wind speed into the trained residual correction model, the power loss correction amount is obtained.

[0028] The power loss is obtained by adding the icing thickness fusion result, the target wind turbine's operating variables, and the base power loss corresponding to the current wind speed to the power loss correction.

[0029] Obtain the target wind turbine's ice-free reference power under operating variables;

[0030] The output loss rate is obtained based on the power loss and the reference power without icing.

[0031] Optionally, the above-mentioned method of obtaining the power loss rate and icing probability based on the icing thickness fusion result also includes: obtaining the icing probability through a trained residual correction model, or inputting the icing thickness fusion result, the dual power output deviation observation, the icing meteorological triggering index obtained from meteorological variables, and the power output deviation duration of the target wind turbine into a probability mapping function to obtain the icing probability.

[0032] Secondly, another embodiment of this application provides a device for predicting wind turbine icing and assessing output loss, comprising:

[0033] The acquisition module is used to acquire meteorological variables, operational variables, and terrain variables of the target wind turbine;

[0034] The prior module is used to calculate the prior data of physical icing based on the operating variables and meteorological variables;

[0035] The terrain component correction module is used to calculate the terrain correction coefficient of the target wind turbine based on terrain variables; and to calculate the component exposure coefficient of the target wind turbine based on operating variables.

[0036] The correction module is used to correct the prior data of physical icing based on the terrain correction coefficient and the component exposure coefficient to obtain the physical icing thickness of the target wind turbine.

[0037] The observation module is used to obtain the dual output deviation observation based on the output data of multiple adjacent wind turbines;

[0038] The inversion fusion module is used to obtain the inverted icing thickness based on the dual output deviation observations and operating variables; and to fuse the physical icing thickness and the inverted icing thickness to obtain the fused icing thickness result.

[0039] The output module is used to obtain the output loss rate and icing probability based on the fusion results of the icing thickness.

[0040] In another aspect, this application provides a computer device including at least one connected processor, memory, and transceiver, wherein the memory is used to store program code, and the processor is used to call the program code in the memory to execute the methods described in the above aspects.

[0041] In another aspect, this application provides a computer storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the above aspects.

[0042] Compared with existing technologies, the wind turbine icing prediction and output loss assessment method provided in this application has the following beneficial effects: First, this application uses terrain correction coefficients calculated from terrain variables to express the icing differences between complex mountain turbine locations, and uses component exposure coefficients calculated from operating variables to express the influence of speed, yaw, pitch angle, blade orientation, and component geometric differences on icing. By using terrain correction coefficients and component exposure coefficients to correct the physical icing thickness obtained from prior physical icing data, it can express the differences in icing risk between complex mountain turbine locations and the differences in icing exposure of components such as wind turbine blades, nacelles, and towers. Second, the dual output deviation observations obtained from the output data of multiple adjacent wind turbines are introduced into the inversion process of operating variables, reducing misjudgments caused by power curtailment, faults, and local wind condition differences. Finally, the mutually independent physical icing thickness and inverted icing thickness are fused and output, avoiding the repeated propagation of prediction errors and inversion misjudgments in the final fusion, thus improving the accuracy of output loss rate and icing probability. Attached Figure Description

[0043] Figure 1 This is a flowchart of a method for predicting wind turbine icing and assessing power output loss according to this application;

[0044] Figure 2 This is a structural diagram of a wind turbine icing prediction and output loss assessment device according to this application; Detailed Implementation

[0045] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein.

[0046] Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The module divisions appearing in the embodiments of this application are merely logical divisions; in actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual couplings or direct couplings or communication connections may be through some interfaces, and the indirect couplings or communication connections between modules may be electrical or other similar forms, none of which are limited in the embodiments of this application. Moreover, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0047] Due to the diverse micro-topography of mountain wind farms in mid- and low-latitude regions, including ridges, passes, windward slopes, leeward slopes, wind tunnel convergence zones, and high-altitude transition sections, the temperature, humidity, wind conditions, and cloud / water supply capacity vary significantly among different turbine locations within the same wind farm. Existing icing prediction schemes, which utilize site-level meteorological station data or coarse-resolution numerical weather prediction data, struggle to reflect the local icing environment at hub height and blade swept layer. They also fail to adequately express the differences in icing risk between complex mountain turbine locations and the differences in icing exposure of components such as turbine blades, nacelles, and towers, leading to icing prediction errors. Simultaneously, wind turbine operating data contains information on the response of icing to actual output. This information can be used to invert the current icing state and stitch it together with the icing prediction results. However, if non-icing factors such as power curtailment, faults, maintenance, abnormal yaw, and communication anomalies are not excluded, misjudgments can easily occur during the inversion process, ultimately leading to repeated propagation of errors after stitching.

[0048] To resolve the above issues, please refer to [link / reference]. Figure 1 This application provides a method for predicting wind turbine icing and assessing power output loss, mainly including:

[0049] Step S1: Obtain the meteorological variables, operational variables, and terrain variables of the target wind turbine.

[0050] Meteorological variables include numerical weather prediction variables and actual meteorological variables. Numerical weather prediction variables can be output by WRF (Mesoscale Weather Research and Forecasting Model) or equivalent mesoscale models, and at least include temperature, relative humidity, wind speed, wind direction, precipitation, cloud-water mixing ratio, cloud-ice mixing ratio, and air pressure.

[0051] Operating variables include SCADA (Supervisory Control and Data Acquisition) operating variables, including at least actual output, turbine status code, speed, pitch angle, yaw angle, nacelle wind speed, and nacelle wind direction.

[0052] Terrain variables include the altitude, slope, aspect, neighborhood elevation, width of the upstream terrain cross section, and shading angle of the incoming flow direction of the DEM (Digital Elevation Model).

[0053] Optionally, the method further includes:

[0054] Step S11: Perform quality control on the acquired meteorological variables, operational variables, and topographic variables.

[0055] Specifically, quality control is performed on various data variables to remove missing measurements, abnormal negative values, temperature, humidity and wind records that exceed physical limits, communication interruption records, and obviously abnormal power records.

[0056] Step S12: Interpolate meteorological variables based on topographic variables, and perform spatiotemporal registration of the interpolated meteorological variables according to the location of the target wind turbine and the preset time step.

[0057] Specifically, meteorological variables are first interpolated to the hub height based on the model layer height, model terrain height, turbine location DEM elevation, hub height, and blade radius, or interpolated to the upper, middle, and lower heights of the blade swept layer and then weighted averaged. Then, spatiotemporal registration is performed according to the turbine location and preset time step.

[0058] Step S2: Calculate the prior data of physical icing based on the operating variables and meteorological variables.

[0059] For the target wind turbine, calculate the freezing rain icing increase ΔH. fr ΔH, the increase in rime ice formation rime and the increase in wet snow icing ΔH wet And calculate the ablation amount ΔH melt and shedding amount ΔH shed .

[0060] Prior data for physical icing are derived using the following formula:

[0061] H t =max[0,H_{t-1}+ΔH fr +ΔH rime+ΔH wet -ΔH melt -ΔH shed ]

[0062] The increase in ice cover due to freezing rain, rime, and wet snow is mainly calculated from the registered meteorological variables; the ablation is determined by air temperature or wet-bulb temperature, duration of positive temperature, radiation, and wind speed; the amount of detachment is related to the existing ice thickness and can also use operational information such as speed recovery, sudden wind speed increase, blade vibration, yaw change, and pitch angle change; the ice thickness H_{t-1} at the previous moment is a state recursive quantity. Specifically, ΔH fr Determined based on precipitation phase, precipitation per unit time, wet-bulb temperature, wind speed, water density, ice density, and rime density; ΔH rime Determined based on liquid water content, cloud droplet diameter, wind speed, component equivalent radius, collision rate, capture rate, and freezing rate; ΔH wet ΔH is determined based on wet snow particle mass concentration, wind speed, terminal velocity of falling snow particles, capture rate, and snow density. melt Determined based on one or more of the following: duration of positive temperature, wet-bulb temperature, solar radiation, longwave radiation, and wind speed; ΔH shed It is determined based on one or more of the following: sudden increase in wind speed, speed recovery, blade vibration index, yaw change, pitch angle change, and icing thickness threshold.

[0063] Liquid water content (LWC) can be calculated from the cloud-water mixing ratio (WRF) combined with air density, or, when conditions permit, provided by cloud droplet or liquid water content monitoring equipment; cloud droplet diameter (MVD) can be calculated from liquid water content and microphysical spectrum parameters, or obtained from a cloud droplet spectrometer. Furthermore, hub height, impeller radius, blade leading edge equivalent radius, nacelle / tower characteristic dimensions, rated power, and icing-free power curves are static technical parameters of the wind turbine and can be obtained from wind turbine technical manuals, design documents, or equipment ledgers.

[0064] Step S3: Calculate the terrain correction factor for the target wind turbine based on the terrain variables.

[0065] Specifically, the terrain correction factor K terrain Used to characterize the amplifying or attenuating effect of complex mountain locations on icing growth. K terrain Normalization factor I based on altitude elev Windward slope index I windward 1. Air duct convergence index I conv Ridge Exposure Index I ridge and leeward shading index I shelter At least two of the following can be obtained.

[0066] As a more stable implementation method, K terrainIt should include at least the windward slope index and the wind channel convergence index, or at least the windward slope index, the ridge exposure index, and the leeward shielding index.

[0067] K terrain =clip(1+a*I elev +b*I windward +c*I conv +d*I ridge -e*I shelter ,K min ,K max )

[0068] I elev =clip((Z turbine -Z mean ) / max(Z std ,ε),-1,1)

[0069] I windward =max[0,cos(θ aspect - θ wind )]

[0070] I conv =clip((W up -W site ) / max(W up ,ε),0,1)

[0071] I ridge =clip((Z turbine -Z neighmean ) / max(Z neighstd ,ε),0,1)

[0072] I shelter =clip(sin(α shelter ),0,1)

[0073]

[0074] Among them, Z turbine For the target wind turbine location elevation, Z mean and Z std These are the mean and standard deviation of the altitude of the site or adjacent airfields, respectively; θ wind θ is the incoming flow direction angle, derived from the WRF wind direction after hub height correction, the wind direction from the meteorological tower / automatic weather station, or the SCADA nacelle wind direction controlled by quality control. aspect W is the windward normal direction angle of the slope. up W is the effective width of the cross-section along the contour line or valley at a preset distance upstream of the aircraft position. site Z is the effective width of the cross-section at the machine position; neighmean and Z neighstdThese are the mean and standard deviation of the DEM elevation of the aircraft station's neighborhood, respectively; α shelter The maximum terrain shielding angle is preset within the search distance in the direction of the incoming flow; clip indicates the upper and lower limits for truncation, and ε is a positive number to prevent the denominator from being zero. Z turbine Z mean Z std Slope aspect, W site Z neighmean Z neighstd and α shelter All can be calculated from the DEM, wind turbine coordinates, and a preset neighborhood / upstream search window. a, b, c, d, and e are terrain influence coefficients, which can be constrained to be non-negative; in one embodiment, K can be taken as... min =0.5, K max =3.0.

[0075] Step S4: Calculate the component exposure coefficient of the target wind turbine based on the operating variables.

[0076] Specifically, the component exposure coefficient K part Used to characterize the impact of wind turbine operating status and geometric differences of different components on icing capture efficiency, ice cross-section or thickness conversion. K part The indicator R is determined by the shutdown or low speed. stop yaw error normalization quantity Y err Pitch angle icing response quantity B pitch Leaf azimuth exposure B azimuth And the component equivalent radius correction amount R eq At least two of the constructions are obtained. As a more stable implementation, K... part This includes at least the low-speed indication, blade orientation exposure, and component equivalent radius correction.

[0077] K part =clip(1+m1*R stop +m2*Y err +m3*B pitch +m4*B azimuth +m5*R eq ,P min ,P max )

[0078] Among them, R stop Used to characterize the localized, continuous upwind exposure effect of blades caused by low speed or shutdown, calculated from SCADA speed and operating status code; Y err The deviation of the frontal surface caused by yaw error is used to characterize the difference between the cabin orientation / yaw angle and the direction of the incoming wind at the same time; B pitch Used to characterize the change in the windward projection of the blade leading edge due to the pitch angle, calculated from the SCADA pitch angle using a preset response function; B azimuthTo characterize differences in blade azimuth exposure, when the fan speed is below the speed threshold, it is determined based on the cosine response of the angle between the blade's stationary azimuth and the incoming flow direction (i.e., the blade azimuth angle). If there is no direct azimuth angle, it can be estimated based on the stationary azimuth / speed phase record under low speed or shutdown conditions. When the fan speed is above or equal to the speed threshold, B... azimuth Take the time average of the azimuth response within one rotation cycle; R eq This refers to the geometric correction values ​​for wind turbine components, used to characterize the equivalent radius corrections for different components such as the blade leading edge, nacelle, or tower. It does not originate from real-time SCADA; its original equivalent radius or leading edge curvature radius should be obtained from the wind turbine technical manual, blade / nacelle / tower design data, equipment ledger, or CAD geometric parameters, and then normalized according to the radius of the reference component. m1-m5 are component exposure influence coefficients, which can be constrained to be non-negative and controlled via P... min P max For K part Cut off.

[0079] Step S5: Correct the prior physical icing data based on the terrain correction coefficient and component exposure coefficient to obtain the physical icing thickness of the target wind turbine. Specifically, correct the prior physical icing data H... phy,corr Multiplicative correction, additive correction, or piecewise correction methods can be used. A preferred implementation is as follows:

[0080] H phy,corr =H t *K terrain *K part

[0081] Step S6: Obtain the dual output deviation observation based on the output data of multiple adjacent wind turbines.

[0082] Among them, the dual output deviation observation includes the relative air density corrected power curve deviation ΔP of the target wind turbine. curve and the deviation ΔP of the relative neighboring reference fan neighbor .

[0083] The above-mentioned dual output deviation observation, obtained based on the output data of multiple adjacent wind turbines, includes:

[0084] Step S61: Select multiple adjacent wind turbines according to preset screening rules to obtain at least one reference wind turbine. Specifically, before constructing the dual output deviation observation, samples with power outages, faults, maintenance, communication anomalies, and non-icing power anomalies need to be eliminated. Therefore, the reference wind turbine must meet the following conditions: it is not currently triggering the low temperature and high humidity and dual output deviation icing conditions. It must also meet at least two of the following preset screening rules: distance from the target wind turbine is less than a preset distance, altitude difference is less than a preset threshold, wind direction angle is less than a preset threshold, wind speed difference is less than a preset threshold, and power correlation coefficient over a preset time period is higher than a correlation threshold.

[0085] Step S62: Obtain the weighted output of the neighboring reference wind turbine based on the output data of the reference wind turbine.

[0086] Weighted output P of the nearest reference wind turbine ref is the weighted average of the output of each reference wind turbine. The weights are positively correlated with the historical power correlation coefficient and negatively correlated with one or more of the following: turbine distance, altitude difference, wind direction angle difference, and wind speed difference. A preferred calculation formula is:

[0087] P ref =sum(w i *P i ) / sum(w i )

[0088] w i =ρ i / [(d i +ε)*(1+abs(ΔZ i ))*(1+abs(ΔΘ i ))]

[0089] Among them, P i For the output data of the i-th reference wind turbine, w i Let ρ be the weight of the i-th reference wind turbine. i Let d be the historical power correlation coefficient between the target wind turbine and the i-th reference wind turbine. i Let ΔZ be the distance between the target wind turbine and the i-th reference wind turbine. i Let ΔΘ be the altitude difference between the target wind turbine and the i-th reference wind turbine. i Let ε be the angle difference between the wind direction of the target fan and the i-th reference fan, and let ε be a positive number to prevent the denominator from being zero.

[0090] Step S63: Calculate the relative air density correction power curve deviation and the relative adjacent reference fan deviation based on the target fan's theoretical reference power, actual active power, and weighted output of the adjacent reference fan.

[0091] In the specific implementation process, the calculation of the dual output deviation observation is as follows:

[0092] ΔP curve =(P curve -P obs ) / max(P curve ,ε)

[0093] ΔP neighbor =(P ref -P obs ) / max(P ref ,ε)

[0094] Among them, P curve This represents the theoretical reference power of the target wind turbine under current hub height wind speed, air density, and comparable operating conditions, obtained from the non-icing power curve. The unit is kW or MW. Its calculation should be corrected for air density and can be adapted according to conditions such as blade pitch and rated power range. P obs This represents the actual active power of the target wind turbine as recorded by SCADA at the same moment, with units equal to P. curve Consistent.

[0095] Step S7: Obtain the inverted icing thickness based on the dual output deviation observation and operating variables.

[0096] Inverting the ice thickness H inv Primarily derived from dual-output bias observations, this method incorporates operational and meteorological variables during the inversion process, but excludes a correction term for physical icing thickness. This is to prevent physical prediction errors from propagating repeatedly during the inversion of icing thickness and subsequent fusion. A robust implementation method is as follows:

[0097] H inv =max[0,c0+c1*ΔP curve +c2*ΔP neighbor +c3*ln(1+D dur )+c4*I ice ]

[0098] I ice =f t (Tw)*f RH (RH)*f LWC (LWC)*f U (U)

[0099] Among them, I ice It is a dimensionless icing weather triggering index derived from meteorological variables, with a value range of 0-1; D durThe duration of the output deviation is obtained from the time series statistics of the continuous satisfaction of the dual output deviation conditions; Tw is the wet-bulb temperature, calculated from air temperature, relative humidity, and air pressure; RH is the relative humidity, obtained from the WRF, weather station, or anemometer tower; LWC is the liquid water content, which can be converted from the cloud-water mixing ratio of the WRF to the air density, or obtained from on-site liquid water content observation; U is the wind speed, which can be obtained from the height-corrected WRF, anemometer tower, or the SCADA nacelle wind speed controlled by quality; f t f RH f LWC and f U These are the collision response functions for wet-bulb temperature, relative humidity, liquid water content, and wind speed, respectively, with values ​​ranging from 0 to 1. c1-c4 correspond to output deviation, duration, and meteorological triggering amount. c1-c4 can be constrained to be non-negative, and c0 can be a signed intercept. Calibration is performed using non-negative constrained regression, ridge regression, or minimizing the thickness error using historical samples.

[0100] Step S8: Fuse the physical icing thickness and the inverted icing thickness to obtain the fused icing thickness result.

[0101] Specifically, the above-mentioned fusion of physical icing thickness and inverted icing thickness to obtain the icing thickness fusion result includes: fusing physical icing thickness and inverted icing thickness based on a preset fusion rule using data quality weights to obtain the icing thickness fusion result. The data quality weights are obtained based on numerical weather forecast confidence, real-time meteorological data quality, operational data quality, and operational observation reliability.

[0102] H final =λ*H phy,corr +(1-λ)*H inv

[0103] Where λ is the data quality weight, which can be calculated using the following formula:

[0104] λ=clip(A / (A+B+ε),0,1)

[0105] A=w1*Q nwp +w2*Q met

[0106] B=w3*Q scada +w4*Q run

[0107] w1-w4 are non-negative confidence weights, which can be normalized individually or uniformly satisfy w1+w2+w3+w4=1. Q nwp Q represents the confidence level of numerical weather prediction. met To improve the quality of real-time meteorological data, Q scada For the quality of SCADA operational data, Q runTo represent the reliability of operational-side observations, values ​​are taken from 0 to 1; Q run The determination is based on the persistence of dual output deviation, the availability of the reference wind turbine, the non-power-limited and non-faulty state of the target wind turbine, and the stability of the power curve deviation.

[0108] Step S9: Based on the icing thickness fusion results, obtain the output loss rate and icing probability.

[0109] The above-mentioned output loss rate and icing probability, obtained from the fusion results based on icing thickness, include:

[0110] Step S91: Input the icing thickness fusion result, the target wind turbine's operating variables, and the current wind speed into the trained residual correction model to obtain the power loss correction amount.

[0111] For training the residual correction model, multiple input samples and corresponding sample labels can be constructed. The input samples include the running variable S. State Wind speed U and icing thickness H observed .

[0112] Sample label ΔP label The observed power loss ΔP of the target wind turbine under normal conditions loss,observed The difference between the base power loss of the target wind turbine and the base power loss of the target wind turbine under the input sample:

[0113] ΔP label =ΔP loss,observed -F(H observed ,U, S State )

[0114] Among them, H observed The input sample contains the ice thickness in mm, which can be obtained from icing sensors, manual / shutdown inspection records, image recognition, pre- and post-de-icing inspections, or other verified field records.

[0115] ΔP loss,observed To observe the power loss, the unit is kW or MW. It can be calculated as the difference between the reference power without icing and the actual power under comparable conditions such as no power curtailment and no faults.

[0116] F is the base power loss function for calculating base power loss, satisfying F=P clean -P ice .

[0117] A residual correction model is trained using each input sample and its corresponding label. After verifying model convergence, the icing thickness fusion result, the target wind turbine's operating variables, and the current wind speed are input into the trained residual correction model to obtain the power loss correction ΔP output by the model. corr .

[0118] Step S92: Add the icing thickness fusion result, the target wind turbine's operating variables, and the base power loss corresponding to the current wind speed to the power loss correction to obtain the power loss.

[0119] Step S93: Obtain the reference power of the target wind turbine without icing under the operating variables.

[0120] Step S94: Obtain the output loss rate based on the power loss and the reference power without icing.

[0121] ΔP loss =F(H final ,U, S State )+ΔP corr

[0122] F=P clean (U, S State )-P ice (U, H final , S State )

[0123] R loss =ΔP loss / max(P clean ,ε)

[0124] Wherein, ΔP loss R represents power loss, expressed in kW or MW. loss S represents the output loss rate, which is the ratio of power loss to the reference power without icing. State This represents one or more of the following operating variables: turbine status code, speed, pitch angle, yaw angle, power limit status, shutdown status, fault status, maintenance status, and communication status. clean (U,S state ) represents the wind speed U and the running variable S state Reference power without icing; P ice (U,H final ,S state ) indicates that under the same conditions and with an ice thickness of H final Power under icing conditions.

[0125] In another embodiment, the residual correction model can be trained by adding input samples and sample labels related to physical icing thickness, so that it simultaneously learns the physical icing thickness residual and the power loss residual, and outputs the physical icing thickness correction ΔH along with the power loss correction. corr (When fusing physical icing thickness and inverted icing thickness using a preset fusion rule based on data quality weights, ΔH) corr If the default value is 0, then the physical icing thickness and the inverted icing thickness are fused to obtain the fused icing thickness result as follows:

[0126] H final =λ*H phy,corr +(1-λ)*H inv +ΔH corr

[0127] Furthermore, the above-mentioned output loss rate and icing probability obtained based on the icing thickness fusion results also include: obtaining the icing probability through a trained residual correction model, or inputting the icing thickness fusion results, dual output deviation observations, icing meteorological triggering index obtained from meteorological variables, and the output deviation duration of the target wind turbine into a probability mapping function to obtain the icing probability.

[0128] It should be noted that the residual correction model can be trained using a multi-task structure. After inputting data such as the fusion results of icing thickness, the regression output of the residual correction model gives ΔH. corr and / or ΔP corr The classification output is given by the sigmoid function, which provides the ice accumulation probability P. iceprob .

[0129] If the residual correction model is not trained using a multi-task structure, the icing probability can be obtained from the icing thickness fusion result, the dual output deviation observation, and the icing meteorological triggering index through a probability mapping function.

[0130] Specifically, when the residual correction model is not used, the icing probability P iceprob Based on the probability mapping function according to H final ΔP curve ΔP neighbor I ice and D dur The calculation yielded:

[0131] P iceprob =1 / [1+exp(-S)]

[0132] S = r0 + r1 * H norm +r2*ΔP curve +r3*ΔP neighbor +r4*I ice +r5*D norm

[0133] Among them, I ice H is a dimensionless icing weather triggering index. norm D represents the fusion result of normalized or standardized icing thickness. normThe normalized or standardized output deviation duration is represented by r0-r5, which are probability mapping coefficients. r0 is the intercept, and r1-r5 are constrained to be non-negative under the definition that all existing variables are positively correlated with icing risk. They are obtained through training using logistic regression or cross-entropy loss.

[0134] Compared with existing technologies, the wind turbine icing prediction and output loss assessment method provided in this application has the following beneficial effects: First, this application uses terrain correction coefficients calculated from terrain variables to express the icing differences between complex mountain turbine locations, and uses component exposure coefficients calculated from operating variables to express the influence of speed, yaw, pitch angle, blade orientation, and component geometric differences on icing. By using terrain correction coefficients and component exposure coefficients to correct the physical icing thickness obtained from prior physical icing data, it can express the differences in icing risk between complex mountain turbine locations and the differences in icing exposure of components such as wind turbine blades, nacelles, and towers. Second, the dual output deviation observations obtained from the output data of multiple adjacent wind turbines are introduced into the inversion process of operating variables, reducing misjudgments caused by power curtailment, faults, and local wind condition differences. Finally, the mutually independent physical icing thickness and inverted icing thickness are fused and output, avoiding the repeated propagation of prediction errors and inversion misjudgments in the final fusion, thus improving the accuracy of output loss rate and icing probability.

[0135] In a historical sample validation example, manual inspection records of icing, shutdown inspection records, and SCADA power anomaly records were used as control samples to compare the meteorological threshold method, the pure physical model, the physical plus terrain correction model, and the method of this application. The icing identification evaluation indicators in the table below can be accuracy, recall, F1 score, or a combination thereof; accuracy is used as an example here.

[0136]

[0137] In some embodiments, the method may further include:

[0138] After obtaining the power loss rate and icing probability, the risk level is obtained by querying the risk level table based on the icing thickness fusion result, power loss rate, and icing probability.

[0139] Specifically, please refer to the table below. In the risk level table, each icing risk level is set with an icing probability threshold, an icing thickness threshold, and an output loss rate threshold. When the three indicators are inconsistent, the final level is determined by the rule that at least two of them reach the same level, or the strictest rule is used to output the highest risk level.

[0140]

[0141] The following engineering calculation example illustrates the application of this application in practical work:

[0142] Taking the L3-05 wind turbine in a mountain wind farm as an example, at a certain moment, the hub temperature is -2.1 degrees Celsius, the wet-bulb temperature is -0.4 degrees Celsius, the relative humidity is 96%, the wind speed is 8.2 m / s, the liquid water content is 0.18 g / m³, and the median volume diameter of cloud droplets is 18 micrometers. The physical model calculates the hourly freezing rain icing increase as 0.8 mm, rime icing increase as 1.6 mm, and wet snow icing increase as 0.2 mm, with 0 mm of ablation and 0 mm of detachment. The ice thickness at the previous moment was 3.0 mm. Therefore, H... t =5.6mm.

[0143] The wind turbine is located 180m above the average elevation of the site, with a windward slope index of 0.82, a wind tunnel convergence index of 0.30, a ridge exposure index of 0.45, and a leeward shading index of 0.05. K was obtained through calibration using historical samples. terrain =1.23; During this period, the wind turbine was operating at low speed and the yaw error was small, so the component exposure coefficient K was 1.23. part =1.08, then H phy,cor r = 5.6 * 1.23 * 1.08 = 7.44 mm. After excluding power outages and faults, P... curve For 1800kW, P obs For 1250kW, P ref If the power is 1700kW, then ΔP curve The value is 0.306, ΔP neighbor H is 0.265. inv The value is 10.8 mm. The residual correction model output ΔH corr =0.9mm, λ=0.45, then H final =10.19mm. If P clean The power output is 1800kW, with a power loss of 324kW at F, and ΔP. corr If the power is 54kW, then ΔP loss For 378kW, R loss =21%.

[0144] Several steps in the above embodiments of this application can be implemented in alternative ways. For example, numerical weather prediction data can be output from the WRF model, or other mesoscale models, ensemble forecasts, reanalysis data, or short-term forecast data from field stations. The physical icing model used to calculate the prior physical icing data can be the Makkonen model, Jones model, freezing rain icing model, rime icing model, wet snow icing model, or equivalent models. The terrain correction coefficient can also be calculated using linear weighting, piecewise functions, lookup tables, or nonlinear functions calibrated from historical samples. The component exposure coefficient can be set separately for different objects such as blades, nacelles, towers, or sensors.

[0145] The inverted icing thickness can be calculated using linear mapping, piecewise mapping, logistic regression, tree models, or other regression models. The residual correction model is not limited to random forests, gradient boosting trees, neural networks, support vector machines, or Bayesian models; its key lies in learning the physical icing thickness residual and the power loss residual. If the fusion output of the physical icing thickness and the inverted icing thickness does not use a residual correction model, a preset fusion rule based on data quality weights can also be used to complete the fusion of the physical icing thickness and the inverted icing thickness.

[0146] Please refer to Figure 2 Another embodiment of this application provides a device for predicting wind turbine icing and assessing power output loss, which mainly includes the following modules:

[0147] The acquisition module 101 is used to acquire the meteorological variables, operating variables and terrain variables of the target wind turbine.

[0148] The prior module 102 is used to calculate the prior data of physical icing based on the operating variables and meteorological variables.

[0149] The terrain component correction module 103 is used to calculate the terrain correction coefficient of the target wind turbine based on terrain variables and to calculate the component exposure coefficient of the target wind turbine based on operating variables.

[0150] The correction module 104 is used to correct the prior data of physical icing based on the terrain correction coefficient and the component exposure coefficient to obtain the physical icing thickness of the target wind turbine.

[0151] The observation module 105 is used to obtain a dual output deviation observation based on the output data of multiple adjacent wind turbines.

[0152] The inversion fusion module 106 is used to obtain the inverted icing thickness based on the dual output deviation observation and operating variables; and to fuse the physical icing thickness and the inverted icing thickness to obtain the fused icing thickness result.

[0153] Output module 107 is used to obtain the output loss rate and icing probability based on the fusion results of icing thickness.

[0154] In some embodiments, the device may further include:

[0155] The risk level assessment module is used to query the risk level table based on the icing thickness fusion result, output loss rate, and icing probability after obtaining the output loss rate and icing probability, and obtain the icing risk level.

[0156] In some embodiments, the device may further include:

[0157] The quality control module is used to perform quality control on the meteorological variables, operational variables and topographic variables acquired by the acquisition module 101; it interpolates the meteorological variables based on the topographic variables, and performs spatiotemporal registration of the interpolated meteorological variables according to the location of the target wind turbine and the preset time step.

[0158] The specific limitations of the wind turbine icing prediction and power output loss assessment device provided in this embodiment can be found in the embodiment of the wind turbine icing prediction and power output loss assessment method described above, and will not be repeated here. Each module in the above-described wind turbine icing prediction and power output loss assessment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0159] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of a wind turbine icing prediction and output loss assessment method as described in any of the above embodiments. The working process, details, and technical effects of the computer device provided in this embodiment can be found in the above embodiments regarding a wind turbine icing prediction and output loss assessment method, and will not be repeated here.

[0160] This application provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the steps of a wind turbine icing prediction and output loss assessment method as described in any of the above embodiments. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The working process, details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiments of a wind turbine icing prediction and output loss assessment method described above, and will not be repeated here.

[0161] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0162] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for predicting wind turbine icing and assessing power output loss, characterized in that, include: Obtain the meteorological variables, operational variables, and terrain variables of the target wind turbine; Calculate prior data on physical icing based on operational and meteorological variables; Calculate the terrain correction factor for the target wind turbine based on the terrain variables; Calculate the component exposure coefficient of the target wind turbine based on the operating variables; The physical icing thickness of the target wind turbine is obtained by correcting the prior physical icing data based on the terrain correction coefficient and the component exposure coefficient. A dual output deviation observation was obtained based on the output data of multiple adjacent wind turbines; The inverted icing thickness is obtained based on the dual output deviation observation and the operating variables. By fusing the physical icing thickness and the inverted icing thickness, the icing thickness fusion result is obtained; The power loss rate and icing probability are obtained based on the fusion results of the icing thickness.

2. The method for predicting wind turbine icing and assessing power loss according to claim 1, characterized in that, Also includes: After obtaining the power loss rate and icing probability, the risk level is obtained by querying the risk level table based on the icing thickness fusion result, the power loss rate, and the icing probability.

3. The method for predicting wind turbine icing and assessing power loss according to claim 1, characterized in that, Also includes: Quality control is performed on the acquired meteorological variables, operational variables, and topographic variables; The meteorological variables are interpolated based on the topographic variables, and the interpolated meteorological variables are spatiotemporally registered according to the location of the target wind turbine and a preset time step.

4. The method for predicting wind turbine icing and assessing power loss according to claim 1, characterized in that, The dual output deviation observations include the relative air density corrected power curve deviation of the target fan and the deviation relative to the nearest reference fan; The dual output deviation observation obtained based on the output data of multiple adjacent wind turbines includes: Multiple adjacent wind turbines are filtered according to preset filtering rules to obtain at least one reference wind turbine; The weighted output of the neighboring reference wind turbines is obtained based on the output data of the reference wind turbine. The relative air density correction power curve deviation and the relative adjacent reference fan deviation are calculated based on the theoretical reference power, actual active power, and weighted output of the target fan and the adjacent reference fan.

5. The method for predicting wind turbine icing and assessing power loss according to claim 1, characterized in that, The process of fusing the physical icing thickness and the retrieved icing thickness to obtain the icing thickness fusion result includes: fusing the physical icing thickness and the retrieved icing thickness based on a preset fusion rule with data quality weights to obtain the icing thickness fusion result. The data quality weights are obtained based on the confidence level of numerical weather forecasts, the quality of real-time meteorological data, the quality of operational data, and the reliability of operational observations.

6. The method for predicting wind turbine icing and assessing power loss according to claim 1, characterized in that, The process of obtaining the output loss rate and icing probability based on the icing thickness fusion result includes: The power loss correction amount is obtained by inputting the fused ice thickness result, the operating variables of the target wind turbine, and the current wind speed into the trained residual correction model. The power loss is obtained by adding the icing thickness fusion result, the operating variables of the target wind turbine, the base power loss corresponding to the current wind speed, and the power loss correction. Obtain the ice-free reference power of the target wind turbine under the operating variables; The output loss rate is obtained based on the power loss and the reference power without icing.

7. The method for predicting wind turbine icing and assessing output loss according to claim 6, characterized in that, The process of obtaining the output loss rate and icing probability based on the icing thickness fusion result also includes: The icing probability can be obtained by using a trained residual correction model, or by inputting the icing thickness fusion result, the dual output deviation observation, the icing meteorological triggering index obtained from meteorological variables, and the output deviation duration of the target wind turbine into a probability mapping function to obtain the icing probability.

8. A device for predicting wind turbine icing and assessing output loss, characterized in that, include: The acquisition module is used to acquire meteorological variables, operational variables, and terrain variables of the target wind turbine; The prior module is used to calculate the prior data of physical icing based on the operating variables and meteorological variables; A terrain component correction module is used to calculate the terrain correction coefficient of the target wind turbine based on the terrain variables; Calculate the component exposure coefficient of the target wind turbine based on the operating variables; The correction module is used to correct the physical icing prior data according to the terrain correction coefficient and the component exposure coefficient to obtain the physical icing thickness of the target wind turbine; The observation module is used to obtain the dual output deviation observation based on the output data of multiple adjacent wind turbines; The inversion fusion module is used to obtain the inverted icing thickness based on the dual output deviation observation and the operating variables; and to fuse the physical icing thickness and the inverted icing thickness to obtain the icing thickness fusion result. The output module is used to obtain the output loss rate and icing probability based on the icing thickness fusion result.

9. A computer device, characterized in that, It includes at least one processor, a memory, and a transceiver; wherein the memory is used to store program code, and the processor is used to call the program code stored in the memory to perform the method as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, It includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-7.