Mountain wind power plant fan icing intelligent forecasting method and system

Abstract

The invention provides a mountain wind power plant fan icing intelligent forecasting method and system, and relates to the field of fan icing forecasting. The method comprises the following steps: determining a meteorological element threshold value of a fan in an icing scene and a meteorological element threshold value of the fan in an ice melting scene; establishing a temperature-wind speed-power dynamic response model; constructing a three-layer linkage model for fan icing prediction; the first-layer model judges an icing condition in a preset range around the station based on a meteorological element threshold value in an icing scene; the second-layer model is started before the first-layer model judges the icing condition, and a convection radiation ice melting scheme is coupled when the meteorological condition reaches a meteorological element threshold value in an ice melting scene; and on the premise of the second-layer model, the third-layer model timely triggers graded early warning based on a temperature-wind speed-power dynamic response model, and outputs a forecasting and early warning result. According to the method, the regional adaptability and precision of fan icing forecasting are improved, risk level early warning is optimized, and about 10% of power generation loss of a wind power plant is reduced.

Description

Technical Field

[0001] The present invention relates to the field of wind turbine icing forecasting, and in particular to an intelligent forecasting method and system for wind turbine icing in a mountain wind farm. Background Art

[0002] As the global energy transition accelerates, wind power, as one of the most promising renewable energy sources, plays a key role in achieving carbon neutrality. Compared to traditional energy sources, wind power offers significant advantages: First, wind energy resources are widely distributed, wind power generation has minimal environmental impact, and wind farms occupy relatively little land, offering significant environmental advantages. Second, wind farm construction cycles are relatively short, wind power generation costs have declined significantly over the past decade, and the volatility of wind power grid connection has been effectively mitigated, offering significant economic advantages. In recent years, in particular, with the application of high-tower, long-blade technology, and intelligent forecasting systems, the volatility of wind power grid connection has been significantly improved, further enhancing its dominant energy source. In today's global energy transition and the pursuit of carbon neutrality, wind power has become the most competitive renewable energy source.

[0003] However, under the low temperature and high humidity climate conditions in winter, the icing problem of wind turbine blades seriously restricts the efficient use of wind energy. Icing increases the mechanical load on the blades and causes a decrease in the aerodynamic performance of the blades, significantly reducing the power generation efficiency of the wind turbine. In extreme cases, it may cause blade breakage or tower collapse, posing a serious safety hazard. This phenomenon is particularly prominent in mountainous wind farms with significant altitude gradients. According to previous studies, the rate of change of ground wind speed is strongly correlated with altitude. When the wind speed weakens or increases, the rate of change of wind speed in high-altitude areas is higher than that in low-altitude areas. The particularity of the mountainous environment is the main reason for the low efficiency of forecasting the thickness and intensity of icing on wind turbines in wind farms. Henan Province is located in the mid-latitude north-south climate transition zone, with hilly mountains in the west and plains in the east. It is surrounded by Taihang Mountains, Funiu Mountains, Tongbai Mountains and Dabie Mountains on the north, west and south, and has a complex terrain. According to statistics, by the end of December 2024, Henan Province had 23.34 million kilowatts of installed wind power capacity, accounting for 15.7% of the province's total. Nearly half of this capacity is located in the western mountainous areas of the province, which are at high risk of icing. The complex and volatile geographical environment makes Henan's mountainous wind farms, which primarily rely on localized winds for their energy, extremely susceptible to the low temperatures and high humidity of winter. This can lead to increased ice accumulation on the blades, resulting in reduced power generation efficiency, equipment damage, and even widespread shutdowns.

[0004] Unlike the relatively well-researched icing of transmission lines, icing on mountain wind turbine blades occurs on three-dimensional rotating blades. This icing occurs in an environment influenced by complex airflow and centrifugal forces, leading to diverse damage mechanisms. These include: reduced turbine output due to blade profile changes caused by icing; increased vibration and damage to turbine components due to uneven ice distribution; and significantly increased centrifugal forces on blades as the ice thickens, challenging blade pressure loads. For many years, wind turbine icing research has focused on experiments and numerical simulations. While these studies address the practical needs of wind farm operations and maintenance, they still face challenges such as poor regional adaptability, insufficient data support, and low forecast accuracy, making it difficult to achieve precise forecasts for specific regions. Summary of the Invention

[0005] The purpose of the invention is to propose an intelligent forecasting method for wind turbine icing in mountainous wind farms and a system that can implement the method. By determining region-specific meteorological element thresholds and combining them with an icing growth model, the system can accurately predict the start and end time of icing and the extent of icing, thereby solving the above-mentioned problems existing in the prior art. The specific technical solution is as follows:

[0006] In a first aspect of the present invention, a method for intelligently predicting wind turbine icing in a mountainous wind farm is provided, comprising the following steps:

[0007] Based on the historical icing operation logs of the wind turbines at the test site, the icing period was determined, the critical value distribution characteristics of typical meteorological elements during the critical icing period were analyzed, and the meteorological element thresholds for wind turbines in icing scenarios and ice melting scenarios were determined.

[0008] Based on the meteorological element thresholds in the icing scenario, machine learning is used to distinguish the original power data of the wind turbine's actual operation and establish a temperature-wind speed-power dynamic response model.

[0009] A three-layer linkage model for wind turbine icing prediction is constructed; the first layer model determines the icing situation within a predetermined range around the station based on the meteorological element thresholds in the icing scenario; the second layer model is started at a predetermined time before the first layer model begins to determine the icing situation, and couples the convective radiation de-icing scheme when the meteorological conditions reach the meteorological element thresholds in the de-icing scenario; the third layer model, based on the second layer model, triggers graded warnings in a timely manner based on the temperature-wind speed-power dynamic response model and the monitored real-time data of the wind turbine;

[0010] Output forecast and warning results; the forecast and warning results include dynamically displaying the icing risk level on the GIS map, outputting the icing growth curve and the changing trend of key meteorological elements for a predetermined period of time in the future, and outputting the percentage of power generation attenuation.

[0011] In a further embodiment of the first aspect, the meteorological element thresholds in the icing scene include: temperature -7°C to 2°C, relative humidity ≥80%, wind speed 2 to 10 m / s, and precipitation ≥0.1 mm / 6 h;

[0012] The meteorological element thresholds in the ice melting scenario include: temperature > 2°C, relative humidity ≤ 80%, and wind speed 1-6 m / s.

[0013] In a further embodiment of the first aspect, the raw power data of the actual operation of the wind turbine is divided into three categories: power anomalies caused by icing, power noise caused by mixed sensor failures, power grid power restrictions, etc., and normal wind turbine power data;

[0014] The raw power data is differentiated through machine learning, where the machine learning methods include one or more of 3sigma, LOF, iForest, and DBSCAN.

[0015] In a further embodiment of the first aspect, the temperature-wind speed-power dynamic response model is used to calculate the influence characteristics of temperature and wind speed changes on power generation under mountain micro-topography, and determine the theoretical power P 理论 (t):

[0016] P 理论 (t)=0.5·ρ(T)·C p ·A·v(t) 3 ·η(T)

[0017] Where A is the swept area of ​​the wind wheel; v(t) is the wind speed; ρ(T) is the air density temperature correction term; η(T) is the blade aerodynamic efficiency temperature compensation coefficient; C p is the wind energy utilization coefficient.

[0018] In a further embodiment of the first aspect, the first-layer model judges the icing conditions of grid points within ten kilometers around the station based on the meteorological element thresholds in the icing scenario, marks 1 if there is icing and 0 if there is no icing, and preliminarily judges the possible icing timing of the wind farm in the next ten days by fusing the spatial probability density function weighting with the terrain correction coefficient.

[0019] In a further embodiment of the first aspect, the method of preliminarily judging the possible icing time sequence of the wind farm in the next ten days by fusing the weighted spatial probability density function with the terrain correction coefficient specifically includes:

[0020] Use Gaussian kernel function to calculate the horizontal distance weight W distance Simulate the spatial attenuation effect of meteorological elements:

[0021]

[0022] Where di is the horizontal distance between the grid point and the station; σ is the terrain bandwidth correlation coefficient;

[0023] Using the height weight W height Correct the elevation difference ΔH between the grid and the station i , constraining grid points whose elevation differences are greater than a predetermined value:

[0024]

[0025] Comprehensive calculation weight W i =W height ×W distance ;

[0026] Using comprehensive calculation weight W i Statistical effective ice coverage probability P ice :

[0027]

[0028] Where, if the grid point is covered with ice, then S i =1; if the grid point is not covered with ice, then S i =0; i is the current grid point, and n is the total number of grid points.

[0029] In a further embodiment of the first aspect, the second-layer model is started in a timely manner 2 hours before the first-layer model starts to judge the icing situation. The second-layer model is based on the Makkonen icing growth model. When the meteorological conditions reach the melting conditions, the convection radiation melting scheme is coupled with the Makkonen icing growth model to form an icing and melting nested model.

[0030] In a further embodiment of the first aspect, two hours before the first layer model starts to determine the icing situation, the Makkonen icing growth model is started to describe the ice growth process, and the calculation formula is as follows:

[0031]

[0032] Where M is the mass of ice; t is the freezing time; α1 is the collision rate; α2 is the capture rate, α3 is the freezing rate; v is the velocity of the particle relative to the fan; ω is the liquid water content; S is the effective cross-sectional area of ​​the object that the water droplet collides with;

[0033] The collision rate α1 is determined based on experience; the capture rate α2 = 1; when all captured supercooled water droplets are frozen on the surface of the object, it is a dry growth state with a freezing rate of α3; when the supercooled water droplets are partially frozen and the unfrozen part is lost on the surface of the object, it is a wet growth state with a freezing rate of α3:

[0034]

[0035] Where F is the water flux density of the ice-covered surface; λ is the unfrozen portion of the surface, that is, the proportion of liquid water; h is the convective heat exchange coefficient; σ is the Stefan-Boltzmann constant; α is the radiation constant; ε is the molar molecular ratio of water vapor; r is an empirical parameter; P is the air pressure; e a 、e s are water vapor pressure and saturated water vapor pressure respectively; t s , t a , t d are ice surface temperature, air temperature, and droplet collision temperature respectively; L f 、L e are the latent heat of freezing and the latent heat of evaporation of water respectively; C p 、C w are the specific heats of air and water, respectively.

[0036] In a further embodiment of the first aspect, when meteorological conditions reach ice melting conditions, a convection radiation ice melting scheme is coupled with a Makkonen ice growth model;

[0037] Calculate the melting rate of ice under the convection radiation melting scheme

[0038]

[0039] Where σ is the Stefan-Boltzmann constant; α is the radiation constant; c i is the specific heat of ice; ρ i is the density of melted ice;

[0040] When simulating ice thickness, the blade thickness of the wind turbine is set to a cylinder for simulation, with a diameter of D0, and the ice shape is regarded as cylindrical ice. When the ice melting condition is reached, the ice accumulation diameter D t and ice thickness C t They are as follows:

[0041]

[0042] Where, M is the mass of ice; t The ice cover weight at the tth moment output by the ice formation and ice melting nested model.

[0043] The second aspect of the present invention proposes an intelligent prediction system for wind turbine icing in a mountain wind farm, the system comprising at least one processor and at least one memory communicatively connected to the processor; wherein: the memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement an intelligent prediction method for wind turbine icing in a mountain wind farm as described in the first aspect.

[0044] Compared with the prior art, the present invention has the following significant beneficial effects:

[0045] (1) High-precision ice forecast: The present invention significantly improves the forecast accuracy of ice start and end time and ice thickness growth by constructing a mountain-specific meteorological element threshold library, combining the Makkonen ice growth model based on the heat balance principle, coupling the ice melting module, superimposing the mountain turbulent wind speed field correction and the introduction of temperature vertical gradient.

[0046] (2) Regional adaptation and dynamic response: The present invention dynamically adjusts model parameters (such as mountain wind speed field, vertical temperature gradient, and radiation effect) through localization. The model uses the forecast results of numerical models as the initial field to quickly identify the time of ice cover mutation and dynamically predict the start and end time of ice cover and the degree of ice cover.

[0047] (3) Real-time warning and operation and maintenance support: The present invention links the forecast results with the wind farm management system, and integrates and visualizes geographic information overlay (GIS map dynamically displays the icing risk level), time series charts (icing growth curve and changing trends of key meteorological elements in the next 240 hours), and wind turbine status panels (real-time data such as power generation efficiency loss percentage), to achieve real-time warning and reduce operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 4 is a flow chart of the method for intelligently predicting wind turbine icing in a mountain wind farm in an embodiment.

[0049] Figure 2 It is the distribution frequency of temperature, wind speed, relative humidity and precipitation during the period when wind turbine icing occurs.

[0050] Figure 3 Schematic diagram of the geographical locations of wind farms A and B in the embodiment. DETAILED DESCRIPTION

[0051] In the following description, numerous specific details are provided to provide a more thorough understanding of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced without one or more of these details. In other instances, certain technical features well known in the art have not been described to avoid confusion with the present invention.

[0052] This embodiment provides a method for intelligently predicting wind turbine icing in a mountainous wind farm. The specific technical solution is as follows:

[0053] S1. Based on the historical icing operation log of the wind turbine at the test site, determine the time period when icing occurs, analyze the critical value distribution characteristics of typical meteorological elements during the critical period of icing, and determine the threshold distribution range of meteorological elements under the wind turbine icing scenario.

[0054] Specifically, based on historical icing events, the probability density function was used to statistically analyze the temperature, relative humidity, wind speed and other variables during the occurrence and melting of icing at the test site, and the critical value distribution characteristics of key meteorological elements during the occurrence and melting of icing events on wind turbine blades in mountain wind farms in Henan Province were analyzed.

[0055] Based on the icing conditions, the authors determined the icing thresholds for four key meteorological factors: temperature (-7°C to 2°C), relative humidity (≥80%), wind speed (2-10 m / s), and precipitation (≥0.1 mm / 6 hours) for different wind farm sites. A regression model was used to verify the weighted contribution of each factor to icing occurrence. Furthermore, based on the environmental conditions, the distribution characteristics of meteorological factors related to ice ablation were analyzed, and the distribution of the main meteorological factors suitable for ice ablation was determined: temperature (>2°C), relative humidity (≤80%), and wind speed (1-6 m / s).

[0056] Industry data such as the start and end time of wind turbine icing, air temperature and wind speed at the height of the wind turbine hub, and wind turbine parameters are obtained from historical records of two experimental wind farms in southern Henan (Wind Farm A in Xinxian County, Xinyang, and Wind Farm B in Biyang County, Zhumadian) during the period 2020-2022; temperature and wind speed are derived from the wind turbine's own recorded values, and relative humidity and precipitation are derived from the station difference results of the ERA5 reanalysis data (horizontal resolution of 0.25°×0.25°, time resolution of 1h).

[0057] S2, under the premise of the threshold of icing meteorological elements, accurately distinguish the actual operating power data of the wind turbine through machine learning, establish a "temperature-wind speed-power" dynamic response model, and obtain the dynamic attenuation law of icing power.

[0058] Specifically, the original wind turbine power data is divided into three categories: power anomalies caused by icing, power noise caused by mixed sensor failures, grid power restrictions, etc., and normal wind turbine power data.

[0059] By integrating meteorological condition thresholds to initially screen icing data, a three-level time series processing strategy (change point detection, dynamic time window merging, and duration filtering) is designed to extract power anomalies that may be caused by icing.

[0060] Machine learning methods include 3sigma, LOF, iForest, and DBSCAN. Each site is tested one by one, and multiple models are processed collaboratively to select the most suitable abnormal data cleaning method for the current site and identify power noise caused by objective factors such as sensor failure and grid power restrictions.

[0061] Through the above two steps, the wind turbine power data is cleaned. The cleaned normal wind turbine power historical data is deeply coupled with the high-precision temperature and wind speed data provided by the wind turbine itself to establish a "temperature-wind speed-power" dynamic response model exclusive to the experimental wind farm station. This model dynamically reflects the impact of temperature and wind speed changes on power generation under specific mountain micro-topography.

[0062] The calculation of theoretical power in the dynamic response model takes into account the influence of temperature and wind speed. On the one hand, it decouples the direct impact of wind speed changes on power fluctuations, and on the other hand, it corrects the indirect impact of low temperature environment on the power baseline value. This provides a more accurate reference benchmark for the "temperature-wind speed-power" linked icing warning and supports the extraction of power attenuation characteristics in the icing model.

[0063] In the process of model establishment, historical data is needed to determine the parameters in the theoretical formula, such as the wind energy utilization coefficient C p And the blade aerodynamic efficiency temperature compensation coefficient η(T):

[0064] Specifically, the core formula of theoretical power is as follows:

[0065] P 理论 (t)=0.5·ρ(T)·C p ·A·v(t) 3 ·η(T)

[0066] Where A is the swept area of ​​the wind wheel (A=πR 2 , R is the blade radius); v(t) is the wind speed.

[0067] ρ(T) is the air density temperature correction term (air density increases as temperature decreases, and the air mass per unit volume is higher at low temperatures, which directly affects the amount of wind energy captured).

[0068]

[0069] η(T) is the temperature compensation coefficient of blade aerodynamic efficiency (T ref =0℃ is the reference temperature):

[0070] η(T)=1-k·(T ref -T(t))

[0071] Low temperatures typically cause changes in blade material stiffness, increase surface friction, and reduce aerodynamic efficiency, necessitating calibration based on historical temperature and power data. The empirical value of k in the temperature compensation coefficient is typically 0.0035. Considering that the actual k in mountainous wind farms is typically higher than that in plain wind farms, this paper uses linear regression to calibrate the coefficient. Calibration results show that k in the test wind farm is typically between 0.0042 and 0.0051.

[0072] C p The wind energy utilization coefficient is usually the inherent parameter of the wind turbine provided by the manufacturer (0.35-0.45). However, in actual operation, it may change due to factors such as equipment aging and micro-topography, and needs to be recalibrated based on historical data. This paper uses cleaned historical normal data (including power, wind speed, and temperature during periods when the temperature is greater than 0 degrees and there is no ice) to reversely infer the actual C p , and found that the C p 5% to 15% lower than the nominal value of 0.35.

[0073] S3, based on the multi-source fusion of numerical forecast fields and measured wind speeds of wind turbines, innovatively develops a three-layer linkage model for wind turbine icing prediction, breaking through the technical limitations of the traditional Makkonen icing model and improving the accuracy of mountain wind turbine icing.

[0074] The numerical simulation data are forecast data from the China Meteorological Administration's Wind and Solar Energy Forecasting System (CMA-WSP). The forecast meteorological elements include temperature, wind speed, humidity, rainfall, etc. The forecast time is from 20:00 every day to 240 hours in the future, with a temporal and spatial resolution of 15 minutes / 9 kilometers.

[0075] The first-level model (meteorological element threshold test) uses the model data and the icing meteorological element threshold for the station screened in Part 1 to calculate the icing conditions of grid points within a ten-kilometer radius of the station (with icing marked as 1 and no icing marked as 0). By integrating the weighted spatial probability density function with the terrain correction coefficient, it preliminarily determines the possible icing time series of the wind farm in the next ten days.

[0076] Step 1: Calculate the grid weights using the adaptive Gaussian kernel function.

[0077] a. Use the Gaussian kernel function to calculate the horizontal distance weight to simulate the spatial attenuation effect of meteorological elements. The closer the grid point is to the station, the greater the impact.

[0078]

[0079] Among them, d i is the horizontal distance between the grid point and the station, and σ is the terrain bandwidth correlation coefficient, which is taken as 3km here. The main reasons are:

[0080] 1. Affected by the terrain shielding effect, the effective radius of the mountain micrometeorological field is usually 2-4km (such as the vortex area on the leeward slope and the uplift area on the windward slope).

[0081] 2. The original resolution of the CMA-WSP model is 9 km, which is downscaled to a 3 km bandwidth (9 km ÷ 3 = 3), which not only preserves the integrity of the model physical processes but also enhances local adaptation.

[0082] b. Correct elevation differences and constrain grid points with large elevation differences to avoid incorrect effects on valleys / mountaintops.

[0083]

[0084] Where ΔH i is the elevation difference between the grid point and the station.

[0085] c. Comprehensive calculation weights.

[0086] W i =W height ×W distance

[0087] Step 2: Use comprehensive calculation weight W i Statistical effective ice coverage probability P ice :

[0088]

[0089] When P ice When ≥0.65, it is considered that there is an icing risk at this station. ice = 0.65, the ice hit rates (the probability of the model triggering an alarm when ice is actually covered) for the two stations are approximately 82% and 85%, the false alarm rates (the probability of a false alarm when ice is not covered) are approximately 14% and 12%, and the missed alarm rates (the probability of the model missing a warning when ice is actually covered) are approximately 4% and 3%. Setting the threshold too stringent reduces the false alarm rate but increases the missed alarm rate. Selecting this threshold balances the false alarm and missed alarm rates for ice coverage while ensuring the hit rate.

[0090] The purpose of this model is to reduce the ice miss rate as much as possible while ensuring the ice hit rate, and at the same time improve the operating efficiency of the linkage model.

[0091] The second-layer model (nested ice formation and ice-melting model) starts the Makkonen ice growth model in time three hours before the first-layer model initially judges the onset of ice cover. When the meteorological conditions meet the ice-melting conditions, the convection-radiation ice-melting scheme is coupled.

[0092] In the first step, the threshold method is used to determine 3 hours before the onset of icing, and the Makkonen icing growth model is activated to describe the ice growth process.

[0093] The model deeply analyzes the microphysical mechanism and thermal balance process of the collision between hydroparticles and transmission lines. Under the premise of relatively accurate input meteorological factors, the model can accurately describe the growth process of ice accumulation. Its calculation formula is as follows:

[0094]

[0095] Where M is the mass of ice (g); t is the freezing time (s); is the mass growth of ice per unit time (g / s); α1 is the collision rate; α2 is the capture rate; α3 is the freezing rate; v is the effective particle velocity (m / s), that is, the speed of the particles relative to the fan; ω is the water content of the particle group (g / m3), that is, the liquid water content LWC (Liquid Water Content); S is the effective cross-sectional area of ​​the object that the water droplet collides with (m 2 ); the collision rate α1 is:

[0096] α1=A-0.028-C(B-0.0454)

[0097] Where A, B, and C are empirical parameters.

[0098] The capture rate α2 is set to 1. The freezing rate α3 describes the probability of a droplet freezing on the surface of an object. The calculation of α3 should consider both dry growth and wet growth states. When all captured supercooled water droplets are frozen on the surface of an object, it is a dry growth state with a freezing rate of 1. When some supercooled water droplets are frozen and the unfrozen part flows away from the surface of an object, it is a wet growth state with a freezing rate of α3:

[0099]

[0100] Where F = α1α2ωv, is the water flux density of the ice-covered surface; λ is the unfrozen portion of the surface, that is, the proportion of liquid water; h = Nu·λ a / D, is the convective heat exchange coefficient, where λ a is the thermal conductivity of air, which is 2.5362×10 -5 J / (m·s·K), Nu is the Nusselt number, Nu=0.032Re 0.85 ; σ is the Stefan-Boltzmann constant, which is 5.6696×10 -8 W / (m 2 k 4 ); α is the radiation constant, which is 8.1×10 7 K 3 ; ε is the molar ratio of water vapor, which is taken as 0.62; r is an empirical parameter, which is taken as 0.79; P is the air pressure (Pa), e a ,e s are water vapor pressure and saturated water vapor pressure (Pa), respectively; t s , t a , t d are ice surface temperature, air temperature and droplet collision temperature (℃); L f , L e are the latent heat of freezing and latent heat of evaporation of water (J / (kg·K)); C p , Cw are the specific heats of air and water (J / (kg·K)), respectively.

[0101] Step 2: After the meteorological field reaches the ice melting conditions, the ice melting model is coupled with the Makkonen model.

[0102] Ice melting mainly considers the convection radiation ice melting scheme, and the melting rate of ice is:

[0103]

[0104] Where dD / dt is the ice melting rate, that is, the change in ice thickness D (cm) per unit time; σ is the Stefan-Boltzmann constant; c i is the specific heat of ice (J / (kg·K)); ρ i is the density of melted ice, take 0.9g / cm 3 .

[0105] When simulating ice thickness, the blade thickness is assumed to be a cylinder with a diameter of D0, and the ice shape is regarded as an idealized cylindrical ice. When the ice melting condition is reached, the ice melting model is coupled with the Makkonen model, and the ice weight M at the tth moment output by the model is t , ice accumulation diameter D t , ice thickness C t for:

[0106]

[0107] The purpose of this layer of model is to achieve accurate prediction of the start and end time and degree of icing based on the first layer of model, so as to improve the icing hit rate.

[0108] The third-layer model (wind turbine monitoring supplements short-term and impending warning) combines the "temperature-wind speed-power" dynamic response model and real-time wind turbine monitoring to build an intelligent early warning system of "temperature perception-wind speed coordination-power attenuation" to improve the reliability of impending warnings.

[0109] The first step is to ensure that the data source for real-time monitoring of wind turbines is complete and to quantify the power attenuation characteristics.

[0110] Under the premise of the second-layer model, based on the "temperature-wind speed-power" model of the power station, the power attenuation under icing conditions is dynamically determined. The formula is as follows:

[0111]

[0112] Step 2: Filter threshold conditions based on historical conditions and trigger graded warnings at the appropriate time.

[0113] Triggering a clear and concise three-level warning allows non-technical personnel to quickly understand the risk status of wind turbine icing.

[0114] When the power reduction is greater than 10% and lasts for 10 minutes, a level 3 icing warning is triggered and blade video monitoring is activated;

[0115] When power attenuation exceeds 15% and lasts for 30 minutes, a Level 2 icing warning is triggered, notifying operation and maintenance personnel for on-site verification.

[0116] When the power reduction is greater than 20% and lasts for 60 minutes, a level 1 warning is triggered and de-icing preparations are required.

[0117] S4, verification of ice cover forecast effect.

[0118] A cross-validation method was used to predict the icing process at typical mountain wind farms (sites A and B) in Henan Province during the winter of 2022. The hit rate and false alarm rate of icing events compared with actual observations were statistically analyzed to evaluate the icing reporting rate. A sliding time window method was used to identify critical deviation periods and calculate the mean absolute error of icing start and end times. A multi-objective genetic algorithm was used to optimize model parameters to improve forecast accuracy (the reporting rate of icing for more than one day was increased to over 70%, and the time error was controlled within 3 hours).

[0119] S5, forecast and warning result output.

[0120] The forecast results are output by integrating geographic information overlay (GIS map dynamically displays the icing risk level), time series charts (icing growth curve and changing trends of key meteorological elements in the next 240 hours), and wind turbine status panels (real-time data such as power generation efficiency loss percentage) into the visualization platform. With the help of interfaces, real-time data interaction with the wind farm system is achieved, and the forecast results are pushed to the wind farm operation and maintenance system to provide technical support for de-icing decisions and scientific scheduling.

[0121] Taking the mountain wind farm in Henan Province as an example, Figure 1 The figure is a flow chart of the method described in the present invention. The collected data was first preprocessed. Then, based on the icing operation log data from wind farms A and B from 2020 to 2022, this study conducted preliminary statistics on the frequency of temperature, relative humidity, wind speed, and precipitation during icing events at wind farms A and B in southern Henan Province. The critical value distribution characteristics of key meteorological elements during wind turbine blade icing events at mountain wind farms in Henan Province were analyzed. Figure 2 Table 1 shows the thresholds of meteorological factors suitable for wind turbine icing in mountainous areas of Henan Province: the temperature at hub height is between -7 and 2°C, the wind speed is between 2 and 10 m / s, the relative humidity is above 80%, and there is slight precipitation.

[0122] Table 1 Distribution of threshold values ​​of main meteorological factors suitable for icing

[0123] Meteorological elements Ice cover meteorological element threshold temperature -7～2℃ relative humidity 80％～100％ wind speed 2~10m / s precipitation trace precipitation

[0124] Figure 2In the figure, the four sub-figures (a), (b), (c) and (d) respectively show the distribution frequencies of temperature, wind speed, relative humidity and precipitation during the period of wind turbine icing.

[0125] Figure 3 A schematic diagram of the geographical locations of wind farms A and B is given (the color is the terrain height), wind farm A (in Xinyang Xinxian County, 380-400m above sea level), and wind farm B (in Zhumadian Biyang County, 800-900m above sea level).

[0126] Next, based on the threshold distribution of meteorological factors for wind turbine icing (hereinafter referred to as the threshold method), combined with the Makkonen icing growth model, we developed a forecast for the start and end times of wind turbine icing. We used wind farms A and B as the research objects and simulated their icing processes in 2022.

[0127] Tables 2 and 3 show the effectiveness of the threshold method and the Makkonen model for predicting the start and end times of icing events at wind farms A and B, respectively. The results show that the threshold method achieved a 60% prediction rate for icing events, while both Makkonen model prediction schemes achieved a 70% prediction rate. Regarding the accuracy of icing onset prediction, the three prediction schemes (threshold method, Makkonen 1-day-ahead, and 2-day-ahead) had average deviations from the actual icing onset time of -6.2 hours, -3.6 hours, and -3 hours, respectively, with mean absolute errors of 7.5 hours, 5 hours, and 6.7 hours, respectively. High-precision forecasts with mean absolute errors of less than 3 hours were achieved four, four, and two times, respectively.

[0128] Table 2 Verification of the forecast start time of the icing process for wind farms A and B in 2022

[0129]

[0130]

[0131] Table 3 Verification of the forecast time for the end of the icing process in 2022 for wind farms A and B

[0132]

[0133] The method for intelligent prediction of wind turbine icing in a mountain wind farm disclosed in the above embodiment can, in actual application, have its operating logic written into a program product, which is written into a storage medium and runs on an electronic device. More specific examples of the computer-readable storage medium mentioned in this embodiment may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The readable signal medium may also be any readable medium other than a readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0134] In summary, the present invention addresses the key issues that restrict existing wind turbine icing forecasts: first, the traditional method has insufficient forecast accuracy. Relying on a static model, it results in significant dynamic errors in the start and end times of icing and ice thickness growth, making it difficult to accurately determine the timing for de-icing. Second, the existing model has poor regional adaptability and does not consider the coupling effect between mountain micrometeorological characteristics and wind turbine rotation effects, making it difficult to accurately predict micro-topography icing phenomena. Third, the dynamic response of the forecast is delayed, lacking the support of high-temporal and spatial resolution local meteorological data, making it difficult to respond to sudden changes in icing in a timely manner. The present invention, based on the terrain characteristics of Henan Province, designs an intelligent forecast method and system for wind turbine icing in mountain wind farms. By constructing a mountain-specific meteorological element threshold library, combining neural network optimization of Makkonen model parameters, and introducing a terrain elevation correction factor, a technical breakthrough is achieved in forecasting the start and end times of icing for wind turbines in mountain wind farms with an accuracy rate of more than 90%. This invention helps wind farms reduce power generation losses by about 10% by improving the regional adaptability and accuracy of wind turbine icing forecasts, optimizing risk level warnings, and providing operation and maintenance advice outputs, thus achieving an operation and maintenance upgrade from passive de-icing to active defense.

[0135] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes may be made to it in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. An intelligent prediction method for wind turbine icing in a mountainous wind farm, characterized in that: The steps include: Based on the historical icing operation logs of the wind turbines at the test site, the icing period was determined, the critical value distribution characteristics of typical meteorological elements during the critical icing period were analyzed, and the meteorological element thresholds for wind turbines in icing scenarios and ice melting scenarios were determined. Based on the meteorological element thresholds in the icing scenario, machine learning is used to distinguish the original power data of the wind turbine's actual operation and establish a temperature-wind speed-power dynamic response model. A three-layer linkage model for wind turbine icing prediction is constructed; the first layer model determines the icing situation within a predetermined range around the station based on the meteorological element thresholds in the icing scenario; the second layer model is started at a predetermined time before the first layer model begins to determine the icing situation, and couples the convective radiation de-icing scheme when the meteorological conditions reach the meteorological element thresholds in the de-icing scenario; the third layer model, based on the second layer model, triggers graded warnings in a timely manner based on the temperature-wind speed-power dynamic response model and the monitored real-time data of the wind turbine; Output forecast and warning results; the forecast and warning results include dynamically displaying the icing risk level on the GIS map, outputting the icing growth curve and the changing trend of key meteorological elements for a predetermined period of time in the future, and outputting the percentage of power generation attenuation.

2. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 1, characterized in that: The meteorological element thresholds in the ice cover scenario include: temperature -7°C to 2°C, relative humidity ≥ 80%, wind speed 2 to 10 m / s, and precipitation ≥ 0.1 mm / 6 h; The meteorological element thresholds in the ice melting scenario include: temperature > 2°C, relative humidity ≤ 80%, and wind speed 1-6 m / s.

3. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 1, characterized in that: The raw power data from actual wind turbine operation is divided into three categories: power anomalies caused by icing, power noise from sensor failures and grid power restrictions, and normal wind turbine power data. The raw power data is differentiated through machine learning, where the machine learning methods include one or more of 3sigma, LOF, iForest, and DBSCAN. The temperature-wind speed-power dynamic response model is used to calculate the influence characteristics of temperature and wind speed changes on power generation under mountain micro-topography, and determine the theoretical power P 理论 (t): P 理论 (t)=0.5·ρ(T)·C p ·A·v(t) 3 ·η(T) Where A is the swept area of ​​the wind wheel; v(t) is the wind speed; ρ(T) is the air density temperature correction term; η(T) is the blade aerodynamic efficiency temperature compensation coefficient; C p is the wind energy utilization coefficient.

4. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 1, characterized in that: The first-layer model determines the icing conditions of grid points within a ten-kilometer radius around the station based on the meteorological element thresholds under the icing scenario. Ice coverage is marked as 1 and no ice coverage is marked as 0. By integrating the weighted spatial probability density function with the terrain correction coefficient, the possible icing sequence of the wind farm in the next ten days is preliminarily determined.

5. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 4, characterized in that: The method of integrating the weighted spatial probability density function with the terrain correction coefficient to preliminarily judge the possible icing time sequence of the wind farm in the next ten days includes: Use Gaussian kernel function to calculate the horizontal distance weight W distance Simulate the spatial attenuation effect of meteorological elements: Where di is the horizontal distance between the grid point and the station; σ is the terrain bandwidth correlation coefficient; Using the height weight W height Correct the elevation difference ΔH between the grid and the station i , constraining grid points whose elevation differences are greater than a predetermined value: Comprehensive calculation weight W i =W height ×W distance ; Using comprehensive calculation weight W i Statistical effective ice coverage probability P ice : Where, if the grid point is covered with ice, then S i =1; if the grid point is not covered with ice, then S i =0; i is the current grid point, and n is the total number of grid points.

6. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 1 or 4, characterized in that: Three hours before the first-layer model begins to judge the icing situation, the second-layer model is started in time. The second-layer model is based on the Makkonen ice growth model. When the meteorological conditions reach the melting conditions, the convection radiation melting scheme is coupled with the Makkonen ice growth model to form a nested ice formation and melting model.

7. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 6, characterized in that: Three hours before the first-layer model begins to judge the icing situation, the Makkonen icing growth model is started to describe the icing growth process. The calculation formula is as follows: Where M is the mass of ice; t is the freezing time; α1 is the collision rate; α2 is the capture rate, α3 is the freezing rate; v is the velocity of the particle relative to the fan; ω is the liquid water content; S is the effective cross-sectional area of ​​the object that the water droplet collides with; The collision rate α1 is determined based on experience; the capture rate α2 = 1; when all captured supercooled water droplets are frozen on the surface of the object, it is a dry growth state with a freezing rate of α3; when the supercooled water droplets are partially frozen and the unfrozen part is lost on the surface of the object, it is a wet growth state with a freezing rate of α3: Where F is the water flux density of the ice-covered surface; λ is the unfrozen portion of the surface, that is, the proportion of liquid water; h is the convective heat exchange coefficient; σ is the Stefan-Boltzmann constant; α is the radiation constant; ε is the molar molecular ratio of water vapor; r is an empirical parameter; P is the air pressure; e a 、e s are water vapor pressure and saturated water vapor pressure respectively; t s , t a , t d are ice surface temperature, air temperature, and droplet collision temperature respectively; L f , L e are the latent heat of freezing and the latent heat of evaporation of water respectively; C p 、C w are the specific heats of air and water, respectively.

8. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 6, characterized in that: When meteorological conditions reach the melting conditions, the convection-radiation melting scheme is coupled with the Makkonen ice growth model; Calculate the melting rate of ice under the convection radiation melting scheme Where σ is the Stefan-Boltzmann constant; α is the radiation constant; c i is the specific heat of ice; ρ i is the density of melted ice; When simulating ice thickness, the blade thickness of the wind turbine is set to a cylinder for simulation, with a diameter of D0, and the ice shape is regarded as cylindrical ice. When the ice melting condition is reached, the ice diameter D at time t is t and ice thickness C t They are as follows: Where, M is the mass of ice; t The ice cover weight at the tth moment output by the ice formation and ice melting nested model.

9. The intelligent prediction method for wind turbine icing in a mountainous wind farm according to claim 3, characterized in that: The third layer model dynamically determines the power attenuation D under icing conditions based on the temperature-wind speed-power dynamic response model under the premise of the second layer model. P(t) : According to the power attenuation D P(t) Triggering graded warnings: When the power is reduced D P(t) >10% and lasts for 10 minutes, triggering a Level 3 ice cover warning and starting blade video monitoring; When the power is reduced D P(t) >15% and lasts for 30 minutes, triggering a Level 2 icing warning and notifying operation and maintenance personnel for on-site verification; When the power is reduced D P(t) >20% and lasts for 60 minutes, triggering a Level 1 warning and preparing for de-icing.

10. An intelligent forecasting system for wind turbine icing in mountainous wind farms, characterized in that: include: at least one processor and at least one memory in communication with the processor; wherein: The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the intelligent prediction method for wind turbine icing in a mountain wind farm according to any one of claims 1 to 9.

Images