Multi-source meteorological data dynamic fusion wind power plant simulation method

Abstract

The invention relates to the technical field of wind energy development and utilization, in particular to a multi-source meteorological data dynamic fusion wind power plant simulation method, which comprises the following steps of: accessing satellite remote sensing, a ground observation station, numerical forecasting and wind power plant historical operation data, and outputting standardized meteorological data through format analysis, standardized processing and cleaning; a credibility score is calculated based on multi-dimensional indexes such as data integrity and time consistency, and abnormal data restoration or standby data source switching is achieved; in combination with wind power plant DEM topographic features and a self-adaptive sliding window, the weight of a data source is adjusted in real time through reinforcement learning, and a dynamic weight value is generated; an improved Windpowerlib framework adapts to fusion data, fan parameters are dynamically configured, and an initial simulation result is generated; and collecting real-time data of the wind power plant, and correcting model parameters by using Kalman filtering. The wind power plant simulation precision and timeliness can be improved, the data utilization rate and the model adaptability are enhanced, and the deviation between the simulation result and the actual working condition is reduced.

Description

Technical Field

[0001] This invention belongs to the field of wind energy development and utilization technology, specifically involving a method for simulating wind farms by dynamically fusing multi-source meteorological data. Background Technology

[0002] With the global energy structure accelerating its transformation towards clean energy, wind power, with its clean and renewable characteristics, has become an important part of the renewable energy system. Its development and utilization scale is continuously expanding, playing an increasingly crucial role in the energy transition process.

[0003] Wind farm simulation technology, as a core support tool for the entire process management of wind power projects, runs through the planning and design stage (such as site selection and turbine layout optimization), the operation optimization stage (such as power generation efficiency improvement and fault prediction), and the benefit evaluation stage (such as investment return cycle calculation and long-term benefit forecasting). Its technical level and application effectiveness directly affect the investment return efficiency and overall operational quality of wind farms, and have a significant impact on the large-scale and high-quality development of the wind power industry. However, current wind farm simulation methods still have significant limitations in practical applications, mainly concentrated in two core areas: meteorological data processing and model adaptation. (1) Most existing simulation methods rely on a single meteorological data source (such as numerical weather prediction data) or adopt a multi-source data fusion mode with fixed weights. This method cannot fully reflect the dynamic changes of meteorological elements such as wind speed, wind direction, and air pressure. In the time dimension, it is difficult to adapt to the meteorological fluctuations caused by the alternation of day and night and the changing of seasons. In the spatial dimension, it cannot accurately match the local meteorological differences caused by topographic differences and regional climate characteristics. Ultimately, under complex meteorological conditions (such as strong gusts and local circulation), the simulation results deviate greatly from the actual situation, making it difficult to support accurate decision-making.

[0004] (2) For multi-source heterogeneous meteorological data such as satellite remote sensing data, ground station data, and numerical forecast data, the existing technical system lacks a systematic quality assessment mechanism and cannot effectively distinguish the accuracy differences and reliability levels of different data sources. Once low-precision and high-noise data is directly input into the simulation model, it is very easy to introduce simulation bias, which greatly reduces the credibility of the model output results and restricts the scientific nature of wind farm planning and design and the effectiveness of operation optimization.

[0005] (3) Current mainstream wind farm simulation frameworks (such as Windpowerlib) have significant shortcomings in data source compatibility, and the types of data sources they natively support are relatively limited. Faced with the need to access multi-source heterogeneous data, a lot of customized development work is often required, which not only increases the technical cost and implementation cycle, but may also affect the stability of the simulation system due to compatibility issues during the development process. At the same time, the interaction between these frameworks and the real-time operating data of wind farms is significantly lagging, and it is impossible to dynamically correct the simulation model based on the real-time operating data, which makes it difficult for the simulation results to match the actual operating state of the wind farm in real time, further weakening the application value of simulation technology.

[0006] In view of this, the present invention is hereby proposed. Summary of the Invention

[0007] To address the aforementioned technical problems in existing technologies, this invention provides a method for simulating wind farms by dynamically fusing multi-source meteorological data. By constructing a dynamic fusion mechanism, a quality assessment model, and a real-time correction process, the accuracy and timeliness of wind farm simulation are improved.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: A method for simulating wind farms by dynamically fusing multi-source meteorological data includes: S1. Access satellite remote sensing data, ground station data, numerical forecast data and historical wind farm operation data through dynamic interface and generate multi-source data. After performing format adaptive parsing on the multi-source data, perform data standardization processing and data cleaning, and output standardized meteorological data. S2. Construct a multi-dimensional evaluation index system using the standardized meteorological data, calculate the credibility score of the standardized meteorological data using a weighted algorithm, mark the data as abnormal, perform Kriging interpolation repair or trigger the switching of backup data sources based on the score, and output the data credibility score. S3. Based on the data credibility score, a spatial correlation matrix is ​​established by combining the terrain features extracted from the wind farm digital elevation model. The time trend of meteorological data is extracted by the adaptive sliding window algorithm. The weights of each data source are adjusted in real time by the reinforcement learning algorithm, and the dynamic weight values ​​are output. S4. Based on the meteorological data after weighted fusion of the dynamic weight values, the simulation framework is further developed to adapt to the fused data, the wind turbine parameters are dynamically configured, and the wind turbine power curve calculation module, wake effect simulation module and air density correction module of the model are called to generate the initial simulation results of the wind farm. S5. Collect real-time operating data of the wind farm and calculate the deviation between the data and the initial simulation results of the wind farm. When the deviation exceeds the set threshold, dynamically correct the key parameters of the simulation framework model through the Kalman filter algorithm until the deviation is reduced to within the threshold and output the final simulation results of the wind farm.

[0009] Furthermore, the formats of the multi-source data include NetCDF, CSV, and JSON, and the adaptive format parsing is implemented using the xarray library, pandas library, and json library; The data standardization process includes: secondary development of the Windpowerlib framework, adding a data adaptation layer, and converting multi-source heterogeneous data into a unified format; The data cleaning includes: using Python's Pandas library to clean the data, removing outliers and duplicates, and using linear interpolation or spline interpolation to fill in missing data to ensure data continuity.

[0010] Furthermore, the collection of multi-source data in step S1 is implemented by writing a collection script in Python. The collection script integrates the requests library to support HTTP / HTTPS protocol data requests and uses the APScheduler scheduling framework to set the collection cycle to 1 minute / time. The sources of the multi-source data include: satellite data platform NASAEarthdata, ground station data center NPC API, and numerical weather prediction database ECMWFWebAPI.

[0011] Furthermore, the multi-dimensional evaluation indicators include: data integrity, temporal consistency, spatial correlation, and historical error rate; The specific weighting for the credibility score is as follows: data integrity 30%, temporal consistency 25%, spatial correlation 25%, and historical error rate 20%; the credibility score calculation formula is: The result is rounded to one decimal place.

[0012] Furthermore, the method is characterized in that the establishment of the spatial correlation matrix in step S3 includes: Based on wind farm digital elevation model (DEM) data, 30m resolution DEM data was read using the GDAL library, and terrain features were extracted using the rasterio library. These terrain features included slope and aspect, and the slope calculation formula was as follows:

[0013] The formula for calculating slope aspect is:

[0014] Based on the aforementioned terrain features, a spatial attenuation coefficient model for wind speed is constructed. The formula for the attenuation coefficient model is as follows:

[0015] in, The terrain influence coefficient is used in mountainous scenes. In the plains scene A spatial correlation matrix of a 50m×50m grid is established based on the attenuation coefficient.

[0016] Furthermore, the window size adjustment rule of the adaptive sliding window algorithm in step S3 is as follows: calculate the wind speed change rate, specifically using the following formula:

[0017] When the wind speed change rate is greater than 10%, the sliding window size is set to 10 minutes to capture gust data at a high frequency; when the wind speed change rate is less than 5% and less than 10%, the sliding window size is set to 20 minutes; when the wind speed change rate is less than 5%, the sliding window size is set to 30 minutes to reduce the amount of calculation during stable periods.

[0018] Furthermore, the reinforcement learning algorithm described in step S3 is the Q-learning algorithm, and the parameter configuration of the Q-learning algorithm includes: The state space S = (credibility score vector, spatial correlation coefficient, time rate of change), the action space A is the weight adjustment amount of each data source, with an adjustment range of -5% to +5%, and the reward function R is calculated using the following formula:

[0019] A reinforcement learning environment was built using the gym library, with a learning rate of α=0.1 and a discount factor of γ=0.9. Iterative training was stopped when the weight change was less than 0.1% for 100 consecutive steps. The weight limit for a single data source is set to 80% to avoid simulation bias caused by a single data source dominating the data.

[0020] Furthermore, the simulation framework mentioned in step S4 is the improved Windpowerlib framework, and the dynamic configuration of the fan parameters includes: Call the WindTurbine class of Windpowerlib to configure the wind turbine model, power curve and hub height. The hub height is set according to the actual parameters of the wind farm, and the value range is 80m~150m. In the wind turbine power curve calculation module, wind speed-output data provided by the wind turbine manufacturer is imported through pandas.DataFrame, and cubic polynomial fitting is performed using numpy.polyfit(degree=3) to optimize the smoothness of the power curve.

[0021] Furthermore, the wake effect simulation module in step S4 adopts the Jensen wake model. The parameters of the Jensen wake model include a wake radius expansion coefficient of 0.07 and a wind speed attenuation coefficient of 0.8 at the hub height. The overall power output calculation formula for the wind farm is as follows:

[0022] The air density correction module calculates air density using the ideal gas law, which is as follows:

[0023] in, The air pressure is hPa. For temperature K, .

[0024] Furthermore, the real-time operation data of the wind farm mentioned in step S5 is collected by connecting to the wind farm SCADA system through the pymodbus library. The collection cycle is 1 minute / time, and the collected data includes the actual output of the unit, the nacelle temperature and the gearbox status.

[0025] Compared with existing technologies, the present invention provides a method for dynamic fusion of multi-source meteorological data in wind farm simulation. The method includes: accessing satellite remote sensing, ground station, numerical weather prediction, and historical wind farm operation data; performing format parsing, standardization processing, and cleaning to output standardized meteorological data; calculating a reliability score based on multi-dimensional indicators such as data integrity and time consistency to achieve abnormal data repair or backup data source switching; combining wind farm DEM terrain features and an adaptive sliding window, adjusting data source weights in real time through reinforcement learning to generate dynamic weight values; improving the Windpowerlib framework to adapt to the fused data, dynamically configuring wind turbine parameters, and generating initial simulation results; collecting real-time wind farm data and using Kalman filtering to correct model parameters. This invention can improve the accuracy and timeliness of wind farm simulation, enhance data utilization and model adaptability, and reduce the deviation between simulation results and actual operating conditions. Attached Figure Description

[0026] Figure 1 A flowchart of a wind farm simulation method provided in an embodiment of the present invention; Figure 2 A flowchart for dynamic weight calculation provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the data processing of the windpowerlib extension interface provided in an embodiment of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0028] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0029] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0030] Example 1 See Figure 1 , Figure 1 This invention proposes a multi-source meteorological data dynamic fusion wind farm simulation method. This method, through the coordinated operation of five core steps, forms a complete wind farm simulation process, applicable to scenarios such as early planning (turbine layout optimization), operation scheduling (output prediction), and fault early warning (extreme weather load simulation) for onshore and offshore wind farms. Specific steps may include: S1. Multi-source meteorological data acquisition and preprocessing: Satellite remote sensing data, ground station data, numerical weather prediction data, and historical wind farm operation data are accessed via a dynamic interface. After adaptive parsing of the multi-source data, data standardization and cleaning are performed to output standardized meteorological data. Specifically, this includes: S11. Multi-source data integration: Access satellite remote sensing data (such as Sunflower satellite wind speed products), ground station data (wind speed, wind direction, temperature, air pressure, etc.), numerical forecast data (such as ECMWF forecast data), and historical wind farm operation data through dynamic interfaces, and support adaptive parsing of multiple data formats such as NetCDF, CSV, and JSON.

[0031] Specifically, Python is used to write the data acquisition script, and the requests library is integrated to implement HTTP / HTTPS protocol data requests. The APScheduler scheduling framework is used to set the acquisition cycle to 1 minute / time, and data is obtained from satellite data platforms (such as NASAEarthdata), ground station data centers (such as the National Meteorological Information Center API), and numerical forecast databases (such as ECMWFWebAPI).

[0032] Customized data parsing logic was developed to suit the interface characteristics of different data sources: satellite remote sensing data (NetCDF format) was parsed using the xarray library, ground station data (CSV format) was read using the pandas library, and numerical weather prediction data (JSON format) was directly parsed using the json library.

[0033] S12. Data standardization processing: The Windpowerlib framework is further developed to add a data adaptation layer, which converts multi-source heterogeneous data into a unified format, including coordinate system 1 (converted to WGS84 coordinate system), unit conversion (wind speed is unified to m / s), and timestamp synchronization (unified to UTC time).

[0034] Specifically, see Figure 3 Develop an extended interface adaptation layer, using custom functions. Achieving heterogeneous data transformation: Coordinate transformation: Based on the Transformer class of the pyproj library, the latitude and longitude of satellite data (WGS84) and the local coordinate system of the ground station are converted to the UTM projected coordinate system. The formula is as follows: (XX is the UTM partition number).

[0035] Unit conversion: Wind speed units will be standardized to m / s (e.g., converting km / h to m / s: wind speed (m / s) = wind speed (km / h) / 3.6). Timestamps will be standardized to UTC format (via...). accomplish).

[0036] S13. Data Cleaning: The Pandas library in Python is used to clean the data, removing outliers (such as wind speed values ​​that exceed the physical reasonable range) and duplicate values. Missing data is filled in using linear interpolation or spline interpolation to ensure data continuity.

[0037] Specifically, outlier detection: Outliers are identified using the IQR (interquartile range) method, calculated as follows:

[0038] in, The lower quartile, Upper quartile, Remove them using the pandas.DataFrame.drop() method.

[0039] Duplicate value handling: Call the pandas.DataFrame.drop.duplicates(subset=[timestamp','latitude and longitude']) method to remove duplicates based on time and spatial dimensions.

[0040] Missing value completion: For continuous meteorological data (such as wind speed and temperature), cubic spline interpolation is used to complete the missing value calculation. To ensure the continuity of the time series, nearest neighbor interpolation is used to complete discrete data (such as weather types).

[0041] S2. Meteorological Data Quality Assessment: Using the standardized meteorological data output from step S1 as input, a multi-dimensional assessment index system is constructed based on data integrity, temporal consistency, spatial correlation, and historical error rate. A weighted algorithm is used to calculate the reliability score of the standardized meteorological data. Based on the score, anomaly marking, Kriging interpolation repair, or triggering backup data source switching are performed on the data, and the data reliability score is output. Specifically, this includes: S21. Construction of Evaluation Index System: Evaluation indicators are constructed from four dimensions: data integrity, temporal consistency, spatial correlation, and historical error rate. Data integrity is measured by the missing rate; temporal consistency is judged by detecting data mutation points; spatial correlation is determined by deviation analysis with data from surrounding stations; and the historical error rate is calculated based on the historical deviation between the data source and the actual observation value.

[0042] Specifically, data integrity is assessed using a data integrity score, calculated using the following formula:

[0043] The scoring rules are as follows: missing rate ≤ 5%: 100 points; 5% < missing rate ≤ 10%: 80 points; 10% < missing rate ≤ 15%: 60 points; missing rate > 15%: 40 points (triggering a switch to a backup data source).

[0044] Temporal consistency was assessed using a temporal consistency score, and a mutation point was detected using the Grubbs test. Achievement (significance level α = 0.05). If the p-value < 0.05, it is considered a mutation point. Scoring rules: Base score 100 points, deduct 10 points for each mutation point detected, with a minimum score of 40 points.

[0045] Spatial correlation was assessed using a spatial correlation score, selecting the three nearest stations around the target data point (distances were calculated using the Haversine formula, and stations less than 5 km were selected), and the average deviation was calculated.

[0046] Scoring rules: Average deviation < 1 m / s: 100 points; 1 m / s ≤ average deviation ≤ 2 m / s: 85 points; Average deviation > 2 m / s: 70 points.

[0047] Historical error is assessed using a historical error score, which calculates the mean absolute percentage error (MAPE) based on data from the past 30 days. The specific formula is as follows:

[0048] Scoring rules: MAPE < 5%: 100 points; 5% ≤ MAPE ≤ 10%: 80 points; 10% < MAPE ≤ 15%: 60 points; MAPE > 15%: 40 points.

[0049] S22. Credibility score calculation: Use a weighted algorithm to calculate each indicator and generate a credibility score ranging from 0 to 100 points. Among them, the data integrity accounts for 30%, the time consistency accounts for 25%, the spatial correlation accounts for 25%, and the historical error rate accounts for 20%. The credibility score adopts a weighted calculation formula, and the specific formula is: , retain 1 decimal place.

[0050] S23. Data screening and repair: Mark the data with a score lower than the threshold (such as 60 points). For slightly abnormal data, combine it with the Kriging interpolation method for repair; for severely abnormal data, trigger the backup data source switching mechanism.

[0051] S3. Spatiotemporal dynamic modeling and adaptive weight calculation: Take the data credibility score output in step S2 as the core input, combine the terrain features extracted from the digital elevation model of the wind farm to establish a spatial correlation matrix, use the adaptive sliding window algorithm to extract the time trend of meteorological data, and adjust the weights of each data source in real time through the reinforcement learning algorithm to output the dynamic weight value; refer to Figure 2 , specifically including: S31. Spatial correlation analysis: Based on the digital elevation model (DEM) data of the wind farm, extract terrain features (slope, aspect, terrain undulation), analyze the spatial distribution characteristics of meteorological data (such as wind speed gradient caused by terrain), and establish a spatial correlation matrix of meteorological data in different regions.

[0052] Terrain data processing: Use the GDAL library to read 30m resolution DEM data, and extract the slope ( ), aspect ( ) and other features through the rasterio library.

[0053] Calculation of wind speed spatial attenuation coefficient: Based on terrain features, construct an exponential attenuation model, and the specific formula is:

[0054] Among them, is the terrain influence coefficient, k = 0.05 in mountainous areas, k = 0.02 in plains, and establish a spatial correlation matrix of 50m × 50m grid.

[0055] S32. Time Series Analysis: A sliding window algorithm is used (the window size is adaptively adjusted according to the rate of change of meteorological data; for example, the window is set to 10 minutes when wind speed changes drastically and 30 minutes when it is stable) to extract the time trend of meteorological data (such as intraday wind speed variation patterns and seasonal fluctuation characteristics). The sliding window adaptively adjusts the window size based on the rate of change of wind speed, and the specific formula is as follows:

[0056] When the rate of change is >10%, the window size is 10 minutes (for high-frequency capture of gusts); when the rate of change is 5% ≤ 10%, the window size is 20 minutes; when the rate of change is <5%, the window size is 30 minutes (to reduce computation during stable periods).

[0057] Time trend extraction: through Apply moving average filtering to extract intraday / seasonal trend features of wind speed and direction.

[0058] S33. Dynamic Weight Generation: Based on data quality assessment results and spatiotemporal correlation analysis, the weights of each data source are adjusted in real time using reinforcement learning algorithms. For example, under clear sky conditions, the weight of satellite remote sensing data is increased (60%); in complex terrain areas, the weight of ground station data is increased (50%); and during periods of high accuracy in numerical weather prediction data, its weight is increased (e.g., 40% at night). Simultaneously, upper and lower limits are set for weights (the weight of a single data source should not exceed 80%) to avoid bias caused by a single data source dominating the data.

[0059] Constructing a weight adjustment model based on the Q-learning algorithm: State space: S = (credibility score vector, spatial correlation coefficient, rate of change over time); Action space: A = Weight adjustment amount for each data source (-5% to +5%); Reward function: (The smaller the deviation, the higher the reward).

[0060] A reinforcement learning environment was built using the gym library, with a learning rate of α=0.1 and a discount factor of γ=0.9. The training was iteratively conducted until the weights converged (the weight change was less than 0.1% for 100 consecutive steps).

[0061] S4. Wind Farm Simulation Based on Simulation Framework: Using the dynamically weighted and fused meteorological data output from step S3 as input, the simulation framework is further developed to adapt to the fused data. Wind turbine parameters are dynamically configured, and the model's wind turbine power curve calculation module, wake effect simulation module, and air density correction module are called to generate initial simulation results for the wind farm; specifically including: S41. Input of fused data: Input the weighted and fused meteorological data into the improved Windpowerlib model, call the core functions of the model such as the wind turbine power curve calculation module, the wake effect simulation module, and the air density correction module, and generate simulation results such as wind farm output, unit load, and power generation efficiency.

[0062] S42. Dynamic Configuration of Model Parameters: Based on the geographical location of the wind farm, the type of wind turbine (e.g., horizontal axis wind turbine, vertical axis wind turbine), and the installed capacity, dynamically configure the parameters of the Windpowerlib model, such as hub height, blade length, cut-in wind speed, and cut-out wind speed, to ensure that the model matches the actual scenario. Specifically, this includes: S421. Call the WindTurbine class in Windpowerlib, with the following parameter settings: Fan model: such as "VestasV164-9.5MW"; Power curves: Import power curve data (wind speed-output relationship) provided by the manufacturer using pandas.DataFrame, and fit it with a cubic polynomial. Optimize curve smoothness; Hub height: Set according to the actual parameters of the wind farm (e.g., 130m).

[0063] S422, Single Fan Output Calculation: Input the merged meteorological data (wind speed, air density) The method involves correcting air density using the ideal gas law, with the specific formula as follows:

[0064] in, The air pressure is hPa. For temperature K, .

[0065] S423. Overall wind farm output correction: The WakeLosses class is called, and the Jensen wake model is used to calculate the wake effect. Parameters include: Wake radius expansion coefficient: 0.07; Wind speed attenuation coefficient at wheel hub height: 0.8; Overall output calculation formula: Total output = Σ single fan output × (1 - wake loss rate).

[0066] S5. Dynamic Correction of Simulation Results: Collect real-time operating data of the wind farm and calculate the deviation between the data and the initial simulation results output in step S4. When the deviation exceeds a set threshold, dynamically correct the key parameters of the simulation framework model using a Kalman filter algorithm until the deviation is reduced to within the threshold, and output the final wind farm simulation results. Specifically, this includes: S51. Real-time Data Introduction: Collect real-time operating data of the wind farm (such as actual turbine output, nacelle temperature, gearbox status, etc.) and calculate the deviation from the simulation results. Connect to the wind farm's SCADA system via the pymodbus library, collecting actual turbine output data at a frequency of 1 minute / time, and calculate the relative error. The specific formula is as follows:

[0067] S52. Parameter Correction: When the deviation exceeds a set threshold (e.g., 5%), the key parameters of the Windpowerlib model (such as air density coefficient and surface roughness coefficient) are dynamically corrected using the Kalman filter algorithm until the deviation is reduced to within the threshold, forming a closed-loop optimization of "fusion data - simulation results - actual data - parameter correction". Specifically, Kalman filter parameter correction includes: When the deviation is greater than 5%, the Kalman filter is activated: State equation: x(k) = A × x(k-1) + w(k) (x is the parameter vector to be corrected, A is the state transition matrix, and w is the process noise); Observation equation: z(k) = H × x(k) + v(k) (z is the biased observation value, H is the observation matrix, and v is the observation noise); Corrected parameters: air density coefficient (initial value 1.225 kg / m³), surface roughness (initial value 0.03 m), iterate until the deviation is ≤5% and stop.

[0068] S53. Simulation Result Update: Re-enter the corrected parameters into the Windpowerlib model and call... The WakeLosses class is recalculated to generate corrected wind farm simulation results.

[0069] In summary, the present invention has the following advantages: 1. Significantly improved simulation accuracy: Breaking through the traditional static fusion mode, the weights of each data source are adjusted in real time based on data quality assessment and spatiotemporal correlation analysis, realizing the dynamic adaptation of meteorological data with time and space, reducing the simulation error under complex meteorological conditions (such as typhoons, gusts, and areas with complex terrain) by more than 30%, and solving the problem of insufficient accuracy of existing methods in dynamic meteorological scenarios. 2. Improved data utilization efficiency and reliability: A multi-dimensional meteorological data quality assessment system (including completeness, temporal consistency, spatial correlation, and historical error rate) was constructed. Data quality was quantified through credibility scoring, reducing interference from low-quality data. Data utilization rate increased from 60% to 90%. At the same time, the reliability of input data was further guaranteed through abnormal data repair and backup data source switching mechanisms. 3. Enhanced model adaptability and scalability: The Windpowerlib framework has been further developed, and a new multi-source heterogeneous data dynamic access adaptation layer has been added. It supports standardized conversion of multiple data formats such as NetCDF, CSV, and JSON, as well as minute-level high-frequency data access. This expands the data source adaptation capability of the framework, reduces the technical threshold for multi-source data fusion, and can flexibly adapt to different types of wind farms (onshore and offshore) and meteorological data sources. 4. Improved consistency between simulation results and actual operating conditions: Real-time wind farm operation data is introduced as a feedback signal, and key model parameters (such as air density coefficient and surface roughness coefficient) are dynamically corrected through Kalman filtering algorithm, forming a closed-loop optimization of "fusion data - simulation results - actual data - parameter correction", so that the deviation between simulation results and actual output is controlled within 5%, and the ability to reflect the dynamic operating status of wind farm is improved.

[0070] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for simulating wind farms by dynamically fusing multi-source meteorological data, characterized in that, include: S1. Access satellite remote sensing data, ground station data, numerical forecast data and historical wind farm operation data through dynamic interface and generate multi-source data. After performing format adaptive parsing on the multi-source data, perform data standardization processing and data cleaning, and output standardized meteorological data. S2. Construct a multi-dimensional evaluation index system using the standardized meteorological data, calculate the credibility score of the standardized meteorological data using a weighted algorithm, mark the data as abnormal, perform Kriging interpolation repair or trigger the switching of backup data sources based on the score, and output the data credibility score. S3. Based on the data credibility score, a spatial correlation matrix is ​​established by combining the terrain features extracted from the wind farm digital elevation model. The time trend of meteorological data is extracted by the adaptive sliding window algorithm. The weights of each data source are adjusted in real time by the reinforcement learning algorithm, and the dynamic weight values ​​are output. S4. Based on the meteorological data after weighted fusion of the dynamic weight values, the simulation framework is further developed to adapt to the fused data, the wind turbine parameters are dynamically configured, and the wind turbine power curve calculation module, wake effect simulation module and air density correction module of the model are called to generate the initial simulation results of the wind farm. S5. Collect real-time operating data of the wind farm and calculate the deviation between the data and the initial simulation results of the wind farm. When the deviation exceeds the set threshold, dynamically correct the key parameters of the simulation framework model through the Kalman filter algorithm until the deviation is reduced to within the threshold and output the final simulation results of the wind farm.

2. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 1, characterized in that, The formats of the multi-source data include NetCDF, CSV and JSON. Adaptive parsing of the format is implemented using the xarray library, pandas library and json library. The data standardization process includes: secondary development of the Windpowerlib framework, adding a data adaptation layer, and converting multi-source heterogeneous data into a unified format; The data cleaning includes: using Python's Pandas library to clean the data, removing outliers and duplicates, and using linear interpolation or spline interpolation to fill in missing data to ensure data continuity.

3. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 2, characterized in that, The collection of multi-source data in step S1 is implemented by writing a collection script in Python. The collection script integrates the requests library to support HTTP / HTTPS protocol data requests and uses the APScheduler scheduling framework to set the collection cycle to 1 minute / time. The sources of the multi-source data include: satellite data platform NASAEarthdata, ground station data center NPC API, and numerical weather prediction database ECMWFWebAPI.

4. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 1, characterized in that, The multi-dimensional evaluation indicators include: data integrity, temporal consistency, spatial correlation, and historical error rate; The specific weighting for the credibility score is as follows: data integrity 30%, temporal consistency 25%, spatial correlation 25%, and historical error rate 20%; the credibility score calculation formula is: The result is rounded to one decimal place.

5. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 1, characterized in that, The establishment of the spatial correlation matrix in step S3 includes: Based on wind farm digital elevation model (DEM) data, 30m resolution DEM data was read using the GDAL library, and terrain features were extracted using the rasterio library. These terrain features included slope and aspect, and the slope calculation formula was as follows: The formula for calculating slope aspect is: Based on the aforementioned terrain features, a spatial attenuation coefficient model for wind speed is constructed. The formula for the attenuation coefficient model is as follows: in, The terrain influence coefficient is used in mountainous scenes. In the plains scene A spatial correlation matrix of a 50m×50m grid is established based on the attenuation coefficient.

6. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 5, characterized in that, The window size adjustment rule of the adaptive sliding window algorithm in step S3 is as follows: calculate the wind speed change rate, and the specific calculation formula is: When the wind speed change rate is greater than 10%, the sliding window size is set to 10 minutes to capture gust data at a high frequency; when the wind speed change rate is less than 5% and less than 10%, the sliding window size is set to 20 minutes; when the wind speed change rate is less than 5%, the sliding window size is set to 30 minutes to reduce the amount of calculation during stable periods.

7. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 6, characterized in that, The reinforcement learning algorithm mentioned in step S3 is the Q-learning algorithm, and the parameter configuration of the Q-learning algorithm includes: State space S = (credibility score vector, spatial correlation coefficient, time rate of change), action space A represents the weight adjustment of each data source, with an adjustment range of -5% to +5%, and the reward function R is calculated using the following formula: A reinforcement learning environment was built using the gym library, with a learning rate of α=0.1 and a discount factor of γ=0.

9. Iterative training was stopped when the weight change was less than 0.1% for 100 consecutive steps. The weight limit for a single data source is set to 80% to avoid simulation bias caused by a single data source dominating the data.

8. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 1, characterized in that, The simulation framework mentioned in step S4 is the improved Windpowerlib framework, and the dynamic configuration of the wind turbine parameters includes: Call the WindTurbine class of Windpowerlib to configure the wind turbine model, power curve and hub height. The hub height is set according to the actual parameters of the wind farm, and the value range is 80m~150m. In the wind turbine power curve calculation module, wind speed-output data provided by the wind turbine manufacturer is imported through pandas.DataFrame, and cubic polynomial fitting is performed using numpy.polyfit(degree=3) to optimize the smoothness of the power curve.

9. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 8, characterized in that, The wake effect simulation module in step S4 uses the Jensen wake model. The parameters of the Jensen wake model include a wake radius expansion coefficient of 0.07 and a wind speed attenuation coefficient of 0.8 at the hub height. The overall power output calculation formula for the wind farm is as follows: The air density correction module calculates air density using the ideal gas law, which is as follows: in, The air pressure is hPa. For temperature K, .

10. The method for dynamic fusion of multi-source meteorological data to simulate wind farms according to claim 1, characterized in that, The real-time operation data of the wind farm mentioned in step S5 is collected by connecting to the wind farm SCADA system through the pymodbus library. The collection cycle is 1 minute / time. The collected data includes the actual output of the unit, the nacelle temperature and the gearbox status.

Images