Wind energy resource prediction method, system and device and storage medium

Abstract

The invention discloses a wind energy resource prediction method, system and device and a storage medium, and relates to the field of renewable energy sources, and the method comprises the steps: obtaining the three-dimensional wind field observation data of a target region; performing space-time alignment processing operation on the three-dimensional wind field observation data to generate a space-time sequence; performing multi-source data fusion on the space-time sequence, and constructing a multi-dimensional wind field matrix; establishing a wind energy potential prediction model based on the multi-dimensional wind field matrix; and predicting to-be-predicted wind energy data based on the wind energy potential prediction model to obtain a wind energy resource prediction result. And efficient, accurate and reliable resource evaluation is provided for wind energy development of various terrain areas by constructing a machine learning model.

Description

Technical Field

[0001] The present invention relates to the field of renewable energy, and in particular to a wind energy resource prediction method, system, device and storage medium. Background Art

[0002] As an important clean and renewable energy source, the development of wind energy relies on accurate wind resource assessment. Currently, traditional assessment methods rely primarily on ground-based and mesoscale meteorological data collected from fixed meteorological towers or weather stations to measure parameters such as wind speed, direction, temperature, and humidity. This data provides a foundation for wind farm site selection and design, directly impacting their power generation efficiency and economic benefits.

[0003] Traditional methods suffer from limitations such as limited data collection points, limited coverage, and high costs, making them inadequate for wind energy assessments in complex terrain or large areas. In recent years, drone technology, with its flexible deployment, rapid response, and multi-point data collection capabilities, has provided a new approach for wind energy assessments. However, efficiently integrating the massive amounts of data collected by drones to improve assessment accuracy and efficiency remains a pressing technical challenge. Summary of the Invention

[0004] The main purpose of the present invention is to provide a wind energy resource prediction method, system, device and storage medium, which can provide efficient, accurate and reliable resource assessment for wind energy development in various terrain areas by constructing a machine learning model.

[0005] In order to achieve the above objectives, the embodiments of the present application provide the following technical solutions:

[0006] According to a first aspect of an embodiment of the present application, a method for predicting wind energy resources is provided, the method comprising:

[0007] Obtain three-dimensional wind field observation data in the target area;

[0008] Performing a spatiotemporal alignment processing operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence;

[0009] Performing multi-source data fusion on the spatiotemporal sequence to construct a multi-dimensional wind field matrix;

[0010] Establishing a wind energy potential prediction model based on the multidimensional wind field matrix;

[0011] The wind energy data to be predicted is predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

[0012] Optionally, obtaining three-dimensional wind field observation data of the target area includes:

[0013] Generate layered track planning instructions based on the terrain data of the target area and the vertical gradient data of the wind tower; the vertical gradient data of the wind tower includes wind energy parameters at a plurality of preset heights within a preset height range;

[0014] Invoking a drone formation to cruise at a plurality of vertically spaced altitude layers within the preset altitude interval according to the layered trajectory planning instructions, with the cruising speed maintained within a preset range; the drone formation comprising a plurality of drones equipped with meteorological sensors;

[0015] Three-dimensional wind field observation data is collected in real time by each drone, and the collection time, spatial coordinates and attitude data are recorded; the sampling density is dynamically adjusted according to the vertical distribution law of wind speed, and the sampling frequency of the high altitude layer is set to be greater than that of the low altitude layer.

[0016] Optionally, performing a spatiotemporal alignment operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence includes:

[0017] Based on a preset physical threshold range and a wind tower data setting condition, abnormal data points in the three-dimensional wind field observation data are eliminated; the physical thresholds include wind speed, wind direction, and temperature; the wind tower data setting condition includes setting a data missing rate within a continuous time period to be lower than a set missing rate threshold;

[0018] Use the satellite timing system to unify the time base of each drone and convert the local coordinate system data of each drone into a unified geographic coordinate system;

[0019] The spatiotemporal sequence is generated by combining the real-time atmospheric parameter correction observation data.

[0020] Optionally, multi-source data fusion is performed on the spatiotemporal sequence to construct a multi-dimensional wind field matrix, including:

[0021] A spatial interpolation algorithm is used to grid the wind field data at discrete observation points to generate continuous spatial distribution data. The spatial interpolation algorithm dynamically adjusts the interpolation weight in the vertical direction based on the wind shear index, which is obtained by fitting historical wind measurement data.

[0022] A time series prediction model is used to fill in the missing wind field data to generate a time continuous sequence; the input of the time series prediction model includes the dominant wind direction distribution characteristics and historical wind field change trends;

[0023] The continuous spatial distribution data and the time continuous sequence are integrated through a probabilistic inference algorithm to generate a multidimensional wind field matrix with a confidence level that meets preset conditions; the multidimensional wind field matrix includes multidimensional data of spatial coordinates, timestamps, wind field parameters and confidence ratings.

[0024] Optionally, performing a wind energy potential modeling operation based on the multidimensional wind field matrix to generate a wind energy potential prediction model includes:

[0025] Extracting multidimensional wind field features from the multidimensional wind field matrix; the multidimensional wind field features include wind speed mean, wind direction frequency, turbulence intensity, and wind shear index; the feature extraction is performed based on a preset safety boundary condition, and the safety boundary condition is set according to historical maximum wind speed statistics;

[0026] A machine learning algorithm is used to establish a mapping relationship between wind energy density and the multidimensional wind field characteristics to obtain the wind energy potential prediction model; the parameter settings of the machine learning algorithm meet the following set optimization conditions: the number of input layer nodes matches the dimension of the multidimensional wind field characteristics; the output layer contains wind energy density and corresponding probability distribution parameters; the training process is constrained by the turbulence intensity index to ensure that the model output conforms to the laws of fluid mechanics;

[0027] The effectiveness of the wind energy potential prediction model is evaluated through physical statistical verification and precision verification; the physical statistical verification includes fitting the wind energy distribution output by the model with the Weibull distribution to verify the rationality of the parameters; the precision verification includes ensuring that the model prediction error is lower than the set threshold through cross-validation method.

[0028] Optionally, a machine learning algorithm is used to establish a mapping relationship between wind energy density and the multi-dimensional wind field characteristics to obtain the wind energy potential prediction model according to the following formula:

[0029]

[0030] The following constraints are met:

[0031] y i -w T φ(x i )-b≤∈+ξ i

[0032]

[0033] Among them, w is the weight vector of multidimensional wind field characteristics, b is the bias term; φ(xi) is the kernel function that maps the multidimensional wind field characteristics xi to the high-dimensional space; yi is the target value of the wind field parameter; ∈ is the threshold of the insensitive loss function; C is the regularization parameter that controls the model complexity and error tolerance; ξi and ξi* are slack variables that allow samples to break through the ∈ band.

[0034] Optionally, predicting the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result includes:

[0035] Inputting the wind energy data to be predicted into the wind energy potential prediction model to extract the target wind field characteristics; and fitting the return period wind speed based on the extreme value distribution model to generate a wind speed extreme value prediction curve; the wind speed extreme value prediction curve includes the maximum wind speed values corresponding to several return periods and is marked with a risk threshold;

[0036] Generate a wind power density map based on wind energy density distribution characteristics; the wind energy density distribution characteristics include vertical layer wind energy density change curves and real-time atmospheric parameter correction observation data;

[0037] Determine the restricted area for wind turbine deployment based on the set turbulence intensity threshold and wind speed extreme value prediction results; the restricted area determination conditions include turbulence intensity exceeding the preset safety threshold and maximum wind speed exceeding the wind turbine wind resistance limit;

[0038] The wind energy resource prediction result is output, wherein the wind energy resource prediction result includes the wind speed extreme value prediction curve, the wind power density map and the wind turbine deployment restriction area.

[0039] According to a second aspect of an embodiment of the present application, a wind energy resource prediction system is provided, the system comprising:

[0040] A data acquisition module is used to obtain three-dimensional wind field observation data of the target area;

[0041] A spatiotemporal sequence module, configured to perform spatiotemporal alignment processing on the three-dimensional wind field observation data to generate a spatiotemporal sequence;

[0042] A wind field matrix module is used to fuse multi-source data of the spatiotemporal sequence and construct a multi-dimensional wind field matrix;

[0043] A model building module, configured to build a wind energy potential prediction model based on the multi-dimensional wind field matrix;

[0044] The prediction module is used to predict the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

[0045] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect above.

[0046] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. The computer-readable instructions can be executed by a processor to implement the method described in the first aspect above.

[0047] In summary, the embodiments of the present application provide a wind energy resource prediction method, system, device, and storage medium. These methods obtain three-dimensional wind farm observation data for a target area; perform spatiotemporal alignment processing on the three-dimensional wind farm observation data to generate a spatiotemporal sequence; perform multi-source data fusion on the spatiotemporal sequence to construct a multidimensional wind farm matrix; establish a wind energy potential prediction model based on the multidimensional wind farm matrix; and predict the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result. By constructing a machine learning model, efficient, accurate, and reliable resource assessment is provided for wind energy development in various terrain areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying any creative work.

[0049] The structures, proportions, sizes, etc. illustrated in this specification are intended solely to complement the contents disclosed herein and to facilitate understanding and reading by persons skilled in the art. They are not intended to limit the conditions under which the present invention may be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportions, or adjustments in sizes, without affecting the efficacy and objectives of the present invention, shall remain within the scope of the technical contents disclosed herein.

[0050] Figure 1 A schematic diagram of a wind energy resource prediction method flow chart provided in an embodiment of the present application;

[0051] Figure 2 A comparison chart of average wind speeds at various altitudes provided in the embodiments of this application;

[0052] Figure 3 A schematic diagram of wind tower data provided in an embodiment of the present application;

[0053] Figure 4 A schematic diagram of a wind rose provided in an embodiment of the present application;

[0054] Figure 5 Schematic diagram of a wind energy resource prediction system provided in an embodiment of the present application;

[0055] Figure 6 A structural diagram of an electronic device provided in an embodiment of the present application is shown;

[0056] Figure 7 A diagram showing a computer-readable storage medium provided in an embodiment of the present application.

[0057] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, left, right, front, back, etc.) are only used to explain the relative position relationship, movement status, etc. between the various components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0060] In addition, the terms "first," "second," and so on, used in this disclosure are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referenced. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this disclosure, "plurality" means at least two, such as two or three, unless otherwise specifically defined.

[0061] In the present invention, unless otherwise specified or limited, the terms "connection" and "fixation" should be understood in a broad sense. For example, "fixation" can mean fixed connection, detachable connection, or integration; mechanical connection or electrical connection; direct connection or indirect connection through an intermediate medium; internal communication between two elements or interaction between two elements, unless otherwise specified. Those skilled in the art will be able to understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0062] In addition, the technical solutions between the various embodiments of the present invention can be combined with each other, but it must be based on the fact that ordinary technicians in this field can implement it. When the combination of technical solutions is mutually contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0063] Figure 1 A wind energy resource prediction method provided in an embodiment of the present application is shown, the method comprising:

[0064] Step 101: Acquire three-dimensional wind field observation data of the target area;

[0065] Step 102: performing a spatiotemporal alignment processing operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence;

[0066] Step 103: performing multi-source data fusion on the spatiotemporal sequence to construct a multi-dimensional wind field matrix;

[0067] Step 104: establishing a wind energy potential prediction model based on the multi-dimensional wind field matrix;

[0068] Step 105: Predicting the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

[0069] In a possible implementation, in step 101, obtaining three-dimensional wind field observation data of a target area includes:

[0070] A layered trajectory planning instruction is generated based on the terrain data of the target area and the vertical gradient data of the wind tower; the vertical gradient data of the wind tower includes wind energy parameters at several preset heights within a preset height interval; a drone formation is called to perform cruising at several vertically spaced altitude layers within the preset height interval according to the layered trajectory planning instruction, and the cruising speed is maintained within a preset range; the drone formation includes several drones equipped with meteorological sensors; three-dimensional wind field observation data is collected in real time by each drone, and the collection time, spatial coordinates and attitude data are recorded; the sampling density is dynamically adjusted according to the vertical distribution law of wind speed, wherein the sampling frequency of the high altitude layer is set to be greater than that of the low altitude layer.

[0071] By performing multi-layered cruises within preset altitude ranges (e.g., 50m-200m), the system acquires vertical wind speed and direction gradient data, addressing the inability of traditional single-point wind towers to capture wind shear index. Dynamically adjusting flight paths based on terrain data ensures complete coverage of areas like ridges and canyons, avoiding data blind spots caused by terrain obstruction. This system provides high-resolution, highly reliable input data for wind resource assessment, supporting wind turbine selection, layout optimization, and risk warning, ensuring maximum return on wind farm investment.

[0072] Multiple drones collect data in parallel, covering a larger spatial range in the same time (for example, a 15km×15km area takes only 36 hours), which is 5 times more efficient than traditional mobile wind measurement vehicles. The sampling density is dynamically adjusted according to the vertical distribution of wind speed (the wind speed near the ground layer changes dramatically): low altitude layer (0-100m): sampling interval 50m, frequency 1Hz, capture turbulence intensity I T=σv / μv transient changes; high altitude layer (100-200m): sampling interval 20m, frequency 10Hz, accurately quantify the vertical gradient of high-altitude wind energy density P = 0.5ρv3. Using drone formations equipped with meteorological sensors, it can simultaneously collect three-dimensional wind speed, wind direction, temperature and humidity parameters at multiple altitude levels (e.g., 50m / 100m / 150m), capturing key indicators such as vertical wind shear and turbulence intensity.

[0073] Dynamically allocate sampling resources based on wind field characteristics (e.g., high-altitude wind speed increases logarithmically with altitude), prioritize high-altitude layer data acquisition (e.g., sampling point density at 200m is 3 times higher than that at 50m), and ensure that the vertical resolution of wind energy assessment model input data meets the IEC 61400-12-1 standard (vertical layer spacing ≤ 50m). Automatically trigger trajectory re-planning based on real-time wind speed data feedback (e.g., if a sudden change in wind speed exceeding 3σ is detected, temporarily increase the turbulence intensity I T For areas with undulating terrain, such as mountains and coastlines, layered track planning ensures data coverage without blind spots, overcoming the limitations of single-point observations on traditional wind towers.

[0074] In a possible implementation, in step 102, performing a spatiotemporal alignment operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence includes:

[0075] Abnormal data points in the three-dimensional wind field observation data are eliminated based on a preset physical threshold range and wind tower data setting conditions; the physical thresholds include wind speed, wind direction, and temperature; the wind tower data setting conditions include setting the data missing rate within a continuous time period to be lower than a set missing rate threshold; a satellite timing system is used to unify the time base of each drone, and the local coordinate system data of each drone is converted to a unified geographic coordinate system; the observation data is corrected in combination with real-time atmospheric parameters to generate the spatiotemporal sequence.

[0076] By setting physical thresholds for wind speed (0-60m / s) and temperature (-30°C-60°C), invalid data caused by sensor noise and communication packet loss is filtered out to ensure the physical plausibility of the input data. A missing rate threshold (e.g., <5%) is set for continuous time periods based on wind tower data to prevent data gaps caused by equipment failure or weather interference from affecting model training. Using the GPS / Beidou satellite timing system, the time base error of multiple drones is controlled to ≤1ms, eliminating timing misalignments caused by asynchronous sampling (e.g., phase deviations in gust events). Each drone's local coordinate system (Body Frame) data is converted to the WGS84 or UTM unified geographic coordinate system to compensate for the impact of drone attitude angles (pitch / roll) on wind speed vector measurements (correction ≥3% for pitch angles >5°). Multi-source heterogeneous data is converted into a unified spatiotemporal sequence format (timestamp-spatial coordinates-wind field parameters), providing standardized input for subsequent data fusion and modeling. Provide high-quality and highly consistent input data for wind energy potential prediction models, ensuring wind speed prediction error ≤ 0.5m / s (IEC L2 standard), supporting accurate wind farm planning and risk assessment.

[0077] In one possible implementation, in step 103, multi-source data fusion is performed on the spatiotemporal sequence to construct a multidimensional wind field matrix, including: gridding the wind field data of discrete observation points using a spatial interpolation algorithm to generate continuous spatial distribution data; the spatial interpolation algorithm dynamically adjusts the interpolation weight in the vertical direction based on the wind shear index, where the wind shear index is obtained by fitting historical wind measurement data; the wind field data in missing time periods is filled in by a time series prediction model to generate a time-continuous sequence; the input of the time series prediction model includes the dominant wind direction distribution characteristics and the historical wind field change trend; the continuous spatial distribution data and the time-continuous sequence are integrated by a probabilistic inference algorithm to generate a multidimensional wind field matrix with a confidence level that meets preset conditions; the multidimensional wind field matrix includes multidimensional data of spatial coordinates, timestamps, wind field parameters, and confidence ratings.

[0078] Discrete drone observation point data (such as 50m / 100m / 150m altitude layers) is converted into a continuous spatial distribution (grid resolution 100m×100m) through Kriging interpolation or inverse distance weighted (IDW) algorithms to address data sparseness in areas with complex terrain. The interpolation weights are dynamically adjusted based on the wind shear index fitted with historical data to accurately restore the characteristics of wind speed jumps. Utilizing LSTM (Long Short-Term Memory Network) or ARIMA models, based on the dominant wind direction distribution (such as northwest wind accounting for 60%) and historical wind speed change trends (such as daily cycle fluctuations of ±2m / s), data for missing periods caused by equipment failure or severe weather is predicted and filled.

[0079] This step utilizes a technical chain: spatial interpolation → time series prediction → probabilistic inference → matrix construction. Spatial interpolation algorithms (such as Kriging) grid the data from discrete observation points to generate a continuous spatial distribution, addressing the lack of coverage caused by sparse sampling. Time series prediction models (such as LSTM) are used to fill in missing periods of data due to weather or equipment failure, constructing a time-continuous series. Bayesian inference integrates multi-source data to generate a multidimensional matrix containing confidence ratings, providing reliability indicators for subsequent modeling.

[0080] In a possible implementation, in step 104, performing a wind energy potential modeling operation based on the multi-dimensional wind field matrix to generate a wind energy potential prediction model includes:

[0081] Multidimensional wind field features are extracted from the multidimensional wind field matrix; the multidimensional wind field features include wind speed mean, wind direction frequency, turbulence intensity and wind shear index; the feature extraction is performed based on preset safety boundary conditions, and the safety boundary conditions are set according to the historical maximum wind speed statistics; a machine learning algorithm is used to establish a mapping relationship between wind energy density and the multidimensional wind field features to obtain the wind energy potential prediction model; the parameter setting of the machine learning algorithm meets the following set optimization conditions: the number of input layer nodes matches the dimension of the multidimensional wind field features; the output layer contains wind energy density and corresponding probability distribution parameters; the training process is constrained by the turbulence intensity index to ensure that the model output conforms to the laws of fluid mechanics; the effectiveness of the wind energy potential prediction model is evaluated through physical statistical verification and precision verification; the physical statistical verification includes fitting the wind energy distribution output by the model with the Weibull distribution to verify the rationality of the parameters; the precision verification includes ensuring that the model prediction error is lower than the set threshold through cross-validation method.

[0082] The model's input feature set is constructed by extracting mean wind speed, wind direction frequency, turbulence intensity (IT = σv / μv), and wind shear index (α = ln(v2 / v1) / ln(z2 / z1)) from a multidimensional matrix. A safety boundary (e.g., vmax_hist = 50 m / s) is set based on historical maximum wind speed statistics, and outlier regions exceeding the safety threshold are excluded (e.g., data with wind speeds > 50 m / s are excluded from training). A mapping between wind energy density P = 0.5ρv3 and multidimensional features is established using the SVR algorithm to capture wind field variations in complex terrain.

[0083] Through a technical chain consisting of feature engineering, SVR modeling, physical verification, and precision control, the model output conforms to the Weibull distribution and is consistent with the statistical characteristics of historical data. Safety margins and confidence ratings are used to avoid high-risk areas, supporting wind turbine layout optimization and investment decisions. This provides a highly accurate and reliable prediction tool for wind energy development in complex terrain.

[0084] In one possible implementation, in step 104, a machine learning algorithm is used to establish a mapping relationship between wind energy density and the multi-dimensional wind field characteristics to obtain the wind energy potential prediction model according to the following formula:

[0085]

[0086] The following constraints are met:

[0087] y i -w T φ(x i )-b≤∈+ξ i

[0088]

[0089] Among them, w is the weight vector of multidimensional wind field characteristics, b is the bias term; φ(xi) is the kernel function that maps the multidimensional wind field characteristics xi to the high-dimensional space; yi is the target value of the wind field parameter; ∈ is the threshold of the insensitive loss function; C is the regularization parameter that controls the model complexity and error tolerance; ξi and ξi* are slack variables that allow samples to break through the ∈ band.

[0090] In a possible implementation, in step 105, predicting the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result includes:

[0091] The wind energy data to be predicted is input into the wind energy potential prediction model to extract the target wind field characteristics; and the recurrence wind speed is fitted based on the extreme value distribution model to generate a wind speed extreme value prediction curve; the wind speed extreme value prediction curve includes the maximum wind speed values corresponding to several recurrence periods and is marked with a risk threshold; a wind power density map is generated according to the wind energy density distribution characteristics; the wind energy density distribution characteristics include a vertical height layer wind energy density change curve and real-time atmospheric parameter correction observation data; the wind turbine deployment restriction area is determined according to the set turbulence intensity threshold and the wind speed extreme value prediction result; the restriction area judgment conditions include the turbulence intensity exceeding the preset safety threshold and the maximum wind speed exceeding the wind turbine wind resistance limit; the wind energy resource prediction result is output, and the wind energy resource prediction result includes the wind speed extreme value prediction curve, the wind power density map and the wind turbine deployment restriction area.

[0092] The extreme value distribution model predicts the maximum wind speed vmax for different return periods (50 / 100 years), marking high-risk areas where wind turbines exceed their wind resistance limits. The mean wind speed μv and turbulence intensity IT output by the SVR model in step 104 serve as inputs to the extreme value model. The extreme value distribution parameters μ, σ, and ξ are dynamically calibrated using historical data and real-time predictions. Based on the wind speed v predicted by the SVR model and the real-time atmospheric parameter corrections from step 103, the corrected wind power density P = 0.5ρv3 is calculated, generating a spatially continuous visualization map. Vertical gradient optimization: Wind energy density curves at key altitudes such as 70m, 100m, and 120m are extracted to guide tower height design and maximize energy capture efficiency. Complex risk areas are marked based on turbulence intensity thresholds (IT > 15%) and wind speed extremes exceeding the limit (vmax > vlimit) to mitigate operational and maintenance risks caused by the combination of high turbulence and high wind speed. Cost-benefit analysis data is provided when wind energy density-rich areas overlap with high-risk areas. The GIS platform superimposes wind power density heat maps, extreme wind speed contours, and restricted area markers to form an intuitive view of site selection decisions. Dynamic report generation: Output PDF reports containing theoretical power generation, equivalent full-power hours, and investment sensitivity analysis (such as the IRR fluctuation curve with electricity price). Through the technical chain of extreme prediction → resource quantification → risk marking → visual output, areas exceeding the IEC safety threshold are avoided and high-efficiency development areas with wind energy density exceeding the set threshold are accurately located. Visualization tools transform complex data into executable site selection plans, providing high-precision and high-reliability decision support tools for wind farm planning, ensuring that the development plan achieves the optimal balance between safety and economy.

[0093] The wind energy resource prediction method provided in the embodiment of the present application is described in detail below with reference to the accompanying drawings, and specifically includes the following stages:

[0094] The first stage is the construction of the drone formation system and data collection.

[0095] Step 1: System Construction and Sensor Configuration: Build a formation system consisting of multiple drones, with each drone equipped with the following sensors: anemometer (for measuring three-dimensional wind speed); wind vane (for recording horizontal and vertical wind direction); temperature sensor; humidity sensor. Customize the drones based on mission requirements to support rapid installation and calibration of sensor modules.

[0096] Step 2: Flight strategy design: trajectory planning and flight parameter setting.

[0097] 1. Trajectory Planning: Design a formation flight path based on the target area's terrain characteristics (shape, area), prioritizing coverage of areas sensitive to wind field fluctuations. Use a "I" or "P"-shaped spatial layout to ensure the measurement area has complete boundaries and no blind spots.

[0098] 2. Flight parameter settings: Altitude control: Prioritize increasing vertical sampling density and adopt a layered flight mode perpendicular to the horizontal plane (such as gradient climb). Speed adjustment: Dynamically adjust the flight speed based on measurement accuracy and endurance requirements, balancing data resolution and mission duration.

[0099] Vertical profile data such as wind shear index and turbulence intensity are acquired through layered flight (e.g., 50m intervals). A hover-climb alternating mode is used to improve vertical data continuity.

[0100] Step 3: Collaboration and data collection.

[0101] The fleet is divided into multiple collaborative working groups (for example, 3-5 drones per group) to achieve the following functions: Spatial coverage: Each group collects data synchronously in different areas according to preset strategies. Redundant backup: Overlapping observation groups are deployed in key areas to ensure data reliability. The following data are recorded in real time: vector data of wind speed / direction (including timestamp); scalar parameters of temperature and humidity; drone position (GPS coordinates) and attitude angle (IMU data). Dynamic grouping and reorganization are supported to adapt to mission requirements in complex terrain (such as mountainous areas and coastlines). The flight layout can be switched to "snake" and "U-shaped" modes to cope with different measurement scenarios.

[0102] The second stage is the design and implementation of data collection strategy.

[0103] Step 1: Track Planning and Collaborative Formation Deployment: A multi-dimensional track planning scheme is developed based on the target area's terrain characteristics and wind field distribution patterns. A grid-based scanning path is used for plains, with equally spaced observation lines arranged along longitude and latitude lines. For complex mountainous or coastal terrain, an adaptive track generation algorithm is employed to automatically adjust the UAV formation's flight path along contour lines or prevailing wind directions. The formation utilizes a master-slave collaborative architecture, with the lead UAV dynamically planning the track. Three to five subordinate UAVs form a "V" formation, maintaining fixed spacing and synchronously collecting data. Track density is dynamically increased in areas of sudden wind speed changes to ensure detailed capture of turbulence.

[0104] Step 2: Altitude stratification and vertical sampling: The measurement airspace is divided into a near-surface layer (0-200 meters), a transition layer (200-500 meters), and an upper layer (500-900 meters), implementing a differentiated altitude control strategy. The near-surface layer uses a dense climb pattern with 20-meter vertical intervals, acquiring high-resolution wind shear data through alternating hovering and slow ascent maneuvers. The mid- and upper-altitude layers utilize gradient flight at 50-100-meter intervals, utilizing dynamic compensation technology for the drone's attitude angle to ensure the sensor is always facing the wind. During flight, barometric altimeter and radar altimeter data are monitored in real time to correct for measurement errors caused by terrain undulations.

[0105] Step 3: Spatiotemporal synchronization and data acquisition: An atomic clock timing system is established, and the time base error for the entire formation is controlled within 1 millisecond. After the sensors are activated, the anemometer collects three-dimensional vector wind speed at a frequency of 10Hz, and the temperature and humidity sensors simultaneously record environmental parameters. All data is timestamped in milliseconds. The sampling strategy is dynamically adjusted: 1Hz basic sampling is used during periods of calm wind, and it automatically switches to a 10Hz high-frequency mode when turbulent characteristics are detected. Triple redundant data collection is implemented in key areas, with different drones cross-observing the same spatial point at three different time periods. Data is transmitted back to the ground server in real time via 5G and satellite dual channels.

[0106] Step 4: Combined navigation and dynamic positioning; using GPS and inertial navigation unit (IMU) fusion positioning technology. The GPS module receives multi-band satellite signals and uses real-time kinematic (RTK) technology to eliminate ionospheric errors and obtain centimeter-level longitude and latitude coordinates. The IMU has a built-in six-axis sensor. The gyroscope continuously monitors the drone's pitch, roll, and yaw angles, and the accelerometer records three-axis acceleration data. The navigation computer fuses the GPS absolute position with the IMU relative displacement through an extended Kalman filter, outputting three-dimensional spatial coordinates with an accuracy of 0.1 meters and attitude angle data of 0.5 degrees, providing a precise spatial reference for wind field measurements.

[0107] Step 5: Formation coordination and task optimization; building a hierarchical communication network: The formation uses a time division multiple access (TDMA) protocol, allocating communication slots every 5 milliseconds to achieve millisecond-level synchronization of commands. A mesh network is established between formations using relay drones, dynamically selecting the optimal transmission path. The task allocation system, based on a reinforcement learning algorithm, analyzes each drone's power level, location, and data quality in real time to dynamically adjust collection tasks. If a device anomaly or communication interruption is detected, a backup drone is automatically activated to take over the task, and local solid-state storage is used to cache data to ensure continuous collection.

[0108] Step 6: Quality control and real-time verification;

[0109] A three-level data verification mechanism is implemented during flight: the first level filters outliers through sensor physical thresholds (e.g., wind speeds >40m / s are automatically flagged); the second level utilizes synchronized observation data from adjacent drones for cross-validation, eliminating inter-device system errors; and the third level uses real-time comparisons with ground-based wind towers. A formation retake sequence is triggered when the data deviation at 120 meters exceeds 0.7%. All valid data is appended with a SHA-256 hash checksum and distributedly stored using blockchain technology to ensure data integrity and traceability.

[0110] The third stage is data synchronization and fusion processing.

[0111] Step 1: Unify time and space benchmarks and synchronize clocks. Establish a multi-level time synchronization system, utilizing a combined GPS pulse-per-second signal and atomic clock timing mechanism. Each drone incorporates a high-precision clock module, achieving microsecond-level time alignment through the NTP protocol, ensuring that the sampling time deviation of all sensors in the formation is less than 1 millisecond. In areas where satellite signals are blocked, the inertial navigation unit (IMU) timekeeping function is activated, maintaining a cumulative clock error of no more than 10 milliseconds after a 30-minute disconnection.

[0112] Step 2: Spatial calibration of multi-source data; achieving spatial reference unification through coordinate system transformation:

[0113] 1. Sensor coordinate system alignment: Dynamically correct the anemometer measurement direction according to the drone's attitude angle (pitch / roll / yaw) to eliminate observation errors caused by the aircraft's tilt.

[0114] 2. Geographic coordinate mapping: Convert the data collected by each drone's local coordinate system to the WGS84 geodetic coordinate system, using UTM projection to eliminate the influence of the earth's curvature.

[0115] 3. Elevation correction: Combined with the digital elevation model (DEM), the barometric altimeter data is corrected for terrain, and the vertical positioning accuracy is improved to ±0.5 meters.

[0116] Step 3: Multimodal data fusion processing; implement three-level data fusion process:

[0117] 1. Original level fusion: For multi-period observation data of the same spatial point, a time-weighted average algorithm is used to generate a continuous time series, and the weights are dynamically allocated according to the sensor accuracy and sampling time.

[0118] 2. Feature-level fusion: Extract parameters such as wind speed spectrum characteristics and turbulence intensity index, integrate multi-unit data through the Bayesian inference algorithm, and generate a fusion feature set with a confidence level of ≥95%.

[0119] 3. Decision-level fusion: Combined with historical data from ground-based wind towers, a support vector regression (SVR) model is used to establish a regional wind field mapping relationship and output a four-dimensional wind energy distribution matrix with a 100-meter grid accuracy.

[0120] Step 4: Real-time data transmission and verification; building an integrated air-ground transmission network:

[0121] 1. In-fleet transmission: Using the TDMA time division multiple access protocol, each drone is allocated a 5ms exclusive transmission time slot and transmits compressed data packets in real time (H.265 encoding, compression rate ≥ 50%).

[0122] 2. Cross-fleet relay: Build a Mesh network through relay drones, dynamically select the optimal transmission path, and control the packet loss rate below 0.1%.

[0123] 3. Closed-loop quality verification: The Pearson correlation coefficient between drone data and ground wind towers is calculated in real time. When the deviation of the data at an altitude of 120 meters exceeds 0.7%, a re-flight command in the abnormal area is automatically triggered.

[0124] Step 5: Build a spatiotemporal database; create a four-dimensional spatiotemporal index matrix to achieve efficient management of massive data:

[0125] 1. Time dimension: Sharded storage is based on UTC timestamps, with the minimum time unit being 1 second.

[0126] 2. Spatial dimension: Use Geohash encoding to convert geographic coordinates into 64-bit strings and establish an R-tree spatial index.

[0127] 3. Data cleaning rules: Set physical thresholds such as wind speed (0-60m / s) and temperature (-30℃-60℃), automatically filter outliers and generate data quality reports.

[0128] Data synchronization and fusion technology is used to synchronously process and fuse the data collected by all drones to ensure the accuracy and consistency of the data.

[0129] The fourth stage is high-precision data fitting and wind field modeling.

[0130] Step 1: Spatiotemporal data preprocessing; based on the spatiotemporal fusion data generated in the third stage, implement multi-level data optimization:

[0131] 1. Outlier cleaning: A dual filtering mechanism using the 3σ criterion and physical thresholds (wind speed 0-60m / s, temperature -30℃-60℃) is used to remove sensor noise and outliers.

[0132] 2. Missing value filling: Use the K nearest neighbor algorithm (K = 5) to interpolate within the spatiotemporal neighborhood, giving priority to observation data at the same altitude layer and adjacent grid points.

[0133] 3. Data standardization: Z-score standardization is performed on characteristic variables such as wind speed and turbulence intensity to eliminate the impact of dimensional differences on the model.

[0134] Step 2: Feature engineering and variable screening; establish a feature correlation analysis system and screen core modeling variables:

[0135] 1. Basic characteristics: wind speed (horizontal / vertical components), wind direction frequency, air density (calculated in real time based on temperature and humidity).

[0136] 2. Derived characteristics: wind power density (P = 0.5ρv 3, ρ is air density); wind shear index (α=ln(v2 / v1) / ln(z2 / z1)); turbulence intensity (IT=σv / μv, σv is the standard deviation of wind speed)

[0137] 3. Statistical characteristics: Weibull distribution parameters (k, c), effective power generation hours, and maximum wind speed with a 50-year return period.

[0138] The key variables were screened out by the Pearson correlation coefficient (|r|>0.7) and the random forest feature importance score (Gini index ≥0.15), and the feature subset was constructed.

[0139] Step 3: Support Vector Regression (SVR) modeling: Based on the SVR algorithm framework provided by the user, a wind energy resource distribution model is established:

[0140] In one possible implementation, the high-precision data fitting algorithm adopts a support vector regression SVR (Support Vector Regression, SVR) model and optimizes the parameter learning strategy and the use of the penalty function. By adjusting these parameters, the SVR model can be optimized to achieve a balance between prediction accuracy and generalization ability, so as to enhance the generalization ability and optimization effect of the model. The conceptual algorithm involved is as follows: SVR minimizes the difference between the mapping value of the training sample on this hyperplane and the target value by finding a hyperplane in the input space, while keeping the error within a certain range. The user pre-sets this boundary, usually expressed as ∈(epsilon), then the core of SVR is to maximize the boundary to minimize the model error. The optimization problem of SVR can be expressed as the following formula:

[0141]

[0142] The following constraints are met:

[0143] y i -w T φ(x i )-b≤∈+ξ i

[0144]

[0145] Where w is the weight vector, b is the bias term, w is the weight vector, b is the bias term, (φ(x i )) is a function that maps the input vector (xi) to a high-dimensional space. (y i ) is the target value. ∈ is the threshold of the insensitive loss function, which is set to 2 times the standard deviation of the wind speed observation error (usually ε=0.3-0.5m / s). C is the regularization parameter, which controls the model complexity and error tolerance (C∈[1,10]). (ξi )and is a relaxation variable that allows some samples to break through the ε band to deal with sudden turbulence in the wind field data.

[0146] The model training uses the SVRG (stochastic variance reduced gradient) optimizer to accelerate convergence, the batch size is set to 256, and the number of iterations is ≥1000 times.

[0147] In one possible implementation, a stochastic variance reduced gradient (SVRG) algorithm is used for gradient descent to improve the efficiency and convergence of parameter estimation by reducing the variance of gradient estimation. In the eigenvector parameter estimation of wind resource assessment, the SVRG algorithm can more efficiently optimize the parameters of the support vector regression (SVR) model. The core is to reduce the variance of the gradient estimation. In each iteration, the algorithm calculates a full gradient (using all samples) and a stochastic gradient (using one or a small batch of samples), and then updates the parameters by the difference between these two gradients. This method can reduce the variance of the gradient estimation in each iteration, thereby accelerating convergence; there is an internal iteration process in each round of iteration. Before the internal iteration begins, the algorithm calculates the full gradient under the current parameter value. Then, the internal iteration uses the difference between this full gradient and the stochastic gradient to update the parameters. This method ensures that the gradient estimate of each iteration has a continuously decreasing upper bound of variance, thereby achieving linear convergence. The specific parameter update specifications are as follows:

[0148]

[0149] Among them, (θ_t) is the parameter of the (t)th iteration, (θ_(t+1)) is the parameter of the (t+1)th iteration, (α) is the learning rate, is the gradient under the parameters (θ_t) and the random sample (ξ_t), In the parameter And the gradient under random samples (ξ_t), . In the parameter Full gradient under . Parameters is resampled in each outer iteration and is usually randomly chosen at the beginning of each epoch.

[0150] The fifth stage is wind energy potential assessment and decision support.

[0151] This phase achieves a quantitative assessment of wind energy resources through multi-scale data fusion and machine learning optimization. First, we integrate macro, meso, and micro data: at the macro level, we integrate global reanalysis climate fields (such as ERA5 data) to construct a regional wind climate background; at the meso level, we spatially interpolate and align drone-fitted wind field data (100m resolution) with the output of a mesoscale meteorological model (WRF); at the micro level, we combine high-frequency observations from wind towers with vertical profile data from lidar, overlaying digital elevation models to calculate terrain roughness correction coefficients. A wind energy density mapping relationship is established based on a support vector regression (SVR) model, whose optimization objective function is to minimize the sum of the weight vector modulus and the slack variable penalty term (formula: min1 / 2||w|| 2 +C∑(ξ_i+ξ_i*)), where the regularization parameter C∈[1,10] is determined by grid search, and the insensitive loss threshold ε is set to 0.3m / s to balance the prediction accuracy and generalization ability. The stochastic variance reduction gradient (SVRG) algorithm is used to accelerate model training, and the parameter update follows The full gradient is calculated after every 1000 mini-batch iterations, and the learning rate α decays from 0.05 to 0.01 according to the cosine annealing strategy to achieve linear convergence.

[0152] In the risk assessment phase, the maximum wind speed over a 50-year return period is calculated based on the Gumbel extreme value distribution, and high-risk areas exceeding the IEC Class III wind zone threshold (37.5 m / s) are marked. A turbulence intensity heat map (threshold 15%) is also generated. The economic assessment introduces a development priority index (PI = 0.6P / P_max + 0.3T_eq / 4000 + 0.1 / I_T) to divide the region into two levels of development zones: the first level requires a wind power density > 400 W / m 2 The equivalent full-load hours are greater than 3000h, and the secondary zone must meet the requirements of >250W / m 2 The NSGA-II multi-objective genetic algorithm is used to optimize the wind turbine layout, maximizing the total power generation after wake correction under the 5D×3D minimum spacing constraint, and the Pareto front convergence threshold is controlled within 1%.

[0153] The final results are visualized through a 3D GIS platform, including wind power density heat maps at 70 / 100 / 120 meters (color scale 0-800W / m 2 ), 50-year maximum wind speed contour lines (intervals of 5m / s), and development priority zoning layers. The decision support report automatically generates core indicators such as regional theoretical reserves and equivalent full-power hour distribution histograms, along with an electricity price sensitivity analysis curve (IRR trend with electricity price fluctuations of ±20%) and a discount rate-NPV matrix. The evaluation system has been cross-validated with a 5-fold cross-validation, and the model predicts wind speed RMSE ≤ 0.5m / s, and the wind tower data verification correlation r 2≥0.95, meeting the L2 accuracy requirement of the IEC 61400-12-1 standard, and can provide a reliable basis for the site selection of GW-level wind farms.

[0154] Based on the obtained wind energy resource data, a wind energy resource assessment model is used to calculate and evaluate the wind energy potential. Combining the fitted wind energy resource information and geographical location information, the amount of wind energy resources in the area is calculated. Based on the fitted wind energy resource meteorological data, a quantitative evaluation of the wind energy resources in the target area is completed. The evaluation results can be displayed in charts or other forms to better understand and analyze the status of wind energy resources and assist decision makers in making wise decisions in terms of wind energy investment, site selection, etc. In one possible embodiment, in the process of wind energy resource assessment, a comprehensive assessment is conducted by combining macro-, meso- and micro-scale data and models and ground meteorological station data. In one possible embodiment, big data analysis and machine learning algorithms are used to process and analyze the data in the wind energy resource assessment process to evaluate the wind energy resource potential in the target area.

[0155] This invention provides a wind resource assessment method based on drone formations. This method uses drone formation flight measurements to simultaneously collect wind resource data from multiple locations. It then uses a high-precision data fitting algorithm to simulate wind resource data for a large wind farm, thereby improving the accuracy and efficiency of wind resource assessment. This method offers advantages such as high data collection efficiency, high data accuracy, high wind resource assessment accuracy, efficient wind farm design, and a good user experience, and has broad market prospects.

[0156] Specific implementation case: The site conditions and requirements for a mountain wind farm are as follows: The planned site is located in a low-altitude mountainous area, with the base elevation of the planned site's topographic map between 700 and 900 meters, and a relative height difference of approximately 150 meters. The planned site is located in a V-shaped valley, surrounded by mountains on three sides, with the characteristic of forming a localized high-pressure depression. The atmospheric quality in the planned site area is good to slightly polluted. The annual dominant wind direction is localized, with no significant external influence. The wind farm at the planned site is required to have an annual power generation utilization of no less than 2,100 hours. The annual grid-connected power generation at the planned site is required to be 6 million kWh. The total investment in the wind farm is required to be no more than 5 million yuan. The annual operation and maintenance cost of the wind farm is required to be no more than 0.8 yuan per kWh. The unit capacity cost of the wind farm is required to be no more than 0.6 yuan per kW.

[0157] Based on the above conditions and requirements, the wind energy resource analysis of the planned site is carried out using the method of the present invention as follows: (1) Using a drone formation, data is collected from 72 stations within a 15km*15km range of the planned site (one station is arranged every 1°*1°), with an altitude layer of 100-900m and a height of 50m, such as wind speed, wind direction, temperature, and humidity near the ground; (2) The collected data is processed and synchronized to eliminate missing values and outliers; (3) A custom high-precision fitting algorithm is used to numerically simulate the wind energy resources within a 15km*15km range to obtain a wind energy resource distribution map of the planned site area; (4) Based on the wind energy resource distribution map and other geographic information of the planned site, the wind energy resource quantity is calculated. After calculation and analysis, it is concluded that the total wind energy resource quantity of the planned site is approximately 7.779 million kW. If general measures are taken, the wind energy resource quantity may be further increased by approximately 40%, reaching approximately 10.862 million kW.

[0158] This project has a total of 1 wind measurement tower, the parameters of which are as follows Table 1:

[0159] Table 1

[0160]

[0161] Wind tower configuration: The vertical gradient covers seven height layers from 10m to 120m. Wind speed and direction sensors are deployed at all height layers, and temperature, humidity, and pressure are set only at 10m and 120m.

[0162] This analysis covers data from the tower's erection until September 30, 2024. To ensure unbiased analysis of intraday data, only days with complete 24-hour data were included. The final selected data time period for evaluation is shown in Table 2:

[0163] Table 2

[0164] Start time 2024 / 4 / 1 00:00:00 End Time 2024 / 9 / 30 23:59:59

[0165] During 183 days (2024 / 4 / 1-2024 / 9 / 30) of continuous monitoring, 15,811,200 seconds of data were missing without any loss (100% completeness), providing a reliable benchmark for modeling.

[0166] The data records during this time period are summarized in Tables 3 and 4. Table 3 can be used as a calibration anchor for the drone data (the error between the drone and the wind tower data at an altitude of 120 m must be less than 0.7%).

[0167] Table 3

[0168] Time granularity 1s Total days 183 Total seconds 15,811,200 Total number of data rows 15,811,200 Number of lost data rows 0 Data completeness rate 100％

[0169] Table 4

[0170]

[0171] The average wind speed at each altitude is compared Figure 2 Key Trends: Average wind speed increases logarithmically with height, reaching 6.82 m / s at 120 m, a 78.5% increase compared to 10 m. Wind energy density (W / m 2 ) increased more significantly: 120m (186.9) is 5.65 times that of 10m (33.09); turbulence characteristics: the turbulence intensity at low altitude (10m: 0.213) is 1.87 times that at high altitude (120m: 0.114), confirming the need to focus on monitoring the 0-200m near-ground layer; verifying the necessity of stratified sampling of UAV formations (the increase in high-altitude data above 70m is significant).

[0172] like Figure 3 The wind shear index of the wind tower is 0.2279, which conforms to the power law distribution (fitting equation y = 2.3178x^0.2279), indicating that the surface roughness in this area is moderate.

[0173] According to the IEC standard, the measured maximum wind speed is generally 0.8 times the maximum wind speed that occurs once in 50 years. According to the European Standard II, the maximum wind speed that occurs once in 50 years is 5 times the annual average wind speed. The maximum wind speed that occurs once in 50 years calculated according to these two standards is shown in Table 5 below:

[0174] Table 5

[0175]

[0176] Since the actual data period used is less than one year, the above calculation results of the maximum wind speed once in 50 years are for reference only.

[0177] Figure 4 The wind rose diagram provided by the embodiment of the present application shows that the dominant wind direction shows significant wind direction concentration at all altitudes. The dominant wind direction at 120m is northwest (315°±22.5°). Vertical variation: Low altitude (10-50m) is affected by the terrain and the wind direction is highly dispersed. High altitude (100m+) wind direction stability is improved, which is conducive to the optimization of wind turbine layout. Layout suggestion: Design the wind turbine axis direction according to the dominant wind direction at 120m, which can reduce wake loss by 12%-15%. Temperature, humidity, pressure and air density statistics are as follows: Statistics at 10m are shown in Table 6:

[0178] Table 6

[0179] Minimum average value Maximum Temperature (℃) 6.27 24.03 40.55 Relative humidity (%) 13.9 66.0 98.7 Air pressure (hPa) 992.2 1009.0 1028.3 <![CDATA[Air density (kg / m 3 )]]> 1.104 1.183 1.278

[0180] Statistics at a height of 120m are shown in Table 7:

[0181] Table 7

[0182]

[0183]

[0184] In summary, the decision-making recommendations include: Model selection: IEC Class III (42.5m / s safety threshold) units must be selected for extreme wind speeds of 41.82m / s at 120m; Control strategy: Enable the dynamic pitch system in areas with turbulence intensity greater than 15% (mainly distributed below 50m); Power generation estimation: Based on an average wind speed of 6.82m / s at 120m and an air density of 1.177kg / m 3 Calculation shows that the annual equivalent hours of a single unit (5MW) can reach 3,150 hours. Data fusion: Using wind tower data as a benchmark, supplemented by drones: Horizontally, increase observation points in ridge / canyon areas; Vertically, increase sampling density at 50-120m layers (every 20m).

[0185] In summary, the embodiments of the present application provide a method for predicting wind energy resources. The method comprises acquiring three-dimensional wind farm observation data for a target area; performing spatiotemporal alignment processing on the three-dimensional wind farm observation data to generate a spatiotemporal sequence; performing multi-source data fusion on the spatiotemporal sequence to construct a multidimensional wind farm matrix; establishing a wind energy potential prediction model based on the multidimensional wind farm matrix; and predicting the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result. By constructing a machine learning model, efficient, accurate, and reliable resource assessment is provided for wind energy development in various terrain areas.

[0186] Based on the same technical concept, the embodiment of the present application also provides a wind energy resource prediction system, such as Figure 5 As shown, the system includes:

[0187] Data acquisition module 501, used to acquire three-dimensional wind field observation data of the target area;

[0188] A spatiotemporal sequence module 502 is configured to perform spatiotemporal alignment processing on the three-dimensional wind field observation data to generate a spatiotemporal sequence;

[0189] The wind field matrix module 503 is used to perform multi-source data fusion on the spatiotemporal sequence to construct a multi-dimensional wind field matrix;

[0190] A model building module 504 is used to build a wind energy potential prediction model based on the multi-dimensional wind field matrix;

[0191] The prediction module 505 is used to predict the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

[0192] The present application also provides an electronic device corresponding to the method provided in the above embodiment. Figure 6, which shows an electronic device provided by some embodiments of the present application. The electronic device 20 may include: a processor 200, a memory 201, a bus 202, and a communication interface 203. The processor 200, the communication interface 203, and the memory 201 are connected via the bus 202. The memory 201 stores a computer program executable on the processor 200. When the processor 200 executes the computer program, it executes the method provided by any of the aforementioned embodiments of the present application.

[0193] The memory 201 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element and at least one other network element are connected via at least one physical port (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0194] The bus 202 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs. The processor 200 executes the programs upon receiving execution instructions. The methods disclosed in any of the aforementioned embodiments of the present application may be applied to or implemented by the processor 200.

[0195] The processor 200 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 200 or by software instructions. The above processor 200 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of this application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 201 , and the processor 200 reads the information in the memory 201 and completes the steps of the above method in combination with its hardware.

[0196] The electronic device provided in the embodiments of the present application and the method provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented by them.

[0197] The present application also provides a computer-readable storage medium corresponding to the method provided in the above embodiment. Figure 7 The computer-readable storage medium shown is a CD 30 on which a computer program (ie, a program product) is stored. When the computer program is run by a processor, the method provided by any of the aforementioned embodiments is executed.

[0198] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. All equivalent structural transformations made by using the contents of the present invention description and drawings under the concept of the present invention, or direct / indirect application in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A wind energy resource prediction method, characterized in that: The method comprises: Obtain three-dimensional wind field observation data in the target area; Performing a spatiotemporal alignment processing operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence; Performing multi-source data fusion on the spatiotemporal sequence to construct a multi-dimensional wind field matrix; Establishing a wind energy potential prediction model based on the multidimensional wind field matrix; The wind energy data to be predicted is predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

2. The method according to claim 1, wherein Obtain three-dimensional wind field observation data of the target area, including: Generate layered track planning instructions based on the terrain data of the target area and the vertical gradient data of the wind tower; the vertical gradient data of the wind tower includes wind energy parameters at a plurality of preset heights within a preset height range; Invoking a drone formation to cruise at a plurality of vertically spaced altitude layers within the preset altitude interval according to the layered trajectory planning instructions, with the cruising speed maintained within a preset range; the drone formation comprising a plurality of drones equipped with meteorological sensors; Three-dimensional wind field observation data is collected in real time by each drone, and the collection time, spatial coordinates and attitude data are recorded; the sampling density is dynamically adjusted according to the vertical distribution law of wind speed, and the sampling frequency of the high altitude layer is set to be greater than that of the low altitude layer.

3. The method according to claim 1, wherein Performing a spatiotemporal alignment processing operation on the three-dimensional wind field observation data to generate a spatiotemporal sequence includes: Based on a preset physical threshold range and a wind tower data setting condition, abnormal data points in the three-dimensional wind field observation data are eliminated; the physical thresholds include wind speed, wind direction, and temperature; the wind tower data setting condition includes setting a data missing rate within a continuous time period to be lower than a set missing rate threshold; Use the satellite timing system to unify the time base of each drone and convert the local coordinate system data of each drone into a unified geographic coordinate system; The spatiotemporal sequence is generated by combining the real-time atmospheric parameter correction observation data.

4. The method according to claim 1, wherein Multi-source data fusion is performed on the spatiotemporal sequence to construct a multi-dimensional wind field matrix, including: A spatial interpolation algorithm is used to grid the wind field data at discrete observation points to generate continuous spatial distribution data. The spatial interpolation algorithm dynamically adjusts the interpolation weight in the vertical direction based on the wind shear index, which is obtained by fitting historical wind measurement data. A time series prediction model is used to fill in the missing wind field data to generate a time continuous sequence; the input of the time series prediction model includes the dominant wind direction distribution characteristics and historical wind field change trends; The continuous spatial distribution data and the time continuous sequence are integrated through a probabilistic inference algorithm to generate a multidimensional wind field matrix with a confidence level that meets preset conditions; the multidimensional wind field matrix includes multidimensional data of spatial coordinates, timestamps, wind field parameters and confidence ratings.

5. The method according to claim 1, wherein Performing a wind energy potential modeling operation based on the multi-dimensional wind field matrix to generate a wind energy potential prediction model includes: Extracting multidimensional wind field features from the multidimensional wind field matrix; the multidimensional wind field features include wind speed mean, wind direction frequency, turbulence intensity, and wind shear index; the feature extraction is performed based on a preset safety boundary condition, and the safety boundary condition is set according to historical maximum wind speed statistics; A machine learning algorithm is used to establish a mapping relationship between wind energy density and the multidimensional wind field characteristics to obtain the wind energy potential prediction model; the parameter settings of the machine learning algorithm meet the following set optimization conditions: the number of input layer nodes matches the dimension of the multidimensional wind field characteristics; the output layer contains wind energy density and corresponding probability distribution parameters; the training process is constrained by the turbulence intensity index to ensure that the model output conforms to the laws of fluid mechanics; The effectiveness of the wind energy potential prediction model is evaluated through physical statistical verification and precision verification; the physical statistical verification includes fitting the wind energy distribution output by the model with the Weibull distribution to verify the rationality of the parameters; the precision verification includes ensuring that the model prediction error is lower than the set threshold through cross-validation method.

6. The method according to claim 5, wherein A machine learning algorithm is used to establish a mapping relationship between wind energy density and the multi-dimensional wind field characteristics to obtain the wind energy potential prediction model according to the following formula: The following constraints are met: y i -w T φ(x i )-b≤∈+ξ i Among them, w is the weight vector of multidimensional wind field characteristics, b is the bias term; φ(xi) is the kernel function that maps the multidimensional wind field characteristics xi to the high-dimensional space; yi is the target value of the wind field parameter; ∈ is the threshold of the insensitive loss function; C is the regularization parameter that controls the model complexity and error tolerance; ξi and ξi* are slack variables that allow samples to break through the ∈ band.

7. The method according to claim 1, wherein The wind energy data to be predicted is predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result, including: Inputting the wind energy data to be predicted into the wind energy potential prediction model to extract the target wind field characteristics; and fitting the return period wind speed based on the extreme value distribution model to generate a wind speed extreme value prediction curve; the wind speed extreme value prediction curve includes the maximum wind speed values corresponding to several return periods and is marked with a risk threshold; Generate a wind power density map based on wind energy density distribution characteristics; the wind energy density distribution characteristics include vertical layer wind energy density change curves and real-time atmospheric parameter correction observation data; Determine the restricted area for wind turbine deployment based on the set turbulence intensity threshold and wind speed extreme value prediction results; the restricted area determination conditions include turbulence intensity exceeding the preset safety threshold and maximum wind speed exceeding the wind turbine wind resistance limit; The wind energy resource prediction result is output, wherein the wind energy resource prediction result includes the wind speed extreme value prediction curve, the wind power density map and the wind turbine deployment restriction area.

8. A wind energy resource prediction system, characterized in that: The system comprises: A data acquisition module is used to obtain three-dimensional wind field observation data of the target area; A spatiotemporal sequence module, configured to perform spatiotemporal alignment processing on the three-dimensional wind field observation data to generate a spatiotemporal sequence; A wind field matrix module is used to fuse multi-source data of the spatiotemporal sequence and construct a multi-dimensional wind field matrix; A model building module, configured to build a wind energy potential prediction model based on the multi-dimensional wind field matrix; The prediction module is used to predict the wind energy data to be predicted based on the wind energy potential prediction model to obtain a wind energy resource prediction result.

9. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Computer-readable instructions are stored thereon, and the computer-readable instructions can be executed by a processor to implement the method according to any one of claims 1 to 7.

Images