Dynamic optimization method for wake flow correction coefficient of fan

Abstract

The invention discloses a dynamic optimization method for a fan wake flow correction coefficient, and particularly relates to the technical field of fan wake flow correction, and the method comprises the steps: S1, collecting the multi-working-condition operation data of a wind power plant in real time; s2, constructing a nonlinear mapping model of wake flow correction coefficients and multi-working-condition parameters; s3, a multi-objective optimization function with the minimum wake flow prediction error, the maximum total output power of the wind power plant and the minimum total load loss of the wind turbine generator set as objectives is established; s4, an improved adaptive PSO-SVM algorithm is adopted to solve a multi-objective optimization function; and S5, constructing a closed-loop feedback adjustment mechanism. According to the dynamic optimization method for the fan wake flow correction coefficient, the correction coefficient can continuously adapt to working condition changes in long-term operation of a wind field, correction precision reduction is avoided, it is ensured that the wind field is kept in an efficient and safe operation state for a long time, and the service life of a unit is prolonged.

Description

Technical Field

[0001] This invention relates to the field of wind turbine wake correction technology, and in particular to a dynamic optimization method for wind turbine wake correction coefficients. Background Technology

[0002] In wind power generation, the wake effect of wind turbines can lead to a decrease in wind speed and an increase in load on downstream wind turbines, which directly affects the total power output of the wind farm and the service life of the units. The rationality of the wake correction coefficient is the core of accurately controlling the wake effect.

[0003] However, existing wind turbine wake correction technologies have significant shortcomings and are difficult to adapt to the complex operating conditions of wind farms: 1. Poor adaptability of static coefficients: Existing technologies mostly use fixed wake correction coefficients without considering real-time operating parameters such as wind speed gradient, turbulence intensity, and ambient temperature. This results in wake prediction errors exceeding the control requirement of 0.3 m² / s² under complex wind conditions, making it impossible to match actual wake patterns.

[0004] 2. Insufficient multi-objective coordination: Existing solutions often focus on a single objective and fail to balance the requirements of "small wake prediction error, large wind farm power, and small unit load". Full-load operation can easily lead to unit load exceeding limits, increased power loss when wind speed changes suddenly, and increased operation and maintenance costs and safety risks.

[0005] 3. Insufficient real-time performance of the algorithm: Traditional solution algorithms converge slowly and are prone to getting trapped in local optima. The solution time often exceeds 0.5 seconds, which cannot meet the requirements of real-time wind field control for rapid updates of correction coefficients and is difficult to respond to dynamic changes in wind conditions.

[0006] Lack of closed-loop feedback mechanism: The training samples of the existing model are not updated for a long time and run with the wind field. The SVM fitting accuracy R² is easy to drop below 0.92, and the correction coefficient is out of touch with the actual wake.

[0007] In summary, existing technologies suffer from drawbacks such as static nature, single objective, poor real-time performance, and lack of dynamic updates, resulting in significant wind farm power losses, high turbine operation and maintenance costs, and prominent safety risks. There is an urgent need for a dynamically optimized method for correcting wind turbine wake coefficients to address these issues. Summary of the Invention

[0008] The main objective of this invention is to provide a dynamic optimization method for the wake correction coefficient of a wind turbine, which can effectively solve the problems mentioned above.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for dynamically optimizing the wake correction coefficient of a wind turbine includes the following steps: S1: Real-time acquisition of multi-condition operation data of wind farm, including wind farm environmental parameters, wind turbine operating status parameters and measured wake parameters, to provide basic data support for dynamic optimization of correction coefficients; S2: Construct a nonlinear mapping model between the wake correction coefficient and multiple operating parameters, clarify the correlation between the primary correction coefficient, attenuation factor, wake initial radius correction coefficient and real-time operating parameters, and break through the limitations of traditional static correction coefficients. S3: Establish a multi-objective optimization function with the objectives of minimizing wake prediction error, maximizing total wind farm output power, and minimizing total wind turbine load loss, and coordinate the priorities of multiple objectives through dynamic weighting coefficients; S4: An improved adaptive PSO-SVM algorithm is used to solve the multi-objective optimization function and output the optimal wake correction coefficient of each wind turbine in real time; S5: Construct a closed-loop feedback adjustment mechanism, compare the measured parameters of the wake with the predicted parameters of the optimized model, and when the deviation exceeds the preset threshold, update the training samples of the mapping model and iterate and optimize again.

[0010] Preferably, the multi-condition operating data collected in real time in S1 specifically includes: Wind field environmental parameters: instantaneous wind speed, average wind speed, wind speed gradient, turbulence intensity, wind direction angle and wind direction change rate are collected by a three-dimensional ultrasonic anemometer and lidar wind speed monitoring system deployed in the wind field, with a sampling frequency of not less than 1Hz; Wind turbine operating status parameters: Output power, tower bending moment, blade bending moment, yaw angle, tip speed ratio, and axial induction factor of each wind turbine are acquired in real time through the wind turbine controller and load sensors; Wake velocity measurements: The wake velocity distribution at a distance of 1-3 times the rotor diameter in front of the downstream wind turbine is dynamically measured by lidar along the plane swept by the rotor.

[0011] Preferably, the specific construction process of the nonlinear mapping model between the wake correction coefficient and the multi-condition parameters in S2 includes: S2.1: Define the core correction factor: primary correction factor Attenuation factor initial radius correction factor for wake ; S2.2: Establish the correlation between β and multiple operating condition parameters: in, Basic correction factor, For wind speed gradient, Turbulence intensity; S2.3: Establish Correlation formula with multiple operating parameters: in, Based on the attenuation factor, Yaw angle downstream distance, The diameter of the wind turbine; S2.4: Establish Correlation formula with multiple operating parameters: in, For real-time tip speed ratio, For the optimal tip speed ratio, It is the axial induction factor; S2.5: The SVM algorithm is used to fit the nonlinear relationship of the above correlation, with historical working condition data as training samples, and the fitting accuracy R²≥0.92.

[0012] Preferably, the multi-objective optimization function in S3 is specifically constructed as follows: S3.1: Define the objective function for wake prediction error : in, The number of wake measurement points. For the first The measured wind speed at each measurement point For the predicted wind speed based on the correction coefficient, the objective is E→min; S3.2: Define the objective function for the total output power of the wind farm : in, Where is the radius of the wind turbine. air density, For the first The incoming wind speed of the typhoon fan For the first The axial induction factor of the typhoon is targeted to be →max; S3.3: Define the objective function for total load loss of wind turbine units : in, , The bending moment of the blade, the target is →min; S3.4: Constructing a multi-objective comprehensive optimization function: in, , , For dynamic weighting coefficients: + =1, To allow for the maximum wake error, This represents the total power output of the wind farm when a single unit is operating at full capacity. To the maximum permissible total load; The weighting coefficient is dynamically adjusted according to the operating conditions: when the wind speed changes abruptly. ≥0.4, under full power conditions ≥0.4, under normal operating conditions ≥0.3.

[0013] Preferably, the specific solution process of the improved adaptive PSO-SVM algorithm in S4 includes: S4.1: Algorithm Initialization: Set the particle count to 30-50, the number of iterations to 50-100, and the inertia weight. The initial value is 0.8, which decreases linearly to 0.2 with the number of iterations. The kernel function of SVM uses radial basis functions, with a penalty coefficient. The initial value was 10, and it was optimized through cross-validation. S4.2: Particle Coding: [This part is incomplete and requires further context to translate accurately.] , , As a three-dimensional variable of a particle, the range of its values ​​is: ∈[0.1,0.8], ∈[0.05,0.5], ∈[0.9,1.5]; S4.3: Fitness Calculation: Substitute the correction coefficient corresponding to each particle into the multi-objective optimization function. , The value represents the fitness of a particle; the higher the fitness, the better the particle. S4.4: Adaptive Update: Dynamically adjusts the learning factor based on the particle's fitness value. , The top 30% of particles in terms of fitness =1.2、 =1.8, the last 30% of particles =1.8、 =1.2; S4.5: Convergence Judgment: When the rate of change of the optimal fitness after 5 consecutive iterations is ≤0.5%, output the value corresponding to the current particle. , , As the optimal correction coefficient in real time, the solution time is ≤0.5s, which meets the requirements of real-time control.

[0014] Preferably, the closed-loop feedback adjustment mechanism in S5 specifically includes: S5.1: Deviation Calculation: Calculate the wake prediction error every 5 minutes. S5.2: Threshold Judgment: Preset Error Threshold =0.3m² / s², if ≤ Maintain the current mapping model; like > This triggers a model update; S5.3: Sample Update: Update the multi-condition data at the trigger time and the corresponding data. Add the training sample set to the mapping model and remove historical samples older than 3 months to ensure the timeliness of the samples; S5.4: Model Retraining: Retrain the SVM mapping model using a new sample set and adjust... , , The correlation parameters were used to verify the fitting accuracy R² ≥ 0.92 after retraining; otherwise, S5.3 was repeated to supplement the samples. S5.5: Iterative optimization: The updated mapping model is used in S2, and the "mapping-optimization-feedback" loop is re-entered to achieve long-term dynamic optimization of the correction coefficient.

[0015] Preferably, the correlation of β in S2.2 also introduces a temperature compensation term: in, To compensate for the effect of ambient temperature on air density, and improve... Adaptability.

[0016] Preferably, the dynamic weighting coefficient in S3.4 , , It can also be achieved by switching the wind farm's operating mode: during the initial grid connection phase. ≥0.5, maintenance period ≥0.4, old units ≥0.5.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention clarifies the primary correction coefficient by constructing a nonlinear mapping model between the wake correction coefficient and multiple operating parameters. Attenuation factor initial radius correction factor for wake The association logic, Correlate the wind speed gradient with turbulence intensity and introduce a temperature compensation term to offset the effect of air density. Correlate yaw angle with downstream distance. By correlating the tip speed ratio with the axial induction factor and combining it with the SVM algorithm, a fitting accuracy of R² ≥ 0.92 is achieved. This design allows the correction coefficient to dynamically follow changes in operating conditions, effectively controlling the wake prediction error within a preset threshold of 0.3 m² / s², accurately matching the actual wake pattern, and providing reliable data and model support for subsequent wind field optimization.

[0018] 2. This invention establishes a multi-objective optimization function that aims to minimize wake prediction error, maximize total wind farm power, and minimize total turbine load. It also designs dynamic weighting coefficients to coordinate priorities: error control is prioritized during sudden wind speed changes to avoid excessive power fluctuations; load reduction is prioritized during full-load operation to prevent turbine overload damage; and power enhancement is prioritized during normal operation to ensure power generation revenue. Simultaneously, it supports switching operating modes according to the initial grid connection period, maintenance period, and aging turbine period. This design achieves coordinated optimization of multiple objectives, reducing power loss and avoiding load overruns, significantly lowering maintenance costs and safety risks, and improving the long-term economic efficiency and stability of the wind farm.

[0019] 3. This invention employs an improved adaptive PSO-SVM algorithm: by setting 30-50 particles and 50-100 iterations, combined with an inertia weight linearly decreasing from 0.8 to 0.2, and dynamically adjusting the learning factor according to fitness, rapid convergence is achieved, with a solution time ≤0.5s. This algorithm can output the optimal correction coefficients for each wind turbine in real time, meeting the requirement for rapid updates of correction coefficients in real-time wind farm control, ensuring stable operation of the wind farm even during wind fluctuations, and avoiding additional power loss due to response lag.

[0020] 4. This invention employs a closed-loop feedback adjustment mechanism, calculating the deviation between the measured and predicted wake values ​​every 5 minutes. If the deviation exceeds a threshold, an update is triggered: current multi-condition data is added to the sample set, while historical samples older than 3 months are removed. The model is then retrained until R² ≥ 0.92, and the "mapping-optimization-feedback" loop is restarted. This design ensures that the correction coefficients continuously adapt to changes in wind farm operating conditions over long periods, preventing a decrease in correction accuracy, ensuring the wind farm maintains efficient and safe operation, and extending the unit's service life. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the overall method of the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1, as Figure 1 As shown, a dynamic optimization method for wind turbine wake correction coefficients includes the following steps: S1: Real-time acquisition of multi-condition operation data of wind farm, including wind farm environmental parameters, wind turbine operating status parameters and measured wake parameters, to provide basic data support for dynamic optimization of correction coefficients; S2: Construct a nonlinear mapping model between the wake correction coefficient and multiple operating parameters, clarify the correlation between the primary correction coefficient, attenuation factor, wake initial radius correction coefficient and real-time operating parameters, and break through the limitations of traditional static correction coefficients. S3: Establish a multi-objective optimization function with the objectives of minimizing wake prediction error, maximizing total wind farm output power, and minimizing total wind turbine load loss, and coordinate the priorities of multiple objectives through dynamic weighting coefficients; S4: An improved adaptive PSO-SVM algorithm is used to solve the multi-objective optimization function and output the optimal wake correction coefficient of each wind turbine in real time; S5: Construct a closed-loop feedback adjustment mechanism, compare the measured parameters of the wake with the predicted parameters of the optimized model, and when the deviation exceeds the preset threshold, update the training samples of the mapping model and iterate and optimize again.

[0024] The implementation steps are as follows: First, S1 real-time data acquisition was performed on a 100MW wind farm (including 20 2.5MW wind turbines with a rotor diameter of 130m). A three-dimensional ultrasonic anemometer was set up at the highest point of the wind farm, and a lidar was set up at the central axis of the wind turbines. The sampling frequency was 1Hz. For example, at 10:00 on April 15, the instantaneous wind speed was 8m / s, the average wind speed was 7.5m / s, the wind speed gradient was 0.2m / s², the turbulence intensity was 0.12, the wind direction angle was 30°, and the wind direction change rate was 5° / min. The wind turbine controller obtained the output power of wind turbine No. 1 as 2.2MW, the tower bending moment as 800kN·m, the blade bending moment as 500kN·m, the yaw angle as 5°, the tip speed ratio as 8, and the axial induction factor as 0.2. The lidar measured the wind speed at 5 points 260m (twice the rotor diameter) in front of wind turbine No. 2 downstream as 5.2, 5.0, 4.8, 5.1, and 4.9m / s. Next, S2 constructs a nonlinear mapping model and defines... , , The three core coefficients, with the basic correction coefficient β0 set to 0.5, are substituted into... =0.2、 =0.12; According to the formula: ; Calculated =0.5×(1+0.3×0.2+0.5×0.12)=0.56, basic attenuation factor Take 0.3 and substitute it into the equation. =5°, x / D=2, according to the formula: The calculation yields: =0.3×e^(-0.1×5)×(1-0.2×ln2)≈0.156, axial induction factor a=0.2, =8、 =8.5, according to the formula: Calculated: =√[(1-0.2) / (1-0.4)]×(1+0.1×(8-8.5))≈1.097, and then use 5000 sets of historical data over 3 months to train the SVM, with a fitting accuracy of R²=0.94; Then, S3 establishes a multi-objective optimization function, using the formula when N=5: Calculations, substituting the measured and predicted wind speeds, yield: E = (1 / 5)[(5.2-5.1)²+(5.0-4.9)²+(4.8-4.8)²+(5.1-5.0)²+(4.9-4.9)²] = 0.006 m² / s², wind turbine radius r = 65 m, ρ = 1.225 kg / m³, according to the formula: Calculate the power of fan 1 (P1) as: P1 = 2π × 65² × 1.225 × 7.5³ × 0.2 × 0.8² ≈ 1.756 MW. (20 units) ≈35.12MW, tower height h=120m, according to the formula: Calculate ≈26kN·m, according to the formula: 20 units =10520kN·m, taken as normal operating condition =0.3、 =0.4、 =0.3, =0.3、 0 = 50MW 0 = 12000 kN·m, according to the formula: Calculate maxF≈0.31195; S4 was then solved using an improved PSO-SVM, initializing 40 particles and performing 80 iterations. The initial value is 0.8 and decreases linearly. The SVM penalty coefficient is C=10 (determined by 5-fold cross-validation). The particle encoding is (0.56, 0.156, 1.097). The fitness is 0.31195. The first 12 particles have c1=1.2 and c2=1.8, and the next 12 particles have c1=1.8 and c2=1.2. The fitness change rate is 0≤0.5% for 5 consecutive iterations. The optimal coefficient is output in 0.4s. Finally, S5 provides closed-loop feedback, calculated every 5 minutes according to the formula: ,Calculate 10:05 =0.008≤0.3 Maintain the model, at 10:30 =0.35>0.3, so add the current data (instantaneous wind speed 9m / s, etc.) to the sample and remove the data from 3 months ago, retrain the SVM to get R²=0.93, update the model and loop again to achieve dynamic optimization of the wind turbine wake correction coefficient.

[0025] In Example 2, further, in S1 of this method, the multi-condition operation data collected in real time, during the wind field environmental parameter acquisition, a three-dimensional ultrasonic anemometer is set at an unobstructed location at the center of the wind field (altitude 100m), and a lidar is set at the wind field entrance. The sampling frequency is 1Hz, and the instantaneous wind speed is recorded once per second at a certain moment. After 60 records, the wind speed is 7.8-8.2m / s. According to the average wind speed calculation formula (average wind speed = sum of 60 instantaneous wind speeds / 60), the average wind speed is calculated to be 7.9m / s. The wind speeds at heights of 10m and 50m are taken as 7.5m / s and 8.3m / s, respectively. The wind speed gradient is calculated according to the wind speed gradient calculation formula (wind speed gradient = (wind speed at 50m height - wind speed at 10m height) / (50 - 10)). =(8.3-7.5) / (50-10)=0.02m / s·m. Using the turbulence intensity calculation formula (turbulence intensity = instantaneous wind speed standard deviation / average wind speed), and calculating 60 times with an instantaneous wind speed standard deviation of 0.8m / s, we obtain the turbulence intensity. 0.8 / 7.9≈0.101, the wind vane measures a wind direction angle of 35°, and the wind direction angle 10 minutes ago was 30°. According to the formula for calculating the rate of change of wind direction (rate of change of wind direction = (current wind direction angle - wind direction angle 10 minutes ago) / 10), the rate of change of wind direction is calculated to be 0.5° / min. When collecting wind turbine operating status parameters, the No. 1 wind turbine controller directly reads the output power as 2.3MW. The strain sensor at the bottom of the tower, using the strain-torque conversion relationship (torque = strain × elastic modulus × section modulus), outputs 0.5V, corresponding to a strain value that corresponds to the tower bending moment of 850kN·m. The fiber optic sensor at the blade root, using the same conversion relationship, outputs 150με, corresponding to a blade bending moment of 520kN·m. The yaw motor encoder, using the pulse-angle conversion relationship (angle = number of pulses × 360° / total number of pulses), with the current pulse corresponding to a yaw angle of 7°, calculates the blade linear velocity using the tip speed ratio calculation formula (tip speed ratio = blade linear velocity / incoming air velocity). =π×130×15 / 60≈102.1m / s, incoming air velocity 7.9m / s, the tip speed ratio λ=102.1 / 7.9≈12.92, from the fan power curve (pre-calibrated power-axial induction factor correspondence table), the axial induction factor a=0.21; During the acquisition of wake parameters, the lidar swept along the wind turbine plane (perpendicular to the ground and passing the center of the wind turbine). Measurement points were set at 10m intervals at a distance of 130-390m downstream (1-3 times the wind turbine diameter). Five measurement points (center and 20m above and below) were set at a distance of 260m downstream in front of wind turbine No. 2. The wind speeds at the five points were directly measured as 5.5, 5.3, 5.1, 5.4, and 5.2 m / s. Through the above equipment layout and parameter calculation, the real-time acquisition of S1 multi-condition operation data was completed.

[0026] Furthermore, the specific construction process of the nonlinear mapping model between the wake correction coefficient and the multi-condition parameters in method S2 includes: The nonlinear mapping model of wake correction coefficients and multi-condition parameters in S2 is constructed by first defining the core correction coefficients. Used to correct the initial attenuation of wake wind speed. Used to correct the rate at which wake wind speed decreases with downstream distance. Used to correct the initial radius of the wake; Next, establish Correlation formula with multiple operating conditions, standard operating condition (wind speed 8m / s, =0.1、 Foundation correction factor at 0.01m / s·m Taking 0.5, and considering wind speeds of 7.6 m / s and 8.4 m / s at heights of 10 m and 60 m respectively, calculate using the wind speed gradient formula ( =(Wind speed at 60m height - Wind speed at 10m height) / (60-10)) is calculated to get =(8.4-7.6) / (60-10)=0.016m / s·m, according to the turbulence intensity calculation formula ( =Standard deviation of instantaneous wind speed in 1 minute / average wind speed in 1 minute), with a standard deviation of instantaneous wind speed of 0.7 m / s and an average wind speed of 7.8 m / s in 1 minute, we can calculate: =0.7 / 7.8≈0.0897, substitute into the formula ,have to =0.5×(1+0.3×0.016+0.5×0.0897)≈0.5248; Re-establish Correlation with multiple operating parameters, no yaw ( =0°), downstream twice the rotor diameter (x / D=2) base attenuation factor Taking 0.3, the yaw encoder is measured according to the pulse-angle conversion relationship. =6°, the distance between wind turbines 1 and 2 is 500m. Using the formula x / D (x / D = distance between two wind turbines / rotor diameter), we get: x / D = 500 / 130 ≈ 3.846. Substituting into the formula: ; have to =0.3×e^(-0.1×6)×(1-0.2×ln3.846)≈0.1203; Then establish The blade linear velocity is calculated using the correlation formula with multiple operating parameters and the blade linear velocity calculation formula (blade linear velocity = π × rotor diameter × rotational speed). π×130×14 / 60≈92.36m / s, incoming air velocity 7.7m / s, calculated using the tip speed ratio formula (λ=blade linear velocity / incoming air velocity). =92.36 / 7.7≈12, the manufacturer provides the optimal tip speed ratio. =12.5, from the fan power curve, the axial induction factor a=0.2, substitute into the formula: ; have to =√[(1-0.2) / (1-0.4)]×(1+0.1×(12-12.5))≈1.097; Finally, the SVM algorithm was used to fit the nonlinear relationship. Five thousand sets of historical data from three months were used, divided into 4,000 training samples and 1,000 validation samples in a 4:1 ratio. When fitting β, the R² formula was used (R² = 1 - Sum of squared deviations of the validation set / Total Sum of squared deviations of the validation set). The calculated Sum of squared deviations of the validation set was 1.2, and the total Sum of squared deviations was 15.2. R² = 1 - (1.2 / 15.2) ≈ 0.921. Similarly, the fitting algorithm was used... and The R² values ​​are 0.93 and 0.925, respectively, both satisfying R²≥0.92, thus completing the construction of the nonlinear mapping model.

[0027] Furthermore, the multi-objective optimization function in method S3 is specifically constructed as follows: The construction of the multi-objective optimization function in S3 begins by defining the wake prediction error objective function. Five measuring points (center, 15m above and below, and 30m above and below) were set up at a distance of twice the rotor diameter downstream of the wind turbine, with the number of measuring points N=5. The actual wind speed at the five points was directly measured using a laser anemometer. =[5.4,5.2,5.0,5.3,5.1] m / s, obtained using the constructed nonlinear mapping model. =0.52、 =0.12、 =1.1, calculate the predicted wind speed for the corresponding 5 points. =[5.3,5.1,5.0,5.2,5.1] m / s, substitute into the formula: ; The calculation yields: E=(1 / 5)[(5.4-5.3)²+(5.2-5.1)²+(5.0-5.0)²+(5.3-5.2)²+(5.1-5.1)²]=0.006m² / s²; Redefining the objective function of total output power of a wind farm The wind turbine radius r = 65m (wind turbine diameter 130m, radius = diameter / 2), air density ρ = 1.225kg / m³ (air density at 25℃ under standard atmospheric pressure), and the incoming wind speed v1 = 7.8m / s measured by an anemometer, are given by the axial induction factor a1 = 0.21 for wind turbine No. 1, which is then substituted into the single-unit wind turbine power calculation formula: The calculated power output of wind turbine #1 is P1 = 2π × 65² × 1.225 × 7.8³ × 0.21 × 0.79² ≈ 2.0236 MW. The wind farm has a total of 20 wind turbines. The total output power is calculated using the formula... Calculated ≈38.5MW; Then define the objective function for the total load loss of the wind turbine. The tower height h = 120m (wind turbine design parameter). Substitute this into the formula for calculating the tower bending moment. Calculation of fan number 1: =2×1.225×120×9.8696×7.8²×0.21×0.79≈29.28kN·m; The bending moment of the blades of wind turbine No. 1 was measured using a blade load sensor. =530kN·m, calculated according to the formula for the total load of a single wind turbine. Get No. 1 wind turbine =559.28kN·m, the formula for calculating the total load of 20 wind turbines: Calculated ≈11200kN·m; Finally, a multi-objective comprehensive optimization function is constructed, and the dynamic weight coefficients are set to satisfy ω1+ω2+ω3=1, allowing for a maximum wake error. =0.3m² / s² (based on historical error statistics of wind farms, taking the upper limit of the 95% confidence interval), total power of the wind farm when a single unit is operating at full capacity. =50MW (total full-power output of 20 2.5MW wind turbines), maximum allowable total load =12000kN·m (90% of the maximum design load of the wind turbine). When the wind speed changes abruptly (the wind speed increases from 7m / s to 10m / s within 10 minutes, and the change in wind speed >2m / s / 10min is considered a sudden change), take ω1=0.4, ω2=0.3, and ω3=0.3, and substitute them into the formula: The calculation yields: =0.4×(1-0.006 / 0.3)+0.3×(38.5 / 50)-0.3×(11200 / 12000)≈0.343. When operating at full capacity (a single wind turbine with a power of ≥2.4MW, i.e., more than 96% of the rated power is considered to be operating at full capacity), we take ω1=0.2, ω2=0.3, and ω3=0.5. Under normal operating conditions (wind speed 6-9m / s, which is the wind speed range for efficient wind turbine operation), we take ω1=0.3, ω2=0.4, and ω3=0.3. This completes the construction of the multi-objective optimization function.

[0028] Furthermore, the specific solution process of the improved adaptive PSO-SVM algorithm in S4 of this scheme includes: The improved adaptive PSO-SVM algorithm in S4 is used for solution. First, the algorithm is initialized, setting the particle swarm size to 40 (by comparing the solution time and accuracy of 30, 40, and 50 particles, 40 particles showed the best accuracy within 0.4-0.5 seconds). The number of iterations is 80 (50 iterations may not converge, 100 iterations take more than 0.5 seconds, 80 iterations balance convergence and time consumption). The initial value of the inertia weight ω_pso is 0.8, and it decreases linearly according to the formula (…). =Initial value - (Initial value - Final value) / Iteration count × Current iteration count), the final value is 0.2, at the 10th iteration. =0.8-(0.8-0.2) / 80×9=0.7325. The kernel function of SVM adopts the radial basis function, and the initial value of the penalty coefficient C is 10. Through 5-fold cross-validation (the training samples are divided into 5 groups, each with 1000 samples, and 4 groups are used for training and 1 group for validation in turn, and the validation error of different C values ​​is calculated), it is determined that the validation error is the smallest (0.02) when C=10. Then perform particle encoding, , , As a three-dimensional variable of particles, referencing the historical correction coefficient distribution of wind farms, The value range is set to [0.1, 0.8] (in historical data). Minimum 0.12, maximum 0.78 (expanding the range to avoid boundary restrictions). The value range is set to [0.05, 0.5] (in historical data). Minimum 0.06, maximum 0.48), The value range is set to [0.9, 1.5] (in historical data). (Minimum 0.92, maximum 1.47), the code for a certain particle is (0.52, 0.12, 1.09). Next, the fitness is calculated. The values ​​of β=0.52, k_d=0.12, and γ_r=1.09 corresponding to this particle are substituted into the multi-objective optimization function F, and then calculated according to the formula... The calculated value is F≈0.245, which is the fitness of the particle. Then, an adaptive update is performed, sorting the fitness of the 40 particles from largest to smallest. The first 12 particles (40×30%=12) have the fitness in the top 30%, and the learning factors are set to c1=1.2 and c2=1.8 (to enhance local search and utilize the current best position). The last 12 particles (the bottom 30% of fitness) have c1=1.8 and c2=1.2 (to enhance global search and explore new positions). The middle 16 particles have c1=1.5 and c2=1.5 (to balance local and global search). Finally, convergence is checked, and the optimal fitness value for each iteration is recorded. The optimal fitness for the 25th iteration is 0.345, and the optimal fitness for iterations 26-29 is also 0.345. Using the fitness rate of change formula (rate of change = |current optimal fitness - previous optimal fitness| / previous optimal fitness × 100%), if the rate of change is 0 ≤ 0.5% for five consecutive iterations, then the value corresponding to the current particle is output. =0.53、 =0.118、 =1.1, and the solution time was recorded as 0.42s≤0.5s by the timer, which meets the real-time control requirements and completes the solution of the optimal wake correction coefficient.

[0029] The closed-loop feedback adjustment mechanism in method S5 specifically includes: The closed-loop feedback adjustment mechanism in S5 first calculates the deviation, setting it to calculate the wake prediction error every 5 minutes (i.e., 00:05, 00:10, 00:15...23:55 every day). Three groups of adjacent wind turbines (No. 1-2, No. 5-6, and No. 15-16) were selected in the wind farm. Five points were measured downstream of each group of turbines, for a total of 15 measurement points (N=15). At 08:00, the measured wind speeds at the 15 points were obtained using lidar. =[5.5,5.3,5.1,5.4,5.2,5.6,5.4,5.2,5.5,5.3,5.4,5.2,5.0,5.3,5.1]m / s, the predicted wind speeds for the corresponding 15 points are calculated using the current mapping model. =[5.4,5.2,5.1,5.3,5.2,5.5,5.3,5.2,5.4,5.3,5.3,5.2,5.0,5.2,5.1]m / s, substitute into the formula Calculated =(0.01+0.01+0+0.01+0+0.01+0.01+0+0.01+0+0.01+0+0+0.01+0) / 15≈0.0053m² / s²; Then, a threshold judgment is performed, with a preset error threshold. =0.3m² / s² (Based on wind farm operation requirements, when When the value is >0.3, the wake prediction error will cause a power loss of more than 5%, hence this threshold is set. (At 08:00) =0.0053≤0.3, keeping the current mapping model unchanged, the result calculated at 14:30 is... =0.35>0.3, triggering the model update process; Then, the sample was updated, collecting multi-condition data at 14:30, including instantaneous wind speed of 11 m / s, average wind speed of 10.5 m / s, wind speed gradient of 0.3 m / s², turbulence intensity of 0.15, No. 1 wind turbine output power of 2.5 MW, tower bending moment of 900 kN·m, etc., as well as the calculated data at this time. =0.35, add these data to the training sample set of the mapping model, and remove historical samples older than 3 months (such as data from January 15th and earlier) from the training sample set in chronological order to ensure the timeliness of the samples; Next, the model was retrained using the updated 5000 training samples. After training, the fitting accuracy of β was verified to be R² = 0.918 < 0.92. Then, two sets of recent high-turbulence data were added (instantaneous wind speed 12 m / s, turbulence intensity 0.18; instantaneous wind speed 9 m / s, turbulence intensity 0.16). After retraining, the R² of β was verified to be 0.923 ≥ 0.92. Similarly, adjustments were made. , The correlation parameters are used to ensure that the fitting accuracy R² of both is ≥0.92; Finally, iterative optimization is performed, and the updated mapping model is used in step S2 to construct the nonlinear mapping relationship between the wake correction coefficient and the multi-condition parameters, re-entering the "mapping-optimization-feedback" loop. Calculation is performed at 14:45. =0.008≤0.3, and then continue to repeat the process of deviation calculation and threshold judgment every 5 minutes to achieve long-term dynamic optimization of the correction coefficient.

[0030] Furthermore, the correlation of β in S2.2 of this method also introduces a temperature compensation term: First, the ambient temperature T is collected using a wind field temperature sensor. The temperature sensor is installed near the three-dimensional ultrasonic anemometer (avoiding direct sunlight to prevent affecting measurement accuracy; the measurement accuracy is ±0.5℃). At a certain moment, the current ambient temperature T=30℃, while the ambient temperature under standard operating conditions is 25℃ (referencing the wind turbine design standard ambient temperature). A basic correction factor is then applied. Taking 0.5, and considering wind speeds of 7.7 m / s and 8.5 m / s at heights of 10 m and 50 m respectively, calculate using the wind speed gradient formula: ( =(50m wind speed - 10m wind speed) / (50 - 10)) is calculated to get =0.02m / s·m, instantaneous wind speed standard deviation within 1 minute is 0.8m / s, average wind speed is 8.0m / s, calculated according to the turbulence intensity formula ( =Calculated by (standard deviation of instantaneous wind speed / average wind speed) =0.1, first substitute into the formula: The calculated value within the parentheses is 1 + 0.3 × 0.02 + 0.5 × 0.1 = 1.056. Next, the temperature compensation term is calculated using the temperature compensation formula (temperature compensation term = 1 + 0.01 × (T - 25)). Substituting T = 30℃, the temperature compensation term = 1 + 0.01 × (30 - 25) = 1.05. Finally, substituting this into the formula... Association The calculated β is 0.5 × 1.056 × 1.05 = 0.5544; If the ambient temperature T = 20℃ at a certain moment, the temperature compensation term is calculated using the same formula as 1 + 0.01 × (20 - 25) = 0.95. Substituting this into the correlation formula, we get β = 0.5 × 1.056 × 0.95 = 0.5016. If the ambient temperature T = 35℃, the temperature compensation term is calculated as 1 + 0.01 × (35 - 25) = 1.1. Substituting this into the correlation formula, we get β = 0.5 × 1.056 × 1.1 = 0.5808. By calculating parameters under different ambient temperatures and introducing the temperature compensation term, the adaptability of β is improved.

[0031] Furthermore, the dynamic weighting coefficients in S3.4 can be switched through the wind farm operation mode. The wind farm operation mode is divided into the initial grid connection period, the operation and maintenance period, and the aging unit period. The initial grid connection period refers to the first 3 months before the wind farm is put into operation (this stage is mainly to verify the power generation capacity). Priority is given to ensuring the total output power of the wind farm. The dynamic weighting coefficients are set to satisfy ω1+ω2+ω3=1, and ω1=0.2, ω2=0.5, and ω3=0.3 are taken, where ω2=0.5≥0.5. At a certain moment, the wake prediction error E=0.007m² / s² and the total output power of the wind farm are calculated. =36MW, total load loss of wind turbine units =10800kN·m, maximum allowable wake error =0.3m² / s², total power of the wind farm when a single unit is operating at full capacity =50MW, maximum allowable total load =12000kN·m, substituting into the formula The calculation yields: =0.2×(1-0.007 / 0.3)+0.5×(36 / 50)-0.3×(10800 / 12000)=0.2×0.9767+0.5×0.72-0.3×0.9≈0.1953+0.36-0.27≈0.2853; The operation and maintenance period refers to 1-5 years of wind farm operation (during which power generation and equipment losses need to be balanced). We set ω1=0.2, ω2=0.4, and ω3=0.4, where ω3=0.4≥0.4. At a certain moment, E=0.006m² / s². =37MW =11000 kN·m, substituting into the formula, we get: =0.2×(1-0.006 / 0.3)+0.4×(37 / 50)-0.4×(11000 / 12000)=0.2×0.98+0.4×0.74-0.4×0.9167≈0.196+0.296-0.3667≈0.1253; The term "old units" refers to wind farms that have been in operation for over 10 years (during this stage, equipment aging necessitates prioritizing load loss control to ensure equipment safety). We set ω1=0.5, ω2=0.2, and ω3=0.3, where ω1=0.5≥0.5, and at a certain moment, E=0.008m² / s². =35MW =11500kN·m, substituting into the formula, we get... =0.5×(1-0.008 / 0.3)+0.2×(35 / 50)-0.3×(11500 / 12000)=0.5×0.9733+0.2×0.7-0.3×0.9583≈0.4867+0.14-0.2875≈0.3392. By setting and calculating the dynamic weight coefficient under different wind farm operation modes, the weight coefficient switching is realized.

[0032] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamically optimizing the wake correction coefficient of a wind turbine, characterized in that, Includes the following steps: S1: Real-time acquisition of multi-condition operation data of wind farm, including wind farm environmental parameters, wind turbine operating status parameters and measured wake parameters, to provide basic data support for dynamic optimization of correction coefficients; S2: Construct a nonlinear mapping model between the wake correction coefficient and multiple operating parameters, clarify the correlation between the primary correction coefficient, attenuation factor, wake initial radius correction coefficient and real-time operating parameters, and break through the limitations of traditional static correction coefficients. S3: Establish a multi-objective optimization function with the objectives of minimizing wake prediction error, maximizing total wind farm output power, and minimizing total wind turbine load loss, and coordinate the priorities of multiple objectives through dynamic weighting coefficients; S4: An improved adaptive PSO-SVM algorithm is used to solve the multi-objective optimization function and output the optimal wake correction coefficient of each wind turbine in real time; S5: Construct a closed-loop feedback adjustment mechanism, compare the measured parameters of the wake with the predicted parameters of the optimized model, and when the deviation exceeds the preset threshold, update the training samples of the mapping model and iterate and optimize again.

2. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 1, characterized in that, The multi-condition operation data collected in real time in S1 specifically includes: Wind field environmental parameters: instantaneous wind speed, average wind speed, wind speed gradient, turbulence intensity, wind direction angle and wind direction change rate are collected by a three-dimensional ultrasonic anemometer and lidar wind speed monitoring system deployed in the wind field, with a sampling frequency of not less than 1Hz; Wind turbine operating status parameters: Output power, tower bending moment, blade bending moment, yaw angle, tip speed ratio, and axial induction factor of each wind turbine are acquired in real time through the wind turbine controller and load sensors; Wake velocity measurements: The wake velocity distribution at a distance of 1-3 times the rotor diameter in front of the downstream wind turbine is dynamically measured by lidar along the plane swept by the rotor.

3. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 1, characterized in that, The specific construction process of the nonlinear mapping model between the wake correction coefficient and the multi-condition parameters in S2 includes: S2.1: Define the core correction factor: primary correction factor Attenuation factor initial radius correction factor for wake ; S2.2: Establish the correlation between β and multiple operating condition parameters: in, Basic correction factor, For wind speed gradient, Turbulence intensity; S2.3: Establish Correlation formula with multiple operating parameters: in, Based on the attenuation factor, Yaw angle downstream distance, The diameter of the wind turbine; S2.4: Establish Correlation formula with multiple operating parameters: in, This is the real-time tip speed ratio. For the optimal tip speed ratio, It is the axial induction factor; S2.5: The SVM algorithm is used to fit the nonlinear relationship of the above correlation, with historical working condition data as training samples, and the fitting accuracy R²≥0.

92.

4. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 1, characterized in that, The multi-objective optimization function in S3 is specifically constructed as follows: S3.1: Define the objective function for wake prediction error : in, The number of wake measurement points. For the first The measured wind speed at each measurement point For the predicted wind speed based on the correction coefficient, the objective is E→min; S3.2: Define the objective function for the total output power of the wind farm : in, Where is the radius of the wind turbine. air density, For the first The incoming wind speed of the typhoon fan For the first The axial induction factor of the typhoon is targeted to be →max; S3.3: Define the objective function for total load loss of wind turbine units : in, , The bending moment of the blade, the target is →min; S3.4: Constructing a multi-objective comprehensive optimization function: in, , , For dynamic weighting coefficients: + =1, To allow for the maximum wake error, This represents the total power output of the wind farm when a single unit is operating at full capacity. To the maximum permissible total load; The weighting coefficient is dynamically adjusted according to the operating conditions: when the wind speed changes abruptly. ≥0.4, under full power conditions ≥0.4, under normal operating conditions ≥0.

3.

5. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 1, characterized in that, The specific solution process of the improved adaptive PSO-SVM algorithm in S4 includes: S4.1: Algorithm Initialization: Set the particle count to 30-50, the number of iterations to 50-100, and the inertia weight. The initial value is 0.8, which decreases linearly to 0.2 with the number of iterations. The kernel function of SVM uses radial basis functions, with a penalty coefficient. The initial value was 10, and it was optimized through cross-validation. S4.2: Particle Coding: [This likely refers to a specific program or code, possibly related to particle coding or encoding.] , , As a three-dimensional variable of a particle, the range of its values ​​is: ∈[0.1,0.8], ∈[0.05,0.5], ∈[0.9,1.5]; S4.3: Fitness Calculation: Substitute the correction coefficient corresponding to each particle into the multi-objective optimization function. , The value represents the fitness of a particle; the higher the fitness, the better the particle. S4.4: Adaptive Update: Dynamically adjusts the learning factor based on the particle's fitness value. , The top 30% of particles in terms of fitness =1.2、 =1.8, the last 30% of particles =1.8、 =1.2; S4.5: Convergence Judgment: When the rate of change of the optimal fitness after 5 consecutive iterations is ≤0.5%, output the value corresponding to the current particle. , , As the optimal correction coefficient in real time, the solution time is ≤0.5s, which meets the requirements of real-time control.

6. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 1, characterized in that, The closed-loop feedback adjustment mechanism in S5 specifically includes: S5.1: Deviation Calculation: Calculate the wake prediction error every 5 minutes. S5.2: Threshold Judgment: Preset Error Threshold =0.3m² / s², if ≤ Maintain the current mapping model; like > This triggers a model update; S5.3: Sample Update: Update the multi-condition data at the trigger time and the corresponding data. Add the training sample set to the mapping model and remove historical samples older than 3 months to ensure the timeliness of the samples; S5.4: Model Retraining: Retrain the SVM mapping model using a new sample set and adjust... , , The correlation parameters were used to verify the fitting accuracy R² ≥ 0.92 after retraining; otherwise, S5.3 was repeated to supplement the samples. S5.5: Iterative optimization: The updated mapping model is used in S2, and the "mapping-optimization-feedback" loop is re-entered to achieve long-term dynamic optimization of the correction coefficient.

7. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 3, characterized in that, The correlation of β in S2.2 also introduces a temperature compensation term: in, To compensate for the effect of ambient temperature on air density, and improve... Adaptability.

8. The method for dynamic optimization of wind turbine wake correction coefficient according to claim 4, characterized in that: The dynamic weighting coefficient in S3.4 , , It can also be achieved by switching the wind farm's operating mode: during the initial grid connection phase. ≥0.5, during maintenance period ≥0.4, old units ≥0.5.

Images