Wind turbine generator wake flow optimization cooperative control system and method

Abstract

The invention relates to the technical field of wind power generation control, in particular to a wind turbine generator wake flow optimization cooperative control system and method. According to the technical scheme, the wind turbine generator wake flow optimization cooperative control system comprises a feedforward LiDAR array which is deployed at the upstream 1-2 km of the prevailing wind direction of a wind power plant and used for collecting upstream three-dimensional wind field data in real time; the unit embedded sensor group comprises an ultrasonic anemograph and an inertial measurement unit IMU which are arranged in a cabin of each wind turbine unit, and the sampling frequency is not lower than 20Hz; the data fusion module is used for performing space-time alignment and noise filtering on the data of the LiDAR array and the unit embedded sensor through a convolutional neural network (CNN) and a Kalman filtering algorithm to generate a dynamic wind field digital twinborn model; and the LSTM wind field predictor predicts the wind speed and wind direction change trend in the future 30 seconds based on the dynamic wind field digital twinborn model. Through multi-source sensing fusion and dynamic game optimization, the operation efficiency and safety of the wind power plant under the dynamic wind condition are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of wind power generation control technology, and in particular to a wind turbine wake optimization coordinated control system and method. Background Art

[0002] In large wind farms, the wake generated by upstream turbines can cause wind speeds in downstream turbines to decrease and turbulence to increase, resulting in power generation losses (typically 10%-20%) and fatigue damage to mechanical components.

[0003] However, existing wake optimization technologies lack sensing capabilities and rely on single-point anemometers and static wake models, making it impossible to perceive changes in the upstream wind field in advance, resulting in control lags. The optimization objectives are single, and traditional methods focus on maximizing power generation, ignoring mechanical load control and exacerbating equipment wear. Dynamic response is slow: centralized optimization algorithms are time-consuming to calculate and cannot respond to sudden changes in wind direction or wind speed fluctuations in a timely manner. The communication load is too heavy: global data transmission is required, which causes communication delays and bandwidth bottlenecks when the number of units increases.

[0004] This results in low operating efficiency of wind farms under complex wind conditions and increased equipment maintenance costs. The present invention aims to overcome these technical barriers. Summary of the Invention

[0005] The present invention proposes a wind turbine wake optimization collaborative control system and method, which solves the problems in the existing technology caused by insufficient accuracy of static wake models, poor real-time performance of centralized optimization, control lag caused by a single sensor, and the inability of a single-objective strategy to coordinate power generation and mechanical load, resulting in low efficiency and short equipment life of wind farms under dynamic wind conditions.

[0006] The technical solutions of the present invention are as follows:

[0007] Wind turbine wake optimization coordinated control system includes:

[0008] A feed-forward LiDAR array is deployed 1-2 km upstream of the wind farm's dominant wind direction to collect real-time upstream three-dimensional wind field data.

[0009] The embedded sensor group of the wind turbine unit includes an ultrasonic anemometer and an inertial measurement unit (IMU) installed in the nacelle of each wind turbine unit, with a sampling frequency of not less than 20Hz;

[0010] The data fusion module uses a convolutional neural network (CNN) and Kalman filter algorithm to perform spatiotemporal alignment and noise filtering on the data from the LiDAR array and the turbine's embedded sensors to generate a dynamic wind farm digital twin model.

[0011] LSTM wind farm predictor, based on the dynamic wind farm digital twin model, predicts the wind speed and direction change trends within the next 30 seconds;

[0012] A dynamic game strategy generator defines upstream and downstream units as non-cooperative game participants, takes maximizing power generation and minimizing mechanical load as the game objectives, and generates a multi-objective optimization function;

[0013] A distributed solver is used to iteratively solve the multi-objective optimization function using an alternating direction multiplier method (ADMM) and output coordinated control instructions for the yaw angle and speed of each unit;

[0014] The adaptive execution module synchronously adjusts the yaw motor and the converter according to the coordinated control instruction to achieve power-load decoupling control.

[0015] Furthermore, the feedforward LiDAR array is deployed in a manner that every five wind turbines share one LiDAR, and the overlap rate of the scanning areas of adjacent LiDARs is not less than 30%.

[0016] Furthermore, the embedded sensor group of the unit also includes:

[0017] The vibration acceleration sensor installed at the bottom of the tower is used to monitor the excitation state of the first-order natural frequency of the tower in real time;

[0018] The strain sensor installed at the hub is used to measure the bending moment fluctuation data at the blade root.

[0019] Furthermore, in the data fusion module, the input of the convolutional neural network (CNN) is the spatiotemporal correlation matrix of the LiDAR point cloud data and the unit ultrasonic anemometer data, and the output layer dynamically compensates for the sensor error through Kalman filtering.

[0020] Furthermore, the training data of the LSTM wind field predictor includes LiDAR scan sequences in historical wind speed mutation events, and the prediction results include confidence intervals for wind speed fluctuation rate and wind direction deviation angle within the next 30 seconds.

[0021] Furthermore, the multi-objective optimization function of the dynamic game strategy generator is:

[0022]

[0023] Among them, the weight coefficients α(t) and β(t) are dynamically adjusted according to the real-time wind speed fluctuation rate:

[0024] When the wind speed fluctuation rate exceeds 2m / s 2 When , the weight of β(t) increases to 1.5 times of α(t);

[0025] When the wind speed fluctuation rate is less than 0.5m / s 2 When , the weight of α(t) increases to twice that of β(t).

[0026] Furthermore, during the iterative solution process of the distributed solver, each unit only exchanges yaw angle setting values ​​and speed limit values ​​with adjacent units, and the communication data is transmitted through the AES-256 encryption protocol.

[0027] Furthermore, in the adaptive execution module, the virtual inertia control curve of the converter satisfies:

[0028]

[0029] Among them, K v is the inertia coefficient, which is adaptively adjusted according to the wake intensity of the downstream unit, and the adjustment range is 0.5-2.0pu.

[0030] Furthermore, the adaptive execution module also includes a load protection submodule, which triggers a torque limit instruction when the tower vibration acceleration exceeds a threshold value, and preferentially reduces the mechanical load to a safe range.

[0031] Furthermore, the wind turbine wake optimization coordinated control method includes the following steps:

[0032] Collect multi-dimensional wind field data through feed-forward LiDAR array and turbine embedded sensor group;

[0033] Generate a dynamic wind farm digital twin model using the CNN-Kalman fusion algorithm;

[0034] Use the LSTM model to predict the wind speed and direction change trends within the next 30 seconds;

[0035] If the predicted wind direction deviation rate exceeds 1° / s, the dynamic game strategy generator is started to construct a multi-objective optimization function;

[0036] The ADMM algorithm is used for distributed iterative solution to generate coordinated control instructions for the yaw angle and speed of each unit;

[0037] The yaw motor and the converter are synchronously adjusted according to the instructions, and a load protection mechanism is triggered when tower resonance is detected.

[0038] The beneficial effects of the present invention are: through multi-source sensor fusion and dynamic game optimization, the operating efficiency and safety of wind farms under dynamic wind conditions are significantly improved.

[0039] Feedforward LiDAR perception and real-time wind field prediction enable accurate prediction of the wake impact range, significantly reducing power generation losses of downstream units; the dynamic game model balances power generation and load targets, reduces blade bending moment fluctuations and tower vibration, and extends equipment service life; the distributed optimization architecture combined with a lightweight communication protocol achieves second-level collaborative control, effectively responding to sudden changes in wind direction; only key control parameters are exchanged between adjacent units, reducing network bandwidth requirements and data latency; vibration monitoring and torque limiting mechanisms prevent mechanical overload, ensuring stable operation of the unit under extreme wind conditions. DETAILED DESCRIPTION

[0040] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0041] Wind turbine wake optimization coordinated control system includes:

[0042] A feed-forward LiDAR array is deployed 1-2 km upstream of the wind farm's dominant wind direction to collect real-time upstream three-dimensional wind field data.

[0043] The embedded sensor group of the wind turbine unit includes an ultrasonic anemometer and an inertial measurement unit (IMU) installed in the nacelle of each wind turbine unit, with a sampling frequency of not less than 20Hz;

[0044] The data fusion module uses a convolutional neural network (CNN) and Kalman filter algorithm to perform spatiotemporal alignment and noise filtering on the data from the LiDAR array and the turbine's embedded sensors to generate a dynamic wind farm digital twin model.

[0045] LSTM wind farm predictor, based on the dynamic wind farm digital twin model, predicts the wind speed and direction change trends within the next 30 seconds;

[0046] A dynamic game strategy generator defines upstream and downstream units as non-cooperative game participants, takes maximizing power generation and minimizing mechanical load as the game objectives, and generates a multi-objective optimization function;

[0047] A distributed solver is used to iteratively solve the multi-objective optimization function using an alternating direction multiplier method (ADMM) and output coordinated control instructions for the yaw angle and speed of each unit;

[0048] The adaptive execution module synchronously adjusts the yaw motor and the converter according to the coordinated control instruction to achieve power-load decoupling control.

[0049] The feedforward LiDAR array is deployed in a manner that every five wind turbines share one LiDAR, and the overlap rate of the scanning areas of adjacent LiDARs is not less than 30%.

[0050] The embedded sensor group of the unit also includes:

[0051] The vibration acceleration sensor installed at the bottom of the tower is used to monitor the excitation state of the first-order natural frequency of the tower in real time;

[0052] The strain sensor installed at the hub is used to measure the bending moment fluctuation data at the blade root.

[0053] In the data fusion module, the input of the convolutional neural network (CNN) is the spatiotemporal correlation matrix of the LiDAR point cloud data and the unit ultrasonic anemometer data, and the output layer dynamically compensates for the sensor error through Kalman filtering.

[0054] The training data of the LSTM wind field predictor includes LiDAR scan sequences in historical wind speed mutation events, and the prediction results include confidence intervals for wind speed fluctuation rate and wind direction deviation angle within the next 30 seconds.

[0055] The multi-objective optimization function of the dynamic game strategy generator is:

[0056]

[0057] Among them, the weight coefficients α(t) and β(t) are dynamically adjusted according to the real-time wind speed fluctuation rate:

[0058] When the wind speed fluctuation rate exceeds 2m / s 2 When , the weight of β(t) increases to 1.5 times of α(t);

[0059] When the wind speed fluctuation rate is less than 0.5m / s 2 When , the weight of α(t) increases to twice that of β(t).

[0060] During the iterative solution process of the distributed solver, each unit only exchanges yaw angle setting values ​​and speed limit values ​​with adjacent units, and communication data is transmitted using the AES-256 encryption protocol.

[0061] In the adaptive execution module, the virtual inertia control curve of the converter satisfies:

[0062]

[0063] Among them, K v is the inertia coefficient, which is adaptively adjusted according to the wake intensity of the downstream unit, and the adjustment range is 0.5-2.0pu.

[0064] The adaptive execution module also includes a load protection submodule, which triggers a torque limit instruction when the tower vibration acceleration exceeds a threshold value, and preferentially reduces the mechanical load to a safe range.

[0065] The wind turbine wake optimization coordinated control method includes the following steps:

[0066] Collect multi-dimensional wind field data through feed-forward LiDAR array and turbine embedded sensor group;

[0067] Generate a dynamic wind farm digital twin model using the CNN-Kalman fusion algorithm;

[0068] Use the LSTM model to predict the wind speed and direction change trends within the next 30 seconds;

[0069] If the predicted wind direction deviation rate exceeds 1° / s, the dynamic game strategy generator is started to construct a multi-objective optimization function;

[0070] The ADMM algorithm is used for distributed iterative solution to generate coordinated control instructions for the yaw angle and speed of each unit;

[0071] The yaw motor and the converter are synchronously adjusted according to the instructions, and a load protection mechanism is triggered when tower resonance is detected.

[0072] This embodiment provides a wind power control system that can perceive dynamic wind fields in real time and quickly coordinate the actions of multiple units. Through multi-objective game optimization and distributed computing, it achieves the dual goals of improving power generation efficiency and extending equipment life.

[0073] System composition

[0074] 1. Feedforward Perception Layer

[0075] LiDAR array: Multiple LiDARs are deployed 1-2 km upstream of the wind farm's dominant wind direction, with each LiDAR covering 5-8 turbines. They scan at a 10 Hz frequency to generate 3D wind field point cloud data (including wind speed, wind direction, and turbulence intensity).

[0076] Synchronization calibration module: aligns the timestamps of each LiDAR scan through the GPS clock to ensure the temporal and spatial consistency of the data.

[0077] 2. Local Sensing Layer

[0078] High-frequency sensor group: Each unit is equipped with the following sensors:

[0079] Ultrasonic anemometer (20Hz sampling): installed on the top of the cabin to measure real-time wind speed and direction;

[0080] Six-axis IMU: fixed in the middle of the tower, monitoring yaw acceleration (range ±5rad / s 2, accuracy 0.01°) and lateral vibration;

[0081] Strain sensor: embedded in the root of the blade to measure bending moment fluctuations (range ±5000μ∈, accuracy 1μ∈).

[0082] 3. Data processing layer

[0083] Multi-source data fusion:

[0084] Input: LiDAR point cloud data (100m×100m grid, 50m vertical layer), local wind speed and vibration data of the unit;

[0085] Processing flow:

[0086] Spatiotemporal alignment: Interpolate the LiDAR data to the crew coordinate system to compensate for the 1.2-second propagation delay (calculation formula: Where D is the distance from LiDAR to the unit, v wind is the current wind speed;

[0087] Noise filtering: Kalman filtering is used to eliminate random sensor errors, and median filtering is additionally added to LiDAR point cloud data;

[0088] Feature extraction: The spatial features of the wake vortex are extracted through a three-layer convolutional neural network (CNN), and a dynamic wind field vector diagram (resolution 0.1 m / s, 0.1°) is output.

[0089] Wind field prediction module:

[0090] LSTM model: Input the wind field sequence of the past 60 seconds (including wind speed, wind direction, and turbulence intensity) and output the predicted value for the next 30 seconds;

[0091] Confidence interval calculation: Based on the Monte Carlo Dropout method, a 95% probability interval of wind speed fluctuation rate and wind direction deviation angle is generated.

[0092] 4. Optimize the decision-making layer

[0093] Dynamic game model:

[0094] Participants: All units in the wind farm are defined as non-cooperative game participants;

[0095] Objective function:

[0096] Among them, P i is the generating power of unit i, which depends on its speed s i and yaw angle θ i ;

[0097] L i is the mechanical load, which is determined by the impeller speed ωi and blade bending moment M i calculate;

[0098] The weight coefficients α(t) and β(t) are dynamically adjusted according to the real-time wind speed fluctuation rate Δv:

[0099] β(t)=1-α(t)(k=0.5)

[0100] Constraints:

[0101] Yaw angle change rate

[0102] Speed ​​fluctuation|s i -s nom |≤10%s nom ,(s nom is the rated speed).

[0103] Distributed solver:

[0104] ADMM algorithm: decomposes the global optimization into local sub-problems of each unit, and the iterative formula is:

[0105]

[0106] where f i (x i ) is the local objective function of unit i, ρ = 1.0 is the penalty coefficient;

[0107] Communication protocol: Only adjacent units need to exchange yaw angle setting value θ i and speed limit value s lim , data is encrypted with AES-256 and transmitted via 5G NR-U.

[0108] 5. Execution Control Layer

[0109] Yaw lead control:

[0110] According to the LSTM prediction results, when the wind direction change rate exceeds 1° / s, the yaw motor is started 1-3 seconds in advance, and the yaw angle adjustment amount Δθ=v wind ·t pred sin(φ), (φ is the predicted wind direction deviation angle, t pred is the predicted duration);

[0111] Virtual inertia adjustment:

[0112] Converter control strategy:

[0113] where K v Adaptive adjustment based on the wake strength of the downstream unit;

[0114] Load protection mechanism:

[0115] The tower vibration acceleration is monitored in real time. If the threshold (0.3g) is exceeded or the first-order natural frequency (typical value: 0.15-0.25Hz) is detected, the torque limit command is immediately triggered to reduce the output power to 80% of the rated value.

[0116] Example

[0117] Coastal wind farm applications

[0118] The scenario parameters are as follows: wind farm scale: 50MW, 25 3MW units, arranged with a spacing of 5D (D is the rotor diameter); prevailing wind speed: 8-12m / s, and annual average wind direction change rate of 2° / s.

[0119] LiDAR array: Five WindSight Pro-3D radars were deployed 1.5 km upstream, each covering five units with a scan overlap of 30%;

[0120] Sensor configuration: Each unit is equipped with a Vaisala WXT536 anemometer, an XSENS MTi-630 IMU, and an HBM 1-LY41 strain sensor.

[0121] Edge computing nodes: Every five units share an industrial-grade server (Intel Xeon 8-core, 32GB RAM) to run data fusion and optimization algorithms.

[0122] Operation process:

[0123] Step 1: LiDAR detected that the upstream wind speed dropped suddenly from 10 m / s to 8 m / s, and the wind direction deflected at an angular rate of 1.5° / s.

[0124] Step 2: The CNN-Kalman module fuses the data and predicts that the wind direction will continue to shift by 6° (95% confidence interval ±1.2°) within the next 20 seconds.

[0125] Step 3: The dynamic game model generates the objective function (α = 0.7, β = 0.3), and the ADMM distributed solver converges after 15 iterations;

[0126] Step 4: The yaw angle of the upstream unit is adjusted by 4° in advance, and the converter of the downstream unit is set to K v = 1.5 to smooth power fluctuations;

[0127] Step 5: The IMU detected a tower vibration acceleration of 0.28g, but the protection mechanism was not triggered.

[0128] The following results were achieved:

[0129] Power generation: After optimization, the total power output increased from 42.3MW to 46.1MW (+9.0%);

[0130] Mechanical load: The maximum bending moment of the tower is reduced from 1.6MN·m to 1.1MN·m (-31.3%);

[0131] Response speed: It takes 1.1 seconds from detecting wind field changes to completing yaw adjustment.

[0132] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. Wind turbine wake optimization coordinated control system, characterized by: include: A feedforward LiDAR array is deployed 1-2 km upstream of the wind farm's dominant wind direction to collect real-time 3D upstream wind field data. The embedded sensor group of the wind turbine unit includes an ultrasonic anemometer and an inertial measurement unit (IMU) installed in the nacelle of each wind turbine unit, with a sampling frequency of no less than 20Hz; The data fusion module uses a convolutional neural network (CNN) and a Kalman filter algorithm to perform spatiotemporal alignment and noise filtering on the data from the LiDAR array and the turbine's embedded sensors, generating a dynamic wind farm digital twin model. LSTM wind farm predictor, based on the dynamic wind farm digital twin model, predicts the wind speed and direction change trends within the next 30 seconds; A dynamic game strategy generator defines upstream and downstream units as non-cooperative game participants, takes maximizing power generation and minimizing mechanical load as the game objectives, and generates a multi-objective optimization function; A distributed solver employs an alternating direction multiplier method (ADMM) to iteratively solve the multi-objective optimization function and output coordinated control instructions for the yaw angle and speed of each unit; The adaptive execution module synchronously adjusts the yaw motor and the converter according to the coordinated control instruction to achieve power-load decoupling control.

2. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: The feedforward LiDAR array is deployed in a manner that every five wind turbines share one LiDAR, and the overlap rate of the scanning areas of adjacent LiDARs is not less than 30%.

3. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: The embedded sensor group of the unit also includes: The vibration acceleration sensor installed at the bottom of the tower is used to monitor the excitation state of the first-order natural frequency of the tower in real time; The strain sensor installed at the hub is used to measure the bending moment fluctuation data at the blade root.

4. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: In the data fusion module, the input of the convolutional neural network (CNN) is the spatiotemporal correlation matrix of the LiDAR point cloud data and the unit ultrasonic anemometer data, and the output layer dynamically compensates for the sensor error through Kalman filtering.

5. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: The training data of the LSTM wind field predictor includes LiDAR scan sequences in historical wind speed mutation events, and the prediction results include confidence intervals for wind speed fluctuation rate and wind direction deviation angle within the next 30 seconds.

6. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: The multi-objective optimization function of the dynamic game strategy generator is: Among them, the weight coefficients α(t) and β(t) are dynamically adjusted according to the real-time wind speed fluctuation rate: When the wind speed fluctuation rate exceeds 2m / s 2 When , the weight of β(t) increases to 1.5 times of α(t); When the wind speed fluctuation rate is less than 0.5m / s 2 When , the weight of α(t) increases to twice that of β(t).

7. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: During the iterative solution process of the distributed solver, each unit only exchanges yaw angle setting values ​​and speed limit values ​​with adjacent units, and communication data is transmitted using the AES-256 encryption protocol.

8. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: In the adaptive execution module, the virtual inertia control curve of the converter satisfies: Among them, K v is the inertia coefficient, which is adaptively adjusted according to the wake intensity of the downstream unit, and the adjustment range is 0.5-2.0pu.

9. The wind turbine wake optimization coordinated control system according to claim 1, characterized in that: The adaptive execution module also includes a load protection submodule, which triggers a torque limit instruction when the tower vibration acceleration exceeds a threshold value, and preferentially reduces the mechanical load to a safe range.

10. A wind turbine wake optimization coordinated control method based on the wind turbine wake optimization coordinated control system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collect multi-dimensional wind field data through feed-forward LiDAR array and turbine embedded sensor group; Generate a dynamic wind farm digital twin model using the CNN-Kalman fusion algorithm; Use the LSTM model to predict the wind speed and direction change trends within the next 30 seconds; If the predicted wind direction deviation rate exceeds 1° / s, the dynamic game strategy generator is started to construct a multi-objective optimization function; The ADMM algorithm is used for distributed iterative solution to generate coordinated control instructions for the yaw angle and speed of each unit; The yaw motor and the converter are synchronously adjusted according to the instructions, and a load protection mechanism is triggered when tower resonance is detected.

Images