Matrix yaw system of wind farm

Abstract

The invention discloses a matrix yaw system of a wind power plant. The matrix yaw system comprises a data acquisition module, a wind plant modeling module, a parameter extraction module and a yaw control module. The data acquisition module acquires a wind field original data set containing three-dimensional space coordinates and timestamps by using a multi-source sensor; the wind field modeling module calculates correlation between points by combining a covariance function through a Gaussian process regression algorithm, and constructs a three-dimensional dynamic wind field model; the parameter extraction module is used for extracting wind regime parameter vectors in the model by adopting a nearest neighbor interpolation algorithm on the basis of actual space coordinates of a fan; and the yaw control module utilizes a depth deterministic strategy gradient reinforcement learning model to generate a yaw angle adjustment instruction in combination with the wind regime parameter vector, the current yaw state of the fan and a preset maximum cumulative reward function. Through multi-module cooperation and intelligent algorithm optimization, accurate matching of wind field dynamic modeling and yaw control is achieved, and the energy efficiency and operation stability of the wind generating set are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation, and in particular relates to a matrix yaw system of a wind power plant. Background Art

[0002] With the rapid development of wind power technology, the scale and intelligence of wind farms have become an industry trend. Efficient and accurate yaw control technology is becoming increasingly important as a core link to improve the energy efficiency of turbines. Existing wind power yaw systems are mostly based on static wind field models and traditional PID control algorithms. They obtain single wind direction data through wind direction sensors and generate yaw commands in combination with fixed parameter logic. They are characterized by simple structure and easy implementation, and are suitable for ideal scenarios with stable wind conditions. In traditional technologies, yaw control mainly relies on the following methods: static wind field modeling: a fixed wind condition model is constructed through limited sensor data, ignoring the spatiotemporal dynamic changes of wind speed and wind direction; single-target control strategy: using wind direction tracking error as the single optimization target, the PID algorithm is used to adjust the yaw angle, without comprehensively considering multi-dimensional factors such as motor power consumption and wake interference; independent control mode: a single wind turbine independently executes yaw commands, and there is a lack of a collaborative processing mechanism for the wake effects of adjacent wind turbines. However, the above traditional methods have exposed significant defects in complex wind farm environments: insufficient wind farm modeling accuracy: the static model cannot represent the wind speed gradient, turbulence intensity changes and time series correlation in three-dimensional space, resulting in large errors in wind condition parameter extraction; poor control strategy robustness: the fixed parameter PID algorithm is difficult to adapt to sudden changes in wind conditions, and frequent adjustments can easily cause problems such as increased mechanical vibration and shortened equipment life; lack of cluster coordination capabilities: the wake interaction between wind turbines is not considered, and single-machine optimization may lead to a decrease in the overall energy efficiency of the cluster, making it difficult to achieve global optimal control of the wind farm. Summary of the Invention

[0003] Based on this, it is necessary to provide a matrix yaw system for a wind farm that can solve the above problems.

[0004] In a first aspect, the present application provides a matrix yaw system for a wind farm, comprising:

[0005] The data acquisition module is used to obtain the original wind field data set containing three-dimensional spatial coordinates and time stamps using multi-source sensors;

[0006] The wind field modeling module is used to process the original wind field data set using the Gaussian process regression algorithm, calculate the inter-point correlation of the original data set through the covariance function, and build a three-dimensional dynamic wind field model based on the original data set;

[0007] A parameter extraction module is used to extract the wind condition parameter vector in the three-dimensional dynamic wind field model based on the actual spatial coordinates of the wind turbine in the power plant using the nearest neighbor interpolation algorithm;

[0008] The yaw control module is used to generate yaw angle adjustment instructions by adopting a yaw control model based on deep deterministic policy gradient reinforcement learning, combining the wind condition parameter vector, the current yaw state of the wind turbine and the preset maximized cumulative reward function.

[0009] In one embodiment, the system further includes an instruction optimization module for:

[0010] Based on the wind parameter vector, the preset multi-source interference model is used to perform interference analysis and generate the command interference feature vector;

[0011] Combining the yaw angle adjustment command with the interference feature vector, a candidate adjustment command set is generated through Monte Carlo simulation method;

[0012] Based on the candidate adjustment instruction set, a multi-objective optimization algorithm is used to screen out the basic optimal instruction set with the goal of minimizing the yaw angle error and maximizing the instruction execution efficiency.

[0013] In one embodiment, the instruction optimization module further includes a wake collaborative optimization unit configured to:

[0014] Based on the wind parameter vector, an improved wake propagation model is used to analyze the wake interaction between the target wind turbine and its adjacent wind turbines, and the wake influence coefficient matrix is ​​calculated.

[0015] Combining the wake influence coefficient matrix with the basic optimal instruction set, the collaborative yaw algorithm is used to generate the anti-wake candidate instruction set through the multi-agent consensus protocol;

[0016] Using the preset multi-source interference model, Monte Carlo simulation uncertainty analysis and performance evaluation are performed on the candidate anti-wake turbulence instruction set. With the goal of minimizing the wake interference loss and maximizing the energy efficiency of the unit cluster, the anti-wake yaw adjustment instruction set is screened.

[0017] In one embodiment, the yaw control module is further configured to construct a deep deterministic policy gradient reinforcement learning algorithm using the following formula:

[0018]

[0019] in, represents the optimal yaw angle adjustment instruction at time t, γ∈[0,1] represents the discount factor, E[· represents the expected cumulative reward calculation based on the deep deterministic policy gradient (DDPG) algorithm, r t (w t ,θ t ) represents the reward function, defined as r t =α·cos(Δθ t )+β·eff(θ t ), where Δθ tIndicates the deviation angle between the current yaw angle and the optimal wind direction angle, eff(θ t ) represents the instruction execution efficiency, α and β represent weight coefficients, and α+β=1.

[0020] In one embodiment, the system further includes a safety detection module for:

[0021] Based on the yaw angle adjustment command, a dynamic yaw adjustment sequence is generated by performing timing matching with a preset reference command template through a dynamic time warping algorithm;

[0022] When the adjustment amplitude of the dynamic yaw adjustment sequence exceeds a preset threshold, the mechanical vibration parameters of the wind turbine are collected in real time, including but not limited to vibration acceleration, velocity and displacement;

[0023] Perform fast Fourier transform spectrum analysis on mechanical vibration parameters to obtain vibration characteristic vectors;

[0024] If the vibration characteristic vector exceeds the preset safety threshold, the adaptive damping control algorithm is used to optimize the speed curve of the yaw actuator based on the proportional-differential control framework, generate a safe yaw speed instruction set containing segmented speed thresholds and buffer zone parameters, and simultaneously trigger the early warning mechanism.

[0025] In one embodiment, the security detection module is further configured to:

[0026] A mechanical vibration feature database for wind turbines is established based on normal vibration spectrum templates at different yaw speeds and angles;

[0027] The dynamic time warping algorithm is used to perform similarity matching between the mechanical vibration parameters and the spectrum template, and the vibration anomaly index is calculated;

[0028] The multi-level warning thresholds set by the fuzzy logic threshold controller are used. When the vibration abnormality index exceeds the corresponding threshold, the weight coefficient in the maximum cumulative reward function is adjusted.

[0029] In one embodiment, the system further includes a visualization module for:

[0030] Based on the 3D dynamic wind farm model, WebGL technology is used to render the wind turbine layout and wake impact, generating a heat map of wind farm energy distribution.

[0031] Combining wind farm energy distribution heat maps, yaw angle adjustment instructions, and vibration anomaly index, the yaw actuator status is visualized using multiple parameters to generate a dynamic risk assessment matrix with early warning prompts.

[0032] Based on the risk assessment matrix, machine learning algorithms are used to analyze the correlation between wind turbine operation trends and energy efficiency improvement space, and an interactive optimization solution recommendation dashboard is generated.

[0033] In a second aspect, the present application further provides a matrix yaw method for a wind farm, comprising:

[0034] Use multi-source sensors to obtain wind field raw data sets containing three-dimensional spatial coordinates and timestamps;

[0035] The original wind field data set is processed using the Gaussian process regression algorithm. The inter-point correlation of the original data is calculated using the covariance function. The three-dimensional dynamic wind field model is constructed by combining the original data set and the inter-point correlation.

[0036] Based on the actual spatial coordinates of the wind turbines in the power plant, the nearest neighbor interpolation algorithm is used to extract the wind condition parameter vector in the three-dimensional dynamic wind field model;

[0037] A yaw control model based on deep deterministic policy gradient reinforcement learning is adopted to generate the initial yaw angle adjustment command by combining the wind condition parameter vector, the current yaw state of the wind turbine and the preset maximum cumulative reward function.

[0038] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned matrix yaw system function of a wind farm when executing the computer program.

[0039] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the matrix yaw system function of the above-mentioned wind farm.

[0040] The matrix yaw system, method, computer device and storage medium for a wind farm described above use multi-source sensors to acquire a wind farm original data set containing three-dimensional spatial coordinates and timestamps through a data acquisition module. The wind farm modeling module uses a Gaussian process regression algorithm combined with a covariance function to calculate the correlation between points to construct a three-dimensional dynamic wind farm model. The parameter extraction module then extracts a wind parameter vector based on the actual spatial coordinates of the wind turbine using a nearest neighbor interpolation algorithm. Finally, the yaw control module uses a deep deterministic policy gradient reinforcement learning model combined with the wind parameter vector, the current yaw state of the wind turbine and a preset maximization cumulative reward function to generate a yaw angle adjustment instruction, forming a complete closed loop from multi-source data acquisition to dynamic modeling, precise parameter extraction to intelligent yaw control. This solves the problems of insufficient accuracy in traditional static wind farm modeling, poor robustness of PID control strategies and lack of cluster coordination in independent control modes, achieves accurate characterization of the spatiotemporal dynamic characteristics of the wind farm, and multi-objective adaptive optimization of the yaw angle and execution efficiency, improves the accuracy of wind parameter extraction and the response speed of yaw control, and enhances the operational stability and energy efficiency of wind turbines in complex wind farm environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 This is an overall structural diagram of the matrix yaw system of the wind farm of the present invention;

[0043] Figure 2 This is a detailed structural diagram of the matrix yaw system of the wind farm of the present invention;

[0044] Figure 3 This is a flow chart of the matrix yaw method for a wind farm according to the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0046] The matrix yaw system of this invention can be implemented in distributed wind farms. Its hardware comprises multi-source sensors, an edge computing terminal or server, and a yaw actuator. The sensors collect wind farm data containing temporal and spatial information. The computing terminal or server then generates yaw commands through modeling and algorithmic calculations, driving the actuator to adjust the angle. This system is suitable for complex terrain and variable wind conditions, enabling precise control and cluster coordination, improving energy efficiency and stability.

[0047] In one embodiment, Figure 1 As shown, a matrix yaw system for a wind farm is provided. This embodiment uses the system deployed on a terminal as an example for illustration. It is understandable that the method can also be deployed on a server, or on a CTCOM architecture including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the system includes:

[0048] The data acquisition module 11 is used to acquire a wind field original data set including three-dimensional spatial coordinates and time stamps using multi-source sensors.

[0049] Multi-source sensors, such as lidar, ultrasonic anemometers, and wind vanes, can be deployed at different locations across the wind farm. These sensors simultaneously collect multidimensional wind farm parameters, including wind speed, direction, and turbulence intensity. Three-dimensional spatial coordinates (x, y, z) and a timestamp are attached to each collected data point to construct a raw wind farm dataset with spatiotemporal dimensions. Multi-source data fusion improves the integrity and accuracy of wind farm information. Spatiotemporal annotated data provides the foundational input for building a three-dimensional dynamic wind farm model, supporting subsequent parameter extraction and the dynamic response capabilities of yaw control.

[0050] The wind field modeling module 12 is used to process the original wind field data set using the Gaussian process regression algorithm, calculate the inter-point correlation of the original data set through the covariance function, and construct a three-dimensional dynamic wind field model based on the original data set.

[0051] Among them, the Gaussian process regression algorithm can use a non-parametric Bayesian model to model the original data of the wind field, and characterize the spatiotemporal uncertainty of parameters such as wind speed and wind direction through probability distribution, so as to realize the prediction and inference of wind conditions at unsampled points. The kernel function of the covariance is introduced (such as the square exponential kernel, Matern kernel, etc.) to define the correlation measure between data points, quantify the similarity of wind parameters at any two points in three-dimensional space, and capture spatial characteristics and time series dynamic changes such as wind speed gradient and turbulence structure. Based on the original data set with spatiotemporal labels, a dynamic model containing x, y, z spatial dimensions and time t dimension is constructed to update the wind field parameter distribution in real time to reflect the evolution of wind conditions over time.

[0052] The parameter extraction module 13 is used to extract the wind condition parameter vector in the three-dimensional dynamic wind field model based on the actual spatial coordinates of the wind turbine in the power plant using the nearest neighbor interpolation algorithm.

[0053] Among them, the physical position (longitude, latitude, and altitude) of each wind turbine in the wind farm is mapped to the coordinate space of the three-dimensional dynamic wind farm model, and the spatial correspondence between the wind turbine and the model grid point is established. The nearest neighbor interpolation algorithm is used to search for the grid node closest to the spatial coordinates of the target wind turbine, and the wind condition parameters of the node (such as wind speed, wind direction, turbulence intensity, etc.) are directly used as the input parameter vector of the target wind turbine, realizing the rapid mapping of wind condition information from the model space to the physical wind turbine. It has the characteristics of low computational complexity and strong real-time performance, and meets the requirements of wind farm cluster control for parameter extraction speed.

[0054] The yaw control module 14 is used to generate a yaw angle adjustment instruction by adopting a yaw control model based on deep deterministic policy gradient reinforcement learning, combining the wind condition parameter vector, the current yaw state of the wind turbine and a preset maximization cumulative reward function.

[0055] The deep deterministic policy gradient reinforcement learning model, based on the actor-critic framework, approximates the policy function (actor) and the value function (critic) through deep neural networks (such as convolutional neural networks or fully connected networks) to achieve yaw angle optimization in a continuous action space. By combining wind parameter vectors (such as wind speed, wind direction, and turbulence intensity) with the wind turbine's current yaw state (such as current yaw angle and actuator speed), a high-dimensional state space is constructed to comprehensively characterize the wind turbine's operating environment and the dynamic characteristics of the controlled object. The actor network inputs the wind parameter vector and current yaw state and outputs yaw angle adjustment instructions; the critic network evaluates the action value and guides iterative policy optimization. The preset maximization cumulative reward function is a multi-objective optimization goal, enabling the system to comprehensively consider factors such as dynamic wind conditions and the physical constraints of the actuators, avoiding local optimality problems caused by single-objective optimization. Leveraging the nonlinear fitting capabilities of deep neural networks, the complex mapping relationship between wind field parameters and yaw angle is effectively modeled, addressing modeling challenges in high-dimensional, unstructured data scenarios and enabling intelligent generation of yaw angle adjustment instructions.

[0056] The matrix yaw system of the above-mentioned wind farm obtains the original wind farm data set containing three-dimensional spatial coordinates and timestamps through the data acquisition module, constructs a three-dimensional dynamic wind farm model through the wind farm modeling module using the Gaussian process regression algorithm combined with the covariance function, extracts the wind condition parameter vector based on the actual spatial coordinates of the wind turbine through the nearest neighbor interpolation algorithm by the parameter extraction module, and generates the yaw angle adjustment instruction through the yaw control module using the deep deterministic policy gradient reinforcement learning model. Through multi-module collaboration and intelligent algorithm optimization, a complete closed loop is formed from multi-source data acquisition to dynamic modeling, precise parameter extraction to intelligent yaw control, solving the problems of insufficient accuracy of traditional static wind farm modeling, poor robustness of PID control strategy and lack of cluster collaboration of independent control mode in the background technology, realizing accurate characterization of the spatiotemporal dynamic characteristics of the wind farm, multi-objective adaptive optimization of yaw angle and execution efficiency, improving the accuracy of wind condition parameter extraction and yaw control response speed, and enhancing the operational stability and energy efficiency of wind turbines in complex wind farm environments.

[0057] In one embodiment, Figure 2 As shown, the system further includes an instruction optimization module 15, which is used to:

[0058] Based on the wind parameter vector, the preset multi-source interference model is used to perform interference analysis and generate the command interference feature vector;

[0059] Combining the yaw angle adjustment command with the interference feature vector, a candidate adjustment command set is generated through Monte Carlo simulation method;

[0060] Based on the candidate adjustment instruction set, a multi-objective optimization algorithm is used to screen out the basic optimal instruction set with the goal of minimizing the yaw angle error and maximizing the instruction execution efficiency.

[0061] Specifically, based on wind parameter vectors (such as wind speed fluctuations and turbulence intensity), a pre-defined multi-source interference model (e.g., Gaussian noise and pulse interference models) is used to quantify the potential impact of uncertainty factors in the wind farm environment on yaw commands. This generates a command interference feature vector that includes characteristics such as interference type, intensity, and frequency. The initial yaw angle adjustment command output by the yaw control module is combined with the interference feature vector. Monte Carlo simulation is then used to randomly perturb and sample the command parameters. This generates a candidate adjustment command set covering a variety of possible interference scenarios, simulating the uncertainty of command execution in a real wind farm. A multi-objective optimization algorithm (e.g., NSGA-II and MOPSO) is used to globally optimize the candidate command set, with the optimization objectives of minimizing yaw angle error (ensuring wind direction tracking accuracy) and maximizing command execution efficiency (reducing actuator energy consumption and mechanical wear). A Pareto front analysis is then used to select the optimal command set that balances accuracy and efficiency. A three-level optimization chain (interference analysis, scenario simulation, and multi-objective optimization) is constructed, forming a closed-loop control system from command generation to command screening, further improving the system's adaptability to complex wind farm environments.

[0062] In one embodiment, the instruction optimization module 15 further includes a wake collaborative optimization unit for:

[0063] Based on the wind parameter vector, an improved wake propagation model is used to analyze the wake interaction between the target wind turbine and its adjacent wind turbines, and the wake influence coefficient matrix is ​​calculated.

[0064] Combining the wake influence coefficient matrix with the basic optimal instruction set, the collaborative yaw algorithm is used to generate the anti-wake candidate instruction set through the multi-agent consensus protocol;

[0065] Using the preset multi-source interference model, Monte Carlo simulation uncertainty analysis and performance evaluation are performed on the candidate anti-wake turbulence instruction set. With the goal of minimizing the wake interference loss and maximizing the energy efficiency of the unit cluster, the anti-wake yaw adjustment instruction set is screened.

[0066] For example, based on the wind condition parameter vector, an improved wake propagation model (such as the Jensen model, CFD correction model) is used to quantify the wake superposition effect between the target wind turbine and the adjacent wind turbines, and the wake influence coefficient matrix is ​​used to characterize the degree of influence of different wind turbine yaw angles on each other's wind speed attenuation. Combining the basic optimal instruction set with the wake influence coefficient matrix, a collaborative yaw algorithm (such as a distributed optimization algorithm) and a multi-agent consensus protocol are used to establish an information interaction mechanism for the wind turbine cluster, so that each wind turbine can synchronously consider the wake feedback of the adjacent units when adjusting the yaw angle, and generate a globally coordinated anti-wake candidate instruction set. A preset multi-source interference model is used to perform Monte Carlo simulation on the anti-wake candidate instruction set to evaluate the impact of uncertainty factors such as wake interference and wind speed fluctuations on the instruction performance; with the dual goals of minimizing wake interference losses (reducing wind speed attenuation of downstream units) and maximizing the energy efficiency of the unit cluster (increasing overall power generation), the optimal anti-wake yaw adjustment instruction set is screened through Pareto optimization. By adding a wake collaborative optimization unit, an optimization link of wake modeling-collaborative instruction generation-uncertainty screening is constructed, filling the gap in traditional technology in dealing with wake effects between wind turbines.

[0067] In one embodiment, the yaw control module 14 is further configured to construct a deep deterministic policy gradient reinforcement learning algorithm using the following formula:

[0068]

[0069] in, represents the optimal yaw angle adjustment instruction at time t, γ∈[0,1] represents the discount factor, E[· represents the expected cumulative reward calculation based on the deep deterministic policy gradient (DDPG) algorithm, r t (w t ,θ t ) represents the reward function, defined as r t =α·cos(Δθ t )+β·eff(θ t ), where Δθ t Indicates the deviation angle between the current yaw angle and the optimal wind direction angle, eff(θ t ) represents the instruction execution efficiency, α and β represent weight coefficients, and α+β=1.

[0070] Specifically, the optimal yaw angle adjustment instruction at time t is Calculated by the DDPG algorithm, it is the deterministic policy output in the continuous action space; the discount factor γ∈[0,1] is used to adjust the weight of the influence of future rewards on the current decision. When γ approaches 1, it strengthens long-term benefits, and when it approaches 0, it focuses on immediate rewards; the expected cumulative reward calculation E[·] based on the deep deterministic policy gradient (DDPG) algorithm is implemented through the Actor-Critic network structure to achieve policy gradient optimization; the reward function r t (w t ,θ t ) is designed as follows: cos(Δθ t ) quantifies the yaw angle deviation Δθ using the cosine function t (the difference between the current yaw angle and the optimal wind direction angle), which is directly related to the wind energy capture efficiency; eff(θ t ) represents the efficiency of command execution, which reflects the balance between angle adjustment speed and mechanical loss. The weight coefficients α and β are dynamically adjusted through the online learning mechanism. When the mechanical vibration exceeds the limit, the β weight is increased to reduce the wear caused by frequent yaw. t As state input, the closed-loop coupling of the spatiotemporal characteristics of the wind field and yaw control is realized; the reward function simultaneously optimizes the wind direction tracking accuracy (cos(Δθ t ) item) and execution efficiency (eff(θ t ) item), avoid mechanical oscillation caused by a single target.

[0071] In one embodiment, the system further includes a safety detection module 16 for:

[0072] Based on the yaw angle adjustment command, a dynamic yaw adjustment sequence is generated by performing timing matching with a preset reference command template through a dynamic time warping algorithm;

[0073] When the adjustment amplitude of the dynamic yaw adjustment sequence exceeds a preset threshold, the mechanical vibration parameters of the wind turbine are collected in real time, including but not limited to vibration acceleration, velocity and displacement;

[0074] Perform fast Fourier transform spectrum analysis on mechanical vibration parameters to obtain vibration characteristic vectors;

[0075] If the vibration characteristic vector exceeds the preset safety threshold, the adaptive damping control algorithm is used to optimize the speed curve of the yaw actuator based on the proportional-differential control framework. Combined with the dynamic yaw adjustment sequence, a safe yaw speed instruction set containing segmented speed thresholds and buffer zone parameters is generated, and the early warning mechanism is triggered simultaneously.

[0076] For example, a dynamic time warping algorithm is used to align the yaw angle adjustment command sequence with a preset reference command template. By minimizing the path cost function, an optimal timing mapping is generated, generating a dynamic yaw adjustment sequence. When the adjustment amplitude of the dynamic yaw adjustment sequence (e.g., an angle change rate >5° / s) exceeds a preset threshold, a triaxial accelerometer is triggered to collect the wind turbine's vibration parameters in real time, covering multiple metrics such as acceleration, velocity, and displacement, forming a complete mechanical vibration parameter monitoring dataset. A short-time Fourier transform is performed on the vibration parameters to extract energy characteristics in specific frequency bands to form a vibration feature vector. These features can be categorized as low-frequency (0-10Hz) tower oscillation energy; mid-frequency (10-100Hz) gearbox meshing frequency; and high-frequency (100-1000Hz) bearing fault characteristic frequency. Adaptive damping control algorithms, such as those based on the PD (Perception and Decision) control framework, use the vibration feature vector as feedback input to dynamically adjust the speed curve. Different PD parameters can be set according to different areas, such as low speed area [0-2° / s], medium speed area [2-4° / s], high speed area [4-6° / s], and acceleration limit in the start and stop phases (such as 0.5° / s 2 ) and an angle overshoot compensation coefficient (e.g., 0.8) are used as feedback inputs. Based on the proportional-derivative control framework, the speed curve is dynamically adjusted to generate a safe yaw speed command set. By comparing the vibration eigenvector with safety thresholds, multi-level warnings are implemented (e.g., yellow warning: eigenenergy increases by 10%; red warning: eigenenergy exceeds 0%), providing a quantitative basis for operation and maintenance decisions. A three-level processing architecture of timing matching, spectrum analysis, and adaptive control establishes a safety assurance mechanism from command generation to execution.

[0077] In one embodiment, the security detection module 16 is further configured to:

[0078] A mechanical vibration feature database for wind turbines is established based on normal vibration spectrum templates at different yaw speeds and angles;

[0079] The dynamic time warping algorithm is used to perform similarity matching between the mechanical vibration parameters and the spectrum template, and the vibration anomaly index is calculated;

[0080] The multi-level warning thresholds set by the fuzzy logic threshold controller are used. When the vibration abnormality index exceeds the corresponding threshold, the weight coefficient in the maximum cumulative reward function is adjusted.

[0081] Specifically, a multidimensional vibration signature database can be established, with operating conditions divided by yaw speed level (e.g., low / medium / high) and angle range (e.g., 0°-30° / 31°-60° / 61°-90°). For each operating condition, vibration signals under normal operating conditions are collected, and frequency domain feature vectors are extracted through Fourier transform to form a standardized spectrum template set, thereby establishing a mechanical vibration signature database for the wind turbine. A DTW algorithm can be used to time-series align the real-time vibration spectrum with the template, calculating the minimum cumulative distance as a similarity metric. A vibration anomaly index is defined as the normalized DTW distance between the measured spectrum and the template spectrum, quantifying the degree to which the equipment state deviates from the normal mode. A two-dimensional fuzzy controller is designed, taking the anomaly index and its rate of change as input and outputting a warning level (no warning / yellow / orange / red). A fuzzy rule base is used to implement nonlinear threshold decision-making. For example, an orange warning is triggered when the anomaly index rises rapidly and exceeds the moderate threshold; a yellow warning is triggered when the anomaly index increases slowly but continuously exceeds the mild threshold. Based on the warning level, the weight coefficients in the reinforcement learning reward function are adjusted in real time. For example, in the normal state (no warning), the wind energy capture efficiency term (α = 0.8) is maintained with a high weight. In the warning state, the execution efficiency term is gradually increased (β from 0.2 to 0.8), while the yaw angle tracking priority is reduced. This achieves a dynamic balance between wind energy capture and equipment safety. This technology extends the maximization of the cumulative reward function, dynamically adjusting the optimization objective through equipment status perception, providing a more specific implementation path for improving the operational stability of wind turbines.

[0082] In one embodiment, the system further includes a visualization module 17 for:

[0083] Based on the 3D dynamic wind farm model, WebGL technology is used to render the wind turbine layout and wake impact, generating a heat map of wind farm energy distribution.

[0084] Combining wind farm energy distribution heat maps, yaw angle adjustment instructions, and vibration anomaly index, the yaw actuator status is visualized using multiple parameters to generate a dynamic risk assessment matrix with early warning prompts.

[0085] Based on the risk assessment matrix, machine learning algorithms are used to analyze the correlation between wind turbine operation trends and energy efficiency improvement space, and an interactive optimization solution recommendation dashboard is generated.

[0086] For example, based on a three-dimensional dynamic wind field model, WebGL technology is used to achieve real-time 3D rendering of wind turbine clusters and wake effects; a wind field energy distribution heat map is constructed, using color gradients to intuitively display the spatial distribution of parameters such as wind speed and turbulence intensity, and the wake area is visualized with a translucent cone to achieve dynamic perception of wind field energy flow. The wind field heat map, yaw control instructions (angle / speed) and vibration anomaly index are aligned in time and space; a dynamic risk assessment matrix is ​​designed, with the horizontal axis being the wind turbine number and the vertical axis being the operating parameters (such as wind speed range, yaw angle), and the matrix cells are color-coded (green / yellow / red) to intuitively display the equipment risk level; an early warning prompt function is integrated, and when the vibration anomaly index exceeds the threshold, the corresponding cell displays a flashing icon and annotates the anomaly type (such as bearing failure, resonance risk). Based on historical operating data and a risk assessment matrix, machine learning algorithms (such as random forests and LSTM) are applied to establish a model for predicting wind turbine operating trends and assessing potential energy efficiency improvements. Feature importance analysis identifies key influencing factors (such as wake turbulence obstruction rate and yaw response delay) to generate an interactive optimization solution dashboard. This dashboard recommends personalized yaw strategy adjustments (such as coordinated yaw angles under specific wind directions) and estimates the potential for energy efficiency improvements (such as a 2.3% increase in annual power generation). Through a technical chain of data visualization, risk assessment, and intelligent recommendation, abstract control algorithms and equipment status are transformed into actionable decision-making foundations, providing key human-computer interaction and decision support.

[0087] The matrix yaw system of the above-mentioned wind farm obtains the original wind farm data set containing spatiotemporal information through the data acquisition module, and constructs a three-dimensional dynamic wind farm model by using Gaussian process regression in the wind farm modeling module. Combined with the precise parameter mapping of the parameter extraction module, the deep reinforcement learning multi-objective optimization of the yaw control module, the anti-interference and wake collaborative processing of the instruction optimization module, the vibration monitoring adaptive control of the safety detection module and the multi-dimensional data intelligent analysis of the visualization module, a complete technical closed loop from dynamic modeling to intelligent control, safety protection to decision support is formed, breaking the limitations of insufficient accuracy of traditional static modeling, poor robustness of PID control strategy and lack of cluster coordination, realizing accurate characterization of the spatiotemporal dynamic characteristics of the wind farm, anti-interference multi-objective optimization of yaw commands, coordinated suppression of wake effects of unit clusters, active early warning of mechanical vibration risks and adaptive adjustment of control strategies, improving the energy efficiency level, operation stability, equipment safety and intelligent operation and maintenance decision-making of wind turbines in complex wind farms, and solving the industry problem of efficient and precise control of large-scale wind farms.

[0088] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0089] Based on the same inventive concept, embodiments of the present application also provide a method for implementing the aforementioned matrix yaw system for a wind farm. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the matrix yaw method for a wind farm provided below can be found in the aforementioned limitations on the matrix yaw system for a wind farm, and will not be further elaborated here.

[0090] In an exemplary embodiment, Figure 3 As shown, a matrix yaw method for a wind farm is provided, comprising:

[0091] S101, using multi-source sensors to obtain a wind field original data set including three-dimensional spatial coordinates and time stamps;

[0092] S102, using a Gaussian process regression algorithm to process the original wind field data set, calculating the inter-point correlation of the original data using a covariance function, and constructing a three-dimensional dynamic wind field model by combining the original data set and the inter-point correlation;

[0093] S103, based on the actual spatial coordinates of the wind turbine in the power plant, using a nearest neighbor interpolation algorithm to extract a wind condition parameter vector in the three-dimensional dynamic wind field model;

[0094] S104 , using a yaw control model based on deep deterministic policy gradient reinforcement learning, combined with the wind condition parameter vector, the current yaw state of the wind turbine and the preset maximization cumulative reward function, generates an initial yaw angle adjustment instruction.

[0095] In one embodiment, the method further comprises:

[0096] S201, based on the wind condition parameter vector, using a preset multi-source interference model to perform interference analysis and generate a command interference feature vector;

[0097] S202, combining the yaw angle adjustment instruction and the interference feature vector, generating a candidate adjustment instruction set through a Monte Carlo simulation method;

[0098] S203 , based on the candidate adjustment instruction sets, a multi-objective optimization algorithm is used to screen and obtain a basic optimal instruction set with the goal of minimizing the yaw angle error and maximizing the instruction execution efficiency.

[0099] In one embodiment, the method further comprises:

[0100] S301, based on the wind condition parameter vector, an improved wake propagation model is used to analyze the wake interaction between the target wind turbine and adjacent wind turbines, and a wake influence coefficient matrix is ​​calculated;

[0101] S302, combining the wake influence coefficient matrix and the basic optimal instruction set, using the cooperative yaw algorithm, and generating an anti-wake candidate instruction set through a multi-agent consensus protocol;

[0102] S303, using the preset multi-source interference model, conducts Monte Carlo simulation uncertainty analysis and performance evaluation on the candidate wake turbulence instruction set, and screens the anti-wake yaw adjustment instruction set with the goal of minimizing wake turbulence interference loss and maximizing the energy efficiency of the unit cluster.

[0103] In one embodiment, the method further includes S401, constructing a deep deterministic policy gradient reinforcement learning algorithm using the following formula:

[0104]

[0105] in, represents the optimal yaw angle adjustment instruction at time t, γ∈[0,1] represents the discount factor, E[· represents the expected cumulative reward calculation based on the deep deterministic policy gradient (DDPG) algorithm, r t (w t ,θ t ) represents the reward function, defined as r t =α·cos(Δθ t )+β·eff(θ t ), where Δθ t Indicates the deviation angle between the current yaw angle and the optimal wind direction angle, eff(θ t ) represents the instruction execution efficiency, α and β represent weight coefficients, and α+β=1.

[0106] In one embodiment, the method further comprises:

[0107] S501, based on the yaw angle adjustment instruction, performing timing matching with a preset reference instruction template through a dynamic time warping algorithm to generate a dynamic yaw adjustment sequence;

[0108] S502, when the adjustment amplitude of the dynamic yaw adjustment sequence exceeds a preset threshold, collecting mechanical vibration parameters of the wind turbine in real time, including but not limited to vibration acceleration, velocity, and displacement;

[0109] S503, performing fast Fourier transform spectrum analysis on the mechanical vibration parameters to obtain a vibration characteristic vector;

[0110] At step S504, if the vibration characteristic vector exceeds a preset safety threshold, an adaptive damping control algorithm is used to optimize the velocity curve of the yaw actuator based on a proportional-differential control framework, generating a safe yaw velocity instruction set including segmented velocity thresholds and buffer zone parameters, and simultaneously triggering an early warning mechanism.

[0111] In one embodiment, the method further comprises:

[0112] S601, establishing a mechanical vibration feature database of the wind turbine based on normal vibration spectrum templates at different yaw speeds and angles;

[0113] S602, using a dynamic time warping algorithm to perform similarity matching between the mechanical vibration parameters and the spectrum template, and calculating a vibration anomaly index;

[0114] S603, using the multi-level warning threshold set by the fuzzy logic threshold controller, when the vibration abnormality index exceeds the corresponding threshold, adjust the weight coefficient in the maximum cumulative reward function.

[0115] In one embodiment, the method further comprises:

[0116] S701, based on the 3D dynamic wind farm model, uses WebGL technology to render the wind turbine layout and wake impact, and generates a wind farm energy distribution heat map;

[0117] S702: Combining the wind farm energy distribution heat map, yaw angle adjustment instructions, and vibration anomaly index, perform multi-parameter visualization of the yaw actuator status to generate a dynamic risk assessment matrix including early warning prompts;

[0118] S703, based on the risk assessment matrix, uses machine learning algorithms to analyze the correlation between wind turbine operation trends and energy efficiency improvement space, and generates an interactive optimization solution recommendation dashboard.

[0119] In one embodiment, a computer device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the aforementioned steps of the matrix yaw system function of a wind farm when executing the computer program.

[0120] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0121] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0122] The above-described embodiments merely represent several implementation methods of the embodiments of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the concept of the embodiments of the present application, and these modifications and improvements fall within the scope of protection of the embodiments of the present application.

Claims

1. A matrix yaw system for a wind farm, characterized in that: The system comprises: The data acquisition module is used to obtain the original wind field data set containing three-dimensional spatial coordinates and time stamps using multi-source sensors; A wind field modeling module is used to process the original wind field data set using a Gaussian process regression algorithm, calculate the inter-point correlation of the original data set using a covariance function, and construct a three-dimensional dynamic wind field model based on the original data set; a parameter extraction module, configured to extract a wind condition parameter vector in the three-dimensional dynamic wind field model using a nearest neighbor interpolation algorithm based on the actual spatial coordinates of the wind turbine in the power plant; The yaw control module is used to generate a yaw angle adjustment instruction by adopting a yaw control model based on deep deterministic policy gradient reinforcement learning, combining the wind condition parameter vector, the current yaw state of the wind turbine and a preset maximized cumulative reward function.

2. The system according to claim 1, wherein: The system further includes an instruction optimization module for: Based on the wind condition parameter vector, interference analysis is performed using a preset multi-source interference model to generate a command interference feature vector; Combining the yaw angle adjustment instruction with the interference feature vector, generating a candidate adjustment instruction set through a Monte Carlo simulation method; Based on the candidate adjustment instruction set, a basic optimal instruction set is screened and obtained by utilizing a multi-objective optimization algorithm with the goals of minimizing the yaw angle error and maximizing the instruction execution efficiency.

3. The system according to claim 2, characterized in that The instruction optimization module further includes a wake collaborative optimization unit for: Based on the wind condition parameter vector, an improved wake propagation model is used to analyze the wake interaction between the target wind turbine and adjacent wind turbines, and a wake influence coefficient matrix is ​​calculated; Combining the wake influence coefficient matrix with the basic optimal instruction set, using the collaborative yaw algorithm, and generating the anti-wake candidate instruction set through the multi-agent consensus protocol; The preset multi-source interference model is used to perform Monte Carlo simulation uncertainty analysis and performance evaluation on the candidate anti-wake instruction set, and the anti-wake yaw adjustment instruction set is screened with the goal of minimizing wake interference losses and maximizing the energy efficiency of the unit cluster.

4. The system according to claim 1, wherein: The yaw control module is also used to construct a deep deterministic policy gradient reinforcement learning algorithm using the following formula: in, represents the optimal yaw angle adjustment instruction at time t, γ∈[0,1] represents the discount factor, E[· represents the expected cumulative reward calculation based on the deep deterministic policy gradient (DDPG) algorithm, r t (w t ,θ t ) represents the reward function, defined as r t =α·cos(Δθ t )+β·eff(θ t ), where Δθ t Indicates the deviation angle between the current yaw angle and the optimal wind direction angle, eff(θ t ) represents the instruction execution efficiency, α and β represent weight coefficients, and α+β=1.

5. The system according to claim 1, wherein: The system further includes a safety detection module for: Based on the yaw angle adjustment instruction, a dynamic yaw adjustment sequence is generated by performing timing matching with a preset reference instruction template through a dynamic time warping algorithm; When the adjustment amplitude of the dynamic yaw adjustment sequence exceeds a preset threshold, collecting the mechanical vibration parameters of the wind turbine in real time, including but not limited to vibration acceleration, velocity and displacement; Performing fast Fourier transform spectrum analysis on the mechanical vibration parameters to obtain a vibration characteristic vector; If the vibration characteristic vector exceeds a preset safety threshold, an adaptive damping control algorithm is used to optimize the speed curve of the yaw actuator based on a proportional-differential control framework, generate a safe yaw speed instruction set including segmented speed thresholds and buffer zone parameters, and simultaneously trigger an early warning mechanism.

6. The system according to claim 5, characterized in that The safety detection module is also used for: A mechanical vibration feature database for wind turbines is established based on normal vibration spectrum templates at different yaw speeds and angles; A dynamic time warping algorithm is used to perform similarity matching between the mechanical vibration parameters and the spectrum template, and a vibration anomaly index is calculated; The multi-level warning threshold value is set by the fuzzy logic threshold controller, and when the vibration abnormality index exceeds the corresponding threshold value, the weight coefficient in the maximum cumulative reward function is adjusted.

7. The system according to claim 6, characterized in that The system further includes a visualization module for: Based on the three-dimensional dynamic wind field model, the wind turbine layout and wake impact are rendered using WebGL technology to generate a wind field energy distribution heat map; Combined with the wind field energy distribution thermal map, yaw angle adjustment instructions and vibration abnormality index, multi-parameter visualization processing is performed on the yaw actuator state to generate a dynamic risk assessment matrix including early warning prompts; Based on the risk assessment matrix, a machine learning algorithm is used to perform correlation analysis between the operation trend and energy efficiency improvement space of the wind turbine, and an interactive optimization solution recommendation dashboard is generated.

8. A matrix yaw method for a wind farm, characterized in that: The method comprises: Use multi-source sensors to obtain wind field raw data sets containing three-dimensional spatial coordinates and timestamps; The original wind field data set is processed using a Gaussian process regression algorithm, the inter-point correlation of the original data is calculated using a covariance function, and a three-dimensional dynamic wind field model is constructed by combining the original data set and the inter-point correlation; Based on the actual spatial coordinates of the wind turbine in the power plant, a nearest neighbor interpolation algorithm is used to extract the wind condition parameter vector in the three-dimensional dynamic wind field model; A yaw control model based on deep deterministic policy gradient reinforcement learning is adopted to generate an initial yaw angle adjustment instruction by combining the wind condition parameter vector, the current yaw state of the wind turbine and a preset maximization cumulative reward function.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images