A wind farm wake cooperative control method, device and equipment

Abstract

The application provides a wind farm wake synergic control method, device and equipment, and belongs to the technical field of wind farm intelligent control. The method comprises the following steps: obtaining fan data of any wind farm in at least two wind farms and observation data of an environment where the fan is located; obtaining instant benefit data and a first state vector according to the fan data and the observation data; obtaining a yaw control instruction according to the first state vector; adjusting a fan angle according to the yaw control instruction to obtain a second state vector; obtaining a time sequence difference error according to the instant benefit data, the first state vector and the second state vector; obtaining a local model parameter according to the time sequence difference error; and synergically controlling wakes of each fan group of the wind farm according to the local model parameter. The scheme can reduce equipment loss, improve portability and realize real-time response to wind speed fluctuation.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for wind farms, and in particular to a method, device, and equipment for coordinated control of wind farm wake. Background Technology

[0002] Due to the renewable nature of wind power, wind energy has come into the public eye as an indispensable alternative to traditional energy sources. Current wind turbines can convert wind power into electricity for residential and commercial use, offering a new approach to future energy development in the face of the depletion crisis of traditional energy sources. However, during the process of converting wind power into electricity, various factors can lead to lower-than-expected energy output. For example, while wind turbines extract energy from the wind, they create a wake downstream where wind speed decreases. If a downstream wind turbine is located within this wake, its input wind speed will be lower than that of the upstream turbine. This wake effect causes uneven wind speed distribution within the wind farm, affecting the operation of each wind turbine and further impacting the wind farm's operating conditions and output.

[0003] Currently, wind farms still face multiple technical challenges and industry pain points in the field of wake control. Many unresolved issues severely restrict the power generation efficiency and equipment reliability of wind farms. For example, current mainstream technical solutions mainly rely on computational fluid dynamics simulation tools, which require solving nanodimensional... Stokes equations for wake modeling result in significant modeling lag, meaning high single-calculation delays make it difficult to respond to second-level wind speed fluctuations. This leads to a severe disconnect between control commands and actual operating conditions. Furthermore, when wind speed changes drastically in a short period, traditional models cannot adjust the turbine yaw angle in time, exacerbating the wake superposition effect and causing power generation losses of over 15%. Additionally, existing technologies often prioritize maximizing power generation while neglecting the core requirement of equipment lifespan protection. Frequent adjustments increase gearbox failure rates by 30%, adding millions of yuan to annual maintenance costs. Moreover, wind turbines are often deployed in clusters, with different brands and construction sites in different regions. Using traditional graph neural networks to fix the turbine layout requires re-collecting data and retraining the model when applying it to new sites with different densities or terrains, resulting in adaptation costs exceeding 500,000 yuan per site. Against this backdrop, developing wake collaborative control technology that combines real-time performance, portability, and equipment protection capabilities has become a key path to overcome the bottlenecks in the wind power industry. Summary of the Invention

[0004] This invention provides a method, device, and equipment for coordinated control of wind farm wake, which solves the problems of asynchronous control commands and actual operating conditions, high equipment wear and tear during use, and difficulty in relocating between wind farms.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] This invention provides a wind farm wake coordinated control method, comprising:

[0007] Obtain wind turbine data and observation data of the environment where the wind turbines are located from at least two wind farms;

[0008] Based on the wind turbine data and the observation data, real-time revenue data and a first state vector are obtained;

[0009] Based on the first state vector, the yaw control command is obtained;

[0010] The fan angle is adjusted according to the yaw control command to obtain the second state vector;

[0011] The time-series difference error is obtained based on the real-time revenue data, the first state vector, and the second state vector.

[0012] Based on the time-series difference error, the local model parameters are obtained;

[0013] Based on the local model parameters, the wake of each wind turbine in the wind farm is controlled collaboratively.

[0014] Optionally, acquire wind turbine data from the wind farm and observational data of the environment in which the turbines are located, including:

[0015] Obtain at least one of the following wind turbine data from the wind farm: blade radius, turbine coordinates, number of faults, number of operating days, risk set, yaw error, wind direction change rate, yaw angular velocity, pitch angle, and historical yaw count.

[0016] Obtain at least one of the following observational data: air density, upstream wind speed, measured power, wind speed sequence, average wind speed, and turbulence enhancement coefficient of the environment where the wind turbine is located.

[0017] Optionally, based on the wind turbine data and the observation data, real-time revenue data and a first state vector are obtained, including:

[0018] The maximum power is obtained by using the air density, the upstream wind speed, and the turbine swept area;

[0019] The power difference is obtained based on the maximum power and the measured power.

[0020] Based on the observed data, the turbulence intensity is obtained;

[0021] Based on the turbulence intensity, the preset wake attenuation coefficient, the fan data, and the observation data, the wake influence matrix is ​​obtained;

[0022] The weighting coefficients are obtained based on the historical yaw counts, the turbulence intensity, and the wake influence matrix.

[0023] Based on the weighting coefficients, the wind turbine data, and the feature vector, a baseline risk function is obtained;

[0024] according to This yields the proportional risk model; where, For proportional risk models; Let t be the baseline risk function. This refers to the number of times a ship has veered off course in history. Turbulence intensity; This is the wake effect matrix, where i and j represent the wind turbine; , and These are the weighting coefficients; exp() is the exponential function.

[0025] according to This allows us to obtain real-time revenue data; among which, For real-time revenue data; The difference in work done; For proportional risk models; The preset dynamic penalty weight has a range of [value]. It is the revenue from electricity generation; It's a fatigue penalty; i and j represent the fan.

[0026] The yaw error, the average wind speed, the rate of change of wind direction, the turbulence intensity, the power difference, the yaw angular velocity, and the pitch angle are vectorized to obtain the state vector;

[0027] Each parameter in the state vector is standardized to obtain the first state vector.

[0028] Optionally, based on the first state vector, a yaw control command is obtained, including:

[0029] Based on the first state vector, the first hidden value is obtained;

[0030] Based on the first hidden value, obtain the second hidden value;

[0031] Based on the second hidden value, the mean yaw angle and the standard deviation of the motion are obtained;

[0032] The yaw control command is obtained by sampling the mean yaw angle and the standard deviation of the action.

[0033] Optionally, the fan angle is adjusted according to the yaw control command to obtain a second state vector, including:

[0034] The fan angle is adjusted according to the yaw control command to obtain the adjusted fresh fan data;

[0035] The second state vector is obtained based on the fresh air machine data and the observation data.

[0036] Optionally, based on the real-time revenue data, the first state vector, and the second state vector, the time-series difference error is obtained, including:

[0037] Based on the first state vector, the first long-term value prediction result is obtained;

[0038] Based on the second state vector, the second long-term value prediction result is obtained;

[0039] according to The timing difference error is obtained; where, This refers to timing difference error; For real-time revenue data; Preset discount factor; This is the first long-term value prediction result; This represents the second long-term value prediction result; t indicates the time step.

[0040] Optionally, based on the time-series difference error, local model parameters are obtained, including:

[0041] according to The first network coefficients are obtained; where, These are the first network coefficients; F is the number of samples; b is the sample index; It is the direction of the policy gradient; It is the action probability density output by the first network; It is the timing differential error; t represents the time step; Indicates yaw control command; This represents the first state vector.

[0042] The updated parameters of the first network are obtained by presetting the learning rate, coefficients, and original parameters of the first network.

[0043] according to The second network coefficients are obtained;

[0044] The updated parameters of the second network are obtained by presetting the learning rate, coefficients, and original parameters of the second network.

[0045] in, It is the second network coefficient; It is a timing difference error.

[0046] This invention also provides a wind farm wake coordination control device, comprising:

[0047] The acquisition module is used to acquire wind turbine data and observation data of the environment where the wind turbines are located from any one of at least two wind farms.

[0048] The processing module is configured to: obtain real-time revenue data and a first state vector based on the wind turbine data and the observation data; obtain a yaw control command based on the first state vector; adjust the wind turbine angle based on the yaw control command to obtain a second state vector; obtain a time-series differential error based on the real-time revenue data, the first state vector, and the second state vector; obtain local model parameters based on the time-series differential error; and perform coordinated control of the wake of each wind turbine in the wind farm based on the local model parameters.

[0049] This invention also provides a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when run by the processor, executes the above-described method.

[0050] This invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method.

[0051] The technical solution of the present invention has at least the following effects:

[0052] The above-described solution of the present invention acquires wind turbine data and environmental observation data of any one of at least two wind farms; obtains real-time revenue data and a first state vector based on the wind turbine data and the observation data; obtains yaw control commands based on the first state vector; adjusts the wind turbine angle based on the yaw control commands to obtain a second state vector; obtains a time-series difference error based on the real-time revenue data, the first state vector, and the second state vector; obtains local model parameters based on the time-series difference error; and performs coordinated control of the wake of each wind turbine in the wind farm based on the local model parameters. This improves the model's generalization ability and portability, enables cross-wind farm data sharing, reduces equipment operating losses, extends equipment lifespan, and achieves real-time response to wind speed fluctuations. Attached Figure Description

[0053] Figure 1 This is a flowchart of the wind farm wake coordinated control method provided in the embodiments of the present invention;

[0054] Figure 2 This is a structural diagram of the wind farm wake coordination control device provided in an embodiment of the present invention;

[0055] Figure 3 This is an interaction diagram between the local wind farm and the server in the wind farm wake collaborative control method provided in this embodiment of the invention;

[0056] Figure 4 This is a schematic diagram of the structure of the computing device provided in an embodiment of the present invention. Detailed Implementation

[0057] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0058] like Figure 1 As shown, an embodiment of the present invention proposes a wind farm wake cooperative control method, including:

[0059] Step 11: Obtain wind turbine data and observation data of the environment where the wind turbines are located from any one of the at least two wind farms;

[0060] Step 12: Based on the wind turbine data and the observation data, obtain real-time revenue data and a first state vector;

[0061] Step 13: Obtain the yaw control command based on the first state vector;

[0062] Step 14: Adjust the fan angle according to the yaw control command to obtain the second state vector;

[0063] Step 15: Obtain the time-series difference error based on the real-time revenue data, the first state vector, and the second state vector;

[0064] Step 16: Obtain the local model parameters based on the time-series difference error;

[0065] Step 17: Based on the local model parameters, perform coordinated control of the wake of each wind turbine in the wind farm.

[0066] In this embodiment, the wind turbine data mainly covers the wind turbine's own operating parameters and fixed data, such as wind turbine coordinates. Each wind turbine is installed in a specific location, and this installation location rarely changes, so it can be considered fixed data. Wind turbine blades are an integral part of the wind turbine, and similarly, the blade specifications of a wind turbine rarely change; therefore, the blade radius can be considered a wind turbine's own operating parameter. Observational data comes from external monitoring equipment, including data acquired using Supervisory Control and Data Acquisition (SCADA) systems and other observation methods, including various wind speed data, as wind speed changes at different heights affect the formation and propagation of the wake; and measured power, the actual power result measured by the SCADA system. By integrating this wind turbine data and observational data, information support can be provided for subsequent analysis and decision-making.

[0067] After acquiring the data, specific algorithms and models are used to perform in-depth analysis and processing, including obtaining immediate benefits and a first state vector. Immediate benefits are a quantitative assessment of the wind turbine's operating performance in its current state, comprehensively considering multiple factors such as current power generation, turbine operating efficiency, and the impact of wake on surrounding turbines. By accurately calculating immediate benefits, the operational efficiency of the wind turbine at the current moment can be intuitively understood. In addition, a first state vector can also be constructed based on the data. The first state vector is a seven-dimensional data set that integrates various key information from wind turbine data and observation data, describing the current operating state of the wind turbine and environmental conditions.

[0068] The first state vector is processed by an Actor network. A three-layer fully connected neural network is used to generate optimal yaw angle suggestions and control the intensity of action exploration. Action sampling is performed to generate optimal yaw control commands. The yaw control commands specify the angle and direction that the wind turbine needs to adjust, aiming to keep the wind turbine in optimal alignment with the wind direction at all times, thereby maximizing wind energy capture, improving power generation efficiency, and reducing the adverse effects of wake on other wind turbines.

[0069] After receiving the yaw control command, the wind turbine's yaw system starts operating, precisely adjusting the turbine's angle according to the command requirements. Once the adjustment is complete, relevant data from the wind turbine is collected again, and combined with the current observation data, a second state vector is constructed. The second state vector reflects the new operating state and environmental conditions of the wind turbine after executing the yaw control command. Compared with the first state vector, the second state vector reflects the state changes brought about by the adjustment of the wind turbine's angle, providing an important basis for subsequent evaluation of the adjustment effect.

[0070] Step 15 involves a Critic network evaluation, which comprehensively assesses the effectiveness of the current adjustments. Immediate benefits, yaw control commands, the first state vector, and the second state vector are used as input parameters. A specific evaluation model is employed for calculation. The time-series difference error is the core evaluation indicator, measuring the difference between the actual and expected operating results of the wind turbine after executing the yaw control command. A small time-series difference error indicates that the yaw control command effectively improved the wind turbine's operating state and increased power generation efficiency; conversely, a large error indicates deficiencies in the control command, requiring further optimization. Accurate calculation of the time-series difference error allows for timely identification of problems in the control process, providing direction for subsequent model parameter updates.

[0071] The parameters of the local control model are updated based on the calculated time-series differential error. According to the magnitude and direction of the error, the parameters in the model are adjusted so that the model can more accurately predict the operating effect of the wind turbine under different conditions, thereby generating better yaw control commands.

[0072] The updated local model parameters are uploaded to the federated server. As the central hub for the coordinated control of the entire wind farm, the federated server collects local model parameters from each wind farm. Through the integration and analysis of these parameters, it obtains the aggregate weights for each wind farm. Based on these weights, a global model is derived that is shared by all wind farms. This global model comprehensively considers the operating characteristics and environmental differences of different wind turbines within the entire wind farm, exhibiting broader applicability and higher accuracy. The server distributes the global model to each wind farm, overwriting the existing local models, thus achieving unified updates and optimization of the entire wind farm's control model. This iterative optimization of the control model improves the effectiveness of wind farm wake coordination control, ultimately enhancing the overall operational efficiency and economic benefits of the wind power plant.

[0073] This technical solution differs from traditional mainstream solutions in its calculation method, reducing computational latency, enabling real-time model control, and incorporating equipment usage and protection into the value system. It balances power generation and equipment lifespan, and eliminates the need for complete retraining when the wind farm changes, thus reducing the cost of solution migration.

[0074] In an optional embodiment of the present invention, step 12 may include:

[0075] Step 121: The maximum power is obtained by using the air density, the upstream wind speed, and the turbine swept area; the turbine swept area is obtained by using the blade radius, which is the area of ​​the circle with the blade as the radius.

[0076] Step 122: Obtain the power difference value based on the maximum power and the measured power;

[0077] Step 123: Obtain the turbulence intensity based on the observed data;

[0078] Step 124: Based on the turbulence intensity, the preset wake attenuation coefficient, the fan data, and the observation data, obtain the wake influence matrix;

[0079] Step 125: Obtain the weighting coefficients based on the historical yaw counts, the turbulence intensity, and the wake influence matrix;

[0080] Step 126: Obtain the baseline risk function based on the weighting coefficients, the wind turbine data, and the feature vector;

[0081] Step 127, according to This yields the proportional risk model; where, For proportional risk models; Let t be the baseline risk function. This refers to the number of times a ship has veered off course in history. Turbulence intensity; This is the wake effect matrix, where i and j represent the wind turbine; , and These are the weighting coefficients; exp() is the exponential function;

[0082] Step 128, according to This allows us to obtain real-time revenue data; among which, For immediate benefits; The difference in work done; For proportional risk models; The preset dynamic penalty weight has a range of [value]. ; It is the revenue from electricity generation; It is a fatigue penalty; n and m represent the total number of samples; i and j represent the fan;

[0083] Step 129: Vectorize the yaw error, the average wind speed, the wind direction change rate, the turbulence intensity, the power difference, the yaw angular velocity, and the pitch angle to obtain the state vector;

[0084] Step 130: Standardize each parameter in the state vector to obtain the first state vector.

[0085] In this embodiment, in step 121, by The swept area of ​​the turbine is obtained, according to To obtain maximum power;

[0086] In step 122, the difference between the calculated maximum power and the actual measured power is calculated, based on... The power difference value is obtained, which reflects the degree of deviation between the actual operating power of the wind turbine and the theoretical maximum power. This data is affected by a variety of factors, such as the mechanical loss of the wind turbine, the efficiency of the control system, and the wake effect. By analyzing the power difference value, the operating status and performance of the wind turbine can be preliminarily assessed.

[0087] in, Where A is the blade radius; A is the turbine swept area; This refers to the air density (approximately 1.225 kg / m³, under standard conditions). The upstream wind speed is the real-time wind speed under ideal conditions (the frontmost fan). For maximum power, that is ; The measured power is the actual power output; i represents the fan.

[0088] Turbulence is a common and complex flow phenomenon in wind fields, which can cause drastic changes in wind speed and direction in a short period of time, adversely affecting the operational stability of wind turbines.

[0089] Step 123: Extract minute-by-minute wind speed sequences from the meteorological database. ,according to Obtain the standard deviation of wind speed, and then according to Accurately obtaining the turbulence intensity of the current wind field is crucial. The magnitude of this turbulence intensity reflects the strength of turbulence in the wind field and is a key indicator for assessing the severity of the wind turbine's operating environment. is the standard deviation of wind speed, n is the total number of samples in the wind speed sequence; z is the index of the current wind speed sample. It is the z-th wind speed sample sequence; It is the average wind speed per unit time. It is the turbulence coefficient.

[0090] The preset wake attenuation coefficient is a parameter pre-set based on the physical characteristics of the wind turbine and the experience of wind farm layout. It describes the attenuation law of the wake during the propagation process. By combining wind turbine data (such as wind turbine location) and observation data (such as wind speed), the degree of wake influence between different wind turbines can be obtained and represented in the form of a matrix, namely the wake influence matrix.

[0091] In step 124, the wind turbine coordinates are used to determine the wind turbine's position. Obtain the distance to the wind turbine; based on the distance to the wind turbine, it is possible to... Obtain the wind direction offset angle; then according to The wake influence matrix is ​​obtained, which can intuitively reflect the mutual influence relationships between wind turbines in the wind farm, providing an important basis for subsequent risk assessment and control strategy formulation. Where (x, y) are the wind turbine coordinates; i and j are the wind turbines; It is the distance between fan i and fan j; It is the wind direction offset angle; It is the preset wake attenuation coefficient, which can be manually set, and its default value can be set to 0.05; It is the average wind speed per unit time; k is the turbulence enhancement coefficient, which is calibrated through wind tunnel experiments, and the turbulence enhancement coefficient is different for each wind turbine. It is the turbulence coefficient.

[0092] In step 125, according to The weighting coefficients are derived, which accurately reflect the combined impact of each factor on the operation of the wind turbine. Among them, It is a partial likelihood function; m is the total number of wind turbine samples; The weighting coefficient is, i.e. ; It is an eigenvector, which is , The number of yaw movements per unit time. For turbulence intensity, The wake effect matrix; It is a fault indicator variable. This indicates a fault. This indicates that there is no fault. It is the set of wind turbines that have not yet failed at time t; i and j are wind turbines.

[0093] The baseline risk function is a mathematical model used to assess the risk of wind turbine failure or performance degradation under specific operating conditions. This model comprehensively considers the influence of various factors reflected by weighting coefficients, equipment status information in the wind turbine data, and operating history and environmental conditions in the feature vector, enabling it to quantitatively calculate the baseline risk level of the wind turbine under its current condition. The baseline risk function can be obtained, where t is the equipment failure time, i.e., the failure time of the fan bearing / gearbox; d is... The number of wind turbine failures at any given time; Let X be the set of wind turbines that have not yet failed at time t; X is the eigenvector, which is... The number of yaw movements per unit time. For turbulence intensity, The wake effect matrix; The weighting coefficient is, i.e. .

[0094] In step 127, according to This yields a proportional risk model, representing the overall risk level of a wind turbine after considering the influence of multiple factors. Among these factors, For proportional risk models; The baseline risk function at time t reflects the risk status of the wind turbine under foundation conditions; The number of historical yaw operations is considered; frequent yaw operations will increase the mechanical wear and tear and the risk of failure of the wind turbine. Due to the intensity of turbulence, a highly turbulent environment will cause greater stress on the structural components and control system of the wind turbine; The wake effect matrix considers the impact of wake interactions between wind turbines on risk, where i and j represent wind turbines; , and , where are weighting coefficients, representing the contribution of historical yaw counts, turbulence intensity, and wake influence matrix to the overall risk, respectively.

[0095] In step 128, according to This provides real-time revenue data, comprehensively reflecting the operational efficiency of the wind turbine at the current moment; among which, For immediate benefits; The difference in work done; For proportional risk models; The preset dynamic penalty weight has a range of [value]. This is used to adjust the weight of fatigue penalty in the calculation of immediate benefits, and should be set reasonably according to the actual operation and management needs of the wind farm. It is the power generation revenue, which is obtained by summing the power difference. It represents the economic benefits brought about by the gap between the actual power generation of the wind turbine and the theoretical maximum power generation. It is a fatigue penalty, which takes into account the impact of fatigue damage caused by various factors (such as yaw, turbulence, wake, etc.) on the life and operational reliability of the wind turbine during operation; n and m represent the total number of samples; i and j represent the wind turbines;

[0096] In step 129, the yaw error, average wind speed, rate of change of wind direction, turbulence intensity, work difference, yaw angular velocity, and pitch angle are vectorized to obtain the state vector. The state vector is a set of vectors containing seven key parameters, which can comprehensively describe the current operating state and environmental conditions of the wind turbine. Yaw error directly reflects the accuracy of wind alignment and has a significant impact on the power generation efficiency of wind turbines. It is the average wind speed per unit time, which determines aerodynamic efficiency. Yaw sensitivity varies at different wind speeds. It is the rate of change of wind direction, which can predict future trends and adjust the yaw speed in advance; It is the turbulence intensity, which reflects the strength of turbulence in the wind field and is related to the operational stability of the wind turbine; It is the power difference, which reflects the gap between the actual power generation of the wind turbine and the theoretical maximum power, and provides direct performance feedback; It is the yaw rate, which can avoid sudden movements and ensure smooth action; It is the pitch angle, used to coordinate and control changes in the wind turbine.

[0097] Since the parameters in the state vector have different dimensions and numerical ranges, directly using the original state vector for analysis and processing can lead to the exaggeration or neglect of the role of some parameters, affecting the accuracy and reliability of the analysis results.

[0098] In step 130, a standardization method is used. Each parameter in the state vector is processed and converted into standardized data with the same dimensions and numerical range to obtain the first state vector. This first state vector more fairly and accurately reflects the relative influence of each parameter on the wind turbine's operating state, providing more reliable data support for subsequent wind turbine control strategy formulation and performance evaluation. It is the first state vector; These are the various parameters; It is the mean of the parameters; It is the standard deviation of the parameter; These are standard parameters.

[0099] In an optional embodiment of the present invention, step 13 may include:

[0100] Step 131: Obtain the first hidden value based on the first state vector;

[0101] Step 132: Obtain the second hidden value based on the first hidden value;

[0102] Step 133: Based on the second hidden value, obtain the mean yaw angle and the standard deviation of the motion;

[0103] Step 134: Sample the mean yaw angle and the standard deviation of the action to obtain the yaw control command.

[0104] In this embodiment, according to Get the first hidden value;

[0105] according to Obtain the second hidden value;

[0106] according to and The mean yaw angle and the standard deviation of the movement are obtained;

[0107] The yaw control commands follow a normal distribution with a mean of the yaw angle and a variance of the squared standard deviation of the actions. and according to The final yaw control command is obtained.

[0108] in, , , and For weight parameters, , The and It is a feature extractor, whose main function is to encode the original state into high-level decision features; and all The It is the core of action decision-making, and its main function is to learn how to adjust the yaw angle according to the state. It is an exploration strategy regulator, whose main function is to increase randomness when environmental uncertainty is high; , , and It is the bias vector; and ; and ; and It is an activation threshold regulator, whose main function is to dynamically control the effectiveness of features through bias. It is the default yaw reference, the default yaw angle when there is no external excitation (usually ≈0); It is the basic exploration intensity, ensuring minimum exploration and preventing the strategy from converging too early; and This is the loss function.

[0109] In an optional embodiment of the present invention, step 14 may include:

[0110] Step 141: Adjust the fan angle according to the yaw control command to obtain the adjusted new fan data;

[0111] Step 142: Obtain the second state vector based on the fresh air machine data and the observation data.

[0112] In this embodiment, the yaw control command includes adjusting the direction (clockwise or counterclockwise yaw) and adjusting the angle. After adjusting the fan angle according to the yaw control command, the fan data is remeasured to obtain the new fan data. Based on the new fan data and the observation data, the seven parameters are re-vectorized to obtain the state vector. The state vector is then standardized again to obtain the second state vector.

[0113] In an optional embodiment of the present invention, step 15 may include:

[0114] Step 151: Obtain the first long-term value prediction based on the first state vector;

[0115] Step 152: Obtain the second long-term value prediction based on the second state vector;

[0116] Step 153, according to The timing difference error is obtained; where, This refers to timing difference error; For real-time revenue data; Preset discount factor; This is the first long-term value prediction result; This represents the second long-term value prediction result; t indicates the time step.

[0117] In this embodiment, according to The first long-term value prediction result was obtained, based on The second long-term value prediction result was obtained, based on The temporal difference error is obtained, which is the deviation between the predicted value and the actual reward. The closer the temporal difference error is to 0, the higher the superiority of the model. It is the first state vector; It is the second state vector; , represents the feature extraction weights of the Critic network, represents the relationship between the learning state and the long-term value, h = the hidden layer dimension, which is 16 here; , representing the bias benchmark for feature extraction; , representing the value mapping weight; Representing state The long-term value prediction, i.e., the first long-term value prediction result. Representing state The long-term value prediction, i.e. the second long-term value prediction result, for example: high wind speed + small yaw error → high value; For timing difference error, A value greater than 0 indicates that actual returns exceeded expectations, thus encouraging current actions. A value less than 0 indicates that the actual return is lower than expected, and the probability of taking that action should be reduced. =0 means that the actual return is the same as the expected return; For immediate benefits; The preset discount factor is set so that a value close to 0 represents a focus on immediate rewards, while a value close to 1 represents a focus on long-term benefits. Since the cost of gearbox damage far outweighs short-term power generation losses, this factor is set to a value close to 1 to prioritize long-term benefits in the system. High discount factor.

[0118] like Figure 3 As shown, in an optional embodiment of the present invention, step 16 may include:

[0119] Step 161, according to The first network coefficients are obtained;

[0120] Step 162: By presetting the first network learning rate, the first network coefficients and the original parameters of the first network, the updated new parameters of the first network are obtained;

[0121] Step 163, according to The second network coefficients are obtained;

[0122] Step 164: By presetting the learning rate, coefficients, and original parameters of the second network, the updated parameters of the second network are obtained.

[0123] Step 165: Upload the local model parameters to the server; the local model parameters include the first new network parameters and the second new network parameters;

[0124] Furthermore, the method also includes:

[0125] Step 166, according to This yields the global model;

[0126] Step 167: Distribute the global model to each wind farm, overriding the local model of the wind farm;

[0127] in, It is a timing difference error; b is the action probability density output by the first network; b is the sample index. It represents the direction of the policy gradient, indicating how quickly the probability of the action increases due to the adjustment; F is the number of samples. It is the first network coefficient; It is the second network coefficient; It is the time decay factor, according to Seeking, is the maximum data age of the wind field, p is an artificially set decay target, which means that the weight of the oldest data is p times that of the newest data; k and d are the wind field; For the local model parameters of wind field k; This refers to the amount of data for wind field k. It is the amount of data for wind field d; It refers to the freshness of the wind field data (k). Q represents the data freshness of wind farm d; Q represents the number of wind farms. and It is an exponentially decaying term; It is a global model.

[0128] In this embodiment, according to and The dual network coefficients are obtained, wherein the dual network coefficients include a first network coefficient and a second network coefficient; according to and The dual network parameters are updated to obtain local model parameters, which include new parameters for the first and second networks. After uploading the local model parameters to the server, the aggregation weights for each wind field are obtained based on data from different wind fields. The global model is then derived based on these aggregation weights. The federated server distributes the global model to each wind farm, which then overwrites the local models in each wind farm, thereby improving the model's generalization ability and portability. It is a timing difference error; b is the action probability density output by the first network; b is the sample index. It represents the direction of the policy gradient, indicating how quickly the probability of the action increases due to the adjustment; F is the number of samples. It is the first network coefficient; It is the second network coefficient; It is the time decay factor, according to Seeking, is the maximum data age of the wind field, p is an artificially set decay target, which means that the weight of the oldest data is p times that of the newest data; k and d are the wind field; For the local model parameters of wind field k; This refers to the amount of data for wind field k. It is the amount of data for wind field d; It refers to the freshness of the wind field data (k). Q represents the data freshness of wind farm d; Q represents the number of wind farms. and It is an exponentially decaying term; It is a global model.

[0129] A specific embodiment of the wind farm wake coordinated control method provided in this invention is as follows:

[0130] Step 1: Acquire wind turbine data and observation data. The wind turbine data mainly includes the wind turbine's own operating parameters and fixed data, such as wind turbine coordinates and blade radius. The observation data comes from external monitoring equipment, including data obtained using SCADA systems and other observation methods, such as various wind speed data and measured power.

[0131] Step 2, through The swept area of ​​the turbine is obtained, according to To obtain the maximum power, according to Obtain the power difference value; extract the minute-by-minute wind speed sequence from the meteorological database. ,according to Obtain the standard deviation of wind speed, and then according to Obtain the turbulence intensity; using the wind turbine coordinates, based on Obtain the distance to the wind turbine, and based on the distance to the wind turbine, according to... Obtain the wind direction offset angle, and then according to Obtain the wake influence matrix; based on The weighting coefficients are derived; based on Obtain the baseline risk function; based on The proportional risk model is obtained; based on Immediate benefits are obtained; the yaw error, average wind speed, wind direction change rate, turbulence intensity, work difference, yaw angular velocity, and pitch angle are vectorized to obtain the state vector. ;according to Data standardization is performed to obtain the first state vector.

[0132] in, Where A is the blade radius; A is the turbine swept area; This refers to the air density (approximately 1.225 kg / m³, under standard conditions). The upstream wind speed is the real-time wind speed under ideal conditions (the frontmost fan). For maximum power, that is ; The measured power is based on the actual power measured by the SCADA system; i represents the fan. is the standard deviation of wind speed, n is the total number of samples in the wind speed sequence; z is the index of the current wind speed sample. It is the z-th wind speed sample sequence; It is the average wind speed per unit time. This is the turbulence coefficient. (x, y) are the wind turbine coordinates; i and j are the wind turbine coordinates. It is the distance between fan i and fan j; It is the wind direction offset angle; It is the preset wake attenuation coefficient, which can be manually set, and its default value can be set to 0.05; It is the average wind speed per unit time; k is the turbulence enhancement coefficient, which is calibrated through wind tunnel experiments, and the turbulence enhancement coefficient is different for each wind turbine. It is the turbulence coefficient. It is a partial likelihood function; m is the total number of samples; The weighting coefficient is, i.e. ; It is an eigenvector, which is , The number of yaw movements per unit time. For turbulence intensity, The wake effect matrix; It is a fault indicator variable. This indicates a fault. This indicates that there is no fault. It is the set of wind turbines that have not yet failed at time t. t is the equipment failure time, i.e., the failure time of the wind turbine bearing / gearbox; d is the number of wind turbine failures at time t; It is the set of wind turbines that have not yet failed at time t. For proportional risk models; Let t be the baseline risk function. , and , where are weighting coefficients, representing the contribution of historical yaw counts, turbulence intensity, and wake influence matrix to the overall risk, respectively. For immediate benefits; The preset dynamic penalty weight has a range of [value]. ; It is the revenue from electricity generation; It is a fatigue penalty; n and m represent the total number of samples. It is yaw error; It is the average wind speed per unit time. It is the rate of change of wind direction; It is the turbulence intensity; It is the difference in work done; It is the yaw rate; It is the propeller pitch angle. It is the first state vector; These are the various parameters; It is the mean of the parameters; It is the standard deviation of the parameter; These are standard parameters.

[0133] Step 3, according to Get the first hidden value; based on Obtain the second hidden value; based on and The mean yaw angle and the standard deviation of the action are obtained; the yaw control command and according to The final yaw control command is obtained. Among them, , , and For weight parameters, , The and It is a feature extractor; and all The It is the core of action decision-making; It is an exploration strategy regulator; , , and It is a bias vector. and , and , and It activates the threshold regulator; It is the default yaw reference. It is the basic exploration intensity; and This is the loss function.

[0134] Step 4: Adjust the fan angle according to the yaw control command to obtain the adjusted fresh fan data; obtain the second state vector based on the fresh fan data and the observation data.

[0135] Step 5, according to Obtain the first long-term value forecast, based on The second long-term value forecast is obtained based on The timing difference error is obtained. Among them, It is the first state vector; It is the second state vector; , represents the feature extraction weights of the Critic network, represents the relationship between the learning state and the long-term value, h = the hidden layer dimension, which is 16 here; , representing the bias benchmark for feature extraction; , representing the value mapping weight; Representing state The long-term value prediction, namely the first long-term value prediction; Representing state The long-term value prediction, namely the second long-term value prediction; For timing difference error, A value greater than 0 indicates that actual returns exceeded expectations, thus encouraging current actions. A value less than 0 indicates that the actual return is lower than expected, and the probability of taking that action should be reduced. =0 means that the actual return is the same as the expected return; For immediate benefits; The preset discount factor is set so that a value close to 0 represents a focus on immediate rewards, while a value close to 1 represents a focus on long-term benefits. Since the cost of gearbox damage far outweighs short-term power generation losses, this factor is set to a value close to 1 to prioritize long-term benefits in the system. High discount factor.

[0136] Step 6, according to and The dual network coefficients are obtained, wherein the dual network coefficients include a first network coefficient and a second network coefficient; according to and The dual network parameters are updated to obtain local model parameters, which include new parameters for the first and second networks. After uploading the local model parameters to the server, the aggregation weights for each wind field are obtained based on data from different wind fields. The global model is then derived based on these aggregation weights. The federated server distributes the global model to each wind farm, thus overriding the local models of each wind farm. It is a timing difference error; b is the action probability density output by the first network; b is the sample index. It represents the direction of the policy gradient, indicating how quickly the probability of the action increases due to the adjustment; F is the number of samples. It is the first network coefficient; It is the second network coefficient; It is the time decay factor, according to Seeking, is the maximum data age of the wind field, p is an artificially set decay target, which means that the weight of the oldest data is p times that of the newest data; k and d are the wind field; For the local model parameters of wind field k; This refers to the amount of data for wind field k. It is the amount of data for wind field d; It refers to the freshness of the wind field data (k). Q represents the data freshness of wind farm d; Q represents the number of wind farms. and It is an exponentially decaying term; It is a global model.

[0137] The wind farm wake collaborative control method proposed in this invention balances the economic situation and lifespan of the wind farm through a dual-network model. It can achieve a high level of power generation while protecting the equipment, thereby extending the equipment lifespan, improving the model's scalability, and minimizing the negative impact of wake effects.

[0138] like Figure 2 As shown, this embodiment of the invention also provides a wind farm wake coordination control device 20, comprising:

[0139] The acquisition module 21 is used to acquire wind turbine data and observation data of the environment where the wind turbines are located from any one of the at least two wind farms.

[0140] The processing module 22 is configured to: obtain real-time revenue data and a first state vector based on the wind turbine data and the observation data; obtain a yaw control command based on the first state vector; adjust the wind turbine angle based on the yaw control command to obtain a second state vector; obtain a time-series differential error based on the real-time revenue data, the first state vector, and the second state vector; obtain local model parameters based on the time-series differential error; and perform coordinated control of the wake of each wind turbine in the wind farm based on the local model parameters.

[0141] Optionally, processing module 22 is specifically used for:

[0142] The maximum power is obtained by using the air density, the upstream wind speed, and the turbine swept area;

[0143] The power difference is obtained based on the maximum power and the measured power.

[0144] Based on the observed data, the turbulence intensity is obtained;

[0145] Based on the turbulence intensity, the preset wake attenuation coefficient, the fan data, and the observation data, the wake influence matrix is ​​obtained;

[0146] The weighting coefficients are obtained based on the historical yaw counts, the turbulence intensity, and the wake influence matrix.

[0147] Based on the weighting coefficients, the wind turbine data, and the feature vector, a baseline risk function is obtained;

[0148] according to This yields the proportional risk model; where, For proportional risk models; Let t be the baseline risk function. This refers to the number of times a ship has veered off course in history. Turbulence intensity; This is the wake effect matrix, where i and j represent the wind turbine; , and These are the weighting coefficients; exp() is the exponential function.

[0149] according to This allows us to obtain real-time revenue data; among which, For real-time revenue data; The difference in work done; For proportional risk models; The preset dynamic penalty weight has a range of [value]. It is the revenue from electricity generation; It's a fatigue penalty; i and j represent the fan.

[0150] The yaw error, the average wind speed, the rate of change of wind direction, the turbulence intensity, the power difference, the yaw angular velocity, and the pitch angle are vectorized to obtain the state vector;

[0151] Each parameter in the state vector is standardized to obtain the first state vector.

[0152] Optionally, the processing module 22 is also specifically used for:

[0153] Based on the first state vector, the first hidden value is obtained;

[0154] Based on the first hidden value, obtain the second hidden value;

[0155] Based on the second hidden value, the mean yaw angle and the standard deviation of the motion are obtained;

[0156] The yaw control command is obtained by sampling the mean yaw angle and the standard deviation of the action.

[0157] Optionally, the fan angle is adjusted according to the yaw control command to obtain a second state vector, including:

[0158] The fan angle is adjusted according to the yaw control command to obtain the adjusted fresh fan data;

[0159] The second state vector is obtained based on the fresh air machine data and the observation data.

[0160] Optionally, based on the real-time revenue data, the first state vector, and the second state vector, the time-series difference error is obtained, including:

[0161] Based on the first state vector, the first long-term value prediction result is obtained;

[0162] Based on the second state vector, the second long-term value prediction result is obtained;

[0163] according to The timing difference error is obtained; where, This refers to timing difference error; For real-time revenue data; Preset discount factor; This is the first long-term value prediction result; This represents the second long-term value prediction result; t indicates the time step.

[0164] Optionally, based on the time-series difference error, local model parameters are obtained, including:

[0165] according to The first network coefficients are obtained; where, These are the first network coefficients; F is the number of samples; b is the sample index; It is the direction of the policy gradient; It is the action probability density output by the first network; It is the timing differential error; t represents the time step; Indicates yaw control command; This represents the first state vector.

[0166] The updated parameters of the first network are obtained by presetting the learning rate, coefficients, and original parameters of the first network.

[0167] according to The second network coefficients are obtained;

[0168] The updated parameters of the second network are obtained by presetting the learning rate, coefficients, and original parameters of the second network.

[0169] in, It is the second network coefficient; It is a timing difference error.

[0170] It should be noted that this device is a device corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0171] like Figure 4 As shown, this embodiment of the invention also provides a computing device 40, including a processor 41, a memory 42, and a program or instructions stored in the memory 42 and executable on the processor 41. When the program or instructions are executed by the processor 41, they implement the various processes of the above-described wind farm wake coordinated control method embodiment and achieve the same technical effects. To avoid repetition, they will not be described again here. It should be noted that the computing device in this embodiment of the invention includes the aforementioned mobile electronic devices and non-mobile electronic devices.

[0172] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0173] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0174] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0175] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0176] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0177] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0178] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0179] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code for implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps for performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0180] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A wind farm wake coordination control method, characterized in that, include: Obtain wind turbine data and observation data of the environment where the wind turbines are located from at least two wind farms; Based on the wind turbine data and the observation data, real-time revenue data and a first state vector are obtained; Based on the first state vector, the yaw control command is obtained; The fan angle is adjusted according to the yaw control command to obtain the second state vector; The time-series difference error is obtained based on the real-time revenue data, the first state vector, and the second state vector. Based on the time-series difference error, the local model parameters are obtained; Based on the local model parameters, the wake of each wind turbine in the wind farm is controlled collaboratively. This includes acquiring wind turbine data from the wind farm and observational data of the environment in which the turbines are located, including: Obtain at least one of the following wind turbine data from the wind farm: blade radius, turbine coordinates, number of faults, number of operating days, risk set, yaw error, wind direction change rate, yaw angular velocity, pitch angle, and historical yaw count. Obtain at least one of the following observational data: air density, upstream wind speed, measured power, wind speed sequence, average wind speed, and turbulence enhancement coefficient of the environment where the wind turbine is located; Specifically, based on the wind turbine data and the observation data, real-time revenue data and a first state vector are obtained, including: The maximum power is obtained by using the air density, the upstream wind speed, and the turbine swept area; The power difference is obtained based on the maximum power and the measured power. Based on the observed data, the turbulence intensity is obtained; Based on the turbulence intensity, the preset wake attenuation coefficient, the fan data, and the observation data, the wake influence matrix is ​​obtained; The weighting coefficients are obtained based on the historical yaw counts, the turbulence intensity, and the wake influence matrix. Based on the weighting coefficients, the wind turbine data, and the feature vector, a baseline risk function is obtained; According to , a proportional hazard model is obtained; wherein, is the proportional hazard model; is the baseline hazard function at time t; is the historical number of deviations; is the turbulence intensity; is the wake effect matrix, i and j represent the wind turbines; , and are weight coefficients respectively; exp() is the exponential function; According to , the instant revenue data is obtained; wherein, is the instant revenue data; is the power difference value; is the proportional hazard model; is the preset dynamic penalty weight, the range is is the power generation revenue; is the fatigue penalty, i and j represent the wind turbine; The yaw error, the average wind speed, the rate of change of wind direction, the turbulence intensity, the power difference, the yaw angular velocity, and the pitch angle are vectorized to obtain the state vector; Standardize each parameter in the state vector to obtain the first state vector; The local model parameters are obtained based on the time-series difference error, including: According to , a first network coefficient is obtained; wherein, is the first network coefficient; F is the number of samples; b is the sample subscript; is the policy gradient direction; is the action probability density of the first network output; is the time difference error; t represents the time step; represents the yaw control command; represents the first state vector; The updated parameters of the first network are obtained by presetting the learning rate, coefficients, and original parameters of the first network. According to , a second network coefficient is obtained; The updated parameters of the second network are obtained by presetting the learning rate, coefficients, and original parameters of the second network. wherein, is a second network coefficient; is a timing difference error.

2. The wind farm wake coordination control method according to claim 1, characterized in that, Based on the first state vector, yaw control commands are obtained, including: Based on the first state vector, the first hidden value is obtained; Based on the first hidden value, obtain the second hidden value; Based on the second hidden value, the mean yaw angle and the standard deviation of the motion are obtained; The yaw control command is obtained by sampling the mean yaw angle and the standard deviation of the action.

3. The wind farm wake coordination control method of claim 1, wherein, The fan angle is adjusted according to the yaw control command to obtain the second state vector, including: The fan angle is adjusted according to the yaw control command to obtain the adjusted fresh fan data; The second state vector is obtained based on the fresh air machine data and the observation data.

4. The wind farm wake coordination control method of claim 1, wherein, Based on the real-time revenue data, the first state vector, and the second state vector, the time-series difference error is obtained, including: Based on the first state vector, the first long-term value prediction result is obtained; Based on the second state vector, the second long-term value prediction result is obtained; according to The timing difference error is obtained; where, This refers to timing difference error; For real-time revenue data; Preset discount factor; This is the first long-term value prediction result; This represents the second long-term value prediction result; t indicates the time step.

5. A wind farm wake coordination control device, implemented using the method described in any one of claims 1 to 4, characterized in that, include: The acquisition module is used to acquire wind turbine data and observation data of the environment where the wind turbines are located from any one of at least two wind farms. The processing module is configured to: obtain real-time revenue data and a first state vector based on the wind turbine data and the observation data; obtain yaw control commands based on the first state vector; adjust the wind turbine angle based on the yaw control commands to obtain a second state vector; obtain a time-series differential error based on the real-time revenue data, the first state vector, and the second state vector; obtain local model parameters based on the time-series differential error; and perform coordinated wake control on the wind turbines of the wind farm based on the local model parameters. The acquisition of wind turbine data and observation data of the wind turbine environment includes: acquiring at least one of the following: blade radius, turbine coordinates, number of faults, number of operating days, risk set, yaw error, wind direction change rate, yaw angular velocity, pitch angle, and historical yaw counts for the wind turbines in the wind farm. Data; acquire at least one of the following observational data for the environment where the wind turbine is located: air density, upstream wind speed, measured power, wind speed sequence, average wind speed, and turbulence enhancement coefficient; wherein, based on the wind turbine data and the observational data, obtain real-time revenue data and a first state vector, including: obtaining the maximum power using the air density, upstream wind speed, and turbine swept area; obtaining the power difference value based on the maximum power and the measured power; obtaining the turbulence intensity based on the observational data; obtaining the wake influence matrix based on the turbulence intensity, a preset wake attenuation coefficient, the wind turbine data, and the observational data; obtaining weighting coefficients based on the historical yaw counts, the turbulence intensity, and the wake influence matrix; obtaining a baseline risk function based on the weighting coefficients, the wind turbine data, and the feature vector; based on... This yields the proportional hazards model; where, For proportional risk models; Let t be the baseline risk function. This refers to the number of times a ship has veered off course in history. Turbulence intensity; This is the wake effect matrix, where i and j represent the wind turbine; , and These are the weighting coefficients; exp() is the exponential function; according to This allows us to obtain real-time revenue data; among which, For real-time revenue data; The difference in work done; For proportional risk models; The preset dynamic penalty weight has a range of [value]. It is the revenue from electricity generation; This is a fatigue penalty, where i and j represent the wind turbine; the yaw error, the average wind speed, the rate of change of wind direction, the turbulence intensity, the power difference, the yaw angular velocity, and the pitch angle are vectorized to obtain a state vector; each parameter in the state vector is standardized to obtain a first state vector; wherein, based on the time-series difference error, the local model parameters are obtained, including: based on The first network coefficients are obtained; where, These are the first network coefficients; F is the number of samples; b is the sample index; It is the direction of the policy gradient; It is the action probability density output by the first network; It is the timing difference error; t represents the time step; Indicates yaw control command; This represents the first state vector; by presetting the first network learning rate, first network coefficients, and the original parameters of the first network, the updated parameters of the first network are obtained; according to... The second network coefficients are obtained; by presetting the second network learning rate, the second network coefficients, and the original parameters of the second network, the updated parameters of the second network are obtained; where, It is the second network coefficient; It is a timing difference error.

6. A computing device, characterized in that, include: A processor, a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The system stores instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 4.

Images