Wind turbine generator torque optimization method and related device

Abstract

The invention belongs to the technical field of wind power generation control, and discloses a wind turbine generator torque optimization method and a related device, and the method comprises the steps: obtaining the state data of a wind turbine generator; according to the state data of the wind turbine generator, a rotating speed control value of the wind turbine generator is obtained through a pre-trained generator torque control model; wherein the generator torque control model is constructed on the basis of a reinforcement learning model, a reward function of the generator torque control model is constructed on the basis of dynamic weighted generating capacity reward and mechanical stress punishment, and the weight of dynamic weighting is determined on the basis of the system health degree of a wind turbine generator. Based on the autonomous optimization potential of the reinforcement learning technology, the method has good adaptability, and can accurately and quickly realize generator torque optimization. By adopting a multi-target collaborative optimization reward function design, equipment health state monitoring data are effectively fused, the balance between power generation efficiency and mechanical loss is realized, and stable operation of a generator of a wind turbine generator is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind power generation control, and relates to a method for optimizing the torque of a wind turbine generator and a related device. Background Art

[0002] As an important equipment for renewable energy development, the operation and control strategy of wind turbines is directly related to power generation efficiency and equipment life. At present, the following methods are mainly used to optimize the generator torque of wind turbines: (1) The classical control algorithm based on PID (Proportional-Integral-Differential) achieves maximum power point tracking through fixed parameter adjustment, but it is difficult to adapt to the complex turbulent conditions at sea and does not consider the time-varying characteristics of mechanical loads, resulting in an intensified contradiction between power generation efficiency and equipment loss; (2) Rule-based control strategies such as IPC (Independent Pitch Control) and torque graded regulation can reduce mechanical stress, but due to the lack of dynamic adaptability of the preset threshold, it is easy to cause power oscillation when the wind speed changes suddenly. Summary of the Invention

[0003] The purpose of the present invention is to overcome the above-mentioned shortcomings of the prior art and provide a method for optimizing the torque of a wind turbine generator and related devices.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a method for optimizing the torque of a wind turbine generator, comprising: obtaining status data of the wind turbine generator; obtaining a speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator; wherein the generator torque control model is constructed based on a reinforcement learning model, the reward function of the generator torque control model is constructed based on a dynamically weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

[0006] Optionally, the status data of the wind turbine generator includes one or more of the following: vibration spectrum, spatial temperature image, unit power, rotation speed and main shaft torque.

[0007] Optionally, the method of obtaining the speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator includes: extracting characteristic data of the status data of the wind turbine generator; wherein, the vibration spectrum uses a one-dimensional convolutional neural network to extract characteristic data; the spatial temperature image uses a two-dimensional convolutional neural network to extract characteristic data; the unit power, speed and main shaft torque all use a graph attention network to extract characteristic data; the characteristic data of the status data of the wind turbine generator is input into the pre-trained generator torque control model, and the speed control value of the wind turbine generator is output through the pre-trained generator torque control model.

[0008] Optionally, the pre-trained generator torque control model is obtained by: constructing a virtual simulation environment for a wind turbine generator; constructing a generator torque control model, and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; obtaining historical operating data of the current wind turbine generator, and fine-tuning the initially trained generator torque control model based on the historical operating data to obtain a pre-trained generator torque control model.

[0009] Optionally, fine-tuning the initially trained generator torque control model according to the historical operating data includes: constructing a fine-tuning training sample according to the historical operating data; and obtaining a training weight of the fine-tuning training sample by the following formula:

[0010]

[0011] Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of .

[0012] The fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

[0013] Optionally, before obtaining the speed control value of the wind turbine generator through the pre-trained generator torque control model, it also includes: obtaining the pre-trained generator torque control model of each wind turbine generator in the set area, and optimizing the model parameters of the pre-trained generator torque control model through federated learning.

[0014] Optionally, before obtaining the status data of the wind turbine generator, the step further includes: obtaining real-time stress impact data of the wind turbine generator; obtaining a real-time stress prediction value of the wind turbine generator through a pre-trained stress prediction model based on the real-time stress impact data of the wind turbine generator; when the real-time stress prediction value is less than a preset stress threshold, continuing the step of obtaining the status data of the wind turbine generator; otherwise, stopping the step of obtaining the status data of the wind turbine generator, and controlling the speed of the wind turbine generator through the controller provided by the wind turbine.

[0015] Optionally, the real-time stress impact data of the wind turbine generator includes one or more of the following: ambient wind speed, vibration amplitude and vibration frequency.

[0016] Optionally, the generator torque control model is constructed based on a hierarchical reinforcement learning model, the top model is used to determine the minute-level speed reference value of the generator, and the bottom model is used to determine the millisecond-level speed control value of the generator.

[0017] Optionally, the reward function R of the generator torque control model t for:

[0018] R t =α(t)·R powr +β(t)·R stress +γ·R constraint

[0019]

[0020] R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -ω)) 2 ]

[0021] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t P is the generator output power at the current moment; ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w i is the stress weight of generator component i; σ i is the real-time stress value of generator component i; ωallow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

[0022] Optionally, the system health of the wind turbine generator is obtained by: obtaining system health impact data of the wind turbine generator; obtaining the system health of the wind turbine generator through a pre-trained system health prediction model based on the system health impact data of the wind turbine generator; wherein the system health prediction model is constructed based on a long short-term memory network.

[0023] Optionally, the system health impact data of the wind turbine generator includes one or more of the following: time-frequency characteristics of vibration signals, metal particle concentration of gearbox oil, and winding temperature change rate.

[0024] Optionally, the weight of the dynamic weighting is determined by the following formula:

[0025]

[0026] α(t)+β(t)=1

[0027] Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

[0028] According to a second aspect of the present invention, a wind turbine generator torque optimization system is provided, comprising: a data acquisition module for acquiring status data of the wind turbine generator; a speed control module for obtaining a speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator; wherein the generator torque control model is constructed based on a reinforcement learning model, the reward function of the generator torque control model is constructed based on a dynamic weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

[0029] Optionally, the pre-trained generator torque control model is obtained by: constructing a virtual simulation environment for a wind turbine generator; constructing a generator torque control model, and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; obtaining historical operating data of the current wind turbine generator, and fine-tuning the initially trained generator torque control model based on the historical operating data to obtain a pre-trained generator torque control model.

[0030] Optionally, fine-tuning the initially trained generator torque control model according to the historical operating data includes: constructing a fine-tuning training sample according to the historical operating data; and obtaining a training weight of the fine-tuning training sample by the following formula:

[0031]

[0032] Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of .

[0033] The fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

[0034] Optionally, before acquiring the status data of the wind turbine generator, the data acquisition module is also used to: acquire real-time stress impact data of the wind turbine generator; obtain the real-time stress prediction value of the wind turbine generator through a pre-trained stress prediction model based on the real-time stress impact data of the wind turbine generator; when the real-time stress prediction value is less than a preset stress threshold, continue the step of acquiring the status data of the wind turbine generator; otherwise, stop the step of acquiring the status data of the wind turbine generator, and control the speed of the wind turbine generator through the controller provided by the wind turbine.

[0035] Optionally, the generator torque control model is constructed based on a hierarchical reinforcement learning model, the top model is used to determine the minute-level speed reference value of the generator, and the bottom model is used to determine the millisecond-level speed control value of the generator.

[0036] Optionally, the reward function R of the generator torque control model t for:

[0037] R t =α(t)·R power +β(t)·R stress +γ·R constraint

[0038]

[0039] R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -ω))2 ]

[0040] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t is the generator output power at the current moment; P ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w i is the stress weight of generator component i; σ i is the real-time stress value of generator component i; σ allow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

[0041] Optionally, the system health of the wind turbine generator is obtained by: obtaining system health impact data of the wind turbine generator; obtaining the system health of the wind turbine generator through a pre-trained system health prediction model based on the system health impact data of the wind turbine generator; wherein the system health prediction model is constructed based on a long short-term memory network.

[0042] Optionally, the weight of the dynamic weighting is determined by the following formula:

[0043]

[0044] α(t)+β(t)=1

[0045] Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

[0046] In a third aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for optimizing the torque of a wind turbine generator when executing the computer program.

[0047] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for optimizing the torque of a wind turbine generator are implemented.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The wind turbine generator torque optimization method of the present invention is based on the state data of the wind turbine generator and adopts a pre-trained generator torque control model based on a reinforcement learning model to achieve effective prediction of the speed control value of the wind turbine generator, thereby achieving effective optimization of the wind turbine generator torque. Based on the autonomous optimization potential of reinforcement learning technology, it has good adaptability, can accurately and quickly achieve generator torque optimization, and effectively solve the problems existing in the existing generator torque optimization method. At the same time, the reward function of the generator torque control model is constructed based on dynamic weighted power generation rewards and mechanical stress penalties, and the dynamic weighted weight is determined based on the system health of the wind turbine generator. By adopting a multi-objective collaborative optimization reward function design, it can effectively integrate equipment health status monitoring data, achieve a balance between power generation efficiency and mechanical loss, and ensure the safe and stable operation of the wind turbine generator. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a method for optimizing the torque of a wind turbine generator according to an embodiment of the present invention.

[0051] Figure 2 This is a diagram of the vibration spectrum amplitude of a wind turbine generator according to an embodiment of the present invention.

[0052] Figure 3 This is a diagram of voltage, current and power of a wind turbine generator according to an embodiment of the present invention.

[0053] Figure 4 This is a structural block diagram of a wind turbine generator torque optimization system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0055] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0056] The present invention is described in further detail below with reference to the accompanying drawings:

[0057] See also Figure 1 In one embodiment of the present invention, a method for optimizing the torque of a wind turbine generator is provided, which can accurately and quickly optimize the generator torque and ensure the safe and stable operation of the wind turbine generator.

[0058] Specifically, the wind turbine generator torque optimization method of the present invention includes the following steps:

[0059] S1: Obtain status data of wind turbine generator.

[0060] S2: According to the state data of the wind turbine generator, the speed control value of the wind turbine generator is obtained through the pre-trained generator torque control model.

[0061] Among them, the generator torque control model is constructed based on a reinforcement learning model, the reward function of the generator torque control model is constructed based on a dynamically weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

[0062] The wind turbine generator torque optimization method of the present invention is based on the state data of the wind turbine generator and adopts a pre-trained generator torque control model based on a reinforcement learning model to achieve effective prediction of the speed control value of the wind turbine generator, thereby achieving effective optimization of the wind turbine generator torque. Based on the autonomous optimization potential of reinforcement learning technology, it has good adaptability, can accurately and quickly achieve generator torque optimization, and effectively solve the problems existing in the existing generator torque optimization method. At the same time, the reward function of the generator torque control model is constructed based on dynamic weighted power generation rewards and mechanical stress penalties, and the dynamic weighted weight is determined based on the system health of the wind turbine generator. By adopting a multi-objective collaborative optimization reward function design, it can effectively integrate equipment health status monitoring data, achieve a balance between power generation efficiency and mechanical loss, and ensure the safe and stable operation of the wind turbine generator.

[0063] In a possible implementation manner, the status data of the wind turbine generator includes one or more of the following: vibration spectrum, spatial temperature image, unit power, rotation speed, and main shaft torque.

[0064] Interpretatively, vibration spectra can reveal the vibration characteristics of mechanical structures, indicating component wear or failure. Spatial temperature images can visually display temperature distribution, identifying anomalies such as localized overheating. Unit power reflects power generation efficiency. Speed ​​is closely related to wind energy conversion and power generation control. Main shaft torque reflects the force applied to the rotor. By comprehensively utilizing this data, we can comprehensively and accurately assess the generator's condition, providing a rich basis for torque optimization, thereby achieving more precise speed control and improving power generation efficiency and unit reliability.

[0065] In a possible embodiment, obtaining the speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator includes: extracting characteristic data of the status data of the wind turbine generator; wherein, the vibration spectrum uses a one-dimensional convolutional neural network to extract characteristic data; the spatial temperature image uses a two-dimensional convolutional neural network to extract characteristic data; the unit power, speed and main shaft torque all use a graph attention network to extract characteristic data; the characteristic data of the status data of the wind turbine generator is input into the pre-trained generator torque control model, and the speed control value of the wind turbine generator is output through the pre-trained generator torque control model.

[0066] For example, a one-dimensional convolutional neural network includes five layers of convolution, with convolution kernels of 64, 32, 16, 8, and 4, respectively, to extract feature data of minute-to-second dynamic characteristics of the vibration spectrum. A two-dimensional convolutional neural network can use the ResNet-18 (Residual Network-18) architecture network to effectively extract the spatial gradient characteristics of the temperature field. The unit power, speed, and spindle torque can all be graph-modeled based on the sensor topology, and then feature data can be extracted based on the graph attention network.

[0067] In one possible embodiment, the pre-trained generator torque control model is obtained by: constructing a virtual simulation environment for a wind turbine generator; constructing a generator torque control model, and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; obtaining historical operating data of the current wind turbine generator, and fine-tuning the initially trained generator torque control model based on the historical operating data to obtain a pre-trained generator torque control model.

[0068] Explanatory, by building a virtual simulation environment of wind turbine generators, comprehensive acquisition of training data is achieved to effectively ensure the training effect of the generator torque control model.

[0069] For example, when building a virtual simulation environment for a wind turbine generator, first, Bladed software is used to build an aerodynamic load and mechanical load simulation module. This module takes key parameters such as wind speed and pitch angle as input, and after precise algorithm calculation and model simulation, outputs data reflecting load characteristics such as rotor thrust and main shaft torque. At the same time, a generator electromagnetic characteristic model is built in the Simulink environment. This model comprehensively covers core elements such as stator / rotor windings and flux equations, and can accurately simulate the electromagnetic characteristics of the generator during operation. After completing the construction of the above two models, they are connected to the RTDS simulation system. With the help of the high-speed real-time simulation core and synchronous clock module of the RTDS simulation system, the data interaction timing between the two models is precisely aligned with microsecond accuracy, thereby successfully realizing real-time coupled simulation of loads and electromagnetic characteristics, providing a reliable virtual simulation platform for performance analysis and optimization of wind turbine generators.

[0070] Explanatory, a reinforcement learning framework is constructed that integrates digital twins with a dynamic reward mechanism. After completing the initial training of the generator torque control model, the generator torque control model in the virtual environment is transferred to the wind turbine generator. This transfer is mainly achieved by fine-tuning the initially trained generator torque control model using historical operating data of the wind turbine generator to improve the model's adaptability to the target wind turbine.

[0071] In one possible implementation, fine-tuning the initially trained generator torque control model based on the historical operating data includes: constructing a fine-tuning training sample based on the historical operating data; and obtaining a training weight for the fine-tuning training sample using the following formula:

[0072]

[0073] Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of .

[0074] The fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

[0075] Explanatory, the migration process of the generator torque control model adopts a virtual-to-reality domain adaptation algorithm based on the maximum mean difference. By calculating the distance between the distribution of virtual environment data and the distribution of real environment data, the domain difference is quantified, and then weights are assigned to each sample of real environment data, namely the fine-tuning training sample. Among them, when fine-tuning the initially trained generator torque control model using fine-tuning training samples with training weights, the training weights of the fine-tuning training samples are incorporated into the loss function of the model, so that the model pays more attention to situations that are more likely to occur in the real environment, so as to achieve efficient migration of the generator torque control model from virtual to reality.

[0076] In a possible embodiment, before obtaining the speed control value of the wind turbine generator through the pre-trained generator torque control model, it also includes: obtaining the pre-trained generator torque control model of each wind turbine generator in the set area, and optimizing the model parameters of the pre-trained generator torque control model through federated learning.

[0077] Explanatory, by obtaining the pre-trained generator torque control model of each wind turbine generator in the set area and optimizing the model parameters with the help of federated learning, it can make full use of the rich local operating data of each wind turbine. These data cover operating information under different working conditions and environmental conditions, enabling the model to learn more comprehensive and representative features, and improve the model's adaptability to complex and changeable actual scenarios.

[0078] For example, when optimizing the model parameters of the pre-trained generator torque control model through federated learning, a weighted average strategy can be adopted, that is, the model parameters of the pre-trained generator torque control model of each wind turbine generator are weighted averaged and then the current model parameters are updated.

[0079] In a possible embodiment, before obtaining the status data of the wind turbine generator, the step further includes: obtaining real-time stress impact data of the wind turbine generator; obtaining a real-time stress prediction value of the wind turbine generator through a pre-trained stress prediction model based on the real-time stress impact data of the wind turbine generator; when the real-time stress prediction value is less than a preset stress threshold, continuing the step of obtaining the status data of the wind turbine generator; otherwise, stopping the step of obtaining the status data of the wind turbine generator, and controlling the speed of the wind turbine generator through the controller provided by the wind turbine.

[0080] Explanatory, a safety control mechanism is designed, that is, safety control is achieved by comparing the real-time stress prediction value with the preset stress threshold. For example, the preset stress threshold can be designed to be 0.9σ allow , where σ allow The allowable stress of the generator material is simplified here, and the allowable stress of the material of all the generator components is set to one value. Of course, different settings can be made according to different components. When the real-time stress prediction value σ of the wind turbine generator is pred ≥0.9σ allow When σ is set, the controller built into the wind turbine is used to control the speed of the wind turbine generator to protect the safe operation of the wind turbine generator. pred <0.9σ allow When the generator torque is optimized, the pre-trained generator torque control model is used to optimize the generator torque of the wind turbine, so that the generator torque is in the optimal state and efficient operation is guaranteed.

[0081] In a possible implementation manner, the real-time stress impact data of the wind turbine generator includes one or more of the following: ambient wind speed, vibration amplitude, and vibration frequency.

[0082] Optionally, a pre-trained stress prediction model is constructed based on a deep neural network, which mines the intrinsic relationship between stress impact data and stress prediction values ​​through the deep neural network, thereby achieving accurate prediction of real-time stress prediction values ​​based on real-time stress impact data to ensure the effectiveness of the safety control mechanism.

[0083] In one possible implementation, the generator torque control model is constructed based on a hierarchical reinforcement learning model, where the top model is used to determine a minute-level speed reference value of the generator, and the bottom model is used to determine a millisecond-level speed control value of the generator.

[0084] Explanatory, a generator torque control model based on a hierarchical reinforcement learning model is constructed. The top-level model is set to determine the generator speed reference value at the minute level, while the bottom-level model determines the speed control value at the millisecond level. This allows the top-level model to grasp the overall operating trend of the generator from a macro perspective, providing stable directional guidance for the bottom-level control, ensuring that the generator operation conforms to the overall plan. The bottom-level model focuses on microscopic rapid response, able to promptly respond to various dynamic changes on the millisecond time scale to ensure stable operation of the generator. This balance of macro-planning and micro-response effectively improves the accuracy and adaptability of generator torque control.

[0085] In one possible implementation, the reward function R of the generator torque control model is t for:

[0086] R t =α(t)·R power +β(t)·R stress +γ·R constraint

[0087]

[0088] R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -ω)) 2 ]

[0089] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t is the generator output power at the current moment; P ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w i is the stress weight of generator component i; σ i is the real-time stress value of generator component i; σ allow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

[0090] For example, the nonlinear penalty index k∈[2, 4], preferably k=3, reflects the cubic accumulation law of material fatigue damage. The fixed value γ of the constraint penalty coefficient is typically 0.5 to 1.0, and is used to strengthen the hard constraint.

[0091] In a possible embodiment, the system health of the wind turbine generator is obtained by: obtaining system health impact data of the wind turbine generator; and obtaining the system health of the wind turbine generator through a pre-trained system health prediction model based on the system health impact data of the wind turbine generator; wherein the system health prediction model is constructed based on a long short-term memory network.

[0092] Optionally, the system health impact data of the wind turbine generator includes one or more of the following: time-frequency characteristics of vibration signals, metal particle concentration of gearbox oil, and winding temperature change rate.

[0093] In a possible implementation, the weight of the dynamic weighting is determined by the following formula:

[0094]

[0095] α(t)+β(t)=1

[0096] Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

[0097] Explanatory, a dynamic weighted weight calculation formula is designed based on the logistic function. When the SOH value changes, the formula can smoothly and continuously adjust the value of α(t). Specifically, when the SOH approaches 0.8, the function value changes significantly, reflecting the sensitive impact of system health on the weight of power generation rewards. If the SOH is higher than 0.8, it indicates that the generator system is in good health. At this time, α(t) will approach a larger value, which means that a higher weight is given to the power generation reward, encouraging the generator to fully utilize its power generation capacity in a healthy state and improve energy output efficiency. Conversely, when the SOH is lower than 0.8, α(t) will gradually decrease, reflecting concerns about the health of the generator, reducing the weight of the power generation reward to avoid potential damage to the generator due to excessive pursuit of power generation. At the same time, setting α(t) + β(t) = 1 creates a dynamic balance between the power generation reward and the mechanical stress penalty in terms of weight distribution. When the generator SOH is high, the weight of the mechanical stress penalty is appropriately reduced to promote more power generation; when the SOH decreases, the weight of the mechanical stress penalty is increased to impose stricter constraints on the operation of the generator to prevent the aging and damage of the generator caused by excessive mechanical stress, thereby effectively extending the service life of the generator and ensuring the long-term stable operation of the wind turbine.

[0098] In one possible implementation, a 5MW offshore wind turbine generator is used as an example to illustrate the method for optimizing the torque of the wind turbine generator of the present invention. The parameter settings of the 5MW offshore wind turbine generator are shown in Table 1:

[0099] Table 1

[0100] parameter Value source <![CDATA[P rated ]]> 5000kW Equipment Technical Specifications <![CDATA[σ allow ]]> 320MPa Gearbox material fatigue test data <![CDATA[T max ]]> 4200kN·m Design safety factor 1.2 times rated value <![CDATA[ω min ]]> 5rpm Aerodynamic stall critical speed

[0101] See also Figure 2 , shows the vibration spectrum amplitude diagram of the 5MW offshore wind turbine generator. Figure 3 , which shows the voltage, current and power diagram of the 5MW offshore wind turbine generator.

[0102] Real-time running data: P t =5,200kW; P ref =4,950kW;ω t =12.3rpm;ω opt =12.8rpm; SOH = 0.85, then α(t) = 0.73, β(t) = 0.27; the stress of the gearbox components is 335MPa, and the stress of the main bearing components is 285MPa. The reward function of the generator torque control model is calculated as:

[0103] Power generation bonus items:

[0104] Mechanical stress penalty:

[0105] Constraint penalty term: Assume the current torque T = 4150 kN·m <T max , speed ω=12.3rpm>ω min , so the constraint penalty is 0.

[0106] The total reward value of the generator torque control model is -0.211. This negative reward guides the generator torque control model to reduce the speed to reduce mechanical stress, achieving a balance between power generation and equipment protection.

[0107] In general, the wind turbine generator torque optimization method of the present invention realizes intelligent control of the generator torque by constructing a closed-loop optimization system that integrates virtual and real elements. First, a high-fidelity digital twin environment is built based on a multi-physics field coupling model, and then a dynamic weighted multi-objective reward function is designed. The weight value is adjusted in real time through the predicted system health to achieve a weight balance between power generation efficiency and mechanical loss. A hierarchical reinforcement learning architecture is used to carry out virtual pre-training, and the generator torque control is decomposed into two cascade control stages. The top-level reinforcement learning network generates a minute-level reference value for the current torque, and the bottom-level reinforcement learning network generates a real-time millisecond-level control value. The two-part network jointly realizes the real-time torque control of the generator set. In the online deployment stage, the stress prediction value is predicted in real time, and the switching between the reinforcement learning strategy and the self-controller strategy is realized by combining the safety action filtering mechanism. At the same time, the virtual model is adapted to the target wind turbine by using transfer learning to improve the adaptability of the model to the target unit, and the model parameter optimization results of multiple target units in the transfer learning process are shared through federated learning to improve the model optimization efficiency of multiple target units and realize multi-machine collaborative optimization.

[0108] The following are device embodiments of the present invention, which can be used to implement the method embodiments of the present invention. For details not disclosed in the device embodiments, please refer to the method embodiments of the present invention.

[0109] See also Figure 4 In another embodiment of the present invention, a wind turbine generator torque optimization system is provided, which can be used to implement the above-mentioned wind turbine generator torque optimization method. Specifically, the wind turbine generator torque optimization system includes a data acquisition module and a speed control module.

[0110] Among them, the data acquisition module is used to obtain the status data of the wind turbine generator; the speed control module is used to obtain the speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator; wherein, the generator torque control model is constructed based on a reinforcement learning model, and the reward function of the generator torque control model is constructed based on a dynamic weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

[0111] In a possible implementation manner, the status data of the wind turbine generator includes one or more of the following: vibration spectrum, spatial temperature image, unit power, rotation speed, and main shaft torque.

[0112] In a possible embodiment, obtaining the speed control value of the wind turbine generator through a pre-trained generator torque control model based on the status data of the wind turbine generator includes: extracting characteristic data of the status data of the wind turbine generator; wherein, the vibration spectrum uses a one-dimensional convolutional neural network to extract characteristic data; the spatial temperature image uses a two-dimensional convolutional neural network to extract characteristic data; the unit power, speed and main shaft torque all use a graph attention network to extract characteristic data; the characteristic data of the status data of the wind turbine generator is input into the pre-trained generator torque control model, and the speed control value of the wind turbine generator is output through the pre-trained generator torque control model.

[0113] In one possible embodiment, the pre-trained generator torque control model is obtained by: constructing a virtual simulation environment for a wind turbine generator; constructing a generator torque control model, and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; obtaining historical operating data of the current wind turbine generator, and fine-tuning the initially trained generator torque control model based on the historical operating data to obtain a pre-trained generator torque control model.

[0114] In one possible implementation, fine-tuning the initially trained generator torque control model based on the historical operating data includes: constructing a fine-tuning training sample based on the historical operating data; and obtaining a training weight for the fine-tuning training sample using the following formula:

[0115]

[0116] Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of ; the fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

[0117] In a possible embodiment, before obtaining the speed control value of the wind turbine generator through the pre-trained generator torque control model, it also includes: obtaining the pre-trained generator torque control model of each wind turbine generator in the set area, and optimizing the model parameters of the pre-trained generator torque control model through federated learning.

[0118] In a possible embodiment, before acquiring the status data of the wind turbine generator, the data acquisition module is further used to: acquire real-time stress impact data of the wind turbine generator; obtain a real-time stress prediction value of the wind turbine generator through a pre-trained stress prediction model based on the real-time stress impact data of the wind turbine generator; when the real-time stress prediction value is less than a preset stress threshold, continue the step of acquiring the status data of the wind turbine generator; otherwise, stop the step of acquiring the status data of the wind turbine generator, and control the speed of the wind turbine generator through the controller provided by the wind turbine.

[0119] In a possible implementation manner, the real-time stress impact data of the wind turbine generator includes one or more of the following: ambient wind speed, vibration amplitude, and vibration frequency.

[0120] In one possible implementation, the generator torque control model is constructed based on a hierarchical reinforcement learning model, where the top model is used to determine a minute-level speed reference value of the generator, and the bottom model is used to determine a millisecond-level speed control value of the generator.

[0121] In one possible implementation, the reward function R of the generator torque control model is t for:

[0122] R t =α(t)·R powr +β(t)·R stress +γ·R constraint

[0123]

[0124] R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -ω)) 2 ]

[0125] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t is the generator output power at the current moment; P ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w iis the stress weight of generator component i; σ i is the real-time stress value of generator component i; σ allow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

[0126] In a possible embodiment, the system health of the wind turbine generator is obtained by: obtaining system health impact data of the wind turbine generator; and obtaining the system health of the wind turbine generator through a pre-trained system health prediction model based on the system health impact data of the wind turbine generator; wherein the system health prediction model is constructed based on a long short-term memory network.

[0127] In a possible implementation, the system health impact data of the wind turbine generator includes one or more of the following: time-frequency characteristics of vibration signals, metal particle concentration in gearbox oil, and winding temperature change rate.

[0128] In a possible implementation, the weight of the dynamic weighting is determined by the following formula:

[0129]

[0130] α(t)+β(t)=1

[0131] Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

[0132] All relevant contents of each step involved in the embodiment of the aforementioned wind turbine generator torque optimization method can be referred to the functional description of the functional modules corresponding to the wind turbine generator torque optimization system in the embodiment of the present invention, and will not be repeated here.

[0133] The module division in the embodiments of the present invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in various embodiments of the present invention may be integrated into a single processor, exist physically as separate modules, or two or more modules may be integrated into a single module. The integrated modules may be implemented in either hardware or software functional modules.

[0134] In another embodiment of the present invention, a computer device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of the wind turbine generator torque optimization method.

[0135] In another embodiment of the present invention, the present invention further provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the wind turbine generator torque optimization method in the above embodiment.

[0136] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0137] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0138] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0139] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for optimizing the torque of a wind turbine generator, characterized in that: include: Obtain status data of wind turbine generators; According to the status data of the wind turbine generator, the speed control value of the wind turbine generator is obtained through the pre-trained generator torque control model; Among them, the generator torque control model is constructed based on a reinforcement learning model, the reward function of the generator torque control model is constructed based on a dynamically weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

2. The method for optimizing the torque of a wind turbine generator according to claim 1, wherein: The status data of the wind turbine generator includes one or more of the following: Vibration spectrum, spatial temperature image, unit power, speed and main shaft torque.

3. The method for optimizing wind turbine generator torque according to claim 2, characterized in that: The step of obtaining the speed control value of the wind turbine generator by using a pre-trained generator torque control model according to the state data of the wind turbine generator includes: Extract characteristic data from the status data of wind turbine generators; a one-dimensional convolutional neural network is used to extract characteristic data from the vibration spectrum; a two-dimensional convolutional neural network is used to extract characteristic data from the spatial temperature image; and a graph attention network is used to extract characteristic data from the unit power, speed, and main shaft torque. The characteristic data of the state data of the wind turbine generator is input into a pre-trained generator torque control model, and the speed control value of the wind turbine generator is output through the pre-trained generator torque control model.

4. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: The pre-trained generator torque control model is obtained in the following way: Construct a virtual simulation environment for wind turbine generators; Constructing a generator torque control model and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; Historical operating data of the current wind turbine generator is obtained, and an initially trained generator torque control model is fine-tuned according to the historical operating data to obtain a pre-trained generator torque control model.

5. The method for optimizing wind turbine generator torque according to claim 4, characterized in that: The fine-tuning of the initially trained generator torque control model according to the historical operating data includes: Constructing fine-tuning training samples based on the historical operation data; The training weights of the fine-tuning training samples are obtained by the following formula: Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of The fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

6. The method for optimizing wind turbine generator torque according to claim 4, characterized in that: Before obtaining the speed control value of the wind turbine generator through the pre-trained generator torque control model, the method further includes: The pre-trained generator torque control model of each wind turbine generator in the set area is obtained, and the model parameters of the pre-trained generator torque control model are optimized through federated learning.

7. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: Before obtaining the status data of the wind turbine generator, the method further includes: Obtain real-time stress impact data of wind turbine generators; According to the real-time stress impact data of the wind turbine generator, the real-time stress prediction value of the wind turbine generator is obtained through the pre-trained stress prediction model; When the real-time stress prediction value is less than the preset stress threshold, the step of obtaining the status data of the wind turbine generator is continued; otherwise, the step of obtaining the status data of the wind turbine generator is stopped, and the speed of the wind turbine generator is controlled by the controller provided by the wind turbine.

8. The method for optimizing wind turbine generator torque according to claim 7, characterized in that: The real-time stress impact data of the wind turbine generator includes one or more of the following: Ambient wind speed, vibration amplitude and vibration frequency.

9. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: The generator torque control model is constructed based on a hierarchical reinforcement learning model, where the top model is used to determine the minute-level speed reference value of the generator, and the bottom model is used to determine the millisecond-level speed control value of the generator.

10. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: The reward function R of the generator torque control model is t for: R t =α(t)·R power +β(t)·R stress +γ·R constraint R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -oh)) 2 ] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t is the generator output power at the current moment; P ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w i is the stress weight of generator component i; σ i is the real-time stress value of generator component i; ω allow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

11. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: The system health of the wind turbine generator is obtained in the following manner: Obtain system health impact data for wind turbine generators; According to the system health impact data of the wind turbine generator, the system health of the wind turbine generator is obtained through the pre-trained system health prediction model; Among them, the system health prediction model is built based on the long short-term memory network.

12. The method for optimizing wind turbine generator torque according to claim 11, characterized in that: The system health impact data of the wind turbine generator includes one or more of the following: Time-frequency characteristics of vibration signals, gearbox oil metal particle concentration and winding temperature change rate.

13. The method for optimizing wind turbine generator torque according to claim 1, characterized in that: The weight of the dynamic weighting is determined by the following formula: α(t)+β(t)=1 Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

14. A wind turbine generator torque optimization system, characterized in that: include: A data acquisition module is used to obtain status data of the wind turbine generator; The speed control module is used to obtain the speed control value of the wind turbine generator through the pre-trained generator torque control model according to the state data of the wind turbine generator; Among them, the generator torque control model is constructed based on a reinforcement learning model, the reward function of the generator torque control model is constructed based on a dynamically weighted power generation reward and a mechanical stress penalty, and the weight of the dynamic weighting is determined based on the system health of the wind turbine generator.

15. The wind turbine generator torque optimization system according to claim 14, characterized in that: The pre-trained generator torque control model is obtained in the following way: Construct a virtual simulation environment for wind turbine generators; Constructing a generator torque control model and training it based on the virtual simulation environment to obtain an initially trained generator torque control model; Historical operating data of the current wind turbine generator is obtained, and an initially trained generator torque control model is fine-tuned according to the historical operating data to obtain a pre-trained generator torque control model.

16. The wind turbine generator torque optimization system according to claim 15, characterized in that: The fine-tuning of the initially trained generator torque control model according to the historical operating data includes: Constructing fine-tuning training samples based on the historical operation data; The training weights of the fine-tuning training samples are obtained by the following formula: Among them, x i is the i-th fine-tuning training sample; w xi is x i The training weight of p real (x i ) is the historical operation data x i The probability density of virtual (x i ) is x in the virtual simulation environment i The probability density of The fine-tuning training samples with training weights are used to fine-tune the initially trained generator torque control model.

17. The wind turbine generator torque optimization system according to claim 14, characterized in that: Before acquiring the status data of the wind turbine generator, the data acquisition module is further used to: Obtain real-time stress impact data of wind turbine generators; According to the real-time stress impact data of the wind turbine generator, the real-time stress prediction value of the wind turbine generator is obtained through the pre-trained stress prediction model; When the real-time stress prediction value is less than the preset stress threshold, continuing with the step of acquiring status data of the wind turbine generator; Otherwise, the step of obtaining the status data of the wind turbine generator is stopped, and the rotation speed of the wind turbine generator is controlled by the controller provided by the wind turbine.

18. The wind turbine generator torque optimization system according to claim 14, characterized in that: The generator torque control model is constructed based on a hierarchical reinforcement learning model, where the top model is used to determine the minute-level speed reference value of the generator, and the bottom model is used to determine the millisecond-level speed control value of the generator.

19. The wind turbine generator torque optimization system according to claim 14, characterized in that: The reward function R of the generator torque control model is t for: R t =α(t)·R power +β(t)·R stress +γ·R constraint R constraint =-[(max(0,TT max )) 2 +(max(0,ω min -oh)) 2 ] Among them, α(t) is the weight of the power generation reward; R power is the power generation reward; β(t) is the weight of the mechanical stress penalty; R stress is the mechanical stress penalty; γ is the fixed value of the constraint penalty coefficient; R constraint To constrain punishment; P t P is the generator output power at the current moment; ref is the reference power value; P rated is the rated power of the generator; ω t is the generator speed at the current moment; ω opt is the optimal speed setting value; σ is the speed deviation tolerance coefficient; w i is the stress weight of generator component i; σ i is the real-time stress value of generator component i; ω allow is the allowable stress of the generator material; k is the nonlinear penalty index; T is the real-time torque of the generator; T max is the maximum torque of the generator, ω is the real-time speed of the generator; ω min is the minimum speed of the generator.

20. The wind turbine generator torque optimization system according to claim 14, characterized in that: The system health of the wind turbine generator is obtained in the following manner: Obtain system health impact data for wind turbine generators; According to the system health impact data of the wind turbine generator, the system health of the wind turbine generator is obtained through the pre-trained system health prediction model; Among them, the system health prediction model is built based on the long short-term memory network.

21. The wind turbine generator torque optimization system according to claim 14, characterized in that: The weight of the dynamic weighting is determined by the following formula: α(t)+β(t)=1 Among them, α(t) is the weight of the power generation reward; β(t) is the weight of the mechanical stress penalty; SOH is the system health of the wind turbine generator.

22. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the wind turbine generator torque optimization method according to any one of claims 1 to 13 are implemented.

23. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for optimizing the torque of a wind turbine generator set according to any one of claims 1 to 13 are implemented.

Images