A wind turbine remaining life prediction method and system

Abstract

The application belongs to the field of wind power generation, and discloses a wind turbine remaining life prediction method and system. The method collects multi-source state data and synchronizes time stamps, inputs a pre-built wind turbine digital twin model, and maps a real-time physical field distribution matrix of key components through multi-physical field coupling calculation. Taking this as the initial boundary, a time sequence iteration solution is obtained based on the component damage evolution equation, and a virtual fault simulation time sequence sample is generated by reconstructing and labeling according to time. The sample is spliced and normalized with real running samples labeled as normal to construct a joint training set. A long short-term memory network model is trained using the training set to generate a health state evaluation model. Finally, real-time state data is input into the model to output the predicted remaining life. The application can solve the technical problem of insufficient prediction accuracy of existing wind turbine life prediction models due to the lack of real fault samples.

Description

Technical Field

[0001] This application belongs to the field of wind power generation, specifically relating to a method and system for predicting the remaining life of wind turbine generators. Background Technology

[0002] In the field of wind power operation and maintenance, health monitoring and lifespan prediction of wind turbine units are crucial for ensuring safe equipment operation. Currently, wind farms generally rely on SCADA systems and sensors deployed on the drivetrain or structural components to collect multi-dimensional operational data. Based on this historical data and machine learning algorithms, they construct condition assessment models to predict equipment lifespan. In operation and maintenance scheduling, the industry largely relies on these basic prediction results to execute fixed, periodic maintenance plans. Furthermore, some intelligent wind power projects have gradually introduced digital twin technology. Its conventional application involves establishing a three-dimensional digital model of the physical unit and achieving remote, visual monitoring of the equipment's operating status through one-way access to measured data.

[0003] The current problem is that the aforementioned lifespan prediction and maintenance methods have significant limitations when dealing with special geographical and climatic environments such as high altitudes and frigid zones. The accuracy of existing data-driven prediction models heavily relies on training with massive amounts of real-world fault samples containing complete degradation cycles. However, in actual engineering projects, the probability of severe failures in core components of wind turbines (such as main bearings and pitch motors) is extremely low. Furthermore, the impact of extremely low temperatures on material stiffness and lubrication characteristics is nonlinear and abrupt, resulting in a severe lack of real-world fault evolution data characteristic of high-altitude environments. This lack of fault samples directly leads to weak generalization ability and insufficient assessment accuracy in prediction models. Simultaneously, existing digital twin technologies are typically limited to surface-level state mapping and data display, failing to deeply integrate and dynamically extrapolate with the underlying physical degradation mechanisms such as mechanical wear and fatigue damage. Summary of the Invention

[0004] This application addresses the technical problem of insufficient prediction accuracy in existing wind turbine life prediction models due to a lack of real fault samples, and provides a method and system for predicting the remaining life of wind turbines.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for predicting the remaining life of a wind turbine includes: Collect multi-source status data of wind turbine units under the target operating environment; After the multi-source state data is time-stamped and synchronized, it is used as a boundary condition input into the pre-constructed digital twin model of the wind turbine. The multi-physics field coupling solution maps the multi-source state data into the real-time physical field distribution matrix of the key components of the wind turbine. In the digital twin model, the real-time physical field distribution matrix is ​​used as the initial health state boundary, and the physical field feature matrix of the key component in multiple progressive degradation stages is obtained by time-series iterative solution based on the preset component damage evolution equation. The physical field feature matrices of the key components in multiple progressive degradation stages are spliced ​​and reconstructed along the time dimension, and corresponding fault categories and degradation degree labels are assigned to generate virtual fault simulation time series samples. The multi-source state data is labeled as normal operation samples, and then combined and normalized with the virtual fault simulation time series samples in the time series dimension to obtain a joint training set. The pre-constructed long short-term memory network model is supervised and trained using the joint training set. The network weights are updated with the goal of minimizing the prediction error until the loss function converges, thereby generating a health status assessment model. The real-time multi-source status data of the wind turbine is input into the health status assessment model, and the predicted remaining life of the key components is output.

[0006] In some implementations, the multi-source state data includes SCADA system data of the wind turbine, vibration data of the mechanical transmission chain, strain data of key load-bearing components, and temperature data of key parts.

[0007] In some embodiments, the real-time physical field distribution matrix includes a three-dimensional temperature field distribution matrix and a three-dimensional stress field distribution matrix; The process of mapping the multi-source state data into a real-time physical field distribution matrix of the key components of the wind turbine through multi-physics coupling solution includes: Based on the multi-source state data synchronized with timestamps, finite element thermo-mechanical coupling calculations are performed in the digital twin model of the wind turbine to obtain the real-time stress deformation and heat conduction state of key components. Based on the real-time stress deformation and heat conduction state of the key components, the geometric nodes of the digital twin model of the wind turbine are driven to generate updated three-dimensional temperature field distribution matrix and three-dimensional stress field distribution matrix, which serve as real-time physical field distribution matrix.

[0008] In some embodiments, the preset component damage evolution equation includes a fatigue damage evolution equation and a mechanical wear evolution equation; The fault categories include main bearing wear faults and pitch motor jamming faults. The time-series iterative solution based on the preset component damage evolution equation includes: Extract local temperature data from the three-dimensional temperature field distribution matrix and local stress data from the three-dimensional stress field distribution matrix; Based on the local temperature data, the material elastic modulus and fatigue limit at the corresponding temperature are obtained, and combined with the local stress data, they are substituted into the fatigue damage evolution equation for iterative calculation to obtain the cumulative damage of the key component in the current degradation stage. Based on the local temperature data, the viscosity and friction coefficient of the lubricating oil at the corresponding temperature are obtained, and combined with the local stress data, they are substituted into the mechanical wear evolution equation for iterative calculation to obtain the surface wear depth of the key component at the current degradation stage. Based on the cumulative damage amount and surface wear depth corresponding to multiple degradation stages, a physical field feature matrix of the key component in multiple progressive degradation stages is constructed.

[0009] In some implementations, the step of labeling the multi-source state data as normal operation samples and performing data concatenation and normalization processing with the virtual fault simulation time-series samples in the time-series dimension to obtain a joint training set includes: Time-frequency domain features are extracted from the multi-source state data to generate a first feature vector; Spatial features are extracted from the virtual fault simulation time series samples to generate a second feature vector; The first feature vector and the second feature vector are concatenated and recombined at the same time node to obtain a high-dimensional state feature vector with a unified dimension. The high-dimensional state feature vector is subjected to redundancy reduction using principal component analysis and mapped to a preset normalization interval to generate the joint training set.

[0010] In some implementations, after inputting the current real-time multi-source status data of the wind turbine into the health status assessment model and outputting the predicted remaining life of the key components, the method further includes: Determine whether the predicted remaining lifetime is less than a preset safety threshold. If it is less than the preset safety threshold, then generate a predictive maintenance strategy.

[0011] Secondly, a wind turbine remaining life prediction system includes: The data acquisition module is used to collect multi-source status data of wind turbines under the target operating environment; The twin mapping module is used to timestamp and synchronize the multi-source state data, and then input it as a boundary condition into the pre-constructed digital twin model of the wind turbine. Through multi-physics coupling solution, the multi-source state data is mapped to the real-time physical field distribution matrix of the key components of the wind turbine. The fault simulation module is used in the digital twin model to take the real-time physical field distribution matrix as the initial health state boundary and perform time-series iterative solution based on the preset component damage evolution equation to obtain the physical field feature matrix of the key component in multiple progressive degradation stages. The sample reconstruction module is used to reconstruct the physical field feature matrix of the key component in multiple progressive degradation stages by splicing it together along the time dimension, and assign corresponding fault categories and degradation degree labels to generate virtual fault simulation time series samples. The training set construction module is used to label the multi-source state data as normal operation samples and perform data splicing and normalization processing on the time-series dimension with the virtual fault simulation time-series samples to obtain a joint training set. The model training module is used to supervise the training of the pre-built long short-term memory network model using the joint training set, update the network weights with the goal of minimizing the prediction error, until the loss function converges, and generate a health status assessment model. The life prediction module is used to input the current real-time multi-source status data of the wind turbine into the health status assessment model and output the predicted remaining life of the key components.

[0012] Thirdly, a computer device includes: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by the processor, implements the method for predicting the remaining life of a wind turbine as described above.

[0013] Fourthly, a computer-readable storage medium storing a computer program adapted to be loaded by a processor and executed as described in the wind turbine remaining life prediction method.

[0014] Fifthly, a computer program product comprising a computer program that, when executed by a processor, implements the wind turbine remaining life prediction method as described above.

[0015] Compared with the prior art, this application has the following beneficial effects: This application achieves multi-physics coupling solution by synchronizing multi-source state data with timestamps and inputting it into a digital twin model. Then, using the real-time physical field distribution matrix as the initial boundary, it performs fault injection simulation based on the component damage evolution equation to generate virtual fault simulation time series samples. Finally, it splices these samples with normal operation samples to construct a joint training set to train the long short-term memory network model. This overcomes the historical limitation of traditional data-driven models that heavily rely on real fault samples. It also makes up for the lack of negative samples of severe faults in extreme environments by using mechanism inference within the digital twin space. By integrating real health data and virtual fault samples, it improves the robustness and generalization ability of the prediction model from the underlying data structure, and achieves high-precision prediction of the remaining life of wind turbine units.

[0016] Furthermore, by performing finite element thermo-mechanical coupling calculations in the digital twin model of the wind turbine to obtain real-time stress deformation and heat conduction states, and then driving the geometric nodes to update to generate a three-dimensional temperature field distribution matrix and a three-dimensional stress field distribution matrix, the abstract one-dimensional sensing signals are visualized into an intuitive three-dimensional spatial physical field. This achieves a deep digital mapping of the thermodynamic and dynamic states of the physical unit, solving the technical problem that existing digital twin technologies only display surface data and lack the support of internal three-dimensional physical mechanisms.

[0017] Furthermore, by extracting local temperature and stress data, the material's elastic modulus, fatigue limit, lubricating oil viscosity, and friction coefficient at the corresponding temperature are dynamically obtained. These values ​​are then substituted into the fatigue damage evolution equation and the mechanical wear evolution equation to iteratively calculate the cumulative damage and surface wear depth. Without introducing artificially created correction coefficients, this method directly follows the constitutive laws of physics and accurately reproduces the nonlinear physical effects of extreme cold and low temperature environments on material embrittlement and lubrication failure. This ensures the absolute reliability and authenticity of the generated virtual samples, such as main bearing wear failure and pitch motor jamming failure, at the engineering mechanics level. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a method for predicting the remaining life of a wind turbine provided in this application embodiment; Figure 2 A structural diagram of a wind turbine remaining life prediction system provided in this application embodiment; Figure 3 A computer device diagram provided for an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] In one embodiment of this application, such as Figure 1 As shown, a method for predicting the remaining life of a wind turbine is provided, including: S1, collects multi-source status data of wind turbine units under the target operating environment; Specifically, the target operating environment in this embodiment is a high-altitude, cold wind farm environment. Multi-source status data is collected in real time through a sensor network deployed at key parts of the wind turbine and the main control system. The collected data items include: wind speed, active power, and generator speed obtained from the SCADA system; triaxial acceleration vibration data of the main bearings and gearbox in the mechanical transmission chain; strain gauge measurements at the blade roots and the bottom of the tower; and ambient temperature outside the nacelle, gearbox oil sump temperature, and generator winding temperature.

[0022] S2, after the multi-source state data is time-stamped and synchronized, it is used as a boundary condition input into the pre-constructed digital twin model of the wind turbine. The multi-physics field coupling solution maps the multi-source state data into the real-time physical field distribution matrix of the key components of the wind turbine. Specifically, using the main control system's reference clock, an interpolation algorithm is employed to perform unified timestamp synchronization processing on all multi-source state data. The wind turbine digital twin model in this embodiment is constructed based on a combination of the Unity3D 3D visualization engine, the MQTT IoT communication protocol, and the InfluxDB time-series database. The synchronized data stream is written to the InfluxDB time-series database in real time via the MQTT protocol and used as boundary load and ambient temperature boundary conditions input into the digital twin model.

[0023] The background finite element solver performs thermo-mechanical coupling calculations to obtain the stress deformation displacement and heat conduction temperature values ​​of the key components of the unit under the current actual operating conditions. Based on this, it drives the geometric nodes in the Unity3D engine to perform spatial mapping and dynamic updates, accurately outputting the three-dimensional temperature field distribution matrix and the three-dimensional stress field distribution matrix. The two together constitute the real-time physical field distribution matrix.

[0024] S3, in the digital twin model, the real-time physical field distribution matrix is ​​used as the initial health state boundary, and the time-series iterative solution is performed based on the preset component damage evolution equation to obtain the physical field feature matrix of the key component in multiple progressive degradation stages. Specifically, the acquired three-dimensional temperature field and stress field distribution matrices are set as the zero-time state of damage evolution. First, local temperature data from the three-dimensional temperature field distribution matrix and local stress data from the three-dimensional stress field distribution matrix are extracted respectively.

[0025] For the fatigue damage evolution equation, based on the extracted extremely low local temperature data and the material constitutive relation, the elastic modulus and fatigue limit of the material at the current low temperature are calculated. Then, the local stress data is substituted into the fatigue damage evolution equation based on Miner's linear cumulative damage theory for single-step iterative calculation to obtain the cumulative damage of the key component at the current degradation stage.

[0026] For the mechanical wear evolution equation, the corresponding lubricating oil viscosity and friction coefficient are calculated based on local temperature data. Combined with local stress data, the mechanical wear evolution equation based on Arcard theory is substituted into the equation for iterative calculation to obtain the surface wear depth.

[0027] By setting a series of incremental iteration time steps, the cumulative damage and surface wear depth calculated at each degradation stage are integrated into the original physical field matrix, thereby constructing multiple physical field feature matrices that show a progressive relationship between the main bearing and the pitch motor under different adverse conditions.

[0028] S4, the physical field feature matrices of the key components in multiple progressive degradation stages are spliced ​​and reconstructed according to the time dimension, and corresponding fault categories and degradation degree labels are assigned to generate virtual fault simulation time series samples; Specifically, the discrete physical field feature matrices are continuously stacked and spliced ​​together according to the time sequence of iterative calculations to reconstruct a fault evolution sequence.

[0029] For this evolution sequence, a category label is assigned to the main bearing wear failure or the pitch motor jamming failure, and a degree label of mild degradation, moderate degradation and severe degradation is assigned according to the specific numerical range of cumulative damage and wear depth, thereby forming a virtual fault simulation time series sample under high cold and low temperature conditions.

[0030] S5, the multi-source state data is labeled as normal operation samples, and the data is spliced ​​and normalized with the virtual fault simulation time series samples in the time series dimension to obtain a joint training set; Specifically, to ensure the consistency of the dimensions of the input features of the long short-term memory network model, a unified feature engineering is performed on the labeled normal operation samples and the virtual fault simulation time series samples.

[0031] The fast Fourier transform algorithm is used to perform time-frequency domain calculations on the one-dimensional multi-source state data at this time node to generate the first feature vector; Simultaneously extract the maximum stress node value and local temperature gradient of the three-dimensional physical field feature matrix at this time node to generate the second feature vector.

[0032] The first and second feature vectors at the same time point are horizontally concatenated and recombined into a unified high-dimensional state feature vector that represents the overall health status of the unit at that time.

[0033] Subsequently, principal component analysis (PCA) was used to reduce the dimensionality of the serialized high-dimensional state feature vectors to eliminate redundant variables. Then, the maximum-minimum normalization method was used to strictly map all feature values ​​to a preset value range. This ensured that the normal operation samples and the virtual fault samples were fully aligned in terms of data structure, generating a joint training set that was adapted to the model input format.

[0034] S6. The pre-constructed long short-term memory network model is supervised and trained using the joint training set. The network weights are updated with the goal of minimizing the prediction error until the loss function converges, thereby generating a health status assessment model. Specifically, a long short-term memory network model containing a forget gate, an input gate, and an output gate is selected. The joint training set is divided into a training set and a validation set, and then input into the long short-term memory network model in batches.

[0035] The model calculates the predicted degradation state using forward propagation, calculates the prediction error between the predicted value and the true label using the mean squared error function, calculates the network gradient using a time-based backpropagation algorithm, and dynamically updates the weight parameters of each gated unit within the network using an adaptive moment estimation optimizer.

[0036] During training, the generalization performance of the model is continuously monitored using the validation set. When the mean squared error curve tends to stabilize and no longer decreases within ten consecutive training cycles, the early stopping mechanism is triggered to stop training, solidify the network parameters, and output a mature health status assessment model.

[0037] S7, input the current real-time multi-source status data of the wind turbine into the health status assessment model, and output the predicted remaining life of the key components.

[0038] Specifically, during the actual operation of the unit, the multi-source status data collected in real time is input into the deployed health status assessment model. The model outputs in real time the predicted remaining lifespan days that the main bearing and pitch motor can continue to operate from the current state to the failure threshold.

[0039] The system backend continuously checks whether the predicted remaining lifespan in days is less than a preset safety threshold. If the result is less than the safety threshold, a predictive maintenance strategy is automatically triggered and generated. The predictive maintenance strategy specifically includes: generating a maintenance scheduling work order containing the suggested maintenance date and pushing it to the smart terminal of the on-site maintenance personnel; and simultaneously outputting the remaining lifespan of the equipment to the backend asset management system, thereby realizing the transformation of wind turbine maintenance from scheduled maintenance to on-demand maintenance.

[0040] In one embodiment of this application, such as Figure 2 As shown, a wind turbine remaining life prediction system is provided, comprising: The data acquisition module is used to collect multi-source status data of wind turbines under the target operating environment; The twin mapping module is used to timestamp and synchronize the multi-source state data, and then input it as a boundary condition into the pre-constructed digital twin model of the wind turbine. Through multi-physics coupling solution, the multi-source state data is mapped to the real-time physical field distribution matrix of the key components of the wind turbine. The fault simulation module is used in the digital twin model to take the real-time physical field distribution matrix as the initial health state boundary and perform time-series iterative solution based on the preset component damage evolution equation to obtain the physical field feature matrix of the key component in multiple progressive degradation stages. The sample reconstruction module is used to reconstruct the physical field feature matrix of the key component in multiple progressive degradation stages by splicing it together along the time dimension, and assign corresponding fault categories and degradation degree labels to generate virtual fault simulation time series samples. The training set construction module is used to label the multi-source state data as normal operation samples and perform data splicing and normalization processing on the time-series dimension with the virtual fault simulation time-series samples to obtain a joint training set. The model training module is used to supervise the training of the pre-built long short-term memory network model using the joint training set, update the network weights with the goal of minimizing the prediction error, until the loss function converges, and generate a health status assessment model. The life prediction module is used to input the current real-time multi-source status data of the wind turbine into the health status assessment model and output the predicted remaining life of the key components.

[0041] For specific limitations regarding the wind turbine remaining life prediction system, please refer to the limitations of the wind turbine remaining life prediction method described above. The corresponding technical effects are equivalent and will not be repeated here. Each module in the aforementioned wind turbine remaining life prediction system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0042] Figure 3 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 3 As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting the remaining life of a wind turbine. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0043] As will be understood by those skilled in the art, computer equipment Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0044] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0045] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0046] In summary, the wind turbine remaining life prediction method, system, computer equipment, and storage medium provided in this application embodiment effectively solve the technical problem of insufficient accuracy of life prediction model due to the scarcity of severe fault samples in actual engineering by establishing a digital twin model for multi-physics coupled calculation and introducing the component damage evolution equation based on physical and mechanical laws into the generation process of virtual fault samples. At the same time, this application performs strict alignment and dimensionality reduction stitching of one-dimensional multi-source state signals and three-dimensional physical field matrices, realizing the structural consistency and deep fusion of multimodal heterogeneous data at the bottom layer of the long short-term memory network, thereby improving the generalization performance of the algorithm.

[0047] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0048] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A method for predicting the remaining life of a wind turbine generator, characterized in that, include: Collect multi-source status data of wind turbines under the target operating environment; After the multi-source state data is time-stamped and synchronized, it is used as a boundary condition input into the pre-constructed digital twin model of the wind turbine. The multi-physics field coupling solution maps the multi-source state data into the real-time physical field distribution matrix of the key components of the wind turbine. In the digital twin model, the real-time physical field distribution matrix is ​​used as the initial health state boundary, and the physical field feature matrix of the key component in multiple progressive degradation stages is obtained by time-series iterative solution based on the preset component damage evolution equation. The physical field feature matrices of the key components in multiple progressive degradation stages are spliced ​​and reconstructed along the time dimension, and corresponding fault categories and degradation degree labels are assigned to generate virtual fault simulation time series samples. The multi-source state data is labeled as normal operation samples, and then combined and normalized with the virtual fault simulation time series samples in the time series dimension to obtain a joint training set. The pre-constructed long short-term memory network model is supervised and trained using the joint training set. The network weights are updated with the goal of minimizing the prediction error until the loss function converges, thereby generating a health status assessment model. The real-time multi-source status data of the wind turbine is input into the health status assessment model, and the predicted remaining life of the key components is output.

2. The method for predicting the remaining life of a wind turbine generator according to claim 1, characterized in that, The multi-source status data includes SCADA system data of the wind turbine, vibration data of the mechanical transmission chain, strain data of key load-bearing components, and temperature data of key parts.

3. The method for predicting the remaining life of a wind turbine according to claim 1, characterized in that, The real-time physical field distribution matrix includes a three-dimensional temperature field distribution matrix and a three-dimensional stress field distribution matrix; The process of mapping the multi-source state data into a real-time physical field distribution matrix of the key components of the wind turbine through multi-physics coupling solution includes: Based on the multi-source state data synchronized with timestamps, finite element thermo-mechanical coupling calculations are performed in the digital twin model of the wind turbine to obtain the real-time stress deformation and heat conduction state of key components. Based on the real-time stress deformation and heat conduction state of the key components, the geometric nodes of the digital twin model of the wind turbine are driven to generate updated three-dimensional temperature field distribution matrix and three-dimensional stress field distribution matrix, which serve as real-time physical field distribution matrix.

4. The method for predicting the remaining life of a wind turbine according to claim 3, characterized in that, The preset component damage evolution equations include fatigue damage evolution equations and mechanical wear evolution equations. The fault categories include main bearing wear faults and pitch motor jamming faults. The time-series iterative solution based on the preset component damage evolution equation includes: Extract local temperature data from the three-dimensional temperature field distribution matrix and local stress data from the three-dimensional stress field distribution matrix; Based on the local temperature data, the material elastic modulus and fatigue limit at the corresponding temperature are obtained, and combined with the local stress data, they are substituted into the fatigue damage evolution equation for iterative calculation to obtain the cumulative damage of the key component in the current degradation stage. Based on the local temperature data, the viscosity and friction coefficient of the lubricating oil at the corresponding temperature are obtained, and combined with the local stress data, they are substituted into the mechanical wear evolution equation for iterative calculation to obtain the surface wear depth of the key component at the current degradation stage. Based on the cumulative damage amount and surface wear depth corresponding to multiple degradation stages, a physical field feature matrix of the key component in multiple progressive degradation stages is constructed.

5. The method for predicting the remaining life of a wind turbine generator according to claim 1, characterized in that, The process involves labeling the multi-source state data as normal operation samples and concatenating and normalizing them with the virtual fault simulation time-series samples along the time-series dimension to obtain a joint training set, including: Time-frequency domain features are extracted from the multi-source state data to generate a first feature vector; Spatial features are extracted from the virtual fault simulation time series samples to generate a second feature vector; The first feature vector and the second feature vector are concatenated and recombined to obtain a high-dimensional state feature vector of the same dimension; The high-dimensional state feature vector is subjected to redundancy reduction using principal component analysis and mapped to a preset normalization interval to generate the joint training set.

6. The method for predicting the remaining life of a wind turbine according to claim 1, characterized in that, After inputting the current real-time multi-source status data of the wind turbine into the health status assessment model and outputting the predicted remaining life of the key components, the method further includes: Determine whether the predicted remaining lifetime is less than a preset safety threshold. If it is less than the preset safety threshold, then generate a predictive maintenance strategy.

7. A wind turbine remaining life prediction system, characterized in that, include: The data acquisition module is used to collect multi-source status data of wind turbines under the target operating environment; The twin mapping module is used to timestamp and synchronize the multi-source state data, and then input it as a boundary condition into the pre-constructed digital twin model of the wind turbine. Through multi-physics coupling solution, the multi-source state data is mapped to the real-time physical field distribution matrix of the key components of the wind turbine. The fault simulation module is used in the digital twin model to take the real-time physical field distribution matrix as the initial health state boundary and perform time-series iterative solution based on the preset component damage evolution equation to obtain the physical field feature matrix of the key component in multiple progressive degradation stages. The sample reconstruction module is used to reconstruct the physical field feature matrix of the key component in multiple progressive degradation stages by splicing it together along the time dimension, and assign corresponding fault categories and degradation degree labels to generate virtual fault simulation time series samples. The training set construction module is used to label the multi-source state data as normal operation samples and perform data splicing and normalization processing on the time-series dimension with the virtual fault simulation time-series samples to obtain a joint training set. The model training module is used to supervise the training of the pre-built long short-term memory network model using the joint training set, update the network weights with the goal of minimizing the prediction error, until the loss function converges, and generate a health status assessment model. The life prediction module is used to input the current real-time multi-source status data of the wind turbine into the health status assessment model and output the predicted remaining life of the key components.

8. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements a method for predicting the remaining life of a wind turbine as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements a method for predicting the remaining life of a wind turbine as described in any one of claims 1 to 6.

Images