Health prediction method, device and equipment for wind generating set

Abstract

The invention provides a health prediction method, device and equipment for a wind generating set, and the method comprises the steps: constructing a double-layer physical topological structure according to a mapping relation between a functional unit and a sensor node of the wind generating set; aggregating the sensor features corresponding to the sensor nodes to the corresponding functional unit nodes according to the mapping relation; based on the double-layer physical topological structure, fusing the sensor features aggregated to the functional unit nodes to obtain fused features; and obtaining a health prediction result of the wind generating set based on the fusion features. The problem that heterogeneous sensor data is difficult to perform collaborative modeling is solved, and comprehensive expression of cross-component and cross-modal health information is realized; the fusion feature fuses the features of different sensors, and the features of a plurality of related sensors are gathered, so that a fault signal can be amplified, the fault can be identified at the early stage of the fault, and the accuracy of a prediction result is improved.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and specifically to a method, apparatus, and equipment for predicting the health of wind turbine generator sets. Background Technology

[0002] As a typical complex electromechanical system, the operating status of wind turbine generators is affected by multiple factors, including mechanical structure, electrical system, control strategy, and external operating conditions. Existing wind turbine health monitoring systems are mostly based on SCADA data or single vibration signals, and use preset thresholds or rule models to determine the equipment status and trigger alarms.

[0003] However, such methods typically only output discrete judgments such as "normal / abnormal," which are insufficient to reflect the specific health status of the wind turbine. Furthermore, due to the complex internal structure of wind turbine generators, the variety of sensor types, and significant differences in sampling frequencies, traditional methods are inadequate in multimodal data fusion, cross-component state correlation modeling, and early fault identification, resulting in inaccurate prediction results. Summary of the Invention

[0004] In view of this, the present invention aims to provide a method, apparatus and equipment for predicting the health of wind turbine generator sets, so as to solve the problem that the existing technology has shortcomings in multimodal data fusion, cross-component state correlation modeling and early fault identification, resulting in inaccurate prediction results.

[0005] This invention provides a method for predicting the health of a wind turbine generator set, the method comprising: A two-layer physical topology is constructed based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes. Based on the mapping relationship, the sensor features corresponding to the sensor nodes are aggregated to the corresponding functional unit nodes; Based on the aforementioned two-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features; Based on the fusion features, the health prediction results of the wind turbine generator are obtained.

[0006] In one possible embodiment, constructing a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes includes: Obtain the functional units included in the wind turbine generator set, and the sensor nodes included in the functional units; The functional units are mapped as nodes in a two-layer physical topology structure; Based on the physical structure and energy transfer relationship between the functional units, the connection relationship between the nodes is constructed to form a functional unit node network; The sensor nodes contained in each functional unit are assigned to the corresponding functional unit as child nodes of the corresponding functional unit, forming the two-layer physical topology.

[0007] In one possible embodiment, the step of aggregating the sensor features corresponding to the sensor node to the corresponding functional unit node according to the mapping relationship includes: Based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, determine the sensor nodes included in the functional unit nodes; The sensor features corresponding to the identified sensor nodes are aggregated through attention and converged to the functional unit nodes to form structured features that reflect the operating state of the functional unit.

[0008] In one possible embodiment, the process of fusing sensor features aggregated to the functional unit nodes based on the two-layer physical topology to obtain fused features includes: Feature extraction is performed on each sensor feature contained in the functional unit node to obtain the feature vector of each sensor; The feature vectors are fused to obtain a fused feature vector.

[0009] In one possible embodiment, obtaining the health prediction result of the wind turbine based on the fused features includes: The fused features are mapped to the constructed low-dimensional health space through a two-layer physical topology structure to obtain the state points corresponding to the fused features. Calculate the first distance between the state point and the center of the healthy cluster, and calculate the second distance between the state point and the center of the faulty cluster; The health index is calculated based on the first distance and the second distance; Based on the health index, the health prediction results of the wind turbine generator are obtained.

[0010] In one possible embodiment, after constructing the two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensors, the method further includes: Positive and negative sample data are selected from historical operational data, and sample pairs are formed based on the positive and negative sample data; A twin network is constructed based on the health space mapping network and the aforementioned two-layer physical topology; The sample pairs are input into the two-layer physical topology to obtain the first structured feature and the second structured feature; The first structured feature and the second structured feature are input into the health space mapping network to obtain the first health space embedding vector and the second health space embedding vector. The Euclidean distance between the first health space embedding vector and the second health space embedding vector is calculated using the distance metric module of the twin network. The loss function of the twin network is calculated based on the Euclidean distance. Add a health center constraint term to the loss function to obtain the target loss function; The two-layer physical topology is trained using the target loss function to obtain the trained two-layer physical topology.

[0011] In one possible embodiment, before aggregating the sensor features corresponding to the sensor to the corresponding functional unit node according to the mapping relationship, the method further includes: The wind turbine generator sets collect monitoring data during operation through their sensors, including low-frequency SCADA operation data and high-frequency CMS vibration monitoring data. The monitoring data is time-axis aligned to obtain aligned monitoring data; The aligned monitoring data will undergo outlier identification and cleaning to obtain cleaned monitoring data. The dimensional differences in the monitoring data after cleaning are eliminated by filtering or normalization to form multi-channel time-series data segments; Based on the time series of the multi-channel time series data segments, local features are extracted from the multi-channel time series data segments to obtain local features; The local features are input into the Transformer encoder, and the Transformer encoder outputs a context-aware feature sequence of the local features. The context-aware feature sequence is aggregated into a fixed-dimensional feature vector and set as the sensor feature of the corresponding sensor.

[0012] In one possible embodiment, after obtaining the health prediction result of the wind turbine based on the fusion features, the method further includes: Record the health prediction results; Based on the health prediction results obtained from records at different times, the evolution trajectory of the wind turbine generator in the health space is formed.

[0013] In a second aspect, the present invention provides a health prediction device for wind turbine generator sets, the device comprising: The building module is used to construct a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator and the sensor nodes; The aggregation module is used to aggregate the sensor features corresponding to the sensor nodes to the corresponding functional unit nodes according to the mapping relationship. The fusion module is used to fuse the sensor features aggregated to the functional unit nodes based on the two-layer physical topology to obtain fused features; The acquisition module is used to obtain the health prediction results of the wind turbine generator set based on the fused features.

[0014] Thirdly, this application provides an electronic device, the device comprising: a memory and a processor; the memory being used to store relevant program code; the processor being used to call the program code to execute the health prediction method for wind turbine generator sets described in any of the implementations of the first aspect.

[0015] Fourthly, this application provides a computer-readable storage medium for storing a computer program for executing the health prediction method for wind turbine generators described in any implementation of the first aspect.

[0016] Fifthly, this application provides a computer program product, which includes a computer program / instruction that, when executed by a processor, implements the health prediction method for wind turbine generator sets described in any of the implementations of the first aspect.

[0017] In the above implementation of the present invention, a two-layer physical topology is constructed based on the mapping relationship between the functional units and sensors of the wind turbine generator set. This achieves structured fusion of sensor-level information to functional unit-level states, ensuring both physical rationality and good interpretability and engineering applicability. According to the mapping relationship, sensor features corresponding to the sensors are aggregated to the corresponding functional unit nodes. Based on the two-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features, and based on these fused features, the health prediction result of the wind turbine generator set is obtained. By fusing the sensor features aggregated to the functional unit nodes, the problem of collaborative modeling of heterogeneous sensor data is solved, achieving a comprehensive expression of cross-component and cross-modal health information. The fused features integrate features from different sensors; by aggregating features from multiple related sensors, fault signals can be amplified, enabling fault identification in the early stages and improving the accuracy of the prediction results. Attached Figure Description

[0018] Figure 1 A flowchart of a health prediction method for wind turbine generator sets provided in an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of a two-layer physical topology provided in an embodiment of the present invention.

[0020] Figure 3 A schematic diagram of a health prediction device for a wind turbine generator set provided in an embodiment of the present invention.

[0021] Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] One embodiment of the present invention provides a health prediction method for wind turbine generator sets. By constructing a two-layer physical topology based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, a structured fusion of sensor-level information to functional unit-level states is achieved. This two-layer physical topology ensures good interpretability and engineering applicability while maintaining physical rationality. According to the mapping relationship, sensor features corresponding to the sensor nodes are aggregated to the corresponding functional unit nodes. Based on the two-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features. Based on these fused features, the health prediction result of the wind turbine generator set is obtained. By fusing the sensor features aggregated to the functional unit nodes, the problem of difficult collaborative modeling of heterogeneous sensor data is solved, achieving a comprehensive expression of cross-component and cross-modal health information. The fused features integrate features from different sensors, and by aggregating the changing trends of multiple related sensors, fault signals can be amplified, enabling fault identification in the early stages and improving the accuracy of the prediction results.

[0024] Please see Figure 1 In one exemplary embodiment, a health prediction method for wind turbine generator sets is provided and applied to wind turbine generator sets. The method may include the following steps: S101: Construct a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes.

[0025] Specifically, the functional units of a wind turbine generator set are used to describe the key components of the wind turbine generator set and their physical functional divisions. Sensors are deployed on each component of the wind turbine generator set, i.e., on the functional units. The sensors are used to monitor parameters, and each sensor node corresponds to a specific monitoring parameter or signal channel, such as vibration, temperature, pressure, current, voltage, speed, etc.

[0026] Specifically, the sensor nodes included in a functional unit can be obtained based on the parameters it contains, thus establishing a mapping relationship between functional units and sensor nodes. Specifically, a functional unit includes at least the following nodes: pitch system, gearbox, nacelle unit, generator, converter, main shaft, hub, and output power unit. The pitch system includes parameters such as pitch angle and hub temperature; therefore, the included sensor nodes are angle sensors and temperature sensors. The gearbox includes parameters such as low-speed shaft speed, low-speed shaft torque, gear oil temperature, gearbox inlet temperature, low-speed / high-speed shaft bearing temperature, gearbox inlet pressure, and gearbox oil pump pressure; therefore, the included sensor nodes are temperature sensors, pressure sensors, and acceleration sensors. The nacelle unit includes parameters such as wind speed, wind direction, and ambient temperature; therefore, the included sensor nodes are wind speed sensors, wind direction sensors, and ambient temperature sensors. The converter includes parameters such as control load and converter temperature; therefore, the included sensor nodes are current sensors and resistance temperature detectors (RTDs). The generator includes the following parameters: generator speed, generator bearing temperature, and generator stator winding temperature. Therefore, the corresponding sensor nodes are temperature sensors and acceleration sensors. The spindle includes parameters such as vibration and temperature; therefore, the included sensor nodes are acceleration sensors, angle sensors, and temperature sensors.

[0027] The functional units, parameters, and sensor nodes included in the wind turbine generator sets described above are merely examples and do not constitute a limitation on the wind turbine generator sets.

[0028] Based on the mapping relationship between functional units and sensor nodes, the specific steps for constructing a two-layer physical topology may include: Obtain the functional units included in the wind turbine generator set, and the sensor nodes included in the functional units; The functional units are mapped as nodes in a two-layer physical topology structure; Based on the physical structure and energy transfer relationship between the functional units, the connection relationship between the nodes is constructed to form a functional unit node network; The sensor nodes contained in each functional unit are assigned to the corresponding functional unit as child nodes of the corresponding functional unit, forming the two-layer physical topology.

[0029] In the specific implementation process, the functional units included in the wind turbine generator set are acquired, such as the pitch system, gearbox, nacelle unit, generator, converter, etc., and the sensor nodes included in the functional units are also acquired. For example, the gearbox includes temperature sensors, pressure sensors, and acceleration sensors, wherein the pressure sensors may include a gearbox inlet pressure sensor and a gearbox oil pump pressure sensor. (Refer to...) Figure 2 Functional units are mapped as nodes in a two-layer physical topology. Based on the physical structure and energy transfer relationships between these functional units, connections are constructed between nodes to form a functional unit node network. Specifically, the physical structure and energy transfer relationships between functional units are as follows: nodes related to power transmission are connected sequentially along mechanical transmission paths, forming a transmission chain from the pitch system, main shaft, gearbox to the generator; that is, the pitch system functional unit is connected to the main shaft functional unit, which is also connected to the gearbox, and the gearbox functional unit is also connected to the generator. Nodes related to electrical energy conversion are connected according to the energy flow direction, forming a connection from the generator to the converter; that is, the generator functional unit is connected to the converter functional unit. Simultaneously, based on external environmental influences and condition monitoring requirements, a nacelle unit is introduced. The nacelle unit can be connected to multiple functional units, such as the gearbox and the generator functional unit, to characterize the impact of external disturbances on each functional unit. The aforementioned functional unit layer nodes and their edge relationships together constitute a static physical topology, reflecting the actual structural relationships of the wind turbine generator. Each functional unit's sensor nodes are assigned to the corresponding functional unit as its child nodes, forming the two-layer physical topology. For example, further refer to... Figure 2Angle and current sensors are assigned to the pitch system; temperature, acceleration, and angle sensors are assigned to the main shaft; temperature, pressure, and acceleration sensors are assigned to the gearbox; distance, wind speed, and wind direction sensors are assigned to the nacelle unit; current and RTD sensors are assigned to the converter; and acceleration and temperature sensors are assigned to the generator. Specifically, each sensor node establishes a cross-layer connection only with its corresponding functional unit node. For example, temperature and pressure sensor nodes within the gearbox are only connected to the gearbox functional unit node; current, voltage, and vibration sensor nodes related to the generator are only connected to the generator functional unit node; and vibration or load sensor nodes related to the hub are only connected to the hub functional unit node. This achieves node connectivity between the functional unit layer and the sensor layer through cross-layer mapping. This cross-layer mapping relationship is predefined based on the actual installation relationship and monitoring configuration of the wind turbine and remains fixed during model operation. The design of the above-mentioned two-layer physical topology enables the structured fusion of sensor-level information into functional unit-level states, allowing the model to have good interpretability and engineering applicability while ensuring physical rationality.

[0030] The above sensor node allocation is only an example. The specific allocation can be made according to the sensor nodes contained in the functional unit, and there are no restrictions here.

[0031] Furthermore, after constructing the two-layer physical topology, the two-layer physical topology can be trained based on the Siamese network, thereby achieving joint optimization of the mapping relationship between the parameters of the two-layer physical topology model and the health space.

[0032] Specifically, after constructing a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator and the sensors, it also includes: Positive and negative sample data are selected from historical operational data, and sample pairs are formed based on the positive and negative sample data; A twin network is constructed based on the health space mapping network and the aforementioned two-layer physical topology; The sample pairs are input into the two-layer physical topology to obtain the first structured feature and the second structured feature. The first structured feature and the second structured feature are input into the health space mapping network to obtain the first health space embedding vector and the second health space embedding vector. The Euclidean distance between the first health space embedding vector and the second health space embedding vector is calculated using the distance metric module of the twin network. The loss function of the twin network is calculated based on the Euclidean distance. Add a health center constraint term to the loss function to obtain the target loss function; The two-layer physical topology is trained using the target loss function to obtain the trained two-layer physical topology.

[0033] In the specific implementation process, historical operating data of wind turbine generators is acquired, and positive and negative sample data are obtained from this data. Positive sample data represents long-term stable operation data, while negative sample data represents fault data. Sample pairs are constructed based on these positive and negative sample data. These sample pairs can include similar pairs and dissimilar pairs. Similar pairs consist of two samples from the same sample set, such as both samples coming from a positive sample dataset or both coming from a negative sample dataset. Dissimilar pairs consist of two samples from different sample sets, such as one sample from a positive dataset and the other from a negative sample dataset.

[0034] A Siamese network is constructed, comprising a health spatial mapping network and the two-layer physical topology, wherein the health spatial mapping network and the two-layer physical topology constitute the shared feature extractor G_θ(⋅) of the Siamese network. Sample pairs are input to the shared feature extractor, and the sample pairs are processed through the two-layer physical topology of the shared feature extractor to obtain a first structured feature and a second structured feature. The first and second structured features are input to the health spatial mapping network to obtain a first health spatial embedding vector and a second health spatial embedding vector. The Euclidean distance between the first and second health spatial embedding vectors is calculated using the distance metric module of the Siamese network.

[0035] The specific calculation formula is as follows: ,in. Euclidean distance. Embed vectors in the first health space. Embed vectors for the second health space.

[0036] The loss function of the Siamese network is calculated based on the obtained Euclidean distance. The specific calculation formula is as follows: Among them, X i and X j For sample data, Y ij For X i and X j The sample pair Y ijThe samples can be either similar or dissimilar; that is, Xi and Xj can both be positive or negative samples, or Xi can be positive and Xj can be negative. α is a boundary parameter. For similar samples, the loss is the square of the distance, encouraging the distance to approach 0. For dissimilar samples, the loss is only applied when the distance D... ij Loss will only occur when the distance is less than α, and a distance greater than α is encouraged.

[0037] In the specific implementation process, the cosine distance between the first health space embedding vector and the second health space embedding vector can also be calculated, and the loss function of the Siamese network can be calculated through the cosine distance.

[0038] To more explicitly cluster healthy samples, a health center constraint term can be added to the loss function. This health center constraint term can be calculated using healthy samples, i.e., positive sample data. The specific calculation formula is as follows: Where Xi represents positive sample data. The center of the healthy cluster is defined. A health center constraint term is added to the loss function, i.e., the loss function is added to the health center constraint term to obtain the target loss function. The two-layer physical topology is trained based on the calculated target loss function. In specific implementation, a shared feature extractor can be used to train the two-layer physical topology. Specifically, the shared feature extractor G is initialized. θ The parameters θ are shared parameters between the health space mapping network and the two-layer physical topology. A boundary parameter α is set, for example, 1.0 can be set empirically or through cross-validation. An optimizer is selected, such as Adam. The gradient of the loss function with respect to the shared parameters θ is calculated. Using the calculated gradient, the optimizer is used to update the parameters θ, thereby driving functional units in the same health state to embed closer together, and functional units in different fault states to embed further apart, achieving joint optimization of the relationship between the two-layer physical topology model parameters and the health space mapping.

[0039] S102: Based on the mapping relationship, the sensor features corresponding to the sensor nodes are aggregated to the corresponding functional unit nodes.

[0040] In this embodiment, based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, the sensor nodes included in the functional unit nodes are determined, and the sensor features corresponding to the sensor nodes are aggregated to the corresponding functional unit nodes. The sensor features corresponding to the determined sensor nodes are then aggregated to the functional units. For example, assuming the pitch system includes angle sensors and temperature sensors as sensor nodes, the sensor features collected by the angle sensors and temperature sensors are aggregated to the pitch system.

[0041] Specifically, the step of aggregating the sensor features corresponding to the sensor node to the corresponding functional unit node according to the mapping relationship includes: Based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, determine the sensor nodes included in the functional unit nodes; The sensor features corresponding to the identified sensor nodes are aggregated through attention and converged to the functional unit nodes, forming a structured feature reflecting the operational state of the functional unit. That is, the functional unit is treated as a graph node, and the sensor features are its attributes or associated edge information. Through a message passing mechanism, the functional unit node aggregates information from its associated sensors.

[0042] In the specific implementation process, based on the mapping relationship between the functional units and sensors of the wind turbine generator set, the sensor nodes included in the functional unit nodes are determined. The sensor features corresponding to the determined sensor nodes are then aggregated through attention aggregation to the functional unit nodes, forming a structured feature reflecting the operating state of the functional unit. For example, if the pitch system includes angle sensors and temperature sensors, the sensor features collected by the angle and temperature sensors are aggregated through attention aggregation to the pitch system. By aggregating the features of multiple sensor nodes under the same functional unit through cross-layer mapping relationships, a comprehensive state representation of that functional unit is formed. This achieves a structured fusion of sensor-level information to functional unit-level state, enabling the model to possess good interpretability and engineering applicability while ensuring physical rationality.

[0043] Furthermore, before aggregating sensor features, the data acquired by the sensors can be further processed to improve the reliability of the sensor data. Specifically, before aggregating the sensor features corresponding to the sensor to the corresponding functional unit node according to the mapping relationship, the following steps are also included: The wind turbine generator sets collect monitoring data during operation through their sensors, including low-frequency SCADA operation data and high-frequency CMS vibration monitoring data. The monitoring data is time-axis aligned to obtain aligned monitoring data; The aligned monitoring data will undergo outlier identification and cleaning to obtain cleaned monitoring data. The dimensional differences in the monitoring data after cleaning are eliminated by filtering or normalization to form multi-channel time-series data segments; Based on the time series of the multi-channel time series data segments, local features are extracted from the multi-channel time series data segments to obtain local features; The local features are input into the Transformer encoder, and the Transformer encoder outputs a context-aware feature sequence of the local features. The context-aware feature sequence is aggregated into a fixed-dimensional feature vector and set as the sensor feature of the corresponding sensor.

[0044] In the specific implementation process, monitoring data during the operation of the wind turbine generator is collected through sensors. This monitoring data includes low-frequency SCADA operation data and high-frequency CMS vibration monitoring data. The monitoring data is then time-axis aligned to achieve time-axis uniformity, resulting in aligned monitoring data. Outlier identification and cleaning processing is performed on the aligned monitoring data to obtain cleaned monitoring data. Dimensional differences in the cleaned monitoring data are eliminated using filtering or normalization methods, thereby transforming time-series data collected by different sensors with different physical units and numerical ranges into multi-channel time-series data segments with uniform numerical ranges and regular dimensions. Based on the time series of these multi-channel time-series data segments, local features are extracted to obtain local features. These local features are input into a Transformer encoder, which outputs a context-aware feature sequence of the local features. This context-aware feature sequence is aggregated into a fixed-dimensional feature vector and set as the sensor feature of the corresponding sensor.

[0045] S103: Based on the dual-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features.

[0046] Specifically, through the aforementioned two-layer physical topology, feature extraction is performed on the features of each sensor contained in the functional unit node, such as time-domain statistical features, frequency-domain features, wavelet features, etc., to obtain the feature vector of each sensor. After obtaining the feature vector of each sensor, a consistency check can be performed on the feature vector to improve the reliability of the sensor feature vector. The feature vectors of the sensors that have passed the consistency check are fused to obtain a fused feature vector. Specifically, the feature vectors of multiple sensors can be fused using a weighted average or summation method, where the weight of each sensor node can be allocated according to the importance of the sensor to the health of the functional unit. The fusion process makes full use of the two-layer physical topology to fuse multiple related, possibly weak, sensor signals into a stronger indicator signal that represents the overall functional status, thereby improving the signal-to-noise ratio and sensitivity to early, minor faults.

[0047] S104: Based on the fusion features, obtain the health prediction results of the wind turbine generator set.

[0048] The fused features are processed using a two-layer physical topology structure to obtain the health prediction results of the wind turbine generator set.

[0049] Specifically, the steps for obtaining the health prediction results of the wind turbine generator based on the fusion features include: The fused features are mapped to the constructed low-dimensional health space through a two-layer physical topology structure to obtain the state points corresponding to the fused features. Calculate the first distance between the state point and the center of the healthy cluster, and calculate the second distance between the state point and the center of the faulty cluster; The health index is calculated based on the first distance and the second distance; Based on the health index, the health prediction results of the wind turbine generator are obtained.

[0050] In the specific implementation process, the fused features are mapped onto the constructed low-dimensional health space through a two-layer physical topology structure to obtain the state points corresponding to the fused features. The first distance between the state point and the center of the health cluster is calculated, where the calculation formula is: Where H is the set of all healthy clusters, It is the center of the h-th healthy cluster.

[0051] The second distance between the state point and the fault cluster is further calculated using the following formula: Among them, c f It is the center of the fault cluster.

[0052] The health index HI is calculated based on the first distance between the state point and the center of the healthy cluster, and the second distance to the faulty cluster. The specific calculation formula is as follows: .

[0053] The health prediction result of the wind turbine is determined based on the first distance between the status point and the center of the healthy cluster, the second distance to the faulty cluster, and the health index. Specifically, if the first distance is approximately 0 and the health index is approximately 1, it means the status point is close to the center of the healthy cluster, and the health prediction result of the wind turbine is completely healthy. If the status point is far from the center of the healthy cluster, i.e., the first distance is greater than 0 (e.g., 0.5 or 1), and the health index is less than the first threshold but greater than the second threshold (e.g., less than 0.7 but greater than 0.5), the health prediction result of the wind turbine is in a warning state. If the status point is close to the fault center point, for example, the second distance is approximately 1, and the health index is less than the second threshold, the health prediction result of the wind turbine is that a fault exists. This achieves the health status assessment of the wind turbine.

[0054] Since the final fusion feature input is based on functional units, when a wind turbine generator malfunctions, the potentially faulty functional unit can be quickly located by intuitively comparing the deviation of the output of each functional unit from the center of the healthy cluster, thereby improving the fault location capability.

[0055] Furthermore, after obtaining the health prediction results of the wind turbine generator set, these results can be recorded. That is, after obtaining the health prediction results of the wind turbine generator set based on the fusion features, the following steps are also included: Record the health prediction results; Based on the health prediction results obtained from records at different times, the evolution trajectory of the wind turbine generator in the health space is formed.

[0056] The health prediction results are recorded, and based on the health prediction results recorded at different times, an evolution trajectory in the health space of the wind turbine generator is formed, thereby realizing early fault warning and remaining life prediction, and providing a basis for maintenance decisions. This can better support the optimization of long-term operation and maintenance decisions for wind turbine generators.

[0057] Based on the method provided in the above embodiments, a two-layer physical topology is constructed according to the mapping relationship between the functional units and sensors of the wind turbine generator set. This achieves structured fusion of sensor-level information to functional unit-level states, ensuring both physical rationality and good interpretability and engineering applicability. According to the mapping relationship, sensor features corresponding to the sensors are aggregated to the corresponding functional unit nodes. Based on the two-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features, and based on these fused features, the health prediction result of the wind turbine generator set is obtained. By fusing the sensor features aggregated to the functional unit nodes, the problem of collaborative modeling of heterogeneous sensor data is solved, achieving a comprehensive expression of cross-component and cross-modal health information. The fused features integrate features from different sensors; by aggregating features from multiple related sensors, fault signals can be amplified, enabling fault identification in the early stages and improving the accuracy of the prediction results.

[0058] Based on the above method embodiments, this invention also provides a health prediction device for wind turbine generator sets. See also... Figure 3 The diagram shown is a schematic of a health prediction device for a wind turbine generator set provided in an embodiment of the present invention.

[0059] The device 300 includes: Module 301 is used to construct a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes. The aggregation module 302 is used to aggregate the sensor features corresponding to the sensor nodes to the corresponding functional unit nodes according to the mapping relationship; The fusion module 303 is used to fuse the sensor features aggregated to the functional unit nodes based on the two-layer physical topology to obtain fused features; The module 304 is used to obtain the health prediction result of the wind turbine generator set based on the fused features.

[0060] In one possible implementation, the functional units included in the wind turbine generator set and the sensor nodes included in the functional units are obtained; The functional units are mapped as nodes in a two-layer physical topology structure; Based on the physical structure and energy transfer relationship between the functional units, the connection relationship between the nodes is constructed to form a functional unit node network; The sensor nodes contained in each functional unit are assigned to the corresponding functional unit as child nodes of the corresponding functional unit, forming the two-layer physical topology.

[0061] In one possible implementation, the step of aggregating the sensor features corresponding to the sensor node to the corresponding functional unit node according to the mapping relationship includes: Based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, determine the sensor nodes included in the functional unit nodes; The sensor features corresponding to the identified sensor nodes are aggregated through attention and converged to the functional unit nodes to form structured features that reflect the operating state of the functional unit.

[0062] In one possible implementation, the process of fusing sensor features aggregated to the functional unit nodes based on the two-layer physical topology to obtain fused features includes: Feature extraction is performed on each sensor feature contained in the functional unit node to obtain the feature vector of each sensor; The feature vectors are fused to obtain a fused feature vector.

[0063] In one possible implementation, obtaining the health prediction result of the wind turbine based on the fused features includes: The fused features are mapped to the constructed low-dimensional health space through a two-layer physical topology structure to obtain the state points corresponding to the fused features. Calculate the first distance between the state point and the center of the healthy cluster, and calculate the second distance between the state point and the center of the faulty cluster; The health index is calculated based on the first distance and the second distance; Based on the health index, the health prediction results of the wind turbine generator are obtained.

[0064] In one possible implementation, after constructing the two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator and the sensors, the method further includes: Positive and negative sample data are selected from historical operational data, and sample pairs are formed based on the positive and negative sample data; A twin network is constructed based on the health space mapping network and the aforementioned two-layer physical topology; The sample pairs are input into the two-layer physical topology to obtain the first structured feature and the second structured feature; The first structured feature and the second structured feature are input into the health space mapping network to obtain the first health space embedding vector and the second health space embedding vector. The Euclidean distance between the first health space embedding vector and the second health space embedding vector is calculated using the distance metric module of the twin network. The loss function of the twin network is calculated based on the Euclidean distance. Add a health center constraint term to the loss function to obtain the target loss function; The two-layer physical topology is trained using the target loss function to obtain the trained two-layer physical topology.

[0065] In one possible implementation, before aggregating the sensor features corresponding to the sensor to the corresponding functional unit node according to the mapping relationship, the method further includes: The wind turbine generator sets collect monitoring data during operation through their sensors, including low-frequency SCADA operation data and high-frequency CMS vibration monitoring data. The monitoring data is time-axis aligned to obtain aligned monitoring data; The aligned monitoring data will undergo outlier identification and cleaning to obtain cleaned monitoring data. The dimensional differences in the monitoring data after cleaning are eliminated by filtering or normalization to form multi-channel time-series data segments; Based on the time series of the multi-channel time series data segments, local features are extracted from the multi-channel time series data segments to obtain local features; The local features are input into the Transformer encoder, and the Transformer encoder outputs a context-aware feature sequence of the local features. The context-aware feature sequence is aggregated into a fixed-dimensional feature vector and set as the sensor feature of the corresponding sensor.

[0066] In one possible implementation, after obtaining the health prediction result of the wind turbine based on the fusion features, the method further includes: Record the health prediction results; Based on the health prediction results obtained from records at different times, the evolution trajectory of the wind turbine generator in the health space is formed.

[0067] See Figure 4 , Figure 4 This is a schematic diagram of an electronic device provided in an embodiment of this application.

[0068] The device 400 includes a memory 401 and a processor 402; the memory 401 is used to store relevant program code; the processor 402 is used to call the program code to execute the health prediction method for wind turbine generator sets described in the above method embodiments.

[0069] Furthermore, embodiments of the present invention also provide a computer-readable storage medium for storing a computer program for executing the health prediction method for wind turbine generator sets described in the above method embodiments.

[0070] This invention also provides a computer program product, which includes a computer program / instruction. When the computer program / instruction is executed by a processor, it implements the health prediction method for wind turbine generator sets described in the above method embodiments.

[0071] It should be noted that the computer-readable medium described above in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0072] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0073] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. In particular, for system or device embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The device embodiments described above are merely illustrative. The units or modules described as separate components may or may not be physically separate. The components shown as units or modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the units or modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0074] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions, and operations that may be implemented according to various embodiments of the invention, including methods, apparatus, and devices. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0075] It should be understood that in this invention, "at least one (item)" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0076] It should also be noted that, in this invention, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0077] The steps of the methods or algorithms described in conjunction with the embodiments disclosed in this invention can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the health of a wind turbine generator set, characterized in that, The method includes: A two-layer physical topology is constructed based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes. Based on the mapping relationship, the sensor features corresponding to the sensor nodes are aggregated to the corresponding functional unit nodes; Based on the aforementioned two-layer physical topology, the sensor features aggregated to the functional unit nodes are fused to obtain fused features; Based on the fusion features, the health prediction results of the wind turbine generator are obtained.

2. The method according to claim 1, characterized in that, The construction of a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensor nodes includes: Obtain the functional units included in the wind turbine generator set, and the sensor nodes included in the functional units; The functional units are mapped as nodes in a two-layer physical topology structure; Based on the physical structure and energy transfer relationship between the functional units, the connection relationship between the nodes is constructed to form a functional unit node network; The sensor nodes contained in each functional unit are assigned to the corresponding functional unit as child nodes of the corresponding functional unit, forming the two-layer physical topology.

3. The method according to claim 1, characterized in that, The step of aggregating the sensor features corresponding to the sensor nodes to the corresponding functional unit nodes according to the mapping relationship includes: Based on the mapping relationship between the functional units and sensor nodes of the wind turbine generator set, determine the sensor nodes included in the functional unit nodes; The sensor features corresponding to the identified sensor nodes are aggregated through attention and converged to the functional unit nodes to form structured features that reflect the operating state of the functional unit.

4. The method according to claim 1, characterized in that, The process of fusing sensor features aggregated to the functional unit nodes based on the dual-layer physical topology to obtain fused features includes: Feature extraction is performed on each sensor feature contained in the functional unit node to obtain the feature vector of each sensor; The feature vectors are fused to obtain a fused feature vector.

5. The method according to claim 1, characterized in that, The process of obtaining the health prediction result of the wind turbine based on the fused features includes: The fused features are mapped to the constructed low-dimensional health space through a two-layer physical topology structure to obtain the state points corresponding to the fused features. Calculate the first distance between the state point and the center of the healthy cluster, and calculate the second distance between the state point and the center of the faulty cluster; The health index is calculated based on the first distance and the second distance; Based on the health index, the health prediction results of the wind turbine generator are obtained.

6. The method according to claim 1, characterized in that, After constructing a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator set and the sensors, the method further includes: Positive and negative sample data are selected from historical operational data, and sample pairs are formed based on the positive and negative sample data; A twin network is constructed based on the health space mapping network and the aforementioned two-layer physical topology; The sample pairs are input into the two-layer physical topology to obtain the first structured feature and the second structured feature; The first structured feature and the second structured feature are input into the health space mapping network to obtain the first health space embedding vector and the second health space embedding vector. The Euclidean distance between the first health space embedding vector and the second health space embedding vector is calculated using the distance metric module of the twin network. The loss function of the twin network is calculated based on the Euclidean distance. Add a health center constraint term to the loss function to obtain the target loss function; The two-layer physical topology is trained using the target loss function to obtain the trained two-layer physical topology.

7. The method according to claim 1, characterized in that, Before aggregating the sensor features corresponding to the sensor to the corresponding functional unit node according to the mapping relationship, the method further includes: The wind turbine generator sets collect monitoring data during operation through their sensors, including low-frequency SCADA operation data and high-frequency CMS vibration monitoring data. The monitoring data is time-axis aligned to obtain aligned monitoring data; The aligned monitoring data will undergo outlier identification and cleaning to obtain cleaned monitoring data. The dimensional differences in the monitoring data after cleaning are eliminated by filtering or normalization to form multi-channel time-series data segments; Based on the time series of the multi-channel time series data segments, local features are extracted from the multi-channel time series data segments to obtain local features; The local features are input into the Transformer encoder, and the Transformer encoder outputs a context-aware feature sequence of the local features. The context-aware feature sequence is aggregated into a fixed-dimensional feature vector and set as the sensor feature of the corresponding sensor.

8. The method according to claim 1, characterized in that, After obtaining the health prediction result of the wind turbine based on the fusion features, the method further includes: The health prediction results are recorded; Based on the health prediction results obtained from records at different times, the evolution trajectory of the wind turbine generator in the health space is formed.

9. A health prediction device for wind turbine generator sets, characterized in that, The device includes: The building module is used to construct a two-layer physical topology based on the mapping relationship between the functional units of the wind turbine generator and the sensor nodes; The aggregation module is used to aggregate the sensor features corresponding to the sensor nodes to the corresponding functional unit nodes according to the mapping relationship. The fusion module is used to fuse the sensor features aggregated to the functional unit nodes based on the two-layer physical topology to obtain fused features; The acquisition module is used to obtain the health prediction results of the wind turbine generator set based on the fused features.

10. An electronic device, characterized in that, The device includes: a memory and a processor; the memory is used to store relevant program code; the processor is used to call the program code to execute the health prediction method for wind turbine generator sets according to any one of claims 1 to 8.

Images