Cloud-edge collaborative self-supervised predictive control system and method for thermal management of wind turbine units

Abstract

This invention discloses a cloud-edge collaborative self-supervised predictive control system and method for wind turbine thermal management. The system includes a data acquisition module, a data preprocessing module, a self-supervised detection module, and an execution module located at the edge layer, all interconnected via communication. A cloud-edge collaborative interaction module is located at the transmission layer, communicating with the execution module at the edge layer. A thermal fault diagnosis unit and a global thermal management optimization unit are located at the cloud layer; the thermal fault diagnosis unit communicates with the cloud-edge collaborative interaction module, and the global thermal management optimization unit communicates with the thermal fault diagnosis unit. This invention, employing the aforementioned cloud-edge collaborative self-supervised predictive control system and method for wind turbine thermal management, reduces downtime caused by thermal management failures and effectively lowers power generation losses and operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a cloud-edge collaborative self-monitoring predictive control system and method for thermal management of wind turbine generators. Background Technology

[0002] Wind energy is a clean and pollution-free renewable energy source with enormous reserves. Wind power generation has become one of the core development directions in the global energy sector and has received high attention from countries around the world.

[0003] Wind turbine generators (hereinafter referred to as wind turbines) are systems that convert the kinetic energy of wind into electrical energy. Their safe and stable operation directly determines the quality of wind power supply. As the operating conditions of wind turbines become increasingly demanding, ensuring their safe and stable operation and improving their operational reliability have become hot topics in wind power research. The core components of a wind turbine are mainly the gearbox, generator, and converter. However, wind turbines operate under complex conditions for extended periods, and environmental factors (such as wind, sand, or humidity) can easily lead to problems such as scale buildup in the cooling system or fan performance degradation, directly resulting in a decline in thermal management capabilities. The proportion of downtime caused by thermal management failures is increasing, not only causing power generation losses but also increasing operation and maintenance costs and equipment wear and tear.

[0004] Therefore, there is an urgent need for a cloud-edge collaborative self-supervised predictive control system and method for the thermal management of wind turbine units. Summary of the Invention

[0005] The purpose of this invention is to provide a cloud-edge collaborative self-supervised predictive control system and method for thermal management of wind turbines. By making real-time decisions at the edge layer and optimizing globally at the cloud layer, it solves the latency problem of traditional systems and reduces downtime caused by thermal management failures. It also avoids the local optimum trap of a single edge system and can be adapted to wind turbines in different scenarios such as onshore or offshore, effectively reducing power generation losses and operation and maintenance costs.

[0006] To achieve the above objectives, the present invention provides a cloud-edge collaborative self-supervised predictive control system for wind turbine thermal management, comprising: an edge layer, a transmission layer, and a cloud layer;

[0007] The edge layer includes: a data acquisition module, a data preprocessing module, a self-supervised detection module, and an execution module; the data acquisition module, data preprocessing module, self-supervised detection module, and execution module are connected via communication.

[0008] The transport layer includes a cloud-edge collaborative interaction module; the cloud-edge collaborative interaction module is communicatively connected to the execution module of the edge layer.

[0009] The cloud layer includes a thermal fault diagnosis unit and a global thermal management optimization unit; the thermal fault diagnosis unit is communicatively connected to the cloud-edge collaborative interaction module; the global thermal management optimization unit is communicatively connected to the thermal fault diagnosis unit.

[0010] Preferably, the data acquisition module is used to acquire multi-dimensional data of the wind turbine in real time, including wind turbine operating data and environmental parameters;

[0011] (1) Wind turbine operating data includes thermal status data and operating condition data;

[0012] The thermal status data includes: gearbox, generator stator / rotor temperature and converter IGBT module temperature;

[0013] Operating data includes wind speed, engine speed, torque, and grid power.

[0014] (2) Environmental parameters include ambient temperature, humidity and precipitation.

[0015] Preferably, the data preprocessing module is used to preprocess the multidimensional data of the wind turbine collected by the data acquisition module to obtain standardized multidimensional data; based on the historical operation records without anomalies, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset;

[0016] The preprocessing process includes the following steps:

[0017] First, high-frequency noise reduction is performed on the multidimensional data of the wind turbine using wavelet transform.

[0018] Secondly, the mean imputation method was used to fill in the missing data in the multidimensional data of the denoised wind turbine units;

[0019] Finally, the missing multidimensional data of the wind turbine was mapped to using the max-min normalization method. Interval.

[0020] Preferably, the self-supervised detection module includes a self-supervised pre-training unit, a feature extraction unit, and an anomaly scoring unit;

[0021] Self-supervised pre-training units are used to build and train CNN-Transformer models based on pre-training datasets to generate a normal working condition feature library.

[0022] The feature extraction unit, based on the trained CNN-Transformer model, performs real-time feature vector extraction on the preprocessed standardized multidimensional data.

[0023] The anomaly scoring unit compares the real-time feature vector with the feature vector in the normal operating condition feature library to obtain the feature similarity. The feature similarity is negatively correlated with the degree of anomaly of the real-time data and is used to characterize the degree of anomaly of the real-time data.

[0024] Preferably, the CNN-Transformer model includes: a CNN subunit for extracting local dimensional coupling features of the data;

[0025] The Transformer subunit extracts global temporal features from the local dimensional coupling features output by the CNN subunit through a self-attention mechanism.

[0026] The feature fusion subunit fuses the local dimensional coupling features of the CNN subunit and the global temporal features extracted by the Transformer subunit to generate a comprehensive feature vector.

[0027] Preferably, the CNN-Transformer model is self-supervised pre-trained based on the pre-trained dataset, and the comprehensive feature vectors generated by the self-supervised pre-training are used to construct a normal working condition feature library.

[0028] Preferably, the anomaly scoring unit classifies the thermal state level based on feature similarity, according to the following rules:

[0029] Based on the operating scenarios and engineering practice experience of wind turbine units, a first threshold A and a second threshold B are set. The first threshold A is the minimum matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The second threshold B is set based on the feature vectors corresponding to the thermal safety critical conditions of the gearbox, generator stator / rotor, and converter, and is the critical matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The first threshold A... Second threshold B;

[0030] (1) Feature similarity The first threshold A is used to determine the normal operating condition.

[0031] (2) Second threshold B Feature similarity The first threshold A determines that the condition is mildly abnormal;

[0032] (3) Feature similarity The second threshold B determines a severe abnormality and triggers an alert.

[0033] Preferably, the execution module is connected to the self-supervised detection module, and performs differential regulation based on the output thermal state level of the self-supervised detection module.

[0034] Normal operating conditions: Adaptive cooling is achieved by adjusting fan speed; and normal operating condition data is stored locally;

[0035] Minor anomalies: By dynamically allocating the microchannel coolant flow rate, the semiconductor cooling power is adaptively adjusted, and an anomaly data message is generated and uploaded to the transport layer;

[0036] Severe anomaly: Emergency control, locate the faulty component, trigger an alarm, generate an abnormal data message, and upload the abnormal data message to the transport layer.

[0037] Preferably, the thermal fault diagnosis unit in the cloud layer receives abnormal data packets uploaded from the transport layer, combines them with the historical fault case library, completes fault tracing and predicts the remaining service life, and outputs maintenance suggestions;

[0038] The global thermal management optimization unit constructs an optimization model by aggregating the thermal state level and operating condition data of the wind turbine, generates optimization parameters, and transmits them to the execution module and the self-monitoring detection module.

[0039] A cloud-edge collaborative self-supervised predictive control method for wind turbine thermal management includes the following steps:

[0040] S1. Collect multi-dimensional data of the wind turbine in real time through the data acquisition module and input it to the data preprocessing module;

[0041] S2. The collected multidimensional data of the wind turbine is preprocessed using the data preprocessing module to obtain standardized multidimensional data; based on the historical operation records without anomalies, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset.

[0042] S3. Based on the pre-trained dataset obtained in S2, the CNN-Transformer model is constructed and trained through the self-supervised pre-training unit of the self-supervised detection module to generate a normal working condition feature library.

[0043] S4. Through the feature extraction unit of the self-supervised detection module, the trained CNN-Transformer model is called to extract real-time feature vectors from the standardized multidimensional data preprocessed in S2; and through the anomaly scoring unit of the self-supervised detection module, the feature similarity between the real-time feature vectors and the feature vectors of the normal working condition feature library obtained in S3 is calculated; and the thermal state level is classified.

[0044] S5, based on the thermal state levels divided by S4, performs differentiated control through the execution module, and generates abnormal data messages, which are then uploaded to the cloud layer via the transport layer;

[0045] S6. After receiving abnormal data messages, the cloud-layer thermal fault diagnosis unit completes fault tracing and predicts the remaining service life. The cloud-layer global thermal management optimization unit generates optimization parameters, which are transmitted to the execution module and the self-supervised detection module via the transmission layer.

[0046] Therefore, the present invention adopts the above-mentioned cloud-edge collaborative wind turbine thermal management self-supervised predictive control system and method. Through real-time decision-making at the edge layer and global optimization at the cloud layer, it not only solves the latency problem of traditional systems and reduces downtime caused by thermal management failures, but also avoids the local optimum trap of a single edge system. It can be adapted to wind turbines in different scenarios such as onshore or offshore, effectively reducing power generation losses and operation and maintenance costs.

[0047] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0048] Figure 1 This is a structural diagram of the cloud-edge collaborative wind turbine thermal management self-supervised predictive control system of the present invention;

[0049] Figure 2 This is a flowchart of the cloud-edge collaborative self-supervised predictive control method for thermal management of wind turbine units in this embodiment of the invention. Detailed Implementation

[0050] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] Example

[0052] like Figure 1 As shown, the present invention provides a cloud-edge collaborative self-supervised predictive control system for wind turbine thermal management, comprising: an edge layer, a transmission layer, and a cloud layer.

[0053] The edge layer includes a data acquisition module, a data preprocessing module, a self-supervised detection module, and an execution module, which are connected to each other through communication.

[0054] The data acquisition module is used to collect multi-dimensional data of the wind turbine in real time, including wind turbine operation data and environmental parameters.

[0055] (1) Wind turbine operating data includes thermal status data and operating condition data.

[0056] Thermal status data includes real-time temperature data for the gearbox, generator stator / rotor, and converter IGBT modules.

[0057] Operating condition sensors include operating condition data such as wind speed, rotational speed, torque, and grid power.

[0058] (2) Environmental sensors are used to acquire external data such as ambient temperature, humidity and precipitation.

[0059] The data preprocessing module is used to preprocess the multidimensional data of the wind turbine collected by the data acquisition module to obtain standardized multidimensional data. Based on the historical operation records without anomalies, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset.

[0060] The preprocessing process includes the following steps:

[0061] First, the wind turbine multidimensional data is subjected to high-frequency noise reduction using wavelet transform to retain the effective features of the wind turbine multidimensional data.

[0062] Secondly, the mean imputation method was used to fill in the missing data in the multidimensional data of the denoised wind turbine.

[0063] Finally, the missing multidimensional data of the wind turbine was mapped to using the max-min normalization method. Interval.

[0064] The self-supervised detection module includes a self-supervised pre-training unit, a feature extraction unit, and an anomaly scoring unit.

[0065] Self-supervised pre-training units are used to build and train CNN-Transformer models based on pre-training datasets, generating a normal operating condition feature library.

[0066] The CNN-Transformer model includes: a CNN subunit for extracting local dimensional coupling features of the data; a Transformer subunit for extracting global temporal features from the local dimensional coupling features output by the CNN subunit through a self-attention mechanism; and a feature fusion subunit for fusing the local dimensional coupling features of the CNN subunit and the global temporal features extracted by the Transformer subunit to generate a comprehensive feature vector.

[0067] Based on the pre-trained dataset, the CNN-Transformer model is self-supervised pre-trained, and the comprehensive feature vectors generated by the training are used to construct a normal working condition feature library.

[0068] The feature extraction unit, based on the trained CNN-Transformer model, performs real-time feature vector extraction on the preprocessed standardized multidimensional data.

[0069] The anomaly scoring unit compares the real-time feature vector with the feature vector in the normal operating condition feature library to obtain the feature similarity. The feature similarity is negatively correlated with the degree of anomaly of the real-time data and is used to characterize the degree of anomaly of the real-time data.

[0070] The thermal state levels are determined based on feature similarity, according to the following rules:

[0071] Based on the operating scenarios of wind turbine units (e.g., onshore or offshore) and engineering practice experience, a first threshold A and a second threshold B are set. The first threshold A is the minimum matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The second threshold B is set based on the feature vectors corresponding to the thermal safety critical conditions of the gearbox, generator stator / rotor, and converter, and is the critical matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The first threshold A... Second threshold B;

[0072] (1) Feature similarity The first threshold A is used to determine the normal operating condition.

[0073] (2) Second threshold B Feature similarity The first threshold A determines that the condition is mildly abnormal.

[0074] (3) Feature similarity The second threshold B determines a severe abnormality and triggers an alert.

[0075] The execution module is connected to the self-supervised detection module. Based on the output thermal state level of the self-supervised detection module, it performs differential regulation.

[0076] (1) Normal operating conditions: adaptive heat dissipation by adjusting the fan speed; and storing normal operating condition data locally.

[0077] (2) Mild anomaly: By dynamically allocating the microchannel coolant flow rate, the semiconductor cooling power is adaptively adjusted, and an abnormal data message is generated and uploaded to the transport layer.

[0078] (3) Severe anomaly: Emergency control, lock the location of the faulty component, trigger the alarm, generate an abnormal data message, and upload the abnormal data message to the transport layer.

[0079] The transport layer includes a cloud-edge collaborative interaction module, which communicates with the execution module and is used to upload abnormal data messages to the cloud layer.

[0080] The cloud layer includes a thermal fault diagnosis unit and a global thermal management optimization unit, and the modules are connected through communication.

[0081] The thermal fault diagnosis unit receives abnormal data packets uploaded from the transport layer, combines them with a historical fault case library, completes fault tracing and predicts remaining service life, and outputs maintenance suggestions.

[0082] The global thermal management optimization unit constructs an optimization model by aggregating the thermal state level and operating condition data of the wind turbine, generates optimization parameters, and transmits them to the execution module and the self-monitoring detection module.

[0083] Based on the above system, such as Figure 2 As shown, the cloud-edge collaborative self-supervised predictive control method for wind turbine thermal management includes the following steps:

[0084] S1. The data acquisition module collects multi-dimensional data of the wind turbine in real time and inputs it into the data preprocessing module.

[0085] S2. The collected multidimensional data of the wind turbine is preprocessed using the data preprocessing module to obtain standardized multidimensional data. Based on the historical operation records without abnormalities, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset.

[0086] S3. Based on the pre-trained dataset obtained in S2, the CNN-Transformer model is constructed and trained through the self-supervised pre-training unit of the self-supervised detection module to generate a normal working condition feature library.

[0087] S4. The feature extraction unit of the self-supervised detection module calls the trained CNN-Transformer model to extract real-time feature vectors from the standardized multidimensional data preprocessed in S2; and the anomaly scoring unit of the self-supervised detection module calculates the feature similarity between the real-time feature vectors and the feature vectors of the normal working condition feature library obtained in S3; and classifies the thermal state level.

[0088] S5, based on the thermal state levels defined by S4, performs differentiated control through the execution module, and generates abnormal data messages, which are then uploaded to the cloud layer via the transport layer.

[0089] S6. After receiving abnormal data messages, the cloud-based thermal fault diagnosis unit completes fault tracing and predicts the remaining service life. The cloud-based global thermal management optimization unit generates optimization parameters and transmits them to the execution module and the self-supervised detection module.

[0090] Therefore, the present invention adopts the above-mentioned cloud-edge collaborative wind turbine thermal management self-supervised predictive control system and method. Through real-time decision-making at the edge layer and global optimization at the cloud layer, it not only solves the latency problem of traditional systems and reduces downtime caused by thermal management failures, but also avoids the local optimum trap of a single edge system. It can be adapted to wind turbines in different scenarios such as onshore or offshore, effectively reducing power generation losses and operation and maintenance costs.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management, characterized in that, include: Edge layer, transport layer, and cloud layer; The edge layer includes: a data acquisition module, a data preprocessing module, a self-supervised detection module, and an execution module; the data acquisition module, data preprocessing module, self-supervised detection module, and execution module are connected via communication. The execution module is connected to the self-supervised detection module. Based on the output thermal state level of the self-supervised detection module, it performs differential regulation. Normal operating conditions: Adaptive cooling is achieved by adjusting fan speed; and normal operating condition data is stored locally; Minor anomalies: By dynamically allocating the microchannel coolant flow rate, the semiconductor cooling power is adaptively adjusted, and an anomaly data message is generated and uploaded to the transport layer; Severe anomaly: Emergency control, locate the faulty component, trigger an alarm, generate an abnormal data message, and upload the abnormal data message to the transport layer; The transport layer includes a cloud-edge collaborative interaction module; the cloud-edge collaborative interaction module is communicatively connected to the execution module of the edge layer. The cloud layer includes: a thermal fault diagnosis unit and a global thermal management optimization unit; the thermal fault diagnosis unit is communicatively connected to the cloud-edge collaborative interaction module; the global thermal management optimization unit is communicatively connected to the thermal fault diagnosis unit. The cloud-based thermal fault diagnosis unit receives abnormal data packets uploaded from the transport layer, combines them with a historical fault case library to complete fault tracing and predict remaining service life, and outputs maintenance suggestions. The global thermal management optimization unit constructs an optimization model by aggregating the thermal state level and operating condition data of the wind turbine, generates optimization parameters, and transmits them to the execution module and the self-monitoring detection module.

2. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 1, characterized in that, The data acquisition module is used to collect multi-dimensional data of the wind turbine in real time, including wind turbine operation data and environmental parameters. (1) Wind turbine operating data includes thermal status data and operating condition data; The thermal status data includes real-time temperature data of the gearbox, generator stator, generator rotor, and converter IGBT module. Operating data includes wind speed, engine speed, torque, and grid power. (2) Environmental parameters include ambient temperature, humidity and precipitation.

3. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 2, characterized in that, The data preprocessing module is used to preprocess the multidimensional data of the wind turbine collected by the data acquisition module to obtain standardized multidimensional data; based on the historical operation records without anomalies, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset. The preprocessing process includes the following steps: First, high-frequency noise reduction is performed on the multidimensional data of the wind turbine using wavelet transform. Secondly, the mean imputation method was used to fill in the missing data in the multidimensional data of the denoised wind turbine units; Finally, the missing multidimensional data of the wind turbine was mapped to using the max-min normalization method. Interval.

4. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 3, characterized in that, The self-supervised detection module includes a self-supervised pre-training unit, a feature extraction unit, and an anomaly scoring unit. Self-supervised pre-training units are used to build and train CNN-Transformer models based on pre-training datasets to generate a normal working condition feature library. The feature extraction unit, based on the trained CNN-Transformer model, performs real-time feature vector extraction on the preprocessed standardized multidimensional data. The anomaly scoring unit compares the real-time feature vector with the feature vector in the normal operating condition feature library to obtain the feature similarity. The feature similarity is negatively correlated with the degree of anomaly of the real-time data and is used to characterize the degree of anomaly of the real-time data.

5. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 4, characterized in that, The CNN-Transformer model includes: CNN sub-units, used to extract local dimensional coupling features of the standardized multidimensional data; The Transformer subunit extracts global temporal features from the local dimensional coupling features output by the CNN subunit through a self-attention mechanism. The feature fusion subunit fuses the local dimensional coupling features of the CNN subunit and the global temporal features extracted by the Transformer subunit to generate a comprehensive feature vector.

6. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 5, characterized in that, Based on the pre-trained dataset, the CNN-Transformer model is self-supervised pre-trained, and the comprehensive feature vectors generated by the self-supervised pre-training are used to construct a normal working condition feature library.

7. The cloud-edge collaborative self-monitoring predictive control system for wind turbine thermal management according to claim 6, characterized in that, The anomaly scoring unit classifies thermal state levels based on feature similarity, according to the following rules: Based on the operating scenarios and engineering practice experience of wind turbine units, a first threshold A and a second threshold B are set. The first threshold A is the minimum matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The second threshold B is set based on the feature vectors corresponding to the thermal safety critical conditions of the gearbox, generator stator, generator rotor, and converter, and is the critical matching degree threshold between the real-time feature vector and the feature vector under normal operating conditions. The first threshold A... Second threshold B; (1) Feature similarity The first threshold A is used to determine the normal operating condition. (2) Second threshold B Feature similarity The first threshold A determines that the condition is mildly abnormal; (3) Feature similarity The second threshold B determines a severe abnormality and triggers an alert.

8. A cloud-edge collaborative self-supervised predictive control method for wind turbine thermal management, applied to the cloud-edge collaborative self-supervised predictive control system for wind turbine thermal management as described in any one of claims 1-7, characterized in that, The method includes the following steps: S1. Collect multi-dimensional data of the wind turbine in real time through the data acquisition module and input it to the data preprocessing module; S2. The collected multidimensional data of the wind turbine is preprocessed using the data preprocessing module to obtain standardized multidimensional data; based on the historical operation records without anomalies, the standardized multidimensional data under normal operating conditions is selected to form a pre-training dataset. S3. Based on the pre-trained dataset obtained in S2, the CNN-Transformer model is constructed and trained through the self-supervised pre-training unit of the self-supervised detection module to generate a normal working condition feature library. S4. Through the feature extraction unit of the self-supervised detection module, the trained CNN-Transformer model is called to extract real-time feature vectors from the standardized multidimensional data preprocessed in S2; and through the anomaly scoring unit of the self-supervised detection module, the feature similarity between the real-time feature vectors and the feature vectors of the normal working condition feature library obtained in S3 is calculated; and the thermal state level is classified. S5, based on the thermal state levels divided by S4, performs differentiated control through the execution module, and generates abnormal data messages, which are then uploaded to the cloud layer via the transport layer; S6. After receiving abnormal data messages, the cloud-layer thermal fault diagnosis unit completes fault tracing and predicts the remaining service life. The cloud-layer global thermal management optimization unit generates optimization parameters, which are transmitted to the execution module and the self-supervised detection module via the transmission layer.

Images