Wind generating set health management system based on deep learning and application method

Through the deep learning-based wind turbine health management system, the problem that the wind turbine status monitoring system cannot establish complex coupling relationship models and long-term predictions is solved, and the intelligent health management of wind turbines is realized, which improves detection accuracy and early warning capabilities. , reducing operation and maintenance costs.

CN118934482APending Publication Date: 2024-11-12HARBIN NENGCHUANG DIGITAL TECH CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202410824827.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-25
Publication Date
2024-11-12

AI Technical Summary

Technical Problem

The existing wind turbine status monitoring system cannot establish a complex coupling relationship model between the various components of the unit, it is difficult to achieve detailed diagnosis at the component level, and it is impossible to conduct long-term status prediction, and lacks the function of analyzing the development trend of the fault.

Method used

Adopt a wind turbine health management system based on deep learning, including data acquisition module, data processing module, model building module, status evaluation module, early warning module and human-computer interaction interface, and build models through deep neural networks and recurrent neural networks to perform data Processing and status evaluation to realize intelligent health management of wind turbines.

Benefits of technology

It improves the accuracy and sensitivity of state detection, realizes the intelligence of wind turbine condition monitoring and fault prediction, reduces the operation and maintenance costs of wind farms, and improves power generation efficiency and economic benefits.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118934482A_ABST
    Figure CN118934482A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of wind power generation, and discloses a wind generating set health management system based on deep learning and an application method. The system integrates a deep learning algorithm and operation data of a wind generating set; the invention relates to a power generator set key component automatic evaluation and fault early warning system, which realizes automatic evaluation and fault early warning of the working state of a power generator set key component, and comprises a data acquisition module, a data processing module, a model construction module, a state evaluation module and an early warning module, and compared with the existing method based on a physical model and threshold judgment, the system uses an end-to-end data driving method. According to the system, an accurate physical model does not need to be constructed manually, the dynamic prediction capability of a deep learning model is utilized, early warning of faults can be achieved, the accuracy and sensitivity of state detection are improved, intelligentization of condition monitoring and fault prediction of the wind generating set is achieved, the operation and maintenance cost of a wind power plant is reduced, and the working efficiency is improved. And the power generation efficiency and the economic benefit are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a wind turbine health management system and an application method based on deep learning. Background Art

[0002] With the rapid development of wind power generation, my country's wind power installed capacity is increasing steadily. In order to adapt to the continuous improvement of power levels, super-large wind turbines above megawatt level are becoming mainstream products. These large wind turbines are mainly composed of hubs, main shafts, gearboxes, generators, control systems, wind measurement systems and other components. They have complex structures and very harsh operating conditions. They operate in harsh offshore environments for a long time, which can easily lead to mechanical failures, system out of control and other accidents, causing huge economic losses and safety hazards. Therefore, establishing an automated wind turbine health management system to realize real-time monitoring of the unit's operating status and fault prediction is of great significance to ensuring the safe and efficient operation of wind farms.

[0003] At present, the traditional wind turbine condition monitoring system mainly relies on sensors to collect physical signals such as vibration and temperature, and conducts simple comparisons based on the thresholds set by the project. When the signal exceeds the threshold, an alarm is generated. This condition assessment method based on physical models and empirical judgment has obvious defects: first, it is impossible to establish a complex coupling relationship model between the various components of the unit, and it is difficult to achieve detailed diagnosis at the component level; second, it is impossible to perform long-term condition prediction and lacks the function of fault development trend analysis.

[0004] In response to the above problems, deep learning technology provides a new idea for building an intelligent state assessment and fault prediction system. Deep learning is a cutting-edge technology in machine learning. By simulating the neural network structure of the human brain, it can achieve deep abstraction and learning of complex data features. Compared with traditional methods, deep learning has the versatility of learning complex data representation and powerful feature extraction capabilities. By using deep neural networks for end-to-end training, it can directly learn the mapping relationship of state assessment from massive multi-source data without the need to manually build a physical model.

[0005] In addition, the recurrent neural network (RNN) in deep learning can process long-sequence time data and realize modeling of the system state evolution process. Therefore, deep learning technology provides the possibility for building an automated and intelligent wind turbine health management system. In short, the development of wind turbines towards intelligence and informatization is an inevitable trend. The application of deep learning technology, full use of wind turbine operation data, and establishment of a health management system are of great significance to improving wind power reliability and economic benefits. Summary of the invention

[0006] In order to solve the technical problems that the existing complex coupling relationship model between the various components of the unit cannot be established, the detailed diagnosis at the component level is difficult to achieve, the long-term state prediction cannot be performed, and the fault development trend analysis function is lacking, the present invention provides a wind turbine health management system based on deep learning and an application method.

[0007] The present invention is implemented by the following technical scheme: a wind turbine health management system based on deep learning and an application method, including a data acquisition module, a data processing module, a model building module, a status assessment module, an early warning module and a human-computer interaction interface. First, the data collected by the data acquisition module will be transmitted to the data processing module for processing and stored in a time series database IoTDB, etc. Then, the model building module reads the time series data in IoTDB, constructs various neural network models, and continuously learns and optimizes. After that, the status assessment module determines the danger level of the health status of the component according to the real-time input data, and the assessment results are fed back to the early warning module.

[0008] Preferably, the data acquisition module collects operating data such as pressure, temperature, and current of the wind turbine generator set through a sensor network configured in key components of the wind turbine generator set. The sensor network includes but is not limited to a cabin vibration sensor, a gearbox oil pressure sensor, a generator temperature sensor, a mechanical control system signal acquisition, etc. The acquisition frequency ranges from seconds to minutes. The collected data needs to be time-synchronized and then sent to the back-end data processing module in a streaming manner.

[0009] Preferably, the data processing module performs format verification on the collected data, performs quality assessment on the data, outputs data quality details of different measuring points every day, stores the real-time collected data stream in a time series database such as IoTDB, and stores wind farm and equipment metadata in PostgreSQL.

[0010] Preferably, the model building module uses a deep learning framework such as TensorFlow to build various neural network models, including convolutional neural networks, recurrent neural networks, etc., by reading the time series data in IoTDB. The model is continuously optimized by adjusting parameters and incremental training, and finally a state assessment model of components such as the cabin and gearbox is obtained. The model supports incremental learning and online updates.

[0011] Preferably, the status assessment module uses a trained model to predict the current health status of the component based on real-time input data, monitor whether the status parameters of different components are within the normal range, predict the remaining service life of the component, and determine the danger level of the health status of the component. The assessment results are fed back to the early warning module in JSON format. At the same time, the collected data is analyzed and processed, and potential failure modes and abnormal behaviors are identified using technologies such as data mining and machine learning. Through the fault diagnosis algorithm, possible failures can be discovered and predicted in advance, and maintenance measures can be taken in time to reduce failure risks and maintenance costs.

[0012] Preferably, the early warning module uses rule engine technology to integrate the status assessment results of each model of each component, perform statistical analysis of failure risks, output early warning information, divide the level of warning according to the type and severity of the failure, send different levels of early warning information to management personnel through pop-up windows, and record all early warning logs.

[0013] Preferably, the human-computer interaction interface module is presented in a Web format, displaying the real-time status information, health scores, and various statistical analysis results of the components. The user can configure system parameters, browse historical data, and support confirmation feedback for early warning information. The above is a detailed description of the present invention. Each module integrates innovative algorithms and software and hardware to achieve an end-to-end solution for intelligent evaluation of the status of wind turbines and fault prediction.

[0014] Preferably, the sensor network includes a cabin vibration sensor, a gearbox oil pressure sensor, a generator temperature sensor, a strain sensor, an anemometer and a wind vane.

[0015] Preferably, the neural network model includes a convolutional neural network and a recurrent neural network.

[0016] Preferably, the warning module sets multiple warning thresholds according to the fault type and severity, sends warning information of different levels to management personnel through pop-up windows, and records all warning logs.

[0017] Compared with the prior art, the present invention has the following beneficial effects: When the present invention is in use, the system uses an end-to-end data-driven approach and does not require manual construction of an accurate physical model. At the same time, by utilizing the dynamic prediction capabilities of the deep learning model, early warning of faults can be achieved. The system improves the accuracy and sensitivity of state detection, realizes the intelligent monitoring of wind turbine condition and fault prediction, reduces the operation and maintenance costs of wind farms, and improves power generation efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of a system flow chart of the present invention; Figure 2 A sensor network for key parts of the wind turbine generator set of the system of the present invention; Figure 3 The present invention provides a flow chart for constructing and evaluating the model construction module and the state evaluation module. DETAILED DESCRIPTION

[0019] The present invention is further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form a new embodiment.

[0020] At present, the traditional wind turbine condition monitoring system is unable to establish a complex coupling relationship model between the various components of the unit, and it is difficult to achieve detailed diagnosis at the component level. Secondly, it is impossible to perform long-term condition prediction and lacks the function of fault development trend analysis.

[0021] Example 1: Please refer to Figure 1-3 , a wind turbine health management system and application method based on deep learning in this embodiment include a data acquisition module, a data processing module, a model building module, a state evaluation module, an early warning module and a human-computer interaction interface. First, the data acquisition module collects the pressure, temperature, current and other data of the wind turbine, and transmits them to the data processing module for processing and stores them in the time series database IoTDB, etc. Then, the model building module reads the time series data in IoTDB, builds various neural network models, uses LSTM-RNN and other recurrent neural networks to train historical time series data, models the state sequence of components, integrates XGBoost, Random Forest and other algorithms, establishes a multi-source heterogeneous data prediction model, uses a multi-task learning method, customizes a deep learning model for each key component, verifies and optimizes the trained model, and then the state evaluation module determines the danger level of the health status of the component according to the real-time input data, and the evaluation result is fed back to the early warning module. Finally, the real-time status information of the component is displayed through the human-computer interaction interface; Furthermore, the data acquisition module collects the operating data of the wind turbine generator set, such as pressure, temperature, and current, through the sensor network configured in each key component of the wind turbine generator set. The sensor network includes but is not limited to the cabin vibration sensor, gearbox oil pressure sensor, generator temperature sensor, mechanical control system signal acquisition, etc. The acquisition frequency ranges from seconds to minutes. The collected data needs to be time-synchronized and then sent to the back-end data processing module in a streaming manner; Among them, sensors are installed at key parts of wind turbines such as main shaft, gearbox, pitch system, generator, etc. to monitor operation data in real time. Vibration sensors and temperature sensors are installed at the main shaft to measure the vibration acceleration, temperature and other data of the main shaft. Oil pressure and temperature sensors are installed at the gearbox to monitor the pressure, temperature and other parameters in the gearbox lubricating oil. Strain sensors are installed at the root of the blades of the pitch system to measure the angle of the blades and temperature sensors are installed to measure the temperature of the blades. Current sensors, temperature sensors and axial displacement sensors are installed at the generator to measure the current, temperature, axial displacement and other parameters of the generator. Anemometers and wind vanes are installed at the tower to measure wind speed and direction. Vibration sensors are installed to measure the vibration data of the tower. Furthermore, the data processing module performs format verification on the collected data, evaluates the data quality, and outputs the data quality details of different measuring points every day. The real-time collected data stream is stored in the time series database IoTDB, etc. The metadata of the wind farm and equipment is stored in PostgreSQL. Apache Kafka is used as the data transmission and storage pipeline to cache the data stream and write it to IoTDB in real time. In addition, it also provides offline data import function, which can import the exported offline SCADA data into IoTDB, develop data quality models, calculate the data integrity rate (actual data volume / theoretical data volume) and data efficiency (effective data volume / actual data volume) of each measuring point every day, and the data efficiency is calculated according to the upper and lower limits of the measuring point value, which are set according to business experience. The data quality results are returned to the front end in JSON, and PostgreSQL is used to store the static information of the wind turbine, such as rated speed, model, coordinates, name and other metadata. The front-end editing function is provided, and users can upload and edit metadata online on the front end. Furthermore, the model building module uses deep learning frameworks such as TensorFlow to build various neural network models, including convolutional neural networks and recurrent neural networks, by reading time series data in IoTDB. The model is continuously optimized through parameter adjustment and incremental training, and finally the state assessment model of components such as cabin and gearbox is obtained. The model supports incremental learning and online update, uses recurrent neural networks such as LSTM-RNN to train historical time series data, model the state sequence of components, integrates algorithms such as XGBoost and Random Forest, establishes a multi-source heterogeneous data prediction model, uses multi-task learning methods, customizes deep learning models for each key component, and verifies and optimizes the trained model; Furthermore, by analyzing and processing the collected data, using technologies such as data mining and machine learning, potential failure modes and abnormal behaviors can be identified. Through the fault diagnosis algorithm, possible failures can be discovered and predicted in advance, and maintenance measures can be taken in time, which can reduce the risk of failure and maintenance costs. The trained model is used in the status assessment module to predict the current health status of the component based on real-time input data, monitor whether the status parameters of different components are within the normal range, predict the remaining service life of the component, and determine the danger level of the component health status. The assessment results are fed back to the early warning module in JSON format; Among them, the status assessment module provides a new task function, that is, by selecting a specific wind farm, model, F (execution frequency), D (delay time), R (data range), model, number of CPUs, and memory size, the corresponding JSON parameters can be generated to call multiple pre-trained deep learning models. The model reads the parameters for prediction, monitors the health status of each part of the unit in real time, and gives operation and maintenance suggestions. This module also provides a model online update function, which supports the addition, editing, updating, and deletion of models. Furthermore, the early warning module uses rule engine technology to integrate the status assessment results of each component and each model, conduct statistical analysis of fault risks, output early warning information, divide the warning level according to the fault type and severity, set multiple warning thresholds according to the fault type and severity, send different levels of warning information to managers through pop-up windows, and record all warning logs; Furthermore, the human-computer interaction interface module is presented in a Web form, showing the real-time status information, health scores, and various statistical analysis results of the components. Users can configure system parameters, browse historical data, and support confirmation feedback for warning information. The above is a detailed description of the present invention. Each module integrates innovative algorithms and software and hardware to achieve an end-to-end solution for intelligent evaluation of wind turbine status and fault prediction. The human-computer interaction interface uses Python Web frameworks such as Django to build a Web interface, display monitoring information, support user management and authority control, and use visualization components such as ECharts to draw status monitoring charts. The above specifically explains the details of the technical solution of the present invention, which can effectively realize intelligent health management of wind turbines, achieve the purpose of status evaluation and fault warning, and improve the safety and economic benefits of wind farms. Working principle: The system uses data acquisition module, data processing module, model building module, state assessment module, early warning module and human-computer interaction interface. By using an end-to-end data-driven approach, there is no need to manually build an accurate physical model. At the same time, the dynamic prediction ability of the deep learning model can achieve early warning of faults. The system improves the accuracy and sensitivity of state detection, realizes the intelligent monitoring of wind turbine condition and fault prediction, reduces the operation and maintenance costs of wind farms, and improves power generation efficiency and economic benefits.

[0022] The above-mentioned embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and substitutions made by technicians in this field on the basis of the present invention shall fall within the scope of protection required by the present invention.

Claims

1. A wind turbine health management system and application method based on deep learning, characterized in that: It includes a data acquisition module, a data processing module, a model building module, a status assessment module, an early warning module and a human-computer interaction interface. First, the data collected by the data acquisition module will be transmitted to the data processing module for processing and stored in the time series database IoTDB, etc. Then the model building module reads the time series data in IoTDB, builds various neural network models, and continuously learns and optimizes. After that, the status assessment module determines the danger level of the component health status based on real-time input data, and the assessment results are fed back to the early warning module.

2. According to claim 1, a wind turbine health management system and application method based on deep learning is characterized in that: The data acquisition module collects operating data such as pressure, temperature, and current of the wind turbine generator set through a sensor network configured in key components of the wind turbine generator set.

3. According to a wind turbine health management system and application method based on deep learning according to claim 1, it is characterized in that: The data processing module performs format verification on the collected data, performs quality assessment on the data, and outputs quality details of the data at different measuring points every day.

4. According to claim 1, a wind turbine health management system and application method based on deep learning is characterized in that: The model building module uses deep learning frameworks such as TensorFlow to build various neural network models by reading time series data in IoTDB, and uses machine learning algorithms to train evaluation models of components such as the cabin, variable pitch system, and generator.

5. A wind turbine health management system and application method based on deep learning according to claim 1, characterized in that: The state assessment module uses the trained model to predict the current health state of the component based on real-time input data.

6. A wind turbine health management system and application method based on deep learning according to claim 1, characterized in that: The early warning module uses rule engine technology to integrate the status assessment results of each model of each component, conduct statistical analysis of failure risks, and output early warning information.

7. A wind turbine health management system and application method based on deep learning according to claim 1, characterized in that: The human-computer interaction interface module is presented in a Web format, displaying the real-time status information, health scores, and various statistical analysis results of the components.

8. A wind turbine health management system and application method based on deep learning according to claim 2, characterized in that: The sensor network includes a nacelle vibration sensor, a gearbox oil pressure sensor, a generator temperature sensor, a strain sensor, an anemometer and a wind vane.

9. A wind turbine health management system and application method based on deep learning according to claim 4, characterized in that: The neural network model includes a convolutional neural network and a recurrent neural network.

10. A wind turbine health management system and application method based on deep learning according to claim 6, characterized in that: The warning module sets multiple warning thresholds according to the fault type and severity, sends warning information of different levels to management personnel through pop-up windows, and records all warning logs.

Citation Information

Cited By

  • Wind power intelligent protection control system and method based on artificial intelligence

    CN120120182A

  • Power station equipment state trend prediction method based on deep learning

    CN120596899A

  • Online monitoring method and system for outlet pressure of gas compressor of gas turbine

    CN120744356A