Wind turbine generator set abnormal state detection method and system based on extreme learning machine

Abstract

This invention discloses a method and system for detecting abnormal states of wind turbine generators based on Extreme Learning Machine (ELM). The method includes: real-time acquisition of multi-dimensional operational data from the wind turbine generator; data aggregation and timestamping; preprocessing of the multi-dimensional operational data including cleaning, feature engineering, and normalization to obtain a feature dataset; construction of an ELM model comprising an input layer, hidden layer, and output layer; construction of a training set and a validation set based on pre-set historical operational data of the wind turbine generator; training the ELM model using the training set and optimizing hyperparameters using the validation set; deployment of the trained ELM model at the edge of the wind turbine generator; real-time reading of the feature dataset and execution of inference to obtain anomaly judgment results, thus completing the detection of abnormal states of the wind turbine generator. This invention utilizes the ability to learn from large amounts of data to identify abnormal patterns such as excessively high temperatures and abnormal vibrations, improving detection accuracy and flexibility.

Description

Technical Field

[0001] This invention relates to the technical field of wind turbine generator set testing, and in particular to a method and system for detecting abnormal states of wind turbine generator sets based on extreme learning machine. Background Technology

[0002] Traditional wind turbine abnormality detection mainly relies on the real-time acquisition and monitoring of single or limited physical quantities such as vibration analysis, temperature monitoring, and oil contamination detection. These methods pre-set fixed thresholds based on on-site maintenance experience and equipment factory parameters, and then rely on manual periodic review, comparison, and analysis of operational data. Only when monitored parameters such as vibration amplitude, temperature, or oil concentration exceed preset normal ranges will the system trigger an alarm signal, prompting maintenance personnel to conduct on-site investigation and handling.

[0003] The key advantages of this type of traditional detection method are that it is technically mature, intuitive in principle, low in cost and easy to implement. It does not have high requirements for hardware equipment and algorithm environment, and can meet basic safety monitoring needs when the operating conditions of wind turbine generators are relatively stable and the fault modes are relatively simple.

[0004] However, the fixed threshold and experience-driven approach also have significant limitations, including insufficient flexibility, poor adaptability, slow response, and low diagnostic accuracy. In actual operating conditions, wind turbine generators often face complex operating conditions such as load fluctuations, ambient temperature changes, and airflow disturbances. Traditional methods struggle to distinguish between normal fluctuations and genuine fault characteristics, easily leading to false alarms and missed alarms, and failing to meet the requirements of modern industry for high reliability, long lifespan, and intelligent operation and maintenance of wind turbine generators. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a method and system for detecting abnormal states of wind turbine generators based on Extreme Learning Machine (ELM). Utilizing the ELM method to detect abnormal states of wind turbines, it can learn from a large amount of data and identify potential abnormal patterns, such as excessively high temperatures, abnormal vibrations, and overloads. Compared with traditional methods, this significantly improves detection accuracy and flexibility.

[0006] The objective of this invention is achieved through the following technical solution: a method for detecting abnormal states of wind turbine generator sets based on extreme learning machines, comprising the following steps:

[0007] S1. Real-time acquisition of multi-dimensional operation data of wind turbine generator sets, after data aggregation and timestamp synchronization, and preprocessing of multi-dimensional operation data such as cleaning, feature engineering and normalization to obtain feature dataset;

[0008] S2. Construct an Extreme Learning Machine (ELM) model containing an input layer, a hidden layer, and an output layer; construct a training set and a validation set based on the preset historical operating data of wind turbine generators; train the ELM model using the training set and optimize the hyperparameters using the validation set.

[0009] S3. Deploy the trained Extreme Learning Machine (ELM) model at the edge of the wind turbine generator set, read the feature dataset in real time and perform inference to obtain the anomaly judgment result, and complete the anomaly state detection of the wind turbine generator set.

[0010] Furthermore, step S1 includes:

[0011] First, define the abnormal state types of the monitored wind turbine generators, the real-time indicators, accuracy indicators, and coverage indicators for abnormal detection, determine the types of sensors used for data collection and their corresponding monitoring parameters, and obtain multi-dimensional operational data.

[0012] The abnormal state types include mechanical abnormalities, electrical abnormalities, performance abnormalities, and operational abnormalities; the real-time performance indicator is set to a data acquisition to abnormality judgment delay of ≤1 second; the accuracy indicator is set to an abnormality identification accuracy rate of ≥95% and a false alarm rate of ≤3%; and the coverage indicator is set to support at least 10 abnormal state types.

[0013] The sensors include a vibration sensor, a temperature sensor, an electrical sensor, an environmental sensor, and a status sensor. The vibration sensor has a sampling frequency of 1 kHz, and its monitoring parameters are the acceleration and vibration frequency of the wind turbine generator's X, Y, and Z axes. The temperature sensor has a sampling frequency of 1 Hz, and its monitoring parameters are the generator temperature, gearbox temperature, and bearing temperature. The electrical sensor has a sampling frequency of 50 Hz, and its monitoring parameters are voltage, current, and power factor.

[0014] Furthermore, step S1 includes:

[0015] The data cleaning process involves removing sensor fault data that exceeds the physical measurement range, using wavelet transform to denoise multi-dimensional operational data, and filling in missing values ​​through interpolation of data from adjacent time points.

[0016] The feature engineering involves extracting time-domain features, frequency-domain features, and fused features. The time-domain features include the mean, peak value, kurtosis, and variance of multi-dimensional operational data. The frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operational data. The fused features include the ratio of wind speed to power and the rate of temperature change.

[0017] The normalization process involves standardizing the feature-engineered data to the range of [0,1] or [-1,1] to obtain the final feature dataset.

[0018] Furthermore, step S2 includes:

[0019] The dimension of the input layer is the number of preprocessed features, the number of nodes in the hidden layer ranges from 50 to 500, the activation function of the Extreme Learning Machine (ELM) model is Sigmoid or ReLU, and the output layer uses One-Hot encoding to represent the anomaly type.

[0020] Training and validation sets are constructed based on preset historical operating data, which includes normal operating data and abnormal operating data. The ratio of the training set to the validation set is 7:3. The SMOTE algorithm is used to perform sample augmentation on the category with fewer abnormal samples. The weights and biases from the input layer to the hidden layer are randomly initialized. The weights from the hidden layer to the output layer are calculated by Moore-Penrose generalized inverse. Hyperparameter optimization is achieved through cross-validation.

[0021] Furthermore, step S2 includes:

[0022] After the Extreme Learning Machine (ELM) model has been trained, its classification performance is evaluated by setting evaluation metrics, including accuracy, precision, recall, and F1 score. The inference time of the ELM model for a single input data point is tested. If the inference time is ≤50ms, the real-time requirement is met; if the inference time is >50ms, features are added to misclassified samples, and the output weights of the ELM model are updated through incremental learning for newly emerging anomaly types.

[0023] Furthermore, step S3 includes:

[0024] The cloud-based monitoring and visualization platform receives anomaly judgment results and related data uploaded from the edge of the wind turbine generator set, triggers graded alarms based on the severity of the anomaly, and automatically records anomaly information.

[0025] An abnormal state detection system for wind turbine generator sets based on extreme learning machines is used to implement the aforementioned abnormal state detection method for wind turbine generator sets based on extreme learning machines, including:

[0026] The data acquisition module is used to collect vibration, temperature, electrical, environmental and operating status data as multi-dimensional operating data, and transmit the multi-dimensional operating data through industrial bus or wireless transmission.

[0027] The data preprocessing module is used to aggregate, timestamp synchronize and cache multi-dimensional running data, perform data cleaning, feature engineering and normalization preprocessing on multi-dimensional running data, and output a standardized feature dataset.

[0028] The ELM model training module is used to build the Extreme Learning Machine (ELM) model and solve the weights of the ELM model based on a preset historical dataset.

[0029] The inference module integrates an Extreme Learning Machine (ELM) model and is configured in the control system of the wind turbine generator set. The control system of the wind turbine generator set acts as an edge device, performing ELM inference in real time and outputting the anomaly type judgment result.

[0030] The anomaly response module is configured in the preset monitoring visualization platform. The monitoring visualization platform, as the cloud, receives the output results from the edge terminal, judges the severity of the result according to the anomaly type, and issues alarms accordingly, and synchronously records logs to the time series database.

[0031] The evaluation iteration module optimizes the model based on the output of the inference module and feeds it back to the ELM model training module.

[0032] Furthermore, the data acquisition module includes several sensors deployed on the gearbox, generator, blade root, and tower of the wind turbine generator set. The sensors include vibration sensors, temperature sensors, electrical sensors, environmental sensors, and status sensors. The industrial bus is Modbus or Profinet, and the wireless transmission method is LoRa or 5G.

[0033] Furthermore, the data preprocessing module uses Redis caching; the data cleaning involves removing sensor fault data that exceeds the physical range, using wavelet transform to denoise multi-dimensional running data, and filling in missing values ​​through interpolation of data from adjacent time points.

[0034] The feature engineering involves extracting time-domain features, frequency-domain features, and fused features. The time-domain features include the mean, peak value, kurtosis, and variance of multi-dimensional operational data. The frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operational data. The fused features include the ratio of wind speed to power and the rate of temperature change.

[0035] The normalization process involves standardizing the feature-engineered data to the range of [0,1] or [-1,1] to obtain the final feature dataset.

[0036] Furthermore, the anomaly response module adopts a tiered alarm system, which is divided into orange alarms and red alarms according to the severity of the anomaly, further triggering local audible and visual alarms, remotely pushing to the operation and maintenance platform, and automatically triggering shutdown protection; the log includes the time of anomaly occurrence, characteristic data, and anomaly judgment results.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] 1. This invention presents an abnormal state detection method for wind turbine generators based on Extreme Learning Machine (ELM). Compared to traditional machine learning methods such as SVM, BP neural networks, and deep learning, this method offers faster training and inference speeds to meet real-time requirements, more accurate nonlinear fitting to adapt to complex operating conditions, lower computational requirements to adapt to edge deployment, and more flexible incremental updates to adapt to dynamic scenarios. These characteristics make it a preferred solution for balancing performance and cost, especially suitable for the low-cost, high-reliability monitoring needs of large-scale wind farms.

[0039] 2. This invention can provide a second-level response to sudden faults such as abnormal vibration and sudden temperature rise, reducing the minute-level delay of traditional SCADA systems by more than 90%.

[0040] 3. This invention eliminates environmental interference by associating feature data from various sensors, thus solving the problem of frequent false alarms in single-sensor monitoring.

[0041] 4. This invention can provide early warnings to avoid losses due to sudden malfunctions and daily downtime; at the same time, it can accurately locate abnormal parts, and maintenance personnel can directly replace the parts upon arrival, reducing blind repairs.

[0042] 5. This invention is compatible with both new and old wind turbine generator sets. By adjusting the feature weights of the Extreme Learning Machine (ELM) model, it solves the problem of high foundation vibration values ​​caused by aging.

[0043] 6. This invention uses an Extreme Learning Machine (ELM) model to detect abnormal states of wind turbine generators, which can integrate automated processes, achieve less manpower for inspection, and form a positive cycle of operation and maintenance.

[0044] In summary, this invention, through the high efficiency of the Extreme Learning Machine (ELM) model and the comprehensiveness of multi-sensor fusion, breaks through the limitations of traditional monitoring technologies in terms of real-time performance, accuracy, and economy, providing intelligent protection for the safe and stable operation of wind turbine generators and significantly improving the operational efficiency and lifecycle value of wind power assets. Attached Figure Description

[0045] Figure 1 This is a flowchart of the training process for the Extreme Learning Machine (ELM) model.

[0046] Figure 2 This is an interactive hierarchical architecture diagram of a wind turbine generator abnormal state detection system based on extreme learning machine. Detailed Implementation

[0047] The present invention will be further described below with reference to specific embodiments.

[0048] Example 1

[0049] The abnormal state detection method for wind turbine generator sets based on extreme learning machine provided in this embodiment includes the following steps:

[0050] S1. Real-time acquisition of multi-dimensional operational data from wind turbine generators. After data aggregation and timestamp synchronization, the multi-dimensional operational data undergoes preprocessing including cleaning, feature engineering, and normalization to obtain a feature dataset, including:

[0051] First, define the abnormal state types of the monitored wind turbine generators, the real-time indicators, accuracy indicators, and coverage indicators for abnormal detection, determine the types of sensors used for data collection and their corresponding monitoring parameters, and obtain multi-dimensional operational data.

[0052] The abnormal state types include mechanical abnormalities, electrical abnormalities, performance abnormalities, and operational abnormalities; the real-time performance indicator is set to a data acquisition to abnormality judgment delay of ≤1 second; the accuracy indicator is set to an abnormality identification accuracy rate of ≥95% and a false alarm rate of ≤3%; and the coverage indicator is set to support at least 10 abnormal state types.

[0053] The sensors include a vibration sensor, a temperature sensor, an electrical sensor, an environmental sensor, and a status sensor. The vibration sensor has a sampling frequency of 1 kHz, and its monitoring parameters are the acceleration and vibration frequency of the wind turbine generator's X, Y, and Z axes. The temperature sensor has a sampling frequency of 1 Hz, and its monitoring parameters are the generator temperature, gearbox temperature, and bearing temperature. The electrical sensor has a sampling frequency of 50 Hz, and its monitoring parameters are voltage, current, and power factor.

[0054] The data cleaning process involves removing sensor fault data that exceeds the physical measurement range, using wavelet transform to denoise multi-dimensional operating data, and filling in missing values ​​through interpolation of data from adjacent time points.

[0055] The feature engineering involves extracting time-domain features, frequency-domain features, and fused features. The time-domain features include the mean, peak value, kurtosis, and variance of multi-dimensional operating data. The frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operating data. The fused features include the ratio of wind speed to power and the rate of temperature change.

[0056] The normalization process involves standardizing the feature-engineered data to the range of [0,1] or [-1,1] to obtain the final feature dataset.

[0057] S2. Construct an Extreme Learning Machine (ELM) model containing an input layer, hidden layers, and an output layer; build a training set and a validation set based on pre-set historical operating data of wind turbine generators; train the ELM model using the training set and optimize the hyperparameters using the validation set, including:

[0058] The dimension of the input layer is the number of preprocessed features, the number of nodes in the hidden layer ranges from 50 to 500, the activation function of the Extreme Learning Machine (ELM) model is Sigmoid or ReLU, and the output layer uses One-Hot encoding to represent the anomaly type.

[0059] Training and validation sets are constructed based on preset historical operating data, which includes normal operating data and abnormal operating data. The ratio of the training set to the validation set is 7:3. The SMOTE algorithm is used to perform sample augmentation on the category with fewer abnormal samples. The weights and biases from the input layer to the hidden layer are randomly initialized. The weights from the hidden layer to the output layer are calculated by Moore-Penrose generalized inverse. Hyperparameter optimization is achieved through cross-validation.

[0060] After the Extreme Learning Machine (ELM) model has been trained, its classification performance is evaluated by setting evaluation metrics, including accuracy, precision, recall, and F1 score. The inference time of the ELM model for a single input data point is tested. If the inference time is ≤50ms, the real-time requirement is met; if the inference time is >50ms, features are added to misclassified samples, and the output weights of the ELM model are updated through incremental learning for newly emerging anomaly types.

[0061] See Figure 1 As shown, the training process for an Extreme Learning Machine (ELM) model includes the following steps:

[0062] S2.1 Dataset preparation: Input parameter is X, label parameter is T;

[0063] S2.2 Data Preprocessing: Normalize and standardize the input dataset, and divide it into training and validation sets;

[0064] S2.3. Define the network structure: the input dimension is n, the hidden layer nodes are L, and the output dimension is m;

[0065] S2.4 Random initialization: Input weights are W, and hidden layer biases are b;

[0066] S2.5 Select the activation function g(*);

[0067] S2.6 Calculate the hidden layer output matrix: H = g(XW + b);

[0068] S2.7 Solve for the output weight β = H↑T (generalized least squares);

[0069] S2.8, Model training complete;

[0070] S2.9 Test set prediction: Calculate H_test × β;

[0071] S2.10 Evaluation metrics: Evaluation shall be conducted using MSE or accuracy calculations.

[0072] S3. Deploy the trained Extreme Learning Machine (ELM) model at the edge of the wind turbine generator set. Read the feature dataset in real time and perform inference to obtain anomaly judgment results, thus completing the anomaly detection of the wind turbine generator set. The cloud-based monitoring and visualization platform receives the anomaly judgment results and related data uploaded from the edge of the wind turbine generator set, triggers tiered alarms based on the severity of the anomaly, and automatically records the anomaly information.

[0073] Example 2

[0074] This embodiment provides an abnormal state detection system for wind turbine generator sets based on Extreme Learning Machine (ELM), used to implement the abnormal state detection method for wind turbine generator sets based on ELM described in Embodiment 1, including:

[0075] The data acquisition module is used to collect vibration, temperature, electrical, environmental, and operational status data as multi-dimensional operational data, and transmits the multi-dimensional operational data through an industrial bus or wireless transmission. The data acquisition module includes several sensors deployed on the gearbox, generator, blade root, and tower of the wind turbine generator set. The sensors include vibration sensors, temperature sensors, electrical sensors, environmental sensors, and status sensors. The industrial bus is Modbus or Profinet, and the wireless transmission method is LoRa or 5G.

[0076] The data preprocessing module is used to aggregate, timestamp synchronize, and cache multi-dimensional operational data. It performs data cleaning, feature engineering, and normalization preprocessing on the multi-dimensional operational data, outputting a standardized feature dataset. The data preprocessing module uses Redis for caching. Data cleaning involves removing sensor fault data exceeding the physical range, using wavelet transform to denoise the multi-dimensional operational data, and filling in missing values ​​through interpolation of data from adjacent time points. Feature engineering involves extracting time-domain features, frequency-domain features, and fused features. The time-domain features include the mean, peak value, kurtosis, and variance of the multi-dimensional operational data; the frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operational data; and the fused features include the ratio of wind speed to power and the rate of temperature change. Normalization standardizes the feature-engineered data to the [0,1] or [-1,1] interval, ultimately obtaining the feature dataset.

[0077] The ELM model training module is used to build the Extreme Learning Machine (ELM) model and solve the weights of the ELM model based on a preset historical dataset.

[0078] The inference module integrates an Extreme Learning Machine (ELM) model and is configured in the control system of the wind turbine generator set. The control system of the wind turbine generator set acts as an edge device, performing ELM inference in real time and outputting the anomaly type judgment result.

[0079] An anomaly response module is configured in a pre-defined monitoring visualization platform. This platform, acting as a cloud-based system, receives output from the edge devices, determines the severity of the anomaly based on its type, and issues tiered alarms. Logs are simultaneously recorded in a time-series database. The anomaly response module employs tiered alarms, classifying anomalies into orange and red alerts based on severity. Further actions include triggering local audible and visual alarms, remote push notifications to the operations and maintenance platform, and automatic shutdown protection. The logs include the anomaly occurrence time, characteristic data, and anomaly determination results.

[0080] The evaluation iteration module optimizes the model based on the output of the inference module and feeds it back to the ELM model training module.

[0081] See Figure 2 The diagram shows the interactive hierarchical architecture of the wind turbine generator abnormal state detection system based on Extreme Learning Machine provided in this embodiment. It forms a hierarchical relationship of hardware perception layer - edge processing layer - cloud application layer. The hardware perception layer includes vibration sensors, temperature sensors, electrical sensors, environmental sensors, and status sensors; the edge processing layer includes an edge data gateway for Redis caching, a data preprocessing module, and an inference module; the cloud application layer includes a monitoring visualization platform, a historical data storage database, and an operation and maintenance management system. The relationships between each layer are shown in Table 1 below.

[0082] Table 1 Interaction Hierarchy Relationship Table

[0083] Example 3

[0084] This embodiment discloses a non-transitory computer-readable medium storing instructions that, when executed by a processor, perform the steps of the wind turbine generator abnormal state detection method based on extreme learning machine as described in Embodiment 1.

[0085] In this embodiment, the non-transitory computer-readable medium can be a disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), USB flash drive, portable hard drive, etc.

[0086] Example 4

[0087] This embodiment discloses a computing device, including a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the wind turbine generator abnormal state detection method based on extreme learning machine described in Embodiment 1.

[0088] The computing device described in this embodiment may be a desktop computer, laptop computer, smartphone, PDA handheld terminal, tablet computer, programmable logic controller (PLC), or other terminal device with processor function.

[0089] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for detecting abnormal states of wind turbine generator sets based on extreme learning machines, characterized in that, Includes the following steps: S1. Real-time acquisition of multi-dimensional operation data of wind turbine generator sets, after data aggregation and timestamp synchronization, and preprocessing of multi-dimensional operation data such as cleaning, feature engineering and normalization to obtain feature dataset; S2. Construct an Extreme Learning Machine (ELM) model containing an input layer, a hidden layer, and an output layer; construct a training set and a validation set based on the preset historical operating data of wind turbine generators; train the ELM model using the training set and optimize the hyperparameters using the validation set. S3. Deploy the trained Extreme Learning Machine (ELM) model at the edge of the wind turbine generator set, read the feature dataset in real time and perform inference to obtain the anomaly judgment result, and complete the anomaly state detection of the wind turbine generator set.

2. The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine according to claim 1, characterized in that, Step S1 includes: First, define the abnormal state types of the monitored wind turbine generators, the real-time indicators, accuracy indicators, and coverage indicators for abnormal detection, determine the types of sensors used for data collection and their corresponding monitoring parameters, and obtain multi-dimensional operational data. The abnormal state types include mechanical abnormalities, electrical abnormalities, performance abnormalities, and operational abnormalities; the real-time performance indicator is set to a data acquisition to abnormality judgment delay of ≤1 second; the accuracy indicator is set to an abnormality identification accuracy rate of ≥95% and a false alarm rate of ≤3%; and the coverage indicator is set to support at least 10 abnormal state types. The sensors include a vibration sensor, a temperature sensor, an electrical sensor, an environmental sensor, and a status sensor. The vibration sensor has a sampling frequency of 1 kHz, and its monitoring parameters are the acceleration and vibration frequency of the wind turbine generator's X, Y, and Z axes. The temperature sensor has a sampling frequency of 1 Hz, and its monitoring parameters are the generator temperature, gearbox temperature, and bearing temperature. The electrical sensor has a sampling frequency of 50 Hz, and its monitoring parameters are voltage, current, and power factor.

3. The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine according to claim 1, characterized in that, Step S1 includes: The data cleaning process involves removing sensor fault data that exceeds the physical measurement range, using wavelet transform to denoise multi-dimensional operational data, and filling in missing values ​​through interpolation of data from adjacent time points. The feature engineering involves extracting time-domain features, frequency-domain features, and fused features. The time-domain features include the mean, peak value, kurtosis, and variance of multi-dimensional operational data. The frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operational data. The fused features include the ratio of wind speed to power and the rate of temperature change. The normalization process involves standardizing the feature-engineered data to the range of [0,1] or [-1,1] to obtain the final feature dataset.

4. The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine according to claim 1, characterized in that, Step S2 includes: The dimension of the input layer is the number of preprocessed features, the number of nodes in the hidden layer ranges from 50 to 500, the activation function of the Extreme Learning Machine (ELM) model is Sigmoid or ReLU, and the output layer uses One-Hot encoding to represent the anomaly type. Training and validation sets are constructed based on preset historical operating data, which includes normal operating data and abnormal operating data. The ratio of the training set to the validation set is 7:

3. The SMOTE algorithm is used to perform sample augmentation on the category with fewer abnormal samples. The weights and biases from the input layer to the hidden layer are randomly initialized. The weights from the hidden layer to the output layer are calculated by Moore-Penrose generalized inverse. Hyperparameter optimization is achieved through cross-validation.

5. The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine according to claim 1, characterized in that, Step S2 includes: After the Extreme Learning Machine (ELM) model has been trained, its classification performance is evaluated by setting evaluation metrics, including accuracy, precision, recall, and F1 score. The inference time of the ELM model for a single input data point is tested. If the inference time is ≤50ms, the real-time requirement is met; if the inference time is >50ms, features are added to misclassified samples, and the output weights of the ELM model are updated through incremental learning for newly emerging anomaly types.

6. The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine according to claim 1, characterized in that, Step S3 includes: The cloud-based monitoring and visualization platform receives anomaly judgment results and related data uploaded from the edge of the wind turbine generator set, triggers graded alarms based on the severity of the anomaly, and automatically records anomaly information.

7. An abnormal state detection system for wind turbine generator sets based on extreme learning machine, characterized in that, The method for detecting abnormal states of wind turbine generator sets based on extreme learning machine as described in any one of claims 1-6 includes: The data acquisition module is used to collect vibration, temperature, electrical, environmental and operating status data as multi-dimensional operating data, and transmit the multi-dimensional operating data through industrial bus or wireless transmission. The data preprocessing module is used to aggregate, timestamp synchronize and cache multi-dimensional running data, perform data cleaning, feature engineering and normalization preprocessing on multi-dimensional running data, and output a standardized feature dataset. The ELM model training module is used to build the Extreme Learning Machine (ELM) model and solve the weights of the ELM model based on a preset historical dataset. The inference module integrates an Extreme Learning Machine (ELM) model and is configured in the control system of the wind turbine generator set. The control system of the wind turbine generator set acts as an edge device, performing ELM inference in real time and outputting the anomaly type judgment result. The anomaly response module is configured in the preset monitoring visualization platform. The monitoring visualization platform, as the cloud, receives the output results from the edge terminal, judges the severity of the result according to the anomaly type, and issues alarms accordingly, and synchronously records logs to the time series database. The evaluation iteration module optimizes the model based on the output of the inference module and feeds it back to the ELM model training module.

8. The wind turbine generator abnormal state detection system based on extreme learning machine according to claim 7, characterized in that, The data acquisition module includes several sensors deployed on the gearbox, generator, blade root, and tower of the wind turbine generator set. The sensors include vibration sensors, temperature sensors, electrical sensors, environmental sensors, and status sensors. The industrial bus is Modbus or Profinet, and the wireless transmission method is LoRa or 5G.

9. The wind turbine generator abnormal state detection system based on extreme learning machine according to claim 7, characterized in that, The data preprocessing module uses Redis caching; the data cleaning involves removing sensor fault data that exceeds the physical range, using wavelet transform to denoise multi-dimensional running data, and filling in missing values ​​through interpolation of data from adjacent time points. The feature engineering involves extracting time-domain features, frequency-domain features, and fusion features. The time-domain features include the mean, peak value, kurtosis, and variance of the multi-dimensional operating data; the frequency-domain features are obtained by performing Fourier transform on the multi-dimensional operating data; and the fusion features include the ratio of wind speed to power and the rate of temperature change. The normalization process involves standardizing the feature-engineered data to the range of [0,1] or [-1,1] to obtain the final feature dataset.

10. The wind turbine generator abnormal state detection system based on extreme learning machine according to claim 7, characterized in that, The anomaly response module adopts a tiered alarm system, which is divided into orange and red alarms according to the severity of the anomaly. It further triggers local audible and visual alarms, remote pushes to the operation and maintenance platform, and automatically triggers shutdown protection. The log includes the time of anomaly occurrence, characteristic data, and anomaly judgment results.

Images