Method and system for detecting early deterioration state of main bearing of wind turbine generator

Abstract

The invention belongs to the technical field of wind power generation, and provides a wind turbine generator main bearing early degradation state detection method and system, and the method comprises the steps: obtaining a historical SCADA data set of a wind turbine generator in a normal operation state and an online SCADA data set of the wind turbine generator in a current operation state; respectively carrying out earlier stage processing to obtain a sample data set and a to-be-input data set; training and testing the multi-head self-attention neural network model through the sample data set to obtain a main bearing temperature prediction model; inputting a to-be-input data set into the main bearing temperature prediction model to obtain a temperature prediction value; according to the temperature predicted value and the temperature measured value, a reverse normalization prediction residual error is determined, and if the reverse normalization prediction residual error exceeds a set residual error threshold value, it is judged that the main bearing of the wind turbine generator set is in an early abnormal state; and carrying out sliding window abnormity statistics on the main bearing of the wind turbine generator in the early abnormal state to obtain an early degradation state detection result. And the detection precision and the feasibility of a detection link are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method and system for detecting an early degradation state of a main bearing of a wind turbine generator set. Background Art

[0002] The main shaft system of a wind turbine is a key component of a wind turbine. It primarily includes the cage, main shaft, inner and outer bearings, and rolling elements, among other components. It plays a vital role in converting wind energy into mechanical energy and transmitting mechanical energy. Due to the complex internal structure of the main shaft system, it is subjected to constant impact and alternating loads, resulting in frequent failures and long downtimes, which seriously restricts the operational stability of wind power systems.

[0003] In related technologies, there are two main approaches to detecting the condition of wind turbine main bearings: one is based on time-frequency analysis of the main bearing's axial or radial vibration signals, combined with expert experience to diagnose faults. Another approach uses the vibration signal's time-frequency characteristics or modal decomposition results to build a diagnostic model using traditional machine learning algorithms and shallow neural networks.

[0004] However, expert diagnosis methods based on vibration signals rely on specialized knowledge, are highly subjective, and have limited applicability. Traditional machine learning and shallow neural networks are insufficiently capable of identifying fault characteristics in vibration signals, and vibration signal acquisition requires additional hardware, increasing equipment installation and maintenance costs.

[0005] It can be seen that the traditional wind turbine main bearing status detection scheme has technical problems such as low detection accuracy and insufficient feasibility. Summary of the Invention

[0006] The present invention provides a method and system for detecting the early degradation state of a main bearing of a wind turbine generator set, which are used to solve the defects of low detection accuracy and insufficient feasibility of a traditional wind turbine generator set main bearing state detection scheme.

[0007] In one aspect, the present invention provides a method for detecting early degradation of a main bearing of a wind turbine generator set, comprising:

[0008] Obtain the historical SCADA data set of the wind turbine in normal operation and the online SCADA data set in the current operation state;

[0009] Preliminary processing is performed on the historical SCADA data set and the online SCADA data set to obtain a sample data set and a data set to be input;

[0010] The pre-built multi-head self-attention neural network model is trained and tested using the sample data set to obtain a main bearing temperature prediction model;

[0011] Inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model;

[0012] Determining an inverse normalized prediction residual based on the temperature prediction value and a pre-obtained temperature measured value, and determining that an early abnormal state occurs in a main bearing of the wind turbine generator set if the inverse normalized prediction residual exceeds a set residual threshold;

[0013] Sliding window abnormality statistics are performed on the main bearings of wind turbines that have early abnormal conditions to obtain early degradation state detection results.

[0014] According to the method for detecting the early degradation state of the main bearing of a wind turbine provided by the present invention, the historical SCADA data set is pre-processed to obtain a sample data set, including:

[0015] Performing data cleaning on the historical SCADA data set to obtain a SCADA healthy data set;

[0016] Performing feature extraction on the SCADA health data set to obtain model input feature variables;

[0017] Data standardization is performed on the model input feature variables to obtain a sample data set.

[0018] According to the method for detecting the early degradation state of the main bearing of a wind turbine provided by the present invention, data cleaning is performed on the historical SCADA data set to obtain a SCADA healthy data set, including:

[0019] The historical SCADA data set is initially cleaned using the quartile algorithm to obtain a preliminary healthy data set;

[0020] The preliminary health data set is cleaned twice using the quartile algorithm to obtain the SCADA health data set.

[0021] According to the method for detecting the early degradation state of the main bearing of a wind turbine provided by the present invention, feature extraction is performed on the SCADA health data set to obtain model input feature variables, including:

[0022] Preliminarily extracting state parameters associated with the main bearing temperature from the SCADA health data set based on a fusion grey correlation analysis algorithm to obtain preliminary characteristic variables;

[0023] A feature screening condition is constructed based on the correlation degree and the abnormal frequency, and the preliminary feature variables are secondary extracted according to the feature screening condition to obtain the model input feature variables.

[0024] According to the method for detecting early degradation of a main bearing of a wind turbine generator system provided by the present invention, the multi-head self-attention neural network model includes: a plurality of self-attention submodules;

[0025] Each self-attention submodule consists of a data input layer, a multi-head self-attention layer, a residual connection and normalization layer, and a fully connected layer.

[0026] According to the method for detecting the early degradation state of a main bearing of a wind turbine generator system provided by the present invention, before inputting the data set to be input into the main bearing temperature prediction model, the method further includes:

[0027] Obtaining actual scene information corresponding to the main bearing of the wind turbine to be tested;

[0028] Determining a startup frequency of the main bearing temperature prediction model based on the actual scenario information;

[0029] The main bearing temperature prediction model is controlled to start periodically according to the starting frequency.

[0030] According to the method for detecting the early degradation state of a main bearing of a wind turbine generator system provided by the present invention, after obtaining the main bearing temperature prediction model, the method further includes:

[0031] Regularly obtain the recent SCADA data set of wind turbines within a set time from the current moment;

[0032] The model parameters of the main bearing temperature prediction model are updated according to the recent SCADA data set.

[0033] According to the method for detecting the early degradation state of the main bearing of a wind turbine provided by the present invention, the set residual threshold is determined by the following process:

[0034] Denormalizing the temperature prediction value output by the main bearing temperature prediction model in the training and testing phases to obtain a denormalized temperature prediction value;

[0035] Subtracting the true value of each sample in the training and test phase sample data sets from the denormalized temperature prediction value to obtain a denormalized model prediction residual;

[0036] The kernel density estimation algorithm is used to perform probability density distribution analysis on the denormalized model prediction residuals to obtain probability density distribution data;

[0037] According to the probability density distribution data, the denormalized model prediction residual corresponding to the target confidence level is used as the residual threshold.

[0038] According to the method for detecting the early degradation state of the main bearing of a wind turbine generator set provided by the present invention, sliding window abnormality statistics are performed on the main bearing of the wind turbine generator set that has an early abnormal state to obtain the early degradation state detection result, including:

[0039] A sliding time window with a set width is used to collect statistics on the input data set of the main bearing of the wind turbine generator set with early abnormal conditions, and the amount of data with continuous abnormalities and the total amount of data within the sliding time window are determined;

[0040] Determine the proportion of continuous abnormal data in the sliding time window based on the amount of data with continuous abnormalities and the total amount of data in the sliding time window;

[0041] If the proportion of the continuous abnormal data exceeds a preset proportion threshold, the early degradation state detection result indicates that the main bearing is in an early degradation state.

[0042] On the other hand, the present invention also provides a system for detecting early degradation of a main bearing of a wind turbine generator set, comprising:

[0043] An acquisition module is used to obtain the historical SCADA data set of the wind turbine in normal operation and the online SCADA data set in the current operation state;

[0044] A processing module is used to perform preliminary processing on the historical SCADA data set and the online SCADA data set to obtain a sample data set and a data set to be input;

[0045] A modeling module is used to train and test a pre-built multi-head self-attention neural network model using the sample data set to obtain a main bearing temperature prediction model;

[0046] a prediction module, configured to input the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model;

[0047] an initial judgment module, configured to determine an inverse normalized prediction residual based on the temperature prediction value and a pre-obtained temperature measured value, and to determine that an early abnormal state occurs in the main bearing of the wind turbine generator set if the inverse normalized prediction residual exceeds a set residual threshold;

[0048] The detection module is used to perform sliding window abnormality statistics on the main bearings of wind turbines that have early abnormal conditions to obtain early degradation state detection results.

[0049] The present invention provides a method and system for detecting an early degradation state of a main bearing of a wind turbine generator set. The method and system respectively perform preliminary processing on a historical SCADA data set and an online SCADA data set to obtain a sample data set and a data set to be input; a pre-built multi-head self-attention neural network model is trained and tested using the sample data set to obtain a main bearing temperature prediction model; the data set to be input is input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; an inverse normalized prediction residual is determined based on the temperature prediction value and a pre-obtained actual temperature value; if the inverse normalized prediction residual exceeds a set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state; and sliding window abnormality statistics are performed on the main bearing of the wind turbine generator set that has an early abnormal state to obtain an early degradation state detection result. Since the detection link uses a multi-head self-attention neural network model to achieve accurate prediction of the main bearing temperature, and performs early abnormal state detection based on the denormalized prediction residual, and finally realizes the detection of early degradation state through sliding window anomaly statistics, the detection accuracy is improved. At the same time, the detection accuracy is improved. The entire detection link does not require the addition of auxiliary equipment. The detection can be achieved by using the SCADA data set obtained from the data acquisition and monitoring control system deployed by the wind farm itself. The detection process is easy to implement, which improves the feasibility of the detection link. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0051] Figure 1 1 is a flow chart of a method for detecting an early degradation state of a main bearing of a wind turbine generator set provided by an embodiment of the present invention;

[0052] Figure 2 This is a flow chart of the offline training phase of the model;

[0053] Figure 3 It is a flow chart of the online detection stage;

[0054] Figure 4 1 is a schematic structural diagram of a system for detecting early degradation of a main bearing of a wind turbine generator system provided by an embodiment of the present invention;

[0055] Figure 5 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0057] The following combination Figures 1 to 5 The present invention describes a detailed solution of a method and system for detecting early degradation of a main bearing of a wind turbine generator set provided by an embodiment of the present invention.

[0058] Figure 1 It is a flow chart of a method for detecting an early degradation state of a main bearing of a wind turbine generator set provided by an embodiment of the present invention.

[0059] like Figure 1 As shown, the method for detecting the early degradation state of the main bearing of a wind turbine provided by an embodiment of the present invention can be performed by a computer or server with data receiving and processing capabilities. The method mainly includes the following steps:

[0060] Step 110: Acquire a historical SCADA data set of the wind turbine in a normal operating state and an online SCADA data set in a current operating state.

[0061] In this embodiment, SCADA stands for Supervisory Control And Data Acquisition, which refers to a data acquisition and monitoring control system. In practical applications, a data acquisition and monitoring control system is usually deployed in a wind farm. The system can collect, store, and perform statistical analysis on the operating status parameters and environmental parameters of key equipment in the wind turbine in real time based on different sampling frequencies (such as seconds or minutes). Among them, the operating status parameters include temperature, pressure, speed, torque, power, current, and voltage. Environmental parameters include wind speed, wind direction, temperature, air pressure, and humidity.

[0062] The time series data in the SCADA dataset has the advantages of massive high-dimensionality, multi-source heterogeneity, and dynamic time-varying. It contains the complex nonlinear correlation and dynamic time series characteristics between the main bearing state parameters of wind turbines and is easy to obtain.

[0063] Step 120: Pre-process the historical SCADA dataset and the online SCADA dataset respectively to obtain a sample dataset and a dataset to be input.

[0064] In this embodiment, the preliminary processing specifically includes pre-processing processes such as data cleaning, feature extraction, and data standardization.

[0065] Step 130: Train and test the pre-built multi-head self-attention neural network model using the sample data set to obtain a main bearing temperature prediction model.

[0066] In this embodiment, the main bearing temperature prediction model based on the multi-head self-attention neural network can deeply mine the spatiotemporal characteristics contained in the time series data in the online SCADA data set, and can achieve accurate prediction of the main bearing temperature.

[0067] Step 140: inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model.

[0068] During the online detection stage, the main bearing temperature prediction model can be used to efficiently and accurately predict the main bearing temperature.

[0069] Step 150: Determine an inverse normalized prediction residual based on the temperature prediction value and the previously obtained temperature measured value. If the inverse normalized prediction residual exceeds a set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state.

[0070] This embodiment can effectively judge the early abnormal state of the main bearing of the wind turbine generator set by combining the denormalized prediction residual with the setting of the residual threshold.

[0071] Step 160: Perform sliding window abnormality statistics on the main bearings of the wind turbine generator set that have an early abnormal state to obtain an early degradation state detection result.

[0072] This embodiment uses a sliding window abnormality statistics strategy to avoid the problem of misjudging abnormal conditions caused by temperature jumps in the main bearings, thereby achieving accurate identification of the early degradation state of the main bearings of the wind turbine generator set.

[0073] In one embodiment, the historical SCADA data set is pre-processed to obtain a sample data set, including:

[0074] The first step is to clean the historical SCADA dataset to obtain the SCADA healthy dataset.

[0075] In a specific implementation, the historical SCADA dataset is cleaned to obtain a healthy SCADA dataset, which specifically includes:

[0076] First, the quartile algorithm is used to perform initial rough cleaning on the historical SCADA dataset to obtain a preliminary healthy dataset.

[0077] Then, the quartile algorithm is used to perform secondary fine cleaning on the preliminary health dataset to obtain the SCADA health dataset.

[0078] It can be understood that the quartile algorithm is a statistical method for identifying data distribution and outliers by sorting data by size and dividing it into four equal parts.

[0079] During the data cleaning phase, this embodiment first performs a rough cleansing of the historical SCADA dataset using the quartile algorithm to largely eliminate abnormal data caused by wind curtailment, power rationing, fluctuating operating conditions, and data collection equipment or communication line failures. The resulting preliminary healthy dataset is then subjected to a second fine cleansing using the quartile algorithm to obtain a healthy SCADA dataset.

[0080] This embodiment implements data cleaning through an improved progressive quartile algorithm, which can effectively reduce the interference of erroneous data and improve the data quality of the sample data set.

[0081] In some embodiments, clustering methods such as density-based spatial clustering algorithms or statistical methods such as the Thompson-tau method may be used to clean the historical SCADA data sets to improve data quality.

[0082] In the second step, feature extraction is performed on the SCADA health data set to obtain the model input feature variables.

[0083] In a specific implementation, feature extraction is performed on the SCADA health data set to obtain model input feature variables, including:

[0084] Firstly, the state parameters associated with the main bearing temperature in the SCADA health data set are preliminarily extracted based on the fusion grey relational analysis algorithm to obtain preliminary characteristic variables.

[0085] It can be understood that the grey relational analysis algorithm is a systematic analysis method that evaluates the strength of the correlation between factors by quantifying the similarity of the geometric shapes between sequences.

[0086] In practical applications, the grey correlation analysis algorithm can be integrated with the experience of site operation and maintenance experts to perform feature selection on nearly 100 state parameters related to wind turbines, and obtain state parameters with a strong correlation with the main bearing temperature as preliminary feature variables.

[0087] Then, feature screening conditions are constructed based on the correlation degree and abnormal frequency, and the preliminary feature variables are secondary extracted according to the feature screening conditions to obtain the model input feature variables.

[0088] In the secondary extraction stage, based on the experience of operation and maintenance experts, state parameters with high correlation but prone to abnormalities in the preliminary feature variables can be deleted, such as gearbox oil pool temperature, gearbox front / rear bearing temperature, and generator set front / rear bearing temperature, etc., thereby reducing the impact of abnormal fluctuations in the status of other equipment, and ultimately achieving the goal of reducing model complexity and improving model training efficiency.

[0089] That is to say, in this embodiment, high correlation and high abnormal frequency can be used as feature screening conditions to achieve secondary extraction of preliminary feature variables.

[0090] In some embodiments, correlation analysis algorithms such as Pearson and Spearman may also be used to complete feature extraction operations to achieve the effect of feature selection.

[0091] The third step is to perform data standardization on the model input feature variables to obtain a sample data set.

[0092] In practical applications, the model input feature variables can be normalized to their maximum and minimum values, and the values ​​of different variables can be scaled to the range of [0, 1] to eliminate the dimensional differences between different state parameters in the historical SCADA data set.

[0093] In one embodiment, the residual threshold may be determined by the following process:

[0094] First, the temperature prediction values ​​output by the main bearing temperature prediction model in the training and testing phases are denormalized to obtain the denormalized temperature prediction values.

[0095] Then, the true value of each sample in the sample data set of the training and testing phase is subtracted from the denormalized temperature prediction value to obtain the denormalized model prediction residual.

[0096] Subsequently, the kernel density estimation algorithm is used to perform probability density distribution analysis on the denormalized model prediction residuals to obtain probability density distribution data.

[0097] Finally, based on the probability density distribution data, the corresponding denormalized model prediction residual at the target confidence level is used as the residual threshold.

[0098] It can be understood that the kernel density estimation algorithm is a non-parametric statistical method used to estimate the probability density function of a random variable.

[0099] In this embodiment, data normalization is introduced during the initial data processing phase to reduce model complexity, accelerate model convergence, and improve the efficiency of model training and testing. Therefore, during model training and testing, the temperature prediction values ​​output by the model need to be denormalized to obtain state parameters with different value ranges and physical meanings. The denormalized model prediction residuals are then determined, and residual analysis is performed using a kernel density estimation algorithm to obtain set residual thresholds at different confidence levels.

[0100] In practical applications, the inverse normalized model prediction residual corresponding to the 95.45% confidence level can be set as the residual threshold to achieve a large-scale preliminary judgment of early abnormal conditions. The inverse normalized model prediction residual corresponding to the 99.73% confidence level can also be set as the residual threshold to achieve a small-scale refined judgment of early abnormal conditions, so as to achieve accurate detection of early abnormal conditions of the main bearings of wind turbines.

[0101] It can be understood that the method for detecting the early degradation state of the main bearing of a wind turbine provided in this embodiment requires the cooperation of a machine learning model and can be divided into two stages: an offline model training and testing stage and an online detection stage.

[0102] like Figure 2 As shown in the figure, the offline training phase of the model specifically includes the following processes:

[0103] Step 210: Obtain historical SCADA data sets.

[0104] Step 220: Use the quartile algorithm to perform data cleaning, specifically implementing the above data cleaning process through the improved progressive quartile algorithm.

[0105] Step 230: Fusing the grey relational analysis algorithm with expert experience to perform feature selection, specifically implementing feature extraction through the above-mentioned feature extraction process.

[0106] Step 240: Data maximum and minimum normalization, that is, achieving data standardization through data maximum and minimum normalization processing.

[0107] Step 250: Divide the sample data set obtained after the above preliminary processing into a test set and a training set. Specifically, the division can be performed according to a preset ratio. The training set is used for model training, and the test set is used for model testing.

[0108] Step 260: A main bearing temperature prediction model is obtained through training and testing. The main bearing temperature prediction model uses a multi-head self-attention neural network model as its main architecture.

[0109] Step 270: Denormalization of the prediction residuals. Denormalization can be performed to obtain denormalized model prediction residuals.

[0110] Step 280: In the residual analysis phase based on kernel density estimation, residual analysis can be performed based on a kernel density estimation algorithm.

[0111] Step 290: After performing residual analysis based on the kernel density estimation algorithm, set residual thresholds at different confidence levels can be obtained.

[0112] In one embodiment, the multi-head self-attention neural network model specifically includes: multiple self-attention sub-modules.

[0113] Each self-attention submodule consists of a data input layer, a multi-head self-attention layer, a residual connection and normalization layer, and a fully connected layer.

[0114] Among them, multiple self-attention sub-modules are set up in parallel in the multi-head self-attention neural network model, which can capture the dependencies between different positions in the sequence respectively, and can enhance the model's ability to learn complex patterns.

[0115] In one embodiment, before inputting the data set to be input into the main bearing temperature prediction model, the above method may further include:

[0116] First, the actual scene information corresponding to the main bearing of the wind turbine to be tested is obtained.

[0117] In this embodiment, the actual scenario information specifically includes information such as the layout area of ​​the wind turbine, the station to which it belongs, manufacturer information, model information, actual operating status, and historical health status.

[0118] Then, the starting frequency of the main bearing temperature prediction model is determined based on the actual scenario information.

[0119] In this embodiment, the startup frequency of the main bearing temperature prediction model can be reasonably set according to the detection requirements of the wind turbine generator system, for example, it can be set to start every ten minutes, thirty minutes or one hour, so as to achieve the purpose of balancing computing resources and abnormality detection accuracy.

[0120] Finally, the main bearing temperature prediction model is started periodically according to the start-up frequency.

[0121] This embodiment can provide a more targeted automatic detection function by setting the startup frequency, and can better meet the anomaly detection needs in different application scenarios.

[0122] In one embodiment, after obtaining the main bearing temperature prediction model, the above method may further include:

[0123] First, the recent SCADA dataset of wind turbines within a set time period from the current moment is regularly obtained.

[0124] Then, the model parameters of the main bearing temperature prediction model are updated based on the recent SCADA data set.

[0125] In this embodiment, the SCADA data within the last 30 days can be used to regularly establish a recent SCADA data set, and the model parameters of the online main bearing temperature prediction model can be fine-tuned to ensure that the main bearing temperature prediction model is always in the best performance state.

[0126] In one embodiment, a sliding window abnormality statistics is performed on the main bearing of the wind turbine generator set that has an early abnormal state to obtain an early degradation state detection result, specifically including:

[0127] Firstly, a sliding time window with a set width is used to count the input data set of the main bearing of the wind turbine with early abnormal state, and the amount of data with continuous abnormalities and the total amount of data within the sliding time window are determined.

[0128] Then, based on the amount of data with continuous anomalies and the total amount of data in the sliding time window, the proportion of continuous abnormal data in the sliding time window is determined.

[0129] Finally, if the proportion of continuous abnormal data exceeds the preset proportion threshold, the early degradation state detection result is that the main bearing has an early degradation state.

[0130] This embodiment uses a sliding time window with a set width to perform abnormality statistical analysis, and can decide whether to trigger an early warning of the abnormal state of the main bearing based on the proportion of continuous abnormal data in the window.

[0131] In practical applications, the set width of the sliding time window can be set from 1 minute to 5 minutes, and the specific value can be reasonably determined based on the detection accuracy requirements.

[0132] In some embodiments, an on-site warning can be achieved by emitting sound and light prompt information on site, or a remote warning can be achieved by sending prompt information to the terminal device of the supervisor.

[0133] like Figure 3 As shown in the figure, the online detection stage specifically includes the following processes:

[0134] Step 310: Obtain online SCADA data set.

[0135] Step 320: Perform data preprocessing, feature selection, and data normalization on the online SCADA dataset. This step is generally consistent with the pre-processing of the historical SCADA dataset and can be achieved through data cleaning, feature extraction, and data normalization operations.

[0136] Step 330: After the preliminary processing, the obtained data set to be input may be input into the main bearing temperature prediction model, and the main bearing temperature prediction model is used to predict the main bearing temperature to obtain a temperature prediction value.

[0137] Step 340: Denormalizing the prediction residual. The denormalized prediction residual can be determined based on the predicted temperature value and the previously obtained measured temperature value. Specifically, the predicted temperature value can be denormalized to obtain a denormalized predicted temperature value. The denormalized predicted temperature value can then be subtracted from the measured temperature value to obtain the denormalized prediction residual.

[0138] Step 350: In the early abnormal state judgment link, the abnormality judgment result can be obtained by judging whether it exceeds the threshold. Specifically, if the inverse normalized prediction residual exceeds the set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state; if the inverse normalized prediction residual does not exceed the set residual threshold, it is determined that the main bearing of the wind turbine generator set is normal.

[0139] Step 360: During the degradation state detection phase, a sliding window anomaly statistics system can be used to determine whether a trigger has occurred, thereby enabling early degradation state detection and early warning. Specifically, if the proportion of consecutive abnormal data exceeds a preset threshold, it indicates that the main bearing is experiencing early degradation. If this condition is met, the early warning mechanism is triggered. Otherwise, no warning is triggered.

[0140] In summary, the method for detecting the early degradation state of the main bearing of a wind turbine provided in this embodiment can realize automatic and accurate identification of the degradation state of the main bearing of a wind turbine, so that supervisors can take timely measures to avoid further deterioration of the equipment state, thereby improving the safety and reliability of wind turbine operation.

[0141] Based on the same general inventive concept, the present invention also protects a system for detecting the early degradation state of the main bearing of a wind turbine generator set. The system for detecting the early degradation state of the main bearing of a wind turbine generator set provided by the present invention is described below. The system for detecting the early degradation state of the main bearing of a wind turbine generator set described below and the method for detecting the early degradation state of the main bearing of a wind turbine generator set described above can be referred to each other.

[0142] like Figure 4 As shown, the wind turbine main bearing early degradation state detection system provided by the embodiment of the present invention specifically includes:

[0143] An acquisition module 410 is configured to acquire a historical SCADA data set of a wind turbine in a normal operating state and an online SCADA data set in a current operating state;

[0144] The processing module 420 is used to perform preliminary processing on the historical SCADA data set and the online SCADA data set to obtain a sample data set and a data set to be input;

[0145] A modeling module 430 is used to train and test a pre-built multi-head self-attention neural network model using a sample data set to obtain a main bearing temperature prediction model;

[0146] A prediction module 440 is configured to input the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model;

[0147] The initial judgment module 450 is used to determine the inverse normalized prediction residual based on the temperature prediction value and the previously obtained temperature measured value. If the inverse normalized prediction residual exceeds a set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state;

[0148] The detection module 460 is used to perform sliding window abnormality statistics on the main bearings of the wind turbine generator set that have an early abnormal state to obtain an early degradation state detection result.

[0149] Regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated again here.

[0150] Figure 5 It is a structural diagram of an electronic device provided by an embodiment of the present invention.

[0151] like Figure 5As shown, the electronic device may include: a processor (processor) 510, a communication interface (Communications Interface) 520, a memory (memory) 530 and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call the logic instructions in the memory 530 to execute the early degradation state detection method of the main bearing of the wind turbine generator set, which includes: obtaining a historical SCADA data set of the wind turbine generator set in normal operating state and an online SCADA data set in current operating state; performing pre-processing on the historical SCADA data set and the online SCADA data set respectively to obtain a sample data set and a data set to be input; training and testing a pre-built multi-head self-attention neural network model through the sample data set to obtain a main bearing temperature prediction model; inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; determining the inverse normalized prediction residual based on the temperature prediction value and the pre-obtained temperature measured value; if the inverse normalized prediction residual exceeds the set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state; performing sliding window abnormality statistics on the main bearing of the wind turbine generator set that has an early abnormal state to obtain an early degradation state detection result.

[0152] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0153] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute a method for detecting the early degradation state of the main bearing of a wind turbine generator set, the method including: obtaining a historical SCADA data set of the wind turbine generator set in a normal operating state and an online SCADA data set in a current operating state; performing preliminary processing on the historical SCADA data set and the online SCADA data set respectively to obtain a sample data set and a data set to be input; training and testing a pre-constructed multi-head self-attention neural network model through the sample data set to obtain a main bearing temperature prediction model; inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; determining an inverse normalized prediction residual based on the temperature prediction value and the pre-obtained actual temperature value; if the inverse normalized prediction residual exceeds a set residual threshold, it is determined that the main bearing of the wind turbine generator set has an early abnormal state; performing sliding window abnormality statistics on the main bearing of the wind turbine generator set that has an early abnormal state to obtain an early degradation state detection result.

[0154] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for detecting an early degradation state of a main bearing of a wind turbine generator set, the method comprising: obtaining a historical SCADA data set of the wind turbine generator set in a normal operating state and an online SCADA data set in a current operating state; performing preliminary processing on the historical SCADA data set and the online SCADA data set respectively to obtain a sample data set and a data set to be input; training and testing a pre-constructed multi-head self-attention neural network model using the sample data set to obtain a main bearing temperature prediction model; inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; determining an inverse normalized prediction residual based on the temperature prediction value and the pre-obtained actual temperature value, and if the inverse normalized prediction residual exceeds a set residual threshold, it is determined that an early abnormal state has occurred in the main bearing of the wind turbine generator set; performing sliding window abnormality statistics on the main bearing of the wind turbine generator set that has occurred an early abnormal state to obtain an early degradation state detection result.

[0155] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0156] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0157] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting early degradation of a main bearing of a wind turbine generator set, characterized in that: include: Obtain the historical SCADA data set of the wind turbine in normal operation and the online SCADA data set in the current operation state; Preliminary processing is performed on the historical SCADA data set and the online SCADA data set to obtain a sample data set and a data set to be input; The pre-built multi-head self-attention neural network model is trained and tested using the sample data set to obtain a main bearing temperature prediction model; Inputting the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; Determining an inverse normalized prediction residual based on the temperature prediction value and a pre-obtained temperature measured value, and determining that an early abnormal state occurs in a main bearing of the wind turbine generator set if the inverse normalized prediction residual exceeds a set residual threshold; Sliding window abnormality statistics are performed on the main bearings of wind turbines that have early abnormal conditions to obtain early degradation state detection results.

2. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1, characterized in that: The historical SCADA data set is pre-processed to obtain a sample data set, including: Performing data cleaning on the historical SCADA data set to obtain a SCADA healthy data set; Performing feature extraction on the SCADA health data set to obtain model input feature variables; Data standardization is performed on the model input feature variables to obtain a sample data set.

3. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 2, wherein: The historical SCADA dataset is cleaned to obtain a SCADA healthy dataset, including: The historical SCADA data set is initially cleaned using the quartile algorithm to obtain a preliminary healthy data set; The preliminary health data set is cleaned twice using the quartile algorithm to obtain the SCADA health data set.

4. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 2, wherein: Feature extraction is performed on the SCADA health data set to obtain model input feature variables, including: Preliminarily extracting state parameters associated with the main bearing temperature from the SCADA health data set based on a fusion grey correlation analysis algorithm to obtain preliminary characteristic variables; A feature screening condition is constructed based on the correlation degree and the abnormal frequency, and the preliminary feature variables are secondary extracted according to the feature screening condition to obtain the model input feature variables.

5. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1, wherein: The multi-head self-attention neural network model includes: multiple self-attention submodules; Each self-attention submodule consists of a data input layer, a multi-head self-attention layer, a residual connection and normalization layer, and a fully connected layer.

6. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1, wherein: Before inputting the to-be-input data set into the main bearing temperature prediction model, the method further includes: Obtaining actual scene information corresponding to the main bearing of the wind turbine to be tested; Determining a startup frequency of the main bearing temperature prediction model based on the actual scenario information; The main bearing temperature prediction model is controlled to start periodically according to the starting frequency.

7. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1 or 6, characterized in that: After obtaining the main bearing temperature prediction model, the method further includes: Regularly obtain the recent SCADA data set of wind turbines within a set time from the current moment; The model parameters of the main bearing temperature prediction model are updated according to the recent SCADA data set.

8. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1, wherein: The residual threshold is determined by the following process: Denormalizing the temperature prediction value output by the main bearing temperature prediction model in the training and testing phases to obtain a denormalized temperature prediction value; Subtracting the true value of each sample in the training and test phase sample data sets from the denormalized temperature prediction value to obtain a denormalized model prediction residual; The kernel density estimation algorithm is used to perform probability density distribution analysis on the denormalized model prediction residuals to obtain probability density distribution data; According to the probability density distribution data, the denormalized model prediction residual corresponding to the target confidence level is used as the residual threshold.

9. The method for detecting early degradation of a main bearing of a wind turbine generator set according to claim 1, characterized in that: Sliding window abnormality statistics are performed on the main bearings of wind turbines that have early abnormal conditions to obtain early degradation status detection results, including: A sliding time window with a set width is used to collect statistics on the input data set of the main bearing of the wind turbine generator set with early abnormal conditions, and the amount of data with continuous abnormalities and the total amount of data within the sliding time window are determined; Determine the proportion of continuous abnormal data in the sliding time window based on the amount of data with continuous abnormalities and the total amount of data in the sliding time window; If the proportion of the continuous abnormal data exceeds a preset proportion threshold, the early degradation state detection result indicates that the main bearing is in an early degradation state.

10. A system for detecting early degradation of main bearings of a wind turbine generator set, characterized in that: include: An acquisition module is used to obtain the historical SCADA data set of the wind turbine in normal operation and the online SCADA data set in the current operation state; A processing module is used to perform preliminary processing on the historical SCADA data set and the online SCADA data set to obtain a sample data set and a data set to be input; A modeling module is used to train and test a pre-built multi-head self-attention neural network model using the sample data set to obtain a main bearing temperature prediction model; a prediction module, configured to input the data set to be input into the main bearing temperature prediction model to obtain a temperature prediction value output by the main bearing temperature prediction model; an initial judgment module, configured to determine an inverse normalized prediction residual based on the temperature prediction value and a pre-obtained temperature measured value, and to determine that an early abnormal state occurs in the main bearing of the wind turbine generator set if the inverse normalized prediction residual exceeds a set residual threshold; The detection module is used to perform sliding window abnormality statistics on the main bearings of wind turbines that have early abnormal conditions to obtain early degradation state detection results.

Images