Wind turbine generator nonlinear process state monitoring method, system, equipment and medium

Abstract

The present application discloses a method, system, device and medium for monitoring the nonlinear process state of a wind turbine generator, and relates to the technical field of wind turbine monitoring. The method includes: collecting multi-dimensional operating state data of the wind turbine generator and performing preprocessing; extracting the normal operating data of the generator from the preprocessed data and performing normalization processing, using this data as a sample to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and perform model training; using the trained nonlinear state assessment model of the generator to perform real-time monitoring of the nonlinear process of the generator. The present application fuses SAE, which can effectively extract the nonlinear characteristics of process data, and CCA, which can effectively ensure the correlation between random variables, to construct a generator operating state assessment model based on SAE-CCA fusion, which can realize the effective assessment of the operating state of the wind turbine generator and improve the accuracy and reliability of the generator state assessment results.

Description

Technical Field

[0001] The present application relates to the technical field of wind turbine monitoring, and in particular to a method, system, device and medium for monitoring the nonlinear process state of a wind turbine. Background Art

[0002] Wind turbines can convert natural wind energy into mechanical energy, and then convert mechanical energy into electrical energy through the unit transmission chain (main shaft, gearbox and generator), thereby achieving the purpose of power generation. In order to fully and effectively utilize wind resources, wind turbines are generally installed in sparsely populated areas with complex environments but abundant wind resources, such as Gobi, desert, and sea. Generators are affected by complex and changeable environmental factors during the operation and power generation process, and their operating status is not easy for engineering and technical personnel to grasp in time. Once a fault occurs, the unit will be shut down for maintenance and repair at the least, or the unit will be damaged at the worst, causing serious economic losses.

[0003] In order to reduce unnecessary downtime and maintenance time caused by generator failures and improve power generation efficiency, the current common predictive maintenance method for generators is to conduct online status monitoring and fault diagnosis research on the driving end and free end of the generator.

[0004] Online monitoring is mainly achieved through the online condition monitoring system (CMS). By monitoring the real-time vibration of the generator, the status of the components is recorded and analyzed regularly, hidden dangers are discovered in time, the causes of the problems are found, and corresponding measures are taken to solve them. The condition monitoring system is mainly composed of CMS data acquisition station, vibration sensor, CMS field server and CMS remote server. The data acquisition station and vibration sensor are installed inside the wind turbine, responsible for collecting and processing vibration data, and uploading data to the CMS field server. The CMS field server is located in the wind farm monitoring center, and is mainly responsible for receiving the data uploaded by the acquisition station, judging whether the amplitude of the vibration vector exceeds the limit, and generating an alarm message if it exceeds the limit; running the visual interface program on the CMS field server to display the vibration of the unit in real time. The CMS remote server accesses the CMS field server through the Internet to obtain the vibration data of the rotating parts, and realizes remote monitoring of the vibration of the rotating parts of the wind turbine.

[0005] The fault diagnosis research mainly installs acceleration sensors on the driving end and free end of the wind turbine generator, sets sampling parameters, collects vibration data of the driving and free ends of the generator, and uses time domain analysis, spectrum analysis and envelope demodulation analysis to perform signal analysis on the collected vibration data. In combination with expert experience, characteristic signals are found to evaluate the operating status of key components in the generator, promptly identify signs of generator bearing failure, and grasp the gradual development degree of bearing failure in real time, thereby saving staff time for on-site troubleshooting, optimizing operation and maintenance strategies, and improving the operating reliability of the entire machine.

[0006] The above monitoring and diagnosis processes are based on the generator vibration data for evaluation, and the relationship between the generator operating state and the vibration signal variable is processed as a linear relationship. The vibration data of the generator is collected by the acceleration sensor, and then its operating state is monitored and evaluated based on the linear theory. However, in fact, due to the complex and changeable operating environment of the wind turbine, the generator system is not a single mechanical system during operation. Its operating state is not only affected by structural vibration factors, but also by a variety of environmental factors and equipment mechanical-electrical-hydraulic multi-scale coupling factors. Many process variables often show nonlinear correlation characteristics. There are certain limitations in using a single vibration signal analysis method to evaluate the generator operating state. There are certain differences between the fault diagnosis result identification and fault level judgment and the actual operating state of the unit generator. It is not only unable to effectively guide the on-site staff to implement reasonable operation and maintenance work on the unit, but may even cause excessive maintenance, bringing double losses to manpower and economy. Therefore, how to reveal the complex nonlinear relationship between the generator operating state and many process variables is the key to improving the reliability of nonlinear process state monitoring of wind turbines. Summary of the invention

[0007] In order to effectively monitor the nonlinear process of wind turbine generators and improve the accuracy of generator operation status assessment and judgment, the present application proposes a wind turbine generator nonlinear process state monitoring method, system, equipment and medium. Based on the real-time collected multi-dimensional state data of wind turbine generators, the present application uses stacked autoencoder (SAE) and canonical correlation analysis (CCA) methods to respectively realize the nonlinear characterization and correlation characterization of generator process data, establish a state analysis model based on SAE-CCA fusion and train the model, and mine the deep characteristics of its operation process for state assessment and analysis, which effectively improves the complex nonlinear relationship characterization capability between the state variables of wind turbine generators and realizes effective monitoring and evaluation of the generator operation status.

[0008] This application is implemented through the following technical solutions:

[0009] A method for monitoring the nonlinear process state of a wind turbine generator, the method comprising:

[0010] Collect multi-dimensional operating status data of wind turbine generators and pre-process the collected data;

[0011] Extract the normal operation data of the generator from the preprocessed data and perform normalization processing. Use the normalized data as samples to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and perform model training.

[0012] The trained generator nonlinear state assessment model is used to monitor the generator nonlinear process in real time.

[0013] In some embodiments, the collecting of multi-dimensional operating status data of wind turbine generators and preprocessing of the collected data specifically includes:

[0014] The vibration data of the generator drive end and free end are collected through the acceleration sensor;

[0015] The temperature data of the collector ring and brush surface at the rear side of the generator is collected by infrared thermometer;

[0016] The speed data of the generator input end is collected through the speed sensor;

[0017] The temperature data of the generator drive end bearing, the generator free end bearing and the generator coil are collected by temperature sensors;

[0018] The current data and voltage data of the generator rotor are collected respectively by a current sensor and a voltage sensor;

[0019] The temperature data, pressure data and oil quality data of the generator bearing lubricating oil are collected through the oil sensor;

[0020] The collected data is processed by outlier removal and data filtering.

[0021] In some embodiments, the generator nonlinear state assessment model establishment and training process based on SAE-CCA fusion includes:

[0022] Normalizing the extracted normal operation data of the generator;

[0023] The normal operation data of the generator after normalization is divided into independent variable data and dependent variable data;

[0024] Using the independent variable data to establish a SAE and train it to obtain SAE1, which is used to extract the representation of the process input and establish the input representation statistics;

[0025] Using the dependent variable data to establish a SAE and train it to obtain SAE2, which is used to extract the representation of the process output and establish the output representation statistics;

[0026] Based on the SAE1 and SAE2 and combined with CCA training, a state assessment model based on SAE-CCA fusion is obtained.

[0027] In some embodiments, the state assessment model based on SAE1 and SAE2 is obtained in combination with CCA training based on SAE-CCA fusion, which specifically includes independent learning and joint learning. The residual acquisition process of the two learning processes is:

[0028] Independent learning residuals: A fully connected neural network is added to the top layer of SAE1 and SAE2 to respectively realize the reconstruction of process input variables and process output variables, and two deep neural networks are established. On this basis, the reconstructed values ​​of input samples and output samples are obtained respectively, and then the reconstructed residuals of the input samples and output samples are obtained, and the residual statistics of the input and output are obtained according to the reconstructed residuals of the input samples and output samples respectively;

[0029] Joint learning residual: Add a fully connected neural network to the SAE1 and SAE2, extract deep correlation representations, add a CCA layer between the deep correlation representations and execute, and obtain the input and output paradigm vectors and correlation coefficient matrices.

[0030] In some implementations, the specific process of joint learning residuals includes:

[0031] Obtaining process input fault detection residuals related to process outputs and constructing monitoring statistics to detect faults in the residuals;

[0032] Obtaining process output fault detection residuals related to process inputs and constructing monitoring statistics to detect faults in the residuals;

[0033] Kernel density estimation is used to determine the thresholds of input and output characterization statistics, residual statistics, and monitoring statistics.

[0034] In some embodiments, the use of kernel density estimation to determine the thresholds of input and output characterization statistics, residual statistics, and monitoring statistics specifically includes:

[0035] Get various statistics of normal operation data of the generator;

[0036] The density of each statistic is calculated using a kernel density estimation method;

[0037] The value of the area where the statistic of 97% of the samples is located is determined as the threshold to control the false alarm rate within 3%.

[0038] In some embodiments, the generator nonlinear state assessment model obtained by training is used to perform real-time monitoring of the generator nonlinear process, specifically including:

[0039] Collect multi-dimensional operating status data of generators in real time and pre-process the collected data;

[0040] Standardize the real-time collected data after preprocessing;

[0041] The standardized real-time collected data is divided into independent variable data and dependent variable data and input into the trained generator nonlinear state assessment model to calculate the relevant statistics;

[0042] The statistic is compared with its corresponding threshold value. If it exceeds the threshold value, it is determined that the generator is faulty, otherwise the generator is operating normally.

[0043] In a second aspect, the present application proposes a wind turbine generator nonlinear process state monitoring system, the system comprising:

[0044] A data acquisition and processing module, which is used to collect multi-dimensional operating status data of wind turbine generators and pre-process the collected data;

[0045] A model building module, wherein the model building module extracts the normal operation data of the generator from the preprocessed data and performs normalization processing, uses the normalized data as a sample to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and performs model training;

[0046] And, a state assessment module, which uses the generator nonlinear state assessment model trained by the model construction module to perform real-time monitoring of the generator nonlinear process.

[0047] In a third aspect, the present application proposes a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0048] In a fourth aspect, the present application proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0049] The present application proposes a method, system, device and medium for monitoring the nonlinear process state of a wind turbine generator, which integrates SAE, which can effectively extract the nonlinear characteristics of process data, and CCA, which can effectively ensure the correlation between random variables, to construct a nonlinear operating state assessment model of a wind turbine generator based on SAE-CCA fusion. The model can be used to effectively assess the operating state of a wind turbine generator and improve the accuracy and reliability of the generator state assessment results.

[0050] The present application proposes a method, system, device and medium for monitoring the nonlinear process state of a wind turbine generator. The SAE is established based on the historical operation data of the wind turbine generator. The input variables and output variables are reconstructed by adding a fully connected neural network to obtain residuals and residual statistics, and the dominant information representation used to reconstruct the input and output is extracted. This process can effectively characterize the nonlinear characteristics between the input variables and the output variables, and is more consistent with the actual state of the generator; and CCA is performed between the above-mentioned deep representations, which can effectively explore the correlation between various state variables of the generator, realize comprehensive consideration and evaluation of multiple factors, and make the generator state evaluation result more accurate;

[0051] The present application proposes a method, system, device and medium for monitoring the nonlinear process state of a wind turbine generator, which realizes model training in an offline modeling process, and only stores variable functions instead of all reference samples during training and detection, so that online monitoring is more efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The drawings described herein are used to provide a further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0053] Figure 1 A flow chart of the state monitoring method proposed in the embodiment of the present application;

[0054] Figure 2 This is a state assessment model training diagram based on SAE-CCA fusion according to an embodiment of the present application;

[0055] Figure 3 A schematic diagram of a generator nonlinear state monitoring process according to an embodiment of the present application;

[0056] Figure 4 This is a principle block diagram of the state monitoring system proposed in the embodiment of the present application;

[0057] Figure 5 This is a schematic diagram of the installation of the generator vibration sensor;

[0058] Figure 6 This is a schematic diagram of fault data. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with examples and drawings. The illustrative implementation scheme of the present application and its description are only used to explain the present application and are not intended to limit the present application.

[0060] Example 1: Existing wind turbine generator state monitoring and evaluation technology usually uses a single vibration signal analysis method to evaluate the generator operating state, and does not consider the complex nonlinear relationship between the generator operating state and many process variables, resulting in differences between the fault diagnosis result identification and fault level determination and the actual operating state of the wind turbine generator. Not only can it not effectively guide on-site staff to implement reasonable operation and maintenance work on the unit, it may even cause excessive maintenance. In view of this, this embodiment proposes a nonlinear process state monitoring method for wind turbine generators. The method proposed in this embodiment integrates SAE, which can effectively extract nonlinear characteristics of process data, and CCA, which can effectively characterize the correlation between random variables, and adopts a nonlinear operating state evaluation technology for wind turbine generators based on SAE-CCA fusion to achieve effective monitoring and evaluation of the operating state of wind turbine generators.

[0061] like Figure 1 As shown, the method proposed in this embodiment specifically includes the following steps:

[0062] Step 100: Collect multi-dimensional operating status data of the wind turbine generator and pre-process the collected data.

[0063] In step 100, this embodiment uses an acceleration sensor to collect vibration data of the generator drive end and free end, and the sensor can be installed in the axial and / or radial position of the generator drive end bearing and the axial and / or radial position of the generator free end bearing; this embodiment uses an infrared thermometer to collect the surface temperature data of the collector ring (slip ring) and brush on the rear side of the generator; this embodiment uses a speed sensor to collect the speed data of the generator input end; this embodiment uses a temperature sensor to collect the generator bearing temperature and generator coil temperature data; this embodiment uses current and voltage sensors to collect current and voltage data of the generator rotor; this embodiment uses an oil sensor to collect the temperature, pressure and oil quality data of the generator bearing lubricating oil.

[0064] The preprocessing process mainly includes: data point removal processing and data filtering processing. Among them, this embodiment uses the Laida criterion to remove abnormal points in the above-mentioned collected data, and performs filtering and noise reduction processing by a composite filtering method. Among them, the filtering processing process is as follows:

[0065] Sort the collected data: x 1< x 2<… x N;

[0066] Processing results: x =( x 2+ x 3+… x N-1 ) / ( N -2). That is, first remove the minimum and maximum values, and then take the average of the remaining data.

[0067] Step 200, extracting the normal operation data of the generator from the preprocessed data and performing normalization processing, using this data as a sample to establish a generator nonlinear state assessment model based on SAE-CCA fusion and perform model training.

[0068] The specific process of step 200 is as follows:

[0069] Step 201, normalizing the extracted generator normal operation data, the normalization process is expressed as:

[0070]

[0071] in, is the sample data of the normal operation of the generator, is the minimum sample, is the maximum sample.

[0072] In step 202, the normalized generator normal operation data is divided into independent variable data (input data) and dependent variable data (output data), and a SAE is established using the independent variable data and trained to obtain SAE1 for extracting the representation of the process input, and a SAE is established using the dependent variable data and trained to obtain SAE2 for extracting the representation of the process output, thereby establishing the representation statistics of the input and output.

[0073] Specifically, the independent variable data and dependent variable data are defined as I and D , usage data I Establish a SAE and train it to obtain SAE1, which is used to extract the representation of the process input, denoted as f i ; Usage Data D Establish a SAE and train it to obtain SAE2, which is used to extract the representation of the process output, denoted as f d ; This establishes the input and output characterization statistics C They are:

[0074]

[0075] Among them, Σfi for The covariance of fd for The covariance of .

[0076] Step 203: Based on SAE1 and SAE2, and combined with CCA training, a state assessment model based on SAE-CCA fusion is obtained.

[0077] Specifically, the training of the state assessment model based on SAE-CCA fusion includes two learning processes, namely independent learning and joint learning, such as Figure 2 As shown in Figure 2, the residual acquisition process of the two learning processes is as follows:

[0078] (1) Independent learning residual: Add a fully connected neural network to the top layer of SAE1 and SAE2 to reconstruct the process input variables and output variables respectively, and establish two deep neural networks (DNNs). On this basis, the input samples can be obtained. ( i 1, i 2, ..., i m ) and output samples ( d 1, d 2, ..., d n ) ( , , …, )and ( , , …, ), and then obtain the reconstruction residuals of the input sample and the output sample, which are expressed as:

[0079]

[0080] in, is the reconstructed residual of the input sample, is the reconstructed residual of the output sample, m is the number of independent variables, n is the number of dependent variables.

[0081] The residual statistics can be constructed as:

[0082]

[0083] in, is the residual of the input sample Q Statistics, is the residual of the output sample QStatistics.

[0084] (2) Joint learning residual: A fully connected neural network is added to SAE1 and SAE2 to extract the deep correlation representation of the input variables and output variables respectively: zi ∈ℝ Si and zd ∈ℝ Sd , add the CCA layer between the deep correlation representations and execute it to obtain the input and output paradigm vectors and correlation coefficient matrices. The specific process is:

[0085] First, the process input fault detection residuals related to the process output are obtained, and the process input monitoring statistics are constructed to test the faults in the residuals. The relevant formula is as follows:

[0086] Process input fault detection residual: ;

[0087] Process input monitoring statistics: ,in, .

[0088] in, is the deep correlation representation of the input variables, is the depth-dependent representation of the output variable, is the input paradigm vector, is the output normal form vector, is the correlation coefficient matrix, For the order Si The identity matrix of .

[0089] Then, the process output fault detection residuals related to the process input are obtained, and the monitoring statistics are constructed to test the faults in the residuals. The relevant formula is as follows:

[0090] Process output fault detection residual: ;

[0091] Process output monitoring statistics: ,in, .

[0092] in, For the order Sd The identity matrix of .

[0093] Finally, the kernel density estimation (KDE) method is used to determine the statistics (including the input and output characterization statistics) , , the residual statistic between the input and output , And monitoring statistics of input and output , ). The normal operation data of the generators were selected to calculate the statistics, and the density of each statistic was calculated using KDE. The value of the area where the 97% sample statistics were located was determined as the threshold, and the false alarm rate was controlled within 3%.

[0094] Step 300: Use the trained generator nonlinear state assessment model to perform real-time monitoring of the generator nonlinear process.

[0095] In step 300, the generator nonlinear process monitoring process is specifically as follows:

[0096] First, the multi-dimensional operating status data of the generator is collected in real time, and the collected operating status data of the generator is preprocessed and standardized, and divided into independent variables and dependent variables. The specific data collection, preprocessing, and standardization processes are as described in the above steps 100 and 200, and will not be repeated here.

[0097] Then, the nonlinear state assessment model of the generator trained in step 300 is used to calculate relevant statistics, and the statistics are compared with the corresponding thresholds. If the threshold is exceeded, it is determined that the generator is faulty, otherwise the generator is operating normally. Figure 3 shown.

[0098] The method proposed in this embodiment integrates SAE, which can effectively extract the nonlinear characteristics of process data, and CCA, which can effectively characterize the correlation between random variables, to construct a wind turbine generator operating status assessment model based on SAE-CCA fusion. The model takes into account the nonlinear characteristics and correlation of various factors affecting the generator operating status. The model can realize effective assessment of the generator operating status and improve the accuracy and reliability of the generator status assessment results.

[0099] Based on the same technical concept as above, this embodiment also proposes a wind turbine generator nonlinear process state monitoring system, such as Figure 4 As shown, the system proposed in this embodiment includes:

[0100] The data acquisition and processing module is used to collect multi-dimensional operating status data of the wind turbine generator and pre-process the collected data.

[0101] The model building module extracts the normal operation data of the generator from the preprocessed data and performs normalization processing, uses this data as a sample to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and conducts model training.

[0102] And, a state assessment module, which uses the generator nonlinear state assessment model trained by the model building module to perform real-time monitoring of the generator nonlinear process.

[0103] It should be noted that the specific implementation process of each functional module in the system proposed in this embodiment is as described in the above steps 100 to 300, and will not be repeated here.

[0104] Example 2: This example verifies the effectiveness of the method or system proposed in Example 1. The specific process is as follows:

[0105] Collecting multi-dimensional operating status data of the generator: This embodiment uses four SW2005H acceleration sensors with a frequency range of 0.1~40kHz, which are installed on the axial direction of the generator drive end ( )、Generator drive end radial ( )、Axial direction of the free end of the generator ( ) and the radial direction of the free end of the generator ( ), such as Figure 5 As shown in the figure, the sensor sampling frequency is set to 25600Hz, the number of spectral lines is 102400, and the sampling time is 4s to collect the vibration acceleration signals at both ends of the generator; two non-contact infrared thermometers are used to collect the surface temperature data of the collector ring and brush at the rear side of the generator; a speed sensor is installed at the output end of the high-speed shaft of the gearbox to collect the speed data of the input end of the generator; three temperature sensors are used to collect the bearing temperature at the drive end and free end of the generator and the temperature data of the generator coil; current sensors and voltage sensors are used to collect the current and voltage data of the generator rotor respectively; oil sensors are used to measure the temperature, pressure and oil quality data of the lubricating oil of the generator bearing. 13 variables are measured and recorded, which are divided into independent variables and dependent variables.

[0106] Pre-process the collected data.

[0107] 1000 normal operation data and 1 fault data of the generator are extracted from the preprocessed data and normalized. The normal operation data are used to establish the SAE-CCA fusion generator nonlinear state assessment model and train the model.

[0108] Fault 1: The vibration amplitude of the generator drive end suddenly changes at about the 53rd sample point, such as Figure 6 After normalizing the fault data, the fault data is inserted into the generator nonlinear state assessment model based on SAE-CCA fusion, and the statistics are calculated and compared with the threshold. The results show that the data statistics here all exceed the threshold, indicating the presence of fault characteristics, verifying the effectiveness of the wind turbine generator nonlinear process monitoring method or system proposed in the above embodiment 1.

[0109] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0113] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the specific implementation method of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for monitoring the nonlinear process state of a wind turbine generator, characterized in that: The state monitoring method comprises: Collect multi-dimensional operating status data of wind turbine generators and pre-process the collected data; Extract the normal operation data of the generator from the preprocessed data and perform normalization processing. Use the normalized data as samples to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and perform model training. The trained generator nonlinear state assessment model is used to monitor the generator nonlinear process in real time; The establishment and training process of the generator nonlinear state assessment model based on SAE-CCA fusion includes: Normalizing the extracted normal operation data of the generator; The normal operation data of the generator after normalization is divided into independent variable data and dependent variable data; Using the independent variable data to establish a SAE and train it to obtain SAE1, which is used to extract the representation of the process input and establish the input representation statistics; Using the dependent variable data to establish a SAE and train it to obtain SAE2, which is used to extract the representation of the process output and establish the output representation statistics; Based on the SAE1 and SAE2, and combined with CCA training, a state assessment model based on SAE-CCA fusion is obtained; The state assessment model based on SAE1 and SAE2 and combined with CCA training is obtained based on SAE-CCA fusion, which specifically includes independent learning and joint learning. The residual acquisition process of the two learning processes is: Independent learning residuals: A fully connected neural network is added to the top layer of SAE1 and SAE2 to respectively realize the reconstruction of process input variables and process output variables, and two deep neural networks are established. On this basis, the reconstructed values ​​of input samples and output samples are obtained respectively, and then the reconstructed residuals of the input samples and output samples are obtained, and the residual statistics of the input and output are obtained according to the reconstructed residuals of the input samples and output samples respectively; Joint learning residual: Add a fully connected neural network to the SAE1 and SAE2, extract deep correlation representations, add a CCA layer between the deep correlation representations and execute, and obtain the input and output paradigm vectors and correlation coefficient matrices.

2. A method for monitoring the nonlinear process state of a wind turbine generator according to claim 1, characterized in that: The collecting of multi-dimensional operating status data of wind turbine generators and preprocessing of the collected data specifically include: The vibration data of the generator drive end and free end are collected through the acceleration sensor; The temperature data of the collector ring and brush surface at the rear side of the generator is collected by infrared thermometer; The speed data of the generator input end is collected through the speed sensor; The temperature data of the generator drive end bearing, the generator free end bearing and the generator coil are collected by temperature sensors; The current data and voltage data of the generator rotor are collected respectively by a current sensor and a voltage sensor; The temperature data, pressure data and oil quality data of the generator bearing lubricating oil are collected through the oil sensor; The collected data is processed by outlier removal and data filtering.

3. A method for monitoring the nonlinear process state of a wind turbine generator according to claim 1, characterized in that: The specific process of the joint learning residual includes: Obtaining process input fault detection residuals related to process outputs and constructing monitoring statistics to detect faults in the residuals; Obtaining process output fault detection residuals related to process inputs and constructing monitoring statistics to detect faults in the residuals; Kernel density estimation is used to determine the thresholds of input and output characterization statistics, residual statistics, and monitoring statistics.

4. A method for monitoring the nonlinear process state of a wind turbine generator according to claim 3, characterized in that: The kernel density estimation method is used to determine the thresholds of input and output characterization statistics, residual statistics, and monitoring statistics, specifically including: Get various statistics of normal operation data of the generator; The density of each statistic is calculated using a kernel density estimation method; The value of the area where the statistic of 97% of the samples is located is determined as the threshold to control the false alarm rate within 3%.

5. A method for monitoring the nonlinear process state of a wind turbine generator according to claim 4, characterized in that: The method of using the trained generator nonlinear state assessment model to perform real-time monitoring of the generator nonlinear process specifically includes: Collect multi-dimensional operating status data of generators in real time and pre-process the collected data; Standardize the real-time collected data after preprocessing; The standardized real-time collected data is divided into independent variable data and dependent variable data and input into the trained generator nonlinear state assessment model to calculate the relevant statistics; The statistic is compared with its corresponding threshold value. If it exceeds the threshold value, it is determined that the generator is faulty, otherwise the generator is operating normally.

6. A wind turbine generator nonlinear process state monitoring system, characterized in that: The system comprises: A data acquisition and processing module, which is used to collect multi-dimensional operating status data of wind turbine generators and pre-process the collected data; A model building module, wherein the model building module extracts the normal operation data of the generator from the preprocessed data and performs normalization processing, uses the normalized data as a sample to establish a nonlinear state assessment model of the generator based on SAE-CCA fusion and performs model training; and, a state assessment module, wherein the state assessment module uses the generator nonlinear state assessment model trained by the model building module to perform real-time monitoring of the generator nonlinear process; The establishment and training process of the generator nonlinear state assessment model based on SAE-CCA fusion includes: Normalizing the extracted normal operation data of the generator; The normal operation data of the generator after normalization is divided into independent variable data and dependent variable data; Using the independent variable data to establish a SAE and train it to obtain SAE1, which is used to extract the representation of the process input and establish the input representation statistics; Using the dependent variable data to establish a SAE and train it to obtain SAE2, which is used to extract the representation of the process output and establish the output representation statistics; Based on the SAE1 and SAE2, and combined with CCA training, a state assessment model based on SAE-CCA fusion is obtained; The state assessment model based on SAE1 and SAE2 and combined with CCA training is obtained based on SAE-CCA fusion, which specifically includes independent learning and joint learning. The residual acquisition process of the two learning processes is: Independent learning residuals: A fully connected neural network is added to the top layer of SAE1 and SAE2 to respectively realize the reconstruction of process input variables and process output variables, and two deep neural networks are established. On this basis, the reconstructed values ​​of input samples and output samples are obtained respectively, and then the reconstructed residuals of the input samples and output samples are obtained, and the residual statistics of the input and output are obtained according to the reconstructed residuals of the input samples and output samples respectively; Joint learning residual: Add a fully connected neural network to the SAE1 and SAE2, extract deep correlation representations, add a CCA layer between the deep correlation representations and execute, and obtain the input and output paradigm vectors and correlation coefficient matrices.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images