Fan component fault early warning method and system and computer equipment
By combining Gaussian hybrid model and convolutional neural network, the problem of long model learning time and high false alarm rate in the wind turbine fault warning method is solved, and accurate and efficient early warning of fan faults is achieved to ensure stable operation of the fan.
Patent Information
- Application Number
- CN202411193199.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-28
- Publication Date
- 2024-11-26
AI Technical Summary
The existing wind turbine fault warning methods have the problem of too long learning time, low learning efficiency, and cannot effectively deal with the diversity of fan operating conditions and noise interference, resulting in high false alarm rates.
Combining Gaussian mixed model and convolutional neural network, the fan data is classified through the Gaussian mixed clustering model, the working condition category is determined, and the corresponding convolutional neural network model is used to predict key measurement points data, and the data difference is calculated to judge the operating state. , reduce false alarms and missed reports.
It improves the accuracy and efficiency of fan fault warning, reduces the occurrence of false alarms, ensures the stable operation of the fan and reduces unexpected downtime.
Smart Images

Figure CN119025973A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of fault warning, and more specifically, to a method, system and computer equipment for warning fault of a fan component. Background Art
[0002] At present, deep neural network methods are used to predict the key measurement point data of large components of wind turbines for early warning. Gaussian mixture models are used to process complex information with multiple states of wind turbines, fit residuals and output a fault early warning method for large components of wind turbines based on degradation index. A fault early warning strategy based on the similarity of multi-dimensional features of a group is proposed by closely combining Gaussian mixture models, dynamic time warping and entropy weight algorithms to achieve fault prediction of large components. All of them have certain limitations. The monitoring model established by deep neural network algorithms has the problems of long model learning time and low learning efficiency, which is not suitable for online engineering applications. Most of the current high fault early warning models directly use the Gaussian mixture model (GMM) algorithm to perform outlier detection and clustering on the real-time monitoring data of large components, obtain the probability distribution and cluster center of each cluster, and then use the 3Sigma standard or other abnormal threshold standards to judge whether the on-site working conditions are abnormal. It does not take into account the diverse operating conditions of fans on site. The operating conditions are affected by wind speed and load size. The variable operating conditions lead to non-stationary and nonlinear real-time monitoring signals and more serious noise interference. The on-site operating data is complex and has missing values and abnormal values. The operating conditions are judged by the abnormal probability distribution of a single key measuring point data or multiple related measuring point data, the mean 3Sigma law and the set empirical threshold, which is prone to false alarms. Summary of the invention
[0003] In order to overcome the above-mentioned defects, the present invention provides a fan component fault warning method, system and computer equipment, which adopts a combination of Gaussian mixture model and convolutional neural network to monitor the actual operating status of the fan in real time, timely discover potential failure risks, and issue warnings, thereby improving the accuracy of fan component fault warnings and reducing the occurrence of fan false alarms.
[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:
[0005] The present invention provides a fan component failure early warning method, comprising:
[0006] Obtain real-time operating data of relevant measuring points and key measuring points of fan components within a preset period;
[0007] Input the real-time operation data of the relevant measuring points and key measuring points of the fan components within the preset time period into the trained Gaussian mixture clustering model to obtain the corresponding working condition category of the fan components at each moment;
[0008] Based on the working condition category corresponding to the fan component at each moment, a trained convolutional neural network model corresponding to the working condition category is selected, and real-time operating data of relevant measuring points and key measuring points of the fan component within a preset time period are input to obtain predicted operating data of the key measuring points of the fan component at each moment;
[0009] Calculate the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determine the operating state of the key measuring points of the fan components at each moment according to the difference;
[0010] Based on the operating status of the key measuring points of the fan components at each moment, the fault status of the key measuring points of the fan components within the preset time period is determined.
[0011] Preferably, obtaining a trained Gaussian mixture clustering model includes:
[0012] Obtain historical operation data of relevant measurement points and key measurement points of wind turbine components and build a historical operation data set;
[0013] Preprocessing the historical operation data set to obtain a preprocessed historical operation data set;
[0014] A K-means algorithm is used to aggregate the number of clusters of each preprocessed historical operation data set to obtain an optimal number of cluster centers of the historical operation data set;
[0015] The optimal number of cluster centers of the historical running data set is initialized and applied to the constructed Gaussian mixture clustering model. The preprocessed historical running data set is input for training until the preset number of training times is reached to obtain a trained Gaussian mixture clustering model.
[0016] Preferably, obtaining a trained convolutional neural network model includes:
[0017] Inputting the preprocessed historical operation data set into the trained Gaussian mixture clustering model for classification, and obtaining the operating condition category corresponding to each historical data of the fan component;
[0018] Aggregate historical operation data with the same operating condition category to obtain several historical operation data sets with different operating condition categories;
[0019] Several convolutional neural network models are established, and the historical data of a corresponding working condition category is input for training. The mean square error loss function is set for back propagation and optimization to obtain the trained convolutional neural network model corresponding to each working condition category.
[0020] Preferably, the convolutional neural network model includes an input layer, a first convolutional layer, a first activation layer, a first pooling layer, a second convolutional layer, a second activation layer, a second pooling layer, a third convolutional layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth activation layer, a fourth pooling layer, a fifth convolutional layer, a fifth activation layer, a fifth pooling layer, a sixth convolutional layer, a sixth activation layer, a sixth pooling layer, a fully connected layer and an output layer, which are connected in sequence.
[0021] Preferably, the formula of the mean square error loss function is:
[0022]
[0023] Where N is the number of input data tensors, y i is the regression prediction value of the i-th input data tensor, The actual value of the key measurement point in each row of the k-th input data tensor.
[0024] Preferably, calculating the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determining the operating state of the key measuring points of the fan components at each moment according to the difference comprises:
[0025] A residual threshold is set, and the residual of the key measuring point of the fan component at each moment is calculated based on the predicted operation data and real-time operation data of the key measuring point of the fan component at each moment, and the residual is compared with the residual threshold. If the residual of the key measuring point does not fall within the residual threshold range, the key measuring point is marked as an abnormal operating state, otherwise it is marked as a normal operating state.
[0026] Preferably, setting the residual threshold comprises:
[0027] Inputting the historical operation data set of each operating condition category into the corresponding trained convolutional neural network model to obtain the historical operation prediction values of the key measuring points in the historical operation data set of each operating condition category;
[0028] The historical operation data of the key measuring point in the historical operation data set of each working condition category is used as the historical operation target value of the key measuring point;
[0029] Subtract the historical operation prediction value of the key measuring point in the historical operation data set of each operating condition category from the historical operation target value to obtain the residual range data set of the key measuring point of each operating condition category;
[0030] The maximum and minimum values in the residual range data set of the key measuring points are used as the upper and lower limit residual values of the residual range to obtain the residual threshold of the working condition category.
[0031] Preferably, determining the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment includes:
[0032] The number of moments when the operating status of the key measuring points of the fan components within the preset time period is an abnormal operating state is counted, and the ratio is calculated with the total number of moments within the preset time period to obtain the abnormality rate of the key measuring points of the fan components within the preset time period; when the abnormality rate is greater than the preset abnormality rate threshold, it is determined that the key measuring points of the fan components within the preset time period are abnormal, and an early warning message is generated.
[0033] The present invention also includes a fan component failure early warning system for implementing the above method, comprising:
[0034] Data acquisition module, to obtain real-time operation data of relevant measuring points and key measuring points of fan components within a preset period;
[0035] The operating condition identification module inputs the real-time operating data of the relevant measuring points and key measuring points of the fan components within the preset period into the trained Gaussian mixture clustering model to obtain the operating condition category corresponding to the fan components at each moment;
[0036] The prediction operation module selects the trained convolutional neural network model corresponding to the working condition category of the fan component at each moment, inputs the real-time operation data of the relevant measuring points and key measuring points of the fan component within a preset time period, and obtains the prediction operation data of the key measuring points of the fan component at each moment;
[0037] A state monitoring module, which calculates the difference of the key measuring points of the fan components at each moment based on the predicted operation data and the real-time operation data of the key measuring points of the fan components at each moment, and determines the operation state of the key measuring points of the fan components at each moment according to the difference;
[0038] The fault detection module determines the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment.
[0039] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the above steps when executing the computer program.
[0040] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:
[0041] The present invention provides a method, system and computer equipment for early warning of faults of fan components. The method collects real-time operating data of fan components in a preset time period, classifies the data using a Gaussian mixture clustering model, determines the working condition of the fan at each moment, and improves the ability to identify the fan in different operating states; according to the working condition category, uses a corresponding convolutional neural network model to predict the operating data of key measuring points, so that fault prediction is more accurate and personalized; calculates the difference between the predicted data and the real-time data, and compares it with a set residual threshold to determine the operating state of the measuring point; real-time monitoring and fault early warning help to ensure the stable operation of the fan, reduce unexpected downtime, and improve the reliability of the fan operation; determines the fault state of the fan component within the preset time period according to the operating state of the measuring point; through dynamic monitoring and analysis of real-time data, it is possible to detect abnormalities in time and issue early warnings, reducing the underreporting and false alarms of potential faults. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] Figure 1 This is a flow chart of the fan component failure early warning method described in Example 1;
[0043] Figure 2 A schematic diagram of the operating condition categories after classification by the Gaussian mixture clustering model described in Example 2;
[0044] Figure 3 This is a schematic diagram of the structure of the fan component failure warning system described in Example 3;
[0045] Figure 4 This is a schematic diagram of the structure of a computer device described in Example 4. DETAILED DESCRIPTION
[0046] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;
[0047] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;
[0048] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0049] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0050] Example 1
[0051] This embodiment provides a fan component failure early warning method, such as Figure 1 As shown, including:
[0052] Obtain real-time operating data of relevant measuring points and key measuring points of fan components within a preset period;
[0053] Input the real-time operation data of the relevant measuring points and key measuring points of the fan components within the preset time period into the trained Gaussian mixture clustering model to obtain the corresponding working condition category of the fan components at each moment;
[0054] Based on the working condition category corresponding to the fan component at each moment, a trained convolutional neural network model corresponding to the working condition category is selected, and real-time operating data of relevant measuring points and key measuring points of the fan component within a preset time period are input to obtain predicted operating data of the key measuring points of the fan component at each moment;
[0055] Calculate the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determine the operating state of the key measuring points of the fan components at each moment according to the difference;
[0056] Based on the operating status of the key measuring points of the fan components at each moment, the fault status of the key measuring points of the fan components within the preset time period is determined.
[0057] The workflow of this embodiment is as follows: by collecting real-time operating data of fan components in a preset time period, the real-time operating data is classified using a Gaussian mixture clustering model to determine the operating conditions of the fan at each moment; according to the operating condition category, the corresponding convolutional neural network model is used to predict the operating data of key measuring points; the residual between the predicted data and the real-time data is calculated, and compared with the warning threshold to determine the operating status of the measuring point; according to the operating status of the measuring point, the fault status of the fan component within the preset time period is determined; the time cost of the convolutional neural network model can be effectively reduced, the limitations of the outlier detection method using the GMM model are avoided, and the efficiency and accuracy of fault warning are improved.
[0058] Example 2
[0059] This embodiment provides a fan component failure early warning method, such as Figure 1 As shown, including:
[0060] Obtain real-time operating data of relevant measuring points and key measuring points of fan components within a preset period;
[0061] The relevant measuring points and key measuring points of the fan components in this embodiment are shown in Table 1.
[0062]
[0063]
[0064] Input the real-time operation data of the relevant measuring points and key measuring points of the fan components within the preset time period into the trained Gaussian mixture clustering model to obtain the corresponding working condition category of the fan components at each moment;
[0065] Obtaining a trained Gaussian mixture clustering model includes:
[0066] Obtain historical operation data of relevant measurement points and key measurement points of wind turbine components and build a historical operation data set;
[0067] The historical operation data set includes: hydraulic pressure of the fan component, water cooling pressure of the fan component, temperature of the fan component, acceleration of the fan component, wind speed of the fan, power of the fan, voltage of the fan, current of the fan, speed of the fan generator and speed of the fan impeller;
[0068] The model is applied to extract the historical raw data of the measuring points collected by the wind turbine SCADA system for 1 to 3 years at intervals of 5 to 10 seconds, and then the historical data of the wind turbine under normal power generation status (the power measuring point data at the moment is greater than 0) and normal data communication status (the wind speed, power and other main measuring point data have changed within 10 minutes) are screened out according to the historical downtime records of the wind turbine. According to the historical raw data of the relevant measuring points and key measuring points, a two-dimensional data set is formed for each interval of time;
[0069] Preprocessing the historical operation data set to obtain a preprocessed historical operation data set;
[0070] Preprocessing the historical operation data set includes: performing data cleaning, data normalization and data partitioning operations on the historical operation data set in sequence;
[0071] The data cleaning operation includes: missing value filling, outlier detection and invalid data detection;
[0072] The purpose of data cleaning is to improve the quality of data, reduce the noise and deviation of model training, and improve the performance and accuracy of the model. This method uses the following cleaning steps for the original collected data in sequence:
[0073] Missing value filling: For missing values in historical data, use the previous moment data of each column to fill it.
[0074] Outlier detection: For outliers in the matrix data, the 3Sigma criterion is applied (based on the normal distribution principle, the probability that the data set is distributed in (μ-3σ,μ+3σ) is 0.9973, where μ is the mean of the data set and σ is the standard deviation of the data set) to detect outliers in the data set. If an abnormality is detected in the data at that moment, all the measurement point data at that moment will be deleted.
[0075] Invalid data detection: Use the data characteristics of each type of wind turbine measurement point to formulate detection rules based on the actual operating conditions, and detect invalid data in the data set. If the data at that moment is detected to be abnormal, all the measurement point data at that moment will be deleted. Data that is not outside this normal value range is an invalid value.
[0076] As shown in Table 2, the normal ranges of the physical data of the fan components are as follows:
[0077] Physical quantity Normal range unit temperature -30~200 Celsius Current 0~1000 install Voltage 0~1500 Volt power 0~1.1 times rated power kilowatt Acceleration 0~1 Gravity acceleration Hydraulic pressure 0~300 Pa Water cooling pressure 0~10 Pa Generator speed 0~3000 RPM Impeller speed 0~25 RPM Wind speed 0~60 m / s
[0078] The data normalization operation includes: performing data normalization operation using Z-Score normalization, and the formula of Z-Score normalization is:
[0079]
[0080] Among them, v i is the i-th data, v i ° is the normalized data of the ith data, is the data mean, S A is the data standard deviation, n is the number of data;
[0081] The data partitioning operation includes: using a holdout method to partition normal historical data and fault historical data of the fan components, and randomly sampling the normal historical data and the fault historical data to obtain a training set and a test set;
[0082] A K-means algorithm is used to aggregate the number of clusters of each preprocessed historical operation data set to obtain an optimal number of cluster centers of the historical operation data set;
[0083] Initialize the optimal number of cluster centers of the historical running data set and apply it to the constructed Gaussian mixture clustering model, input the preprocessed historical running data set for training until the preset number of training times is reached, and obtain the trained Gaussian mixture clustering model;
[0084] like Figure 2 As shown, it is the operating condition category after classification by the Gaussian mixture clustering model; based on the operating condition category corresponding to the fan component at each moment, the trained convolutional neural network model corresponding to the operating condition category is selected, and the real-time operating data of the relevant measuring points and key measuring points of the fan component within the preset time period are input to obtain the predicted operating data of the key measuring points of the fan component at each moment;
[0085] Obtaining a trained convolutional neural network model includes:
[0086] Inputting the preprocessed historical operation data set into the trained Gaussian mixture clustering model for classification, and obtaining the operating condition category corresponding to each historical data of the fan component;
[0087] Aggregate historical operation data with the same operating condition category to obtain several historical operation data sets with different operating condition categories;
[0088] Several convolutional neural network models are established, and the historical data of a corresponding working condition category is input for training. The mean square error loss function is set for back propagation and optimization to obtain the trained convolutional neural network model corresponding to each working condition category.
[0089] The convolutional neural network model of this embodiment includes an input layer, a first convolutional layer, a first activation layer, a first pooling layer, a second convolutional layer, a second activation layer, a second pooling layer, a third convolutional layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth activation layer, a fourth pooling layer, a fifth convolutional layer, a fifth activation layer, a fifth pooling layer, a sixth convolutional layer, a sixth activation layer, a sixth pooling layer, a fully connected layer and an output layer, which are connected in sequence. After multiple training experiments, it is obtained that under the existing wind turbine warning working condition data, the entire wind turbine component fault warning model is set with 6 layers of neural networks, and the regression model can obtain better results and the training and prediction time meet the application requirements of the field engineering. The entire convolutional neural network is set with 6 layers of convolutional layers and 6 layers of pooling layers. In order to improve the efficiency of model training, the input layer receives the input two-dimensional tensor data, and according to the number of characteristics of the wind turbine operation data, the input tensor is 36 columns and 36 rows, and the tensor data with less than 36 features is filled with 0. The convolution kernel uses a 2*2 tensor. Each convolution layer completes the convolution operation between the convolution kernel and the features from the previous layer output, and then inputs it to the pooling layer through the activation function of the activation layer. After the downsampling operation is completed in the pooling layer, it is output to the next convolution layer. After the entire convolutional neural network completes 6 convolution and pooling operations, the output of the last pooling layer is connected to the fully connected layer, and after activation by the activation function, the regression prediction value is obtained;
[0090] The parameters of the convolutional neural network model in this embodiment are:
[0091] Set the model learning rate to 1e-6;
[0092] Set the model batch sample size to: 36;
[0093] Set the number of epochs to: 200;
[0094] The model optimizer settings are: Nadam;
[0095] Activation function:
[0096] Pooling layer: Average Pooling method is used.
[0097] Loss function: Improved mean square error loss function, as follows:
[0098]
[0099] Where N is the number of input data tensors, y i is the predicted value of the ith input data tensor regression, The actual value of the key measurement point in each row of the k-th input data tensor.
[0100] Fully connected layer formula:
[0101] Among them, n is the number of neurons in the previous layer of the fully connected layer, x i The output tensor of the convolutional layer above the fully connected layer of the i-th data input tensor, w i The weight of the convolutional layer output tensor of the i-th data input tensor fully connected layer, b i The bias term of the convolutional layer output of the fully connected layer of the i-th data input tensor.
[0102] According to the relevant measurement point column data as input feature data, the key measurement point column data as the target value (label value), the new data sets gathered according to different operating conditions are read into the fan component fault warning model corresponding to each operating condition category for training. After the convolution layer, pooling layer, and fully connected layer, the predicted value is obtained to complete the forward propagation; the loss function is used to calculate the error between the sum of the target actual value of the number of input tensor rows and the product of the number of input tensor rows and the predicted value, and then the neural network back propagation and optimization algorithm Nadam are applied to complete the gradient update and optimize the parameters of the neural network.
[0103] Load the test set data into the trained model and parameters to test and verify its accuracy. If the accuracy does not reach the expected target value, the grid search parameter optimization and model parameter optimization in the above steps are required until the model reaches the expected target accuracy. If the prediction accuracy of the model on the test set reaches the expected target value, the trained fan component fault warning model for each category of each unit is saved for the model to be used online;
[0104] Calculate the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determine the operating state of the key measuring points of the fan components at each moment according to the difference;
[0105] Calculating the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determining the operating state of the key measuring points of the fan components at each moment according to the difference includes:
[0106] A residual threshold is set, and the residual of the key measuring point of the fan component at each moment is calculated based on the predicted operation data and real-time operation data of the key measuring point of the fan component at each moment, and the residual is compared with the residual threshold. If the residual of the key measuring point does not fall within the residual threshold range, the key measuring point is marked as an abnormal operating state, otherwise it is marked as a normal operating state.
[0107] Setting the residual threshold includes:
[0108] Inputting the historical operation data set of each operating condition category into the corresponding trained convolutional neural network model to obtain the historical operation prediction values of the key measuring points in the historical operation data set of each operating condition category;
[0109] The historical operation data of the key measuring point in the historical operation data set of each working condition category is used as the historical operation target value of the key measuring point;
[0110] Subtract the historical operation prediction value of the key measuring point in the historical operation data set of each operating condition category from the historical operation target value to obtain the residual range data set of the key measuring point of each operating condition category;
[0111] The maximum and minimum values in the residual range data set of the key measuring points are used as the upper and lower limit residual values of the residual range to obtain the residual threshold of the working condition category;
[0112] Determining the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment;
[0113] Determining the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment includes:
[0114] The number of moments when the operating status of the key measuring points of the fan components within the preset time period is an abnormal operating state is counted, and the ratio is calculated with the total number of moments within the preset time period to obtain the abnormality rate of the key measuring points of the fan components within the preset time period; when the abnormality rate is greater than the preset abnormality rate threshold, it is determined that the key measuring points of the fan components within the preset time period are abnormal, and an early warning message is generated.
[0115] Example 3
[0116] This embodiment provides a fan component failure warning system, which is used to implement the method of embodiment 1 or 2, such as Figure 3 As shown, including:
[0117] Data acquisition module, to obtain real-time operation data of relevant measuring points and key measuring points of fan components within a preset period;
[0118] The operating condition identification module inputs the real-time operating data of the relevant measuring points and key measuring points of the fan components within the preset period into the trained Gaussian mixture clustering model to obtain the operating condition category corresponding to the fan components at each moment;
[0119] The prediction operation module selects the trained convolutional neural network model corresponding to the working condition category of the fan component at each moment, inputs the real-time operation data of the relevant measuring points and key measuring points of the fan component within a preset time period, and obtains the prediction operation data of the key measuring points of the fan component at each moment;
[0120] A state monitoring module, which calculates the difference of the key measuring points of the fan components at each moment based on the predicted operation data and the real-time operation data of the key measuring points of the fan components at each moment, and determines the operation state of the key measuring points of the fan components at each moment according to the difference;
[0121] The fault detection module determines the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment.
[0122] Example 4
[0123] This embodiment provides a computer device, such as Figure 4 As shown, at least one processor 01 , at least one communication interface 02 , at least one memory 03 and at least one communication bus 04 .
[0124] In the embodiment of the present application, the number of the processor 01 , the communication interface 02 , the memory 03 , and the communication bus 04 is at least one, and the processor 01 , the communication interface 02 , and the memory 03 communicate with each other through the communication bus 04 .
[0125] Processor 01 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. Processor 01 may be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). Processor 01 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, processor 01 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, processor 01 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.
[0126] The memory 03 may include one or more computer-readable storage media, which may be non-transitory. The memory 03 may also include a high-speed random access memory and a non-volatile memory, such as one or more disk storage devices and flash memory storage devices.
[0127] The memory 03 stores a program, and the processor 01 can call the program stored in the memory 03, and the program is used to execute the steps of the fan component failure warning method described in Example 1 or 2.
[0128] The same or similar reference numerals correspond to the same or similar components;
[0129] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;
[0130] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.
Claims
1. A fan component failure early warning method, characterized in that: include: Obtain real-time operating data of relevant measuring points and key measuring points of fan components within a preset period; Input the real-time operation data of the relevant measuring points and key measuring points of the fan components within the preset time period into the trained Gaussian mixture clustering model to obtain the corresponding working condition category of the fan components at each moment; Based on the working condition category corresponding to the fan component at each moment, a trained convolutional neural network model corresponding to the working condition category is selected, and real-time operating data of relevant measuring points and key measuring points of the fan component within a preset time period are input to obtain predicted operating data of the key measuring points of the fan component at each moment; Calculate the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determine the operating state of the key measuring points of the fan components at each moment according to the difference; Based on the operating status of the key measuring points of the fan components at each moment, the fault status of the key measuring points of the fan components within the preset time period is determined.
2. The fan component failure early warning method according to claim 1 is characterized in that: Obtaining a trained Gaussian mixture clustering model includes: Obtain historical operation data of relevant measurement points and key measurement points of wind turbine components and build a historical operation data set; Preprocessing the historical operation data set to obtain a preprocessed historical operation data set; A K-means algorithm is used to aggregate the number of clusters of each preprocessed historical operation data set to obtain an optimal number of cluster centers of the historical operation data set; The optimal number of cluster centers of the historical running data set is initialized and applied to the constructed Gaussian mixture clustering model. The preprocessed historical running data set is input for training until the preset number of training times is reached to obtain a trained Gaussian mixture clustering model.
3. The fan component failure early warning method according to claim 2 is characterized in that: Obtaining a trained convolutional neural network model includes: Inputting the preprocessed historical operation data set into the trained Gaussian mixture clustering model for classification, and obtaining the operating condition category corresponding to each historical data of the fan component; Aggregate historical operation data with the same operating condition category to obtain several historical operation data sets with different operating condition categories; Several convolutional neural network models are established, and the historical data of a corresponding working condition category is input for training. The mean square error loss function is set for back propagation and optimization to obtain the trained convolutional neural network model corresponding to each working condition category.
4. The wind turbine component failure early warning method according to claim 3 is characterized in that: The convolutional neural network model includes an input layer, a first convolutional layer, a first activation layer, a first pooling layer, a second convolutional layer, a second activation layer, a second pooling layer, a third convolutional layer, a third activation layer, a third pooling layer, a fourth convolutional layer, a fourth activation layer, a fourth pooling layer, a fifth convolutional layer, a fifth activation layer, a fifth pooling layer, a sixth convolutional layer, a sixth activation layer, a sixth pooling layer, a fully connected layer and an output layer, which are connected in sequence.
5. The wind turbine component failure early warning method according to claim 3, characterized in that: The formula of the mean square error loss function is: Where N is the number of input data tensors, y i is the regression prediction value of the i-th input data tensor, The actual value of the key measurement point of each row of the k-th input data tensor.
6. The wind turbine component failure early warning method according to claim 1, characterized in that: Calculating the difference of the key measuring points of the fan components at each moment based on the predicted operating data and the real-time operating data of the key measuring points of the fan components at each moment, and determining the operating state of the key measuring points of the fan components at each moment according to the difference includes: A residual threshold is set, and the residual of the key measuring point of the fan component at each moment is calculated based on the predicted operation data and real-time operation data of the key measuring point of the fan component at each moment, and the residual is compared with the residual threshold. If the residual of the key measuring point does not fall within the residual threshold range, the key measuring point is marked as an abnormal operating state, otherwise it is marked as a normal operating state.
7. The wind turbine component failure early warning method according to claim 6, characterized in that: Setting the residual threshold includes: Inputting the historical operation data set of each operating condition category into the corresponding trained convolutional neural network model to obtain the historical operation prediction values of the key measuring points in the historical operation data set of each operating condition category; The historical operation data of the key measuring point in the historical operation data set of each working condition category is used as the historical operation target value of the key measuring point; Subtract the historical operation prediction value of the key measuring point in the historical operation data set of each operating condition category from the historical operation target value to obtain the residual range data set of the key measuring point of each operating condition category; The maximum and minimum values in the residual range data set of the key measuring points are used as the upper and lower limit residual values of the residual range to obtain the residual threshold of the working condition category.
8. The wind turbine component failure early warning method according to claim 6, characterized in that: Determining the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment includes: The number of moments when the operating status of the key measuring points of the fan components within the preset time period is an abnormal operating state is counted, and the ratio is calculated with the total number of moments within the preset time period to obtain the abnormality rate of the key measuring points of the fan components within the preset time period; when the abnormality rate is greater than the preset abnormality rate threshold, it is determined that the key measuring points of the fan components within the preset time period are abnormal, and an early warning message is generated.
9. A fan component failure warning system, used to implement the method of claims 1-8, characterized in that: include: Data acquisition module, to obtain real-time operation data of relevant measuring points and key measuring points of fan components within a preset period; The operating condition identification module inputs the real-time operating data of the relevant measuring points and key measuring points of the fan components within the preset period into the trained Gaussian mixture clustering model to obtain the operating condition category corresponding to the fan components at each moment; The prediction operation module selects the trained convolutional neural network model corresponding to the working condition category of the fan component at each moment, inputs the real-time operation data of the relevant measuring points and key measuring points of the fan component within a preset time period, and obtains the prediction operation data of the key measuring points of the fan component at each moment; A state monitoring module, which calculates the difference of the key measuring points of the fan components at each moment based on the predicted operation data and the real-time operation data of the key measuring points of the fan components at each moment, and determines the operation state of the key measuring points of the fan components at each moment according to the difference; The fault detection module determines the fault status of the key measuring points of the fan components within the preset time period based on the operating status of the key measuring points of the fan components at each moment.
10. 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 wind turbine component failure warning method described in any one of claims 1-8 are implemented.
Citation Information
Patent Citations
Draught fan fault recognition method and device
CN105863970A
System fault prediction method based on Gaussian mixture hidden Markov model
CN113342597A
Equipment working condition early warning method and related device
CN113743531A
Intelligent monitoring method for offshore wind turbine generator based on perception calculation collaboration
CN117989074A
Cross grouping pruning lightweight heterogeneous cloud load prediction method and system
CN118133107A
Cited By
Fault prediction method and device, electronic equipment, storage medium and program product
CN119312232A
Fault prediction method, device, electronic device, storage medium and program product
CN119312232B
Cylinder switching and bypass heat supply multi-mode working condition self-adaptive switching decision-making system and method
CN122106697A
