Fan gearbox temperature prediction method and device and storage medium
By using the feature extraction layer of the pre-trained neural network model to construct the fan gearbox temperature prediction model and using the temperature value sequence difference training, the local minimization problem is solved and the accuracy of temperature prediction is improved.
Patent Information
- Application Number
- CN202310502192.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-06
- Publication Date
- 2024-11-08
AI Technical Summary
The existing fan gearbox temperature prediction model training methods are prone to local minimization, resulting in poor parameter optimization effect and affecting the accuracy of temperature prediction.
The feature extraction layer is extracted by a pre-trained neural network model, a prediction model is constructed, and the difference between the target sample temperature value sequence and the actual temperature value sequence is trained to avoid the local minimization problem caused by random initial hyperparameters.
The training effect and accuracy of the fan gearbox temperature prediction model is improved, local minimization is avoided, and prediction accuracy is improved.
Smart Images

Figure CN118916690A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device and storage medium for predicting the temperature of a fan gearbox. Background Art
[0002] As global warming further intensifies, clean energy such as wind energy and solar energy can significantly reduce greenhouse gas emissions and slow down the process of global warming. In recent years, my country's wind power industry has developed rapidly, and the high failure rate of wind turbines has become a major shackle restricting the further development of the wind power industry. Among the wind turbine failures, gearbox failure is the most common. During the operation of the wind turbine, the operating temperature of the wind turbine gearbox is an important indicator for monitoring the health status of the wind turbine, which can reflect the operating status of the wind turbine. For example, if the wind turbine gearbox operates under abnormal temperature conditions, the viscosity of the lubricating oil inside the wind turbine gearbox will decrease, which will in turn increase the wear of the components during the operation of the wind turbine gearbox, and then continue to cause abnormal temperature rise, which will eventually cause serious damage to the components. At the least, the machine will be shut down for maintenance, resulting in power generation loss, and in severe cases, it may cause huge economic losses to the enterprise and even endanger the lives of the operation and maintenance personnel. Therefore, prior maintenance and processing of the wind turbine gearbox equipment can timely and efficiently detect the potential faults in the mechanical equipment and prevent the equipment failure from aggravating.
[0003] In the related art, the constructed prediction model is usually trained so as to use the trained prediction model for identification. However, in the process of adjusting parameters in the existing model training method, local minimization is prone to occur, and local minimization will seriously affect the parameter optimization process, resulting in poor model training effect. Summary of the invention
[0004] The present application aims to solve one of the technical problems in the related art at least to some extent.
[0005] To this end, the present application proposes a fan gearbox temperature prediction method, device and storage medium to improve the accuracy of model training, thereby improving the accuracy of temperature prediction.
[0006] In one aspect, an embodiment of the present application proposes a model training method, wherein the model is used to predict the temperature of a wind turbine gearbox, and the method comprises:
[0007] Obtaining a first feature extraction layer extracted from a pre-trained first neural network model and a second feature extraction layer extracted from a pre-trained second neural network model;
[0008] Building a prediction model based on the first feature extraction layer and the second feature extraction layer;
[0009] Acquire a target sample temperature value sequence and a corresponding actual temperature value sequence; wherein a sampling time corresponding to at least one temperature value in the actual temperature value sequence is after a sampling time corresponding to at least one temperature value in the target sample temperature value sequence;
[0010] Inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain a predicted temperature value sequence output by the prediction model;
[0011] The prediction model is trained according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
[0012] Another aspect of the present application provides a method for predicting the temperature of a wind turbine gearbox, including:
[0013] Obtain a temperature value sequence at the current moment; wherein the temperature value sequence includes temperature values collected at the current moment and at historical moments before the current moment;
[0014] The temperature value sequence is input into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence; wherein the prediction model is trained using the model training method described in the aforementioned aspect to obtain the trained prediction model.
[0015] Another aspect of the present application provides a model training device, wherein the model is used to predict the temperature of a wind turbine gearbox, and the device includes:
[0016] A first acquisition module, used to acquire a first feature extraction layer extracted from a pre-trained first neural network model, and a second feature extraction layer extracted from a pre-trained second neural network model;
[0017] A construction module, configured to construct a pre-trained prediction model according to the first feature extraction layer and the second feature extraction layer;
[0018] A second acquisition module is used to acquire a target sample temperature value sequence and a corresponding actual temperature value sequence; wherein a sampling time corresponding to at least one temperature value in the actual temperature value sequence is after a sampling time corresponding to at least one temperature value in the target sample temperature value sequence;
[0019] A prediction module, used for inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain a predicted temperature value sequence output by the prediction model;
[0020] The training module is used to train the prediction model according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
[0021] Another aspect of the present application provides a wind turbine gearbox temperature prediction device, comprising:
[0022] An acquisition module is used to acquire a temperature value sequence at the current moment; wherein the temperature value sequence includes temperature values collected at the current moment and at historical moments before the moment;
[0023] A prediction module is used to input the temperature value sequence into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence; wherein the prediction model is trained using the model training device described in the aforementioned embodiment to obtain the trained prediction model.
[0024] Another aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the foregoing method embodiment is implemented.
[0025] Another aspect of the present application is a non-temporary computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method described in the foregoing method embodiment is implemented.
[0026] Another aspect of the present application provides a computer program product on which a computer program is stored. When the program is executed by a processor, the method described in the above method embodiment is implemented.
[0027] The fan gearbox temperature prediction method, device and storage medium proposed in the present application obtain a first feature extraction layer extracted from a pre-trained first neural network model and a second feature extraction layer extracted from a pre-trained second neural network model, and construct a prediction model based on the first feature extraction layer and the second feature extraction layer, obtain a target sample temperature value sequence and a corresponding actual temperature value sequence, input the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model, obtain a predicted temperature value sequence output by the prediction model, train the prediction model according to the difference between the predicted temperature value sequence and the actual temperature value sequence, and construct a prediction model by obtaining the pre-trained first feature extraction layer and the second feature extraction layer, thereby avoiding the occurrence of local minimization caused by random initialization of initial hyperparameters during prediction model training, improving the effect of prediction model training, and thus improving the accuracy of temperature prediction.
[0028] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:
[0030] Figure 1 A flowchart of a model training method provided in an embodiment of the present application;
[0031] Figure 2 A flowchart of another model training method provided in an embodiment of the present application;
[0032] Figure 3 A schematic diagram of the structure of a second neural network model provided in an embodiment of the present application;
[0033] Figure 4 A schematic diagram of the structure of a prediction model provided in an embodiment of the present application;
[0034] Figure 5 A schematic diagram of temperature collection of a fan gearbox provided in an embodiment of the present application;
[0035] Figure 6 A schematic diagram of dividing a training sample set provided in an embodiment of the present application;
[0036] Figure 7 A schematic flow chart of a method for predicting the temperature of a fan gearbox provided in an embodiment of the present application;
[0037] Figure 8 A schematic diagram showing a comparison between the predicted results and the actual results of temperature prediction using the prediction model provided in the embodiment of the present application;
[0038] Fig. 9 A schematic diagram of the structure of a model training device provided in an embodiment of the present application;
[0039] Fig.10 A schematic diagram of the structure of a fan gearbox temperature prediction device provided in an embodiment of the present application;
[0040] Fig.11 A block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0041] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.
[0042] The following describes the fan gearbox temperature prediction method, device and storage medium of the embodiments of the present application with reference to the accompanying drawings.
[0043] Figure 1 A flowchart of a model training method provided in an embodiment of the present application.
[0044] The executor of the model training method in the embodiment of the present application is a model training device, which can be set in an electronic device. The electronic device can be a server or a terminal device, which is not limited in this embodiment.
[0045] like Figure 1 As shown, the method may include the following steps:
[0046] Step 101, obtaining a first feature extraction layer extracted from a pre-trained first neural network model, and a second feature extraction layer extracted from a pre-trained second neural network model.
[0047] The first neural network is a one-dimensional convolutional neural network, for example, a one-dimensional CNN network. The first feature extraction layer is used to extract the spatial features of the temperature data, and the spatial features are local features, for example, temperature anomalies, temperature changes in different spaces, etc. The first feature extraction layer includes at least one one-dimensional convolutional layer, and each one-dimensional convolutional layer is connected to a corresponding pooling layer.
[0048] The second neural network model is a time series model or a time series prediction model, which is used to extract periodic features in temperature features, such as a long short-term memory neural network model (Long Short-Term Memory, LSTM), a recurrent neural network model (Recurrent Neural Network, RNN) and a bidirectional long short-term memory neural network model (Bi-directional Long Short-Term Memory, BiLSTM).
[0049] In the embodiment of the present application, the first neural network and the second neural network are pre-trained separately in advance. Since the first neural network and the second neural network have simple network structures, pre-training separately can better control the parameters of the pre-training. In the embodiment of the present application, the first feature extraction layer is extracted from the first neural network model obtained by pre-training, and the second feature extraction layer is extracted from the second neural network model obtained by pre-training.
[0050] The essence of the pre-training method is that the model parameters are not randomly initialized. The model training process is divided into commonality learning and feature learning. By pre-training the model, the model obtains initial parameters. In the subsequent training process of the prediction model, the initial parameters will be fine-tuned to avoid the problem of local minimization during the training process, so as to achieve higher prediction accuracy.
[0051] Step 102, construct a prediction model based on the first feature extraction layer and the second feature extraction layer. In the embodiment of the present application, a prediction model is constructed based on the first feature extraction layer and the second feature extraction layer, so that the constructed prediction model can obtain the hyperparameters of the first feature extraction layer obtained by pre-training, and the hyperparameters of the second feature extraction layer, that is, the hyperparameters of the prediction model are not the hyperparameters set initially, which avoids the local minimization problem in the prediction model training process, finds the optimal optimization path, and greatly improves the effect of prediction model training.
[0052] It should be noted that when constructing the prediction model, the first feature extraction layer is extracted from the first neural network obtained by pre-training, that is, the convolution layer and the pooling layer, and the second feature extraction layer is extracted from the second neural network, such as the BiLSTM layer, while other processing layers, such as the input layer, the regularization layer, the activation layer, etc. are not extracted, that is, they are all constructed using new layers. In this way, the most core parts of the first neural network model and the second neural network model obtained by pre-training are retained, and the initial hyperparameters of the most core parts are obtained. In the subsequent prediction model training, the occurrence of local minimization caused by random initialization of the initial hyperparameters can be prevented, which greatly improves the effect of the prediction model training, so that the trained prediction model has a higher accuracy of temperature prediction.
[0053] Step 103: Obtain a target sample temperature value sequence and a corresponding actual temperature value sequence.
[0054] The sampling time corresponding to at least one temperature value in the actual temperature value sequence is after the sampling time corresponding to at least one temperature value in the target sample temperature value sequence. As an example, the target sample temperature value sequence and the actual temperature value sequence of the embodiment of the present application refer to the temperature value sequence of the wind turbine gearbox obtained by collection.
[0055] Among them, the training sample set includes multiple sample temperature value sequences, and the actual temperature value sequences corresponding to each sample temperature value sequence in the multiple sample temperature value sequences. The sample temperature value sequence currently used to train the prediction model is the target sample temperature value sequence. The target sample temperature value sequence includes temperature values collected by the temperature sensor at multiple times, and is arranged in the order of time collection. Among them, the actual temperature value sequence corresponding to the target sample temperature value sequence is the annotation data of the target sample temperature value sequence. The actual temperature value sequence is the true value corresponding to the predicted temperature value sequence, and the actual temperature value sequence can be the temperature value of one or more times after the last sampling time in the target sample temperature value sequence.
[0056] Step 104: input the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain a predicted temperature value sequence output by the prediction model.
[0057] In an embodiment of the present application, the target sample temperature value sequence and the corresponding actual temperature value sequence are input into the prediction model for feature extraction. Since the sampling time corresponding to at least one temperature value in the actual temperature value sequence is after the sampling time corresponding to at least one temperature value in the target sample temperature value sequence, the correlation relationship between the temperature value sequence to be predicted and the input target sample temperature value sequence is extracted to improve the accuracy of the predicted temperature value sequence.
[0058] Step 105: training the prediction model according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
[0059] Among them, the trained prediction model is used to predict the temperature of the fan gearbox.
[0060] In the embodiment of the present application, since the core layers in the prediction model, namely the first feature extraction layer and the second feature extraction layer, are extracted from the pre-trained first neural network and the second neural network, the pre-trained hyperparameters are already available through pre-training. Then, the prediction model is trained using the target sample temperature value sequence and the corresponding actual temperature value sequence. Through fine-tuning, the model can converge quickly, avoiding the occurrence of local minimization caused by random initialization of the initial hyperparameters during the prediction model training, thereby greatly improving the effect of the prediction model training.
[0061] It should be noted that the aforementioned steps 101 to 105 need to be repeated multiple times, and different training samples can be used each time, so that the training is stopped when the loss function is less than the threshold, or the training is stopped when the number of repeated executions is greater than the threshold. The prediction model obtained after the last model parameter adjustment is used as the trained prediction model.
[0062] In the model training method of the embodiment of the present application, since the core layers in the prediction model, namely the first feature extraction layer and the second feature extraction layer, are extracted from the pre-trained first neural network and the second neural network, the pre-trained hyperparameters are already available through pre-training. Then, the prediction model is trained using the target sample temperature value sequence and the corresponding actual temperature value sequence. Through fine-tuning, the model can converge quickly, avoiding the occurrence of local minimization caused by random initialization of the initial hyperparameters during the prediction model training, thereby greatly improving the effect of the prediction model training.
[0063] Based on the above embodiments, Figure 2 A flow chart of another model training method provided in an embodiment of the present application is shown as follows: Figure 2 As shown, the method comprises the following steps:
[0064] Step 201, obtaining a first feature extraction layer extracted from a pre-trained first neural network model, and a second feature extraction layer extracted from a pre-trained second neural network model.
[0065] In the embodiment of the present application, the first neural network and the second neural network are pre-trained respectively, and the training method is as follows.
[0066] Among them, the pre-training method of the first neural network is:
[0067] The first neural network model is trained using multiple sample temperature value sequences in the training sample set, that is, one sample temperature value sequence is used each time and input into the first neural network model, and the loss function is determined according to the difference between the predicted temperature value output by the first neural network model and the true value, and the parameters are adjusted using the gradient descent method based on the loss function. In response to the number of times the first neural network model is trained reaching the set number of times, the pre-training of the first neural network model is completed, and the pre-trained first neural network model obtains the pre-trained hyperparameters.
[0068] As an example, the first neural network model is a single one-dimensional convolutional neural network, such as a 1DCNN model, wherein the model includes an input layer, a regularized Dropout layer, two convolutional layers, two pooling layers, an activation layer, a fully connected layer (including Flatten and activation layers) and an output layer. The first neural network model is a one-dimensional convolutional neural network model, which is the same as a traditional convolutional neural network model. The one-dimensional convolutional neural network model includes an input layer, a first feature extraction layer and an output layer. The input layer receives data in a training set and inputs the data into the model for model training. The first feature extraction layer is also a hidden layer, which is a key layer of the convolutional neural network, and mainly includes a convolutional layer and a pooling layer, wherein the convolutional layer is described as follows:
[0069]
[0070] in represents the i-th feature of the l-th layer; is the jth feature of the (1-1)th layer; Represents the weight of the convolution kernel connected to the jth feature in layer l; is the bias value of this feature. In this function, the ReLU activation function is used, which is described as follows:
[0071] σ(x)=max(0,x)
[0072] The pooling layer mainly performs pooling operations, which can reduce network parameters in network calculations and reduce the amount of calculations. Commonly used pooling operations include maximum pooling and average pooling. Maximum pooling is used in this application, where maximum pooling can be described as:
[0073]
[0074] Among them, Ω i Represents the pooling region with index i.
[0075] Among them, the pre-training method of the second neural network model is:
[0076] The second neural network model is trained using multiple sample temperature value sequences in the training sample set, that is, one sample temperature value sequence is used each time and input into the second neural network model, and the loss function is determined according to the difference between the predicted temperature value output by the second neural network model and the true value, and the parameters are adjusted using the gradient descent method based on the loss function. In response to the number of times the second neural network model is trained reaching the set number of times, the pre-training of the second neural network model is completed, and the pre-trained second neural network model obtains the pre-trained hyperparameters.
[0077] As an example, a BiLSTM network model is constructed, which includes an input layer, a BiLSTM layer, an activation layer and an output layer.
[0078] Figure 3 A schematic diagram of the structure of a second neural network model provided for an embodiment of the present application. Among them, the long short-term memory network LTSM, with its unique gate structure, can effectively solve the gradient vanishing and gradient exploding problems that occur when the recurrent neural network (RNN) processes time series signals, and can better process time series signals. The long short-term memory network includes three gate structures: input gate, output gate and forget gate. The three gate structures can determine the signal input, output and deletion based on the current state information and the previous state information, combined with the sigmoid function and the tanh function. It can effectively extract the timing features in the time series signal of the temperature value sequence, wherein the description of the three gate structures is as follows:
[0079] Input gate: i t =Sigmoid(w f ·[h t-1 , x t ]+b f )
[0080] Forget gate: ft = Sigmoid(w t .[h t-1 , x t ]+b i )
[0081] Output gate: o t =Sigmoid(w o .[h t-1 , x t ]+b o )
[0082] Among them, x t is the input at the current moment, i.e., the temperature value sequence, h t-1 The state of the hidden layer at the previous moment, w o 、w t 、w f is the weight, b o , b f , b i is the bias term.
[0083] Although the long short-term memory network can effectively extract the time series features in the time series signal, the time series signal often contains periodic signals. Effectively extracting the periodic signal can improve the accuracy of the prediction. Therefore, based on the long short-term memory network, a bidirectional long short-term memory network BiLSTM is developed. Its bidirectional extraction of feature signals in the data set can effectively consider the connection between the previous and next features in the information, so as to achieve the purpose of extracting the periodic signal of the data set. That is to say, in the embodiment of the present application, the second neural network is a bidirectional long short-term memory network, which is applied to the sample data set of the fan gearbox temperature, and can effectively extract the periodic feature signal in the temperature signal, thereby effectively improving the prediction performance of the model.
[0084] In the embodiment of the present application, the first temperature feature carrying the spatial feature of temperature output by the pooling layer of the previous layer is further extracted with a BiLSTM layer to extract periodic features, wherein the feature extraction formula of BiLSTM is:
[0085]
[0086]
[0087]
[0088] in, is the output vector of the forward LSTM hidden layer at time t, which is the input vector x at the current time t and the forward LSTM output vector at the previous moment jointly determined; among which, is the output vector of the reverse LSTM hidden layer at time t, which is the input vector x at the current time t and the reverse LSTM output vector of the previous moment Jointly determine; t is the output of the LSTM model, where ω t is the weight matrix of the forward LSTM output; v t is the weight matrix of the reverse LSTM output; b t is the bias of the weight matrix, x tis the input vector corresponding to the input temperature value sequence composed of the sample temperature value sequence and the actual temperature value sequence input at time t, O t is the output predicted temperature value sequence. As an example, Figure 3 As shown, I1, I2, I3 and I4 are multiple temperature values in an input temperature value sequence composed of a sample temperature value sequence input in an example scenario and a corresponding actual temperature value sequence. The corresponding input vector is obtained by vectorization processing, and the corresponding O1, O2, O3 and O4 are multiple temperature values in the corresponding predicted temperature value sequence output.
[0089] Step 202: construct a prediction model based on the first feature extraction layer and the second feature extraction layer.
[0090] As an example, Figure 4 As shown, the first neural network model is 1DCNN, the second neural network model is BiLSTM model, and the first feature extraction layer is extracted from the first neural network model obtained by pre-training, that is, it includes two convolution layers, and each convolution layer is connected to a corresponding pooling layer, that is, two convolution layers and pooling layers are alternately stacked, and the convolution kernels of the two convolution layers can be different or the same. Extract the BiLSTM layer from the second neural network model obtained by pre-training, and build a new prediction model based on the extracted first feature extraction layer and the second feature extraction layer, that is, the 1DCNN-BiLSTM model. The prediction model also includes an input layer, a regularization layer (Dropout layer), and an output layer, wherein the output layer includes a fully connected layer, and the fully connected layer includes an activation layer and a Flatten layer. The Flatten layer is used to convert two-dimensional data into one-dimensional data, that is, to reduce the dimension, so as to predict the output of temperature data. These layers are not pre-trained layers.
[0091] Among them, the role of the activation layer is to introduce nonlinear factors, limit the output of neurons to a specific range, improve the robustness and stability of the network, and also play a role in alleviating the gradient to ensure the learning ability of the network.
[0092] It should be noted that a single convolutional neural network can extract spatial features from signals well, but it is difficult to effectively extract periodic features from time series signals; while a bidirectional long short-term memory network can effectively extract periodic feature signals, but it is difficult to extract spatiotemporal signals from data. This makes it difficult to extract all features of multi-feature fusion signals such as temperature signals through a single network, making it difficult to further improve the prediction accuracy. Therefore, this application uses a single neural network and the core layer of a bidirectional long short-term memory network to construct a prediction model, which can improve the effect of model training and the accuracy of model prediction.
[0093] Step 203, obtaining multiple temperature values collected by the temperature sensor during the operation of the fan, and normalizing the multiple temperature values.
[0094] In the embodiment of the present application, in the wind turbine, the gearbox plays the role of accelerating the rotation speed of the fan blades and transmitting it to the wind turbine. Through the action of the gearbox, the low-speed rotation of the fan blades can be converted into high-speed rotation and transmitted to the wind turbine, thereby increasing the power generation efficiency. However, the gearbox will generate a large amount of heat during operation. If there is poor heat dissipation or abnormal contact, it will cause abnormal temperature inside the gearbox. At the least, it will affect the viscosity of the lubricating oil and cause increased wear inside the gearbox, further aggravating the temperature abnormality and causing a vicious cycle; at the worst, it will cause the gearbox to overheat and catch fire, causing more serious economic and even energy losses. Therefore, based on the temperature of the gearbox during the operation of the wind turbine, the training samples are determined to train the prediction model for temperature prediction, so that the hazards can be predicted in advance and equipment losses can be avoided.
[0095] In one implementation of the embodiment of the present application, Figure 5 As shown, a set of sensors for measuring the temperature of the fan gearbox is installed at the embedded hole of the gear of the fan gearbox, and the temperature sensor signal line is connected to the selected PLC measurement module port in the cabin control cabinet, and the fan program is upgraded to open and define the selected module port. When the gearbox is running, the embedded temperature sensor probe can directly detect the temperature of the gearbox gear. Then, the temperature data of the gearbox collected by the sensor is preliminarily eliminated after the problem data is preprocessed and divided. The data set is divided into a training set and a test set in a ratio of 9:1, and the two are normalized. Data normalization can process the data to between 0-1, which can prevent the large numerical difference between the temperature data from covering the decimal and eliminating the influence of the decimal. The formula for normalizing the temperature data is as follows:
[0096]
[0097] Among them, x is a temperature data before normalization, x scale is the normalized temperature data, x min is the minimum temperature, x max is the maximum value of temperature.
[0098] Step 204: determine a training sample set according to the time information of the multiple temperature values after normalization.
[0099] The training sample set includes a plurality of sample temperature value sequences and a plurality of corresponding actual temperature value sequences.
[0100] As an implementation method, a sliding window of a set size is obtained, and the time information of multiple temperature values after normalization is sorted to obtain a temperature value time series. The sliding window is moved in the temperature value sequence according to a set step size to determine multiple sample temperature value sequences and multiple actual temperature value sequences corresponding to each sample temperature value sequence to form a training sample set, and then, the target sample temperature value sequence and the corresponding actual temperature value sequence are obtained from the training sample set.
[0101] As an example, according to the time information of multiple temperature values after normalization, multiple temperature values are sorted to obtain a temperature value time series. For the temperature value time series, the sliding window method is used to divide the sample data. Starting from the first temperature value in the temperature value time series, the first n data in the temperature value time series are used as the first training sample, that is, the sample temperature value sequence, and the n+1th data is used as the label of the first n temperature data, that is, the actual temperature value sequence corresponding to the first training sample. Then, continue to slide with a step size of m, and the second group of data from the 1+m to 1+m+n data is used as the training sample, and the next data is used as the label of the second training sample, that is, the actual temperature value sequence. The temperature value time series is divided into training sample sets by continuously sliding the sliding window. In the case of insufficient training samples, as many sufficient training samples with training significance as possible can be obtained, which can solve the problems of insufficient training samples and sample accuracy.
[0102] For example, Figure 6 As shown, the temperature value time series is A1-An, including n temperature values, the step size of the sliding window is 1, a sample temperature value sequence includes 5 temperature values, and the actual temperature value sequence includes one temperature value. For example, sample 1, that is, the temperature values included in the first sample temperature value sequence are A1, A2, A3, A4 and A5, and the corresponding actual temperature value sequence is A6, also called the annotated label; sample 2, that is, the temperature values included in the second sample temperature value sequence are A2, A3, A4, A5 and A6, and the corresponding actual temperature value sequence is A7, that is, the annotated label, which is not listed here one by one. Similarly, by continuously sliding the sliding window, the division of multiple temperature value time series can be completed, the training sample set can be determined, and the diversity of samples can be increased.
[0103] For example, sample 1, that is, the first sample temperature value sequence and the corresponding actual temperature value sequence are input into the prediction model, and the output is the predicted temperature value corresponding to the moment A6.
[0104] It should be understood that in the above example, n temperature values are taken as a sample temperature value sequence each time, and n+1 temperature value is taken as the actual temperature value sequence, that is, the predicted temperature value is the temperature value at a moment after each input training sample. In actual application, the actual temperature value sequence can also be the temperature values at multiple moments, which can be flexibly adjusted according to needs.
[0105] Step 205: determine a target sample temperature value sequence and a corresponding actual temperature value sequence from the training sample set.
[0106] The sample training set includes multiple sample temperature value sequences and corresponding actual temperature value sequences, and the target sample temperature value sequence is any one of the multiple sample temperature value sequences.
[0107] Step 206: Obtain a target sample temperature value sequence and a corresponding actual temperature value sequence.
[0108] The sampling time corresponding to at least one temperature value in the actual temperature value sequence is after the sampling time corresponding to at least one temperature value in the target sample temperature value sequence.
[0109] Among them, step 205 and step 206 can refer to the explanation in the above-mentioned embodiment, and the principles are the same, so they will not be repeated here.
[0110] Step 207: input the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain the predicted temperature value sequence output by the prediction model.
[0111] In one implementation of the embodiment of the present application, the target sample temperature value sequence and the corresponding actual temperature value sequence are input into the input layer of the prediction model for vectorization processing, and a vector including the target sample temperature value sequence and the corresponding actual temperature value sequence is obtained. Based on the vector of the target sample temperature value sequence and the vector of the corresponding actual temperature value sequence, feature extraction can be performed to better learn the relationship between the previous temperature data and the subsequent temperature data, that is, the temperature data to be predicted, thereby improving the training effect. Further, the vector of the target sample temperature value sequence and the vector of the corresponding actual temperature value sequence are input into the first feature extraction layer of the prediction model for feature extraction to obtain the first temperature feature, wherein the first temperature feature includes the spatial feature of the temperature, the first temperature feature is input into the second feature extraction layer for feature extraction to obtain the second temperature feature, wherein the second temperature feature includes the spatial feature of the temperature and the periodic feature of the temperature, and the second temperature feature is input into the output layer of the prediction model to obtain the predicted temperature value sequence. The second temperature feature includes the spatial feature of the temperature and the periodic feature of the temperature, which realizes the acquisition of multi-dimensional features, and performs the fusion of spatiotemporal features, so that the extracted temperature features are more complete, which can improve the effect of the prediction model training and the prediction accuracy of the prediction model obtained by training.
[0112] Among them, the first feature extraction layer is specifically used to process the one-dimensional data of temperature data, which successfully retains the timing characteristics between the data and avoids the loss of feature information caused by forced conversion into two-dimensional signals.
[0113] Step 208: training the prediction model based on the difference between the predicted temperature value sequence and the actual temperature value sequence.
[0114] In an embodiment of the present application, the prediction model has initial hyperparameters after pre-training. In the subsequent model training process, its initial hyperparameters are no longer random. This method can effectively avoid the problem of local minimization in the training process and greatly improve the accuracy of temperature prediction by the prediction model.
[0115] This application uses a deep learning algorithm to predict the temperature of a fan gearbox. The method is compared with a traditional mathematical method. The traditional mathematical method selects a suitable mathematical regression model for temperature prediction through the connection between data. This method is more dependent on professional knowledge and prior experience, and for indicators such as temperature that are greatly affected by environmental factors, the prediction accuracy of the traditional mathematical method is low. The fan gearbox temperature prediction method based on a deep learning algorithm can automatically extract characteristic signals from the learning data through the model and give a more accurate temperature prediction value. This method can better handle nonlinear temperature signals, can greatly improve the prediction accuracy compared to traditional methods, and can get rid of the dependence on prior knowledge and expert experience.
[0116] After the prediction model training is completed, it can be tested based on the test set. If the error of the predicted temperature value meets the error requirement, the prediction model training is completed and can be used for temperature prediction. Otherwise, continue to use the training sample set for training.
[0117] In the model training method of the present application, the first neural network model and the second neural network model are pre-trained separately through the gearbox temperature data set. The one-dimensional convolution pooling layer in the first neural network model can better process the one-dimensional time series signal without losing its temporal sequence, and can extract more complete signal features. The second neural network model can extract the periodic features contained in the temperature signal with its unique feature extraction method, and improve the accuracy of temperature prediction. The first neural network model and the second neural network model both contain all parts of the complete deep learning model. Through model pre-training, the network obtains initial hyperparameters, and then based on the hidden layers in the two networks, namely the first feature extraction layer and the second feature extraction layer, a pre-trained prediction model is constructed, such as a 1DCNN-BiLSTM model, so that the model can obtain initial hyperparameters, avoid local minimization problems during model training, find the optimal optimization path, and greatly improve the model prediction accuracy.
[0118] Based on the above embodiments, Figure 7 A schematic diagram of a flow chart of a fan gearbox temperature prediction method provided in an embodiment of the present application, such as Figure 7 As shown, the method comprises the following steps:
[0119] Step 701, obtaining the temperature value sequence at the current moment.
[0120] The temperature value sequence includes the temperature values collected at the current moment and the historical moments before the current moment. As an example, the temperature value sequence refers to the temperature value sequence of the wind turbine gearbox.
[0121] Step 702: Input the temperature value sequence into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence.
[0122] The prediction model is trained using the model training method described in the above method embodiment to obtain a trained prediction model. The specific model training method can refer to the explanation in the above embodiment, the principle is the same, and it will not be repeated here.
[0123] As an example, the temperature value sequence contains 5 temperature values, which are divided into temperature data collected at t1, t2, t3, t4 and t5, where the interval between the moments can be set according to the sampling accuracy requirements, for example, 2 minutes, t5 is the current moment, and the others are historical moments before t5, and the predicted temperature values are at t6, t7 and t8.
[0124] As an example, according to the wind turbine gearbox temperature collected by the sensor, the wind turbine will be disturbed by the wind speed during the sampling process, so the temperature data will fluctuate to a certain extent during the rising process. Figure 8 The curve comparison chart of the predicted data value of the temperature of gearbox P2 and the test data set is shown. It can be seen from the image that the prediction accuracy of this method is extremely high and its prediction error is extremely small, which meets the requirements for temperature prediction error in actual engineering and can accurately predict temperature in actual engineering applications.
[0125] In the fan gearbox temperature prediction method of the embodiment of the present application, the temperature of the fan gearbox is predicted by the prediction model obtained through training, which can better process the linear data of the temperature and accurately predict the temperature.
[0126] In order to implement the above embodiment, the embodiment of the present application also proposes a model training device, wherein the model is used to predict the temperature of the wind turbine gearbox.
[0127] Fig. 9 A schematic diagram of the structure of a model training device provided in an embodiment of the present application.
[0128] like Fig. 9 As shown, the device may include:
[0129] The first acquisition module 901 is used to acquire a first feature extraction layer extracted from a pre-trained first neural network model and a second feature extraction layer extracted from a pre-trained second neural network model.
[0130] The construction module 902 is used to construct a pre-trained prediction model according to the first feature extraction layer and the second feature extraction layer.
[0131] The second acquisition module 903 is used to acquire a target sample temperature value sequence and a corresponding actual temperature value sequence; wherein a sampling time corresponding to at least one temperature value in the actual temperature value sequence is after a sampling time corresponding to at least one temperature value in the target sample temperature value sequence.
[0132] The prediction module 904 is used to input the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain the predicted temperature value sequence output by the prediction model.
[0133] The training module 905 is used to train the prediction model according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
[0134] Furthermore, in one implementation of the embodiment of the present application, the device includes:
[0135] A processing module, used to obtain multiple temperature values collected by the temperature sensor during the operation of the fan; and normalize the multiple temperature values;
[0136] A determination module is used to determine a training sample set according to the time information of the multiple temperature values after normalization; wherein the training sample set includes multiple sample temperature value sequences and corresponding multiple actual temperature value sequences; and determine the target sample temperature value sequence and the corresponding actual temperature value sequence from the training sample set.
[0137] In one implementation of the embodiment of the present application, the determination module is specifically configured to:
[0138] Acquire a sliding window of a set size; sort the time information of the multiple temperature values after the normalization process to obtain a temperature value time series; move the sliding window in the temperature value sequence according to a set step size to determine multiple sample temperature value sequences and corresponding multiple actual temperature value sequences.
[0139] In one implementation of the embodiment of the present application, the prediction module 904 is specifically configured to:
[0140] Inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the input layer of the prediction model for vectorization processing to obtain a vector of the target sample temperature value sequence and a vector of the corresponding actual temperature value sequence;
[0141] Inputting the vector of the target sample temperature value sequence and the vector of the corresponding actual temperature value sequence into the first feature extraction layer of the prediction model for feature extraction to obtain a first temperature feature; wherein the first temperature feature includes a spatial feature of temperature;
[0142] Inputting the first temperature feature into the second feature extraction layer for feature extraction to obtain a second temperature feature; wherein the second temperature feature includes a spatial feature of the temperature and a periodic feature of the temperature;
[0143] The second temperature feature is input into the output layer of the prediction model to obtain the predicted temperature value sequence.
[0144] In one implementation of the embodiment of the present application, the first feature extraction layer includes at least one one-dimensional convolution layer, and each of the one-dimensional convolution layers is connected to a corresponding pooling layer.
[0145] In one implementation of the embodiment of the present application, the device further includes:
[0146] The first pre-training module is used to train the first neural network model using multiple sample temperature value sequences in the training sample set; in response to the number of times the first neural network model is trained reaching a set number of times, the pre-training of the first neural network model is completed.
[0147] In one implementation of the embodiment of the present application, the device further includes:
[0148] The second pre-training module is used to train the second neural network model using multiple sample temperature value sequences in the training sample set; in response to the number of times the second neural network model is trained reaching the set number of times, the pre-training of the second neural network model is completed.
[0149] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and will not be repeated here.
[0150] In the model training device of the embodiment of the present application, a first feature extraction layer extracted from a first neural network model obtained through pre-training and a second feature extraction layer extracted from a second neural network model obtained through pre-training are obtained; a prediction model is constructed based on the first feature extraction layer and the second feature extraction layer; a target sample temperature value sequence and a corresponding actual temperature value sequence are obtained; the target sample temperature value sequence and the corresponding actual temperature value sequence are input into the prediction model to obtain a predicted temperature value sequence output by the prediction model; the prediction model is trained based on the difference between the predicted temperature value sequence and the actual temperature value sequence; a prediction model is constructed by obtaining the first feature extraction layer and the second feature extraction layer after pre-training, thereby avoiding the occurrence of local minimization caused by random initialization of initial hyperparameters during prediction model training and improving the effect of prediction model training.
[0151] In order to implement the above embodiment, the embodiment of the present application also proposes a temperature prediction device for a wind turbine gearbox.
[0152] Fig.10 A schematic diagram of the structure of a fan gearbox temperature prediction device provided in an embodiment of the present application.
[0153] like Fig.10 As shown, the device may include:
[0154] The acquisition module 1001 is used to acquire a temperature value sequence at the current moment; wherein the temperature value sequence includes temperature values collected at the current moment and at historical moments before the current moment.
[0155] The prediction module 1002 is used to input the temperature value sequence into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence; wherein the prediction model is trained using the model training device described in the above embodiment to obtain the trained prediction model.
[0156] It should be noted that the above explanation of the method embodiment is also applicable to the device of this embodiment, and will not be repeated here.
[0157] In the fan gearbox temperature prediction device of the embodiment of the present application, temperature prediction is performed through the prediction model obtained by training in the above embodiment, which can better process one-dimensional temperature data, extract linear data, and improve the accuracy of temperature prediction.
[0158] In order to implement the above embodiments, the present application also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the above method embodiments is implemented.
[0159] In order to implement the above embodiments, the present application also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the above method embodiments is implemented.
[0160] In order to implement the above embodiments, the present application also proposes a computer program product on which a computer program is stored. When the computer program is executed by a processor, the method described in the above method embodiments is implemented.
[0161] Fig.11 This is a block diagram of an electronic device provided in an embodiment of the present application. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.
[0162] Reference Fig.11 , the electronic device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .
[0163] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
[0164] The memory 804 is configured to store various types of data to support operations on the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
[0165] The power component 806 provides power to the various components of the electronic device 800. The power component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.
[0166] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.
[0167] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.
[0168] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
[0169] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the electronic device 800. For example, the sensor assembly 814 can detect the open / closed state of the electronic device 800, the relative positioning of the components, such as the display and keypad of the electronic device 800, and the sensor assembly 814 can also detect the position change of the electronic device 800 or a component of the electronic device 800, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and the temperature change of the electronic device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
[0170] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 4G or 5G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
[0171] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above methods.
[0172] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, and the instructions can be executed by a processor 820 of an electronic device 800 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0173] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.
[0174] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0175] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.
[0176] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.
[0177] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.
[0178] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.
[0179] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.
[0180] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A model training method, It is characterized in that The model is used to predict the temperature of the wind turbine gearbox, including: Obtaining a first feature extraction layer extracted from a pre-trained first neural network model and a second feature extraction layer extracted from a pre-trained second neural network model; Building a prediction model based on the first feature extraction layer and the second feature extraction layer; Acquire a target sample temperature value sequence and a corresponding actual temperature value sequence; wherein a sampling time corresponding to at least one temperature value in the actual temperature value sequence is after a sampling time corresponding to at least one temperature value in the target sample temperature value sequence; Inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain a predicted temperature value sequence output by the prediction model; The prediction model is trained according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
2. The method according to claim 1, It is characterized in that Before obtaining the target sample temperature value sequence and the corresponding actual temperature value sequence, the method includes: Acquire multiple temperature values collected by the temperature sensor during the operation of the fan; normalizing the multiple temperature values; Determine a training sample set according to the time information of the plurality of temperature values after the normalization process; wherein the training sample set includes a plurality of sample temperature value sequences and a plurality of corresponding actual temperature value sequences; The target sample temperature value sequence and the corresponding actual temperature value sequence are determined from the training sample set.
3. The method according to claim 2, It is characterized in that The step of determining a training sample set according to the time information of the plurality of temperature values after the normalization process comprises: Get a sliding window of a set size; Sorting the time information of the plurality of temperature values after the normalization process to obtain a temperature value time series; The sliding window is moved in the temperature value sequence according to a set step length to determine the multiple sample temperature value sequences and the corresponding multiple actual temperature value sequences.
4. The method according to claim 1, It is characterized in that The step of inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain the predicted temperature value sequence output by the prediction model comprises: Inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the input layer of the prediction model for vectorization processing to obtain a vector of the target sample temperature value sequence and a vector of the corresponding actual temperature value sequence; Inputting the vector of the target sample temperature value sequence and the vector of the corresponding actual temperature value sequence into the first feature extraction layer of the prediction model for feature extraction to obtain a first temperature feature; wherein the first temperature feature includes a spatial feature of temperature; Inputting the first temperature feature into the second feature extraction layer for feature extraction to obtain a second temperature feature; wherein the second temperature feature includes a spatial feature of the temperature and a periodic feature of the temperature; The second temperature feature is input into the output layer of the prediction model to obtain the predicted temperature value sequence.
5. The method according to claim 4, It is characterized in that The first feature extraction layer includes at least one one-dimensional convolution layer, and each of the one-dimensional convolution layers is connected to a corresponding pooling layer.
6. The method according to claim 2, It is characterized in that The method further comprises: Using a plurality of sample temperature value sequences in the training sample set to train the first neural network model; In response to the number of times the first neural network model is trained reaching a set number of times, the pre-training of the first neural network model is completed.
7. The method according to claim 2, It is characterized in that The method further comprises: Using a plurality of sample temperature value sequences in the training sample set to train the second neural network model; In response to the number of times the second neural network model is trained reaching the set number of times, the pre-training of the second neural network model is completed.
8. A method for predicting fan gearbox temperature. It is characterized in that include: Get the temperature value sequence at the current moment; In the calculation, the temperature value sequence includes temperature values collected at the current moment and at historical moments before the current moment; The temperature value sequence is input into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence; wherein the prediction model is trained using the model training method described in any one of claims 1-7 to obtain the trained prediction model.
9. A model training device, It is characterized in that The model is used to predict the temperature of the wind turbine gearbox, including: A first acquisition module, used to acquire a first feature extraction layer extracted from a pre-trained first neural network model, and a second feature extraction layer extracted from a pre-trained second neural network model; A construction module, configured to construct a pre-trained prediction model according to the first feature extraction layer and the second feature extraction layer; A second acquisition module is used to acquire a target sample temperature value sequence and a corresponding actual temperature value sequence; wherein a sampling time corresponding to at least one temperature value in the actual temperature value sequence is after a sampling time corresponding to at least one temperature value in the target sample temperature value sequence; A prediction module, used for inputting the target sample temperature value sequence and the corresponding actual temperature value sequence into the prediction model to obtain a predicted temperature value sequence output by the prediction model; The training module is used to train the prediction model according to the difference between the predicted temperature value sequence and the actual temperature value sequence.
10. A fan gearbox temperature prediction device, It is characterized in that include: The acquisition module is used to obtain the temperature value sequence at the current moment; In the calculation, the temperature value sequence includes temperature values collected at the current moment and at historical moments before the moment; A prediction module is used to input the temperature value sequence into the trained prediction model to obtain a predicted temperature value sequence corresponding to the temperature value sequence; wherein the prediction model is trained using the model training device as described in claim 9 to obtain the trained prediction model.
11. An electronic device, It is characterized in that The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 7 is implemented, or the method according to any one of claim 8 is implemented.
12. A non-transitory computer-readable storage medium having stored thereon a computer program, It is characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented, or the method according to any one of claim 8 is implemented.
Citation Information
Cited By
Method for predicting wear of friction plate of fan coupler
CN119309805A
