Train wheel set axle temperature prediction grading early warning method and system based on recursive multivariable interaction network

Abstract

The invention provides a method and a system for predicting, grading and early warning the axle temperature of a train wheel set based on a recursive multivariable interactive network, and belongs to the technical field of train state monitoring and early warning. Temperature data of each wheel set bearing in the same train bogie is used as input, and correlation characteristics among different monitoring quantities are mined through multivariable interactive modeling; and a dynamic dependency relationship of bearing temperature evolving along with time is constructed by combining recursive time sequence modeling, accurate prediction of future temperature of the wheel set bearing is realized, and effective discrimination of predicted temperature rise is realized by setting a grading early warning mechanism. According to the method, the problems of insufficient utilization of multivariable information, insufficient time-dependent description and the like are solved, relatively high prediction precision can be kept under different time spans, and accurate judgment of graded early warning is realized; compared with a traditional single network, the method has the advantages of being high in prediction precision, good in early warning effect, high in robustness, wide in application range and the like, and reliable technical support can be provided for train bearing health management and operation safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of train state monitoring and early warning, and in particular to a train wheel set axle temperature prediction and hierarchical early warning method and system based on a recursive multivariate interaction network. BACKGROUND

[0002] High-speed rail operating conditions are complex and diverse, and vehicles are prone to wear and tear and component aging during long-term operation, which will bring different degrees of hidden dangers to the safe operation of the train. Bearings, as an important part of the transmission system of the train bogie axle box, gear box and motor, play a key role in ensuring the safe operation of the train, and bearings are also one of the components that are most prone to damage during operation. Therefore, the bearing state must be monitored in real time to make timely maintenance and repair decisions. The temperature of the bearing can directly reflect its working state. During train operation, abnormal temperature rise of the bearing not only may cause emergency braking, resulting in economic losses, but also may lead to train derailment in more serious cases, threatening the safety of passengers. Therefore, realizing real-time prediction and hierarchical early warning of bearing temperature and grasping the trend of bearing temperature in advance is of great significance for train operation maintenance and spare parts management, and provides a reliable basis for optimizing and adjusting train operation strategies.

[0003] Train axle temperature prediction is a typical multivariate time series prediction task, and traditional statistical modeling and empirical analysis methods have certain limitations in dealing with multivariate time series prediction tasks. In contrast, deep learning technology has shown more obvious advantages due to its stronger feature expression capability. Recurrent neural networks improve the accuracy of temperature prediction by introducing time step information, and then gradually produce gated recurrent units, long short-term memory networks, and graph attention networks for prediction. However, there are still problems to be solved in the actual application of existing single networks for train axle temperature prediction: the existing methods still lack an effective integration mechanism for the temperature states of different bearings within the same bogie, and fail to fully build temperature correlations between bearings; at the same time, traditional time series prediction models only model from a single time dimension when dealing with different types and scales of temperature sequences, and cannot capture both the time-dependent and spatial-coupled features of temperature changes. These shortcomings cause obvious deviations in the prediction of bearing temperature, making it difficult to predict key risk points such as sudden temperature rise and abnormal heat state in a timely manner, affecting the prediction accuracy and safety and reliability in actual applications. SUMMARY

[0004] The present application aims to provide a train wheel set axle temperature prediction and hierarchical early warning method and system based on a recursive multivariate interaction network to solve at least one of the technical problems in the background.

[0005] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0006] In a first aspect, the present application provides a train wheelset axle temperature prediction and hierarchical warning method based on a recursive multivariate interaction network, comprising:

[0007] Obtaining train running speed data;

[0008] Processing the obtained speed data using a pre-trained train wheelset axle temperature prediction and hierarchical model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheelset axle temperature prediction and hierarchical model comprises: a multivariate dependent structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient preservation of feature information; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time series prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; a double-branch fusion structure is introduced for prediction mapping, fully mining the nonlinear relationship between time dependence and candidate features;

[0009] Based on the automatic hierarchical warning mechanism of the predicted temperature rise, whether the axle bearing temperature prediction value exceeds the corresponding threshold value is judged according to the set warning threshold value under different running speeds, and if the threshold value is exceeded, a warning process is performed.

[0010] As a further limitation of the first aspect of the present application, the multivariate correlation weight matrix is constructed, and the standardized interaction weight is obtained through normalization processing, comprising: regarding each time series in the data set as a variable sequence, establishing a connection relationship between variables according to correlation, forming an interaction structure composed of a variable set and its dependence relationship; according to the correlation between time series, the correlation weight element is an exponential decay weight between time steps, representing the influence of variable and on the weight of time step .

[0011] As a further limitation of the first aspect of the present application, the recursive multivariate interaction network more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient preservation of feature information, comprising:

[0012] For the input multivariate features, the i-th variable and its associated variables The single-head attention coefficients are normalized, and the sum of all attention coefficients related to the i-th variable is normalized to ensure that the sum of the attention coefficients is 1, resulting in standardized attention weights. Through weighted summation and self-loop compensation, each temporal variable can absorb the correlation information of its neighbors while retaining its own features during updates, achieving stable and flexible temporal feature aggregation. The vectors transmitted by the single-head information in each layer are aggregated, and a multi-head attention mechanism is implemented using multiple sets of attention weights in parallel computation. This allows variables to learn neighborhood dependencies in different feature subspaces, achieving multi-scale feature aggregation. The network is enhanced from local representation to global modeling, enabling variables to aggregate features of far-distance related variables. A normalized interaction weight matrix is ​​also included. For the characteristics of the current variable Perform feature aggregation and then perform linear transformation matrix Matrix multiplication is used to map the feature space, and then features from the previous layer are fused together. ,use Adjust the dimensions to align the feature space.

[0013] As a further limitation of the first aspect of the present invention, the spatiotemporal sequence encoding module simultaneously characterizes the dependencies between multiple variables and their temporal evolution, including: performing time series encoding on the obtained variable feature matrix; and encoding the matrix into a time series matrix of length [missing information]. The time windows are stacked into a three-dimensional tensor of "time, variables and features", introducing time series dependencies as input to the next time series prediction network; the change trajectory of each variable within the time window is extracted, and the three-dimensional tensor is treated as a multivariate time series for batch input.

[0014] As a further limitation of the first aspect of the present invention, the encoded multivariate feature sequence is gradually mapped to the predicted bearing temperature value at future time by the recursive time series prediction module, so as to realize the modeling and prediction of the temperature change trend of train wheelset bearings; for each time step, the state is updated by combining the layer normalized gated loop unit, and the feature dimension of the sample at the current time step is standardized.

[0015] As a further limitation of the first aspect of the invention, a dual-branch fusion structure is introduced for prediction mapping to fully explore the nonlinear relationship between time dependence and candidate features; this structure, on the one hand, affects the final hidden state Layer normalization and nonlinear function activation are performed to extract stationary features; on the other hand, candidate hidden states are utilized. go through The activation captures its short-term dynamic changes, and the two parts are weighted and fused by linear transformation to obtain the final prediction output.

[0016] Secondly, the present invention provides a train wheelset axle temperature prediction and hierarchical early warning system based on a recursive multivariate interaction network, comprising:

[0017] an acquisition module configured to acquire train running speed data;

[0018] a prediction module configured to process the acquired speed data by using a pre-trained train wheelset axle temperature prediction hierarchical model to obtain an axle bearing temperature prediction value under the running speed, wherein the train wheelset axle temperature prediction hierarchical model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient feature information retention; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and the evolution law thereof in time; a recursive time sequence prediction module, which gradually maps the encoded multivariate feature sequence into an axle bearing temperature prediction value at a future time; and a double-branch fusion structure is introduced for prediction mapping, so as to fully mine the nonlinear relationship between time dependence and candidate features;

[0019] a warning module configured to judge whether the axle bearing temperature prediction value exceeds a corresponding threshold value according to a set warning threshold value under different running speeds based on a prediction temperature rise automatic hierarchical warning mechanism, and perform a warning process if the threshold value is exceeded.

[0020] In a third aspect, the present application provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implement the train wheelset axle temperature prediction hierarchical warning method based on the recursive multivariate interaction network as described in the first aspect.

[0021] In a fourth aspect, the present application provides a computer device comprising a memory and a processor, the processor and the memory being in communication with each other, the memory storing program instructions executable by the processor, and the processor invoking the program instructions to execute the train wheelset axle temperature prediction hierarchical warning method based on the recursive multivariate interaction network as described in the first aspect.

[0022] In a fifth aspect, the present application provides an electronic device comprising a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the instructions for implementing the train wheelset axle temperature prediction hierarchical warning method based on the recursive multivariate interaction network as described in the first aspect.

[0023] The application has the advantages that the multivariate interaction modeling is combined with the time sequence prediction network, the correlation between bearing temperature data and the dependent characteristics of the evolution over time can be described at the same time, and the hierarchical early warning is carried out on the basis of the prediction result.

[0024] The advantages of the additional aspects of the application will be more apparent from the following description part or be understood through the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0026] Figure 1 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0027] Figure 2 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0028] Figure 3 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0029] Figure 4 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0030] Figure 5 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0031] Figure 6 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application.

[0032] Figure 7 The train wheel bearing temperature prediction and hierarchical early warning method based on the recursive multivariate interaction network is described in the embodiment of the application. DETAILED DESCRIPTION

[0033] Embodiments of the present application are described in detail below with reference to the attached drawing figures, wherein like or similar constituent elements or features may be denoted by like reference characters throughout the drawing figures and text. The embodiments described below are merely exemplary in nature and are not intended to limit the application, as defined by the appended claims.

[0034] As would be apparent to those skilled in the art, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0035] It should also be understood that terms such as those defined in a general dictionary, should be interpreted in the same manner as understood by those skilled in the art in the context of the present technology and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0036] In order to facilitate the understanding of the present application, the present application will be further explained and described below in conjunction with the accompanying drawings and specific embodiments, and the specific embodiments do not constitute a limitation on the embodiments of the present application.

[0037] Those skilled in the art should understand that the drawings are only schematic representations of embodiments, and the components in the drawings are not necessarily essential for the implementation of the present application.

[0038] The present application proposes a train wheelset axle temperature prediction and hierarchical early warning method based on recursive multivariate interaction network. In terms of temperature prediction, multivariate interaction modeling and time series prediction network are combined. First, a recursive multivariate interaction network is used to weight and fuse different wheelset bearings and their related monitoring quantities within the same bogie, and to extract multivariate features representing the state transmission and coupling relationship between the bearings. Then, the features are input into a time series encoding module to introduce the dependence of temperature evolution on time, and to complete the joint modeling of multivariate state information and time variation information. Finally, the encoded time series features are input into a time series prediction network to obtain the temperature prediction results of the train wheelset bearings. On this basis, a hierarchical early warning mechanism is constructed, which takes the difference between the bearing temperature prediction value and the outdoor temperature as the temperature rise prediction quantity, sets multiple threshold values based on the average and standard deviation of the temperature rise under different speed conditions, and realizes the hierarchical early warning of the bearing overheating risk. The present application effectively mines the correlation and interaction information between multivariate while fully preserving the long-term dependence of temperature sequence, and realizes the risk assessment of bearing operation based on the predicted temperature rise, providing a safe early warning and decision basis for the operation and maintenance personnel.

[0039] Example 1

[0040] In this embodiment 1, first provide a train wheel pair axle temperature prediction hierarchical warning system based on recursive multivariate interaction network, comprising: an acquisition module for acquiring train running speed data. A prediction module for processing the acquired speed data using a pre-trained train wheel pair axle temperature prediction hierarchical model to obtain the bearing temperature prediction value under the running speed; wherein the train wheel pair axle temperature prediction hierarchical model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; recursive multivariate interaction network, under the premise of ensuring sufficient retention of feature information, more comprehensively depict the complex multivariate dependence within the time window; a space-time sequence encoding module simultaneously depicts the dependence relationship between multivariate and its evolution law in time; the recursive time sequence prediction module gradually maps the encoded multivariate feature sequence into the bearing temperature prediction value at the future time; introduce a double-branch fusion structure for prediction mapping, and fully excavate the nonlinear relationship between time dependence and candidate features. A warning module for automatically classifying and warning based on the predicted temperature rise mechanism, determining whether the bearing temperature prediction value exceeds the corresponding threshold according to the set warning threshold under different running speeds, and performing warning processing if the threshold is exceeded.

[0041] In this embodiment, a train wheel pair axle temperature prediction hierarchical warning method is realized based on the above system, comprising: using the acquisition module to acquire train running speed data; using the prediction module to process the acquired speed data using a pre-trained train wheel pair axle temperature prediction hierarchical model to obtain the bearing temperature prediction value under the running speed; wherein the train wheel pair axle temperature prediction hierarchical model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; recursive multivariate interaction network, under the premise of ensuring sufficient retention of feature information, more comprehensively depict the complex multivariate dependence within the time window; a space-time sequence encoding module simultaneously depicts the dependence relationship between multivariate and its evolution law in time; the recursive time sequence prediction module gradually maps the encoded multivariate feature sequence into the bearing temperature prediction value at the future time; introduce a double-branch fusion structure for prediction mapping, and fully excavate the nonlinear relationship between time dependence and candidate features. Finally, using the warning module based on the predicted temperature rise automatic hierarchical warning mechanism, determining whether the bearing temperature prediction value exceeds the corresponding threshold according to the set warning threshold under different running speeds, and performing warning processing if the threshold is exceeded.

[0042] The multivariate correlation weight matrix is constructed, and the standardized interaction weight is obtained through normalization processing, comprising: regarding each time series in the data set as a variable sequence, establishing a connection relationship between variables according to correlation, forming an interaction structure composed of a variable set and its dependence relationship; according to the correlation between time series, the correlation weight element is an exponential decay weight between time steps, representing the variable and the interaction weight of variable at time step .

[0043] Specifically, first, each time series in the dataset is regarded as a variable sequence, and the connection relationship between variables is established according to the correlation, forming an interaction structure composed of a variable set and its dependency relationship. The structure is defined as , where represents the variable set, represents the connection relationship between variables, is the feature sequence, i.e., the time point sequence in the dataset, and each time step is represented as , represents the feature matrix of the variable at time step , is the feature dimension of each time step , the variable set , if two variables are associated at a certain time step, it is recorded as , and the dependency relationship is determined by the correlation between variables:

[0044] ;

[0045] Secondly, the adaptive construction and normalization processing of the multivariate correlation weight matrix are carried out. The specific process includes three parts: correlation weight calculation, weighted degree matrix calculation and symmetric normalization. First, according to the correlation between time series, the correlation weight element is defined as is the exponential decay weight between time steps, representing the interaction weight of variable and at time step :

[0046]

[0047] where is a regulation coefficient for controlling the smoothness of the weight distribution between variables, and the weighted degree matrix is introduced to measure the "overall connection degree" between each time step and its adjacent time steps, and the correlation weight matrix is symmetrically normalized to obtain the standardized interaction weight matrix.

[0048]

[0049] ( )

[0050] where represents the total connection strength of variable , This indicates the generation of a diagonal matrix where only the elements on the main diagonal are non-zero, and all other elements are 0. Preserve matrix symmetry.

[0051] Recursive multivariate interaction networks, while ensuring sufficient preservation of feature information, more comprehensively characterize complex multivariate dependencies within a time window, including: for input multivariate features, calculating the i-th variable and its associated variables. The single-head attention coefficients are normalized, and the sum of all attention coefficients related to the i-th variable is normalized to ensure that the sum of the attention coefficients is 1, resulting in standardized attention weights. Through weighted summation and self-loop compensation, each temporal variable can absorb the correlation information of its neighbors while retaining its own features during updates, achieving stable and flexible temporal feature aggregation. The vectors transmitted by the single-head information in each layer are aggregated, and a multi-head attention mechanism is implemented using multiple sets of attention weights in parallel computation. This allows variables to learn neighborhood dependencies in different feature subspaces, achieving multi-scale feature aggregation. The network is enhanced from local representation to global modeling, enabling variables to aggregate features of far-distance related variables. A normalized interaction weight matrix is ​​also included. For the characteristics of the current variable Perform feature aggregation and then perform linear transformation matrix Matrix multiplication is used to map the feature space, and then features from the previous layer are fused together. ,use Adjust the dimensions to align the feature space.

[0052] Specifically, a recursive multivariate interaction network is constructed to more comprehensively characterize complex multivariate dependencies within a time window while ensuring sufficient preservation of feature information, including:

[0053] S2-1, Input Multivariate Features , ∈ , It is the number of variables. Indicates the dimension of the input features. This represents the dimension of the output feature, and the corresponding feature vector of its variable is... , ∈ . No. Variables and their related variables The single-head attention coefficient is calculated as follows:

[0054] ;

[0055] in It is a variable For variables Attention weight coefficients; by implementing a masking attention mechanism, they are calculated only between pre-determined pairs of variables that are correlated. Ignore irrelevant or weakly correlated variable combinations. Introduce learnable bias terms. Enhance the ability to distinguish between variables. This is a normalization term to prevent gradient explosion and make interactive feature modeling more efficient and stable.

[0056] S2-2, Attention weight normalization, for the first... All attention coefficients related to each variable are normalized to ensure that the sum of the attention coefficients is 1, resulting in standardized attention weights:

[0057]

[0058] in The standard attention mechanism obtained through softmax is improved by introducing a regularization balance term. Smooth attention distribution to avoid gradients that are too large or too small, thereby increasing network stability and generalization ability.

[0059] S2-3, Attention-based Neighborhood Information Fusion. Through weighted summation and self-loop compensation, each time variable can absorb the correlation information of its neighbors while retaining its own characteristics during the update, achieving stable and flexible aggregation of temporal features.

[0060]

[0061] in It is an activation function that maintains linearity when the input is greater than 0 and sets it to zero when the input is less than 0. In order to be with the first A set of variables related to a variable. This is the weight matrix calculated in the previous step. Each is the current variable and related variables Input features and variables The aggregated output features and Each is the current variable and related variables The linear transformation matrix, These are self-loop weighting coefficients used to control the proportion of a variable's own features in the aggregation, so that it can retain its own feature information.

[0062] S2-4. Aggregate the vectors transmitted by the single-head information in each layer, and use multiple sets of attention weights to perform parallel computation to implement a multi-head attention mechanism. This enables variables to learn neighborhood dependencies in different feature subspaces, achieving multi-scale feature aggregation.

[0063]

[0064] wherein, is the number of multi-head attention, is the mechanism of the th attention variable under the attention weight of the variable , respectively, is the feature transformation matrix and the self-loop transformation matrix of the th attention head, corresponds to the self-loop fusion coefficient of the th attention head, represents vector splicing, connecting the multi-head results to form a high-dimensional representation, is an activation function.

[0065] S2-5, multi-layer recursive propagation and residual enhancement. This step enhances the network from local representation to global modeling of features, allowing variables to aggregate features of long-distance related variables. The normalized interaction weight matrix performs feature aggregation on the current variable feature performs matrix multiplication to realize feature space mapping, and then fuses the features of the upper layer , uses to adjust the dimension and align the feature space. The specific formula is as follows:

[0066]

[0067] wherein represents the strength of controlling the residual injection, and finally fuses the current layer propagation feature with the historical layer residual information after nonlinear activation to output high-order features.

[0068] The space-time sequence encoding module simultaneously depicts the dependency relationship between multiple variables and its evolution law in time, including: time sequence encoding is performed on the obtained variable feature matrix; a time window with a length of is stacked into a three-dimensional tensor of “time, variable and feature”, and time sequence dependency is introduced as the input of the next step of time series prediction network; the change trajectory of each variable in the time window is extracted, and the three-dimensional tensor is regarded as a batch input of multivariate time series.

[0069] Specifically, time sequence encoding is performed on the obtained variable feature matrix. The current feature dimension is , is the number of variables, is the feature dimension of each variable. A time window with a length of The time window stack is a three-dimensional tensor of "time, variable, and feature", which introduces time series dependence as the input of the next step time series prediction network:

[0070]

[0071] wherein represents the three-dimensional tensor of the current time window, represents the stacking operation, which stacks the features of each time point in chronological order to form a time series feature block. Then, the change trajectory of each variable within the time window is extracted, and the three-dimensional tensor is regarded as a multivariate time series for batch input:

[0072]

[0073]

[0074] wherein represents the feature vector of the th variable at time , is a time series sample, is a multivariate time series set.

[0075] The encoded multivariate feature sequence is gradually mapped to the bearing temperature prediction value at the future time point through the recurrent time series prediction module, realizing the modeling and prediction of the change trend of the train wheel bearing temperature. For each time step, the state is updated by combining the layer normalized gated recurrent unit, and the feature dimension of the current time step sample is standardized.

[0076] Specifically, the encoded multivariate feature sequence is gradually mapped to the bearing temperature prediction value at the future time point through the recurrent time series prediction module, realizing the modeling and prediction of the change trend of the train wheel bearing temperature. For each time step, the state is updated by combining the layer normalized gated recurrent unit, and the feature dimension of the current time step sample is standardized. Layer normalization , calculates the mean of the input vector, and the denominator calculates the standard deviation, the scale normalization, is a small value to prevent the denominator from being 0, are the learnable scaling coefficient and the translation coefficient, respectively, which are used to restore the optimal distribution scale required by the network and adjust the mean. The specific formula is as follows:

[0077]

[0078]

[0079]

[0080]

[0081] in Controlling the degree to which the network forgets the past, Controlling the ratio of historical information to candidate new information, This is a candidate hidden state. For variables At time step The hidden state; These are all learnable parameters of the network. Representation layer normalization, It is a non-linear activation function. The hyperbolic tangent function maps the input to the interval (−1, 1). Compared to the zero-centered sigmoid function, training is more stable. This indicates element-wise multiplication.

[0082] A dual-branch fusion structure is introduced for prediction mapping to fully explore the nonlinear relationship between time dependence and candidate features; this structure, on the one hand, affects the final hidden state. Layer normalization and nonlinear function activation are performed to extract stationary features; on the other hand, candidate hidden states are utilized. go through The activation captures its short-term dynamic changes, and the two parts are weighted and fused by linear transformation to obtain the final prediction output.

[0083] Specifically, a dual-branch fusion structure is introduced for prediction mapping to fully explore the nonlinear relationship between time dependence and candidate features. This structure, on the one hand, affects the final hidden state. Layer normalization and nonlinear function activation are performed to extract stationary features; on the other hand, candidate hidden states are utilized. go through The activation captures its short-term dynamic changes, and the two parts are then weighted and fused through a linear transformation to obtain the final prediction output:

[0084]

[0085] in These are the linear mapping matrices for the final hidden state and the candidate hidden states, respectively. These are the bias terms for the two branches. Describe the output layer weight matrix. This represents the bias term of the output layer. For output dimensions, This represents the network's predicted output.

[0086] Construct an automatic graded early warning mechanism for predicted temperature rise, and set early warning thresholds for different operating speeds, specifically including:

[0087] S6-1, set the screening condition. Since the absolute temperature of the bearing is greatly affected by environmental temperature, seasonal changes and other factors, in order to improve the stability of the early warning threshold, the difference between the bearing temperature and the outdoor temperature is used as the temperature rise index, that is , the temperature rise of the bearing temperature of the th train relative to the outdoor temperature at the time is the measured temperature, and is the outdoor temperature; then the train running speed is divided into working conditions, that is , and the uniform speed condition is set. In the time window with a length of centered on , the difference between the maximum and minimum running speeds is calculated , and when is less than the threshold , it is in a uniform speed state. The specific formula is as follows:

[0088]

[0089]

[0090] S6-2, establish the train temperature rise data under different running speeds. Select the time set that meets the speed range and has small fluctuations for different train sets, and calculate the maximum temperature rise of all cars for each time . If the maximum temperature rise does not exceed the upper limit , it is considered that the temperature rise at this time is a healthy sample, and the specific formula is as follows:

[0091]

[0092] For each time of the healthy sample, calculate the median of the temperature rise of each car axle as the representative value of the temperature rise at this time, and at the same time, the absolute deviation of the actual temperature rise of each train set from the median is required to be not more than , otherwise it is removed twice. Under the constraints of speed and temperature rise, the final time set that is considered to be in a healthy running condition is obtained, which is used for subsequent hierarchical threshold calculation. At the same time, the time set is divided according to different speeds, that is

[0093]

[0094] S6-3, multi-level early warning threshold calculation and predicted temperature rise determination. For running speed , calculate the temperature rise mean and standard deviation of the time sequence at :​​

[0095]

[0096]

[0097] wherein is the total number of samples under the operating condition, is the operating condition the first time point temperature rise data. Using criteria, three-level early warning threshold values are constructed within the operating condition

[0098]

[0099] The temperature rise of the S5 prediction result is calculated , wherein is the prediction result at time , the early warning level of the current time point is defined, 0 is normal; 1 is a first-level early warning; 2 is a second-level early warning; 3 is a third-level early warning, and the classification judgment is as follows:

[0100] .

[0101] Example 2

[0102] As shown in Figure 1 , Figure 2 In this embodiment, a train wheelset axle temperature prediction and hierarchical early warning method based on recursive multivariate interaction network is provided, which specifically includes the following steps:

[0103] S1, time series reconstruction and sampling consistency of the original data are performed, and the data set is divided, specifically including the following steps:

[0104] S1-1, time series reconstruction and sampling consistency. The first step is to fill in the blank data. The original data has multiple blank data, so that the data chain is broken and the information is not complete. In addition, the time intervals of data collection are uneven, and the timestamps cannot be consistent, which affects the accuracy of the data and the effectiveness of the subsequent analysis work. Therefore, it is necessary to delete the abnormal time interval, adjust the time interval, and ensure that it is collected at a time interval of 30s; at the same time, repeated data needs to be deleted, and for repeated timestamps and repeated axle temperature data columns in the data, the data structure needs to be reorganized to ensure the accuracy of the data.

[0105] ​1) Linear interpolation for missing value filling. Linear interpolation between adjacent observation points is used to restore the broken time-value chain, so that subsequent resampling and statistical aggregation can be performed on the "complete trajectory", avoiding bias caused by holes. The specific formula is shown in Example 1. On this basis, time interval adjustment and standardization are performed to place uneven time stamps on the same grid, eliminating the discretization error introduced by uneven sampling. Consistent with linear interpolation, the measurement values are time-aligned without changing the physical meaning. , the original time series is resampled to a fixed interval . . Aggregated resampling can also be performed. Let the Kth time window be , and the observation set be . The following are the formulas for mean aggregation and median aggregation, respectively:

[0106]

[0107] 2) Abnormal time interval screening. Define the timestamp difference as:

[0108]

[0109]

[0110] where , are adjacent time intervals. If the time interval is , the index is considered abnormal, where is the tolerance for recording jitter. By depicting the over-dense, over-dense and jump time segments in the recording time interval, the collection anomaly is identified. By comparing the multiple period relationship of , single-step deviation can be detected, and multiple period jumps can also be detected. Specifically, if , the sampling is too dense, and it is mostly repeated or jittered, which needs to be de-duplicated or merged; if , the sampling is too sparse and is considered missing, which needs to be interpolated in the resampling stage, and the interpolation is prohibited for gaps exceeding the threshold.

[0111] 3) Repeat value deletion and consistency reconstruction. For the same timestamp , there are multiple records . By compressing multiple readings at the same time into a unique representative value, the "multiple value conflict" caused by repeated writing at the collection layer is eliminated, ensuring time alignment and uniqueness of the index. The specific formula is as follows:

[0112]

[0113] S1-2, preparing the data set. The data set is processed in batches, The eigenvalues of 2 rows are set as a group, The setting is The 20th row is used as verification, and the first group of data is formed, and the train set is obtained after the train set is processed in batches. The specific calculation formula is as follows:

[0114]

[0115]

[0116] Let the original sequence be , and the sliding window with a length of = 20 is used as input, and the corresponding output is taken , and the data of the first row is obtained, and thus the training sample set

[0117]

[0118] , wherein is the total number of sample groups, , and the first group, the second group, and the group of data are sequentially taken from the original sequence.

[0119] S1-3, dividing the data set. The preprocessed data is divided into a training set and a test set, wherein the test set is further divided into a short-time data set and a long-time data set according to different time spans, and is used to evaluate the prediction accuracy and generalization ability of the model under different prediction scenarios.

[0120] S2, the definition of the multivariate dependent structure specifically includes the following steps:

[0121] S2-1, each time series in the data set is regarded as a variable sequence, and the connection relationship between variables is established according to the correlation, forming an interactive structure composed of a variable set and its dependent relationship. For details, see the embodiment 1.

[0122] S2-2, adaptive construction and normalization processing of the multivariate correlation weight matrix. The specific process includes three parts of correlation weight calculation, weighted degree matrix calculation, and symmetric normalization. First, according to the correlation between time series, the correlation weight element is defined as an exponential decay weight between time steps, representing the influence of variable and at time step The role weight of each variable. A weighted measurement matrix is introduced to measure the "overall connection degree" between each time step and its adjacent time step, and the correlation weight matrix is symmetrically normalized to obtain a standardized interaction weight matrix.

[0123] S3, constructing a recursive multivariate interaction network, the specific steps are as follows:

[0124] S3-1, inputting multivariate features. The i-th variable and its associated variables The single-head attention coefficient calculation method is as shown in embodiment 1.

[0125] S3-2, attention weight normalization, all attention coefficients related to the i-th variable are normalized to ensure that the sum of the attention coefficients is 1, and a standardized attention weight is obtained.

[0126] S3-3, attention-based neighborhood information fusion. Through weighted summation and self-loop compensation, each time variable can absorb the associated information of the neighbors and retain its own features when updating, realizing stable and flexible time sequence feature aggregation.

[0127] S3-4, gathering the vectors of single-head information transmission of each layer, using multiple sets of attention weights to realize multi-head attention mechanism, so that the variable can learn the neighborhood dependency relationship in different feature subspaces, and realize multi-scale feature aggregation.

[0128] S3-5, multi-layer recursive propagation and residual enhancement. This step enhances the features from local representation to global modeling, so that the variable can aggregate the features of the related variables at a long distance. The normalized interaction weight matrix performs feature aggregation on the current variable feature performs matrix multiplication with the linear variation matrix to realize feature space mapping, and then fuses the features of the upper layer , uses to adjust the dimension, aligns the feature space, and the specific formula is as shown in embodiment 1.

[0129] S4, space-time sequence encoding module: after the multi-layer stacking of the recursive multivariate interaction network, the variable feature matrix obtained can only reflect the information of the current time, and cannot reflect the dynamic characteristics of the variable evolution over time, so time sequence encoding is needed, and the specific steps are as shown in embodiment 1.

[0130] S5, through the recursive time sequence prediction module, the encoded multivariate feature sequence is gradually mapped to the bearing temperature prediction value at the future time, realizing modeling and prediction of the change trend of the train wheel bearing temperature, and the specific steps are as shown in embodiment 1.

[0131] ​S6, the variable features obtained by the recursive time series prediction module have fully fused the spatial dependence and time evolution information. However, the hidden state still needs to be further mapped to the specific prediction target space, so in order to fully excavate the nonlinear relationship between the time dependence and the candidate features, a double-branch fusion structure is introduced for prediction mapping. The structure performs layer normalization and nonlinear function activation on the final hidden state , extracts stable features, and on the other hand, uses the candidate hidden state to capture its short-term dynamic changes through activation, and after linear transformation and weighted fusion of the two parts, the final prediction output is obtained.

[0132] In this embodiment, a prediction temperature rise automatic grading early warning mechanism is constructed, and the specific steps are as shown in embodiment 1, which will not be repeated here.

[0133] As shown in Figures 3 to 7 , in this embodiment, the data used is collected from the bogie bearing data of a certain motor train set G1-16 from July 1, 2023 to August 29, 2023 for a total of 60 days, and the time data of 12 features of each train is measured from the temperature sensor installed on each axle box of the motor train set, as shown in Table 1 below.

[0134] Table 1. Time feature table of experimental data

[0135]

[0136] In the specific data analysis and network verification, 8 groups of bogie bearing temperature data are mainly used, i.e. the feature number is 8, and the train axle temperature data is shown in Table 2 and Figure 3 .

[0137] Table 2. Experimental data examples

[0138]

[0139] In the specific experiment process, the train set G16 train is divided into a training set and a test set, in order to fully illustrate the effectiveness of the method, a long-time span state prediction task and a short-time span state prediction task are constructed, and the specific introduction is as follows: the axle temperature data of cars 1-3 is the training set, and the test set is divided into long and short time two parts: the short-time test set selects the axle temperature data of cars 4, 5 and 6 for 3 days, and the long-time test set selects the data of cars 7, 8 and 9 for 7 days, and the specific time range and data size are shown in Table 3.

[0140] Table 3. Training and test data set size

[0141]

[0142] In order to quantify the performance of different networks on the test training set, and measure the generalization ability and prediction accuracy of the network, the commonly used evaluation indicators for prediction are used, that is, mean absolute error (MAE), mean absolute percentage error (MAPE) and root mean square error (RMSE) are used to evaluate the prediction effect of the network.

[0143] a) MAE measures the absolute difference between the predicted value and the true value in each dimension, and takes the overall average. This indicator can intuitively reflect the prediction error level, and has strong robustness to abnormal values:

[0144]

[0145] wherein is the predicted value of the i-th row and the j-th column, is the true value of the i-th row and the j-th column, and are the number of rows and columns, respectively. b) MAPE measures the proportion of the prediction error relative to the true value, which can provide more comparable results on data of different dimensions and different scales:

[0146] c) RMSE squares the error term in the calculation process, thereby amplifying the influence of large deviation samples, and can more sensitively reflect the large errors in prediction. It is an important indicator when emphasizing the robustness of the network:

[0147]

[0148] The method proposed in the present application is used for rolling bearing temperature prediction task in the experiment, and is compared with three commonly used time series prediction methods, that is, gated recurrent unit (GRU), bidirectional long short-term memory network (BiLSTM) and graph attention network (GAT). The parameter settings are shown in Table 4.

[0149] Table 4. Parameter settings

[0150]

[0151]

[0152] ​​​​

[0153] 1) Shaft temperature prediction results: The experimental data is substituted into the network of the embodiment and the comparative network, and the average value of the three indexes of eight characteristic values is taken as the evaluation result for comparison. The experimental results are shown in Tables 5 and 6.

[0154] Table 5. Comparison results of short-term prediction

[0155]

[0156] Table 6. Comparison results of long-term prediction

[0157]

[0158] The experimental results show that the method is significantly better than the single comparative network in the shaft temperature prediction task of different time spans, and can more effectively depict the complex patterns and multiple dependence relationships in the sequence. In addition, the network advantage of the method is more obvious under the condition of long time sequence, which may be due to the fact that the long sequence contains more rich time dependence and potential structure characteristics, which is beneficial to the network to accumulate and utilize cross-scale information. By jointly modeling the multivariate interaction information and the time sequence characteristics, the method can more fully capture the long-term dependence relationship, and significantly improve the accuracy and reliability of shaft temperature prediction under various prediction steps and data distributions.

[0159] 2) Automatic grading warning of predicted temperature rise

[0160] First, the temperature rise data set under the health condition is constructed according to the train running speed and temperature rise constraint. In the load condition of 480-550KN, a certain time window with small speed fluctuation is selected for uniform speed running period, and the abnormal data whose shaft temperature rise exceeds the upper limit is removed. At the same time, the shaft temperature rise of cars 1-16 in the same time period is compared horizontally, and the median of the temperature rise of the multi-car group is taken as the reference. The data deviating more than 10℃ from the median is regarded as suspicious abnormality and is removed again, so as to construct the temperature rise data set. On this basis, the mean and standard deviation under different running speeds are calculated, and the grading warning threshold is constructed. The calculation results are shown in Table 7.

[0161] Table 7. Calculation results of warning threshold under different running speeds

[0162]

[0163] The short-time prediction result of No. 5 vehicle and the long-time prediction result of No. 7 vehicle are respectively substituted into the constructed temperature rise grading early warning threshold discrimination mechanism. In combination with the train running speed conditions, the prediction temperature rise data of No. 5 vehicle in the high-speed condition is selected for early warning discrimination, and the prediction temperature rise data of No. 7 vehicle in the low-speed condition is selected for early warning discrimination. Through the grading discrimination of the above two types of prediction results under the early warning threshold system corresponding to different speed conditions, the applicability and effectiveness of the method under different running conditions are verified. The discrimination result shows that with the train running, the predicted axle temperature rise of No. 5 vehicle and No. 7 vehicle both triggers a first-level warning.

[0164] Embodiment 3

[0165] The embodiment 3 provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implements a train wheelset axle temperature prediction grading early warning method based on a recursive multivariate interaction network as described above, the method comprising:

[0166] obtaining train running speed data;

[0167] processing the obtained speed data by using a pre-trained train wheelset axle temperature prediction grading model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheelset axle temperature prediction grading model comprises: a multivariate dependent structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient feature information retention; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time series prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; introducing a double-branch fusion structure for prediction mapping, fully mining the nonlinear relationship between time dependence and candidate features;

[0168] Based on the prediction temperature rise automatic grading early warning mechanism, it is judged whether the axle bearing temperature prediction value exceeds the corresponding threshold value according to the set early warning threshold value under different running speeds, and if the threshold value is exceeded, early warning processing is performed.

[0169] Embodiment 4

[0170] The embodiment 4 provides a computer device, comprising a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions executable by the processor, and the processor invokes the program instructions to execute a train wheelset axle temperature prediction grading early warning method based on a recursive multivariate interaction network as described above, the method comprising:

[0171] obtaining train running speed data;

[0172] The obtained speed data is processed by using a pre-trained train wheelset axle temperature prediction grading model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheelset axle temperature prediction grading model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient retention of feature information; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time series prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; a double-branch fusion structure is introduced for prediction mapping, and the nonlinear relationship between time dependence and candidate features is fully mined;

[0173] Based on the automatic grading early warning mechanism of the predicted temperature rise, whether the axle bearing temperature prediction value exceeds the corresponding threshold is judged according to the set warning threshold under different running speeds, and if the threshold is exceeded, the early warning processing is performed.

[0174] Embodiment 5

[0175] The embodiment 5 provides an electronic device, comprising: a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the instructions for realizing the train wheelset axle temperature prediction grading early warning method based on the recursive multivariate interaction network as described above, which comprises:

[0176] Obtaining train running speed data;

[0177] The obtained speed data is processed by using a pre-trained train wheelset axle temperature prediction grading model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheelset axle temperature prediction grading model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient retention of feature information; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time series prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; a double-branch fusion structure is introduced for prediction mapping, and the nonlinear relationship between time dependence and candidate features is fully mined;

[0178] Based on the automatic grading early warning mechanism of the predicted temperature rise, whether the axle bearing temperature prediction value exceeds the corresponding threshold is judged according to the set warning threshold under different running speeds, and if the threshold is exceeded, the early warning processing is performed.

[0179] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0180] The present application is described in relation to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present application. It is to be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 means for performing each of the functions specified in the flow diagram and / or block diagram block or blocks.

[0181] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 means for performing each of the functions specified in the flow diagram and / or block diagram block or blocks.

[0182] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow diagram and / or block diagram block or blocks. Figure 1 one or more functions specified in the flow diagram and / or block diagram block or blocks. Figure 1 means for performing each of the functions specified in the flow diagram and / or block diagram block or blocks.

[0183] Although the present application has been described in connection with the preferred embodiments thereof with reference to the drawings, it will be apparent to those of ordinary skill in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. Accordingly, it is intended that all such changes come within the scope of the following claims.

Claims

1. A train wheelset axle temperature prediction hierarchical early warning method based on recursive multivariate interaction network, characterized in that, The method comprises the following steps: obtaining train running speed data; processing the obtained speed data by using a pre-trained train wheel set axle temperature prediction hierarchical model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheel set axle temperature prediction hierarchical model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and standardized interaction weights are obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient feature information retention; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time sequence prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; a double-branch fusion structure is introduced for prediction mapping, and the nonlinear relationship between time dependence and candidate features is fully mined; based on the automatic hierarchical early warning mechanism of the predicted temperature rise, whether the axle bearing temperature prediction value exceeds the corresponding threshold value is judged according to the set early warning threshold value under different running speeds, and if the threshold value is exceeded, early warning processing is performed.

2. The method according to claim 1, wherein the method is characterized by, A multivariate correlation weight matrix is constructed, and a standardized interaction weight is obtained through normalization processing, including: regarding each time series in the data set as a variable sequence, establishing a connection relationship between variables according to correlation to form an interaction structure composed of a variable set and its dependency relationship; defining a correlation weight element is an exponential decay weight between time steps, representing the variable and is the action weight of the variable at time step .

3. The method of claim 1, wherein the method is a hierarchical warning method for train wheelset axle temperature prediction based on recursive multivariable interaction networks. The recursive multivariate interaction network more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient feature information retention, which comprises: For the input multivariate features, the single-head attention coefficient of the ith variable and its associated variables is calculated , all attention coefficients related to the ith variable are normalized to ensure that the sum of the attention coefficients is 1, and standardized attention weights are obtained; through weighted summation and self-loop compensation, each time variable can absorb the associated information of the neighbors and retain its own characteristics when updating, achieving stable and flexible time series feature aggregation; the vectors of single-head information transmission of each layer are converged, multiple sets of attention weights are used for parallel calculation to realize the multi-head attention mechanism, so that the variables can learn the neighborhood dependency relationship in different feature subspaces, and multi-scale feature aggregation is realized; the network is enhanced from local representation to global modeling, so that the variables can aggregate the features of long-distance related variables; the normalized interaction weight matrix is used to perform feature aggregation on the current variable features , and then multiplied by the linear transformation matrix to realize feature space mapping, and then the features of the upper layer are fused , the dimension is adjusted using to align the feature space.

4. The method of claim 1, wherein the method is a hierarchical warning method for train wheelset axle temperature prediction based on recursive multivariable interaction networks. The spatio-temporal sequence encoding module simultaneously depicts the dependency relationship among the multivariate and the evolution law thereof in time, including: performing time sequence encoding on the obtained variable feature matrix; stacking time windows with a length of 3 into a three-dimensional tensor of "time, variable and feature", introducing time sequence dependency, and taking the three-dimensional tensor as an input of a next step time sequence prediction network; and extracting a change trajectory of each variable in the time window, taking the three-dimensional tensor as a batch input of multivariate time sequence.

5. The method of claim 1, wherein the method is a hierarchical warning method for train wheelset axle temperature prediction based on recursive multivariable interaction networks. The recursive time sequence prediction module gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time, realizes the modeling and prediction of the train wheel set axle bearing temperature change trend; for each time step, the state is updated by combining the layer normalized gated recurrent unit, and the feature dimension of the sample at the current time step is standardized.

6. The method of claim 1, wherein the method is a hierarchical warning method for train wheelset axle temperature prediction based on recursive multivariable interaction networks. A double-branch fusion structure is introduced for prediction mapping, which fully excavates the time dependence and the nonlinear relationship of candidate features. The structure performs layer normalization and nonlinear function activation on the final hidden state to extract stationary features. Another aspect utilizes candidate hidden states After The short-term dynamic changes of the captured are activated, and the two parts are fused after linear transformation and weighting to obtain the final prediction output.

7. A train wheelset axle temperature prediction and hierarchical warning system based on recursive multivariate interaction network, characterized in that, The method comprises the following steps: an acquisition module, configured to acquire train running speed data; a prediction module, configured to process the acquired speed data by using a pre-trained train wheel set axle temperature prediction hierarchical model to obtain an axle bearing temperature prediction value under the running speed; wherein the train wheel set axle temperature prediction hierarchical model comprises: a multivariate dependence structure, a multivariate correlation weight matrix is constructed, and standardized interaction weights are obtained through normalization processing; a recursive multivariate interaction network, which more comprehensively depicts the complex multivariate dependence within a time window under the premise of ensuring sufficient feature information retention; a space-time sequence encoding module, which simultaneously depicts the dependence relationship between multivariate and its evolution law in time; a recursive time sequence prediction module, which gradually maps the encoded multivariate feature sequence into the axle bearing temperature prediction value at the future time; a double-branch fusion structure is introduced for prediction mapping, and the nonlinear relationship between time dependence and candidate features is fully mined; an early warning module, configured to judge whether the axle bearing temperature prediction value exceeds the corresponding threshold value based on the automatic hierarchical early warning mechanism of the predicted temperature rise according to the set early warning threshold value under different running speeds, and perform early warning processing if the threshold value is exceeded.

8. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is used to store computer instructions, which are executed by a processor to implement the train wheel set axle temperature prediction hierarchical early warning method based on the recursive multivariate interaction network according to any one of claims 1-6.

9. A computer device, comprising: The application discloses a train wheel set axle temperature prediction and hierarchical early warning method based on a recursive multivariate interaction network.

10. An electronic device, comprising: The application discloses a train wheel set axle temperature prediction and hierarchical early warning method based on a recursive multivariate interaction network. The application discloses a train wheel set axle temperature prediction and hierarchical early warning method based on a recursive multivariate interaction network.

Images