Method, device, equipment and medium for monitoring operation state of electromagnetic suspension system

Abstract

The present application discloses a method, device, equipment and medium for monitoring the operating state of an electromagnetic levitation system, which relates to the cross technical field of transportation and artificial intelligence, and includes: constructing an attribute configuration table containing the corresponding relationship between the train operating state and state attributes based on the engineering knowledge and experience data of the maglev train; determining a preset feature extractor based on the attention mechanism and wavelet function, and combining with the sliding window function to extract features from the test samples, and performing multi-scale analysis on the feature segments based on the wavelet transform technology; training an initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table, and judging the test samples based on the gating model constructed during the training process; using the target attribute learner to predict the attributes of the feature vector, determining the monitoring result based on the judgment result, attribute value and nearest neighbor search rule, and taking warning and maintenance measures. To improve the accuracy of monitoring the operating state of the electromagnetic levitation system.

Description

Technical Field

[0001] The present invention relates to the cross - technical field of transportation and artificial intelligence, and particularly relates to a method, device, equipment and medium for monitoring the operation state of an electromagnetic levitation system. Background Art

[0002] Maglev trains are fast, have low noise, and are safe, environmentally friendly, and are becoming a research hotspot in the field of modern transportation. The electromagnetic levitation system needs to build a complete state monitoring system to ensure the safe and stable operation of the train. However, the current state monitoring research discusses the problem of extremely unbalanced samples less. The occurrence probability of some states of the maglev train is extremely low, and the number of samples available for training the model is extremely low (or even none). The system may be affected by external environmental interference and present new states; moreover, traditional monitoring methods cannot effectively monitor them. At the same time, the time - series information of the electromagnetic levitation system is large in quantity and low in information density, making it difficult to capture the features contained in the sequence.

[0003] As can be seen from the above, how to prevent inaccurate monitoring of the operation state of the electromagnetic levitation system due to extremely unbalanced samples with low information density is an urgent problem to be solved at present. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for monitoring the operation state of an electromagnetic levitation system, which can prevent inaccurate monitoring of the operation state of the electromagnetic levitation system due to extremely unbalanced samples with low information density. The specific solutions are as follows:

[0005] In a first aspect, the present application provides a method for monitoring the operation state of an electromagnetic levitation system, including:

[0006] Constructing an attribute configuration table containing the correspondence between the train operation state and state attributes based on the engineering knowledge and experience data of the maglev train; the train operation states include normal operation, passing through joints, low - frequency swaying, high - frequency vibration, and rail pounding;

[0007] Obtaining test samples by using the operation state data of the maglev train during operation, then extracting features from the test samples based on a preset feature extractor and a sliding window function to obtain feature segments, and performing multi - scale analysis on the feature segments based on wavelet transform technology to obtain feature vectors; the operation state data includes known operation states and unknown operation states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function;

[0008] Train the initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model, so as to obtain a judgment result;

[0009] Use the target attribute learner to predict the attributes of the feature vector to obtain corresponding attribute values, and then determine the monitoring result based on the judgment result and the attribute values, and combine the nearest neighbor search rule to take corresponding warning and maintenance measures based on the monitoring result.

[0010] Optionally, the obtaining of the test sample by using the operating state data of the maglev train during operation includes:

[0011] Use a suspension data acquisition system to collect the operating state data of the maglev train during operation, and determine each operating state data as a test sample;

[0012] Wherein, the suspension data acquisition system includes a data collector, a suspension gap sensor, a current sensor, an acceleration sensor, a temperature sensor, a WIFI network center and an embedded upper computer software center; the operating state data includes the suspension gap, suspension acceleration, lateral acceleration, working current of the suspension electromagnet and the surface temperature of the suspension electromagnet coil of the maglev train during operation.

[0013] Optionally, the feature extraction of the test sample based on a preset feature extractor and a sliding window function to obtain a feature segment includes:

[0014] Segment the test sample based on a preset sliding window length and a preset sliding step to obtain each sliding window, and use the attention mechanism to calculate the mean value, the first energy value and the difference between the first peaks of each sliding window to obtain corresponding calculation results;

[0015] Use the calculation results to determine a first attention weighting function to determine the target window segment with the largest weight value based on the first attention weighting function;

[0016] Determine a second attention weighting function based on the second energy value and the difference between the second peaks of the first-order difference sequence corresponding to the sliding window, and use the second attention weighting function to determine the key features in the target window segment to obtain a feature segment.

[0017] Optionally, the multi-scale analysis of the feature segment based on the wavelet transform technology to obtain a feature vector includes:

[0018] The preset wavelet function is dilated and translated using a preset scale parameter to obtain a target wavelet function, and the inner product is calculated based on the target wavelet function and the feature segment to obtain wavelet coefficients at different scales. The feature vector is obtained by extracting the wavelet coefficients.

[0019] Optionally, training the initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and constructing a gating model during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model to obtain a judgment result, including:

[0020] Training the initial attribute learner based on the generalized zero-shot learning method, machine learning algorithm, and the attribute configuration table to obtain a target attribute learner;

[0021] Constructing a gating model during the training of the initial attribute learner, and obtaining the feature center values of each of the known operating states in the feature space of the gating model;

[0022] Calculating the distribution direction similarity and the distribution distance similarity between the test sample and the feature center value respectively to obtain a first similarity calculation result and a second similarity calculation result;

[0023] Combining the first similarity calculation result and the second similarity calculation result to obtain a comprehensive similarity score, determining the highest comprehensive similarity score as the target similarity score, and judging whether the target similarity score is greater than a preset similarity threshold;

[0024] If the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state;

[0025] If the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state.

[0026] Optionally, the calculating the distribution direction similarity and the distribution distance similarity between the test sample and the feature center value respectively to obtain a first similarity calculation result and a second similarity calculation result includes:

[0027] Determining a first vector and a second vector based on the feature center value and the test sample;

[0028] Calculating the cosine value of the first vector and the second vector to obtain a first similarity calculation result representing the distribution direction similarity between the feature center value and the test sample;

[0029] By calculating the ratio of the norms of the first vector and the second vector, a second similarity calculation result is obtained to characterize the similarity of the distance between the characteristic central value and the distribution of the test samples.

[0030] Optionally, using the target attribute learner to perform attribute prediction on the feature vector to obtain the corresponding attribute value, and then based on the judgment result and the attribute value, and combining the nearest neighbor search rule to determine the monitoring result, and taking corresponding warning and maintenance measures based on the monitoring result, including:

[0031] Using the target attribute learner to perform attribute prediction on the feature vector to obtain the corresponding attribute value. If the judgment result indicates that the test sample is in a known operating state, calculate the first distance between the attribute value and the attribute vectors corresponding to each known operating state, and select the known operating state with the smallest first distance as the target known operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target known operating state;

[0032] If the judgment result indicates that the test sample is in an unknown operating state, calculate the second distance between the attribute value and the attribute vectors corresponding to each unknown operating state, and select the unknown operating state with the smallest second distance as the target unknown operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target unknown operating state.

[0033] In a second aspect, the present application provides an electromagnetic suspension system operating state monitoring device, including:

[0034] An attribute configuration table construction module, configured to construct an attribute configuration table including the correspondence between the train operating state and the state attributes based on the engineering knowledge and experience data of the maglev train; the train operating states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail pounding;

[0035] A feature extraction module, configured to use the operating state data of the maglev train during operation to obtain test samples, and then perform feature extraction on the test samples based on a preset feature extractor and a sliding window function to obtain feature segments, and perform multi-scale analysis on the feature segments based on wavelet transform technology to obtain feature vectors; the operating state data includes known operating states and unknown operating states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function;

[0036] An attribute learner training module, configured to train an initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model, so as to obtain a judgment result;

[0037] An attribute value prediction unit, configured to perform attribute prediction on the feature vector by using the target attribute learner to obtain corresponding attribute values, and then determine a monitoring result based on the judgment result and the attribute values in combination with the nearest neighbor search rule, so as to take corresponding warning and maintenance measures based on the monitoring result.

[0038] In a third aspect, the present application provides an electronic device, including:

[0039] A memory, configured to store a computer program;

[0040] A processor, configured to execute the computer program to implement the foregoing method for monitoring the operating state of an electromagnetic suspension system.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium, configured to store a computer program, where the computer program, when executed by a processor, implements the foregoing method for monitoring the operating state of an electromagnetic suspension system.

[0042] The present application constructs an attribute configuration table including the correspondence between the train operating state and the state attributes based on the engineering knowledge and experience data of the maglev train; the train operating states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail pounding; the test samples are obtained by using the operating state data of the maglev train during operation, and then feature extraction is performed on the test samples based on a preset feature extractor and a sliding window function to obtain feature segments, and multi-scale analysis is performed on the feature segments based on wavelet transform technology to obtain feature vectors; the operating state data includes known operating states and unknown operating states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function; an initial attribute learner is trained based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and a gating model is constructed during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model, so as to obtain a judgment result; the target attribute learner is used to perform attribute prediction on the feature vector to obtain corresponding attribute values, and then a monitoring result is determined based on the judgment result and the attribute values in combination with the nearest neighbor search rule, so as to take corresponding warning and maintenance measures based on the monitoring result.

[0043] As can be seen from the above, in this application, an attribute configuration table is constructed by integrating the engineering knowledge and experience data of the maglev train. Then, a preset feature extractor determined by the attention mechanism and wavelet function and wavelet transform technology are used to extract and perform multi-scale analysis on the operation state data during the operation of the maglev train to obtain feature vectors. Then, based on the generalized zero-shot learning method and the attribute configuration table, the initial attribute learner is trained to obtain a target attribute learner, and a gating model is constructed during the training process to use the gating model to determine whether the test sample is a known operation state or an unknown operation state. Then, based on the attribute learner, attribute prediction is performed on the feature vector to obtain an attribute value. In this way, based on the judgment result and the attribute value, and combined with the nearest neighbor search rule, the final monitoring result is determined. Through the monitoring result, potential faults or abnormal conditions of the maglev train can be discovered in time, and corresponding early warning and maintenance measures can be taken, greatly reducing the fault risk and operation cost of the maglev train and improving the operation safety and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0045] Figure 1 Flowchart of a method for monitoring the operation state of an electromagnetic suspension system disclosed in this application;

[0046] Figure 2 Schematic structural diagram of an EMS-type maglev train provided by this application;

[0047] Figure 3 Schematic diagram of an attribute configuration provided by this application;

[0048] Figure 4 Schematic diagram of the attribute meanings corresponding to different working conditions provided by this application;

[0049] Figure 5 Schematic structural diagram of a suspension data acquisition system provided by this application;

[0050] Figure 6 Flowchart of capturing feature segments by an attention mechanism provided by this application;

[0051] Figure 7 Schematic diagram of the attribute generalized zero-shot learning process provided by this application;

[0052] Figure 8A schematic diagram of the feature space provided for this application;

[0053] Figure 9 A specific method for monitoring the operating state of an electromagnetic suspension system disclosed in this application;

[0054] Figure 10 A schematic diagram of the evaluation index provided for this application;

[0055] Figure 11 A schematic diagram of the online evaluation accuracy rate for monitoring the operating state based on traditional machine learning provided for this application;

[0056] Figure 12 A schematic diagram of the online evaluation accuracy rate for monitoring the operating state based on generalized zero-shot learning provided for this application;

[0057] Figure 13 A schematic diagram of the confusion matrix of the online evaluation results for monitoring the operating state based on traditional machine learning provided for this application;

[0058] Figure 14 A schematic diagram of the confusion matrix of the online evaluation results for monitoring the operating state based on generalized zero-shot learning provided for this application;

[0059] Figure 15 A schematic diagram of the scores of the online evaluation results for monitoring the operating state based on generalized zero-shot learning provided for this application;

[0060] Figure 16 A schematic diagram of the structure of a device for monitoring the operating state of an electromagnetic suspension system disclosed in this application;

[0061] Figure 17 A structural diagram of an electronic device disclosed in this application. Detailed implementation manners

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0063] At present, there is little discussion on the problem of extremely unbalanced samples in the research on the state monitoring of electromagnetic suspension systems. The occurrence probability of some states of maglev trains is extremely low, and the number of samples available for training the model is extremely low. The system may be affected by external environmental interference and present new states. Moreover, traditional monitoring methods cannot effectively monitor them. The time series information of electromagnetic suspension systems is large in quantity and low in information density, making it difficult to capture the features contained in the sequence. For this reason, this application provides a method for monitoring the operating state of an electromagnetic suspension system. Based on the judgment result and the attribute value, and combined with the nearest neighbor search rule, the final monitoring result is determined. Through the monitoring result, potential faults or abnormal conditions of the maglev train can be discovered in time, and corresponding early warning and maintenance measures can be taken, greatly reducing the fault risk and operating cost of the maglev train, and improving the safety and reliability of operation.

[0064] See Figure 1 As shown, an embodiment of the present invention discloses a method for monitoring the operating state of an electromagnetic suspension system, including:

[0065] Step S11, constructing an attribute configuration table based on the engineering knowledge and empirical data of the maglev train, which includes the correspondence between the train operating state and the state attributes; the train operating states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail hitting.

[0066] In this embodiment, Figure 2 is a schematic structural diagram of an EMS-type maglev train provided in this embodiment. A maglev train consists of three carriages and multiple suspension frame modules, and realizes suspension through the interaction between electromagnets and rails. When the train is running, it will encounter abnormal states such as passing through joints, low-frequency swaying, high-frequency vibration, and rail hitting. These states will cause different changes in parameters such as clearance, acceleration, and current, thereby affecting the stability and safety of the train. Figure 3 is a schematic diagram of an attribute configuration provided in this embodiment. An attribute configuration table including the correspondence between the train operating state and the state attributes is constructed based on the engineering knowledge and empirical data of the maglev train. Among them, grade#1, grade#2, grade#3, grade#4, and grade#5 respectively represent five working conditions of the train in normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail hitting. Figure 4 is a schematic diagram of the attribute meanings corresponding to different working conditions provided in this embodiment, which summarizes seven characteristics of the suspension system, including abnormal clearance, jitter conditions, and strong correlations among clearance, current, and acceleration, for evaluating the system state.

[0067] Step S12: Obtain test samples using the operating state data of the maglev train during operation. Then, based on a preset feature extractor and a sliding window function, perform feature extraction on the test samples to obtain feature segments, and perform multi-scale analysis on the feature segments based on wavelet transform technology to obtain feature vectors; the operating state data includes known operating states and unknown operating states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function.

[0068] In this embodiment, Figure 5 is a schematic structural diagram of a suspension data acquisition system provided in this embodiment. Near the bogie and guide rail of the maglev train, a plurality of sensors are arranged to monitor the operating state of the train. Among them, represents the air gap value of the maglev train, with the unit of mm, is the acceleration of the maglev train, with the unit of , is the current in the electromagnet, with the unit of A, and u is the voltage on the electromagnet, with the unit of V. The suspension data acquisition system composed of a data collector, a suspension gap sensor, a current sensor, an acceleration sensor, a temperature sensor, a WIFI network center, and an embedded upper computer software center is used to collect the suspension gap, vertical acceleration, lateral acceleration, working current of the electromagnet, and surface temperature of the electromagnet coil during the debugging process of the maglev train to obtain the operating state data, and use the operating state data as test samples. Specifically, the obtaining of test samples using the operating state data of the maglev train during operation includes: using the suspension data acquisition system to collect the operating state data of the maglev train during operation, and determining each of the operating state data as a test sample; among them, the suspension data acquisition system includes a data collector, a suspension gap sensor, a current sensor, an acceleration sensor, a temperature sensor, a WIFI network center, and an embedded upper computer software center; the operating state data includes the suspension gap, suspension acceleration, lateral acceleration, working current of the suspension electromagnet, and surface temperature of the suspension electromagnet coil of the maglev train during operation.

[0069] Further, after obtaining the test sample, segment the test sample based on a preset sliding window length and a preset sliding step size to obtain each sliding window, and use the attention mechanism to calculate the mean value, the first energy value, and the difference between the first peaks of each sliding window, so as to determine a first attention weighting function based on the mean value, the first energy value, and the difference between the first peaks of the sliding window, and determine the target window segment with the largest weight value based on the first attention weighting function. Then, calculate the second energy value and the difference between the second peaks of the first-order difference sequence corresponding to the sliding window, so as to determine a second attention weighting function based on the second energy value and the difference between the second peaks of the first-order difference sequence corresponding to the sliding window, and use the second attention weighting function to determine the key features in the target window segment to obtain the feature segment.

[0070] Specifically, the feature extraction of the test sample based on a preset feature extractor and a sliding window function to obtain a feature segment includes: segmenting the test sample based on a preset sliding window length and a preset sliding step size to obtain each sliding window, and using the attention mechanism to calculate the mean value, the first energy value, and the difference between the first peaks of each sliding window to obtain corresponding calculation results; using the calculation results to determine a first attention weighting function to determine the target window segment with the largest weight value based on the first attention weighting function; determining a second attention weighting function based on the second energy value and the difference between the second peaks of the first-order difference sequence corresponding to the sliding window, and using the second attention weighting function to determine the key features in the target window segment to obtain the feature segment. It should be noted that the preset sliding window length and the preset sliding step size can be adjusted according to the actual situation and are not specifically limited here.

[0071] Figure 6 The flowchart of capturing the feature segment by the attention mechanism provided in this embodiment combines mathematical information such as the mean value and the standard deviation with the attention mechanism, and constructs different types of attention weighting functions and sliding windows to capture regions with significantly different attention weight values and complete the feature segment capture tasks with different requirements. In a specific implementation, for the sliding window , with a length of N, the corresponding mean value calculation formula is: ; the calculation formula for the corresponding first energy value is:

[0072] ;

[0073] where is the mean value. The calculation formula for the difference between the first peaks is:

[0074] ;

[0075] where is the maximum value of the sliding window; is the minimum value of the sliding window. The calculation formula of the first attention weighting function obtained based on the above three formulas is:

[0076] ;

[0077] where and are parameters for dynamically adjusting the power according to the data distribution characteristics and task requirements, and can be adjusted according to actual needs, and no specific limitation is made here. Calculate the first attention weight value of each sliding window using the first attention weighting function, and determine the sliding window with the largest first attention weight value as the target window segment. Then, based on the sliding window, determine the corresponding first-order difference sequence as: ; The calculation formula for calculating the second energy value of the first-order difference sequence is:

[0078] ;

[0079] The calculation formula for calculating the difference between the second peaks of the first-order difference sequence is: ; where is the maximum value in the first-order difference sequence; is the minimum value in the first-order difference sequence. The second attention weighting function obtained by combining the second energy value and the difference between the second peaks of the first-order difference sequence is: ; where and are parameters for dynamically adjusting the power according to the data distribution characteristics and task requirements, and can be adjusted according to actual needs, and no specific limitation is made here.

[0080] In this embodiment, after obtaining the feature segment, a preset wavelet function is scaled and translated using a preset scale parameter to obtain a target wavelet function. Based on the continuous wavelet function in the target wavelet function, the inner product of the continuous wavelet function and the continuous signal in the feature segment is determined to obtain a first wavelet coefficient. The inner product of the discrete wavelet function in the target wavelet function and the discrete signal in the feature segment is determined to obtain a second wavelet coefficient, and a feature vector is obtained by extracting the first wavelet coefficient and the second wavelet coefficient. Among them, common wavelet functions include: Haar (i.e., Haar wavelet function), Daubechies (i.e., Daubechies wavelet function), Biorthogonal (i.e., biorthogonal wavelet function). Specifically, the multi-scale analysis of the feature segment based on the wavelet transform technology to obtain a feature vector includes: scaling and translating a preset wavelet function using a preset scale parameter to obtain a target wavelet function, and performing an inner product calculation based on the target wavelet function and the feature segment to obtain wavelet coefficients at different scales, and extracting the wavelet coefficients to obtain a feature vector.

[0081] It can be understood that the preset scale parameter is determined according to the actual situation, and the preset wavelet function is scaled and translated using the preset scale parameter to obtain a target wavelet function. The formula of the target wavelet function is:

[0082] ;

[0083] Wherein, is the scaling parameter, is the translation parameter, is the translation factor, is the target wavelet function, is the preset scale parameter.

[0084] The formula of the continuous wavelet function is: ;

[0085] Wherein, is the complex conjugate of the target wavelet function; for the discrete signal in the feature segment, common discrete parameter selections are: ; wherein, and are respectively the integer indices of scaling and moving.

[0086] The formula of the discrete wavelet function is: ; Calculate the inner product of the continuous signal in the feature segment and different wavelet functions to obtain corresponding wavelet coefficients; extract the wavelet coefficients to obtain a feature vector. The calculation formula of the feature vector is:

[0087] ;

[0088] wherein, is the energy of the wavelet coefficient, is the standard deviation of the wavelet coefficient, is the mean value of the wavelet coefficient. The corresponding energy feature extraction formula is:

[0089] ;

[0090] wherein, represents the wavelet coefficients within this frequency band; represents the number of wavelet coefficients. It is worth mentioning that, according to actual requirements, some or all of the wavelet coefficients can be selected as the extraction objects. In addition, the wavelet coefficients can also be processed such as dimensionality reduction and normalization, which are not specifically limited herein.

[0091] Step S13: Train the initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model, so as to obtain a judgment result.

[0092] In this embodiment, after obtaining the feature vector, train the initial attribute learner based on the generalized zero-shot learning method, machine learning algorithm, and the attribute configuration table to obtain a target attribute learner. During the training of the initial attribute learner, construct a gating model for operating state shunting, and obtain the feature center values of each known operating state in the feature space in the gating model, so as to calculate the distribution direction similarity and distribution distance similarity between the test sample and the feature center values respectively based on the feature center values, and comprehensively obtain a comprehensive similarity score for the obtained first similarity calculation result and second similarity calculation result, and determine whether the target similarity score with the highest comprehensive similarity score is greater than a preset similarity threshold, so as to determine whether the test sample is in an unknown operating state or a known operating state based on the judgment result.

[0093] Specifically, the initial attribute learner is trained based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and a gating model is constructed during the training process to determine whether the test sample is in a known operating state or an unknown operating state based on the gating model, so as to obtain a judgment result, including: training the initial attribute learner based on the generalized zero-shot learning method, machine learning algorithms, and the attribute configuration table to obtain a target attribute learner; constructing a gating model during the training of the initial attribute learner, and obtaining the feature center values of each of the known operating states in the feature space of the gating model; respectively calculating the distribution direction similarity and the distribution distance similarity between the test sample and the feature center values to obtain a first similarity calculation result and a second similarity calculation result; comprehensively obtaining a comprehensive similarity score from the first similarity calculation result and the second similarity calculation result, and determining the highest comprehensive similarity score as the target similarity score, and determining whether the target similarity score is greater than a preset similarity threshold; if the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state; if the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state.

[0094] It can be understood that the initial attribute learner is trained based on the generalized zero-shot learning method, machine learning algorithms, and the attribute configuration table. The machine learning algorithms include probabilistic naive Bayes (NB, Naive Bayes), polynomial kernel-support vector machine (SVC or SVM, Support Vector Machine with Polynomial Kernel), and decision tree (Decision Tree); the training set in the generalized zero-shot learning method only includes training samples in known operating states, but the test set includes both known operating states and unknown operating states. In a specific implementation, the training set can be denoted as ; the test set can be denoted as ; further, the data set of the training set is denoted as ; 、 are the training set samples and the corresponding state attributes respectively. The data set of the test set is denoted as ; 、 are the test set samples and the corresponding state attributes respectively.

[0095] Figure 7 FIG. is a schematic diagram of a process of attribute generalized zero-shot learning provided in this embodiment, where is the initial attribute learner, are various state attributes. Based on the generalized zero-shot learning method, the known operating states and their corresponding state attributes are transferred to the unknown operating states, so as to learn the known operating states through the known operating states and their corresponding state attributes to all operating states the mapping model between ; The mapping formula of the state attributes corresponding to all operating states is: ; where is the state attribute description matrix of all operating states. The difference between the attribute value predicted by the trained attribute learner and the actual state attribute is . When the difference meets the preset difference condition, the trained attribute learner is determined as the target attribute learner

[0096] In this embodiment, during the training of the initial attribute learner, a gating model for operating state shunting is constructed, and the feature center values of each of the known operating states in the feature space are obtained, so as to determine a first vector and a second vector based on the feature center values and the corresponding test samples. By calculating the cosine value between the first vector and the second vector, a first similarity calculation result representing the similarity of the distribution directions of the feature center value and the test sample is obtained; calculate the ratio of the norms of the first vector and the second vector to obtain a second similarity calculation result representing the similarity of the distribution distances of the feature center value and the test sample

[0097] Specifically, the distribution direction similarity calculation and the distribution distance similarity calculation are respectively performed on the test sample and the feature center value to obtain a first similarity calculation result and a second similarity calculation result, including: determining a first vector and a second vector based on the feature center value and the test sample; by calculating the cosine value between the first vector and the second vector, a first similarity calculation result representing the similarity of the distribution directions of the feature center value and the test sample is obtained; by calculating the ratio of the norms of the first vector and the second vector, a second similarity calculation result representing the similarity of the distribution distances of the feature center value and the test sample is obtained

[0098] Figure 8 is a schematic diagram of the feature space provided in this embodiment and are the coordinate axes in the feature space. Let the mean of the sample features belonging to the i-th known operating state in the training set be ; The feature of the test sample is denoted as . Starting from the origin of the coordinate system, with and Two vectors are made with the corresponding coordinates as the end points to obtain the first vector and the second vector ; Then, the cosine formula is used to determine the first similarity calculation result representing the similarity between the feature central value and the distribution direction of the test sample. By introducing the norm ratio of the first vector and the second vector , the formula corresponding to the norm ratio of the first vector and the second vector is as follows:

[0099] ;

[0100] where, is the first vector; is the second vector; is the norm of the first vector; is the norm of the second vector. The second similarity calculation result representing the similarity between the feature central value and the distribution distance of the test sample is obtained from the above formula; then the first similarity calculation result and the second similarity calculation result are synthesized to obtain the comprehensive similarity score , and the formula corresponding to the comprehensive similarity score is as follows:

[0101] ;

[0102] After obtaining the comprehensive similarity score, the one with the highest comprehensive similarity score is taken as the target similarity score, and the formula corresponding to the target similarity score is as follows:

[0103] ;

[0104] After obtaining the target similarity score, it is judged whether the target similarity score is greater than the preset similarity threshold , and the corresponding judgment formula is:

[0105] ;

[0106] If the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state; if the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state.

[0107] Step S14: Use the target attribute learner to predict the attributes of the feature vector to obtain the corresponding attribute values, and then determine the monitoring result based on the judgment result and the attribute values, and combine the nearest neighbor search rule to take corresponding warning and maintenance measures based on the monitoring result.

[0108] In this embodiment, after obtaining the target attribute learner, the target attribute learner is used to predict the attributes of the feature vector to obtain the attribute values. If the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state. Then, calculate the first distance between the attribute value and the attribute vectors corresponding to each known operating state, and select the known operating state with the smallest first distance as the target known operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target known operating state. If the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state. Then, calculate the second distance between the attribute value and the attribute vectors corresponding to each unknown operating state, and select the unknown operating state with the smallest second distance as the target unknown operating state of the test sample based on the nearest neighbor search rule, so as to take corresponding warning and maintenance measures based on the target unknown operating state. The distances referred to by the first distance and the second distance can be Mahalanobis distances. It should be noted that if the target known operating state is normal operation, no warning and maintenance measures need to be taken.

[0109] Specifically, using the target attribute learner to predict the attributes of the feature vector to obtain the corresponding attribute values, and then determining the monitoring result based on the judgment result and the attribute value, and combining the nearest neighbor search rule to take corresponding warning and maintenance measures based on the monitoring result, includes: using the target attribute learner to predict the attributes of the feature vector to obtain the corresponding attribute values. If the judgment result indicates that the test sample is in a known operating state, calculate the first distance between the attribute value and the attribute vectors corresponding to each known operating state, and select the known operating state with the smallest first distance as the target known operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target known operating state. If the judgment result indicates that the test sample is in an unknown operating state, calculate the second distance between the attribute value and the attribute vectors corresponding to each unknown operating state, and select the unknown operating state with the smallest second distance as the target unknown operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target unknown operating state.

[0110] As can be seen from the above, in this application, an attribute configuration table is constructed by integrating the engineering knowledge and experience data of maglev trains. Then, a preset feature extractor determined by the attention mechanism and wavelet function, as well as wavelet transform technology, are used to extract and perform multi-scale analysis on the operation state data during the operation of maglev trains to obtain feature vectors. Then, based on the generalized zero-shot learning method and the attribute configuration table, the initial attribute learner is trained to obtain the target attribute learner, and a gating model is constructed during the training process to use the gating model to determine whether the test sample is a known operation state or an unknown operation state. Then, based on the attribute learner, the attribute of the feature vector is predicted to obtain an attribute value. In this way, based on the judgment result and the attribute value, combined with the nearest neighbor search rule, the final monitoring result is determined. Through the monitoring result, potential faults or abnormal conditions of maglev trains can be discovered in time, and corresponding early warning and maintenance measures can be taken, greatly reducing the fault risk and operation cost of maglev trains and improving the operation safety and reliability.

[0111] As can be seen from the above embodiments, this application realizes the operation state monitoring of the electromagnetic suspension system based on feature extraction and training of the initial attribute learner. Therefore, the process of realizing the operation state monitoring of the electromagnetic suspension system based on feature extraction and training of the initial attribute learner is described.

[0112] See Figure 9 As shown, an embodiment of the present invention discloses a specific method for monitoring the operation state of an electromagnetic suspension system, including:

[0113] In this embodiment, first, an attribute configuration table including the correspondence between the train operation state and the state attribute is constructed based on the engineering knowledge and experience data of the maglev train. Optionally, data preprocessing and normalization processing can be performed on the train operation state, the training set data including known operation states is fitted with the corresponding attribute labels, and a preset feature extractor is determined based on the obtained fitted data, attention mechanism, and wavelet function. Then, the preset feature extractor is used to reduce the dimension of the fitted data to obtain the training set data features, the training set data features are used to train the initial attribute learner to obtain the target attribute learner, and a gating model is constructed during the training process to calculate the feature center values of each known operation state in the feature space.

[0114] Next, use the operating state data of the maglev train during operation to obtain test samples including known operating states and unknown operating states, and perform feature extraction on the test samples based on the preset feature extractor to obtain feature segments, and perform multi-scale analysis on the feature segments through wavelet transform technology to obtain feature vectors; then use the target attribute learner to predict the feature vectors to obtain corresponding attribute values. Calculate the distribution direction similarity between the test sample and the feature central value to obtain the first similarity calculation result; calculate the distribution distance similarity between the test sample and the feature central value to obtain the second similarity calculation result, synthesize the first similarity calculation result and the second similarity calculation result to obtain a comprehensive similarity score, and determine the highest comprehensive similarity score as the target similarity score, and judge whether the target similarity score is greater than the preset similarity threshold.

[0115] Further, if the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state. Calculate the first distance between the attribute value corresponding to the test sample and the attribute vectors corresponding to each known operating state, and select the known operating state with the smallest first distance as the target known operating state of the test sample based on the nearest neighbor search rule. The target known operating state is the state that the test sample is most likely to belong to among the known operating states, and corresponding warning and maintenance measures are taken based on the target known operating state; if the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state. Then calculate the second distance between the attribute value corresponding to the test sample and the attribute vectors corresponding to each unknown operating state, and select the unknown operating state with the smallest second distance as the target unknown operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target unknown operating state.

[0116] There are mainly six evaluation indicators in this embodiment. Figure 10 It is a schematic diagram of an evaluation indicator provided by this embodiment. Each symbol represents a different meaning and is used to judge the final monitoring result. As Figures 11 to 15 can be seen, the monitoring of the operating state by traditional machine learning has very poor monitoring effects on unknown operating states; but after introducing generalized zero-shot learning, the average value of F1-score for known and unknown operating states has been significantly improved, and the accuracy and recall rate are also relatively high. It can effectively distinguish and accurately monitor known and unknown operating state samples, and the effect is significantly better than the traditional method.

[0117] As can be seen from the above, in this embodiment, the target operating state corresponding to the test sample is determined by using the attribute value predicted by the target attribute learner and combining the judgment result, and corresponding countermeasures are taken based on the target operating state, breaking the limitation of the traditional method that it is difficult to monitor unknown operating states, achieving accurate monitoring of samples in all operating states, and providing strong support for improving operation efficiency, ensuring passenger safety, and reducing maintenance costs.

[0118] Correspondingly, as shown in Figure 16 the present application also provides an operating state monitoring device for an electromagnetic levitation system, including:

[0119] An attribute configuration table construction module 11, configured to construct an attribute configuration table including the correspondence between the train operating state and the state attributes based on the engineering knowledge and experience data of the maglev train; the train operating states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail pounding;

[0120] A feature extraction module 12, configured to obtain a test sample by using the operating state data of the maglev train during operation, then perform feature extraction on the test sample based on a preset feature extractor and a sliding window function to obtain a feature segment, and perform multi-scale analysis on the feature segment based on wavelet transform technology to obtain a feature vector; the operating state data includes known operating states and unknown operating states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function;

[0121] An attribute learner training module 13, configured to train an initial attribute learner based on a generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to judge whether the test sample is a known operating state or an unknown operating state based on the gating model to obtain a judgment result;

[0122] An attribute value prediction module 14, configured to perform attribute prediction on the feature vector by using the target attribute learner to obtain a corresponding attribute value, then determine a monitoring result based on the judgment result and the attribute value, and combine a nearest neighbor search rule, and take corresponding warning and maintenance measures based on the monitoring result.

[0123] As can be seen from the above, in this application, an attribute configuration table is constructed by integrating the engineering knowledge and experience data of the maglev train. Then, a preset feature extractor determined by the attention mechanism and wavelet function and wavelet transform technology are used to extract and perform multi-scale analysis on the operation state data during the operation of the maglev train to obtain feature vectors. Then, based on the generalized zero-shot learning method and the attribute configuration table, an initial attribute learner is trained to obtain a target attribute learner, and a gating model is constructed during the training process to use the gating model to determine whether the test sample is a known operation state or an unknown operation state. Then, based on the attribute learner, attribute prediction is performed on the feature vector to obtain an attribute value. In this way, based on the judgment result and the attribute value, and combined with the nearest neighbor search rule, the final monitoring result is determined. Through the monitoring result, potential faults or abnormal conditions of the maglev train can be discovered in time, and corresponding early warning and maintenance measures can be taken, greatly reducing the fault risk and operation cost of the maglev train, and improving the operation safety and reliability.

[0124] In some specific embodiments, the feature extraction module 12 may specifically include:

[0125] A data acquisition unit, configured to collect the operation state data of the maglev train during operation by using a suspension data acquisition system, and determine each operation state data as a test sample.

[0126] In some specific embodiments, the feature extraction module 12 may specifically include:

[0127] A test sample segmentation unit, configured to segment the test sample based on a preset sliding window length and a preset sliding step to obtain each sliding window, and calculate the mean value, the first energy value, and the difference between the first peaks of each sliding window by using an attention mechanism to obtain corresponding calculation results;

[0128] A first weighted function determination unit, configured to determine a first attention weighted function by using the calculation results, and determine a target window segment with the largest weight value based on the first attention weighted function;

[0129] A second weighted function determination unit, configured to determine a second attention weighted function based on the second energy value and the difference between the second peaks of the first-order difference sequence corresponding to the sliding window, and use the second attention weighted function to determine the key features in the target window segment to obtain a feature segment.

[0130] In some specific embodiments, the feature extraction module 12 may specifically include:

[0131] A feature segment decomposition unit, which is used to perform dilation and translation transformations on a preset wavelet function by using a preset scale parameter to obtain a target wavelet function, and perform inner product calculations based on the target wavelet function and the feature segment to obtain wavelet coefficients at different scales, and extract the wavelet coefficients to obtain feature vectors.

[0132] In some specific embodiments, the attribute learner training module 13 may specifically include:

[0133] A learner training unit, which is used to train an initial attribute learner based on a generalized zero-shot learning method, a machine learning algorithm, and the attribute configuration table to obtain a target attribute learner;

[0134] A gating model construction unit, which is used to construct a gating model during the training of the initial attribute learner and obtain the feature center values of each known operating state in the feature space;

[0135] A similarity calculation unit, which is used to calculate the distribution direction similarity and the distribution distance similarity between the test sample and the feature center value respectively to obtain a first similarity calculation result and a second similarity calculation result;

[0136] A target similarity determination unit, which is used to synthesize the first similarity calculation result and the second similarity calculation result to obtain a comprehensive similarity score, determine the highest comprehensive similarity score as the target similarity score, and determine whether the target similarity score is greater than a preset similarity threshold;

[0137] A first target similarity judgment unit, which is used to represent that the test sample is in an unknown operating state if the target similarity score is less than the preset similarity threshold;

[0138] A first target similarity judgment unit, which is used to represent that the test sample is in a known operating state if the target similarity score is not less than the preset similarity threshold.

[0139] In some specific embodiments, the attribute learner training module 13 may specifically include:

[0140] A vector determination unit, which is used to determine a first vector and a second vector based on the feature center value and the test sample;

[0141] A first similarity calculation unit, which is used to calculate the cosine value of the first vector and the second vector to obtain a first similarity calculation result representing the distribution direction similarity between the feature center value and the test sample;

[0142] A second similarity calculation unit, configured to obtain a second similarity calculation result characterizing the similarity of the distribution distance between the feature center value and the test sample by calculating the ratio of the norms of the first vector and the second vector.

[0143] In some specific embodiments, the attribute value prediction module 14 may specifically include:

[0144] An attribute prediction unit, configured to perform attribute prediction on the feature vector by using the target attribute learner to obtain a corresponding attribute value. If the judgment result indicates that the test sample is in a known operating state, calculate the first distance between the attribute value and the attribute vectors corresponding to the known operating states, and select, based on the nearest neighbor search rule, the known operating state with the smallest first distance as the target known operating state of the test sample, and take corresponding warning and maintenance measures based on the target known operating state;

[0145] A warning and maintenance unit, configured to, if the judgment result indicates that the test sample is in an unknown operating state, calculate the second distance between the attribute value and the attribute vectors corresponding to the unknown operating states, and select, based on the nearest neighbor search rule, the unknown operating state with the smallest second distance as the target unknown operating state of the test sample, and take corresponding warning and maintenance measures based on the target unknown operating state.

[0146] Furthermore, an embodiment of the present application also discloses an electronic device. Figure 17 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment. The content in the figure cannot be considered as any limitation on the scope of use of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the electromagnetic suspension system operating state monitoring method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0147] In this embodiment, the power supply 23 is used to provide operating voltages for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of the present application, and specific limitations are not imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and specific limitations are not imposed here.

[0148] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a magnetic disk, an optical disc, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be transient storage or permanent storage.

[0149] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it can be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the electromagnetic suspension system operation state monitoring method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 can further include computer programs that can be used to complete other specific tasks.

[0150] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the electromagnetic suspension system operation state monitoring method disclosed above. For the specific steps of this method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be repeated here.

[0151] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description of the method part for related parts.

[0152] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the composition and steps of the examples have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0153] The steps of the method or algorithm described in combination with the embodiments disclosed herein can be directly implemented by hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0154] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0155] The technical solutions provided in this application have been introduced in detail above. Specific examples are used in this text to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for monitoring the operating state of an electromagnetic suspension system, characterized in that, Including: Construct an attribute configuration table that includes the correspondence between the train operation states and state attributes based on the engineering knowledge and experience data of the maglev train; The train operation states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail pounding; Use the operation state data of the maglev train during operation to obtain test samples, then perform feature extraction on the test samples based on a preset feature extractor and a sliding window function to obtain feature segments, and perform multi-scale analysis on the feature segments based on wavelet transform technology to obtain feature vectors; the operation state data includes known operation states and unknown operation states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function; Train an initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to judge whether the test sample is a known operation state or an unknown operation state based on the gating model to obtain a judgment result; Use the target attribute learner to perform attribute prediction on the feature vector to obtain corresponding attribute values, then based on the judgment result and the attribute values, and in combination with the nearest neighbor search rule, determine the monitoring result, and take corresponding warning and maintenance measures based on the monitoring result.

2. The method for monitoring the operating state of an electromagnetic suspension system according to claim 1, wherein The obtaining test samples by using the operation state data of the maglev train during operation includes: Use a suspension data acquisition system to collect the operation state data of the maglev train during operation, and determine each operation state data as a test sample; Wherein, the suspension data acquisition system includes a data collector, a suspension gap sensor, a current sensor, an acceleration sensor, a temperature sensor, a WIFI network center, and an embedded upper computer software center; the operation state data includes the suspension gap, suspension acceleration, lateral acceleration, working current of the suspension electromagnet, and surface temperature of the suspension electromagnet coil of the maglev train during operation.

3. The method for monitoring the operating state of the electromagnetic suspension system according to claim 1, characterized in that, The performing feature extraction on the test samples based on a preset feature extractor and a sliding window function to obtain feature segments includes: Segment the test samples based on a preset sliding window length and a preset sliding step to obtain each sliding window, and use the attention mechanism to calculate the mean value, the first energy value, and the difference between the first peak values of each sliding window to obtain corresponding calculation results; Use the calculation results to determine a first attention weighting function to determine a target window segment with the largest weight value based on the first attention weighting function; Determine a second attention weighting function based on the second energy value and the difference between the second peak values of the first-order difference sequence corresponding to the sliding window, and use the second attention weighting function to determine the key features in the target window segment to obtain feature segments.

4. The method for monitoring the operating state of an electromagnetic suspension system according to any one of claims 1 to 3, characterized in that, The performing multi-scale analysis on the feature segments based on wavelet transform technology to obtain feature vectors includes: The preset wavelet function is scaled and translated using a preset scale parameter to obtain a target wavelet function, and the inner product is calculated based on the target wavelet function and the feature segment to obtain wavelet coefficients at different scales. The wavelet coefficients are extracted to obtain feature vectors.

5. The method for monitoring the operating state of the electromagnetic suspension system according to claim 1, characterized in that, The initial attribute learner is trained based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and a gating model is constructed during the training process to judge whether the test sample is in a known operating state or an unknown operating state based on the gating model to obtain a judgment result, including: The initial attribute learner is trained based on the generalized zero-shot learning method, the machine learning algorithm, and the attribute configuration table to obtain a target attribute learner; A gating model is constructed during the training of the initial attribute learner, and the feature center values of each of the known operating states in the feature space are obtained; The distribution direction similarity and the distribution distance similarity are calculated for the test sample and the feature center value respectively to obtain a first similarity calculation result and a second similarity calculation result; The first similarity calculation result and the second similarity calculation result are combined to obtain a comprehensive similarity score, the highest comprehensive similarity score is determined as the target similarity score, and it is judged whether the target similarity score is greater than a preset similarity threshold; If the target similarity score is less than the preset similarity threshold, it indicates that the test sample is in an unknown operating state; If the target similarity score is not less than the preset similarity threshold, it indicates that the test sample is in a known operating state.

6. The method for monitoring the operating state of the electromagnetic suspension system according to claim 5, wherein The distribution direction similarity and the distribution distance similarity are calculated for the test sample and the feature center value respectively to obtain a first similarity calculation result and a second similarity calculation result, including: A first vector and a second vector are determined based on the feature center value and the test sample; By calculating the cosine value of the first vector and the second vector, a first similarity calculation result representing the distribution direction similarity between the feature center value and the test sample is obtained; By calculating the ratio of the norms of the first vector and the second vector, a second similarity calculation result representing the distribution distance similarity between the feature center value and the test sample is obtained.

7. The method for monitoring the operating state of the electromagnetic suspension system according to claim 5, wherein, The target attribute learner is used to predict the attribute of the feature vector to obtain a corresponding attribute value, and then based on the judgment result and the attribute value, combined with the nearest neighbor search rule, the monitoring result is determined, and corresponding early warning and maintenance measures are taken based on the monitoring result, including: Use the target attribute learner to perform attribute prediction on the feature vector to obtain corresponding attribute values. If the judgment result indicates that the test sample is in a known operating state, calculate the first distance between the attribute value and the attribute vectors corresponding to each known operating state, and select the known operating state with the smallest first distance as the target known operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target known operating state; If the judgment result indicates that the test sample is in an unknown operating state, calculate the second distance between the attribute value and the attribute vectors corresponding to each unknown operating state, and select the unknown operating state with the smallest second distance as the target unknown operating state of the test sample based on the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the target unknown operating state.

8. A monitoring device for the operating state of an electromagnetic suspension system, characterized in that, Comprising: An attribute configuration table construction module, configured to construct an attribute configuration table including the corresponding relationship between the train operating state and the state attributes based on the engineering knowledge and empirical data of the maglev train; the train operating states include normal operation, passing through joints, low-frequency swaying, high-frequency vibration, and rail pounding; A feature extraction module, configured to obtain a test sample by using the operating state data of the maglev train during operation, then perform feature extraction on the test sample based on a preset feature extractor and a sliding window function to obtain a feature segment, and perform multi-scale analysis on the feature segment based on wavelet transform technology to obtain a feature vector; the operating state data includes known operating states and unknown operating states; the preset feature extractor is a feature extractor determined based on an attention mechanism and a wavelet function; An attribute learner training module, configured to train an initial attribute learner based on the generalized zero-shot learning method and the attribute configuration table to obtain a target attribute learner, and construct a gating model during the training process to judge whether the test sample is in a known operating state or an unknown operating state based on the gating model to obtain a judgment result; An attribute value prediction unit, configured to use the target attribute learner to perform attribute prediction on the feature vector to obtain corresponding attribute values, then determine a monitoring result based on the judgment result and the attribute values in combination with the nearest neighbor search rule, and take corresponding warning and maintenance measures based on the monitoring result.

9. An electronic device, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to execute the computer program to implement the electromagnetic suspension system operating state monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, For storing a computer program, wherein the computer program, when executed by the processor, implements the electromagnetic suspension system operating state monitoring method according to any one of claims 1 to 7.

Images