A method and system for monitoring the operation status of locomotive bogies based on voiceprint

Abstract

The present invention discloses a locomotive bogie operation status monitoring method and system based on voiceprint, comprising: firstly, obtaining the voiceprint data of the bogie through a voiceprint acquisition sensor installed on the locomotive bogie, and then identifying the basic and detailed feature types of these data to determine the voiceprint representation type. When an abnormal representation type is identified, the system generates an emergency strategy determination request. After receiving this request, the preset emergency strategy is searched in multiple dimensions according to the semantic features of the locomotive bogie analysis results, so as to quickly determine the appropriate emergency strategy. Such a design improves the accuracy and real-time performance of the locomotive bogie status monitoring, and provides a strong guarantee for the safe operation of the locomotive.

Description

Technical Field

[0001] The present invention relates to the field of acoustic measurement, and in particular to a method and system for monitoring the running status of a locomotive bogie based on soundprint. Background Art

[0002] Traditional locomotive bogie monitoring mainly relies on physical quantity detection equipment such as vibration sensors and temperature sensors. Such methods often have problems of insufficient monitoring accuracy and delayed fault warning when facing complex and changeable operating environments. With the expansion of voiceprint recognition technology in multiple fields, its unique non-contact monitoring and highly sensitive feature recognition capabilities provide new ideas for locomotive bogie monitoring. Summary of the invention

[0003] The object of the present invention is to provide a method and system for monitoring the operating status of a locomotive bogie based on voiceprint.

[0004] In a first aspect, an embodiment of the present invention provides a method for monitoring the running state of a locomotive bogie based on a voiceprint, comprising:

[0005] Acquiring the bogie voiceprint data to be analyzed of the locomotive bogie by means of a voiceprint collection sensor arranged on the locomotive bogie;

[0006] Perform basic feature type recognition and detailed feature type recognition on the bogie voiceprint data to be analyzed, and obtain the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed;

[0007] When the voiceprint representation type is characterized as an abnormal representation type, generating an emergency strategy determination request;

[0008] When an emergency strategy determination request including a locomotive bogie analysis result is obtained, a strategy search is performed on the multi-dimensional semantic features corresponding to each preset emergency strategy according to the semantic analysis features of the locomotive bogie analysis result to obtain an emergency strategy determination result.

[0009] In a second aspect, an embodiment of the present invention provides a server system, including a server, wherein the server is used to execute the method described in the first aspect.

[0010] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a locomotive bogie operating status monitoring method and system based on voiceprint disclosed by the present invention, obtaining the voiceprint data of the bogie through a voiceprint acquisition sensor installed on the locomotive bogie, and then identifying the basic and refined feature types of these data to determine the voiceprint representation type. When an abnormal representation type is identified, the system will generate an emergency strategy determination request. After receiving this request, the preset emergency strategy is searched in multiple dimensions according to the semantic features of the locomotive bogie analysis results, so as to quickly determine the appropriate emergency strategy. Such a design improves the accuracy and real-time performance of locomotive bogie status monitoring, and provides a strong guarantee for the safe operation of the locomotive. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0012] Figure 1 A schematic diagram of the steps of a method for monitoring the running state of a locomotive bogie based on voiceprint provided in an embodiment of the present invention;

[0013] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0015] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0016] In order to solve the technical problems in the aforementioned background technology, Figure 1 A flow chart of a locomotive bogie operating status monitoring method based on voiceprint provided in an embodiment of the present disclosure is provided below. The locomotive bogie operating status monitoring method based on voiceprint is introduced in detail.

[0017] Step S201, obtaining the bogie voiceprint data to be analyzed of the locomotive bogie by means of a voiceprint collection sensor arranged on the locomotive bogie;

[0018] Step S202, performing basic feature type recognition and detailed feature type recognition on the bogie voiceprint data to be analyzed, and obtaining the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed;

[0019] Step S203, when the voiceprint representation type is characterized as an abnormal representation type, generating an emergency strategy determination request;

[0020] Step S204, when an emergency strategy determination request including a locomotive bogie analysis result is obtained, a strategy search is performed on the multi-dimensional semantic features corresponding to each preset emergency strategy according to the semantic analysis features of the locomotive bogie analysis result to obtain an emergency strategy determination result.

[0021] In an embodiment of the present invention, exemplarily, the server obtains the soundprint data of the locomotive bogie in real time during operation through a soundprint collection sensor set on the locomotive bogie. For example, when the locomotive bogie is running at high speed, the sensor captures the specific sound patterns generated by the operation of the bearing, the meshing of the gears, etc., and uses these data as the bogie soundprint data to be analyzed. The server processes the collected bogie soundprint data to be analyzed, and first performs basic feature type recognition, such as distinguishing mechanical vibration sound, friction sound, etc. Then, refine feature type recognition is performed, such as further identifying whether it is an abnormal vibration sound of the bearing or a wear sound of the gear, so as to obtain the soundprint representation type corresponding to the soundprint data. For example, through algorithm analysis, the server recognizes that the bogie soundprint contains abnormal bearing vibration sound, which indicates that there may be a problem with the bearing. When the server determines that the soundprint representation type is an abnormal representation, such as recognizing the abnormal vibration sound of the bearing, the server will generate an emergency strategy determination request. This request contains specific information about the abnormal state of the bogie for further analysis and processing. After receiving the request to determine the emergency strategy, the server will perform multi-dimensional semantic feature matching on each preset emergency strategy based on the semantic analysis features of the locomotive bogie analysis results. For example, based on the semantic features of the abnormal vibration sound of the bearing, the server will retrieve the emergency strategy that matches it, such as "check the bearing immediately and prepare for replacement" or "slow down and run to the next station for maintenance." Through strategy retrieval, the server finally determines the most appropriate emergency strategy and sends instructions to relevant personnel or the system for automatic execution. Through the above steps, the server can effectively monitor the operating status of the locomotive bogie based on voiceprints, and take emergency strategies in time when abnormalities are found to ensure the safe operation of the locomotive.

[0022] In the embodiment of the present invention, the basic feature type recognition and the refined feature type recognition are performed on the bogie voiceprint data to be analyzed to obtain the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed, which can be implemented through the following examples.

[0023] Acquire the bogie voiceprint data to be analyzed, perform a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed, and obtain a voiceprint feature vector;

[0024] Extracting basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and identifying basic feature types based on the basic and advanced features to obtain voiceprint representation type confidence corresponding to each basic representation type;

[0025] Extracting refined high-level features from the voiceprint feature vector to obtain refined high-level features, integrating voiceprint features according to the basic high-level features and the refined high-level features to obtain refined voiceprint integrated features, and using the refined voiceprint integrated features to identify refined feature types to obtain voiceprint representation type confidence corresponding to each refined representation type, wherein the classification granularity corresponding to the refined representation type is greater than the classification granularity corresponding to the basic representation type;

[0026] The voiceprint characterization type corresponding to the bogie voiceprint data to be analyzed is obtained according to the voiceprint characterization type confidence corresponding to each basic characterization type and the voiceprint characterization type confidence corresponding to each refined characterization type, wherein the voiceprint characterization type includes a target basic characterization type and a target refined characterization type corresponding to the target basic characterization type.

[0027] In the embodiment of the present invention, the server first obtains the bogie voiceprint data to be analyzed. The voiceprint feature extraction operation is performed on these data, which involves converting the original voiceprint data into a numerical vector that can represent its unique characteristics, namely, a voiceprint feature vector. For example, the server converts the collected bearing operation sound into a series of numerical values ​​through an algorithm, and these numerical values ​​constitute a voiceprint feature vector. Then, the server performs basic high-level feature extraction on the voiceprint feature vector to obtain features that can reflect the basic properties of the voiceprint, such as the frequency and intensity of the sound. Based on these basic high-level features, the server performs basic feature type identification and calculates the corresponding voiceprint representation type confidence of each basic representation type (such as mechanical vibration sound and friction sound). For example, the server identifies that the voiceprint contains mechanical vibration sound and gives a confidence of 80% for this type of voiceprint. Then, the server further performs refined high-level feature extraction on the voiceprint feature vector to obtain more specific features with smaller classification granularity, such as the vibration mode of a specific component (such as a bearing). The server integrates the basic high-level features and the refined high-level features to form refined voiceprint integrated features, and uses these features to perform refined feature type identification. This step can identify more specific voiceprint types, such as abnormal vibration sound of bearings, and give the corresponding voiceprint characterization type confidence. For example, the server identifies that the voiceprint contains abnormal vibration sound of bearings, and gives the confidence of this refined type as 90%. Finally, the server determines the voiceprint characterization type corresponding to the bogie voiceprint data to be analyzed based on the voiceprint characterization type confidence corresponding to each basic characterization type and the voiceprint characterization type confidence corresponding to each refined characterization type. This type includes the target basic characterization type (such as mechanical vibration sound) and the target refined characterization type corresponding to the target basic characterization type (such as abnormal vibration sound of bearings). The server uses this result as an important basis for monitoring the operating status of the bogie, and may trigger further emergency strategy analysis or maintenance instructions. Through the above steps, the server can analyze the bogie voiceprint data in detail, accurately identify the basic feature types and refined feature types, and provide strong guarantees for the safe operation of the locomotive.

[0028] In the embodiment of the present invention, the voiceprint feature integration is performed according to the basic high-level features and the refined high-level features to obtain the refined voiceprint integration features, which can be implemented through the following examples.

[0029] Merging the basic high-level features and the refined high-level features to obtain a voiceprint merged feature;

[0030] Incremental feature extraction is performed based on the voiceprint combined feature and the refined high-level feature to obtain incremental features, and feature normalization processing is performed on the incremental features to obtain the refined voiceprint integrated feature.

[0031] In the embodiment of the present invention, the server first merges the basic high-level features and the refined high-level features. This step involves combining two feature vectors into a more comprehensive feature vector to more accurately describe the characteristics of the bogie soundprint. For example, the server merges the basic high-level features representing the sound frequency and intensity with the refined high-level features representing the vibration mode of a specific component to form a soundprint merge feature containing more information. Then, the server performs incremental feature extraction based on the soundprint merge feature and the refined high-level features. This step aims to further extract new features that can provide more information from the merged features. The server uses an algorithm to analyze the soundprint merge feature, identify additional information associated with the refined high-level features, such as the change in vibration mode at a specific frequency, and thus obtain the incremental features. Finally, the server normalizes the extracted incremental features to obtain the refined soundprint integrated features. The normalization process is to ensure that the numerical ranges of different features are consistent, which is convenient for subsequent analysis and comparison. The server scales the numerical values ​​of the incremental features to a uniform range, such as between 0 and 1, to form the refined soundprint integrated features. These integrated features contain information about basic advanced features and refined advanced features, as well as the relationship between them, and can more accurately describe the characteristics of the bogie voiceprint. Through the above steps, the server can integrate basic advanced features and refined advanced features to obtain refined voiceprint integrated features that contain more comprehensive and detailed information. These features provide a more accurate data basis for subsequent refined feature type identification, which helps the server to more accurately judge the operating status of the bogie and take corresponding emergency strategies.

[0032] In the embodiment of the present invention, the incremental feature extraction is performed based on the voiceprint merge feature and the refined high-level feature to obtain the incremental feature, which can be implemented through the following example.

[0033] The voiceprint merge feature is subjected to feature conversion using a preset incremental parameter to obtain a conversion feature, and the feature sum of the conversion feature and the refined high-level feature is calculated to obtain the incremental feature.

[0034] In the embodiment of the present invention, the server uses a preset incremental parameter to perform feature conversion on the voiceprint merge feature. This step involves subjecting the voiceprint merge feature to a series of mathematical operations or transformations to highlight the key information or patterns therein. For example, the server can apply a weighted matrix as an incremental parameter to weighted sum each component in the voiceprint merge feature to obtain a conversion feature. This weighted matrix is ​​obtained based on previous experience or data training, and can enhance the features related to the bogie operation state. Next, the server calculates the feature sum of the conversion feature and the refined high-level feature. This step is to add the converted feature to the refined high-level feature point by point to obtain a new feature vector, namely the incremental feature. This incremental feature contains both the key information in the voiceprint merge feature and the details of the refined high-level feature, thereby providing a more comprehensive description of the bogie voiceprint. For example, the server adds each component in the conversion feature to the corresponding component in the refined high-level feature to form a new incremental feature vector. Through the above steps, the server can extract incremental features from the voiceprint merge feature and the refined high-level feature using the preset incremental parameters and feature conversion technology. These incremental features integrate basic high-level features, refined high-level features and the associated information between them, providing a more accurate and comprehensive data basis for subsequent refined feature type identification. This helps the server to more accurately determine the operating status of the bogie and trigger corresponding emergency strategies when necessary to ensure the safe operation of the locomotive.

[0035] In the embodiment of the present invention, obtaining the voiceprint characterization type corresponding to the bogie voiceprint data to be analyzed according to the voiceprint characterization type confidence corresponding to each basic characterization type and the voiceprint characterization type confidence corresponding to each refined characterization type can be implemented through the following examples.

[0036] Obtain the granularity information of the voiceprint representation type;

[0037] According to the voiceprint representation type granularity information, the voiceprint representation type confidence corresponding to each basic representation type and the voiceprint representation type confidence corresponding to each refined representation type are merged to obtain each granularity index confidence;

[0038] A target granularity index confidence is determined from each granularity index confidence, and a voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is determined according to the target granularity index confidence.

[0039] In the embodiment of the present invention, exemplarily, the server first obtains the granularity information of the voiceprint characterization type. This step involves determining the hierarchical relationship or classification granularity between different voiceprint characterization types. For example, the server may have a preset voiceprint characterization type library, in which a hierarchical structure from basic characterization types (such as mechanical vibration sound) to refined characterization types (such as abnormal vibration sound of bearings) is defined. Then, the server fuses the voiceprint characterization type confidence corresponding to each basic characterization type and the voiceprint characterization type confidence corresponding to each refined characterization type according to the voiceprint characterization type granularity information. This step involves combining the confidences of different granularities to form a more comprehensive confidence assessment. For example, the server may use a weighted average method to fuse the confidence of the basic characterization type "mechanical vibration sound" with the confidence of the refined characterization type "abnormal vibration sound of bearings" according to the weight of the granularity information to obtain a comprehensive granularity index confidence. Then, the server determines the target granularity index confidence from each fused granularity index confidence. This step involves selecting the most representative or most credible granularity index confidence as the target. For example, the server may select the granularity index with the highest confidence as the target, because this means that the voiceprint representation type corresponding to the index is most likely to be the true type of the bogie voiceprint data to be analyzed. Finally, the server determines the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed based on the target granularity index confidence. This step involves taking the voiceprint representation type corresponding to the target granularity index confidence as the final result. For example, if the target granularity index confidence corresponds to the refined representation type "abnormal vibration sound of the bearing", the server uses this type as the voiceprint representation type of the bogie voiceprint data to be analyzed, and may trigger corresponding emergency strategies or maintenance instructions. Through the above steps, the server can comprehensively consider the confidence of voiceprint representation types of different granularities and accurately determine the voiceprint representation type of the bogie voiceprint data to be analyzed. This provides a strong guarantee for the safe operation of the locomotive, because the server can identify the operating status of the bogie based on accurate voiceprint analysis and take appropriate measures when necessary.

[0040] In the embodiment of the present invention, the voiceprint representation type confidence corresponding to each basic representation type and the voiceprint representation type confidence corresponding to each refined representation type are merged according to the voiceprint representation type granularity information to obtain each granularity index confidence, which can be implemented through the following examples.

[0041] Determining the voiceprint representation type confidence level corresponding to the current basic representation type from the voiceprint representation type confidence levels corresponding to each basic representation type;

[0042] Acquire a current refined characterization type corresponding to the current basic characterization type according to the voiceprint characterization type granularity information, and determine a voiceprint characterization type confidence corresponding to the current refined characterization type from the voiceprint characterization type confidences corresponding to each refined characterization type;

[0043] Calculating a multiplication result of the voiceprint representation type confidence corresponding to the current basic representation type and the voiceprint representation type confidence corresponding to the current refined representation type to obtain a current granularity index confidence;

[0044] The voiceprint representation type confidence corresponding to each basic representation type and the voiceprint representation type confidence corresponding to each refined representation type are polled to obtain each granularity index confidence.

[0045] In the embodiment of the present invention, exemplarily, the server determines the voiceprint characterization type confidence corresponding to the basic characterization type currently being processed from the voiceprint characterization type confidence corresponding to each basic characterization type. For example, the server may have a list containing basic characterization types such as "mechanical vibration sound" and "fluid sound" and their corresponding confidences. In the current iteration, the server will select one of the basic characterization types, such as "mechanical vibration sound", and obtain its corresponding confidence, such as 80%. Then, the server obtains the current refined characterization type corresponding to the current basic characterization type according to the voiceprint characterization type granularity information, and determines the voiceprint characterization type confidence corresponding to the current refined characterization type from the voiceprint characterization type confidence corresponding to each refined characterization type. For example, if the current basic characterization type is "mechanical vibration sound", then its corresponding refined characterization types may include "abnormal vibration sound of bearing", "wear sound of gear", etc. The server will obtain these refined characterization types and their corresponding confidences, such as the confidence of "abnormal vibration sound of bearing" is 90%. Then, the server calculates the multiplication result of the voiceprint representation type confidence corresponding to the current basic representation type and the voiceprint representation type confidence corresponding to the current refined representation type to obtain the current granularity index confidence. For example, the server will multiply the confidence of "mechanical vibration sound" 80% with the confidence of "abnormal vibration sound of bearing" 90% to obtain 72% as the current granularity index confidence. Finally, the server polls the voiceprint representation type confidence corresponding to each basic representation type and the voiceprint representation type confidence corresponding to each refined representation type, and repeats the above steps to obtain each granularity index confidence. For example, the server will process other basic representation types such as "fluid sound" and their corresponding refined representation types in turn, calculate their granularity index confidences, until all basic representation types have been processed. Through the above steps, the server can comprehensively consider the voiceprint representation type confidences of different granularities and merge them into each granularity index confidence. This provides a more accurate data basis for the subsequent determination of the voiceprint characterization type, helps the server to more accurately judge the operating status of the bogie and take appropriate measures to ensure the safe operation of the locomotive.

[0046] In the embodiments of the present invention, the following implementation modes are also provided.

[0047] Loading the bogie voiceprint data to be analyzed into a first target bogie voiceprint recognition model;

[0048] Using the first target bogie voiceprint recognition model, performing a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed to obtain a voiceprint feature vector;

[0049] Using the first target bogie voiceprint recognition model, extracting basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and performing basic feature type recognition based on the basic and advanced features to obtain a voiceprint representation type confidence corresponding to each basic representation type;

[0050] The first target bogie voiceprint recognition model is used to perform refined high-level feature extraction on the voiceprint feature vector to obtain refined high-level features, voiceprint feature integration is performed based on the basic high-level features and the refined high-level features to obtain refined voiceprint integrated features, and the refined voiceprint integrated features are used to perform refined feature type identification to obtain the voiceprint representation type confidence corresponding to each refined representation type.

[0051] In an embodiment of the present invention, exemplarily, the server loads the bogie voiceprint data to be analyzed into the first target bogie voiceprint recognition model. This step involves reading the voiceprint data from a storage location and inputting it into the recognition model for subsequent analysis. For example, the server can retrieve a recording of a bogie from a database and then load it into the model. The first target bogie voiceprint recognition model is used to perform a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed. This step involves extracting key feature information from the voiceprint data for subsequent analysis and recognition. For example, the model can use a series of filters to extract frequency components in the voiceprint to form a feature vector. The server uses the first target bogie voiceprint recognition model to perform basic high-level feature extraction on the voiceprint feature vector. This step involves further extracting high-level features that can describe the basic characteristics of the bogie voiceprint from the feature vector. For example, the model can identify the mechanical vibration mode in the voiceprint as a basic high-level feature. Then, the server performs basic feature type recognition based on these basic high-level features and obtains the voiceprint representation type confidence corresponding to each basic representation type. This step involves matching the basic high-level features with the preset representation types and giving the confidence of each match. For example, the server may determine that the confidence of the mechanical vibration pattern in the voiceprint with the basic representation type of "mechanical vibration sound" is 80%. The server again uses the first target bogie voiceprint recognition model to extract refined high-level features from the voiceprint feature vector. This step involves extracting high-level features that can describe the more detailed characteristics of the bogie voiceprint from the feature vector. For example, the model can identify the vibration pattern of a specific component in the voiceprint as a refined high-level feature. Then, the server performs voiceprint feature integration based on the basic high-level features and the refined high-level features to obtain a refined voiceprint integrated feature. This step involves merging two features to form a more comprehensive feature representation. For example, the server may combine the mechanical vibration pattern with the vibration pattern of a specific component to form a refined voiceprint integrated feature. Finally, the server uses the refined voiceprint integrated feature to identify the refined feature type and obtain the voiceprint representation type confidence corresponding to each refined representation type. This step involves matching the refined voiceprint integrated features with the preset refined characterization types, and giving the confidence level of each match. For example, the server can determine that the confidence level of the vibration mode of a specific component in the voiceprint and the refined characterization type of "abnormal vibration sound of the bearing" is 90%. Through the above steps, the server can use the first target bogie voiceprint recognition model to comprehensively analyze and identify the bogie voiceprint data to be analyzed, and give corresponding confidence assessments from basic features to refined features. This provides a rich data foundation for the subsequent determination of the voiceprint characterization type, helps the server to more accurately judge the operating status of the bogie, and take appropriate measures to ensure the safe operation of the locomotive.

[0052] In an embodiment of the present invention, the first target bogie voiceprint recognition model includes a voiceprint feature extraction component, a basic feature type recognition component and a refined feature type recognition component; the loading of the bogie voiceprint data to be analyzed into the first target bogie voiceprint recognition model can be implemented through the following example.

[0053] Loading the bogie voiceprint data to be analyzed into the voiceprint feature extraction component to perform a voiceprint feature extraction operation to obtain a voiceprint feature vector;

[0054] Loading the voiceprint feature vector into the basic feature type recognition component, using the basic feature type recognition component to extract basic high-level features from the voiceprint feature vector to obtain basic high-level features, and performing basic representation type confidence calculation based on the basic high-level features to obtain the voiceprint representation type confidence corresponding to each basic representation type;

[0055] The voiceprint feature vector and the basic high-level features are loaded into the refined feature type recognition component, and the refined feature type recognition component is used to extract refined high-level features from the voiceprint feature vector to obtain refined high-level features, and voiceprint feature integration is performed based on the basic high-level features and the refined high-level features to obtain refined voiceprint integrated features, and the refined voiceprint integrated features are used to calculate the refined representation type confidence to obtain the voiceprint representation type confidence corresponding to each refined representation type.

[0056] In an embodiment of the present invention, exemplarily, the server loads the bogie voiceprint data to be analyzed into the voiceprint feature extraction component. For example, the server can retrieve a specific bogie recording from the database and use it as input data. The voiceprint feature extraction component performs a voiceprint feature extraction operation to extract key feature information from the recording. This can include frequency components, amplitude changes, etc. in the recording, and finally form a voiceprint feature vector containing these features. The server loads the voiceprint feature vector into the basic feature type identification component. The task of this component is to further extract high-level features that can describe the basic characteristics of the bogie voiceprint from the feature vector. The basic feature type identification component performs basic high-level feature extraction on the voiceprint feature vector to obtain a series of basic high-level features. For example, it can identify mechanical vibration modes, fluid sound modes, etc. in the voiceprint. Then, the component calculates the confidence of the basic representation type based on these basic high-level features. It matches each basic high-level feature with a preset basic representation type and gives the confidence of each matching item. For example, it can determine that the confidence of the mechanical vibration mode in the voiceprint and the basic representation type of "mechanical vibration sound" is 80%. The server loads the voiceprint feature vector and basic high-level features into the refined feature type recognition component. The task of this component is to extract high-level features that can describe the more detailed characteristics of the bogie voiceprint from the feature vector, and perform feature integration and refined representation type recognition. The refined feature type recognition component extracts refined high-level features from the voiceprint feature vector to obtain a series of refined high-level features. For example, it can identify the vibration mode, abnormal sound mode, etc. of a specific component in the voiceprint. Then, the component integrates the voiceprint features based on the basic high-level features and the refined high-level features to form a refined voiceprint integrated feature containing more comprehensive information. Finally, the refined feature type recognition component uses the refined voiceprint integrated features to calculate the confidence of the refined representation type. It matches the refined voiceprint integrated features with the preset refined representation type and gives the confidence of each match. For example, it can determine that the confidence of the vibration mode of a specific component in the voiceprint and the refined representation type of "abnormal vibration sound of the bearing" is 90%. Through the above steps, the server can make full use of the three core components of the first target bogie voiceprint recognition model to comprehensively analyze and identify the bogie voiceprint data to be analyzed. From basic features to refined features, corresponding confidence assessments are given, which provides a rich data basis for the subsequent determination of voiceprint characterization types, helps the server to more accurately judge the operating status of the bogie and take appropriate measures to ensure the safe operation of the locomotive.

[0057] In an embodiment of the present invention, the refined feature type identification component includes a refined high-level feature extraction subcomponent, a voiceprint feature integration subcomponent and a refined feature type identification subcomponent; the loading of the voiceprint feature vector and the basic high-level features into the refined feature type identification component can be implemented through the following example.

[0058] Loading the voiceprint feature vector into the refined high-level feature extraction subcomponent to extract refined high-level features to obtain the refined high-level features;

[0059] Loading the basic high-level features and the refined high-level features into the voiceprint feature integration subcomponent to integrate the voiceprint features, and obtaining refined voiceprint integration features;

[0060] The refined voiceprint integrated features are loaded into the refined feature type identification subcomponent to perform refined characterization type confidence calculation to obtain the voiceprint characterization type confidence corresponding to each refined characterization type.

[0061] In the embodiment of the present invention, illustratively,

[0062] The server loads the voiceprint feature vector into the refined high-level feature extraction subcomponent. The task of this subcomponent is to extract high-level features that can describe the more detailed characteristics of the bogie voiceprint from the voiceprint feature vector. The refined high-level feature extraction subcomponent extracts refined high-level features from the voiceprint feature vector to obtain a series of refined high-level features. For example, it can identify the vibration mode of a specific component (such as a bearing) in the voiceprint, abnormal sound mode (such as wear sound), etc. These features can more specifically reflect the operating status of the bogie. The server loads the basic high-level features and the refined high-level features into the voiceprint feature integration subcomponent. The task of this subcomponent is to integrate the two features to form a refined voiceprint integrated feature containing more comprehensive information. The voiceprint feature integration subcomponent integrates voiceprint features based on the basic high-level features and the refined high-level features to obtain a refined voiceprint integrated feature. For example, it may combine the mechanical vibration mode (basic high-level feature) and the vibration mode of a specific component (refined high-level feature) to form a refined voiceprint integrated feature containing the overall and local vibration information of the bogie. The server loads the refined voiceprint integrated feature into the refined feature type identification subcomponent. The task of this subcomponent is to use the refined voiceprint integrated features to perform the refined characterization type confidence calculation. The refined feature type identification subcomponent uses the refined voiceprint integrated features to perform the refined characterization type confidence calculation. It matches the refined voiceprint integrated features with the preset refined characterization types and gives the confidence of each match. For example, it can determine that the confidence of the vibration mode of a specific component in the voiceprint and the refined characterization type of "abnormal vibration sound of the bearing" is 90%, which means that the server is 90% confident that the abnormal vibration sound of the bearing is contained in the voiceprint. Through the above steps, the server can make full use of the three subcomponents of the refined feature type identification component to conduct in-depth analysis and identification of the bogie voiceprint data to be analyzed. From the extraction of refined high-level features to the integration of voiceprint features, and then to the confidence calculation of the refined characterization type, detailed data support is provided to help the server more accurately judge the operating status of the bogie and take appropriate measures to ensure the safe operation of the locomotive.

[0063] In the embodiments of the present invention, the following implementation modes are also provided.

[0064] Extracting intermediate-granularity advanced features from the voiceprint feature vector to obtain intermediate-granularity advanced features;

[0065] Integrate voiceprint features according to the basic high-level features and the intermediate-level granularity high-level features to obtain intermediate-level integrated features, perform intermediate-level granularity type recognition according to the intermediate-level integrated features, and obtain voiceprint representation type confidence corresponding to each intermediate-level representation type;

[0066] Performing voiceprint feature integration on the basic high-level features, the intermediate integrated features and the refined high-level features to obtain target refined voiceprint integrated features, and using the target refined voiceprint integrated features to perform refined feature type identification to obtain target voiceprint representation type confidence corresponding to each refined representation type, wherein the classification granularity corresponding to the refined representation type is greater than the classification granularity corresponding to the intermediate representation type, and the classification granularity corresponding to the intermediate representation type is greater than the classification granularity corresponding to the basic representation type;

[0067] The target voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is obtained according to the voiceprint representation type confidence corresponding to each basic representation type, the voiceprint representation type confidence corresponding to each intermediate representation type and the target voiceprint representation type confidence corresponding to each refined representation type. The target voiceprint representation type includes a target basic representation type, a target intermediate representation type corresponding to the target basic representation type and a target refined representation type corresponding to the target intermediate representation type.

[0068] In the embodiment of the present invention, the server performs intermediate granularity advanced feature extraction on the voiceprint feature vector. This step involves extracting advanced features that can describe the intermediate granularity characteristics of the bogie voiceprint from the voiceprint feature vector. For example, the server can identify vibration patterns within certain specific frequency ranges in the voiceprint as intermediate granularity advanced features. The server integrates the voiceprint features according to the basic advanced features and the intermediate granularity advanced features to obtain intermediate integrated features. This step involves merging the two features to form a feature representation containing more extensive information. For example, the server may combine the mechanical vibration pattern (basic advanced feature) and the vibration pattern within a specific frequency range (intermediate granularity advanced feature) to form an intermediate integrated feature. Then, the server performs intermediate granularity type identification according to the intermediate integrated features, and obtains the voiceprint representation type confidence corresponding to each intermediate representation type. This step involves matching the intermediate integrated features with the preset intermediate representation types, and giving the confidence of each matching item. For example, the server can determine that the confidence of the vibration pattern within a specific frequency range in the voiceprint and the intermediate representation type of "intermediate frequency vibration sound" is 70%. The server integrates the basic high-level features, the intermediate integrated features, and the refined high-level features into voiceprint features to obtain the target refined voiceprint integrated features. This step involves merging the three features to form a feature representation that contains the most comprehensive information. For example, the server may combine the mechanical vibration mode (basic high-level feature), the vibration mode within a specific frequency range (intermediate granularity high-level feature), and the vibration mode of a specific component (refined high-level feature) to form a target refined voiceprint integrated feature. Then, the server uses the target refined voiceprint integrated feature to identify the refined feature type and obtain the target voiceprint representation type confidence corresponding to each refined representation type. This step involves matching the target refined voiceprint integrated feature with the preset refined representation type and giving the confidence of each match. Since the classification granularity of the refined representation type is larger than that of the intermediate representation type, this step can provide more detailed recognition results. For example, the server can determine that the confidence of the specific component vibration mode in the voiceprint and the refined representation type of "abnormal vibration sound of the bearing" is 90%, which is part of the final target voiceprint representation type. Finally, the server obtains the target voiceprint characterization type corresponding to the bogie voiceprint data to be analyzed according to the voiceprint characterization type confidence corresponding to each basic characterization type, the voiceprint characterization type confidence corresponding to each intermediate characterization type, and the target voiceprint characterization type confidence corresponding to each refined characterization type. This target voiceprint characterization type includes the target basic characterization type, the target intermediate characterization type corresponding to the target basic characterization type, and the target refined characterization type corresponding to the target intermediate characterization type. For example, the server can finally determine that the target voiceprint characterization type of the bogie voiceprint data to be analyzed is "mechanical vibration sound-medium frequency vibration sound-abnormal vibration sound of the bearing", and give the corresponding confidence of each level.Through the above steps, the server can conduct a comprehensive and detailed analysis and identification of the bogie voiceprint data to be analyzed, and give corresponding confidence assessments from basic features to intermediate features and then to detailed features. This helps the server to more accurately judge the operating status of the bogie and take appropriate measures to ensure the safe operation of the locomotive.

[0069] In the embodiment of the present invention, the voiceprint feature integration is performed according to the basic high-level features and the intermediate-level granularity high-level features to obtain the intermediate-level integrated features, which can be implemented through the following examples.

[0070] Merging the basic high-level features and the intermediate-level high-level features to obtain intermediate voiceprint merged features;

[0071] Incremental feature extraction is performed based on the intermediate voiceprint merged feature and the refined high-level feature to obtain the intermediate incremental feature, and the intermediate incremental feature is subjected to feature normalization processing to obtain the intermediate integrated feature.

[0072] In an embodiment of the present invention, exemplarily, the server performs feature merging of the basic high-level features and the intermediate granularity high-level features. This step involves combining feature vectors of two different granularities into a new feature vector. For example, the server may merge the basic high-level features representing the mechanical vibration mode and the intermediate granularity high-level features representing the vibration mode within a specific frequency range to form an intermediate voiceprint merged feature containing both information. Next, the server performs incremental feature extraction based on the intermediate voiceprint merged feature and the refined high-level features. This step involves further extracting incremental information related to the refined high-level features from the intermediate voiceprint merged feature to form a richer feature representation. For example, the server can identify additional vibration modes related to the vibration mode of a specific component (refined high-level features) from the intermediate voiceprint merged feature as intermediate incremental features. Then, the server performs feature normalization on the intermediate incremental features. This step involves scaling the intermediate incremental features to have a uniform scale for more effective comparison and integration with other features. For example, the server can use the maximum and minimum normalization method to scale the numerical range of the intermediate incremental features to between 0 and 1 to obtain the intermediate integrated features. Through the above steps, the server can effectively integrate basic high-level features, intermediate-level granularity high-level features, and refined high-level features to form an intermediate-level integrated feature containing multi-granularity information. This helps the server to more accurately identify the refined representation type of the bogie voiceprint data to be analyzed in subsequent steps, and ultimately determine the target voiceprint representation type. For example, in actual applications, the server can use the intermediate-level integrated features to identify the "abnormal bearing vibration sound" in the bogie voiceprint and give a corresponding confidence assessment, thereby providing a strong guarantee for the safe operation of the locomotive.

[0073] In the embodiment of the present invention, the voiceprint feature integration of the basic high-level features, the intermediate integrated features and the refined high-level features to obtain the target refined voiceprint integrated features can be implemented through the following examples.

[0074] Merging the basic high-level features, the intermediate integrated features, and the refined high-level features to obtain a target voiceprint merged feature;

[0075] Incremental feature extraction is performed based on the target voiceprint merged feature and the refined high-level feature to obtain the target incremental feature, and feature normalization is performed on the target incremental feature to obtain the target refined voiceprint integrated feature.

[0076] In an embodiment of the present invention, exemplarily, the server performs feature merging of basic high-level features, intermediate integrated features, and refined high-level features. This step involves combining feature vectors of three different granularities into a new feature vector. For example, the server may merge basic high-level features representing mechanical vibration modes, intermediate integrated features representing vibration modes within a specific frequency range combined with component vibration modes, and refined high-level features representing detailed vibration modes of specific components to form a target voiceprint merged feature containing these three types of information. Next, the server performs incremental feature extraction based on the target voiceprint merged feature and the refined high-level features. This step involves further extracting incremental information related to the refined high-level features from the target voiceprint merged feature to form a richer and more specific feature representation. For example, the server can identify additional details or changes related to the detailed vibration mode (refined high-level features) of a specific component from the target voiceprint merged feature as the target incremental feature. Then, the server performs feature normalization on the target incremental feature. This step involves scaling the target incremental feature to have a uniform scale for more effective comparison and integration with other features. For example, the server can use the Z-score normalization method to scale the numerical range of the target incremental feature to a range with a uniform mean and standard deviation, thereby obtaining the target refined voiceprint integrated feature. Through the above steps, the server can effectively integrate the basic high-level features, intermediate integrated features, and refined high-level features to form a target refined voiceprint integrated feature containing multi-granularity and rich information. This helps the server to more accurately identify the refined characterization type of the bogie voiceprint data to be analyzed in subsequent steps, and ultimately determine the target voiceprint characterization type. For example, in actual applications, the server can use the target refined voiceprint integrated feature to identify the "abnormal vibration sound of the bearing-specific frequency" in the bogie voiceprint, and give a corresponding confidence assessment, thereby providing a strong guarantee for the safe operation of the locomotive.

[0077] In the embodiments of the present invention, the following implementation modes are also provided.

[0078] Loading the bogie voiceprint data to be analyzed into a second target bogie voiceprint recognition model;

[0079] Using the second target bogie voiceprint recognition model, performing a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed to obtain a voiceprint feature vector;

[0080] Using the second target bogie voiceprint recognition model to extract basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and performing basic feature type identification based on the basic and advanced features to obtain a voiceprint representation type confidence corresponding to each basic representation type;

[0081] Using the second target bogie voiceprint recognition model, extracting intermediate-granularity advanced features from the voiceprint feature vector to obtain intermediate-granularity advanced features, integrating voiceprint features according to the basic advanced features and the intermediate-granularity advanced features to obtain intermediate-level integrated features, and performing intermediate-granularity type identification according to the intermediate-level integrated features to obtain voiceprint representation type confidence corresponding to each intermediate representation type;

[0082] Using the second target bogie voiceprint recognition model to perform voiceprint feature integration on the basic high-level features, the intermediate integrated features and the refined high-level features to obtain target refined voiceprint integrated features, and using the target refined voiceprint integrated features to perform refined feature type identification to obtain target voiceprint representation type confidence corresponding to each refined representation type;

[0083] The target voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is obtained according to the voiceprint representation type confidence corresponding to each basic representation type, the voiceprint representation type confidence corresponding to each intermediate representation type and the target voiceprint representation type confidence corresponding to each refined representation type. The target voiceprint representation type includes a target basic representation type, a target intermediate representation type corresponding to the target basic representation type and a target refined representation type corresponding to the target intermediate representation type.

[0084] In an embodiment of the present invention, exemplarily, the server loads the bogie voiceprint data to be analyzed into the second target bogie voiceprint recognition model. This step involves reading the voiceprint data from the storage medium and inputting it into the recognition model. For example, the server can read a section of the bogie voiceprint data of the locomotive from the database and then load it into the second target bogie voiceprint recognition model. The server uses the second target bogie voiceprint recognition model to perform a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed. This step involves extracting a feature vector that can describe its characteristics from the voiceprint data. For example, the server can use the feature extraction layer in the model to identify the key frequency, amplitude and phase information in the voiceprint, thereby forming a voiceprint feature vector. The server uses the second target bogie voiceprint recognition model to perform basic high-level feature extraction on the voiceprint feature vector. This step involves extracting high-level features that can describe the basic characteristics of the bogie voiceprint from the voiceprint feature vector. For example, the server can identify the mechanical vibration mode in the voiceprint as the basic high-level feature. Then, the server performs basic feature type recognition based on the basic high-level features and obtains the voiceprint representation type confidence corresponding to each basic representation type. This step involves matching the basic high-level features with the preset basic representation types and giving the confidence of each match. For example, the server can determine that the confidence of the mechanical vibration mode in the voiceprint with the basic representation type of "mechanical vibration sound" is 80%. The server uses the second target bogie voiceprint recognition model to extract the voiceprint feature vector with intermediate granularity high-level features. This step involves extracting high-level features that can describe the intermediate granularity characteristics of the bogie voiceprint from the voiceprint feature vector. For example, the server can identify the vibration mode within a specific frequency range in the voiceprint as an intermediate granularity high-level feature. Then, the server integrates the voiceprint features based on the basic high-level features and the intermediate granularity high-level features to obtain intermediate integrated features. This step involves merging the two features to form a feature representation that contains more extensive information. Next, the server performs intermediate granularity type recognition based on the intermediate integrated features and obtains the voiceprint representation type confidence corresponding to each intermediate representation type. This step involves matching the intermediate integrated features with the preset intermediate representation types and giving the confidence of each match. For example, the server may determine that the confidence of the vibration pattern within a specific frequency range in the voiceprint and the intermediate characterization type of "intermediate frequency vibration sound" is 70%. The server uses the second target bogie voiceprint recognition model to integrate the basic high-level features, intermediate integrated features, and refined high-level features into voiceprint features. This step involves merging the three features to form a feature representation that contains the most comprehensive information. The server then uses the target refined voiceprint integrated features to identify the refined feature type and obtains the target voiceprint characterization type confidence corresponding to each refined characterization type. This step involves matching the target refined voiceprint integrated features with the preset refined characterization type and giving the confidence of each match.Since the classification granularity of the refined characterization type is larger than that of the intermediate characterization type, this step can provide more detailed recognition results. For example, the server can determine that the confidence of the vibration mode of a specific component in the voiceprint and the refined characterization type of "abnormal vibration sound of the bearing" is 90%. Finally, the server obtains the target voiceprint characterization type corresponding to the bogie voiceprint data to be analyzed based on the voiceprint characterization type confidence corresponding to each basic characterization type, the voiceprint characterization type confidence corresponding to each intermediate characterization type, and the target voiceprint characterization type confidence corresponding to each refined characterization type. This target voiceprint characterization type includes the target basic characterization type, the target intermediate characterization type corresponding to the target basic characterization type, and the target refined characterization type corresponding to the target intermediate characterization type. For example, the server can finally determine that the target voiceprint characterization type of the bogie voiceprint data to be analyzed is "mechanical vibration sound-intermediate frequency vibration sound-abnormal vibration sound of the bearing", and give the corresponding confidence of each level. Through the above steps, the server can use the second target bogie voiceprint recognition model to conduct a comprehensive and detailed analysis and recognition of the bogie voiceprint data to be analyzed, and give corresponding confidence assessments from basic features to intermediate features and then to detailed features. This helps the server to more accurately judge the operating status of the bogie and take appropriate measures to ensure the safe operation of the locomotive.

[0085] In an embodiment of the present invention, when an emergency strategy determination request including a locomotive bogie analysis result is obtained, a strategy retrieval is performed on the multi-dimensional semantic features corresponding to each preset emergency strategy based on the semantic analysis features of the locomotive bogie analysis result to obtain the emergency strategy determination result, which can be implemented through the following example.

[0086] When an emergency strategy determination request including a locomotive bogie analysis result is obtained, semantic feature extraction is performed on the locomotive bogie analysis result to obtain a semantic analysis feature;

[0087] According to the multi-dimensional semantic features corresponding to each preset emergency strategy and in combination with the semantic analysis features, a strategy search is performed on the locomotive bogie analysis result and each preset emergency strategy to obtain an emergency strategy determination result.

[0088] In an embodiment of the present invention, exemplarily, the server first extracts semantic features from the analysis results of the locomotive bogie. This step involves extracting key information that can describe its meaning and context from the analysis results to form semantic analysis features. For example, the server can extract key semantic information such as "abnormal bearing vibration" and "significant medium-frequency vibration sound" from the analysis results. Then, the server performs strategy retrieval based on the multi-dimensional semantic features corresponding to each preset emergency strategy and the previously extracted semantic analysis features. This step involves matching the semantic features in the analysis results with the multi-dimensional semantic features of each preset emergency strategy to find the emergency strategy that best matches the current analysis results. For example, the server can have a preset emergency strategy whose multi-dimensional semantic features include "emergency inspection for bearing abnormalities" and "special attention to medium-frequency vibration sounds", etc., which are highly matched with the semantic features in the current analysis results. Finally, the server obtains the emergency strategy determination result based on the results of the strategy retrieval. This step involves selecting the strategy that best matches the current analysis result from all preset emergency strategies as a guide for emergency response. For example, the server can finally determine "perform emergency inspection for bearing abnormalities and pay special attention to medium-frequency vibration sounds" as the current emergency strategy. Through the above steps, the server can quickly and accurately determine the corresponding emergency strategy based on the semantic features of the locomotive bogie analysis results, providing strong protection for the safe operation of the locomotive. In practical applications, this helps the server to respond to possible faults or abnormalities of the bogie in a timely manner and take appropriate measures to prevent further development of the situation.

[0089] In an embodiment of the present invention, according to the multi-dimensional semantic features corresponding to each preset emergency strategy, combined with the semantic analysis features, a strategy retrieval is performed on the locomotive bogie analysis results and each preset emergency strategy to obtain an emergency strategy determination result, which can be implemented through the following examples.

[0090] According to the multi-dimensional semantic features corresponding to each preset emergency strategy and the semantic analysis features, the locomotive bogie analysis results and each preset emergency strategy are retrieved to obtain an emergency strategy determination result; or,

[0091] According to the enhanced features corresponding to each preset emergency strategy and the semantic analysis features, the locomotive bogie analysis results and each preset emergency strategy are retrieved to obtain the emergency strategy determination result; wherein each enhanced feature is obtained based on the multi-dimensional semantic features of each associated preset emergency strategy associated with the corresponding preset emergency strategy.

[0092] In an embodiment of the present invention, exemplarily, the server first matches the multidimensional semantic features corresponding to each preset emergency strategy with the semantic analysis features extracted from the analysis results of the locomotive bogie. This step involves comparing the key semantic information in the analysis results with the multidimensional features described by each preset emergency strategy to find the most similar strategy. Then, the server sorts the preset emergency strategies according to the degree of matching and selects the strategy with the highest matching degree as the emergency strategy determination result. For example, if the semantic features in the analysis results show "abnormal bearing vibration" and "significant medium-frequency vibration sound", the server will select those preset emergency strategies that pay special attention to bearing abnormalities and medium-frequency vibration sounds. In this scenario, the server first generates enhanced features for each preset emergency strategy. These enhanced features are obtained based on the multidimensional semantic features of other preset emergency strategies associated with the strategy, representing the correlation and common concerns between the strategies. Then, the server uses these enhanced features to match the semantic analysis features extracted from the analysis results of the locomotive bogie. This step involves analyzing the similarity between the key information in the enhanced features and the semantic features in the analysis results. Finally, the server selects the most appropriate preset emergency strategy as the emergency strategy determination result based on the matching results. For example, if the semantic features in the analysis results are highly matched with the reinforcement features of a preset emergency strategy, indicating that the strategy has a higher priority or effectiveness in dealing with similar situations, the server will select the strategy as a guide for emergency response. Through the above steps, the server can quickly and accurately determine the corresponding emergency strategy based on the semantic features of the locomotive bogie analysis results and the multi-dimensional or reinforcement features of the preset emergency strategy, providing strong guarantees for the safe operation of the locomotive. In practical applications, this helps the server respond to possible faults or abnormalities of the bogie in a timely manner and take appropriate measures to prevent further development of the situation.

[0093] In an embodiment of the present invention, the enhanced features of a preset emergency strategy are obtained in the following manner:

[0094] According to the multi-dimensional semantic features of each preset emergency strategy, from each other preset emergency strategy in each preset emergency strategy except the one preset emergency strategy, extract other preset emergency strategies that meet the preset strategy association criteria as the associated preset emergency strategies of the one preset emergency strategy;

[0095] According to the extracted multi-dimensional semantic features corresponding to each associated preset emergency strategy, feature enhancement processing is performed on the preset emergency strategy to obtain enhanced features of the preset emergency strategy.

[0096] In an embodiment of the present invention, illustratively, the server first analyzes the multi-dimensional semantic features of each preset emergency strategy. Then, for a given preset emergency strategy, the server extracts those strategies that meet the preset strategy association criteria from all other preset emergency strategies as the associated preset emergency strategies of the preset emergency strategy. For example, suppose there is a preset emergency strategy A, whose multi-dimensional semantic features include "urgent inspection for bearing abnormalities" and "special attention to medium-frequency vibration sounds". The server will find that preset emergency strategies B and C also contain similar semantic features, such as "regular inspection of bearings" and "pay attention to vibrations in the medium frequency band", so B and C are used as the associated preset emergency strategies of A. Once the server determines the associated preset emergency strategies of a given preset emergency strategy, it will use the multi-dimensional semantic features of these associated strategies to perform feature enhancement processing on the preset emergency strategy. Continuing with the above example, the server will analyze the multi-dimensional semantic features of the associated preset emergency strategies B and C, and find some additional key information, such as "inspection with specific tools" or "inspection within a specific time period". Then, the server will add these additional key information to the preset emergency strategy A as its enhanced features. After feature enhancement, the server obtains enhanced features for a given preset emergency strategy. These enhanced features not only contain the original multi-dimensional semantic features, but also incorporate additional key information from the associated preset emergency strategy. In the above example, after feature enhancement, the enhanced features of preset emergency strategy A can include "urgent inspection for bearing abnormalities", "special attention to medium-frequency vibration sounds", "inspection using specific tools", and "inspection within a specific time period". Through the above steps, the server is able to generate enhanced features containing rich information for each preset emergency strategy. These features not only reflect the multi-dimensional semantic characteristics of the strategy itself, but also incorporate additional key information related to other associated strategies. In practical applications, this helps the server to more accurately match the analysis results of the locomotive bogie with the preset emergency strategy, thereby selecting more appropriate emergency response measures.

[0097] In an embodiment of the present invention, the feature enhancement processing is performed on the preset emergency strategy according to the extracted multi-dimensional semantic features corresponding to each associated preset emergency strategy to obtain the enhanced features of the preset emergency strategy, which can be implemented through the following examples.

[0098] The semantic description matching degree between each of the associated preset emergency strategies and the one preset emergency strategy is used as the influencing factor corresponding to each of the associated preset emergency strategies; wherein each semantic description matching degree is determined according to the multi-dimensional semantic features of the corresponding associated preset emergency strategy and the one preset emergency strategy;

[0099] According to the influencing factors corresponding to each associated preset emergency strategy, the multi-dimensional mean characteristics corresponding to each associated preset emergency strategy are linearly superimposed to obtain the enhanced characteristics of the one preset emergency strategy.

[0100] In an embodiment of the present invention, illustratively, the server first analyzes the degree of semantic description matching between each associated preset emergency strategy and a given preset emergency strategy. This step involves comparing the multi-dimensional semantic features of the two strategies and calculating the similarity or matching degree between them. For example, assume that preset emergency strategy A is about "urgent inspection of bearing abnormalities" and associated preset emergency strategy B is about "regular inspection of bearings". The server analyzes the multi-dimensional semantic features of the two strategies and determines the degree of matching between them, such as they both focus on the key component "bearing" and therefore have a high degree of semantic description matching. The server uses the degree of semantic description matching between each associated preset emergency strategy and the given preset emergency strategy as the impact factor corresponding to the associated strategy. This impact factor reflects the importance or weight of the associated strategy in strengthening the given strategy. Next, the server averages the multi-dimensional semantic features of each associated preset emergency strategy to ensure that the features between different strategies are comparable. Then, the server linearly superimposes these averaged multi-dimensional semantic features according to the impact factor corresponding to each associated preset emergency strategy. This step involves weighting the features of each associated strategy according to its impact factor and adding the weighted features to form the enhanced features of the given preset emergency strategy. Continuing with the above example, if the impact factor of the associated preset emergency strategy B is higher, then it will make a greater contribution to the enhanced features of the given preset emergency strategy A during the linear superposition process. After the linear superposition process, the server obtains the enhanced features of the given preset emergency strategy. These enhanced features not only contain the multi-dimensional semantic information of the original strategy, but also incorporate the features of the associated strategy, and are weighted according to the degree of semantic description matching and the impact factor between them. In the above example, the enhanced preset emergency strategy A can not only focus on "emergency inspection of bearing abnormalities", but also include some key information about "regular inspection" in the associated strategy B, which is weighted and integrated according to the degree of semantic description matching and the impact factor between them. Through the above steps, the server can generate enhanced features containing rich information and associated strategy features for each preset emergency strategy. In practical applications, this helps the server to more accurately match the analysis results of the locomotive bogie with the preset emergency strategy, so as to select more appropriate emergency response measures.

[0101] In an embodiment of the present invention, according to the multi-dimensional semantic features of each preset emergency strategy, from each other preset emergency strategy in each preset emergency strategy except the one preset emergency strategy, other preset emergency strategies that meet the preset strategy association criteria are extracted as associated preset emergency strategies of the one preset emergency strategy, which can be implemented through the following examples.

[0102] Obtaining the multi-dimensional averaged feature corresponding to each preset emergency strategy; wherein the multi-dimensional averaged feature of each other preset emergency strategy is obtained by averaging multiple semantic features in the corresponding multi-dimensional semantic features;

[0103] According to the multi-dimensional averaged features corresponding to each other preset emergency strategy except the one preset emergency strategy in each preset emergency strategy, obtaining the semantic description matching degree between each other preset emergency strategy and the one preset emergency strategy;

[0104] According to the obtained matching degree of each semantic description, other preset emergency strategies that meet the preset strategy association criteria are extracted from each other preset emergency strategy as the associated preset emergency strategy of the one preset emergency strategy.

[0105] In an embodiment of the present invention, exemplarily, the server first obtains the multi-dimensional averaged features corresponding to each preset emergency strategy. This step involves averaging the multi-dimensional semantic features of each preset emergency strategy to obtain a averaged feature representation. For example, assume that preset emergency strategy A has multiple semantic features, such as "bearing abnormality", "emergency inspection" and "medium frequency vibration". The server averages these semantic features to generate a averaged feature vector as the multi-dimensional averaged feature of strategy A. Next, the server calculates the degree of semantic description matching between each other preset emergency strategy except the given preset emergency strategy and the strategy. This step involves comparing the multi-dimensional averaged features of the two strategies and calculating the similarity or matching degree between them. Continuing with the above example, the server calculates the degree of semantic description matching between preset emergency strategies B, C, etc. and strategy A. If the averaged features of strategy B are highly similar to the averaged features of strategy A, then the degree of semantic description matching between them is high. Finally, the server extracts the strategy that meets the preset strategy association criteria from all other preset emergency strategies according to the degree of matching of each semantic description as the associated preset emergency strategy of the given preset emergency strategy. In the above example, if the preset policy association criteria stipulate that the semantic description matching degree must be higher than a certain threshold, then the server will select a policy from all other policies that matches policy A higher than the threshold, such as policy B, as the associated preset emergency policy of policy A. Through the above steps, the server can extract other associated preset emergency policies for each preset emergency policy. In practical applications, this helps the server consider more relevant information and possible associated policies when determining the emergency policy, so as to make more accurate decisions. For example, when the analysis results of the locomotive bogie match the preset emergency policy A, the server can also consider the information in its associated policy B to formulate more comprehensive emergency response measures.

[0106] In an embodiment of the present invention, according to the multi-dimensional semantic features corresponding to each preset emergency strategy and the semantic analysis features, the locomotive bogie analysis results and each preset emergency strategy are retrieved to obtain the emergency strategy determination result, which can be implemented through the following examples.

[0107] According to a preset feature extraction method, a target feature matching coefficient is extracted from feature matching coefficients of each of a plurality of semantic features in a multi-dimensional semantic feature of a preset emergency strategy and the semantic analysis feature;

[0108] According to the target feature matching coefficient corresponding to each preset emergency strategy, the emergency strategy determination result is obtained.

[0109] In an embodiment of the present invention, exemplarily, the server first processes the relationship between the multi-dimensional semantic features and the semantic analysis features of a preset emergency strategy according to a preset feature extraction method. This step involves calculating the feature matching coefficient between each semantic feature in the multi-dimensional semantic features and the semantic analysis features. For example, assume that the preset emergency strategy A has multi-dimensional semantic features, such as "bearing abnormality", "emergency inspection" and "medium frequency vibration", and the semantic analysis features are extracted from the soundprint analysis of the locomotive bogie, such as "medium frequency vibration is detected". The server will calculate the feature matching coefficient between each semantic feature in strategy A and the semantic analysis feature of "medium frequency vibration is detected". In this process, the server will extract the target feature matching coefficient from these feature matching coefficients. The target feature matching coefficient can be the highest feature matching coefficient, or a feature matching coefficient selected according to a certain criterion (such as a threshold). Then, the server will determine which preset emergency strategy is most suitable for the current locomotive bogie analysis result based on the target feature matching coefficient corresponding to each preset emergency strategy. Continuing with the above example, if the semantic feature "medium frequency vibration" in the preset emergency strategy A has the highest feature matching coefficient with the semantic analysis feature "medium frequency vibration detected", then the server will select strategy A as the emergency strategy determination result. Through the above steps, the server can retrieve and select the most appropriate emergency strategy based on the voiceprint analysis results of the locomotive bogie and the multi-dimensional semantic features of the preset emergency strategy. In practical applications, this helps the server to quickly and accurately respond to abnormal conditions that may occur in the locomotive bogie, thereby ensuring the safe operation of the locomotive.

[0110] In an embodiment of the present invention, according to the enhanced features corresponding to each preset emergency strategy and the semantic analysis features, the locomotive bogie analysis results and each preset emergency strategy are retrieved to obtain the emergency strategy determination result, which can be implemented through the following examples.

[0111] According to the reinforcement feature corresponding to each preset emergency strategy and the semantic analysis feature, the reinforcement feature matching coefficient corresponding to each preset emergency strategy is obtained, and according to the multi-dimensional semantic feature corresponding to each preset emergency strategy and the semantic analysis feature, the multi-dimensional semantic feature matching coefficient corresponding to each preset emergency strategy is obtained;

[0112] The enhanced feature matching coefficients and multi-dimensional semantic feature matching coefficients corresponding to each preset emergency strategy are respectively integrated to obtain the integrated matching coefficients corresponding to each preset emergency strategy, and the emergency strategy determination result is obtained based on each integrated matching coefficient obtained.

[0113] In an embodiment of the present invention, exemplarily, the server first calculates the matching coefficient of the enhanced features corresponding to each preset emergency strategy according to the enhanced features and the semantic analysis features corresponding to each preset emergency strategy. This step involves comparing the similarity or matching degree between the enhanced features and the semantic analysis features. At the same time, the server also calculates the matching coefficient of the multidimensional semantic features corresponding to each preset emergency strategy according to the multidimensional semantic features and the semantic analysis features corresponding to each preset emergency strategy. This step involves comparing the similarity or matching degree between the multidimensional semantic features and the semantic analysis features. For example, assuming that the enhanced features of the preset emergency strategy A include "urgent inspection of bearing abnormalities" and "using medium-frequency vibration detection tools", and the semantic analysis features are extracted from the soundprint analysis of the locomotive bogie, such as "medium-frequency vibration is detected". The server calculates the enhanced feature matching coefficient between the enhanced features and the semantic analysis features of strategy A, and the multidimensional semantic feature matching coefficient between the multidimensional semantic features and the semantic analysis features of strategy A. Then, the server integrates the enhanced feature matching coefficient and the multidimensional semantic feature matching coefficient corresponding to each preset emergency strategy respectively to obtain the integrated matching coefficient corresponding to each preset emergency strategy. This step involves weighted summation or other forms of integration of the two matching coefficients. Finally, the server selects the preset emergency strategy with the highest integrated matching coefficient as the emergency strategy determination result based on each integrated matching coefficient obtained. Continuing with the above example, if the preset emergency strategy A has the highest integrated matching coefficient, the server will select strategy A as the emergency strategy determination result. Through the above steps, the server can retrieve and select the most appropriate emergency strategy based on the voiceprint analysis results of the locomotive bogie, the enhanced features of the preset emergency strategy, and the multi-dimensional semantic features. In practical applications, this helps the server to consider various relevant information more comprehensively, thereby more accurately determining the emergency strategy to ensure the safe operation of the locomotive.

[0114] In an embodiment of the present invention, according to the multi-dimensional semantic features corresponding to each preset emergency strategy, combined with the semantic analysis features, a strategy retrieval is performed on the locomotive bogie analysis results and each preset emergency strategy to obtain an emergency strategy determination result, which can be implemented through the following examples.

[0115] According to the multi-dimensional semantic features corresponding to each preset emergency strategy, combined with the semantic analysis features, the overall feature matching coefficient corresponding to each preset emergency strategy is obtained;

[0116] According to the policy element features of each policy element included in each of the preset emergency strategies, combined with the semantic analysis features, obtaining the element feature matching coefficient corresponding to each of the preset emergency strategies;

[0117] According to the overall feature matching coefficient and the element feature matching coefficient corresponding to each preset emergency strategy, a strategy search is performed on the locomotive bogie analysis result and each preset emergency strategy to obtain an emergency strategy determination result.

[0118] In an embodiment of the present invention, exemplarily, the server first calculates the overall feature matching coefficient corresponding to each preset emergency strategy according to the multi-dimensional semantic features corresponding to each preset emergency strategy, combined with the semantic analysis features. This step involves a comprehensive comparison of the multi-dimensional semantic features with the semantic analysis features to evaluate the overall similarity or matching degree between them. For example, assuming that the multi-dimensional semantic features of the preset emergency strategy A include "bearing abnormality", "emergency inspection" and "medium frequency vibration", and the semantic analysis features are extracted from the soundprint analysis of the locomotive bogie, such as "medium frequency vibration and abnormal sound are detected". The server will calculate the overall feature matching coefficient between the multi-dimensional semantic features of strategy A and the semantic analysis features. Next, the server calculates the element feature matching coefficient corresponding to each preset emergency strategy according to the policy element features of each policy element contained in each preset emergency strategy, combined with the semantic analysis features. This step involves comparing the policy element features with the semantic analysis features to evaluate the specific similarity or matching degree between them. Continuing with the above example, the server will calculate the element feature matching coefficients between the three policy elements "bearing abnormality", "emergency inspection" and "medium frequency vibration" in strategy A and the semantic analysis features respectively. Finally, the server performs strategy retrieval on the locomotive bogie analysis results and each preset emergency strategy according to the overall feature matching coefficient and element feature matching coefficient corresponding to each preset emergency strategy. This step involves comprehensive consideration of the overall matching degree and the specific matching degree to determine the most appropriate emergency strategy. The server performs weighted summation or other forms of integration of the overall feature matching coefficient and the element feature matching coefficient to obtain the comprehensive matching coefficient of each preset emergency strategy. Then, the server selects the preset emergency strategy with the highest comprehensive matching coefficient as the emergency strategy determination result. In the above example, if the preset emergency strategy A has the highest comprehensive matching coefficient, the server will select strategy A as the emergency strategy determination result. Through the above steps, the server can perform comprehensive strategy retrieval and select the most appropriate emergency strategy based on the voiceprint analysis results of the locomotive bogie, the multi-dimensional semantic features of the preset emergency strategy, and the strategy element features. This helps the server determine the emergency strategy more accurately to quickly respond to abnormal situations that may occur in the locomotive bogie, thereby ensuring the safe operation of the locomotive.

[0119] In an embodiment of the present invention, the element feature matching coefficient corresponding to each preset emergency strategy is obtained according to the policy element features of each policy element included in each preset emergency strategy, in combination with the semantic analysis features, and can be implemented through the following examples.

[0120] According to the preset policy analysis criteria, a preset emergency policy is segmented to obtain each policy element, and semantic features are extracted for each policy element to obtain policy element features corresponding to each policy element;

[0121] According to the obtained policy element matching coefficient between each policy element feature and the semantic analysis feature, the element feature matching coefficient corresponding to the preset emergency policy is obtained.

[0122] In an embodiment of the present invention, exemplarily, the server first performs a segmentation process on a preset emergency strategy according to a preset strategy parsing criterion to obtain each strategy element in the strategy. This step involves splitting the preset emergency strategy into smaller strategy elements with specific meanings. Next, the server extracts semantic features for each strategy element respectively to obtain strategy element features corresponding to each strategy element. This step involves extracting key information in the strategy element to form semantic features that can represent the strategy element. For example, assume that the preset emergency strategy A is "conduct an emergency inspection on bearing abnormalities and use a medium frequency vibration detection tool". The server will cut strategy A into two strategy elements, "conduct an emergency inspection on bearing abnormalities" and "use a medium frequency vibration detection tool", according to the preset strategy parsing criterion, and extract their semantic features, such as "bearing abnormalities", "emergency inspection" and "medium frequency vibration", "detection tool". The server calculates the element feature matching coefficient corresponding to the preset emergency strategy based on the strategy element matching coefficient between each strategy element feature obtained and the semantic analysis feature. This step involves comparing the similarity or matching degree between the strategy element features and the semantic analysis features to obtain the matching coefficient of each strategy element, and these matching coefficients can be further integrated to obtain the element feature matching coefficient of the entire preset emergency strategy. Continuing the above example, the server will calculate the strategy element matching coefficient between the four strategy element features of "bearing abnormality", "emergency inspection", "medium frequency vibration", and "detection tool" in strategy A and the semantic analysis features, and can integrate these matching coefficients by weighted summation or other methods to obtain the element feature matching coefficient of strategy A. Through the above steps, the server can calculate the element feature matching coefficient of each preset emergency strategy based on the strategy elements and their features in each preset emergency strategy, combined with the semantic analysis features. This helps the server to consider the various parts of the preset emergency strategy more carefully and more accurately determine the emergency strategy that best matches the analysis results of the locomotive bogie.

[0123] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the above-mentioned locomotive bogie operating state monitoring method based on voiceprint. Figure 2 As shown, Figure 2 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111 , a processor 112 , and a communication unit 113 .

[0124] To achieve data transmission or interaction, the memory 111, the processor 112, and the communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0125] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A method for monitoring the running status of a locomotive bogie based on voiceprint, characterized in that: include: Acquiring the bogie voiceprint data to be analyzed of the locomotive bogie by means of a voiceprint collection sensor arranged on the locomotive bogie; Perform basic feature type recognition and detailed feature type recognition on the bogie voiceprint data to be analyzed, and obtain the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed; When the voiceprint representation type is characterized as an abnormal representation type, generating an emergency strategy determination request; When an emergency strategy determination request including a locomotive bogie analysis result is obtained, a strategy search is performed on the multi-dimensional semantic features corresponding to each preset emergency strategy according to the semantic analysis features of the locomotive bogie analysis result to obtain an emergency strategy determination result; The basic feature type identification and the refined feature type identification are performed on the bogie voiceprint data to be analyzed to obtain the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed, including: Acquire the bogie voiceprint data to be analyzed, perform a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed, and obtain a voiceprint feature vector; Extracting basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and identifying basic feature types based on the basic and advanced features to obtain voiceprint representation type confidence corresponding to each basic representation type; Extracting refined high-level features from the voiceprint feature vector to obtain refined high-level features, and merging the basic high-level features and the refined high-level features to obtain voiceprint merged features; Using a preset incremental parameter to perform feature conversion on the voiceprint merged feature to obtain a conversion feature, and calculating the feature sum of the conversion feature and the refined high-level feature to obtain an incremental feature, and performing feature normalization processing on the incremental feature to obtain a refined voiceprint integrated feature, and using the refined voiceprint integrated feature to perform refined feature type identification to obtain a voiceprint representation type confidence corresponding to each refined representation type, wherein the classification granularity corresponding to the refined representation type is greater than the classification granularity corresponding to the basic representation type; Obtain the granularity information of the voiceprint representation type; Determining the voiceprint representation type confidence level corresponding to the current basic representation type from the voiceprint representation type confidence levels corresponding to each basic representation type; Acquire a current refined characterization type corresponding to the current basic characterization type according to the voiceprint characterization type granularity information, and determine a voiceprint characterization type confidence corresponding to the current refined characterization type from the voiceprint characterization type confidences corresponding to each refined characterization type; Calculating a multiplication result of the voiceprint representation type confidence corresponding to the current basic representation type and the voiceprint representation type confidence corresponding to the current refined representation type to obtain a current granularity index confidence; Polling the voiceprint representation type confidence corresponding to each basic representation type and the voiceprint representation type confidence corresponding to each refined representation type to obtain each granularity index confidence; The target granularity index confidence is determined from each granularity index confidence, and the voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is determined according to the target granularity index confidence, wherein the voiceprint representation type includes a target basic representation type and a target refined representation type corresponding to the target basic representation type.

2. The method according to claim 1, characterized in that The method further comprises: Loading the bogie voiceprint data to be analyzed into a first target bogie voiceprint recognition model; Using the first target bogie voiceprint recognition model, performing a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed to obtain a voiceprint feature vector; Using the first target bogie voiceprint recognition model, extracting basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and performing basic feature type recognition based on the basic and advanced features to obtain a voiceprint representation type confidence corresponding to each basic representation type; The first target bogie voiceprint recognition model is used to perform refined high-level feature extraction on the voiceprint feature vector to obtain refined high-level features, voiceprint feature integration is performed based on the basic high-level features and the refined high-level features to obtain refined voiceprint integrated features, and the refined voiceprint integrated features are used to perform refined feature type identification to obtain the voiceprint representation type confidence corresponding to each refined representation type.

3. The method according to claim 2, characterized in that The first target bogie voiceprint recognition model includes a voiceprint feature extraction component, a basic feature type recognition component and a refined feature type recognition component; the refined feature type recognition component includes a refined high-level feature extraction subcomponent, a voiceprint feature integration subcomponent and a refined feature type recognition subcomponent; The step of loading the bogie voiceprint data to be analyzed into a first target bogie voiceprint recognition model comprises: Loading the bogie voiceprint data to be analyzed into the voiceprint feature extraction component to perform a voiceprint feature extraction operation to obtain a voiceprint feature vector; Loading the voiceprint feature vector into the basic feature type recognition component, using the basic feature type recognition component to extract basic high-level features from the voiceprint feature vector to obtain basic high-level features, and performing basic representation type confidence calculation based on the basic high-level features to obtain the voiceprint representation type confidence corresponding to each basic representation type; Loading the voiceprint feature vector into the refined high-level feature extraction subcomponent to extract refined high-level features to obtain the refined high-level features; Loading the basic high-level features and the refined high-level features into the voiceprint feature integration subcomponent to integrate the voiceprint features, and obtaining refined voiceprint integration features; The refined voiceprint integrated features are loaded into the refined feature type identification subcomponent to perform refined characterization type confidence calculation to obtain the voiceprint characterization type confidence corresponding to each refined characterization type.

4. The method according to claim 1, characterized in that: The method further comprises: Extracting intermediate-granularity advanced features from the voiceprint feature vector to obtain intermediate-granularity advanced features; Merging the basic high-level features and the intermediate-level high-level features to obtain intermediate voiceprint merged features; Performing incremental feature extraction based on the intermediate voiceprint merged features and the refined high-level features to obtain intermediate incremental features, performing feature normalization processing on the intermediate incremental features to obtain intermediate integrated features, performing intermediate granularity type recognition based on the intermediate integrated features to obtain the voiceprint representation type confidence corresponding to each intermediate representation type; Merging the basic high-level features, the intermediate integrated features, and the refined high-level features to obtain a target voiceprint merged feature; Performing incremental feature extraction according to the target voiceprint combined feature and the refined high-level feature to obtain the target incremental feature, performing feature normalization processing on the target incremental feature to obtain the target refined voiceprint integrated feature, and using the target refined voiceprint integrated feature to perform refined feature type identification to obtain the target voiceprint representation type confidence corresponding to each refined representation type, wherein the classification granularity corresponding to the refined representation type is greater than the classification granularity corresponding to the intermediate representation type, and the classification granularity corresponding to the intermediate representation type is greater than the classification granularity corresponding to the basic representation type; The target voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is obtained according to the voiceprint representation type confidence corresponding to each basic representation type, the voiceprint representation type confidence corresponding to each intermediate representation type and the target voiceprint representation type confidence corresponding to each refined representation type. The target voiceprint representation type includes a target basic representation type, a target intermediate representation type corresponding to the target basic representation type and a target refined representation type corresponding to the target intermediate representation type.

5. The method according to claim 4, characterized in that The method further comprises: Loading the bogie voiceprint data to be analyzed into a second target bogie voiceprint recognition model; Using the second target bogie voiceprint recognition model, performing a voiceprint feature extraction operation on the bogie voiceprint data to be analyzed to obtain a voiceprint feature vector; Using the second target bogie voiceprint recognition model to extract basic and advanced features from the voiceprint feature vector to obtain basic and advanced features, and performing basic feature type identification based on the basic and advanced features to obtain a voiceprint representation type confidence corresponding to each basic representation type; Using the second target bogie voiceprint recognition model, extracting intermediate-granularity advanced features from the voiceprint feature vector to obtain intermediate-granularity advanced features, integrating voiceprint features according to the basic advanced features and the intermediate-granularity advanced features to obtain intermediate-level integrated features, and performing intermediate-granularity type identification according to the intermediate-level integrated features to obtain voiceprint representation type confidence corresponding to each intermediate representation type; Using the second target bogie voiceprint recognition model to perform voiceprint feature integration on the basic high-level features, the intermediate integrated features and the refined high-level features to obtain target refined voiceprint integrated features, and using the target refined voiceprint integrated features to perform refined feature type identification to obtain target voiceprint representation type confidence corresponding to each refined representation type; The target voiceprint representation type corresponding to the bogie voiceprint data to be analyzed is obtained according to the voiceprint representation type confidence corresponding to each basic representation type, the voiceprint representation type confidence corresponding to each intermediate representation type and the target voiceprint representation type confidence corresponding to each refined representation type. The target voiceprint representation type includes a target basic representation type, a target intermediate representation type corresponding to the target basic representation type and a target refined representation type corresponding to the target intermediate representation type.

6. The method according to claim 1, characterized in that When an emergency strategy determination request including a locomotive bogie analysis result is obtained, a strategy search is performed on the multi-dimensional semantic features corresponding to each preset emergency strategy according to the semantic analysis features of the locomotive bogie analysis result to obtain an emergency strategy determination result, including: When an emergency strategy determination request including a locomotive bogie analysis result is obtained, semantic feature extraction is performed on the locomotive bogie analysis result to obtain a semantic analysis feature; According to the multi-dimensional semantic features corresponding to each preset emergency strategy and in combination with the semantic analysis features, a strategy search is performed on the locomotive bogie analysis result and each preset emergency strategy to obtain an emergency strategy determination result.

7. The method according to claim 6, characterized in that The step of performing strategy retrieval on the locomotive bogie analysis result and each preset emergency strategy according to the multi-dimensional semantic features corresponding to each preset emergency strategy and combining the semantic analysis features to obtain the emergency strategy determination result includes: According to a preset feature extraction method, a target feature matching coefficient is extracted from feature matching coefficients of each of a plurality of semantic features in a multi-dimensional semantic feature of a preset emergency strategy and the semantic analysis feature; Obtain the emergency strategy determination result according to the target feature matching coefficient corresponding to each preset emergency strategy; or, According to the reinforcement feature corresponding to each preset emergency strategy and the semantic analysis feature, the reinforcement feature matching coefficient corresponding to each preset emergency strategy is obtained, and according to the multi-dimensional semantic feature corresponding to each preset emergency strategy and the semantic analysis feature, the multi-dimensional semantic feature matching coefficient corresponding to each preset emergency strategy is obtained; The enhanced feature matching coefficient and the multi-dimensional semantic feature matching coefficient corresponding to each preset emergency strategy are respectively integrated to obtain the integrated matching coefficient corresponding to each preset emergency strategy, and the emergency strategy determination result is obtained according to each integrated matching coefficient obtained; wherein each enhanced feature is obtained according to the multi-dimensional semantic feature of each associated preset emergency strategy associated with the corresponding preset emergency strategy, and the enhanced feature of a preset emergency strategy is obtained in the following manner: obtaining the multi-dimensional mean feature corresponding to each preset emergency strategy; wherein the multi-dimensional mean feature of each other preset emergency strategy is obtained by averaging multiple semantic features in the corresponding multi-dimensional semantic feature; according to the multi-dimensional mean feature corresponding to each other preset emergency strategy except the one preset emergency strategy in each preset emergency strategy, the enhanced feature of each preset emergency strategy is obtained. the degree of semantic description matching between his preset emergency strategies and the preset emergency strategies respectively; according to each obtained semantic description matching degree, extract other preset emergency strategies that meet the preset strategy association criteria from each other preset emergency strategy as the associated preset emergency strategies of the preset emergency strategy; the degree of semantic description matching between each associated preset emergency strategy and the preset emergency strategy is used as the influencing factor corresponding to each associated preset emergency strategy; wherein each semantic description matching degree is determined according to the multi-dimensional semantic features of the corresponding associated preset emergency strategy and the preset emergency strategy; according to the influencing factor corresponding to each associated preset emergency strategy, linearly superimpose the multi-dimensional mean features corresponding to each associated preset emergency strategy to obtain the enhanced features of the preset emergency strategy.

8. The method according to claim 6, characterized in that The step of performing strategy retrieval on the locomotive bogie analysis result and each preset emergency strategy according to the multi-dimensional semantic features corresponding to each preset emergency strategy and combining the semantic analysis features to obtain the emergency strategy determination result includes: According to the multi-dimensional semantic features corresponding to each preset emergency strategy, combined with the semantic analysis features, the overall feature matching coefficient corresponding to each preset emergency strategy is obtained; According to the preset policy analysis criteria, a preset emergency policy is segmented to obtain each policy element, and semantic features are extracted for each policy element to obtain policy element features corresponding to each policy element; Obtaining an element feature matching coefficient corresponding to the preset emergency strategy according to a strategy element matching coefficient between each obtained strategy element feature and the semantic analysis feature; According to the overall feature matching coefficient and the element feature matching coefficient corresponding to each preset emergency strategy, a strategy search is performed on the locomotive bogie analysis result and each preset emergency strategy to obtain an emergency strategy determination result.

9. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 8.

Images