Sound system fault analysis method and system based on machine learning

By employing a machine learning-based fault analysis method for audio systems, utilizing LSTM and temporal attention vector mining techniques, the accuracy and efficiency issues of traditional audio system fault detection are addressed. This method enables precise identification and efficient handling of potential faults, thereby improving system stability and operational efficiency.

CN118643450BActive Publication Date: 2024-11-19SHENZHEN BBS ELECTRONICS CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411113076.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-11-19
Estimated Expiration
2044-08-14

AI Technical Summary

Technical Problem

Traditional methods for detecting audio system faults rely on manual inspections and experience-based judgment, which makes it difficult to efficiently and accurately identify potential faults in various abnormal modes, leading to difficulties in fault detection and localization.

Method used

A machine learning-based fault analysis method is adopted. By acquiring the operating status log of the audio system, the initial long short-term memory algorithm (LSTM) and temporal attention vector mining are used to identify and predict potential fault modes. Combined with knowledge embedding, mapping and distribution vector extraction modules, in-depth analysis and accurate location of fault status are achieved.

Benefits of technology

It improves the accuracy and efficiency of fault detection, reduces the complexity and cost of operation and maintenance, and ensures the stable operation and fault prevention of the audio system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118643450B_ABST
    Figure CN118643450B_ABST
Patent Text Reader

Abstract

The present application discloses a method and system for analyzing faults of an audio system based on machine learning, and belongs to the field of data analysis technology. The fault analysis system of the present application realizes accurate identification and efficient processing of potential faults by innovatively acquiring and analyzing the operating status log of the audio system. It is capable of processing operating status data containing different abnormal mode keywords, and using advanced fault analysis algorithms to identify specific fault modes from the first operating status detection data, and further associate and analyze the potential faults that may exist in the second operating status detection data, and output them in the form of target abnormal mode keywords. In this way, the accuracy and efficiency of fault detection are significantly improved, the complexity and cost of audio system operation and maintenance are reduced, and strong technical support is provided for the stable operation and fault prevention of the audio system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of data analysis technology, and specifically relates to a method and system for analyzing faults of an audio system based on machine learning. Background Art

[0002] In the operation and maintenance of audio systems, timely detection and handling of potential faults is the key to ensuring stable operation of the system. However, traditional fault detection methods often rely on manual inspections and empirical judgments, which make it difficult to efficiently and accurately identify all potential fault hazards. In particular, when the audio system has different abnormal modes of failure, the fault manifestations vary, making it more difficult to detect and locate the fault. Summary of the invention

[0003] The present application provides a method and system for analyzing faults in an audio system based on machine learning, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The embodiment of the present application provides a method for analyzing faults of an audio system based on machine learning, which is applied to a fault analysis system. The method includes:

[0005] The fault analysis system obtains a sound system operation status log to be analyzed, wherein the sound system operation status log to be analyzed includes first operation status detection data and second operation status detection data; the first operation status detection data and the second operation status detection data correspond to fault hazards of different abnormal modes; the first operation status detection data includes a first fault status event with a keyword for setting an abnormal mode added thereto;

[0006] The fault analysis system inputs the audio system operation status log to be analyzed into a fault analysis algorithm, and processes the first fault status event with the set abnormal mode keyword added in the first operation status detection data according to the fault analysis algorithm, to obtain the target abnormal mode keyword of the second fault status event in the second operation status detection data.

[0007] In one implementation, the debugging method of the fault analysis algorithm includes:

[0008] Acquire a sound system operation status log sample, wherein the sound system operation status log sample includes a first operation status detection data sample and a second operation status detection data sample; the first operation status detection data sample and the second operation status detection data sample correspond to fault hidden danger samples of different abnormal modes; the first operation status detection data sample includes a first fault status event sample with a first abnormal mode keyword added, and the second operation status detection data sample includes a second fault status event sample with a second abnormal mode keyword added;

[0009] Inputting the sound system operation status log sample into an initial long short-term memory algorithm, performing state vector mining on the first fault state event sample to which the first abnormal mode keyword is added according to the state vector mining branch in the initial long short-term memory algorithm, to obtain a fault state vector sample;

[0010] According to the temporal attention vector mining branch in the initial long short-term memory algorithm, temporal attention vector mining is performed on the fault state vector sample to obtain a temporal attention vector sample;

[0011] According to the abnormal pattern keyword discrimination branch in the initial long short-term memory algorithm, the temporal attention vector sample is subjected to abnormal pattern keyword discrimination to obtain an abnormal pattern sample discrimination viewpoint corresponding to the second fault state event sample in the second operating state detection data sample;

[0012] Based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

[0013] In one implementation, the temporal attention vector mining branch includes a knowledge embedding module, a knowledge mapping module, and a distribution vector extraction module. The temporal attention vector mining branch in the initial long short-term memory algorithm performs temporal attention vector mining on the fault state vector sample to obtain a temporal attention vector sample, including:

[0014] Performing a knowledge embedding operation on the fault state vector sample according to the knowledge embedding module to obtain a linear knowledge vector sample;

[0015] Performing a distribution vector extraction operation on the fault state vector sample according to the distribution vector extraction module to obtain a distribution vector sample;

[0016] The linear knowledge vector sample and the distribution vector sample are subjected to knowledge mapping operation according to the knowledge mapping module to obtain the temporal attention vector sample.

[0017] In one implementation, the state vector mining branch includes a residual connection subnet, a quantization processing subnet, and a state parsing subnet. The state vector mining branch in the initial long short-term memory algorithm performs state vector mining on the first fault state event sample to which the first abnormal mode keyword is added to obtain a fault state vector sample, including:

[0018] performing, according to the residual connection subnetwork, mining of an artifact element vector on the first fault state event sample to which the first abnormal mode keyword is added, to obtain an artifact element vector sample;

[0019] Performing a quantized heat map transformation on the artifact element vector sample according to the quantization processing subnet to obtain a quantized heat map vector sample;

[0020] A fault state analysis operation is performed on the quantized heat map vector sample according to the state analysis subnet to obtain the fault state vector sample.

[0021] In one implementation, the first operation status detection data sample further includes a first distribution feature element of the first fault status event sample, the second operation status detection data sample further includes a second distribution feature element of the second fault status event sample, the initial long short-term memory algorithm further includes a distribution feature identification subnet, and the method further includes:

[0022] Performing state vector mining on a first distribution feature element in the first running state detection data sample according to the state vector mining branch to obtain a first fault state vector;

[0023] According to the temporal attention vector mining branch, performing temporal attention vector mining on the first fault state vector to obtain a first temporal attention vector;

[0024] According to the distribution feature recognition subnet, the first temporal attention vector is subjected to distribution feature recognition to obtain a distribution feature recognition result sample corresponding to the second fault state event sample in the second operating state detection data sample;

[0025] The method of determining the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, debugging the initial long short-term memory algorithm, and obtaining a fault analysis algorithm includes:

[0026] Based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

[0027] In one implementation, the initial long short-term memory algorithm is debugged based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element to obtain a fault analysis algorithm, including:

[0028] Determining a first algorithm training error based on the difference between the abnormal pattern sample discrimination viewpoint and the second abnormal pattern keyword;

[0029] Determining a second algorithm training error based on a difference between the distribution feature recognition result sample and the second distribution feature element;

[0030] Determining a target algorithm training error based on the first algorithm training error and the second algorithm training error;

[0031] Based on the target algorithm training error, optimizing the algorithm variables of the initial long short-term memory algorithm until the debugging compliance requirements are met;

[0032] The neural network algorithm composed of the state vector mining branch, the temporal attention vector mining branch and the abnormal pattern keyword discrimination branch during the completion of debugging is determined as the fault analysis algorithm.

[0033] In one implementation, there are multiple sound system operation status log samples, the initial long short-term memory algorithm further includes an asynchronous identification branch, and the method further includes:

[0034] According to the state vector mining branch, each running state detection data sample in the running state detection data sample tuple is subjected to state vector mining to obtain a second fault state vector; the running state detection data sample tuple is determined according to the multiple sound system running state log samples;

[0035] According to the temporal attention vector mining branch, performing temporal attention vector mining on the second fault state vector to obtain a second temporal attention vector;

[0036] According to the asynchronous recognition branch, asynchronously recognize the second temporal attention vector corresponding to the running status detection data sample binary to obtain an asynchronous recognition result;

[0037] Determining a third algorithm training error based on a difference between an asynchronous recognition result corresponding to the running status detection data sample binary and a priori asynchronous authentication annotation;

[0038] The determining a target algorithm training error based on the first algorithm training error and the second algorithm training error includes:

[0039] A target algorithm training error is determined based on the first algorithm training error, the second algorithm training error, and the third algorithm training error.

[0040] In one implementation, the method further includes:

[0041] Generate a positive running state detection data sample tuple according to the running state detection data sample corresponding to the same sound system running state log sample; the positive running state detection data sample tuple is added with a priori asynchronous authentication annotations of the same batch;

[0042] Generate a negative running state detection data sample tuple according to the running state detection data samples corresponding to the running state log samples of different sound systems; the negative running state detection data sample tuple is added with a priori asynchronous authentication annotations of non-same batches;

[0043] The running state detection data sample tuple is generated according to the positive running state detection data sample tuple and the negative running state detection data sample tuple.

[0044] In one implementation, the method further includes:

[0045] The neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch and the distribution feature recognition subnet during debugging is determined as a distribution feature recognition algorithm;

[0046] The neural network algorithm consisting of the state vector mining branch, the temporal attention vector mining branch and the asynchronous recognition branch during debugging is determined as an asynchronous fault analysis algorithm.

[0047] An embodiment of the present application provides a fault analysis system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0048] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0049] The fault analysis system of the present application achieves accurate identification and efficient processing of potential faults by innovatively acquiring and analyzing the operating status log of the audio system. It can process operating status data containing different abnormal mode keywords, and use advanced fault analysis algorithms to identify specific fault modes from the first operating status detection data, and further associate and analyze the potential faults that may exist in the second operating status detection data, and output them in the form of target abnormal mode keywords. In this way, the accuracy and efficiency of fault detection are significantly improved, the complexity and cost of audio system operation and maintenance are reduced, and strong technical support is provided for the stable operation and fault prevention of the audio system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for analyzing sound system faults based on machine learning provided in an embodiment of the present application.

[0051] Figure 2 A schematic diagram of the structure of a fault analysis system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in this application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0054] Figure 1 A method for analyzing faults of an audio system based on machine learning is shown, which is applied to a fault analysis system. The method includes the following steps 110 to 120.

[0055] Step 110: Acquire an audio system operation status log to be analyzed, wherein the audio system operation status log to be analyzed includes first operation status detection data and second operation status detection data; the first operation status detection data and the second operation status detection data correspond to fault hazards of different abnormal modes; the first operation status detection data includes a first fault status event to which a keyword for setting an abnormal mode is added.

[0056] Step 120: input the audio system operation status log to be analyzed into a fault analysis algorithm; and according to the fault analysis algorithm, process the first fault status event to which the set abnormal mode keyword is added in the first operation status detection data to obtain the target abnormal mode keyword of the second fault status event in the second operation status detection data.

[0057] In the operation and maintenance of the audio system, fault analysis is a crucial task. In order to ensure the stable operation of the audio system and to promptly discover and resolve potential faults, it is particularly necessary to adopt advanced fault analysis technology. The following will use a specific application scenario example to detail the implementation process of the audio system fault analysis technical solution involved in the above steps 110 to 120.

[0058] The sound system of a large concert venue is the core equipment of the performance. Its stability and reliability have a vital impact on the performance effect. In order to ensure the smooth progress of each performance, the venue introduced an advanced fault analysis system to monitor and analyze the operating status of the sound system in real time, and promptly discover and solve potential faults.

[0059] 1. The fault analysis system first obtains the operating status log of the audio system to be analyzed. These logs are various data records generated by the audio system during operation, including system status, device parameters, operating time, etc. In this application scenario, the operating status log of the audio system mainly includes the first operating status detection data and the second operating status detection data.

[0060] The first operating state detection data is the data generated by the sound system during normal operation. These data reflect the current state and operating status of the system. In order to facilitate subsequent analysis, the first fault state event with abnormal mode keywords is specially added to these data. These keywords are pre-set based on historical fault data and expert experience to identify possible fault modes.

[0061] The second operating state detection data is the data generated when the sound system has an abnormality or potential fault. These data contain richer fault information and are the main basis for fault analysis. However, due to the diversity and complexity of fault modes, the fault information in the second operating state detection data often exists in an unstructured form and needs to be further processed by a fault analysis algorithm.

[0062] Second, after obtaining the operating status logs of the sound system, the fault analysis system inputs these logs into the pre-built fault analysis algorithm. This algorithm is based on machine learning and data mining technology, and can automatically analyze and process a large amount of operating status data to extract useful fault information.

[0063] When processing the first operating state detection data, the fault analysis algorithm pays special attention to the first fault state events that have added the keywords of the set abnormal mode. These events are pre-identified patterns of possible faults and are the focus of fault analysis. The algorithm attempts to find the hidden fault characteristics and laws by performing deep learning and pattern matching on these events.

[0064] After a series of complex calculations and analyses, the fault analysis algorithm can eventually extract the target abnormal mode keywords of the second fault state event corresponding to the set abnormal mode keywords from the first operation state detection data. These target abnormal mode keywords are the structured representation of the fault information in the second operation state detection data, and are an important basis for subsequent fault location and repair. In detail, steps 110 to 120 also involve the following technical contents.

[0065] (1) Data preprocessing: Before inputting the operating status log of the sound system into the fault analysis algorithm, data preprocessing is required. The purpose of this step is to clean the fault status and redundant information in the data and improve the quality and availability of the data. Specific operations include removing invalid records, filling missing values, and smoothing data fluctuations.

[0066] (2) Feature extraction: After preprocessing, feature extraction is performed next. The purpose of this step is to extract feature information that is useful for fault analysis from the cleaned data. In this application scenario, feature extraction mainly focuses on the key parameters and status indicators of the audio system, such as volume, sound quality, channel balance, etc. By extracting this feature information, it can provide strong data support for subsequent fault analysis.

[0067] (3) Fault pattern recognition: Based on feature extraction, the fault analysis algorithm begins to perform fault pattern recognition. The purpose of this step is to identify possible fault patterns in the audio system based on the extracted feature information. The algorithm uses deep learning and pattern matching technology to classify and cluster the feature information, trying to find the hidden fault characteristics and patterns.

[0068] (4) Keyword matching and extraction: After completing the fault pattern recognition, the algorithm further performs keyword matching and extraction. The purpose of this step is to match the identified fault pattern with the pre-set abnormal pattern keywords to find the target abnormal pattern keywords corresponding to the second fault state event. Through keyword matching and extraction, the algorithm can convert unstructured fault information into a structured representation, which is convenient for subsequent fault location and repair.

[0069] (5) Fault location and repair suggestions: Finally, based on the extracted target abnormal pattern keywords, the fault analysis system can give specific fault location and repair suggestions. The system will locate the specific parts or components in the audio system where the fault may occur according to the instructions of the keywords, and provide corresponding repair suggestions. These suggestions include replacing damaged parts, adjusting system parameters, optimizing the operating environment, etc. By following these suggestions, operation and maintenance personnel can quickly and accurately solve potential fault hazards of the audio system.

[0070] It is worth mentioning that in the sound system of a large concert venue, it is crucial to ensure its stable operation and timely detection of potential faults. For step 110, first, the fault analysis system will obtain the sound system operation status log to be analyzed. These logs are valuable data records generated during the operation of the sound system, which reflect the current state and operation status of the system in detail. In order to more comprehensively grasp the operation status of the system, the fault analysis system specifically divides these logs into two categories: first operation status detection data and second operation status detection data. Among them, the first operation status detection data is the data generated during the normal operation of the sound system. These data not only record the daily operation status of the system, but also specially add the first fault state event with the abnormal mode keyword. These keywords are pre-set based on historical fault data and expert experience. They are like indicator lights. Once lit, it means that a certain specific fault mode may occur in the system. The second operation status detection data is the data generated when the sound system has an abnormality or potential fault. These data contain richer fault information and are the main basis for fault analysis. However, unlike the first operation status detection data, the fault information in the second operation status detection data often exists in an unstructured form and needs to be further processed by subsequent fault analysis algorithms. By acquiring these two types of data at the same time, the fault analysis system can more comprehensively grasp the operating status of the sound system and promptly discover and deal with potential fault hazards. This technical solution not only improves the accuracy of fault detection, but also provides strong data support for subsequent fault location and repair. In the sound system of large concert venues, the implementation of this technical solution will undoubtedly bring a more stable and high-quality sound experience to the audience.

[0071] Furthermore, as the core invention of the present application, the implementation of the technical solution of step 120 is of vital importance for timely discovery and handling of potential faults. Specifically, this step is to input the operating status log of the audio system to be analyzed into the constructed fault analysis algorithm to achieve in-depth analysis and accurate positioning of the potential faults of the system. First, the fault analysis system will input all the previously acquired operating status logs of the audio system to be analyzed, including the first operating status detection data and the second operating status detection data, into the fault analysis algorithm. This algorithm is constructed based on advanced machine learning and data mining technology and has powerful data processing and analysis capabilities. Next, the algorithm will pay special attention to the first fault state events in the first operating state detection data that have added keywords for setting abnormal modes. These keywords are pre-set based on historical fault data and expert experience, and they can indicate specific fault modes that may occur in the system. The algorithm will perform deep learning and pattern matching on these events, and try to find hidden fault features and laws from them. Then, through a series of complex calculations and analyses, the fault analysis algorithm can gradually reveal the intrinsic connection between the first fault state event and the potential fault. It will not only identify the fault modes corresponding to these events, but will also further analyze the subsequent effects that these fault modes may cause, as well as their possible manifestations in the second operating state detection data. Ultimately, the algorithm will output the target abnormal pattern keywords of the second fault state event in the second operating state detection data. These keywords are structured representations of potential faults in the second fault state event, and they can clearly indicate the type, location, and possible causes of the fault. Operation and maintenance personnel can quickly and accurately locate and resolve potential fault hazards based on these keywords, thereby ensuring the stable operation and high-quality output of the sound system. It can be seen that the technical solution recorded in step 120 achieves in-depth analysis and accurate positioning of potential faults in the sound system through advanced fault analysis algorithms. The implementation of this solution provides strong technical support for the sound system of large concert venues, ensuring that the audience can enjoy a stable, high-quality sound experience.

[0072] The fault analysis system of the present application achieves accurate identification and efficient processing of potential faults by innovatively acquiring and analyzing the operating status log of the audio system. It can process operating status data containing different abnormal mode keywords, and use advanced fault analysis algorithms to identify specific fault modes from the first operating status detection data, and further associate and analyze the potential faults that may exist in the second operating status detection data, and output them in the form of target abnormal mode keywords. In this way, the accuracy and efficiency of fault detection are significantly improved, the complexity and cost of audio system operation and maintenance are reduced, and strong technical support is provided for the stable operation and fault prevention of the audio system.

[0073] In some preferred embodiments, the debugging method of the fault analysis algorithm includes: obtaining a sample of the operating status log of the audio system, the sample of the operating status log of the audio system includes a first operating status detection data sample and a second operating status detection data sample; the first operating status detection data sample and the second operating status detection data sample correspond to fault hidden danger samples of different abnormal modes; the first operating status detection data sample includes a first fault status event sample with a first abnormal mode keyword added, and the second operating status detection data sample includes a second fault status event sample with a second abnormal mode keyword added; inputting the sample of the operating status log of the audio system into an initial long short-term memory algorithm, and mining the state vector in the initial long short-term memory algorithm according to the state vector in the initial long short-term memory algorithm. The branch performs state vector mining on the first fault state event sample to which the first abnormal pattern keyword is added to obtain a fault state vector sample; according to the temporal attention vector mining branch in the initial long short-term memory algorithm, the fault state vector sample is mined in a temporal attention vector to obtain a temporal attention vector sample; according to the abnormal pattern keyword discrimination branch in the initial long short-term memory algorithm, the temporal attention vector sample is discriminated by abnormal pattern keywords to obtain an abnormal pattern sample discrimination viewpoint corresponding to the second fault state event sample in the second operating state detection data sample; based on the difference between the abnormal pattern sample discrimination viewpoint and the second abnormal pattern keyword, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

[0074] It can be understood that this embodiment involves a debugging method for a fault analysis algorithm, which aims to improve the algorithm's ability to identify and predict potential faults of an audio system through refined algorithm training and debugging.

[0075] First, the fault analysis system needs to obtain the operating status log samples of the sound system. These log samples are a valuable data source for algorithm training and debugging. They truly record the performance of the sound system under various operating conditions, especially the details under fault conditions. These log samples are divided into two categories: first operating status detection data samples and second operating status detection data samples. Each type of data sample corresponds to fault hidden danger samples with different abnormal modes, which provides the algorithm with a rich and diverse learning sample, helping the algorithm to better understand and identify various fault modes.

[0076] In the first operation state detection data sample, the fault analysis system specifically added the first abnormal mode keyword to mark the first fault state event sample. These keywords are based on the in-depth analysis of historical fault data and the summary of expert experience, and they can accurately indicate the specific fault modes that may occur in the audio system. Similarly, in the second operation state detection data sample, the fault analysis system also added the second abnormal mode keyword to mark the second fault state event sample. These keywords not only provide clear training goals for the algorithm, but also provide important reference for subsequent algorithm debugging and verification.

[0077] Next, the fault analysis system inputs these sound system operation status log samples into the initial long short-term memory algorithm (LSTM). LSTM is an advanced recurrent neural network algorithm that is particularly suitable for processing time series data and can capture long-term dependencies in the data. In the technical solution of the fault analysis system, the LSTM algorithm is cleverly designed into three branches: state vector mining branch, time series attention vector mining branch, and abnormal pattern keyword identification branch.

[0078] First, the state vector mining branch will perform state vector mining on the first fault state event sample with the first abnormal mode keyword added. The purpose of this step is to extract state vector samples that can reflect the fault characteristics from the original operating state data. These state vector samples are the basis for subsequent analysis, and they contain key information when the fault occurs, such as fault type, degree, and occurrence time.

[0079] Then, the temporal attention vector mining branch will further mine the temporal attention vectors of these fault state vector samples. The purpose of this step is to capture the law of fault state changes over time, so as to more accurately predict the development trend of faults. Through the mining of temporal attention vectors, the algorithm can pay more attention to the temporal features that have an important impact on fault prediction, and improve the accuracy and reliability of prediction.

[0080] Finally, the abnormal pattern keyword discrimination branch will perform abnormal pattern keyword discrimination on the temporal attention vector samples. The purpose of this step is to convert the learning results of the algorithm into specific fault prediction results, that is, to discriminate the abnormal pattern sample discrimination viewpoints corresponding to the second fault state event sample in the second operating state detection data sample. These discrimination viewpoints are the algorithm's direct prediction and judgment of potential faults, and they provide important reference for subsequent fault handling and operation and maintenance decisions.

[0081] After obtaining the judgment opinions of the abnormal pattern samples, the fault analysis system needs to debug the initial LSTM algorithm based on the difference between these opinions and the keywords of the second abnormal pattern. This step is a key link in algorithm training and debugging. It continuously optimizes the performance and accuracy of the algorithm through feedback and adjustment of the algorithm prediction results. Specifically, the fault analysis system will analyze the differences and deviations between the algorithm judgment opinions and the actual situation, and find out the problems and shortcomings of the algorithm. Then, the fault analysis system will adjust and optimize the algorithm accordingly for these problems and shortcomings, such as adjusting algorithm parameters, improving algorithm structure, and increasing training samples. Through continuous debugging and optimization, the fault analysis system can eventually obtain a fault analysis algorithm with stable performance and high accuracy.

[0082] In this way, by obtaining the sound system operation status log sample, designing a clever long short-term memory algorithm structure, and conducting refined algorithm training and debugging, an efficient fault analysis algorithm was successfully implemented. The algorithm can accurately identify potential faults in the sound system and predict the development trend and possible impact range of the fault. This not only improves the operation and maintenance efficiency and reliability of the sound system, but also provides a new idea and method for fault analysis of similar systems. In future research, the fault analysis system can further explore and optimize this technical solution, such as introducing more machine learning algorithms, combining expert systems and knowledge graphs, etc., to further improve the accuracy and efficiency of fault analysis.

[0083] In some other possible embodiments, the temporal attention vector mining branch includes a knowledge embedding module, a knowledge mapping module and a distribution vector extraction module, and the temporal attention vector mining branch in the initial long short-term memory algorithm performs temporal attention vector mining on the fault state vector sample to obtain a temporal attention vector sample, including: performing a knowledge embedding operation on the fault state vector sample according to the knowledge embedding module to obtain a linear knowledge vector sample; performing a distribution vector extraction operation on the fault state vector sample according to the distribution vector extraction module to obtain a distribution vector sample; performing a knowledge mapping operation on the linear knowledge vector sample and the distribution vector sample according to the knowledge mapping module to obtain the temporal attention vector sample.

[0084] In actual application, the specific refinement and implementation of the temporal attention vector mining branch is also one of the core invention points. The temporal attention vector mining branch plays a vital role in the fault analysis algorithm. It is responsible for mining the temporal attention vector samples from the fault state vector samples, providing key information and basis for the subsequent abnormal pattern keyword judgment.

[0085] First, the temporal attention vector mining branch includes three important modules: knowledge embedding module, knowledge mapping module and distribution vector extraction module. These three modules each undertake different tasks, but they work together to complete the mining of temporal attention vectors.

[0086] In the specific implementation, the fault analysis system first performs knowledge embedding operations on the fault state vector sample based on the knowledge embedding module. The purpose of this step is to transform the original fault state vector sample into a more abstract and easier to handle linear knowledge vector sample. The knowledge embedding module extracts the key information from the fault state vector sample through a series of complex mathematical transformations and mappings, and expresses it in a linear vector form. This linear knowledge vector sample not only retains the key features of the original data, but also facilitates subsequent processing and calculation.

[0087] Next, the fault analysis system performs a distribution vector extraction operation on the fault state vector sample based on the distribution vector extraction module. The purpose of this step is to extract a distribution vector sample that can reflect the data distribution characteristics from the fault state vector sample. The distribution vector extraction module constructs a vector that can describe the data distribution by calculating the statistical characteristics of the fault state vector sample in various dimensions, such as mean, variance, covariance, etc. This distribution vector sample not only contains the overall distribution of the data, but also reflects the correlation and difference of the data in various dimensions, providing important reference information for subsequent time series attention vector mining.

[0088] Finally, the fault analysis system performs knowledge mapping operations on the linear knowledge vector samples and the distribution vector samples based on the knowledge mapping module. The purpose of this step is to fuse and map the two vector samples obtained above to obtain the final temporal attention vector sample. The knowledge mapping module extracts the key information from the linear knowledge vector sample and the distribution vector sample through a complex mapping relationship and fuses them into a new vector form. This temporal attention vector sample not only contains the key feature information in the fault state vector sample, but also reflects the temporal distribution and correlation characteristics of the data, providing strong support and basis for the subsequent abnormal pattern keyword discrimination. In summary, through the collaborative work of the knowledge embedding module, the knowledge mapping module and the distribution vector extraction module, the fault analysis system successfully realizes the specific refinement and implementation of the temporal attention vector mining branch. This branch can mine key temporal attention vector samples from the fault state vector sample, providing important information and basis for the subsequent abnormal pattern keyword discrimination and fault prediction. This not only improves the accuracy and efficiency of the fault analysis algorithm, but also provides a new idea and method for the temporal data mining of similar systems. In future research, the fault analysis system can further explore and optimize this technical solution, such as introducing more machine learning algorithms and deep learning models to improve the accuracy and efficiency of time series attention vector mining. At the same time, the fault analysis system can also try to apply this technical solution to time series data mining tasks in other fields to verify its universality and effectiveness.

[0089] In a preferred embodiment, the state vector mining branch includes a residual connection subnet, a quantization processing subnet and a state analysis subnet, and the state vector mining branch in the initial long short-term memory algorithm performs state vector mining on the first fault state event sample to which the first abnormal mode keyword is added to obtain a fault state vector sample, including: performing sound loss element vector mining on the first fault state event sample to which the first abnormal mode keyword is added according to the residual connection subnet to obtain a sound loss element vector sample; performing a quantization heat map transformation on the sound loss element vector sample according to the quantization processing subnet to obtain a quantization heat map vector sample; performing a fault state analysis operation on the quantization heat map vector sample according to the state analysis subnet to obtain the fault state vector sample.

[0090] It can be understood that this embodiment involves the specific refinement and implementation of the state vector mining branch, which plays a core role in the fault analysis algorithm and is responsible for mining key fault state vector samples from the first fault state event sample to which the first abnormal mode keyword is added.

[0091] First, the state vector mining branch carefully designs three subnets: residual connection subnet, quantization processing subnet, and state parsing subnet. These three subnets work together to complete the state vector mining task.

[0092] In the specific implementation, the fault analysis system first uses the residual connection subnet to mine the sound damage element vector of the first fault state event sample with the first abnormal mode keyword added. The goal of this step is to extract the key elements related to the sound system failure and express them in vector form, that is, the sound damage element vector sample. Through a series of complex calculations, the residual connection subnet can capture the slight changes in the fault state event sample, which are often the precursors of the failure. By mining these sound damage elements, the fault analysis system can more accurately identify the potential failure of the sound system.

[0093] Next, the quantization processing subnet performs a quantized heat map transformation on the sound artifact vector sample. The purpose of this step is to convert the continuous values ​​in the sound artifact vector sample into discrete heat map vectors, that is, quantized heat map vector samples. The quantization processing subnet uses a special transformation algorithm to map each value in the sound artifact vector sample to a specific color or grayscale level, thereby generating a heat map. This heat map can intuitively show the distribution of sound artifacts in different states, providing an important visualization basis for the subsequent state analysis of the fault analysis system.

[0094] Finally, the fault analysis system uses the state parsing subnet to perform fault state parsing operations on the quantized heat map vector samples. The goal of this step is to parse out the fault state information in the quantized heat map vector samples and express it in vector form, namely, the fault state vector samples. The state parsing subnet uses a complex parsing algorithm to identify specific patterns or features in the quantized heat map vector samples, which are closely related to the fault state of the audio system. By parsing these patterns or features, the fault analysis system can accurately determine the fault type, degree, and possible impact range of the audio system.

[0095] In order to more specifically illustrate the implementation of this technical solution, an example of a numerical feature vector can be given. For example, the fault analysis system has a first fault state event sample with a first abnormal mode keyword added, and the sample contains a set of numerical feature vectors, such as [0.5, 0.7, 0.9, 0.6, 0.8]. This set of numerical feature vectors reflects the operating status of the sound system at different time points.

[0096] First, the residual connection subnet mines the sound artifact vectors for this set of numerical feature vectors. Through a series of calculations, the residual connection subnet may obtain a set of new sound artifact vector samples, such as [0.1, 0.2, 0.3, 0.1, 0.2]. This set of sound artifact vector samples represents the changes in the sound artifacts during the operation of the sound system.

[0097] The quantization processing subnet then performs a quantization heat map transformation on this set of sound loss element vector samples. Through the transformation algorithm, the quantization processing subnet may obtain a heat map in which each value is mapped to a specific color or grayscale level. For example, the value 0.1 may be mapped to light gray, the value 0.2 may be mapped to dark gray, and the value 0.3 may be mapped to black. In this way, the fault analysis system can obtain a heat map that intuitively shows the distribution of sound loss elements.

[0098] Finally, the state parsing subnet performs fault state parsing on this quantized heat map vector sample. Through the parsing algorithm, the state parsing subnet may identify specific patterns or features in the heat map, such as the abnormally dark or light color of a certain area. These patterns or features are closely related to the fault state of the audio system. By parsing these patterns or features, the state parsing subnet may obtain a fault state vector sample, such as [1, 0, 0, 0, 1], indicating that there are two types of faults in the audio system.

[0099] In summary, through the collaborative work of the residual connection subnet, the quantization processing subnet, and the state parsing subnet, the fault analysis system successfully realized the specific refinement and implementation of the state vector mining branch. This branch can mine key fault state vector samples from the first fault state event sample with the first abnormal mode keyword added, providing important information and basis for subsequent abnormal mode keyword discrimination and fault prediction. This not only improves the accuracy and efficiency of the fault analysis algorithm, but also provides a new idea and method for the operation and management of the sound system. In future research, the fault analysis system can further explore and optimize this technical solution, such as introducing more machine learning algorithms and deep learning models to improve the accuracy and efficiency of state vector mining. At the same time, the fault analysis system can also try to apply this technical solution to state monitoring and fault diagnosis tasks in other fields to verify its versatility and effectiveness.

[0100] In some alternative embodiments, the first running state detection data sample also includes the first distribution feature element of the first fault state event sample, the second running state detection data sample also includes the second distribution feature element of the second fault state event sample, the initial long short-term memory algorithm also includes a distribution feature recognition subnet, and the method further includes: performing state vector mining on the first distribution feature element in the first running state detection data sample according to the state vector mining branch to obtain a first fault state vector; performing time series attention vector mining on the first fault state vector according to the time series attention vector mining branch to obtain a first time series attention vector; performing distribution feature recognition on the first time series attention vector according to the distribution feature recognition subnet to obtain a distribution feature recognition result sample corresponding to the second fault state event sample in the second running state detection data sample. Then, based on the difference between the judgment point of view of the abnormal mode sample and the second abnormal mode keyword, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm, including: based on the difference between the judgment point of view of the abnormal mode sample and the second abnormal mode keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

[0101] In this embodiment, the fault analysis system further enriches the content of the operation status detection data sample and correspondingly expands the function of the initial long short-term memory algorithm. Specifically, in addition to including the first fault state event sample, the first operation status detection data sample also covers the first distribution feature element of the sample; similarly, the second operation status detection data sample also includes the second fault state event sample and its corresponding second distribution feature element. In order to process these newly added elements, the fault analysis system introduces a distribution feature recognition subnet in the initial long short-term memory algorithm.

[0102] First, the fault analysis system uses the state vector mining branch to conduct in-depth state vector mining on the first distribution feature element in the first operating state detection data sample. The goal of this step is to extract the fault state information closely related to the first distribution feature element and refine it into a vector form, namely the first fault state vector. Through a series of complex calculations, the state vector mining branch can capture the key changes in the first distribution feature element, which are often a direct reflection of the occurrence of the fault. By mining these state vectors, the fault analysis system can have a more comprehensive understanding of the operating status of the audio system and provide strong support for subsequent fault analysis.

[0103] Next, the fault analysis system uses the temporal attention vector mining branch to perform temporal attention vector mining on the first fault state vector that has been obtained. The purpose of this step is to further refine the temporal characteristics in the first fault state vector, that is, the law of fault change over time, and express it in vector form, namely the first temporal attention vector. The temporal attention vector mining branch uses a special algorithm to identify specific temporal patterns or features in the first fault state vector, which are closely related to the fault development process of the audio system. By mining these temporal attention vectors, the fault analysis system can more accurately grasp the dynamic changes of the audio system fault, providing an important basis for subsequent fault prediction and diagnosis.

[0104] Then, the fault analysis system uses the newly introduced distribution feature recognition subnet to perform in-depth distribution feature recognition on the first temporal attention vector. The goal of this step is to parse the distribution feature information in the first temporal attention vector, that is, the distribution of the fault at a specific time point, and compare it with the second fault state event sample in the second operating state detection data sample to obtain a distribution feature recognition result sample. The distribution feature recognition subnet uses an advanced recognition algorithm to accurately identify specific distribution patterns or features in the first temporal attention vector, which are closely related to the fault distribution state of the audio system. By comparing these distribution features, the fault analysis system can determine whether the audio system has a similar fault distribution state, providing an important reference for subsequent fault location and repair.

[0105] Finally, the fault analysis system carefully debugs and optimizes the initial long short-term memory algorithm based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, as well as the difference between the distribution feature recognition result sample and the second distribution feature element, and finally obtains a more complete fault analysis algorithm. The goal of this step is to ensure that the fault analysis algorithm can accurately identify the abnormal pattern in the sound system and give an accurate fault judgment. Through debugging and optimization, the fault analysis system can improve the accuracy and robustness of the fault analysis algorithm, so that it can maintain stable performance in various complex situations.

[0106] In order to more specifically illustrate the implementation of this technical solution, an example of a numerical feature vector can be given. For example, the fault analysis system has a first operating state detection data sample, which includes a first fault state event sample and its corresponding first distribution feature element. The first fault state event sample can be represented by a numerical feature vector, such as [0.5, 0.7, 0.9, 0.6, 0.8], and the first distribution feature element can be represented by another numerical feature vector, such as [0.1, 0.2, 0.3, 0.2, 0.1].

[0107] First, the state vector mining branch will perform state vector mining on the first distribution feature element. Through a series of calculations, the state vector mining branch may obtain a first fault state vector, such as [0.2, 0.3, 0.4, 0.3, 0.2]. This vector represents the fault state information related to the first distribution feature element during the operation of the sound system.

[0108] Then, the temporal attention vector mining branch will perform temporal attention vector mining on the first fault state vector. Through the mining algorithm, the temporal attention vector mining branch may obtain a first temporal attention vector, such as [0.1, 0.2, 0.3, 0.2, 0.1]. This vector represents the law of the sound system fault changing over time, that is, the temporal feature.

[0109] Next, the distribution feature recognition subnet will perform distribution feature recognition on the first temporal attention vector. Through the recognition algorithm, the distribution feature recognition subnet may obtain a distribution feature recognition result sample, such as [0.05, 0.15, 0.25, 0.15, 0.05]. This vector represents the distribution of the sound system failure at a specific time point.

[0110] Finally, the fault analysis system debugs and optimizes the initial long short-term memory algorithm based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, as well as the difference between the distribution feature recognition result sample and the second distribution feature element. Through comparison and analysis, the fault analysis system can find out whether the sound system has a fault distribution state similar to the second fault state event sample. If there is a similar fault distribution state, the fault analysis system can determine that the sound system may have a similar fault and give corresponding fault judgment and processing suggestions.

[0111] In summary, by introducing the distribution feature recognition subnet and expanding the functions of the state vector mining branch and the time series attention vector mining branch, the fault analysis system successfully realizes the in-depth analysis and processing of the distribution feature elements in the operation status detection data samples. This technical solution can extract key fault state information and time series features from the first operation status detection data sample, and compare and analyze them with the second operation status detection data sample, and finally obtain accurate fault judgment and processing suggestions. This not only improves the accuracy and efficiency of the fault analysis algorithm, but also provides a new idea and method for the operation, maintenance and management of the sound system.

[0112] In the next step, based on the difference between the abnormal pattern sample judgment point of view and the second abnormal pattern keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm, including: determining a first algorithm training error based on the difference between the abnormal pattern sample judgment point of view and the second abnormal pattern keyword; determining a second algorithm training error based on the difference between the distribution feature recognition result sample and the second distribution feature element; determining a target algorithm training error based on the first algorithm training error and the second algorithm training error; optimizing the algorithm variables of the initial long short-term memory algorithm based on the target algorithm training error until they meet the debugging compliance requirements; determining a neural network algorithm consisting of a state vector mining branch, a temporal attention vector mining branch, and an abnormal pattern keyword judgment branch when debugging is completed as the fault analysis algorithm.

[0113] Based on this embodiment, the fault analysis system will further explore how to debug the initial long short-term memory algorithm based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, as well as the difference between the distribution feature recognition result sample and the second distribution feature element, and finally obtain a complete fault analysis algorithm.

[0114] First, the fault analysis system needs to determine the first algorithm training error based on the difference between the abnormal pattern sample judgment point of view and the second abnormal pattern keyword. In this step, the fault analysis system focuses on the accuracy of the algorithm in identifying abnormal patterns. Specifically, the fault analysis system compares the difference between the algorithm's judgment point of view on the abnormal pattern sample and the known second abnormal pattern keyword. If the algorithm can accurately identify the abnormal pattern and is highly consistent with the second abnormal pattern keyword, then the first algorithm training error will be small. Conversely, if the algorithm has deviations or errors in identifying abnormal patterns, then the first algorithm training error will be large.

[0115] Secondly, the fault analysis system needs to determine the second algorithm training error based on the difference between the distribution feature recognition result sample and the second distribution feature element. This step focuses on the accuracy of the algorithm in identifying the distribution feature. The fault analysis system compares the difference between the algorithm's output for the distribution feature recognition result sample and the known second distribution feature element. If the algorithm can accurately identify the distribution feature and is highly consistent with the second distribution feature element, the second algorithm training error will be smaller. Conversely, if the algorithm has deviations or errors in identifying the distribution feature, the second algorithm training error will be larger.

[0116] After determining the first algorithm training error and the second algorithm training error, the fault analysis system can further determine the target algorithm training error based on these two errors. The target algorithm training error is an error indicator that combines abnormal pattern recognition and distribution feature recognition, which reflects the overall performance of the algorithm. By calculating the weighted sum or average of the first algorithm training error and the second algorithm training error, the fault analysis system can obtain a quantified target algorithm training error. This error indicator will serve as the main basis for the subsequent optimization algorithm of the fault analysis system.

[0117] Next, the fault analysis system will optimize the algorithm variables of the initial LSTM algorithm based on the target algorithm training error until it meets the debugging compliance requirements. In this step, the fault analysis system will use various optimization algorithms and techniques, such as gradient descent and back propagation algorithms, to adjust the algorithm variables in the initial LSTM algorithm, such as weights and biases. Through continuous iteration and optimization, the fault analysis system can gradually reduce the target algorithm training error and improve the performance and accuracy of the algorithm. When the target algorithm training error reaches a predetermined threshold or meets other debugging compliance requirements, the fault analysis system can assume that the algorithm has been fully trained and optimized.

[0118] Finally, the fault analysis system determines the neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch and the abnormal pattern keyword discrimination branch during debugging as the fault analysis algorithm. This fault analysis algorithm has been fully trained and optimized, and can accurately identify the abnormal patterns and distribution characteristics in the audio system, and give corresponding fault judgment and processing suggestions. In practical applications, the fault analysis system can deploy this fault analysis algorithm to the monitoring and management platform of the audio system to realize real-time state monitoring and fault diagnosis of the audio system. In order to more specifically illustrate the implementation method of this technical solution, an example of a numerical feature vector can be given. For example, the fault analysis system has a feature vector [0.5, 0.7, 0.9] of the abnormal pattern sample discrimination point of view, which represents the algorithm's discrimination point of view for a certain abnormal pattern. At the same time, the fault analysis system has a feature vector [0.6, 0.8, 0.95] of the second abnormal pattern keyword, which represents the keyword of another known abnormal pattern. By comparing the difference between the two feature vectors, the fault analysis system can calculate the first algorithm training error.

[0119] Similarly, for example, the fault analysis system has a feature vector [0.2, 0.4, 0.6] of a distribution feature recognition result sample, which represents the recognition result of the algorithm for a certain distribution feature. At the same time, the fault analysis system has a feature vector [0.15, 0.35, 0.55] of a second distribution feature element, which represents the element of another known distribution feature. By comparing the difference between these two feature vectors, the fault analysis system can calculate the second algorithm training error.

[0120] Then, the fault analysis system can determine the target algorithm training error based on the first algorithm training error and the second algorithm training error. For example, if the first algorithm training error is 0.1 and the second algorithm training error is 0.2, the fault analysis system can calculate the weighted sum or average value of the two errors to obtain a quantified target algorithm training error. For example, the fault analysis system can simply average the two errors to obtain a target algorithm training error of (0.1+0.2) / 2=0.15.

[0121] Next, the fault analysis system will optimize the algorithm variables of the initial long short-term memory algorithm based on the target algorithm training error of 0.15. Through continuous iteration and optimization, the fault analysis system can gradually reduce the target algorithm training error. For example, the fault analysis system can use the gradient descent method to update the weights and biases of the algorithm so that the target algorithm training error gradually decreases. When the target algorithm training error reaches a predetermined threshold, such as 0.05, the fault analysis system can assume that the algorithm has been fully trained and optimized.

[0122] Finally, the fault analysis system determines the neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch, and the abnormal pattern keyword discrimination branch during debugging as the fault analysis algorithm. This fault analysis algorithm can accurately identify the abnormal patterns and distribution characteristics in the audio system, and give corresponding fault judgments and processing suggestions. In practical applications, the fault analysis system can apply this fault analysis algorithm to the monitoring and management of the audio system, and realize real-time status monitoring and fault diagnosis of the audio system.

[0123] In addition, it is worth mentioning that in the process of debugging and optimizing the initial long short-term memory algorithm, the fault analysis system can also introduce some innovative algorithm training error loss functions to improve the accuracy and efficiency of the algorithm. For example, the fault analysis system can design a loss function based on the attention mechanism, which can make the algorithm pay more attention to those feature vector elements that have an important impact on fault analysis, thereby improving the recognition accuracy of the algorithm. At the same time, the fault analysis system can also adopt an optimization algorithm based on dynamically adjusting the learning rate, which can dynamically adjust the size of the learning rate according to the training situation of the algorithm, thereby accelerating the convergence speed of the algorithm and improving the training efficiency.

[0124] In summary, by determining the algorithm training error based on the difference between the judgment viewpoint of the abnormal pattern sample and the second abnormal pattern keyword and the difference between the distribution feature recognition result sample and the second distribution feature element, and optimizing the algorithm variables of the initial long short-term memory algorithm based on these errors, the fault analysis system can obtain a perfect fault analysis algorithm. This algorithm can accurately identify the abnormal patterns and distribution characteristics in the sound system, and give corresponding fault judgments and processing suggestions. At the same time, by introducing innovative algorithm training error loss functions and optimization algorithms, the fault analysis system can further improve the accuracy and efficiency of the algorithm. This provides a new idea and method for the operation and management of sound systems.

[0125] In other possible embodiments, there are multiple sound system operation status log samples, the initial long short-term memory algorithm further includes an asynchronous recognition branch, and the method further includes: according to the state vector mining branch, each operation status detection data sample in the operation status detection data sample binary is subjected to state vector mining to obtain a second fault state vector; the operation status detection data sample binary is determined according to the multiple sound system operation status log samples; according to the temporal attention vector mining branch, the second fault state vector is subjected to temporal attention vector mining to obtain a second temporal attention vector; according to the asynchronous recognition branch, the second temporal attention vector corresponding to the operation status detection data sample binary is subjected to asynchronous recognition to obtain an asynchronous recognition result; based on the difference between the asynchronous recognition result corresponding to the operation status detection data sample binary and the prior asynchronous authentication annotation, a third algorithm training error is determined. Then, the determining of the target algorithm training error based on the first algorithm training error and the second algorithm training error includes: determining the target algorithm training error based on the first algorithm training error, the second algorithm training error and the third algorithm training error.

[0126] In this embodiment, there are multiple samples of the sound system operation status log, which means that the initial long short-term memory algorithm of the fault analysis system needs to process more complex and diverse data. In order to meet this demand, the initial long short-term memory algorithm of the fault analysis system is also specially designed with an asynchronous recognition branch to enhance the processing capability and accuracy of the algorithm.

[0127] First, the fault analysis system will perform state vector mining on each running state detection data sample in the running state detection data sample tuple based on the state vector mining branch. The purpose of this step is to extract useful features from the original running state detection data to form a second fault state vector. This second fault state vector is the basis for subsequent analysis, and it contains key information about the running state of the audio system. The running state detection data sample tuple is determined based on multiple audio system running state log samples, ensuring the diversity and representativeness of the data.

[0128] Next, the fault analysis system will mine the second fault state vector based on the temporal attention vector mining branch. The purpose of this step is to further extract the temporal features in the second fault state vector to form a second temporal attention vector. The temporal features are very important for analyzing the operating status of the audio system because they can reflect the changing trend of the system status over time.

[0129] Then, the fault analysis system will asynchronously identify the second temporal attention vector corresponding to the binary tuple of the running status detection data sample based on the asynchronous recognition branch. The purpose of this step is to use the processing power of the asynchronous recognition branch to perform a more in-depth analysis and recognition of the second temporal attention vector to obtain the asynchronous recognition result. The design of the asynchronous recognition branch enables the algorithm of the fault analysis system to better process asynchronous data, improving the accuracy and robustness of the algorithm.

[0130] After obtaining the asynchronous recognition result, the fault analysis system will determine the third algorithm training error based on the difference between the asynchronous recognition result corresponding to the running status detection data sample binary and the prior asynchronous authentication annotation. The purpose of this step is to evaluate the performance of the asynchronous recognition branch. By comparing the difference between the asynchronous recognition result and the prior asynchronous authentication annotation, the fault analysis system can obtain the third algorithm training error, which reflects the accuracy of the asynchronous recognition branch in identifying asynchronous data.

[0131] Finally, after determining the first algorithm training error, the second algorithm training error, and the third algorithm training error, the fault analysis system will determine the target algorithm training error based on these three errors. The purpose of this step is to comprehensively consider the performance of the algorithm in all aspects and obtain a comprehensive evaluation index. By optimizing the target algorithm training error, the fault analysis system can further improve the overall performance of the algorithm so that it can better meet the needs in practical applications.

[0132] In order to more specifically illustrate the implementation of this technical solution, an example of a numerical feature vector can be given. For example, the fault analysis system has an operation status detection data sample tuple, which contains two operation status detection data samples, represented by numerical feature vectors [0.5, 0.7, 0.9] and [0.6, 0.8, 0.95] respectively. These two numerical feature vectors represent the operation status of the sound system at different time points.

[0133] First, the fault analysis system will perform state vector mining on these two numerical feature vectors through the state vector mining branch to obtain the second fault state vector. For example, the second fault state vector after mining is [0.55, 0.75, 0.925]. This vector combines the information of the two original numerical feature vectors and performs appropriate processing to highlight the characteristics of the fault state.

[0134] Next, the fault analysis system will perform temporal attention vector mining on the second fault state vector through the temporal attention vector mining branch to obtain the second temporal attention vector. For example, the second temporal attention vector after mining is [0.6, 0.8, 0.9], which reflects the change trend of each element in the second fault state vector over time.

[0135] Then, the fault analysis system will perform asynchronous recognition on the second temporal attention vector through the asynchronous recognition branch to obtain an asynchronous recognition result. For example, if the asynchronous recognition result is "fault", it means that the audio system is in a fault state at the current time point.

[0136] After obtaining the asynchronous recognition result, the fault analysis system will compare it with the prior asynchronous authentication annotation. For example, if the prior asynchronous authentication annotation is also "fault", then the fault analysis system can believe that the recognition result of the asynchronous recognition branch is accurate and the third algorithm training error is small. On the contrary, if the prior asynchronous authentication annotation is "normal", then the fault analysis system can believe that there is an error in the recognition result of the asynchronous recognition branch and the third algorithm training error is large.

[0137] Finally, the fault analysis system will determine the target algorithm training error based on the first algorithm training error, the second algorithm training error, and the third algorithm training error. For example, if the first algorithm training error is 0.1, the second algorithm training error is 0.2, and the third algorithm training error is 0.15, then the fault analysis system can calculate the weighted sum or average value of these three errors to obtain a quantified target algorithm training error. For example, the fault analysis system can simply average these three errors to obtain a target algorithm training error of (0.1+0.2+0.15) / 3=0.15. This target algorithm training error reflects the overall performance of the algorithm and is the main basis for the subsequent optimization algorithm of the fault analysis system.

[0138] In summary, by introducing asynchronous identification branches and refining and supplementing related steps, the technical solution of the fault analysis system has been further improved and enhanced. This enables the algorithm of the fault analysis system to better handle complex sound system operating status data and improve the accuracy and efficiency of fault identification. At the same time, by comprehensively considering the performance of the algorithm in various aspects, the fault analysis system can obtain a comprehensive evaluation index, which provides strong support for the further optimization and application of the algorithm.

[0139] In some possible embodiments, the method further includes: generating a positive operation status detection data sample binary group based on the operation status detection data samples corresponding to the same audio system operation status log samples; adding a priori asynchronous authentication annotations of the same batch to the positive operation status detection data sample binary group; generating a negative operation status detection data sample binary group based on the operation status detection data samples corresponding to different audio system operation status log samples; adding a priori asynchronous authentication annotations of non-same batches to the negative operation status detection data sample binary group; generating the operation status detection data sample binary group based on the positive operation status detection data sample binary group and the negative operation status detection data sample binary group.

[0140] In this embodiment, in order to more comprehensively evaluate and optimize the operating status detection algorithm of the audio system, the fault analysis system introduces the concepts of positive operating status detection data sample tuples and negative operating status detection data sample tuples, and accordingly generates operating status detection data sample tuples for algorithm training and verification.

[0141] First, the fault analysis system generates a pair of positive operation status detection data samples based on the operation status detection data samples corresponding to the same audio system operation status log samples. The purpose of this step is to construct a set of data samples collected under the same or similar operation status for subsequent analysis of the algorithm's performance in this case. In order to ensure the accuracy and consistency of the data, the fault analysis system also adds a priori asynchronous authentication annotations from the same batch to this set of data samples. These annotations are based on historical data and expert experience and are used to guide the algorithm's training and verification process.

[0142] Secondly, the fault analysis system will generate negative operation status detection data sample tuples based on the operation status detection data samples corresponding to the operation status log samples of different audio systems. Unlike the positive operation status detection data sample tuples, this set of data samples is collected under different or highly different operation states. The purpose of this is to evaluate the robustness and generalization ability of the algorithm when facing diverse and complex operation states. Similarly, the fault analysis system will also add non-same batch prior asynchronous authentication annotations to this set of data samples for use in subsequent analysis and comparison.

[0143] After generating the positive and negative running state detection data sample tuples, the fault analysis system will generate the final running state detection data sample tuple based on these two sets of data samples. This tuple is the basis for algorithm training and verification. It contains data samples collected under different running states and the corresponding prior asynchronous authentication annotations. Through this design, the fault analysis system can more comprehensively evaluate the performance of the algorithm to ensure that it performs well in various situations.

[0144] Next, the fault analysis system will input this running status detection data sample binary into the fault analysis system's initial long short-term memory algorithm for processing. First, the state vector mining branch of the algorithm will perform state vector mining on each running status detection data sample in the binary, extract useful feature information, and form a second fault state vector. This vector is the basis for subsequent analysis, and it contains key information about the operating status of the audio system.

[0145] Then, the algorithm's temporal attention vector mining branch performs temporal attention vector mining on the second fault state vector. The purpose of this step is to further extract the temporal features in the second fault state vector to form a second temporal attention vector. The temporal features are very important for analyzing the operating status of the audio system because they can reflect the changing trend of the system status over time. By mining the temporal features, the fault analysis system can more accurately determine the current status of the audio system and its possible future changing trends.

[0146] After obtaining the second temporal attention vector, the fault analysis system will use the asynchronous recognition branch of the algorithm to perform asynchronous recognition on it. The purpose of this step is to use the processing power of the asynchronous recognition branch to perform a more in-depth analysis and recognition of the second temporal attention vector and obtain asynchronous recognition results. The design of the asynchronous recognition branch enables the algorithm of the fault analysis system to better process asynchronous data, improving the accuracy and robustness of the algorithm. Through asynchronous recognition, the fault analysis system can more accurately determine whether there is an abnormality or fault in the operating status of the audio system.

[0147] Finally, the fault analysis system determines the third algorithm training error based on the difference between the asynchronous recognition results corresponding to the binary tuples of the running status detection data samples and the prior asynchronous authentication annotations. The purpose of this step is to evaluate the performance of the asynchronous recognition branch. By comparing the difference between the asynchronous recognition results and the prior asynchronous authentication annotations, the fault analysis system can obtain the third algorithm training error. This error reflects the accuracy of the asynchronous recognition branch in identifying asynchronous data. If the error is large, it means that the algorithm may have misjudgments or missed judgments in some cases, and the fault analysis system needs to further optimize the algorithm model or adjust the parameter settings.

[0148] After determining the first algorithm training error, the second algorithm training error, and the third algorithm training error, the fault analysis system will determine the target algorithm training error based on these three errors. The purpose of this step is to comprehensively consider the performance of the algorithm in all aspects and obtain a comprehensive evaluation index. By optimizing the target algorithm training error, the fault analysis system can further improve the overall performance of the algorithm so that it can better meet the needs in practical applications. Specifically, the fault analysis system can calculate the weighted sum or average value of these three errors according to actual needs and application scenarios to obtain a quantified target algorithm training error. The smaller this error is, the better the performance of the algorithm is, and the more it can meet the needs of practical applications.

[0149] In summary, by introducing the concepts of positive operation status detection data sample tuples and negative operation status detection data sample tuples, and generating operation status detection data sample tuples for algorithm training and verification, the technical solution of the fault analysis system has been further improved and enhanced. This enables the algorithm of the fault analysis system to better handle complex sound system operation status data and improve the accuracy and efficiency of fault identification. At the same time, by comprehensively considering the performance of the algorithm in various aspects, the fault analysis system can obtain a comprehensive evaluation index, which provides strong support for the further optimization and application of the algorithm. In future work, the fault analysis system will continue to deepen the research and exploration of the sound system operation status detection algorithm, continuously improve the performance and accuracy of the algorithm, and provide more effective technical support for the stable operation and fault prevention of the sound system.

[0150] In other possible embodiments, the method also includes: determining the neural network algorithm composed of the state vector mining branch, the timing attention vector mining branch and the distribution feature recognition subnet when completing debugging as a distributed feature recognition algorithm; determining the neural network algorithm composed of the state vector mining branch, the timing attention vector mining branch and the asynchronous recognition branch when completing debugging as an asynchronous fault analysis algorithm.

[0151] Based on this embodiment, in order to more comprehensively evaluate and optimize the operating status detection algorithm of the audio system, the fault analysis system not only introduces the concepts of positive operating status detection data sample tuples and negative operating status detection data sample tuples, but also further defines a distribution feature recognition algorithm and an asynchronous fault analysis algorithm to more accurately process and analyze the operating status data of the audio system.

[0152] First, the fault analysis system generates a pair of positive operation status detection data samples based on the operation status detection data samples corresponding to the same audio system operation status log samples. The purpose of this step is to construct a set of data samples collected under the same or similar operation status for subsequent analysis of the algorithm's performance in this case. In order to ensure the accuracy and consistency of the data, the fault analysis system will also add a priori asynchronous authentication annotations from the same batch to this set of data samples. These annotations are based on historical data and expert experience, and they provide important guidance for the algorithm's training and verification process.

[0153] Secondly, the fault analysis system will generate negative operation status detection data sample tuples based on the operation status detection data samples corresponding to the operation status log samples of different audio systems. Unlike the positive operation status detection data sample tuples, this set of data samples is collected under different or highly different operation states. The purpose of this is to evaluate the robustness and generalization ability of the algorithm when facing diverse and complex operation states. Similarly, the fault analysis system will also add non-same batch prior asynchronous authentication annotations to this set of data samples for use in subsequent analysis and comparison.

[0154] After generating the positive and negative running state detection data sample tuples, the fault analysis system will generate the final running state detection data sample tuple based on these two sets of data samples. This tuple is the basis for algorithm training and verification. It contains data samples collected under different running states and the corresponding prior asynchronous authentication annotations. Through this design, the fault analysis system can more comprehensively evaluate the performance of the algorithm to ensure that it performs well in various situations.

[0155] Next, at the algorithm level, the fault analysis system will complete the neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch, and the distribution feature recognition subnet during debugging, and determine it as a distribution feature recognition algorithm. This algorithm is designed to deeply mine the distribution features in the operating status data of the audio system to more accurately determine the current state of the system. The distribution feature recognition subnet is the core part of this algorithm, which can further analyze and process the mined state vector and time series attention vector to extract more refined feature information.

[0156] At the same time, the fault analysis system will also complete the neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch and the asynchronous recognition branch during debugging, and determine it as an asynchronous fault analysis algorithm. This algorithm is designed to process and analyze asynchronous data and improve the algorithm's sensitivity and accuracy to changes in the operating state of the audio system. The asynchronous recognition branch is a key part of this algorithm. It can asynchronously identify the mined time series attention vector and determine whether there is an abnormality or fault in the operating state of the audio system.

[0157] In actual applications, the fault analysis system will input the running status detection data sample tuple into the distribution feature recognition algorithm and asynchronous fault analysis algorithm for processing and analysis. First, the distribution feature recognition algorithm will mine the state vector and time series attention vector of each running status detection data sample in the tuple, and then use the distribution feature recognition subnet to extract more detailed feature information. This information will be used to determine whether the current running status of the audio system is normal.

[0158] Secondly, the asynchronous fault analysis algorithm will asynchronously identify the mined temporal attention vectors to determine whether there are any abnormalities or faults in the operating status of the audio system. This step is the core part of the algorithm, which can monitor and analyze the operating status of the audio system in real time, and promptly discover and handle potential faults.

[0159] Finally, the fault analysis system determines the algorithm's training error based on the difference between the asynchronous recognition results corresponding to the binary tuples of the operating status detection data samples and the prior asynchronous authentication annotations. The purpose of this step is to evaluate the performance of the algorithm. By comparing the difference between the asynchronous recognition results and the prior asynchronous authentication annotations, the fault analysis system can obtain the algorithm's training error. This error reflects the accuracy of the algorithm in identifying the operating status of the audio system. If the error is large, it means that the algorithm may have misjudgments or missed judgments in some cases, and the fault analysis system needs to further optimize the algorithm model or adjust parameter settings.

[0160] In summary, by introducing the concepts of positive operation status detection data sample tuples and negative operation status detection data sample tuples, and generating operation status detection data sample tuples for algorithm training and verification, and defining the distribution feature recognition algorithm and asynchronous fault analysis algorithm to conduct in-depth analysis and processing of the operation status of the audio system, the technical solution of the fault analysis system has been further improved and enhanced. This enables the algorithm of the fault analysis system to better process complex audio system operation status data and improve the accuracy and efficiency of fault identification. At the same time, by comprehensively considering the performance of the algorithm in various aspects, the fault analysis system can obtain a comprehensive evaluation index, which provides strong support for the further optimization and application of the algorithm. In future work, the fault analysis system will continue to deepen the research and exploration of the audio system operation status detection algorithm, continuously improve the performance and accuracy of the algorithm, and provide more effective technical support for the stable operation and fault prevention of the audio system.

[0161] In the embodiment of the present application, the asynchronous identification branch is the core component of the asynchronous fault analysis algorithm, and its original design intention is to effectively process the asynchronous features in the operating status data of the sound system. In the actual operation of the sound system, due to various external interferences or internal faults, the system status data often presents non-stationary and nonlinear characteristics, and traditional synchronous processing methods often find it difficult to accurately capture these changes. The asynchronous identification branch can analyze these asynchronous data in real time and dynamically by introducing specific neural network structures and algorithms, thereby realizing accurate monitoring of the operating status of the sound system. When constructing the asynchronous identification branch, the fault analysis system uses a recurrent neural network (RNN) or its variants, such as a long short-term memory network (LSTM) or a gated recurrent unit (GRU). These network structures are good at processing sequence data and can capture the time dependency in the data. In the asynchronous identification branch, the fault analysis system specially designs multiple LSTM layers or GRU layers to achieve deep mining of the operating status data of the sound system. Each layer of the network receives the output of the previous layer of the network, and combines the current input data to calculate a more abstract and advanced feature representation.

[0162] As the core of the entire technical solution, the asynchronous fault analysis algorithm works based on the output results of the asynchronous identification branch. After the asynchronous identification branch completes the asynchronous feature extraction of the audio system operating status data, the asynchronous fault analysis algorithm will further analyze these features to determine whether the system has a fault or abnormal state. To achieve this goal, the fault analysis system integrates a variety of advanced machine learning technologies in the asynchronous fault analysis algorithm, including but not limited to support vector machines (SVM), decision trees, random forests, etc. These technologies can build efficient classification or regression models based on the extracted asynchronous features, thereby achieving accurate judgment of the operating status of the audio system.

[0163] In the specific implementation process, the fault analysis system will first convert the operating status log samples of the sound system into corresponding operating status detection data samples. These data samples contain the status information of the system at different time points and are the basis for subsequent algorithm processing. Then, the fault analysis system will generate positive operating status detection data sample tuples and negative operating status detection data sample tuples based on these data samples. These two sets of data samples represent the operating status data of the system in normal and abnormal states, respectively, and they will be used to train asynchronous recognition branches and asynchronous fault analysis algorithms.

[0164] During the training process, the fault analysis system will input these two sets of data samples into the neural network for learning. The asynchronous recognition branch in the network will first extract asynchronous features from the data, and then the asynchronous fault analysis algorithm will perform classification or regression prediction based on these features. By continuously iterating and optimizing network parameters, the fault analysis system can gradually improve the performance of the asynchronous recognition branch and the asynchronous fault analysis algorithm, ultimately achieving the purpose of accurately monitoring the operating status of the audio system.

[0165] It is worth noting that the performance of asynchronous identification branches and asynchronous fault analysis algorithms depends not only on the design and optimization of the network structure, but also on the quality and quantity of input data. Therefore, in practical applications, the fault analysis system needs to ensure the accuracy and completeness of the input data and verify the effectiveness and reliability of the algorithm through a large number of experiments.

[0166] In summary, the asynchronous identification branch and the asynchronous fault analysis algorithm, as the core invention of this application, play a vital role in the neural network algorithm. They achieve real-time, dynamic monitoring and accurate judgment of the operating status data of the audio system by introducing advanced network structure and algorithm technology.

[0167] In some independent embodiments, after obtaining the target abnormal pattern keyword of the second fault status event in the second operating status detection data, the method also includes: determining the audio system maintenance plan corresponding to the audio system operating status log to be analyzed based on the target abnormal pattern keyword of the second fault status event in the second operating status detection data.

[0168] Based on this embodiment, after determining the target abnormal pattern keywords of the second fault state event in the second operating state detection data, its application scope is further expanded, aiming to accurately locate and determine the audio system maintenance plan corresponding to the audio system operating state log to be analyzed through these keywords. This expansion not only improves the response speed of the fault analysis system to the audio system fault, but also greatly enhances the accuracy and efficiency of fault handling.

[0169] First of all, the fault analysis system needs to make it clear that the target abnormal pattern keywords of the second fault state event in the second operating state detection data are obtained through a series of complex algorithms and models based on an in-depth analysis of the sound system operating state data. These keywords accurately reflect the specific performance of the sound system when a fault occurs, and are an important basis for the fault analysis system to perform subsequent maintenance and fault handling.

[0170] After determining these target abnormal pattern keywords, the fault analysis system method will further use these keywords to match and locate the maintenance plan of the audio system. This step is achieved thanks to the fact that the fault analysis system has established a comprehensive and detailed audio system maintenance plan library in advance. In this plan library, the fault analysis system has developed corresponding maintenance plans for various faults and abnormal conditions that may occur in the audio system. These plans include key information such as fault diagnosis methods, processing steps, required tools, and estimated time consumption, which are important references for the fault analysis system to perform audio system maintenance and fault handling.

[0171] Specifically, after the fault analysis system extracts the target abnormal pattern keywords of the second fault state event from the second operating state detection data, the fault analysis system will match these keywords with the plans in the plan library. The matching process is based on the similarity and relevance of the keywords. The fault analysis system will calculate the similarity score between each plan and the keyword, and then sort them according to the score. The plan with the highest score is the maintenance plan that the fault analysis system believes is most suitable for the current audio system fault situation.

[0172] However, it is not enough to simply rely on keyword matching to determine the maintenance plan. In actual applications, the failure of the audio system may be caused by a variety of factors, and the manifestation of the failure may also vary due to different system configurations and operating environments. Therefore, after determining the maintenance plan, the fault analysis system also needs to further verify and adjust the plan in combination with the actual operating status and log data of the audio system.

[0173] In this step, the fault analysis system will compare and analyze the actual operating status data and log data of the sound system with the diagnostic methods and processing steps in the plan. Through comparison, the fault analysis system can verify the accuracy of the plan and find possible problems or deficiencies. If it is found that the plan deviates greatly from the actual situation or cannot effectively solve the problem, the fault analysis system will adjust and optimize the plan in time to ensure that it can truly meet the needs of sound system maintenance and fault handling.

[0174] In addition, when determining the maintenance plan for the sound system, the fault analysis system will also fully consider the feasibility and implementation difficulty of the plan. Because in actual applications, some plans may be feasible in theory, but cannot be actually implemented due to the difficulty of implementation or the excessive resources required. Therefore, the fault analysis system will evaluate and optimize the plan in combination with the actual operating environment and resource conditions of the sound system to ensure that it can be effectively implemented in actual applications.

[0175] In summary, by determining the method of the maintenance plan for the audio system based on the target abnormal pattern keywords of the second fault state event in the second operating state detection data, the fault analysis system not only improves the response speed and processing accuracy of the audio system faults, but also enhances the feasibility and implementation effect of the plan. The application of this method will greatly improve the stability and reliability of the audio system and provide a strong guarantee for the long-term stable operation of the audio system. In future work, the fault analysis system will continue to deepen the research and exploration of this technology, continuously improve its performance and accuracy, and provide more effective technical support for the maintenance and fault handling of the audio system. At the same time, the fault analysis system also looks forward to cooperating with more industry partners to jointly promote the advancement and development of audio system maintenance technology.

[0176] In some other independent embodiments, based on the target abnormal pattern keyword of the second fault state event in the second operating state detection data, determining the audio system maintenance plan corresponding to the audio system operating state log to be analyzed includes: obtaining an operating state trend data sequence based on the target abnormal pattern keyword, wherein the operating state trend data sequence includes uninterrupted X groups of operating state trend data, and X is an integer greater than or equal to 1; obtaining a fault state trend data sequence according to the operating state trend data sequence, wherein the fault state trend data sequence includes uninterrupted X groups of fault state trend data; based on the operating state trend data sequence, generating a maintenance plan for the network through the first A state semantic embedding module obtains an operating state embedded semantic sequence, wherein the operating state embedded semantic sequence includes X operating state embedded semantics; based on the fault state trend data sequence, a fault state embedded semantic sequence is obtained through a second state semantic embedding module included in the maintenance plan generation network, wherein the fault state embedded semantic sequence includes X fault state embedded semantics; based on the operating state embedded semantic sequence and the fault state embedded semantic sequence, a plan matching weight corresponding to the operating state trend data sequence is obtained through a soft threshold activation module included in the maintenance plan generation network; and a sound system maintenance plan for the operating state trend data sequence is determined according to the plan matching weight.

[0177] Based on this embodiment, it involves how to accurately determine the sound system maintenance plan corresponding to the sound system operation status log to be analyzed based on the target abnormal pattern keyword of the second fault state event in the second operation state detection data. This process not only involves in-depth analysis of the operation state trend, but also combines an advanced maintenance plan generation network to ensure that the fault analysis system can formulate the most appropriate and effective maintenance plan.

[0178] First, the fault analysis system extracts target abnormal pattern keywords from the second operating state detection data. These keywords are condensed expressions of the specific patterns exhibited by the audio system when a fault occurs. They are like a key that can open the door for the fault analysis system to understand the causes and mechanisms of audio system failures.

[0179] Next, the fault analysis system uses these keywords as a guide to obtain the operating status trend data sequence of the sound system. This data sequence contains X sets of uninterrupted operating status trend data, where X is an integer greater than or equal to 1. These data are like a long scroll of history, recording the operating status changes of the sound system over a period of time, and are an important basis for the fault analysis system to analyze its future trends and possible faults.

[0180] After obtaining the operating status trend data sequence, the fault analysis system further extracts the fault status trend data sequence from it. This sequence also contains X groups of uninterrupted fault status trend data, which are accurate capture and prediction of possible fault states of the sound system.

[0181] At this point, the fault analysis system has two sets of valuable data series: operating status trend data series and fault status trend data series. To make full use of these two sets of data and develop an accurate maintenance plan for the sound system, the fault analysis system needs to rely on the maintenance plan generation network to achieve this.

[0182] The maintenance plan generation network is a deep learning model built by the fault analysis system. It contains multiple functional modules and can achieve in-depth analysis and precise matching of operating status and fault status. Among them, the first state semantic embedding module and the second state semantic embedding module are the core components of this network.

[0183] The fault analysis system converts the operating state trend data sequence into an operating state embedded semantic sequence through the first state semantic embedding module. This sequence contains X operating state embedded semantics, which are highly abstract and concise expressions of the operating state of the sound system. Similarly, the fault analysis system converts the fault state trend data sequence into a fault state embedded semantic sequence through the second state semantic embedding module. This sequence contains X fault state embedded semantics, which are highly abstract and concise expressions of the fault state of the sound system.

[0184] After having the embedded semantic sequence of the operating state and the embedded semantic sequence of the fault state, the fault analysis system can obtain the plan matching weight corresponding to the operating state trend data sequence through the soft threshold activation module included in the maintenance plan generation network. This weight is a quantitative expression of the degree of matching between different maintenance plans and the current operating state and fault state of the audio system, and is an important basis for the fault analysis system to determine the final maintenance plan.

[0185] Finally, the fault analysis system selects the plan that best matches the current operating status and fault status of the audio system from a large number of maintenance plans based on the plan matching weights as the final audio system maintenance plan. This plan not only includes detailed processing steps for the current fault, but also provides suggestions and maintenance strategies to prevent similar faults from happening again. It is an important guarantee for the fault analysis system to ensure the long-term stable operation of the audio system.

[0186] In summary, by determining the maintenance plan for the audio system based on the target abnormal pattern keywords of the second fault state event in the second operating state detection data, the fault analysis system not only achieves accurate positioning and rapid response to the fault of the audio system, but also formulates the most appropriate and effective maintenance plan through the advanced maintenance plan generation network. The application of this method will greatly improve the stability and reliability of the audio system and provide a strong guarantee for the long-term stable operation of the audio system. In future work, the fault analysis system will continue to deepen the research and exploration of this technology, continuously improve its performance and accuracy, and provide more effective technical support for the maintenance and fault handling of the audio system. At the same time, the fault analysis system also looks forward to cooperating with more industry partners to jointly promote the advancement and development of audio system maintenance technology and contribute the power of the fault analysis system to the stable operation and continuous optimization of the audio system.

[0187] In some other independent embodiments, the method of obtaining the plan matching weight corresponding to the running state trend data sequence through the soft threshold activation module included in the maintenance plan generation network based on the running state embedded semantic sequence and the fault state embedded semantic sequence includes: obtaining X first sound impairment quantization arrays through the first sound impairment detection subnet included in the maintenance plan generation network based on the running state embedded semantic sequence, wherein each first sound impairment quantization array corresponds to an running state embedded semantic; obtaining X second sound impairment quantization arrays through the second sound impairment detection subnet included in the maintenance plan generation network based on the fault state embedded semantic sequence, wherein each second sound impairment quantization array corresponds to a fault state embedded semantic; concatenating the X first sound impairment quantization arrays and the X second sound impairment quantization arrays to obtain X target sound impairment quantization arrays, wherein each target sound impairment quantization array includes a first sound impairment quantization array and a second sound impairment quantization array; and obtaining the plan matching weight corresponding to the running state trend data sequence through the soft threshold activation module included in the maintenance plan generation network based on the X target sound impairment quantization arrays.

[0188] It can be understood that this embodiment further explores how to accurately obtain the plan matching weight corresponding to the operation status trend data sequence based on the operation status embedded semantic sequence and the fault status embedded semantic sequence through the soft threshold activation module in the maintenance plan generation network. This process involves the integration of multiple key steps and advanced technologies to ensure that the fault analysis system can formulate the most appropriate and effective sound system maintenance plan.

[0189] First, the fault analysis system uses the acquired operation state embedded semantic sequence to obtain X first sound loss quantization arrays through the first sound loss detection subnet in the maintenance plan generation network. In this step, each first sound loss quantization array corresponds to an operation state embedded semantic, which is a quantitative expression of the degree of sound quality loss of the sound system under different operation states. Through this step, the fault analysis system can convert the abstract operation state embedded semantics into specific and measurable sound quality loss data, which provides a basis for subsequent analysis and matching.

[0190] Next, the fault analysis system also uses the fault state embedded semantic sequence to obtain X second sound loss quantization arrays through the second sound loss detection subnet in the maintenance plan generation network. Similar to the first sound loss quantization array, each second sound loss quantization array also corresponds to a fault state embedded semantic, which is a quantitative expression of the degree of sound quality loss of the sound system under different fault states. The completion of this step enables the fault analysis system to have sound quality loss data of the sound system in normal operation and fault state at the same time, which makes it possible for subsequent comparison and analysis.

[0191] After obtaining X first artifact quantization arrays and X second artifact quantization arrays, the fault analysis system performs a key splicing processing step. The fault analysis system splices each first artifact quantization array with its corresponding second artifact quantization array to obtain X target artifact quantization arrays. Each target artifact quantization array contains a first artifact quantization array and a second artifact quantization array, which together constitute the complete sound quality loss information of the sound system at the same time point or in the same operating state.

[0192] After having X target sound loss quantization arrays, the fault analysis system can enter the next key step - obtaining the plan matching weight corresponding to the operation status trend data sequence through the soft threshold activation module in the maintenance plan generation network. The soft threshold activation module is a neural network module carefully designed by the fault analysis system. It can automatically adjust and output a plan matching weight that matches the current sound system operation status and fault status based on the input sound quality loss data. This weight is a quantitative expression of the applicability and effectiveness of different maintenance plans in the current state, and is an important basis for the fault analysis system to determine the final maintenance plan.

[0193] In the soft threshold activation module, the fault analysis system uses advanced deep learning algorithms and activation functions to ensure that the output plan matching weights can accurately reflect the matching degree between different maintenance plans and the current sound system status. Through the processing of this module, the fault analysis system can obtain a plan matching weight sequence corresponding to the operating status trend data sequence, in which each weight corresponds to a possible maintenance plan.

[0194] Finally, the fault analysis system selects the plan that best matches the current state of the audio system from a large number of maintenance plans according to the plan matching weight sequence as the final maintenance plan for the audio system. This plan not only includes detailed processing steps for the current fault, but also provides suggestions and maintenance strategies to prevent similar faults from happening again. Through the implementation of this method, the fault analysis system can accurately locate and quickly respond to audio system faults, providing a strong guarantee for the long-term stable operation of the audio system. In summary, by using the soft threshold activation module in the maintenance plan generation network to obtain the plan matching weight corresponding to the running state trend data sequence based on the running state embedded semantic sequence and the fault state embedded semantic sequence, the fault analysis system not only realizes the deep analysis and precise matching of the audio system state, but also formulates the most appropriate and effective maintenance plan. The application of this method will greatly improve the stability and reliability of the audio system and provide strong technical support for the long-term stable operation of the audio system. In future work, the fault analysis system will continue to deepen the research and exploration of this technology, continuously improve its performance and accuracy, and provide more effective solutions for the maintenance and fault handling of the audio system. At the same time, the fault analysis system also looks forward to cooperating with more industry partners to jointly promote the advancement and development of audio system maintenance technology and contribute the power of the fault analysis system to the stable operation and continuous optimization of the audio system.

[0195] Further, Figure 2 FIG. 2 is a schematic diagram of a fault analysis system 200 provided in an embodiment of the present application. Figure 2 The fault analysis system 200 shown includes a processor 210, and the processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0196] Alternatively, if Figure 2 As shown, the fault analysis system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0197] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0198] Alternatively, if Figure 2 As shown, the fault analysis system 200 may further include a transceiver 220. The processor 210 may control the transceiver 220 to interact with other devices. Specifically, the transceiver 220 may send information or data to other devices, or receive information or data sent by other devices.

[0199] Optionally, the fault analysis system 200 may implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application, which will not be described in detail here for the sake of brevity.

[0200] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.

[0201] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0202] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0203] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0204] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0205] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the present application, all of which are within the protection of the present application.

Claims

1. A method for analyzing sound system faults based on machine learning, characterized in that: Applied to a fault analysis system, the method comprises: The fault analysis system obtains a sound system operation status log to be analyzed, wherein the sound system operation status log to be analyzed includes first operation status detection data and second operation status detection data; the first operation status detection data and the second operation status detection data correspond to fault hazards of different abnormal modes; the first operation status detection data includes a first fault status event with a keyword for setting an abnormal mode added thereto; The fault analysis system inputs the operating state log of the audio system to be analyzed into a fault analysis algorithm, and processes the first fault state event to which the set abnormal mode keyword is added in the first operating state detection data according to the fault analysis algorithm, and obtains the target abnormal mode keyword of the second fault state event in the second operating state detection data; The debugging method of the fault analysis algorithm comprises: Acquire a sound system operation status log sample, wherein the sound system operation status log sample includes a first operation status detection data sample and a second operation status detection data sample; the first operation status detection data sample and the second operation status detection data sample correspond to fault hidden danger samples of different abnormal modes; the first operation status detection data sample includes a first fault status event sample with a first abnormal mode keyword added, and the second operation status detection data sample includes a second fault status event sample with a second abnormal mode keyword added; Inputting the sound system operation status log sample into an initial long short-term memory algorithm, performing state vector mining on the first fault state event sample to which the first abnormal mode keyword is added according to the state vector mining branch in the initial long short-term memory algorithm, to obtain a fault state vector sample; According to the temporal attention vector mining branch in the initial long short-term memory algorithm, temporal attention vector mining is performed on the fault state vector sample to obtain a temporal attention vector sample; According to the abnormal pattern keyword discrimination branch in the initial long short-term memory algorithm, the temporal attention vector sample is subjected to abnormal pattern keyword discrimination to obtain an abnormal pattern sample discrimination viewpoint corresponding to the second fault state event sample in the second operating state detection data sample; Based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

2. The method according to claim 1, characterized in that The temporal attention vector mining branch includes a knowledge embedding module, a knowledge mapping module and a distribution vector extraction module. The temporal attention vector mining branch in the initial long short-term memory algorithm performs temporal attention vector mining on the fault state vector sample to obtain a temporal attention vector sample, including: Performing a knowledge embedding operation on the fault state vector sample according to the knowledge embedding module to obtain a linear knowledge vector sample; Performing a distribution vector extraction operation on the fault state vector sample according to the distribution vector extraction module to obtain a distribution vector sample; The linear knowledge vector sample and the distribution vector sample are subjected to knowledge mapping operation according to the knowledge mapping module to obtain the temporal attention vector sample.

3. The method according to claim 1, characterized in that: The state vector mining branch includes a residual connection subnet, a quantization processing subnet and a state analysis subnet. The state vector mining branch in the initial long short-term memory algorithm performs state vector mining on the first fault state event sample with the first abnormal mode keyword added to obtain a fault state vector sample, including: performing, according to the residual connection subnetwork, mining of an artifact element vector on the first fault state event sample to which the first abnormal mode keyword is added, to obtain an artifact element vector sample; Performing a quantized heat map transformation on the artifact element vector sample according to the quantization processing subnet to obtain a quantized heat map vector sample; A fault state analysis operation is performed on the quantized heat map vector sample according to the state analysis subnet to obtain the fault state vector sample.

4. The method according to claim 1, characterized in that: The first operation status detection data sample further includes a first distribution feature element of the first fault status event sample, the second operation status detection data sample further includes a second distribution feature element of the second fault status event sample, the initial long short-term memory algorithm further includes a distribution feature identification subnet, and the method further includes: Performing state vector mining on a first distribution feature element in the first running state detection data sample according to the state vector mining branch to obtain a first fault state vector; According to the temporal attention vector mining branch, performing temporal attention vector mining on the first fault state vector to obtain a first temporal attention vector; According to the distribution feature recognition subnet, the first temporal attention vector is subjected to distribution feature recognition to obtain a distribution feature recognition result sample corresponding to the second fault state event sample in the second operating state detection data sample; The method of determining the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, debugging the initial long short-term memory algorithm, and obtaining a fault analysis algorithm includes: Based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element, the initial long short-term memory algorithm is debugged to obtain a fault analysis algorithm.

5. The method according to claim 4, characterized in that The method of debugging the initial long short-term memory algorithm based on the difference between the abnormal pattern sample judgment viewpoint and the second abnormal pattern keyword, and the difference between the distribution feature recognition result sample and the second distribution feature element, to obtain a fault analysis algorithm includes: Determining a first algorithm training error based on the difference between the abnormal pattern sample discrimination viewpoint and the second abnormal pattern keyword; Determining a second algorithm training error based on a difference between the distribution feature recognition result sample and the second distribution feature element; Determining a target algorithm training error based on the first algorithm training error and the second algorithm training error; Based on the target algorithm training error, optimizing the algorithm variables of the initial long short-term memory algorithm until the debugging compliance requirements are met; The neural network algorithm composed of the state vector mining branch, the temporal attention vector mining branch and the abnormal pattern keyword discrimination branch during the completion of debugging is determined as the fault analysis algorithm.

6. The method according to claim 5, characterized in that The sound system operation status log samples are multiple, the initial long short-term memory algorithm further includes an asynchronous identification branch, and the method further includes: According to the state vector mining branch, each running state detection data sample in the running state detection data sample tuple is subjected to state vector mining to obtain a second fault state vector; the running state detection data sample tuple is determined according to the multiple sound system running state log samples; According to the temporal attention vector mining branch, performing temporal attention vector mining on the second fault state vector to obtain a second temporal attention vector; According to the asynchronous recognition branch, asynchronously recognize the second temporal attention vector corresponding to the running status detection data sample binary to obtain an asynchronous recognition result; Determining a third algorithm training error based on a difference between an asynchronous recognition result corresponding to the running status detection data sample binary and a priori asynchronous authentication annotation; The determining a target algorithm training error based on the first algorithm training error and the second algorithm training error includes: A target algorithm training error is determined based on the first algorithm training error, the second algorithm training error, and the third algorithm training error.

7. The method according to claim 6, characterized in that The method further comprises: Generate a positive running state detection data sample tuple according to the running state detection data sample corresponding to the same sound system running state log sample; the positive running state detection data sample tuple is added with a priori asynchronous authentication annotations of the same batch; Generate a negative running state detection data sample tuple according to the running state detection data samples corresponding to the running state log samples of different sound systems; the negative running state detection data sample tuple is added with a priori asynchronous authentication annotations of non-same batches; The running state detection data sample tuple is generated according to the positive running state detection data sample tuple and the negative running state detection data sample tuple.

8. The method according to claim 7, characterized in that The method further comprises: The neural network algorithm composed of the state vector mining branch, the time series attention vector mining branch and the distribution feature recognition subnet during debugging is determined as a distribution feature recognition algorithm; The neural network algorithm consisting of the state vector mining branch, the temporal attention vector mining branch and the asynchronous recognition branch during debugging is determined as an asynchronous fault analysis algorithm.

9. A fault analysis system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 8.

Citation Information

Patent Citations

  • Fault processing method and device

    CN105635197A

  • Fault identification method applied to refrigeration house management system and artificial intelligence server

    CN116703284A

  • Rail transit operation and maintenance state data processing method and system based on artificial intelligence

    CN117993894A

  • Early warning method and system for fan fault investigation

    CN118013439A

  • Mould opening equipment operation state anomaly detection method and device and storage medium

    CN118484755A