A water pipe leak detection model training method, system, device and storage medium

Abstract

This application relates to a water pipe leak detection model training method, system, device, and storage medium, and relates to the field of pipeline detection technology. The method includes: obtaining historical audio data information of a water supply pipeline; analyzing and processing the historical audio data information of the water supply pipeline according to a preset frequency band separation method to form historical audio frequency band data information of the water pipe; and analyzing and processing the historical audio frequency band data information of the water pipe according to a preset audio detection data model building method to form a water pipe leak detection model. This application facilitates real-time leak detection of water supply pipelines.

Description

Technical Field

[0001] The present application relates to the field of pipeline detection technology, and in particular to a water pipe leakage detection model training method, system, device and storage medium. Background Art

[0002] In daily life, water supply pipe leakage is a common problem, which refers to water leakage in transportation pipes, valves, joints and other parts, which can easily lead to problems such as water pressure drop in the water supply system, affected water quality and waste of a large amount of water resources. Therefore, it is necessary to conduct leakage detection on the water supply pipe.

[0003] In the prior art, leak detection in water supply pipes typically requires dismantling and destroying the pipes, which is costly and disruptive. Therefore, a non-invasive detection method is needed to address this issue. Currently, the mainstream non-invasive method involves water utility personnel manually listening for leaks in water supply pipes.

[0004] Regarding the above-mentioned related technologies, the following defects were found: when using manual leak detection to detect leaks in water supply pipes, it is impossible to detect leaks in water supply pipes in real time because it requires water workers with a lot of accumulated experience to perform manual leak detection on the water supply pipes to make judgments. There is still room for improvement. Summary of the Invention

[0005] In order to facilitate real-time leakage detection of water supply pipes, the present application provides a water pipe leakage detection model training method, system, device and storage medium.

[0006] In a first aspect, the present application provides a water pipe leak detection model training method, which adopts the following technical solutions:

[0007] A water pipe leak detection model training method, comprising:

[0008] Obtain historical audio data information of water supply pipelines;

[0009] Analyzing and processing the historical audio data information of the water supply pipeline according to a preset frequency band separation method to form historical audio frequency band data information of the water pipeline;

[0010] According to the preset audio detection data model building method, the historical audio frequency band data information of the water pipe is analyzed and processed to form a water pipe leakage detection model.

[0011] By adopting the above technical solution, the historical audio data information of the water supply pipeline is acquired, the historical audio data information of the water supply pipeline is analyzed and processed by the frequency band separation method to form the historical audio frequency band data information of the water pipe, and the historical audio frequency band data information of the water pipe is analyzed and processed by the audio detection data model building method to form a water pipe leakage detection model, thereby facilitating real-time leakage detection of the water supply pipeline using the water pipe leakage detection model.

[0012] Optionally, analyzing and processing the water supply pipeline historical audio data information according to a preset frequency band separation method to form the water pipe historical audio frequency band data information includes:

[0013] Retrieving the maximum frequency value of the data corresponding to the historical audio data information of the water supply pipeline according to the historical audio data information of the water supply pipeline; analyzing and obtaining the frequency band frequency interval length value corresponding to the maximum frequency value of the data and the preset number of audio separation frequency bands according to the corresponding relationship between the maximum frequency value of the data, the preset number of audio separation frequency bands and the preset frequency band frequency interval length value;

[0014] According to the correspondence between the historical audio data information of the water supply pipeline, the number of audio separation frequency bands, the frequency band frequency interval length value and the preset audio separation frequency band information, the audio separation frequency band information corresponding to the historical audio data information of the water supply pipeline, the number of audio separation frequency bands and the frequency band frequency interval length value is analyzed and obtained, and the audio separation frequency band information is used as the historical audio frequency band data information of the water pipe.

[0015] By adopting the above technical solution, the maximum frequency value of the data is retrieved through the historical audio data information of the water supply pipeline, and the frequency band frequency interval length value is obtained by analyzing the maximum frequency value of the data and the numerical value of the audio separation frequency band. Then, the audio separation frequency band information is obtained by analyzing the historical audio data information of the water supply pipeline, the numerical value of the audio separation frequency band and the frequency band frequency interval length value, and the audio separation frequency band information is used as the historical audio frequency band data information of the water pipe, thereby improving the accuracy of the obtained audio separation frequency band information and facilitating subsequent processing and analysis.

[0016] Optionally, analyzing and processing historical audio frequency band data information of water pipes according to a preset audio detection data model building method to form a water pipe leakage detection model includes:

[0017] Analyze and process the historical audio frequency band data of the water pipe according to the preset separation frequency band selection method to form the final separation frequency band selection information;

[0018] Analyzing and processing the final selected information of the separated frequency bands according to a preset separated frequency band weighted analysis method to form separated frequency band weighted information;

[0019] According to the preset weighted aggregation training method, the final selection information of the separated frequency bands and the weighted information of the separated frequency bands are analyzed and processed to form a weighted aggregation training model, and the weighted aggregation training model is used as a water pipe leakage detection model.

[0020] By adopting the above technical solution, the historical audio frequency band data information of the water pipe is analyzed and processed by the separation frequency band selection method to form the final selection information of the separation frequency band, the final selection information of the separation frequency band is analyzed and processed by the separation frequency band weighted analysis method to form the separation frequency band weighted information, the final selection information of the separation frequency band and the separation frequency band weighted information are analyzed and processed by the weighted aggregation training method to form a weighted aggregation training model, and the weighted aggregation training model is used as the water pipe leakage detection model. By selecting frequency bands and performing weighted aggregation, the information of each frequency band is integrated to improve the accuracy of the acquired water pipe leakage detection model.

[0021] Optionally, analyzing and processing the water pipe historical audio frequency band data information according to a preset separation frequency band selection method to form final separation frequency band selection information includes:

[0022] Analyzing and processing the audio separation frequency band information according to a preset greedy strategy selection method to form separation frequency band initial subsequence candidate information;

[0023] Analyzing and processing the separated frequency band initial subsequence candidate information according to a preset timing factor analysis method to form separated frequency band timing subsequence candidate information;

[0024] The separation frequency band timing subsequence candidate information is analyzed and processed according to a preset timing factor selection method to form separation frequency band timing subsequence selection information, and the separation frequency band timing subsequence selection information is used as the separation frequency band final selection information.

[0025] By adopting the above technical solution, the audio separation frequency band information is analyzed and processed by the greedy strategy selection method to form the separation frequency band initial subsequence candidate information, the separation frequency band initial subsequence candidate information is analyzed and processed by the timing factor analysis method to form the separation frequency band timing subsequence candidate information, the separation frequency band timing subsequence candidate information is analyzed and processed by the timing factor selection method to form the separation frequency band timing subsequence selection information, and the separation frequency band timing subsequence selection information is used as the separation frequency band final selection information, so that the obtained separation frequency band final selection information contains a subsequence with time perception ability, and the subsequence with time perception ability is better than the ordinary subsequence in distinguishing normal and abnormal samples in the time dimension, thereby improving the accuracy of the obtained separation frequency band final selection information, and being able to generate an evolutionary relationship in the time dimension, which is convenient for subsequent analysis and processing.

[0026] Optionally, analyzing and processing the final separation frequency band selection information according to a preset separation frequency band weighted analysis method to form separation frequency band weighted information includes:

[0027] Analyzing and processing the final selection information of the separated frequency bands according to a preset subsequence evolution graph confirmation method to form separated frequency band subsequence evolution graph information;

[0028] Analyzing and processing the separated frequency band subsequence evolution graph information according to a preset node label classification method to form separated frequency band subsequence label information;

[0029] Analyzing and processing the separated frequency band subsequence label information and the preset network intermediate layer output vector information according to a preset embedding vector analysis method to form separated frequency band embedding vector information;

[0030] Analyzing and processing the separated frequency band embedded vector information according to a preset embedded vector aggregation method to form separated frequency band aggregated feature information;

[0031] According to the historical audio data information of the water supply pipeline, the statistical feature information and Mel-cepstral coefficient feature information corresponding to the historical audio data information of the water supply pipeline are retrieved, and the separated frequency band aggregation feature information, statistical feature information and Mel-cepstral coefficient feature information are used as the separated frequency band weighted information.

[0032] By adopting the above technical scheme, the final selection information of the separated frequency band is analyzed and processed by the subsequence evolution graph confirmation method to form the separated frequency band subsequence evolution graph information, the separated frequency band subsequence evolution graph information is analyzed and processed by the node label classification method to form the separated frequency band subsequence label information, the separated frequency band subsequence label information and the preset network intermediate layer output vector information are analyzed and processed by the embedded vector analysis method to form the separated frequency band embedded vector information, the separated frequency band embedded vector information is analyzed and processed by the embedded vector aggregation method to form the separated frequency band aggregated feature information, the statistical feature information and the Mel cepstral coefficient feature information are retrieved through the historical audio data information of the water supply pipeline, and the separated frequency band aggregated feature information, the statistical feature information and the Mel cepstral coefficient feature information are used as the separated frequency band weighted information, so as to reduce the feature dimension through the separated frequency band weighted information, facilitate subsequent aggregation, and improve the robustness of the model by adding the statistical feature information and the Mel cepstral coefficient feature information.

[0033] Optionally, analyzing and processing the final selection information of the separated frequency bands according to a preset subsequence evolution graph confirmation method to form the separated frequency band subsequence evolution graph information includes:

[0034] Analyzing and processing the final selection information of the separated frequency bands according to a preset adjacent frequency band selection method to form adjacent frequency band information;

[0035] According to the correspondence between the final selection information of the separated frequency band, the adjacent frequency band information and the preset frequency band association probability value, analyzing and obtaining the frequency band association probability value corresponding to the final selection information of the separated frequency band and the adjacent frequency band information;

[0036] According to the preset normalization analysis method, the final selection information of the separated frequency bands and the frequency band associated probability values ​​are analyzed and processed to form the separated frequency band subsequence normalization information, and the separated frequency band subsequence normalization information is used as the separated frequency band subsequence evolution graph information.

[0037] By adopting the above technical solution, the final selection information of the separated frequency band is analyzed and processed by the adjacent frequency band selection method to form the adjacent frequency band information, the frequency band association probability value is obtained by analyzing the final selection information of the separated frequency band and the adjacent frequency band information, and the final selection information of the separated frequency band and the frequency band association probability value are analyzed and processed by the normalization analysis method to form the separated frequency band subsequence normalization information, and the separated frequency band subsequence normalization information is used as the separated frequency band subsequence evolution graph information, so that the obtained separated frequency band subsequence evolution graph information becomes a scalar, which is convenient for subsequent analysis and processing of the separated frequency band subsequence evolution graph information.

[0038] Optionally, analyzing and processing the separated frequency band embedded vector information according to a preset embedded vector aggregation method to form separated frequency band aggregated feature information includes:

[0039] According to the audio separation frequency band information and the final separation frequency band selection information, a distance value between the audio separation frequency band information and the final separation frequency band selection information is analyzed and obtained as the separation frequency band distance value;

[0040] Analyzing and processing the separation frequency band distance values ​​according to a preset fitting result confirmation method to form separation frequency band fitting result information;

[0041] Analyze and obtain the separated frequency band aggregation information corresponding to the separated frequency band fitting result information according to the correspondence between the separated frequency band fitting result information and the preset separated frequency band aggregation information;

[0042] According to the correspondence between the final selection information of the separated frequency band and the preset subsequence loss value, analyzing and obtaining the subsequence loss value corresponding to the final selection information of the separated frequency band;

[0043] Analyzing and processing the subsequence loss value and the separated frequency band embedded vector information according to a preset separated frequency band weight value analysis method to form a separated frequency band weight value;

[0044] According to the correspondence between the separation frequency band weight value, the separation frequency band embedded vector information and the preset embedded vector aggregate feature information, the embedded vector aggregate feature information corresponding to the separation frequency band weight value and the separation frequency band embedded vector information is analyzed and obtained, and the embedded vector aggregate feature information is used as the separation frequency band aggregate feature information.

[0045] By adopting the above technical solution, the distance value between the audio separation frequency band information and the separation frequency band final selection information is analyzed and obtained and the distance value is used as the separation frequency band distance value, the separation frequency band distance value is analyzed and processed by the fitting result confirmation method to form the separation frequency band fitting result information, the separation frequency band aggregation information is obtained by analyzing the separation frequency band fitting result information, the subsequence loss value is obtained by analyzing the separation frequency band final selection information, the subsequence loss value and the separation frequency band embedded vector information are analyzed and processed by the separation frequency band weight value analysis method to form the separation frequency band weight value, the embedded vector aggregation feature information is obtained by analyzing the separation frequency band weight value and the separation frequency band embedded vector information, and the embedded vector aggregation feature information is used as the separation frequency band aggregation feature information, thereby improving the accuracy of the obtained separation frequency band aggregation feature information.

[0046] In a second aspect, the present application provides a water pipe leakage detection model training system, which adopts the following technical solutions:

[0047] A water pipe leak detection model training system, comprising:

[0048] An acquisition module is used to obtain historical audio data information of the water supply pipeline and current audio data information of the water supply pipeline;

[0049] A memory for storing a program of a water pipe leakage detection model training method according to any one of the first aspects;

[0050] The program in the processor memory can be loaded and executed by the processor to implement a water pipe leakage detection model training method as described in any one of the first aspects.

[0051] By adopting the above technical solution, the historical audio data information and current audio data information of the water supply pipeline are acquired through the acquisition module, and then the program in the memory is loaded and executed by the processor, thereby facilitating real-time leakage detection of the water supply pipeline.

[0052] In a third aspect, the present application provides a water pipe leakage detection model training device, which adopts the following technical solution: a water pipe leakage detection model training device, comprising a memory and a processor, the memory storing a computer program that can be loaded by the processor and execute a water pipe leakage detection model training method such as any one of the first aspects.

[0053] By adopting the above technical solution, the processor loads and executes the program in the memory, thereby facilitating real-time leakage detection of the water supply pipeline.

[0054] In a fourth aspect, the present application provides a computer storage medium capable of storing corresponding programs, which has the characteristics of facilitating the convenient and real-time leakage detection of water supply pipelines, and adopts the following technical solutions:

[0055] A computer storage medium stores a computer program capable of being loaded by a processor and executing a water pipe leakage detection model training method as described in any one of the first aspects.

[0056] By adopting the above technical solution, the program is stored and loaded and executed when needed, thereby facilitating real-time leakage detection of the water supply pipeline.

[0057] In summary, this application includes at least one of the following beneficial technical effects:

[0058] 1. By acquiring historical audio data information of the water supply pipeline, analyzing and processing the historical audio data information of the water supply pipeline through the frequency band separation method to form the historical audio frequency band data information of the water pipe, and then analyzing and processing the historical audio frequency band data information of the water pipe through the audio detection data model building method to form a water pipe leakage detection model, and then using the water pipe leakage detection model to facilitate real-time leakage detection of the water supply pipeline;

[0059] 2. Retrieve the maximum frequency value of the data through the historical audio data information of the water supply pipeline, and obtain the frequency interval length value of the frequency band through analysis of the maximum frequency value of the data and the number of audio separation frequency bands. Then, obtain the audio separation frequency band information through analysis of the historical audio data information of the water supply pipeline, the number of audio separation frequency bands and the number of frequency interval lengths of the frequency bands, and use the audio separation frequency band information as the historical audio frequency band data information of the water pipe, thereby improving the accuracy of the obtained audio separation frequency band information.

[0060] 3. The historical audio frequency band data information of the water pipe is analyzed and processed by the separated frequency band selection method to form the final selected information of the separated frequency band. The final selected information of the separated frequency band is analyzed and processed by the separated frequency band weighted analysis method to form the separated frequency band weighted information. The final selected information of the separated frequency band and the separated frequency band weighted information are analyzed and processed by the weighted aggregation training method to form a weighted aggregation training model. The weighted aggregation training model is used as the water pipe leakage detection model, thereby improving the accuracy of the obtained water pipe leakage detection model. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a flow chart of the method for training a water pipe leakage detection model according to an embodiment of the present application.

[0062] Figure 2 This is a flow chart of a method for analyzing and processing historical audio data information of a water supply pipeline according to a preset frequency band separation method to form historical audio frequency band data information of the water pipeline in an embodiment of the present application.

[0063] Figure 3 This is a flow chart of a method for analyzing and processing historical audio frequency band data information of water pipes to form a water pipe leakage detection model based on a preset audio detection data model building method in an embodiment of the present application.

[0064] Figure 4 This is a flow chart of a method for analyzing and processing historical audio frequency band data information of a water pipe according to a preset separation frequency band selection method to form final separation frequency band selection information in an embodiment of the present application.

[0065] Figure 5 This is a flow chart of a method for analyzing and processing the final selection information of the separated frequency bands according to the preset separated frequency band weighted analysis method in an embodiment of the present application to form separated frequency band weighted information.

[0066] Figure 6 This is a flow chart of a method for analyzing and processing the final selection information of the separated frequency bands according to the preset subsequence evolution graph confirmation method to form the separated frequency band subsequence evolution graph information in an embodiment of the present application.

[0067] Figure 7 This is a flow chart of a method for analyzing and processing separated frequency band embedded vector information according to a preset embedded vector aggregation method in an embodiment of the present application to form separated frequency band aggregated feature information.

[0068] Figure 8 This is a system flow chart of the water pipe leakage detection model training in an embodiment of the present application.

[0069] Explanation of the accompanying drawings: 1. Acquisition module; 2. Memory; 3. Processor. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solutions and advantages of this application more clear, the following Figure 1-8 It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0071] The embodiment of the present application discloses a water pipe leakage detection model training method.

[0072] Reference Figure 1 , a water pipe leak detection model training method includes:

[0073] Step S100: Acquire historical audio data information of the water supply pipeline.

[0074] The historical audio data information of the water supply pipeline refers to the historical audio data information stored after audio detection of the water supply pipeline, and the historical audio data information of the water supply pipeline is retrieved from a database storing the historical audio data information of the water supply pipeline.

[0075] Step S200 : Analyze and process the historical audio data of the water supply pipeline according to a preset frequency band separation method to form historical audio frequency band data of the water pipeline.

[0076] The frequency band separation method refers to a method for splitting the historical audio data of the water supply pipeline into frequency bands. The frequency band separation method is retrieved from a database storing frequency band separation methods. The historical water pipe audio frequency band data information refers to the audio information obtained by splitting the historical audio data of the water supply pipeline into frequency bands.

[0077] The historical audio data information of the water supply pipeline is analyzed and processed by the frequency band separation method to form the historical audio frequency band data information of the water pipe, which facilitates the subsequent use of the historical audio frequency band data information of the water pipe.

[0078] Step S300 : Analyze and process historical audio frequency band data of water pipes according to a preset audio detection data model building method to form a water pipe leakage detection model.

[0079] The audio detection data model building method is a method for building a water pipe leak detection model. The audio detection data model building method is retrieved from a database storing audio detection data model building methods. The water pipe leak detection model is a model used to detect leaks in water supply pipes.

[0080] The historical audio frequency band data information of the water pipe is analyzed and processed through the audio detection data model construction method, thereby forming a water pipe leakage detection model, which is convenient for using the water pipe leakage detection model to detect leakage of the water supply pipeline in real time.

[0081] exist Figure 1 In step S200, in order to further ensure the rationality of the historical audio frequency band data information of the water pipe, it is necessary to further analyze and calculate the historical audio frequency band data information of the water pipe. Figure 2 The steps shown are explained in detail.

[0082] Reference Figure 2 , analyzing and processing the historical audio data information of the water supply pipeline according to the preset frequency band separation method to form the historical audio frequency band data information of the water pipe includes the following steps:

[0083] Step S210: retrieve the maximum frequency value of the data corresponding to the historical audio data of the water supply pipeline according to the historical audio data of the water supply pipeline.

[0084] The maximum frequency value of the data refers to the frequency value of the maximum frequency in the historical audio data information of the water supply pipeline, and the maximum frequency value of the data is obtained by querying a database storing the maximum frequency value of the data.

[0085] The maximum frequency value of the data is retrieved through the historical audio data information of the water supply pipeline, so as to facilitate the subsequent use of the maximum frequency value of the data.

[0086] Step S220 , according to the correspondence between the maximum frequency value of the data, the preset number of audio separation frequency bands and the preset frequency band frequency interval length value, analyze and obtain the frequency band frequency interval length value corresponding to the maximum frequency value of the data and the preset number of audio separation frequency bands.

[0087] The number of audio separation frequency bands refers to the number of frequency bands used to split the historical audio data of the water supply pipeline. This number is retrieved from a database storing the number of audio separation frequency bands. The number of frequency band frequency interval lengths refers to the length of the frequency intervals in each frequency band after the historical audio data of the water supply pipeline is split. This number is retrieved from a database storing the number of frequency band frequency interval lengths.

[0088] The frequency interval length value of the frequency band is obtained by analyzing the maximum frequency value of the data and the preset audio separation frequency band values, so as to facilitate the subsequent use of the frequency interval length value of the frequency band.

[0089] Step S230, based on the correspondence between the historical audio data information of the water supply pipeline, the number of audio separation frequency bands, the frequency band frequency interval length value and the preset audio separation frequency band information, analyze and obtain the audio separation frequency band information corresponding to the historical audio data information of the water supply pipeline, the number of audio separation frequency bands and the frequency band frequency interval length value, and use the audio separation frequency band information as the historical audio frequency band data information of the water pipe.

[0090] The audio separation frequency band information refers to information of each separated frequency band after the historical audio data of the water supply pipeline is split into frequency bands. The audio separation frequency band information is obtained by querying a database storing the audio separation frequency band information.

[0091] Audio separation frequency band information is obtained by analyzing the historical audio data information of the water supply pipeline, the number of audio separation frequency bands and the length value of the frequency band frequency interval, and the audio separation frequency band information is used as the historical audio frequency band data information of the water pipe, thereby improving the accuracy of the obtained historical audio frequency band data information of the water pipe.

[0092] For example, the historical audio data of the water supply pipeline is divided into n equally divided frequency bands from low to high, that is, when the value of the audio separation frequency band is n, the relationship between the historical audio data of the water supply pipeline and the split frequency bands is as follows: original data→{data1,data2,data3,…,data n-1 ,data n};

[0093] At this time, the first audio separation frequency band interval is The second audio separation frequency band is Separate the frequency bands by frequency, and the frequency band interval of the i-th audio separation is:

[0094]

[0095] In this embodiment, by setting filters in different frequency ranges, components in different frequency ranges in the original audio signal are separated, thereby facilitating subsequent processing and analysis.

[0096] exist Figure 1 In step S300, in order to further ensure the rationality of the water pipe leakage detection model, it is necessary to further analyze and calculate the water pipe leakage detection model. Figure 3 The steps shown are explained in detail.

[0097] Reference Figure 3 , according to the preset audio detection data model building method, analyzing and processing the historical audio frequency band data information of the water pipe to form a water pipe leakage detection model includes the following steps:

[0098] Step S310 : Analyze and process the historical audio frequency band data of the water pipe according to a preset separation frequency band selection method to generate final separation frequency band selection information.

[0099] The separated frequency band selection method refers to a method for selecting separated frequency bands after splitting the historical audio data of the water supply pipeline. The separated frequency band selection method is retrieved from a database storing separated frequency band selection methods. The final separated frequency band selection information refers to the frequency band information of the final selected separated frequency band.

[0100] The historical audio frequency band data information of the water pipe is analyzed and processed by the separation frequency band selection method to form the final selection information of the separation frequency band, which is convenient for subsequent use of the final selection information of the separation frequency band.

[0101] In step S320 , the final separation frequency band selection information is analyzed and processed according to a preset separation frequency band weighted analysis method to form separation frequency band weighted information.

[0102] The separated frequency band weighted analysis method refers to weighted information for weighting the finally selected separated frequency bands, and the separated frequency band weighted analysis method is retrieved from a database storing the separated frequency band weighted analysis method.

[0103] The final selection information of the separated frequency bands is analyzed and processed by the separated frequency band weighted analysis method, thereby forming separated frequency band weighted information, thereby facilitating subsequent use of the separated frequency band weighted information.

[0104] In step S330 , the final selection information of the separated frequency bands and the weighted information of the separated frequency bands are analyzed and processed according to a preset weighted aggregation training method to form a weighted aggregation training model, and the weighted aggregation training model is used as a water pipe leakage detection model.

[0105] The weighted aggregation training method refers to a training method for performing aggregation training on weighted separated frequency bands, and the weighted aggregation training method is retrieved from a database storing weighted aggregation training methods. The weighted aggregation training model refers to a training model for performing aggregation training on weighted separated frequency bands.

[0106] The final selection information of the separated frequency bands and the weighted information of the separated frequency bands are analyzed and processed through the weighted aggregation training method to form a weighted aggregation training model, which is then used as a water pipe leakage detection model to improve the accuracy of the obtained water pipe leakage detection model.

[0107] exist Figure 3 In step S310, in order to further ensure the rationality of the final selection information of the separated frequency bands, it is necessary to further analyze and calculate the final selection information of the separated frequency bands. Figure 4 The steps shown are explained in detail.

[0108] Reference Figure 4 , analyzing and processing the historical audio frequency band data of the water pipe according to the preset separation frequency band selection method to form the final separation frequency band selection information includes the following steps:

[0109] Step S311 : analyzing and processing the audio separation frequency band information according to a preset greedy strategy selection method to form separation frequency band initial subsequence candidate information.

[0110] The greedy strategy selection method refers to a method for selecting a time-aware subsequence from the audio separation frequency band information of the current time period using a greedy strategy. The greedy strategy selection method retrieves information from a database storing greedy strategy selection methods. The separation frequency band initial subsequence candidate information refers to the initially selected time-aware initial subsequence information.

[0111] The audio separation frequency band information is analyzed and processed by a greedy strategy selection method, thereby forming separation frequency band initial subsequence candidate information, thereby facilitating subsequent use of the separation frequency band initial subsequence candidate information.

[0112] Step S312 : analyzing and processing the separated frequency band initial subsequence candidate information according to a preset timing factor analysis method to form separated frequency band timing subsequence candidate information.

[0113] The timing factor analysis method refers to a method for performing timing factor analysis on initial subsequence candidate information of a separated frequency band, and the timing factor analysis method is retrieved from a database storing timing factor analysis methods. The separated frequency band timing subsequence candidate information refers to the subsequence candidate information obtained after performing timing factor analysis on the separated frequency band.

[0114] The candidate information of the initial subsequence of the separated frequency band is analyzed and processed by the time series factor analysis method, so that each initial selected initial subsequence with time perception ability learns its corresponding time factor w i and u i Based on the subsequence candidate set selected from all segments and a set of labeled time series T, for each candidate v, there is the following objective function:

[0115]

[0116] is the loss function, including -g(S pos (v,T),S neg (v,T)) and λ‖w‖+∈‖u‖, where S * (v,T) represents the value relative to a specific group T * The function g accepts two finite sets as input and returns a scalar value indicating how far apart the two sets are. This can be information gain or some set dissimilarity measure, namely, the KL divergence. λ‖w‖+∈‖u‖, where w is the local time factor, u is the global time factor, and λ and ∈ are preset hyperparameters. This forms candidate information for separated frequency band time series subsequences, facilitating subsequent use of the candidate information.

[0117] In step S313 , the separated frequency band timing subsequence candidate information is analyzed and processed according to a preset timing factor selection method to generate separated frequency band timing subsequence selection information, and the separated frequency band timing subsequence selection information is used as the separated frequency band final selection information.

[0118] The timing factor selection method refers to the separated frequency band information used to further select the separated frequency band timing subsequence candidate information, and the timing factor selection method is obtained by querying from a database storing the timing factor selection method.

[0119] The separation frequency band timing subsequence candidate information is analyzed and processed by the timing factor selection method to form the separation frequency band timing subsequence selection information, and the separation frequency band timing subsequence selection information is used as the separation frequency band final selection information, thereby improving the accuracy of the obtained separation frequency band final selection information.

[0120] exist Figure 3 In step S320, in order to further ensure the rationality of the weighted information of the separated frequency bands, it is necessary to further analyze and calculate the weighted information of the separated frequency bands. Figure 5 The steps shown are explained in detail.

[0121] Reference Figure 5 , analyzing and processing the final selection information of the separated frequency bands according to the preset separated frequency band weighted analysis method to form the separated frequency band weighted information includes the following steps:

[0122] Step S321 : analyzing and processing the final selection information of the separated frequency bands according to a preset subsequence evolution graph confirmation method to form separated frequency band subsequence evolution graph information.

[0123] Among them, the separated frequency band subsequence evolution graph information refers to the evolution graph information used to evolve the subsequence of the separated frequency band, the subsequence evolution graph confirmation method refers to the method used to confirm the separated frequency band subsequence evolution graph information, and the subsequence evolution graph confirmation method is queried and obtained from a database storing the subsequence evolution graph confirmation method.

[0124] The final selection information of the separated frequency bands is analyzed and processed by the subsequence evolution graph confirmation method, thereby forming the separated frequency band subsequence evolution graph information, which is convenient for subsequent use of the separated frequency band subsequence evolution graph information.

[0125] Step S322 : Analyze and process the separated frequency band subsequence evolution graph information according to a preset node label classification method to form separated frequency band subsequence label information.

[0126] The separated frequency band subsequence label information refers to the label information of the nodes in the subsequence evolution graph, the node label classification method refers to the classification method used to classify the labels of the nodes in the subsequence evolution graph, and the node label classification method is obtained by querying a database storing the node label classification method.

[0127] The separated frequency band subsequence evolution graph information is analyzed and processed by a node label classification method, thereby forming separated frequency band subsequence label information, which is convenient for subsequent use of the separated frequency band subsequence label information.

[0128] Step S323 : Analyze and process the separated frequency band subsequence label information and the preset network intermediate layer output vector information according to a preset embedding vector analysis method to form separated frequency band embedding vector information.

[0129] The separated frequency band embedding vector information refers to the separated frequency band information after vector embedding, the network intermediate layer output vector information refers to the output vector information used for vector embedding of the separated frequency band, and the network intermediate layer output vector information is retrieved from a database storing network intermediate layer output vector information. The embedded vector analysis method refers to an analysis method used to embed and analyze the vector in the separated frequency band, and the embedded vector analysis method is retrieved from a database storing embedded vector analysis methods.

[0130] The separated frequency band subsequence label information and the preset network intermediate layer output vector information are analyzed and processed by the embedded vector analysis method, thereby forming the separated frequency band embedded vector information, which is convenient for the subsequent use of the separated frequency band embedded vector information.

[0131] Step S324 : analyzing and processing the separated frequency band embedded vector information according to a preset embedded vector aggregation method to form separated frequency band aggregated feature information.

[0132] The embedded vector aggregation method refers to a method for aggregating the embedded vector information of the separated frequency bands, and the embedded vector aggregation method is retrieved from a database storing the embedded vector aggregation method. The separated frequency band aggregated feature information refers to feature information after the separated frequency band embedded vector information is aggregated.

[0133] The embedded vector information of the separated frequency bands is analyzed and processed by the embedded vector aggregation method, thereby forming the separated frequency band aggregation feature information, which is convenient for subsequent use of the separated frequency band aggregation feature information.

[0134] Step S325, retrieve statistical feature information and Mel-cepstral coefficient feature information corresponding to the historical audio data information of the water supply pipeline according to the historical audio data information of the water supply pipeline, and use the separated frequency band aggregation feature information, statistical feature information and Mel-cepstral coefficient feature information as separated frequency band weighted information.

[0135] The statistical feature information includes the mean information of the time series in the separated frequency bands, the standard deviation information of the time series in the separated frequency bands, the mean information of the time periods in the separated frequency bands, the standard deviation information of the time periods in the separated frequency bands, the standard deviation information of the mean of the time periods in the separated frequency bands, and the mean information of the standard deviation of the time periods in the separated frequency bands. The statistical feature information is retrieved from a database storing statistical feature information. The Mel-frequency cepstral coefficient feature information refers to a one-dimensional feature generated by averaging the MFCC features extracted from the separated frequency bands for each segment. The Mel-frequency cepstral coefficient feature information is retrieved from a database storing Mel-frequency cepstral coefficient feature information.

[0136] The statistical feature information and Mel-cepstral coefficient feature information are retrieved through the historical audio data information of the water supply pipeline, and the separated frequency band aggregation feature information, statistical feature information and Mel-cepstral coefficient feature information are used as the separated frequency band weighted information, thereby improving the accuracy of the obtained separated frequency band weighted information.

[0137] exist Figure 5 In step S321, in order to further ensure the rationality of the evolution graph information of the separated frequency band subsequences, it is necessary to further analyze and calculate the evolution graph information of the separated frequency band subsequences. Specifically, Figure 6 The steps shown are explained in detail.

[0138] Reference Figure 6 , analyzing and processing the final selection information of the separated frequency band according to the preset subsequence evolution graph confirmation method to form the separated frequency band subsequence evolution graph information includes the following steps:

[0139] Step S3211 : analyzing and processing the final selection information of the separated frequency bands according to a preset adjacent frequency band selection method to form adjacent frequency band information.

[0140] The adjacent frequency band selection method refers to a selection method for selecting adjacent frequency bands of the separated frequency bands, and the adjacent frequency band selection method is retrieved from a database storing adjacent frequency band selection methods. The adjacent frequency band information refers to adjacent frequency band information of the separated frequency bands.

[0141] The final selected information of the separated frequency bands is analyzed and processed by the adjacent frequency band selection method, thereby forming adjacent frequency band information, which is convenient for subsequent use of the adjacent frequency band information.

[0142] Step S3212: Analyze and obtain the frequency band association probability value corresponding to the final separation frequency band selection information and the adjacent frequency band information according to the correspondence between the final separation frequency band selection information, the adjacent frequency band information and the preset frequency band association probability value.

[0143] The frequency band association probability value refers to the probability value of another subsequence in the same time series after the subsequence finally selected for separating the frequency bands appears. The frequency band association probability value is obtained by querying from a database storing frequency band association probability values.

[0144] The frequency band association probability value is obtained by analyzing the final selection information of the separated frequency bands and the information of the adjacent frequency bands. In order to measure the rationality of the allocation, we will allocate the probability p i,j Normalized to:

[0145]

[0146] In the above formula, p i,j represents the probability value of the jth assigned subsequence in the i-th segment of the time-aware subsequence in the separated frequency band, s i Denotes the i-th segment in the time-aware subsequence of the separated frequency bands, which we will assign to segment s i Those time subsequences of i,* , v i,j It is i The jth distribution of is the distance from the i-th segment to the j-th subsequence assigned to it. It represents the maximum distance between the i-th segment and all subsequences assigned to it. Similarly, It represents the minimum distance between the i-th segment and all subsequences assigned to it, and is obtained by analyzing the frequency band association probability value, so as to facilitate the subsequent use of the frequency band association probability value.

[0147] In step S3213, the final separation frequency band selection information and the frequency band association probability value are analyzed and processed according to a preset normalization analysis method to form separation frequency band subsequence normalization information, and the separation frequency band subsequence normalization information is used as separation frequency band subsequence evolution graph information.

[0148] The normalized analysis method refers to a method for performing normalized analysis on the final selection information of the separated frequency bands and the frequency band associated probability values. The normalized analysis method is retrieved from a database storing the normalized analysis method.

[0149] The final selection information of the separated frequency bands and the frequency band association probability values ​​are analyzed and processed by the normalization analysis method to form the normalized information of the separated frequency band subsequences, and the normalized information of the separated frequency band subsequences is used as the separated frequency band subsequence evolution graph information, thereby improving the accuracy of the obtained separated frequency band subsequence evolution graph information.

[0150] exist Figure 5In step S324, in order to further ensure the rationality of the separated frequency band aggregation feature information, it is necessary to further analyze and calculate the separated frequency band aggregation feature information. Figure 7 The steps shown are explained in detail.

[0151] Reference Figure 7 , analyzing and processing the separated frequency band embedded vector information according to the preset embedded vector aggregation method to form separated frequency band aggregated feature information includes the following steps:

[0152] Step S3241 : Analyze and obtain a distance value between the audio separation frequency band information and the separation frequency band final selection information based on the audio separation frequency band information and use it as the separation frequency band distance value.

[0153] The separation frequency band distance value refers to the distance value between the finally selected separation frequency band and the subsequence of the corresponding frequency band.

[0154] Through the audio separation frequency band information and the separation frequency band final selection information, the distance value between the audio separation frequency band information and the separation frequency band final selection information is analyzed and obtained, and the distance value between the audio separation frequency band information and the separation frequency band final selection information is used as the separation frequency band distance value, so as to facilitate the subsequent use of the separation frequency band distance value.

[0155] Step S3242 : analyzing and processing the separation frequency band distance values ​​according to a preset fitting result confirmation method to generate separation frequency band fitting result information.

[0156] The fitting result confirmation method refers to a fitting result confirmation method after fitting the separation frequency band distance value, and the fitting result confirmation method is retrieved from a database storing the fitting result confirmation method. The separation frequency band fitting result information refers to fitting result information after fitting the separation frequency band.

[0157] The separation frequency band distance values ​​are analyzed and processed by the fitting result confirmation method, thereby forming the separation frequency band fitting result information, which is convenient for subsequent use of the separation frequency band fitting result information.

[0158] Step S3243 : Analyze and obtain the separated frequency band aggregation information corresponding to the separated frequency band fitting result information according to the correspondence between the separated frequency band fitting result information and the preset separated frequency band aggregation information.

[0159] The separated frequency band aggregation information refers to the aggregation information of the corresponding subsequence embedding vector converted from the fitting result, and the separated frequency band aggregation information is obtained by querying from a database storing the separated frequency band aggregation information.

[0160] The separated frequency bands aggregation information is obtained by analyzing the separated frequency bands fitting result information, so as to facilitate the subsequent use of the separated frequency bands aggregation information.

[0161] Step S3244 : Analyze and obtain the subsequence loss value corresponding to the final selection information of the separated frequency band according to the corresponding relationship between the final selection information of the separated frequency band and the preset subsequence loss value.

[0162] The subsequence loss value refers to the loss value generated after the subsequences of the separated frequency bands are aggregated, and the subsequence loss value is obtained by querying from a database storing the subsequence loss values.

[0163] By separating the frequency bands, the information is finally selected for analysis to obtain the subsequence loss value, thereby facilitating the subsequent use of the subsequence loss value.

[0164] In step S3245 , the subsequence loss value and the separated frequency band embedding vector information are analyzed and processed according to a preset separated frequency band weight value analysis method to form a separated frequency band weight value.

[0165] The separated frequency band weight value analysis method refers to a method for analyzing the weight value when the separated frequency bands are aggregated, and the separated frequency band weight value analysis method is obtained by querying from a database storing the separated frequency band weight value analysis method.

[0166] The subsequence loss values ​​and the separated frequency band embedding vector information are analyzed and processed by the separated frequency band weight value analysis method. According to the discrimination degree of the subsequence set in the training set represented by the loss value of the subsequence set in different frequency bands, the feature vectors obtained on these frequency bands are arranged in ascending order of discrimination degree, and their weights are assigned according to the preset weight value calculation formula:

[0167] Among them, w i is the frequency band weight value, k is the frequency band number, i is the frequency band serial number after rearrangement, and the separation frequency band weight value is calculated by the weight value calculation formula to form the separation frequency band weight value, which is convenient for subsequent use of the separation frequency band weight value.

[0168] Step S3246, according to the correspondence between the separation frequency band weight value, the separation frequency band embedded vector information and the preset embedded vector aggregate feature information, analyze and obtain the embedded vector aggregate feature information corresponding to the separation frequency band weight value and the separation frequency band embedded vector information, and use the embedded vector aggregate feature information as the separation frequency band aggregate feature information.

[0169] The embedded vector aggregate feature information refers to feature information obtained by fitting and aggregating the time series subsequences of each separated frequency band, and the embedded vector aggregate feature information is obtained by querying from a database storing the embedded vector aggregate feature information.

[0170] The embedded vector aggregate feature information is obtained by analyzing the separated frequency band weight value and the separated frequency band embedded vector information, and the embedded vector aggregate feature information is used as the separated frequency band aggregate feature information, thereby improving the accuracy of the obtained separated frequency band aggregate feature information.

[0171] Reference Figure 8 Based on the same inventive concept, an embodiment of the present invention provides a water pipe leakage detection model training system, comprising:

[0172] Acquisition module 1, used to obtain historical audio data information and current audio data information of water supply pipeline;

[0173] Memory 2, used to store Figures 1 to 7 A procedure for a water pipe leak detection model training method according to any one of the preceding claims;

[0174] Processor 3, the program in the memory can be loaded and executed by the processor and realize the following Figures 1 to 7 A water pipe leak detection model training method according to any one of the above.

[0175] Based on the same inventive concept, an embodiment of the present invention provides a water pipe leakage detection model training device, including a memory and a processor, the memory stores data that can be loaded and executed by the processor. Figures 1 to 7 A computer program for a water pipe leak detection model training method according to any one of the preceding claims.

[0176] Those skilled in the art will clearly understand that for the sake of convenience and brevity, the division of the above-mentioned functional modules is only used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working processes of the above-mentioned systems, devices, and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0177] An embodiment of the present invention provides a computer storage medium storing data that can be loaded and executed by a processor. Figures 1 to 7 A computer program for a water pipe leak detection model training method according to any one of the preceding claims.

[0178] Computer storage media include, for example, various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0179] The above are all preferred embodiments of the present application and are not intended to limit the scope of protection of this application. Unless otherwise stated, any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features. In other words, unless otherwise stated, each feature is merely an example of a series of equivalent or similar features.

Claims

1. A water pipe leakage detection model training method, characterized in that: include: Obtain historical audio data information of water supply pipelines; Analyzing and processing the historical audio data information of the water supply pipeline according to a preset frequency band separation method to form historical audio frequency band data information of the water pipeline; Analyze and process historical audio frequency band data of water pipes according to the preset audio detection data model building method to form a water pipe leakage detection model; Analyzing and processing the historical audio data information of the water supply pipeline according to the preset frequency band separation method to form the historical audio frequency band data information of the water pipeline includes: Retrieving the maximum frequency value of the data corresponding to the historical audio data information of the water supply pipeline according to the historical audio data information of the water supply pipeline; Analyze and obtain the frequency band frequency interval length value corresponding to the maximum frequency value of the data and the preset number of audio separation frequency bands according to the corresponding relationship between the maximum frequency value of the data, the preset number of audio separation frequency bands, and the preset frequency band frequency interval length value; According to the correspondence between the historical audio data information of the water supply pipeline, the number of audio separation frequency bands, the length of the frequency interval of the frequency bands and the preset audio separation frequency band information, the audio separation frequency band information corresponding to the historical audio data information of the water supply pipeline, the number of audio separation frequency bands and the length of the frequency interval of the frequency bands is analyzed and obtained, and the audio separation frequency band information is used as the historical audio frequency band data information of the water pipe; According to the preset audio detection data model building method, the historical audio frequency band data information of the water pipe is analyzed and processed to form a water pipe leakage detection model, including: Analyze and process the historical audio frequency band data of the water pipe according to the preset separation frequency band selection method to form the final separation frequency band selection information; Analyzing and processing the final selected information of the separated frequency bands according to a preset separated frequency band weighted analysis method to form separated frequency band weighted information; According to a preset weighted aggregation training method, the final selection information of the separated frequency bands and the weighted information of the separated frequency bands are analyzed and processed to form a weighted aggregation training model, and the weighted aggregation training model is used as a water pipe leakage detection model; The water pipe historical audio frequency band data information is analyzed and processed according to the preset separation frequency band selection method to form the final separation frequency band selection information including: Analyzing and processing the audio separation frequency band information according to a preset greedy strategy selection method to form separation frequency band initial subsequence candidate information; Analyzing and processing the separated frequency band initial subsequence candidate information according to a preset timing factor analysis method to form separated frequency band timing subsequence candidate information; The separation frequency band timing subsequence candidate information is analyzed and processed according to a preset timing factor selection method to form separation frequency band timing subsequence selection information, and the separation frequency band timing subsequence selection information is used as the separation frequency band final selection information.

2. A water pipe leakage detection model training method according to claim 1, characterized in that: Analyzing and processing the final selection information of the separated frequency bands according to the preset separated frequency band weighted analysis method to form separated frequency band weighted information includes: Analyzing and processing the final selection information of the separated frequency bands according to a preset subsequence evolution graph confirmation method to form separated frequency band subsequence evolution graph information; Analyzing and processing the separated frequency band subsequence evolution graph information according to a preset node label classification method to form separated frequency band subsequence label information; Analyzing and processing the separated frequency band subsequence label information and the preset network intermediate layer output vector information according to a preset embedding vector analysis method to form separated frequency band embedding vector information; Analyzing and processing the separated frequency band embedded vector information according to a preset embedded vector aggregation method to form separated frequency band aggregated feature information; According to the historical audio data information of the water supply pipeline, the statistical feature information and Mel-cepstral coefficient feature information corresponding to the historical audio data information of the water supply pipeline are retrieved, and the separated frequency band aggregation feature information, statistical feature information and Mel-cepstral coefficient feature information are used as the separated frequency band weighted information.

3. A water pipe leakage detection model training method according to claim 2, characterized in that: Analyzing and processing the final selection information of the separated frequency bands according to the preset subsequence evolution graph confirmation method to form the separated frequency band subsequence evolution graph information includes: Analyzing and processing the final selection information of the separated frequency bands according to a preset adjacent frequency band selection method to form adjacent frequency band information; According to the correspondence between the final selection information of the separated frequency band, the adjacent frequency band information and the preset frequency band association probability value, analyzing and obtaining the frequency band association probability value corresponding to the final selection information of the separated frequency band and the adjacent frequency band information; According to the preset normalization analysis method, the final selection information of the separated frequency bands and the frequency band associated probability values ​​are analyzed and processed to form the separated frequency band subsequence normalization information, and the separated frequency band subsequence normalization information is used as the separated frequency band subsequence evolution graph information.

4. A water pipe leakage detection model training method according to claim 2, characterized in that: Analyzing and processing the separated frequency band embedded vector information according to the preset embedded vector aggregation method to form separated frequency band aggregated feature information includes: According to the audio separation frequency band information and the final separation frequency band selection information, a distance value between the audio separation frequency band information and the final separation frequency band selection information is analyzed and obtained as the separation frequency band distance value; Analyzing and processing the separation frequency band distance values ​​according to a preset fitting result confirmation method to form separation frequency band fitting result information; Analyze and obtain the separated frequency band aggregation information corresponding to the separated frequency band fitting result information according to the correspondence between the separated frequency band fitting result information and the preset separated frequency band aggregation information; According to the correspondence between the final selection information of the separated frequency band and the preset subsequence loss value, analyzing and obtaining the subsequence loss value corresponding to the final selection information of the separated frequency band; Analyzing and processing the subsequence loss value and the separated frequency band embedded vector information according to a preset separated frequency band weight value analysis method to form a separated frequency band weight value; According to the correspondence between the separation frequency band weight value, the separation frequency band embedded vector information and the preset embedded vector aggregate feature information, the embedded vector aggregate feature information corresponding to the separation frequency band weight value and the separation frequency band embedded vector information is analyzed and obtained, and the embedded vector aggregate feature information is used as the separation frequency band aggregate feature information.

5. A water pipe leakage detection model training system, characterized in that: include: An acquisition module (1) is used to acquire historical audio data information of a water supply pipeline and current audio data information of a water supply pipeline; A memory (2) for storing a program of a water pipe leakage detection model training method according to any one of claims 1 to 4; A processor (3), wherein the program in the memory can be loaded and executed by the processor and implements a water pipe leakage detection model training method as claimed in any one of claims 1 to 4.

6. A water pipe leakage detection model training device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes a water pipe leakage detection model training method according to any one of claims 1 to 4.

7. A computer storage medium, characterized in that The device stores a computer program that can be loaded by a processor and executes a water pipe leakage detection model training method according to any one of claims 1 to 4.

Images