WO2019019709A1

WO2019019709A1 - Method for detecting water leakage of tap water pipe

Info

Publication number: WO2019019709A1
Application number: PCT/CN2018/083481
Authority: WO
Inventors: 刘晓葳; 肖龙源; 李稀敏; 蔡振华
Original assignee: 厦门快商通科技股份有限公司
Priority date: 2017-07-24
Filing date: 2018-04-18
Publication date: 2019-01-31
Also published as: CN107480705B; JP6872077B2; CN107480705A; JP2020519909A

Abstract

Disclosed is a method for detecting water leakage of a tap water pipe, comprising the following steps: S1, acquiring marked leakage/no leakage sample sound wave data; S2, generating a training feature library according to sample data, using an integration model to train the training feature library, and generating a water leakage classifier; S3, generating a discrimination feature library according to detection data; S4, placing the discrimination feature library in the water leakage classifier to carry out measurement and calculation, and generating point-by-point determination results for water leakage/no water leakage; and S5, classifying the point-by-point determination results according to collection points of the sound wave data, establishing time labels and position labels for discrimination arrays, generating a discrimination rule tree in conjunction with the point-by-point determination results, the time labels and the position labels, carrying out voting according to the discrimination rule tree, and generating a final determination result, wherein the final determination result comprises leakage or no leakage and a probability respectively corresponding thereto. The present invention analyzes pipe sound wave data in real time, thereby simplifying the existing data analysis mode, and improving the accuracy of data analysis, such that the work efficiency is high, and the precision can be continuously improved by means of self-learning.

Description

Tap water leakage detection method

Technical field

The invention relates to the field of data processing methods for monitoring or predicting purposes, in particular to a water pipe leak detecting method based on multi-dimensional acoustic wave feature analysis, machine learning and voting method.

Background technique

Each city is equipped with a complicated water pipe network. The water leakage problem of the water pipe has also plagued urban builders. In order to timely repair the water pipes, frequent maintenance is required. At present, the following methods are mainly used:

First, artificial hearing leakage is also the main application of pipeline maintenance method. It requires nighttime patrol line and electronic equipment to listen to pipeline sounds. The labor cost is high, and the judgment result is greatly affected by the service level of the hearing personnel. The corresponding training cost is also high,

The second is the pipeline data analysis method. With the continuous development of hardware technology, there are some methods for online analysis of water flow and sound waves of wall sensors, such as some companies collecting signal data of pipeline sound waves, according to the returned signals. The content is manually step-by-step to convert the array, decompose the signal, manually review the waveform diagram, such as interpreting the frequency domain map, finding the frequency corresponding to the peak value, and judging whether the monitored pipeline leaks water, the method specifically uses the acoustic wave feature to identify the pipeline Whether water leakage is generally used to decompose the signal into the time domain and the frequency domain by fast Fourier transform or wavelet transform, trying to obtain "the sound data of the leaky pipeline exhibits some characteristics different from the non-leakage data in the time/frequency domain" The rules are essentially classification problems or anomaly detection problems, but their efficiency and accuracy are not high.

In short, by manually leaking the method to the method or reading the waveform one by one, it is a serious drawback to judge whether the pipeline represented by this waveform belongs to the leaking or non-leakage method. In the case of a large amount of data, labor alone is used to cause duplication of labor and inefficiency. The classification lacks systematicity and consistency, and does not make full use of the data self-regulation, resulting in high human consumption costs, and it is difficult to form large-scale and timely judgments.

In the prior art, the model is discriminated by using the time series of the value of the signal or the abnormal feature of the original signal, and the method of manually adjusting the model parameters before each discrimination is essentially a semi-manual classification, and finally concludes whether or not the water leaks. For example, subjectively set thresholds and analysis lengths are analyzed one by one, and the results are summarized. However, the semi-manual methods for missing defects include: first, lack of objective unified model criteria; second, at each discriminant Still need a lot of human participation; third, subjective impact accuracy.

In view of this, the inventors have proposed a water pipe leak detection method with high work efficiency and high precision.

Summary of the invention

In order to solve the above problems, the present invention provides a water pipe leakage detection method for real-time comprehensive and intelligent analysis of pipeline acoustic wave data, detecting pipeline leakage, and simplifying existing data analysis methods, improving data analysis accuracy and work efficiency. High, precision can be continuously improved through self-learning.

In order to achieve the above object, the technical solution adopted by the present invention is:

A tap water leak detection method includes the following steps:

S1. Acquiring the sound wave data of the water pipe marked with leak/non-leakage as sample data, the sound wave data includes three fields of time, place and signal;

S2. generating a training feature library according to the sample data, and using the integrated model to train the training feature database to generate a water leakage classifier;

S3. Acquiring the sound data of the leaking/non-leakage of the water pipe as the detection data, and generating the discriminant feature library according to the detected data;

S4. The discriminant feature library is placed in the leak classifier to calculate the point-by-point judgment result of leaking/non-leakage;

S5. classifying the point-by-point judgment result according to the collection point of the sound wave data, and establishing a time label and a place label for the discriminant array, and generating a discriminant rule tree according to the point-by-point judgment result and the time label and the location label, according to the discrimination The rule tree votes to generate the final judgment result, and the final judgment result includes the leakage or non-leakage and their respective corresponding probabilities.

The step S2 specifically includes the following steps:

S21. Convert the sample data into a leak/leak training array;

S22. Calculating a statistical feature of the training array signal field, where the statistical features include a mean, a variance, and a standard deviation;

S23. Calculating a signal characteristic of the training array signal field, where the signal characteristic includes a multi-level autocorrelation function, a statistic of the autocorrelation function, and a statistic and a frequency value corresponding to the highest amplitude of the training array in the frequency domain;

S24. Constructing a statistical feature, a signal feature, and a corresponding leak/non-leakage annotation as a training feature library of the training array;

S25. Pre-training the training feature library with an integrated model integrating Logis regression, support vector machine, random forest, nearest neighbor method, and naive Bayesian classification model;

S26. Determine whether the feature importance in the training feature database satisfies a preset threshold by using the mean square error reduction amount as an index, and select a feature that meets the preset threshold to construct a water leakage classifier.

The step S3 specifically includes the following steps:

S31. Converting the detection data into a discriminant array;

S32. Calculate the statistical characteristics and signal characteristics of the discriminant array signal field. The statistical features include mean, variance, and standard deviation. The signal characteristics include its multi-order autocorrelation function, the statistics of the autocorrelation function, and the statistics of the training array on the frequency domain. The frequency value corresponding to the highest amplitude;

S33. Construct a discriminant feature library according to the characteristics of the discriminant array.

The step S5 specifically includes the following steps:

S51. Establish a corresponding time label and a place label according to the field identifying the time and the location in the array;

S52. Extracting the time label and the location label one by one according to the discriminant array;

S53. Establish a label set according to the time label and the location label, and paste a plurality of pieces of data of the same collection place with the same place label;

S54. The plurality of pieces of data having the same place label are combined into a data set, and the time stamp is set according to the time field data set, and then the data set is arranged in chronological order;

S55. Matching the point-by-point discriminating water leakage/non-leakage result to the location label and the time label, determining a unique data, and assigning the data leakage/non-leakage determination result as a point-by-point discriminating label, generating a triad {location label The discriminant node formed by the time stamp and the point-by-point discriminant label constructs a discriminant rule tree based on the discriminant node.

The step S5 further includes the following steps:

S56. Classify the triples according to the location label, and calculate a ratio of the number of the leaky labels to the total number of the triplet included in each of the same location label categories to obtain a probability discrimination. value;

Determining a magnitude relationship between the probability discriminant value and a preset threshold;

If the probability discriminant value is greater than a preset threshold, the pipeline represented by the triad classification is judged as water leakage, and the value is given as a water leakage probability;

If the probability discriminant value is less than a preset threshold, the pipeline represented by the triad classification is judged to be non-leakage, and the value is given as a non-leakage probability.

In the step S56, if the probability discriminant value is less than the preset threshold, it may also be determined according to the timing, and the specific steps are as follows:

Sorting the triples according to the location label, and searching for a triple point according to a time stamp for each triplet included in the label class of the same location;

According to the midpoint of the time, the triad is classified into a post-statistical group and a pre-group, and the ratio of the number of the leak-proof labels to the total number of the point-by-point discriminative labels included in the pre-group and the post-group is calculated, and the pre-group and the post-group are obtained. Probability discriminant value;

Calculating a ratio of the probability discriminant value of the previous group and the probability discriminant value of the latter group to obtain a timing discriminant value;

Comparing the timing discriminant value with a preset threshold;

If the timing discriminant value is greater than a preset threshold, the discriminating result is a water leakage, and the threshold value is given as a water leakage probability;

If the timing discriminant value is less than the preset threshold, the discriminating result is non-leakage, and the threshold is given as a non-leakage probability.

The sound wave data is a short time series having a length of 256.

The method further includes the step S6. determining whether the pipeline point is leaking according to the final judgment result, and repeating the steps S1-S2 after determining the water leakage, continuously constructing a new training feature database and a new water leakage classifier for the newly acquired sound wave data for self-learning. Update the water leakage classifier to determine the water leakage and output the judgment result.

After adopting the above technical solution, the beneficial effects of the present invention are: establishing a training feature library suitable for classification according to the known leaked/non-leaked sample data, training the training feature library to generate a water leakage classifier, and then detecting according to the unknown leak/leakage. The data establishes a discriminant feature library for discriminating, and the discriminant feature library is placed in the classifier. When discriminating, the point-by-point judging result can be generated, and a discriminant rule tree combining the time series information and processing the sample imbalance problem can be generated. Finally, the judgment result is output, thereby realizing the determination and alarm of the pipeline leakage condition based on the analysis of the acoustic wave data. The method is characterized by: classification for the purpose, strong scalability, strong resistance to other factors such as the environment, the leaking classifier used can update the data and reset, and it is continuously improved in the analysis practice, and the accuracy is tested. Real-time comprehensive objective and intelligent collection of comment content in text content can be realized to quickly adapt to different situations of different pipelines, and simplify existing data analysis methods and improve the accuracy of data analysis.

Detailed ways

In order to make the technical problems, technical solutions and beneficial effects of the present invention more clear and clear, the present invention will be further described in detail below with reference to the embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

A tap water leakage detecting method of the present invention comprises the following steps:

S1. Acquiring the sound wave data of the tap water pipe marked with leak/non-leakage as sample data, the sound wave data is a short time series, and the length thereof is 256; the sampling frequency of these sound wave signals is generally 10 kHz, and the conversion mode is generally binary number transfer. It is a decimal number; the data source used to construct the training feature database of the algorithm must be the acoustic wave data from the acoustic wave sensor installed on the outer wall of the pipeline, and the data should be marked by field inspection, that is, marked as leak/no leak, so It satisfies the ability to establish a classifier and discriminate new data. The data generally includes three fields: time, place and signal;

The step S2 specifically includes the following steps:

S21. Convert the sample data into a leak/leak training array;

S22. Calculating a statistical feature of the training array signal field, the statistical features including mean, variance, standard deviation, and the mean and standard deviation calculation formula are as follows;

Mean:

Where n is the length of the signal array in each sound wave data record, generally n=128, x _i represents the ith value of the signal array; standard deviation:

S23. Calculating a signal characteristic of the training array signal field, where the signal characteristic includes a multi-level self-correlation function, a statistic of the autocorrelation function, and a statistic and a frequency value corresponding to the highest amplitude of the training array in the frequency domain;

The signal characteristics of each sound wave data calculated, including: calculating the autocorrelation function sequence

Where k represents the order of the autocorrelation function sought; the sum of the sequences of autocorrelation functions

The mean of the autocorrelation function sequence:

And the standard deviation of the autocorrelation function sequence

S25. Pre-training the training feature library with an integrated classifier integrated with Logis regression, support vector machine, random forest, nearest neighbor method, and naive Bayesian classification model;

S26. Calculate the average InMSE (Increase in MSE) of each feature of the training feature database after entering the model, and determine the importance of the feature in the training feature database by using the mean square error reduction as an index. Whether the preset threshold is met, and the feature that meets the preset threshold is selected to be constructed as a water leakage classifier.

The step S3 specifically includes the following steps:

S31. Converting the detection data into a discriminant array;

S5. classifying the point-by-point judgment result according to the collection point of the sound wave data, and establishing a time label and a place label for the discriminant array, and generating a discriminant rule tree according to the point-by-point judgment result and the time label and the location label, according to the discrimination The rule tree makes a vote, and the final judgment result is generated, and the final judgment result includes a leak or a non-leakage and a corresponding probability thereof;

S6. Determine whether the pipeline point leaks according to the final judgment result, and repeat steps S1-S2 to determine the water leakage, and continuously construct a new training feature database and a new water leakage classifier for the newly acquired sound wave data to self-learn and update the water leakage classification. If it is determined that there is no water leakage, the judgment result is output.

The step S5 specifically includes the following steps:

S55. Matching the point-by-point discriminating water leakage/non-leakage result to the location label and the time label, determining a unique data, and assigning the data leakage/non-leakage determination result as a point-by-point discriminating label, wherein the point-by-point discriminating label The "0/1" label is used, including the leaky label "1" and the watertight label "0". The "0" means that it is judged as watertight, and the "1" means that it is judged to be leaking, and the triplet {place label, time stamp is generated. a discriminating node formed by a point-by-point discriminating label}, and constructing a discriminating rule tree according to the discriminating node;

S56. Classify the triad according to the location label, and calculate a ratio of the number of the leaky label "1" to the total number of the triads included in each of the same location label categories. Obtain a probability discriminant value;

According to the midpoint of the time, the triad is classified into a post-statistical group and a pre-group, and the ratio of the number of the leak-proof label "1" to the total number of the point-by-point discriminative labels included in the pre-group and the post-group is calculated, and the pre-group is obtained. And the probability value of the latter group;

Comparing the timing discriminant value with a preset threshold;

The above description shows and describes the preferred embodiments of the present invention. It is to be understood that the invention is not to be construed as being limited to the details disclosed herein. And modifications can be made by the above teachings or related art or knowledge within the scope of the inventive concept. All changes and modifications made by those skilled in the art are intended to be within the scope of the appended claims.

Claims

A tap water leakage detecting method, characterized in that the method comprises the following steps:

S1. Acquiring the sound wave data of the water pipe marked with leak/non-leakage as sample data, the sound wave data includes three fields of time, place and signal;

S2. generating a training feature library according to the sample data, and using the integrated model to train the training feature database to generate a water leakage classifier;

S3. Acquiring the sound data of the leaking/non-leakage of the water pipe as the detection data, and generating the discriminant feature library according to the detected data;

S4. The discriminant feature library is placed in the leak classifier to calculate the point-by-point judgment result of leaking/non-leakage;

S5. classifying the point-by-point judgment result according to the collection point of the sound wave data, and establishing a time label and a place label for the discriminant array, and generating a discriminant rule tree according to the point-by-point judgment result and the time label and the location label, according to the discrimination The rule tree votes to generate the final judgment result, and the final judgment result includes the leakage or non-leakage and their respective corresponding probabilities.
The method for detecting water leakage of a water pipe according to claim 1, wherein the step S2 comprises the following steps:

S21. Convert the sample data into a leak/leak training array;

S22. Calculating a statistical feature of the training array signal field, where the statistical features include a mean, a variance, and a standard deviation;

S23. Calculating a signal characteristic of the training array signal field, where the signal characteristic includes a multi-level autocorrelation function, a statistic of the autocorrelation function, and a statistic and a frequency value corresponding to the highest amplitude of the training array in the frequency domain;

S24. Constructing a statistical feature, a signal feature, and a corresponding leak/non-leakage annotation as a training feature library of the training array;

S25. Pre-training the training feature library with an integrated model integrating Logis regression, support vector machine, random forest, nearest neighbor method, and naive Bayesian classification model;

S26. Determine whether the feature importance in the training feature database satisfies a preset threshold by using the mean square error reduction amount as an index, and select a feature that meets the preset threshold to construct a water leakage classifier.
The method for detecting water leakage in a water pipe according to claim 1, wherein the step S3 comprises the following steps:

S31. Converting the detection data into a discriminant array;

S32. Calculate the statistical characteristics and signal characteristics of the discriminant array signal field. The statistical features include mean, variance, and standard deviation. The signal characteristics include its multi-order autocorrelation function, the statistics of the autocorrelation function, and the statistics of the training array on the frequency domain. The frequency value corresponding to the highest amplitude;

S33. Construct a discriminant feature library according to the characteristics of the discriminant array.
The method for detecting water leakage of a water pipe according to claim 1, wherein the step S5 comprises the following steps:

S51. Establish a corresponding time label and a place label according to the field identifying the time and the location in the array;

S52. Extracting the time label and the location label one by one according to the discriminant array;

S53. Establish a label set according to the time label and the location label, and paste a plurality of pieces of data of the same collection place with the same place label;

S54. The plurality of pieces of data having the same place label are combined into a data set, and the time stamp is set according to the time field data set, and then the data set is arranged in chronological order;

S55. Matching the point-by-point discriminating water leakage/non-leakage result to the location label and the time label, determining a unique data, and assigning the data leakage/non-leakage determination result as a point-by-point discriminating label, generating a triad {location label The discriminant node formed by the time stamp and the point-by-point discriminant label constructs a discriminant rule tree based on the discriminant node.
The method for detecting water leakage of a water pipe according to claim 4, wherein the step S5 further comprises the following steps:

S56. Classify the triples according to the location label, and calculate a ratio of the number of the leaky labels to the total number of the triplet included in each of the same location label categories to obtain a probability discrimination. value;

Determining a magnitude relationship between the probability discriminant value and a preset threshold;

If the probability discriminant value is greater than a preset threshold, the pipeline represented by the triad classification is judged as water leakage, and the value is given as a water leakage probability;

If the probability discriminant value is less than a preset threshold, the pipeline represented by the triad classification is judged to be non-leakage, and the value is given as a non-leakage probability.
The tap water leakage detecting method according to claim 5, wherein in the step S56, if the probability discriminating value is smaller than a preset threshold, the step may be determined according to the timing, and the specific steps are as follows:

Sorting the triples according to the location label, and searching for a triple point according to a time stamp for each triplet included in the label class of the same location;

According to the midpoint of the time, the triad is classified into a post-statistical group and a pre-group, and the ratio of the number of the leak-proof labels to the total number of the point-by-point discriminative labels included in the pre-group and the post-group is calculated, and the pre-group and the post-group are obtained. Probability discriminant value;

Calculating a ratio of the probability discriminant value of the previous group and the probability discriminant value of the latter group to obtain a timing discriminant value;

Comparing the timing discriminant value with a preset threshold;

If the timing discriminant value is greater than a preset threshold, the discriminating result is a water leakage, and the threshold value is given as a water leakage probability;

If the timing discriminant value is less than the preset threshold, the discriminating result is non-leakage, and the threshold is given as a non-leakage probability.
A method for detecting water leakage in a water pipe according to claim 1, wherein said sound wave data is a short time series having a length of 256.
The method for detecting water leakage of a water pipe according to any one of claims 1 to 7, further comprising the step S6. determining whether the pipeline point is leaking according to the final judgment result, and repeating the steps S1-S2 after determining the water leakage. The newly acquired sound wave data is continuously constructed into a new training feature library and a new water leakage classifier, and the water leakage classifier is updated by self-learning, and the judgment result is output when it is determined that the water is not leaking.