CN109555979B - Water supply pipe network leakage monitoring method - Google Patents

Abstract

The present disclosure provides a leakage monitoring method for a water supply pipe network, comprising the following steps: S1, acquiring water supply pipe network data; S2, establishing a leakage identification model based on a deep neural network with a heterogeneous double classifier; and S3, using the The water supply pipe network data and the heterogeneous dual-classifier leakage identification model are used to identify the leakage of the water supply pipe network. The leakage monitoring method of the water supply pipe network disclosed in the present disclosure effectively improves the calculation efficiency of pipeline leakage detection and positioning, expands the applicable scope of the pipeline leakage detection method, and can perform high-precision leakage detection and positioning under complex working conditions.

Description

A method for monitoring leakage of water supply pipe network

technical field

The disclosure belongs to the field of leakage detection of urban water supply pipe networks, and in particular relates to a method for monitoring leakage of water supply pipe networks.

Background technique

Water resources are the foundation of human survival and development. my country's total water resources are abundant, ranking fourth in the world, but the per capita water resources are only 2,300 cubic meters, which is equivalent to about 1/4 of the world's average level. Therefore, my country is also a water shortage country. The urban water supply pipe network system is an important area of urban infrastructure construction, which is called "lifeline project". However, the leakage of the water supply pipe network has always plagued major water companies in the country, which not only causes waste of resources and energy, but also causes secondary disasters such as land subsidence, which affects urban security. According to the survey data of the Ministry of Construction, the leakage rate of water supply network in most cities in my country is between 15% and 35%, while the leakage rate in developed countries such as Japan, the United States and Europe has been generally controlled at about 10% at the end of the last century. , or even a lower level, it can be seen that the control and management of pipeline network leakage in my country urgently needs to be strengthened.

At present, the leakage detection methods of water supply pipe network include passive detection method, regional metering method, surface radar leakage detection method, tracer detection method, acoustic detection method, optical fiber sensing technology method, negative pressure wave method, and real-time transient model method. Etc., many of the above-mentioned pipeline leakage detection methods, most of the methods are not ideal due to their harsh use conditions, complex operations, and high leakage detection costs. The acoustic detection method is widely used in pipeline leakage detection and location due to its advantages of simple and reliable, high detection efficiency and wide application range. However, due to insufficient understanding of the generation mechanism and characteristics of the leakage acoustic signal of the water supply pipe network, and insufficient consideration of the complex topology structure of the water supply pipe network in practice, the existing pipeline leakage detection methods are limited in practical application. The leakage identification accuracy is low, and the false alarm rate is high, especially when there are various fixed interference noises on the detection site, the leakage at the leakage point is small, or the pipeline network structure is complex, and some pipeline information is unknown.

Therefore, based on the current situation and future development trend of my country's urban water supply pipeline network, the acoustic wave transmission theory, deep neural network model, and local search and positioning model are used to detect pipeline leakage caused by structural defects of pipelines and internal and external corrosion. The technical method of identifying and locating the leakage of the water supply pipe network provides the basis and guidance for the maintenance, repair and management of the pipe network, reducing the leakage rate of the pipe network and ensuring the safe operation of the water supply pipe network.

SUMMARY OF THE INVENTION

(1) Technical problems to be solved

In view of the above problems, the main purpose of the present disclosure is to provide a leakage monitoring method for a water supply pipe network, so as to solve at least one of the above problems.

(2) Technical solutions

In order to achieve the above purpose, as an aspect of the present disclosure, a method for monitoring leakage of a water supply pipe network is provided, comprising the following steps:

S1, obtain water supply pipe network data;

S2, establish a heterogeneous dual classifier leakage recognition model based on deep neural network; and

S3, using the water supply pipe network data and the heterogeneous dual-classifier leakage identification model to identify water supply pipe network leakage; wherein the water supply pipe network data includes leakage sound signal data propagating along the water medium, water supply Pipe flow data, water supply pipe pressure data, pipe material data and pipe diameter data.

In some embodiments, after the step S3, it further includes:

S4, establishing a local search and positioning model based on the topology of the water supply pipe network; and

S5, using the water supply pipe network leakage identification result and the local search and positioning model to locate leakage.

In some embodiments, the deep neural network-based heterogeneous dual classifier leakage identification model includes a convolutional layer, a max pooling layer, a long and short-term neural network layer, a first fully connected layer, a fusion layer, a second full Connection layers, support vector machine classifiers, logistic regression classifiers, and heterogeneous dual classifiers.

In some embodiments, the step S3 includes:

The convolutional layer receives the leaked acoustic signal data;

The maximum pooling layer divides the output of the convolutional layer into m sub-regions, and extracts the maximum value of each of the sub-regions to form an output, where m is a positive integer;

The long and short-term neural network layer performs nonlinear data processing on the output of the maximum pooling layer;

The first fully connected layer receives flow data, pressure data, pipe material data and pipe diameter data;

The fusion layer receives the output of the long and short-term neural network layer and the first fully connected layer;

The second fully connected layer receives the output of the fusion layer;

The second fully connected layer is connected with the support vector machine classifier and the logistic regression classifier respectively;

The support vector machine classifier receives the output of the second fully connected layer, classifies and identifies the leakage events of the water supply pipe network, and outputs a classification vector Y1; the logistic regression classifier receives the output of the second fully connected layer, and analyzes the water supply pipe network. The network leakage event is classified and identified, and a classification vector Y2 is output, and the classification vector Y1 and the classification vector Y2 are the probability values for each leakage identification result;

The heterogeneous double classifier calculates the leakage identification result Y according to the above-mentioned classification vector Y1 and classification vector Y2 according to formula (1),

Y=β1*Y1+β2*Y2 Formula (1)

Among them, β1+β2=1, 0<β1<1, 0<β2<1.

In some embodiments, the leakage acoustic signal data is obtained using a hydrophone sensor, the flow data is obtained using a flow meter sensor, and the pressure data is obtained using a pressure gauge sensor.

In some embodiments, a leakage identification model is used to determine the corresponding sensor _Sk when a leakage event occurs, and the leakage localization is performed when the number of the corresponding sensors is greater than or equal to 2.

In some embodiments, the step S5 includes:

Taking the sensor _Sk as a base point, a closed loop is formed on the pipe network diagram with the shortest path, and the closed loop includes i pipeline nodes, and the corresponding pipeline node numbers are r _i ; j pipelines, the corresponding pipeline numbers are l _j ; k sensors, the corresponding sensor numbers are S _k , i, j, and k are all positive integers, and the shortest path refers to the shortest perimeter of the closed loop passing through the sensor S _k ;

Use the objective function f _i to search for the pipeline node closest to the leak point in the closed loop, calculate the f _i values of i pipeline nodes in turn, and select the pipeline node c corresponding to the smallest f _i value as the virtual closest to the leak point. leak point v _c ;

Taking the virtual leak point v _c as the center, the search path is the pipeline connected to the virtual leak point v _c , and a virtual leak point v _g (g=1...n) is set every z meters, a total of n, in order Calculate the f _i values of the n virtual leak points, and select the virtual leak point corresponding to the smallest f _i value as the final leak location point.

In some embodiments, the objective function is shown in formula (2),

f _i =∑ _a≠b (|t _a -t _b |-|ω _ia -ω _ib |) ² Equation (2)

In formula (2), f _i represents the squared error value of the time difference between the acoustic leakage signal at the i-th node reaching different sensors, t _a represents the time it takes for the acoustic leakage signal to reach sensor a, and t _b represents the acoustic leakage The time it takes for the signal to reach sensor b, ω _ia is the time it takes for the acoustic leakage signal to reach sensor a from node i, ω _ib is the time it takes for the acoustic leak signal to reach sensor b from node i, and node i can be a pipe Nodes or virtual leaks.

In some embodiments, the step S2 includes:

Acquire the leakage acoustic signal data sequence S(t), flow data sequence Q(t), and pressure data sequence P(t) of a single sensor, and normalize them;

Obtain the pipe diameter and pipe material data D(t), perform one-hot encoding processing on it, and convert the pipe diameter and pipe material data into a sequence containing only numbers "0" and "1";

Randomize the normalized sound leakage signal data sequence S(t), flow data sequence Q(t), pressure data sequence P(t), and corresponding pipe diameter and pipe material data D(t) Divided into training set and test set;

Select the cross-entropy loss function as the classification target of the model, and use the training set to train the deep neural network-based heterogeneous dual classifier leakage recognition model, until the cross-entropy loss function value remains unchanged, stop training;

Use the test set to test the trained model, and use the confusion matrix to evaluate the model effect.

(3) Beneficial effects

It can be seen from the above technical solutions that a technical method for leakage monitoring (including leakage identification and location) of a water supply pipe network disclosed in the present disclosure has at least one of the following beneficial effects:

(1) The present disclosure proposes a heterogeneous dual-classifier leakage recognition model based on a deep neural network to identify leakage accidents. The deep neural network is combined with a heterogeneous dual-classifier to overcome the problem of low recognition accuracy of a single classifier. , effectively improving the effect of pipeline leakage detection in the actual process, especially the identification under the interference of the surrounding complex environment, the present disclosure can identify small leakage events. At the same time, the identification result of the present disclosure can also determine the size of the leakage, which provides a basis for the next processing.

(2) The present disclosure proposes a local search and localization model based on the topology of the water supply pipe network for the location of leakage points, and the local search algorithm based on the topology of the pipe network is used to improve the calculation efficiency of the location of leakage; select nodes to reach different sensors The square value of the error of the time difference is used as the objective function, which improves the sensitivity of the model; fully considers the topology of the pipeline network in practice, expands the scope of application of pipeline leakage detection and identification, and enables it to perform high-efficiency testing under complex working conditions. Accurate leak location.

(3) The present disclosure uses sensors to collect data, including leakage sound signal data, flow data, pressure data, pipe diameter and pipe material data, and realizes the fusion and mining of multi-dimensional data through a deep neural network, further improving the practical application of the present disclosure. effect in use.

(4) The present disclosure combines deep neural network technology and local search and positioning technology to propose a complete pipeline leakage identification and positioning method, which significantly improves the identification accuracy of leakage events and the positioning accuracy of leakage points. The application scope of this method in the actual complex pipe network topology structure is discussed, so that the daily manager can find the leakage of the pipe network and repair it in time, reduce economic losses, save water resources, and assist the water company to make scientific and reasonable decisions.

Description of drawings

FIG. 1 is a flow chart of a method for identifying leakage of a water supply pipe network of the present disclosure.

FIG. 2 is a schematic diagram of a method for identifying and locating leakage of a water supply pipe network according to the present disclosure.

FIG. 3 is a structural diagram of a leakage identification model of a heterogeneous dual classifier based on a deep neural network of the present disclosure.

FIG. 4 is a topological structure diagram of a water supply pipe network in a certain area in an embodiment of the present disclosure.

Detailed ways

For better understanding and implementation of the present disclosure, the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that although the embodiments of the present disclosure have been described, it is obvious that the present disclosure is not limited to the above-described embodiments, and various modifications can be made without departing from the gist of the present disclosure.

In order to solve the deficiencies of the prior art, the present disclosure combines the acoustic wave transmission theory, the deep neural network model, and the local search and positioning model, and aims to provide a monitoring method for identifying and locating the leakage of the urban water supply pipe network. The method can improve the accuracy rate of leakage identification of the water supply pipe network and reduce the false alarm rate, so that the daily manager can timely find the leakage of the pipe network and repair it in time, reduce economic losses, save water resources, and assist the water company to make scientific and reasonable measures. decision making.

As shown in FIG. 1 , the leakage monitoring method (leakage identification) of the water supply pipe network of the present disclosure includes the following steps:

S1, obtain water supply pipe network data;

S2, establish a heterogeneous dual classifier leakage recognition model based on deep neural network; and

S3, using the water supply pipe network data and the heterogeneous dual-classifier leakage identification model to identify the leakage of the water supply pipe network.

Wherein, the water supply pipe network data includes leakage sound signal data propagating along the water medium, flow data in the water supply pipe, pressure data in the water supply pipe, pipe material data and pipe diameter data.

Further, after the step S3, the method for monitoring leakage of a water supply pipe network (leakage location) of the present disclosure further includes:

S4, establishing a local search and positioning model based on the topology of the water supply pipe network; and

S5, using the water supply pipe network leakage identification result and the local search and positioning model to locate leakage.

In general, the method of the present disclosure mainly includes three parts, which are data collection, establishment of a leak identification model, and establishment of a leak location model, as shown in FIG. 2 .

Specifically, the data collection process is as follows: sensors are arranged on the water supply pipe network for data collection, and the sensors include: a hydrophone, a flow meter, and a pressure gauge. The hydrophone receives the acoustic signal propagating along the water medium, the flow meter measures the flow in the pipe, and the pressure gauge measures the pressure in the pipe. The sensor transmits the collected data wirelessly to the control center and stores it in the computer. At the same time, basic information of pipelines, including pipe material data and pipe diameter data, is obtained from the urban water supply network database.

The process of establishing the leakage identification model is as follows: establish a heterogeneous dual-classifier leakage identification model based on a deep neural network, as shown in Figure 3, and the specific steps include:

The convolution layer is used to receive the leaked acoustic signal data, and the convolutional layer belongs to the convolutional neural network and is used to extract the characteristics of the leaked acoustic signal;

The maximum pooling layer is connected to the above-mentioned convolutional layer. The maximum pooling layer divides the output of the above-mentioned convolutional layer into m sub-regions, and extracts the maximum value of each sub-region to form an output, where m is a positive integer;

The long-short-term neural network layer is connected to the above-mentioned maximum pooling layer, and the long-short-term neural network layer belongs to a variant of the recurrent neural network and has a strong ability to process nonlinear data;

The first fully connected layer is used to receive flow data, pressure data, pipe material and pipe diameter data, and is connected to the fusion layer in a tensor series connection with the above-mentioned long and short-term neural network layer;

The fusion layer is connected to the second fully connected layer in a tensor series;

The second fully connected layer is connected with the support vector machine classifier and the logistic regression classifier respectively;

The support vector machine classifier classifies and identifies pipeline leakage events, and outputs a classification vector Y1;

The logistic regression classifier classifies and identifies pipeline leakage events, and outputs a classification vector Y2;

The classification vector Y1=(p1,...,p8) and the classification vector Y2=(q1,...,q8) are the probability values for each leakage identification result, as shown in Table 1;

The heterogeneous dual classifier calculates the leakage identification result Y according to the above-mentioned classification vector Y1 and the classification vector Y2 according to the formula (1), as shown in Table 1.

Y=β1*Y1+β2*Y2 Formula (1)

Among them, β1+β2=1, 0<β1<1, 0<β2<1.

The pipeline leakage event is modeled by the above-mentioned heterogeneous double classifier leakage identification model based on deep neural network. The model input is the leakage sound signal data sequence S(t), the flow data sequence Q(t), the pressure data sequence P(t), the corresponding pipe diameter and pipe material data D(t), the classification result Y1 of the support vector machine classifier, the classification result Y2 of the logistic regression classifier, the model output is the leakage identification result Y, the model output Y1, Y2, and Y are divided into 8 categories according to the size of leakage, and each category corresponds to a probability value, as shown in Table 1.

Use the leakage identification model to analyze each sensor separately, input the data collected by the sensor into the leakage identification model, and classify the leakage identification corresponding to the maximum probability value in the identification result Y output by the leakage identification model as 2~ At 8:00, it is determined that a leakage event has occurred in the pipeline network, and the corresponding sensor is the sensor corresponding to the leakage event.

Table 1 Leakage identification result output

For each sensor, the above-mentioned deep neural network-based heterogeneous dual-classifier leakage identification model is used for modeling. The specific steps are as follows:

1) Acquire the leakage acoustic signal data sequence S(t), the flow data sequence Q(t), and the pressure data sequence P(t) of a single sensor, and perform normalization processing on them. The normalization processing includes: The value of the original data is converted into the range of [0, 1], and the formula for normalization is shown in formula (2),

In formula (2), y represents the normalized data, x represents the input original data, and x _max and x _min represent the maximum and minimum values of the input data, respectively.

Obtain the corresponding pipe diameter and pipe material data D(t), perform one-hot encoding processing on it, and convert discrete variables such as pipe diameter and pipe material into a sequence containing only numbers "0" and "1".

The above normalized leakage acoustic signal data sequence S(t), flow data sequence Q(t), pressure data sequence P(t), and the corresponding pipe diameter and pipe material data D(t) are randomly divided into training set and test set.

2) Use the training set to train the above-mentioned deep neural network-based heterogeneous dual classifier leakage recognition model, and select the cross-entropy loss function as the classification target of the model. The formula of the cross-entropy loss function is shown in formula (3). ,

In formula (3), L represents the value of the cross-entropy loss function, N represents the sample size, h _pq represents the probability value of the sample p belonging to the category q, and y _pq represents the probability value that the model predicts the sample p to belong to the category q.

When the above cross entropy loss function value L no longer changes, the model stops training.

3) Use the test set to test the above trained model, and use the confusion matrix to evaluate the effect of the model, and the confusion matrix is shown in Table 2.

Table 2 Confusion matrix

normal event leakage event Model identified as normal event TP FP Model identified as leakage event FN TN

In Table 2, the TP value represents a normal event in practice, and the model recognition result is also a normal event; FP value represents a leakage event in practice, and the model recognition result is a normal event; FN value represents a normal event in practice, and the model recognition result is a leakage event; the TN value represents the actual leakage event, and the model identification result is also a leakage event.

According to the above four values, the recognition accuracy rate of the calculation model = TN/(FP+TN) and the false alarm rate = FN/(TP+FN). The higher the accuracy rate, the lower the false alarm rate, and the better the model effect.

The process of establishing the leak location model is as follows: establishing a local search location model based on the topology of the pipe network for leak location, and the specific steps are as follows:

1) The data collected by each sensor will establish a leakage identification model for analysis, but only the sensor whose leakage identification result is a leakage event will establish a leakage localization model for analysis. When the leakage identification classification corresponding to the maximum probability value in the identification result Y output by the leakage identification model is 2 to 8, it is determined that a leakage event has occurred in the pipeline network. The leakage identification model is used to determine the corresponding sensors _Sk when leakage events occur. When the number of sensors that are determined to have leakage events is greater than or equal to 2, a leakage localization model is established for analysis. Taking these sensors _Sk as the base point, Forms a closed loop on the pipe network diagram with the shortest path. The closed loop includes i pipeline nodes, and the corresponding pipeline nodes are numbered r _i ; j pipelines, the corresponding pipeline numbers are l _j ; k sensors, the corresponding sensor numbers are S _k , i, j, and k are all positive integer. The shortest path refers to the shortest perimeter of the closed loop passing through the sensor _Sk .

2) Establish a local search algorithm based on the topology of the pipe network, the objective function is as shown in formula (4),

f _i =∑ _a≠b (|t _a -t _b |-|ω _ia -ω _ib |) ² Equation (4)

In formula (4), f _i represents the error square value of the time difference between the acoustic leakage signal at the i-th node reaching different sensors, t _a represents the time it takes for the acoustic leakage signal to reach sensor a, and t _b represents the acoustic leakage The time it takes for the signal to reach sensor b, ω _ia is the time it takes for the acoustic leakage signal to reach sensor a from node i, ω _ib is the time it takes for the acoustic leak signal to reach sensor b from node i, and node i can be a pipe Nodes or virtual leaks;

3) Search the pipeline node closest to the leakage point in the closed loop, calculate the f _i value of the i pipeline nodes according to formula (4), and select the pipeline node c corresponding to the smallest f _i value as the nearest leakage point. the virtual leak point v _c ;

4) Taking the virtual leak point v _c as the center, the search path is the pipeline connected to the virtual leak point v _c , and a virtual leak point v _g (g=1...n) is set every z meters, a total of n , calculate the f _i value of the n virtual leakage points according to formula (4), and select the virtual leakage point corresponding to the smallest f _i value as the final leakage location point.

Figure 4 shows the topological structure of the water supply pipe network in a certain area. Sensors are installed at the pipe nodes or fire hydrants. The pipe network has 6 sensors S ₁ ... S ₆ , 6 pipe nodes r ₁ ... r ₆ , 16 sensors Pipes, the corresponding pipe numbers are l ₁ . . . l ₁₆ , the real leak point b ₁ .

Arrange sensors on the water supply network for data collection, each multi-sensor includes: hydrophone, flowmeter, and pressure gauge. The hydrophone receives the acoustic signal propagating along the water medium, the flow meter measures the flow in the pipe, and the pressure gauge measures the pressure in the pipe. The sensor transmits the collected data wirelessly to the control center and stores it in the computer. At the same time, basic information of pipelines, including pipe material and pipe diameter data, is obtained from the urban water supply network database.

For this embodiment, a deep neural network-based heterogeneous dual-classifier leakage identification model is established, including:

One convolution layer is used to receive the leaked acoustic signal data, the convolutional layer belongs to the convolutional neural network and is used to extract the characteristics of the leaked acoustic signal;

A maximum pooling layer is connected to the above-mentioned convolutional layer, and the maximum pooling layer divides the output of the above-mentioned convolutional layer into 20 sub-regions, and extracts the maximum value of each sub-region to form an output;

A long-short-term neural network layer is connected to the above-mentioned maximum pooling layer, and the long-short-term neural network layer belongs to a variant of the recurrent neural network and has a strong ability to process nonlinear data;

The first fully-connected layer is used to receive flow data, pressure data, pipe material and pipe diameter data, and is connected to one fusion layer in a tensor series connection with the above-mentioned long-short-term neural network layer;

The fusion layer is connected to the second fully connected layer in a tensor series;

The second fully connected layer is connected with the support vector machine classifier and the logistic regression classifier respectively;

The support vector machine classifier classifies and identifies pipeline leakage events, and outputs a classification vector Y1;

The logistic regression classifier classifies and identifies pipeline leakage events, and outputs a classification vector Y2;

The classification vector Y1=(p1,...,p8) and the classification vector Y2=(q1,...,q8) are the probability values for each leakage identification result, as shown in Table 1;

The heterogeneous dual classifier calculates the leakage identification result Y=β1*Y1+β2*Y2 according to the above-mentioned classification vector Y1 and classification vector Y2 according to formula (1), where β1=0.4 and β2=0.6.

The pipeline leakage event is modeled by the above-mentioned heterogeneous double classifier leakage identification model based on deep neural network. The model input is the leakage sound signal data sequence S(t), the flow data sequence Q(t), the pressure data sequence P(t), the corresponding pipe diameter and pipe material data D(t), the classification result Y1 of the support vector machine classifier, the classification result Y2 of the logistic regression classifier, the model output is the leakage identification result Y, the model output Y1, Y2, and Y are divided into 8 categories according to the size of leakage, and each category corresponds to a probability value, as shown in Table 1.

For the six sensors, the above-mentioned deep neural network-based heterogeneous dual-classifier leakage identification model is used for modeling. Here, sensor 1 is used as an example, and the specific steps are as follows:

1) Acquire the leakage acoustic signal data sequence S(t), the flow data sequence Q(t), and the pressure data sequence P(t) in the sensor 1, and normalize them according to formula (2).

Obtain the corresponding pipe diameter and pipe material data D(t), perform one-hot encoding processing on it, and convert discrete variables such as pipe diameter and pipe material into a sequence containing only numbers "0" and "1".

The above-mentioned leakage acoustic signal data sequence S(t), flow data sequence Q(t), pressure data sequence P(t), pipe diameter and pipe material data D(t) are randomly divided into training set and test set. The set has 32000 samples and the test set has 8000 samples.

2) Use the training set to train the above-mentioned deep neural network-based heterogeneous dual-classifier leakage recognition model. When the above-mentioned cross-entropy loss function value L no longer changes, the model stops training.

3) Use the 8000 samples of the test set to test the above-mentioned trained model, and use the confusion matrix to evaluate the model effect, and the confusion matrix is shown in Table 3,

Table 3 Confusion Matrix

normal event leakage event Model identified as normal event 7870 5 Model identified as leakage event 30 95

Therefore, the recognition accuracy of the model is 95/(95+5)=95%, the false alarm rate is 30/(7870+30)=0.38%, and the model effect is good.

To establish a local search and positioning model based on the topology of the pipe network, the specific steps are as follows:

1) According to the above leakage identification model, it is determined that the corresponding sensors when leakage events occur are S ₁ , S ₂ , and S ₃ , and then take the sensors S ₁ , S ₂ , and S ₃ as the base points, and take the shortest path on the pipe network diagram. forming a closed loop comprising sensors S ₁ , S ₂ , S ₃ , pipeline nodes r ₁ , r ₂ , r ₃ , pipelines l ₁ , l ₂ , l ₃ , l ₄ , l ₅ , l ₆ , Table 4 shows the basic data of the pipeline network in this area.

Table 4 Basic data of the pipeline network in a certain area

pipe number Tube length (m) pipe number Tube length (m) pipe number Tube length (m) l₁ 120 l₇ 67 l₁₃ 66 l₂ 140 l₈ 65 l₁₄ 100 l₃ 60 l₉ 83 l₁₅ 95 l₄ 55 l₁₀ 79 l₁₆ 107 l₅ 68 l₁₁ 62 l₆ 63 l₁₂ 49

2) Establish a local search algorithm based on the topology of the pipe network, and calculate the f _i values corresponding to the pipeline nodes r ₁ , r ₂ , and r ₃ according to formula (4), where the error square value f ₁ corresponding to the pipeline node t ₁ is the smallest, so Select the pipeline node t ₁ as the virtual leak point _vc closest to the leak point;

3) With the virtual leak point _vc as the center, along the pipelines l ₂ and l ₃ connected to the virtual leak point _vc , set a virtual leak point v _g every 2 meters (g=1...100) , a total of 100 virtual leakage points, according to formula (4) to calculate the f _i value of these 100 virtual leakage points, among which f ₆₅ is the smallest, so the virtual leakage point v ₆₅ is the point closest to the real leakage point b ₁ , the virtual leakage point The distance between v ₆₅ and the real leak point b ₁ is 0.88m, and the positioning accuracy is high.

The above results show that the present disclosure can more accurately identify the leakage accident of the water supply pipe network, and at the same time can more accurately locate the leakage point, and the method has strong practicability. The present disclosure expands the research content of the existing method for identifying and locating leakage in the pipeline network, and provides a new idea for water companies to make scientific and reasonable decisions. The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present disclosure in detail. It should be understood that the above are only specific embodiments of the present disclosure, and are not intended to limit the present disclosure. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure should be included within the protection scope of the present disclosure.

Claims (7)

1. A water supply pipe network leakage monitoring method comprises the following steps:

s1, acquiring water supply network data;

s2, establishing a heterogeneous double-classifier leakage identification model based on a deep neural network; and

s3, identifying the leakage of the water supply network by using the water supply network data and the heterogeneous dual-classifier leakage identification model;

the water supply network data comprises leakage sound signal data transmitted along a water medium, flow data in the water supply pipe, pressure data in the water supply pipe, pipe data and pipe diameter data;

the heterogeneous double-classifier leakage identification model based on the deep neural network comprises a convolutional layer, a maximum pooling layer, a long-time and short-time neural network layer, a first full-connection layer, a fusion layer, a second full-connection layer, a support vector machine classifier, a logistic regression classifier and a heterogeneous double classifier;

the step S3 includes:

receiving leakage acoustic signal data by the convolution layer;

the maximum pooling layer divides the output of the convolutional layer into m sub-regions, extracts the maximum value of each sub-region to form output, and m is a positive integer;

the long-time neural network layer and the short-time neural network layer perform nonlinear data processing on the output of the maximum pooling layer;

the first full-connection layer receives flow data, pressure data, pipe data and pipe diameter data;

the fusion layer receives the outputs of the long-time and short-time neural network layer and the first full connection layer;

a second fully connected layer receives the output of the fused layer;

the second full connection layer is respectively connected with the support vector machine classifier and the logistic regression classifier;

the support vector machine classifier receives the output of the second full connection layer, classifies and identifies the leakage event of the water supply network and outputs a classification vector Y1; the logistic regression classifier receives the output of the second full connection layer, classifies and identifies the leakage event of the water supply network and outputs a classification vector Y2;

the heterogeneous double classifier utilizes the classification vector Y1 and the classification vector Y2 to calculate a leakage identification result Y according to the following formula,

Y＝β1*Y1+β2*Y2

wherein β 1+ β 2 is 1, 0 < β 1 <1, 0 < β 2< 1.

2. The method of claim 1, further comprising, after said step S3:

s4, establishing a local searching and positioning model based on the topological structure of the water supply network; and

and S5, carrying out leakage positioning by using the water supply pipe network leakage identification result and the local searching and positioning model.

3. The method of claim 2, wherein sensors are arranged on the water supply network for data acquisition to obtain water supply network data, the sensors comprising hydrophones, flow meters, and pressure gauges; and acquiring the leakage acoustic signal data by using a hydrophone, acquiring flow data by using a flowmeter, and acquiring pressure data by using a pressure gauge.

4. The method of claim 3, wherein the sensor S corresponding to the occurrence of a leak event is determined by a leak recognition model_kAnd when the number of the corresponding sensors is more than or equal to 2, carrying out leakage positioning.

5. The method according to claim 4, wherein the step S5 includes:

with the sensor S_kForming a closed loop on the pipe network diagram by using the shortest path as a base point, wherein the closed loop comprises i pipeline nodes, and the number of the corresponding pipeline node is r_i(ii) a j tubes of pipeline with corresponding pipeline number l_j(ii) a k sensors with corresponding sensor number S_kI, j and k are positive integers, and the shortest path is through the sensor S_kThe shortest circumference of the closed loop of (a);

using an objective function f_iSearching the pipeline node closest to the leakage point in the closed loop, and sequentially calculating f of the i pipeline nodes_iValue, choose the smallest f_iThe pipeline node c corresponding to the value is taken as a virtual leakage point v nearest to the leakage point_c；

With virtual missing points v_cAs a center, search a path with a virtual leak point v_cThe connected pipelines are provided with 1 virtual leakage point v every z meters_g(g-1 … n), n, calculating f of the n virtual leakage points in turn_iValue, choose the smallest f_iAnd taking the virtual leakage point corresponding to the value as a final leakage positioning point.

6. The method of claim 5, wherein the objective function is represented by the following formula,

wherein f is_iError square value, t, representing the time difference between arrival of a leaky acoustic signal at the ith node at different sensors_aRepresenting the time, t, required for the leakage acoustic signal to reach the sensor a_bRepresenting the time, ω, required for the leakage acoustic signal to reach the sensor b_iaRepresenting the time, ω, required for the leakage acoustic signal to reach sensor a from node i_ibRepresenting the time required for the leaky acoustic signal to reach sensor b from node i, which may be a pipe node or a virtual leak point.

7. The method according to claim 3, wherein the step S2 includes:

acquiring a leakage sound signal data sequence S (t), a flow data sequence Q (t) and a pressure data sequence P (t) of a single sensor, and normalizing the leakage sound signal data sequence S (t), the flow data sequence Q (t) and the pressure data sequence P (t);

acquiring pipe diameter and pipe data D (t), carrying out independent thermal coding treatment on the pipe diameter and pipe data, and converting the pipe diameter and pipe data into a sequence only containing numbers of 0 and 1;

dividing the leakage sound signal data sequence S (t), the flow data sequence Q (t), the pressure data sequence P (t) and the corresponding pipe diameter and pipe data D (t) into a training set and a testing set at random;

selecting a cross entropy loss function as a classification target of the model, and training the heterogeneous dual-classifier leakage recognition model based on the deep neural network by using a training set until the cross entropy loss function value is unchanged, and stopping training;

and testing the trained model by using a test set, and evaluating the effect of the model by using a confusion matrix.

Images