Water supply network leakage detection method based on time-frequency domain characteristics of ground penetrating radar

Abstract

The invention discloses a water supply network leakage detection method based on time-frequency domain characteristics of a ground penetrating radar, and relates to the technical field of leakage detection. The method comprises the steps that echo signals, collected by the ground penetrating radar, of different areas of the water supply pipe network and leakage detection results of all the areas are obtained; the leakage detection results comprise non-leakage, leakage and leakage positions; extracting time-frequency domain characteristics of the echo signals; performing feature screening in the time-frequency domain features by adopting a random forest algorithm and a principal component analysis method to obtain effective features of the echo signals; forming a training data set by the effective features of all the echo signals and the leakage detection results of different areas; training a back propagation neural network by adopting the training data set; and inputting effective characteristics of echo signals of the water supply network to be detected, which are acquired in real time, into the trained back propagation neural network, and outputting a leakage detection result of the water supply network to be detected. The accuracy and efficiency of leakage detection of the water supply network can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of leakage detection, and in particular to a method for detecting leakage in a water supply network based on time-frequency domain characteristics of a ground penetrating radar. Background Art

[0002] Leakage in urban water supply networks not only wastes resources, but can also cause major safety hazards. When a leak occurs in an urban water supply network, foreign matter and impurities at the ruptured part of the network will flow against the water flow into the network and mix with tap water, and finally be transported to the homes of residents at high pressure. Once the impurities contain toxins, they will inevitably contaminate drinking water and eventually threaten the safety of drinking water for residents. On the other hand, water in the leakage area continues to flow out and continuously washes the soil, making the soil structure loose and changing the geological environment. If the leakage lasts for a long time, the soil will be hollowed out to form water-containing cavities, causing the network to deform or settle. In more serious cases, emergencies such as roadbed settlement may occur.

[0003] Current leak detection technologies can be divided into two categories: internal state detection of the pipeline network and external environment detection. Internal state detection is to determine whether there is a leak by monitoring the pipeline's own indicators such as flow, pressure and flow velocity. There are mainly various methods such as pressure waves, mass balance, and dynamic modeling. The external environment detection method is based on the impact of pipeline leakage on the surrounding environment, and identifies the leak point by detecting the surrounding environment. It mainly includes acoustic detection, satellite method, infrared thermal imaging method, ground penetrating radar and other methods.

[0004] After the water supply network breaks, the friction sound of high-pressure water rubbing against the network produces water leakage friction sound, and the sound of water flow impacting the soil medium produces medium leakage sound, which together constitutes the sound of water leakage. In the current urban tap water pipe leakage, the main technical means used for leak detection are listening method, correlation instrument method and other technical means. The listening method uses the listening rod to capture the sound of water leakage in the network to demarcate the leakage area and lock the leakage location. The listening method is the most common and effective method in leak detection engineering, but it requires a lot of manpower and a long general inspection cycle to complete the general inspection of the entire urban water supply network. The listening method is not effective in noisy environments, and it often needs to work at night, which has high requirements for the working environment. The accuracy of leak detection is greatly affected by the professional quality of the operators. Only those with considerable experience can achieve more accurate leak positioning, which places high demands on the professional quality of the operators. The correlation analysis method, like the listening method, belongs to the acoustic vibration method. When the listening method cannot identify the leak or cannot ensure the accuracy of leak detection because the sound of water leakage is too weak, the correlation analysis method is often used to supplement it. The correlation analysis method uses a correlator to identify leaks. It requires placing multiple probes at different measuring points in the same pipe network. The leak point is locked based on the time difference between the probes receiving the leaking sound at different distances. The correlation analysis method uses probes to collect leaking sounds and can objectively locate leaks by eliminating human experience factors. However, the premise of its work is that the metal pipe network has good sound transmission, so it is not effective in non-metallic pipe networks. At the same time, the method requires that the workplace must have multiple interconnected pipe network points where probes can be placed, which is greatly restricted by the working environment. Summary of the invention

[0005] The purpose of this application is to provide a water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar, which can improve the accuracy and efficiency of water supply network leakage detection.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] The present application provides a water supply network leakage detection method based on the time-frequency domain characteristics of a ground-penetrating radar, comprising: obtaining echo signals of different areas of a water supply network collected by a ground-penetrating radar and leakage detection results of each area; the leakage detection results include no leakage, leakage and leakage location; extracting the time-frequency domain characteristics of the echo signal; using a random forest algorithm and a principal component analysis method to perform feature screening in the time-frequency domain characteristics to obtain effective features of the echo signal; forming a training data set with effective features of all echo signals and leakage detection results of different areas; using the training data set to train a back-propagation neural network; inputting the effective features of the echo signal of the water supply network to be tested, which are collected in real time, into the trained back-propagation neural network, and outputting the leakage detection results of the water supply network to be tested.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects:

[0009] The present application provides a water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar. The ground penetrating radar is simple to operate, quickly covers a large area, and is not significantly affected by environmental noise. It can work under various environmental conditions and can significantly improve the detection efficiency. Furthermore, by combining data processing of feature extraction and feature screening, and machine learning technology of back propagation neural network, it reduces the dependence on operator experience and skills, reduces detection errors caused by human factors, and improves detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0011] Figure 1 A schematic flow chart of a water supply network leakage detection method based on time-frequency domain characteristics of ground penetrating radar provided in one embodiment of the present application;

[0012] Figure 2 A schematic diagram of survey line arrangement provided in another embodiment of the present application;

[0013] Figure 3 A radar image before preprocessing provided by another embodiment of the present application;

[0014] Figure 4 A preprocessed radar image provided by another embodiment of the present application;

[0015] Figure 5 A schematic diagram of a back propagation neural network structure provided in another embodiment of the present application;

[0016] Figure 6 A schematic diagram of the working principle of a water supply network leakage detection method based on time-frequency domain characteristics of ground penetrating radar provided in one embodiment of the present application. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0018] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0019] In order to solve the problem of identifying water supply network leakage, in an exemplary embodiment, as Figure 1 As shown, a water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar is provided, comprising the following steps 101 to 106. Among them:

[0020] Step 101: Acquire the echo signals of different areas of the water supply network collected by the ground penetrating radar and the leakage detection results of each area. The leakage detection results include no leakage, leakage and leakage location.

[0021] Step 102: extracting the time-frequency domain features of the echo signal.

[0022] Step 103: using a random forest algorithm and a principal component analysis method to perform feature screening in the time-frequency domain features to obtain effective features of the echo signal.

[0023] Step 104: All effective features of the echo signals and the leakage detection results in different areas are combined into a training data set.

[0024] Step 105: Use the training data set to train a back propagation neural network.

[0025] Step 106: input the effective features of the echo signal of the water supply network to be tested, which are collected in real time, into the trained back propagation neural network, and output the leakage detection result of the water supply network to be tested.

[0026] By implementing the above steps 101 to 106, based on the BP neural network, a series of processing is performed on the ground penetrating radar data to quickly and accurately identify whether there is a leakage signal in the area to be tested, and finally identify the leakage location. Passive detection is transformed into active detection, which improves detection efficiency and reduces leakage detection costs.

[0027] At present, the detection of hidden leaks in urban water supply pipe networks mainly relies on acoustic detection for inspection. Among them, the listening method relies on manual experience and environmental conditions, has poor effects in noisy environments, requires nighttime operations, and leads to unstable listening leaks. It requires high professional quality of operators, long general inspection cycles, and large investment in manpower and material resources. The application is not affected by environmental noise, and can work stably even in noisy environments without nighttime operations. Compared with the artificial listening method, ground penetrating radar can quickly cover a large area through simple operation and improve detection efficiency. It reduces manpower requirements and shortens the general inspection cycle, thereby reducing the overall investment in manpower and material resources. The correlation analysis method is a traditional hidden leak detection technology that locates the leak point by detecting the propagation of sound in the pipeline. However, this method has some limitations, especially in the application of non-metallic pipe networks, and is easily restricted by the working environment. In contrast, ground penetrating radar does not rely on the sound conductivity of the pipe network material, so it can be applied to metal and non-metallic pipe networks, including plastic pipes such as PVC and PE. It is not significantly affected by environmental noise and can work under various environmental conditions, including daytime or nighttime, without the need for a specific quiet environment. Ground penetrating radar can penetrate a variety of underground media and detect pipe network leaks at different depths, while correlation analysis methods are usually limited by the distance that sound can travel in a pipe. Ground penetrating radar reduces the reliance on operator experience and skills, reduces detection errors caused by human factors, and improves the consistency and reliability of detection.

[0028] In another exemplary embodiment of the present application, the above step 101 may be replaced by the following steps 201 to 202:

[0029] Step 201: Collect original electromagnetic wave signals from different areas of the water supply network through ground penetrating radar.

[0030] Based on the routine inspection in the city, ground penetrating radar is used to complete the actual data collection, and multiple measurement lines are arranged to cover the leakage points as far as possible to obtain a more complete leakage situation. During the data collection process, the obvious abnormal interference and leakage point positions on the measurement line are marked according to the actual situation. The purpose is to locate the interference and leakage point positions on the radar profile for easy processing and analysis. Figure 2 shown.

[0031] Step 202: preprocessing the original electromagnetic wave signal to obtain an echo signal; the preprocessing includes: removing DC drift, static correction, gain, Butterworth bandpass filtering, mean smoothing and background removal.

[0032] DC drift removal eliminates drift by adjusting the time window size so that the echo signal amplitude is symmetrical with the average amplitude. Static correction removes direct wave signal interference caused by air by locating the zero depth value. Gain processing is used to amplify deep signals on the time profile to highlight the signal strength, which is convenient for profile data interpretation and to obtain rich information for later analysis. Butterworth bandpass filtering is used to filter noise according to the noise frequency distribution range to better highlight effective signals. Background removal Background removal first calculates the background value of the interference area, subtracts the background value from each radar echo to reduce the interference in the horizontal direction, and the noise signal of the processed radar profile is eliminated and the overall is cleaner. Sliding average smoothes the data by averaging adjacent echo signals to eliminate strong surface echoes and oscillation signals. Through the preprocessing process, richer and more effective information can be obtained from radar images.

[0033] The radar image before preprocessing is as follows Figure 3 As shown in Figure 2, the preprocessed radar image is as follows: Figure 4 shown. Figure 3 and Figure 4 The depth on the right side of the vertical axis specifically represents the depth at a velocity equal to 0.095 m / ns.

[0034] In another exemplary embodiment of the present application, the time-frequency domain features of the above step 102 include: amplitude statistical attribute features, frequency domain attribute features and instantaneous attribute features.

[0035] In another exemplary embodiment of the present application, the amplitude statistical attribute characteristics include: total amplitude, average amplitude, root mean square amplitude, amplitude square difference, amplitude cube difference, average absolute amplitude, maximum absolute value amplitude, total absolute value amplitude, maximum peak value, average peak amplitude, maximum trough amplitude, average trough amplitude, total energy, average energy, energy square difference and energy cube difference.

[0036] The total amplitude A1 is calculated as: In the formula, x i represents the i-th sampling point on the single-channel signal of the ground penetrating radar echo, and N represents the number of sampling points of the single-channel signal of the ground penetrating radar echo.

[0037] The calculation formula of average amplitude A2 is:

[0038] The calculation formula for the root mean square amplitude A3 is:

[0039] The calculation formula of the amplitude square difference A4 is:

[0040] The calculation formula of the amplitude cube error A5 is:

[0041] The calculation formula of the average absolute amplitude A6 is:

[0042] The calculation formula of the maximum absolute amplitude A7 is: A7 = max(|x i |).

[0043] The total absolute amplitude A8 is calculated as:

[0044] The calculation formula of the maximum peak value A9 is: A9 = max(x 正 );where x 正 Indicates the amplitude of the sample points with positive amplitude.

[0045] Average peak amplitude A 10 The calculation formula is: In the formula, represents the amplitude of the jth sampling point with positive amplitude, and n represents the number of sampling points with positive amplitude.

[0046] Maximum trough amplitude A 11 The calculation formula is: 11 =max(|x 负 |); where x 负 Indicates the amplitude of the sample points where the amplitude is negative.

[0047] Average trough amplitude A 12 The calculation formula is: In the formula, Indicates the amplitude of the kth sampling point with negative amplitude. The number of sampling points with positive amplitude is equal to the number of sampling points with negative amplitude.

[0048] Total energy A 13 The calculation formula is:

[0049] Average energy A 14 The calculation formula is:

[0050] Energy square difference A 15 The calculation formula is:

[0051] Energy cube difference A 16 The calculation formula is:

[0052] 16 amplitude statistics attributes are mainly selected to characterize the amplitude change of leakage echo signal. Among them, the first 12 attribute parameters can better describe the amplitude change of echo signal, and the last 4 attribute parameters are used to characterize the energy change of echo signal.

[0053] In another exemplary embodiment of the present application, the frequency domain attribute features include: main frequency, average frequency, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation.

[0054] The Fourier transform method converts the signal from the time domain to the frequency domain for analysis, and obtains the distribution of different frequencies of the signal. Assume that the Fourier transform process of the ground penetrating radar echo signal y = x(t) is: the time function y(t) is transformed into the frequency domain function F(t), and the expression is:

[0055]

[0056] Main frequency 17 The calculation formula is: 17 =x 横坐标 (F(max)).

[0057] Average frequency P 18 The calculation formula is: Here, F refers to F(t).

[0058] Center of gravity frequency P 19 The calculation formula is: Where P represents the signal power.

[0059] Mean square frequency P 20 The calculation formula is:

[0060] RMS frequency P 21 The calculation formula is:

[0061] Frequency variance P 22 The calculation formula is:

[0062] Frequency standard deviation P 23 The calculation formula is:

[0063] In another exemplary embodiment of the present application, the instantaneous attribute characteristics include: maximum instantaneous amplitude, average instantaneous amplitude, maximum instantaneous amplitude variance, maximum instantaneous frequency, average instantaneous frequency, instantaneous frequency variance, amplitude-weighted instantaneous frequency and energy-weighted instantaneous frequency.

[0064] Assume that the signal expression of the Hilbert-Huang transform of the ground penetrating radar echo signal is yh=H(t), and extract the following instantaneous attributes: maximum instantaneous amplitude S 24 is the maximum instantaneous amplitude of the MF1 component; the average instantaneous amplitude S 25 is the average instantaneous amplitude of the MF1 component; the maximum instantaneous amplitude variance S 26 is the instantaneous amplitude fluctuation of MF1 component; the maximum instantaneous frequency S27 is the maximum instantaneous frequency of the MF1 component; the average instantaneous frequency S 28 is the average instantaneous frequency of MF1 component; the instantaneous frequency variance S 29 Characterizes the discrete degree of the instantaneous frequency of the MF1 component; the amplitude-weighted instantaneous frequency S 30 is the instantaneous amplitude weighting of MF1 component; energy weighting instantaneous frequency S 31 Weighting for the instantaneous energy of MF1 component.

[0065] Maximum instantaneous amplitude S 24 The calculation formula is: 24 =max(a). Where a represents the instantaneous amplitude.

[0066] Average instantaneous amplitude S 25 The calculation formula is:

[0067] Maximum instantaneous amplitude variance S 26 The calculation formula is:

[0068] Maximum instantaneous frequency S 27 The calculation formula is: 27 =max(f). Wherein, f represents the instantaneous frequency.

[0069] Average instantaneous frequency S 28 The calculation formula is:

[0070] Instantaneous frequency variance S 29 The calculation formula is:

[0071] Amplitude-weighted instantaneous frequency S 30 The calculation formula is: Where A represents the instantaneous amplitude.

[0072] Energy-weighted instantaneous frequency S 31 The calculation formula is: Where E represents instantaneous energy.

[0073] In another exemplary embodiment of the present application, random forest and principal component analysis are integrated to extract leakage sensitive attributes, and the above step 103 can be replaced by the following steps 301 to 302:

[0074] Step 301: Calculate the importance of each time-frequency domain feature using the random forest algorithm, and select the time-frequency domain features whose importance is greater than a preset threshold to obtain the screened time-frequency domain features.

[0075] 31 attribute features are extracted from the amplitude statistics, spectrum and instantaneous attributes of the ground penetrating radar given above. These attribute features are specific reflections of the physical properties of the underground medium. However, not all ground penetrating radar attribute parameters can be used to identify leaks. If the number of parameters is too large, it will affect the effect and efficiency of identifying leakage echo signals. The key is to select a few and effective attributes from a large number of attribute parameters. This application uses random forests and principal components to select leakage sensitive attributes.

[0076] Assuming that there are m ground penetrating radar echo signals, random forest extracts n leakage parameters to form an m-dimensional matrix X mn It can be expressed as:

[0077]

[0078] Calculate the importance of each feature and sort them in descending order according to their importance; select features based on their importance to get a new feature set; repeat the above process with the new feature set until the main features are left; compare the out-of-bag error rates of each feature set and the corresponding feature set, and select the feature set with the lowest out-of-bag error rate as the target feature. Use random forest to evaluate the importance of GPR attribute parameters, and finally select 9 attribute parameters as leakage parameters.

[0079] Among them, random forest is used to divide the ground penetrating radar attributes into three different importance levels. Attributes with importance values ​​greater than 1 are important attribute parameters, which have a greater impact on echo signal recognition. The value is between 0 and 1. The attributes are divided into secondary important attribute parameters, which have an impact on echo signal recognition. The remaining importance is 0, which can be regarded as unimportant parameters that have little impact on echo signal recognition. Features with feature importance greater than 1 are selected as new feature sets. The selection criteria for the remaining main features are: the feature importance value of the main features is greater than 1 and the feature importance ranking remains basically unchanged.

[0080] Step 302: using the principal component analysis method, re-screen the filtered time-frequency domain features to obtain effective features of the echo signal.

[0081] To ensure the validity of the result analysis, the leakage attribute is normalized and dimensionless. The covariance matrix is ​​calculated. The eigenvalues ​​and corresponding eigenvectors of the covariance matrix are obtained by eigendecomposition. The cumulative contribution rate of the principal components is calculated in turn according to the size of the eigenvalues. Generally, when the cumulative contribution rate reaches more than 80%, the principal component representing the original variable is selected. After principal component analysis, 6 attribute parameters are selected as leakage sensitive attributes.

[0082] In another exemplary embodiment of the present application, the back-propagation (BP) neural network is a multi-layer feedforward network trained by error back-propagation. It does not need to clarify the functional relationship between input and output. It performs error analysis based on the training results and expected results to modify the weights and thresholds to obtain the expected results. The BP neural network has the advantages of large-scale parallel processing, self-learning and self-adaptation capabilities, strong robustness and fault tolerance, etc. It can solve classification problems in complex engineering and can better identify leakage echo signals. The BP neural network is mainly divided into forward propagation and reverse propagation of errors. Forward propagation refers to the process from the input layer through the hidden layer to the output layer. The back propagation of errors refers to the continuous correction of the network weights from the output layer to the hidden layer along the negative gradient direction of the network function to achieve higher prediction accuracy. After the BP neural network clarifies a set of learning data sets with known target results, it continuously modifies the network weights according to the error results to ultimately minimize the mean square error and complete the training of the neural network. Figure 5 As shown in the figure, a BP neural network contains three different structural levels: input layer, hidden layer, and output layer. Generally, the actual input of the network determines the number of input layers, the purpose of designing the network determines the number of output layers, and the number of hidden layers determines the efficiency and performance of the network. Generally, the optimal number is found based on the results of actual applications.

[0083] In another exemplary embodiment of the present application, there are two hidden layers in the back propagation neural network; the activation function of the first hidden layer is the Log-Sigmoid function, and the activation function of the second hidden layer is the purelin linear function.

[0084] In another exemplary embodiment of the present application, the above step 105 can be replaced by the following steps 401 to 402:

[0085] Step 401: Divide the training data set into a training set and a validation set in a ratio of 8:2.

[0086] Step 402: Using the training set and the validation set to train and validate the back propagation neural network to obtain a trained back propagation neural network.

[0087] From the above, it can be seen that the process of building a BP neural network model for identifying leakage echo signals of urban water supply pipe networks in this application is as follows:

[0088] (1) Establish a training set and a validation set. Select an equal number of leakage and non-leakage echo signals from the measured sample library, and establish a training set and a validation set in a ratio of 8:2. Since the units and dimensions of the numerical values ​​of the leakage sensitive attributes finally determined are different, in order to ensure the effective recognition of the BP neural network, this application uniformly normalizes the network input attribute parameters. Label each echo signal according to its actual situation, where the leakage echo signal label is 1 and the non-leakage signal label is 2. Recombine the single-channel echo signals and labels in the sample set to form a new data group as input for model training and evaluation.

[0089] (2) Construct a BP neural network model.

[0090] This application has been verified by experiments, and setting two hidden layers has a better effect, so the activation function of the first layer is selected as the Log-Sigmoid function, and the second layer is the purelin linear function.

[0091] (3) Training of the neural network. Input the training set and label determined in this application into the network to calculate the difference between the output value and the target value, and back-propagate the gradient to the neural network parameters to update the network weight value. Call the Sim function to identify. If the error between the result and the actual result is large, it is necessary to continuously adjust the learning rate, number of trainings, maximum allowable average error and other parameters until the error no longer decreases. The training of the leakage identification network is completed. Apply the trained network to the ground penetrating radar echo signal to automatically identify the leakage echo signal.

[0092] In another exemplary embodiment of the present application, the output results of the neural network model constructed by the present application are divided into two categories: leakage 1 and non-leakage 2. The confusion matrix is ​​used to calculate four indicators of accuracy, precision, recall rate, and F1 score to evaluate the accuracy of the model. Among them, the F1 score indicator can evaluate the precision and recall rate, and is the most important evaluation indicator.

[0093] The comparison between the true value and the predicted value is shown in Table 1.

[0094] Table 1 Comparison of actual values ​​and predicted values

[0095]

[0096] Accuracy: describes the number of samples correctly identified as leaks as a percentage of the total number of input samples, focusing on the overall recognition results of the samples. Accuracy is the most intuitive criterion for judging the model recognition results. The results are greatly affected by the number of samples. The numerical value calculation formula is:

[0097]

[0098] Precision: refers to the number of samples predicted by the model output as leaks and the number of samples with actual input leaks, describing the accuracy of the model prediction results. Its numerical calculation formula is:

[0099]

[0100] Recall rate: refers to the ratio of samples whose predicted output is also leaky to the number of samples whose output is leaky.

[0101]

[0102] F1 score: The harmonic mean of precision and recall can better coordinate the functional relationship between precision and recall, and can better evaluate the robustness of the recognition model.

[0103]

[0104] Figure 6 The working principle of the water supply network leakage detection method of the present application is shown, including data collection, data preprocessing, feature extraction, leakage sensitive attribute selection (random forest feature screening and principal component analysis feature screening), neural network training and leakage identification. Through the measured data, leakage data and non-leakage data can be obtained to provide data for neural network training. The acquisition data preprocessing step is to improve the signal quality by denoising and filtering the electromagnetic wave signal collected by the ground penetrating radar, and provide clear data for subsequent analysis. The time-frequency domain information feature extraction step extracts feature data that can characterize the leakage signal from the preprocessed data to provide a basis for identifying the leakage signal. The leakage sensitive attribute extraction step adopts the fusion random forest and principal component analysis (Principal Component Analysis, PCA) method to select feature data that significantly characterizes the leakage signal to improve the recognition efficiency and accuracy. BP neural network training, using the feature data as a training set, training the neural network, and obtaining the mapping relationship between the feature data and the leakage signal. Leakage identification, the feature data of the collected data to be identified is input into the trained neural network, and the presence of a leakage signal is identified according to the mapping result, so as to realize the leakage signal identification of the newly collected data.

[0105] The innovation of this application lies in the combination of ground-penetrating radar technology with advanced data processing and machine learning technologies, which improves the accuracy and efficiency of leak detection in urban water supply networks, and utilizes the water-sensitive characteristics of ground-penetrating radar technology to achieve early detection of hidden leaks.

[0106] The beneficial effects of this application are as follows:

[0107] (1) This application can not only significantly improve the quality of ground penetrating radar data and the accuracy of interpretation through carefully designed data preprocessing steps, but also greatly enhance the signal-to-noise ratio, thereby effectively reducing the interference of background noise on data interpretation. This process involves a series of operations such as cleaning, filtering, and correction of the original data, ensuring the reliability and validity of the data. In addition, by optimizing the preprocessing algorithm, it is possible to more accurately identify and distinguish useful signals and interference signals in underground structures, reducing the possibility of misjudgment due to signal confusion. This not only improves the confidence of the detection results, but also lays a solid foundation for subsequent data analysis and interpretation.

[0108] (2) The time-frequency domain feature extraction and principal component analysis methods can effectively identify and extract representative key information from massive data, and can accurately reflect the leakage characteristics of the leakage location. By combining time domain and frequency domain analysis, this application can capture the dynamic changes and periodic characteristics of the signal, thereby providing richer and deeper insights for leak detection. At the same time, combined with the random forest algorithm, the accuracy and robustness of data screening can be further enhanced, making it possible to identify truly critical features from complex data sets. This method of comprehensive application of multiple data analysis techniques has significantly improved the efficiency and accuracy of data processing, and provided strong technical support for leak detection and diagnosis.

[0109] (3) Training and recognition through neural network technology not only significantly improves the generalization ability of the model, enabling it to adapt to various complex data environments, but also greatly enhances the recognition accuracy of the model. This machine learning method can automatically extract complex features in the data without human intervention, thereby reducing dependence on expert knowledge. In addition, by continuously iterating and optimizing the network structure, the model parameters can be further adjusted to adapt to specific application scenarios, thereby achieving higher recognition accuracy and lower misjudgment rate.

[0110] (4) Ground penetrating radar is a detection technology that uses the reflection characteristics of electromagnetic waves to identify interfaces with obvious dielectric differences. Ground penetrating radar is sensitive to water. Slight changes in water content will cause obvious changes in the ground penetrating radar, and it can sense the existence of leaks that acoustic technology cannot detect. Actively discover leaks as early as possible to control leakage losses, thereby avoiding the formation of large leaks and the serious harm caused. At the same time, ground penetrating radar technology can identify the distribution of pipelines to reduce the excavation area and avoid economic losses and damage to road structures caused by mis-excavation. Applying ground penetrating radar technology to leak detection projects can objectively and early detect water supply network leaks, and also improve the current urban leak inspection system.

[0111] (5) The method of the present application can not only reduce the dependence on the environment and the professional quality of the operators, but also improve the detection efficiency and reduce the cost, which is of great significance for ensuring the safety of urban water supply.

[0112] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0113] The present application uses specific examples to illustrate the principles and implementation methods of the present application. The above examples are only used to help understand the method and core ideas of the present application. At the same time, for those skilled in the art, according to the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present application.

Claims

1. A water supply network leakage detection method based on time-frequency domain characteristics of ground penetrating radar, characterized in that: include: Acquire echo signals of different areas of the water supply network collected by ground penetrating radar and leakage detection results of each area; the leakage detection results include no leakage, leakage and leakage location; Extracting time-frequency domain features of the echo signal; Using random forest algorithm and principal component analysis method, feature screening is performed in the time-frequency domain features to obtain effective features of the echo signal; The effective features of all echo signals and the leakage detection results in different areas constitute a training data set; Using the training data set to train a back propagation neural network; The effective features of the echo signal of the water supply network to be tested, which are collected in real time, are input into the trained back propagation neural network, and the leakage detection results of the water supply network to be tested are output.

2. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: Acquire the echo signals of different areas of the water supply network collected by ground penetrating radar, including: The original electromagnetic wave signals from different areas of the water supply network are collected by ground penetrating radar; The original electromagnetic wave signal is preprocessed to obtain an echo signal; the preprocessing includes: removing DC drift, static correction, gain, Butterworth bandpass filtering, mean smoothing and background removal.

3. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: The time-frequency domain features include: amplitude statistical attribute features, frequency domain attribute features and instantaneous attribute features.

4. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 3 is characterized in that: The amplitude statistical attribute characteristics include: total amplitude, average amplitude, root mean square amplitude, amplitude square difference, amplitude cube difference, average absolute amplitude, maximum absolute value amplitude, total absolute value amplitude, maximum peak value, average peak amplitude, maximum trough amplitude, average trough amplitude, total energy, average energy, energy square difference and energy cube difference.

5. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 3 is characterized in that: The frequency domain attribute features include: main frequency, average frequency, center of gravity frequency, mean square frequency, root mean square frequency, frequency variance and frequency standard deviation.

6. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 3 is characterized in that: The instantaneous attribute characteristics include: maximum instantaneous amplitude, average instantaneous amplitude, maximum instantaneous amplitude variance, maximum instantaneous frequency, average instantaneous frequency, instantaneous frequency variance, amplitude-weighted instantaneous frequency and energy-weighted instantaneous frequency.

7. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: The random forest algorithm and the principal component analysis method are used to perform feature screening in the time-frequency domain features to obtain effective features of the echo signal, specifically including: The importance of each time-frequency domain feature is calculated using the random forest algorithm, and the time-frequency domain features whose importance is greater than a preset threshold are selected to obtain the screened time-frequency domain features; The principal component analysis method is used to perform another screening on the screened time-frequency domain features to obtain the effective features of the echo signal.

8. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: The hidden layer in the back propagation neural network is two layers; The activation function of the first hidden layer is the Log-Sigmoid function, and the activation function of the second hidden layer is the purelin linear function.

9. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: The training data set is used to train a back propagation neural network, specifically including: The training data set is divided into a training set and a validation set in a ratio of 8:2; The back propagation neural network is trained and verified using a training set and a verification set to obtain a trained back propagation neural network.

10. The water supply network leakage detection method based on the time-frequency domain characteristics of ground penetrating radar according to claim 1 is characterized in that: The accuracy evaluation indicators of the trained back propagation neural network include: accuracy, precision, recall rate and F1 score.

Images