A method, storage medium, and equipment for early warning of perimeter collapse of buried pipelines based on water hammer wave characteristic parameter sets.

Abstract

This invention discloses a method, storage medium, and method for early warning of perimeter collapse of buried pipelines based on a water hammer wave feature parameter set. By analyzing the vibration characteristics of water hammer during valve closure, a large amount of data was collected. Sensitive features in the time and frequency domains of the vibration signal were extracted and analyzed. Various machine learning algorithms were used for preliminary training and testing. The radial basis function neural network machine learning method with the best recognition performance was selected, and the model parameters were optimized using an ant colony algorithm. Furthermore, wavelet packet technology was used to extract and reconstruct the original signal from the original dataset. The extracted sensitive frequency band feature information parameters were combined with the original parameter set, significantly improving the model accuracy to nearly 98%. This method system can not only be used for safety monitoring of pipeline systems but can also be integrated with climbing robots and other equipment to achieve real-time monitoring and maintenance of key equipment such as solenoid valves. Therefore, it provides effective protection for pipeline transportation safety.

Description

Technical Field

[0001] This invention relates to the field of pipeline defect detection technology, and in particular to a method, storage medium, and device for early warning of perimeter collapse of buried transmission pipelines based on water hammer wave characteristic parameter sets. Background Technology

[0002] Currently, buried pipeline collapse is a common and serious problem in urban infrastructure operation and maintenance. When the pipeline structure is damaged or soil stability changes, the pipeline may collapse or break, leading to the interruption of basic services such as water supply, gas supply, and sewage treatment, causing great inconvenience and economic losses to people's lives and work. Because buried pipelines are usually located underground, their detection and monitoring are relatively difficult, and traditional inspection and maintenance methods often fail to detect changes in pipeline health and potential risks in a timely manner. Therefore, the research and application of intelligent pipeline monitoring systems based on patented technologies is of significant background importance.

[0003] Enhancing safety: Intelligent pipeline monitoring systems can detect pipeline structural problems and soil changes early through real-time monitoring and data analysis, providing warnings and suggestions so that timely maintenance measures can be taken to reduce the risk of pipeline collapse and breakage, thus ensuring public safety.

[0004] Improved reliability: Through the accurate assessment and prediction of intelligent pipeline monitoring systems, maintenance personnel can develop more effective maintenance plans, optimize maintenance work, reduce the probability of pipeline failures, and improve the reliability of basic services such as water and gas supply.

[0005] Cost Reduction: Traditional pipeline maintenance often involves periodic inspections or reactive repairs, which are inefficient and costly. Intelligent pipeline monitoring systems enable remote monitoring and accurate prediction, helping to optimize maintenance plans, reduce unnecessary maintenance frequency and scope, and thus lower maintenance costs.

[0006] Promoting Sustainable Development: Intelligent pipeline monitoring systems can extend pipeline lifespan, reduce the need for new pipeline construction, decrease resource waste and energy consumption, and drive the sustainable development of urban infrastructure. No existing technologies cover the aforementioned research. Summary of the Invention

[0007] The purpose of this invention is to provide a method, storage medium, and device for early warning of perimeter collapse of buried pipelines based on water hammer wave characteristic parameter set, which can accurately identify the early pipeline collapse type and thus provide assistance for pipeline collapse early warning.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for early warning of perimeter collapse of buried pipelines based on water hammer wave characteristic parameter set includes the following steps:

[0010] S1. Based on the different degrees of perimeter collapse of buried pipelines, defects were uniformly defined and classified. Early collapse of buried pipelines was divided into three situations: bottom collapse, surface exposure, and complete suspension. Then, the vibration data of the pipeline system on the experimental platform was collected using an accelerometer. The collected data was water hammer axial vibration data, where the axial direction is the direction of pipeline extension.

[0011] S2, For the vibration data collected from the pipeline system, perform batch preprocessing to obtain sample data. The preprocessing uses noise reduction and data segmentation.

[0012] S3, Feature Extraction: Extract the time-domain sensitive feature information and frequency-domain features of water hammer vibration signals caused by early collapse of different buried pipelines in the preprocessed sample data to form a feature set; among them, the time-domain sensitive feature information is obtained by establishing a finite element simulation model in COMSOL, simulating the valve closing action to generate water hammer vibration waves in the pipeline by setting a valve closing function, and picking up and analyzing the water hammer vibration wave information by point probe.

[0013] S4. According to the COMSOL simulation model results, the closing of the solenoid valve mainly generates water hammer and the impact of the solenoid valve's metal ball on the pipeline will generate vibration waves. Extract the vibration signal characteristics in the vicinity of 450HZ that are sensitive to the early collapse type of the perimeter of the conveying pipeline.

[0014] S5. Construct different basic machine learning models, and train support vector machines, random forests, radial basis function networks and backpropagation neural networks; use cross-validation to verify the accuracy of the models, and then select the most suitable machine learning model for further optimization.

[0015] S6 further optimizes the radial basis function neural network model, mainly by optimizing the parameter settings of the radial basis function neural network through the ant colony algorithm.

[0016] S7 trains the optimized machine learning model to obtain the parameter set with the highest recognition rate;

[0017] S8 tests the parameter set with the highest recognition rate obtained from S7 to identify early collapses and obtain the early collapse type of the pipeline perimeter.

[0018] The time-domain feature parameters and expressions extracted in step S3 are as follows:

[0019]

[0020]

[0021]

[0022] In the formula, x(n) represents the time-domain sequence of the signal, where n = 1, 2, ..., N, and N is the number of sample points. The mean A1 is used to describe the signal, which can detect the energy of the water hammer vibration signal under different conditions. The variance A2 represents the dynamic component of the signal energy, reflecting the degree of dispersion between data, and has good accuracy in model prediction and experimental data description. The effective value A3 describes the energy of the vibration signal.

[0023] The specific formulas for the frequency domain parameters are as follows:

[0024]

[0025] In the formula, S(k) represents the frequency domain sequence of the signal, where k = 1,2,…,K, and K is the number of sample points; frequency domain characteristic statistical parameters are introduced to describe the characteristics of the signal: the spectral mean B1 describes its fluctuation, and the frequency distribution of the pipeline vibration signal is described by the average frequency B2 and the root mean square frequency B3. It represents the square of the frequency corresponding to the frequency domain sequence.

[0026] The feature set in step S3 includes the time and frequency domain feature sets for the three cases: bottom collapse, surface exposure, and complete suspension.

[0027] In step S4, when extracting the features of the sensitive vibration signal, wave packet technology is used to decompose and reconstruct the signal to obtain the desired wavelet packet frequency band, and then frequency domain feature parameters are extracted; specifically, three-layer wavelets are selected, divided into 8 frequency bands, and Haar is the wavelet basis function.

[0028] In step S5, the MATLAB deep learning tool library is used for initial training and testing to select the most suitable machine learning method for testing and optimization. The main steps are as follows:

[0029] Data preparation: Organize the training data into a feature matrix X and a corresponding label vector Y; ensure that the data has been standardized or preprocessed;

[0030] Support Vector Machine Training: Create a Support Vector Machine model object using the `fitcsvm` function; train the model using the `train` method, passing in the feature matrix X and label vector Y; use the `predict` method to perform classification predictions on new data.

[0031] Random Forest Training: Create a random forest object using the TreeBagger function; train the random forest model using the train method, passing in the feature matrix X and label vector Y; use the predict method to perform classification predictions on new data;

[0032] Radial basis function neural network training: Create a radial basis function neural network object using the `newgrnn` function; train the network model using the `train` method, passing in the feature matrix X and label vector Y; use the `sim` method to classify and predict new data.

[0033] BP Neural Network Training: Create a BP neural network object using the feedforwardnet function; set the network structure and training parameters; train the network model using the train method, passing in the feature matrix X and label vector Y; use the sim method to perform classification and prediction on new data.

[0034] In step S6, the radial basis function neural network structure mainly consists of an input layer, radial base layers, and an output layer; the network structure relationship is mainly adjusted based on the training set, and secondly, the center point and width parameters are determined.

[0035] Step S6 specifically includes the following steps:

[0036] S61: Collect and organize pipeline data for training and testing.

[0037] S62: Preprocessing the data and extracting feature values ​​of various types;

[0038] S63: Establish a basic radial basis function neural network, including the network structure, initialization of weights and biases, and radial basis functions. S64: Iteratively optimize the hidden layers and hyperparameters of the radial basis function network using an ant colony optimization algorithm.

[0039] S65: Determine the network structure;

[0040] S66: Further optimization of parameters is performed using the backpropagation algorithm.

[0041] Step S64 specifically includes the following steps:

[0042] Initialize the parameters of the ant colony algorithm, including the number of ants, the pheromone volatilization rate, and the initial concentration of pheromones;

[0043] Set appropriate values ​​based on the actual situation for iterative optimization of ants;

[0044] The specific process is as follows:

[0045] S64.1 Initialize ant colony parameters: Set the number of ants and their initial positions, determine the size of the ant colony, and assign an initial position to each ant randomly or according to a specific strategy; the weights and biases of each ant are randomly initialized and used as the initial parameters of the neural network;

[0046] S64.2 Initialize the pheromone matrix: Create a pheromone matrix corresponding to the problem size. The initial values ​​can be set to the same constant or adjusted according to the characteristics of the problem.

[0047] S64.3 Ant Path Selection: Each ant performs forward propagation on the training dataset based on the current neural network parameters, calculates the value of the loss function, and updates the neural network parameters based on pheromone concentration and heuristic function;

[0048] S64.4 Pheromone Update: Update the pheromone concentration matrix based on the path quality of each ant; better paths will increase pheromone concentration, and poorer paths will decrease pheromone concentration.

[0049] S64.5 Repeatedly perform iterative optimization and determine the termination condition: Set the number of iterations or stopping conditions: Set an appropriate number of iterations or stopping conditions based on the actual problem and experience, such as reaching the maximum number of iterations or meeting the convergence condition; determine whether the termination condition is met.

[0050] S64.6 Update Neural Network Parameters: During the iteration process, record and save the optimal solution in each iteration, and update the weights and biases of the neural network according to the neural network parameters corresponding to the optimal path;

[0051] S64.7 returns the neural network model after deep optimization using the ant colony algorithm.

[0052] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the device on which the computer-readable storage medium is located executes the method for early warning of perimeter collapse of buried transmission pipelines based on water hammer wave characteristic parameter set as described in any one of claims 1-8.

[0053] An electronic device includes a processor and a memory, the memory storing a program executable on the processor, wherein the processor, when executing the program, implements the buried pipeline perimeter collapse early warning method based on water hammer wave feature parameter set machine learning as described in any one of 1-8.

[0054] This invention demonstrates significant effectiveness in classifying sample data using a radial basis function (RBF) neural network (RBN) algorithm model optimized with a parameter feature set. It effectively identifies early-stage collapse types, achieving a recognition accuracy of up to 97%. Furthermore, deep optimization of the RBN using an ant colony algorithm further validates the effectiveness of the selected feature parameter set in identifying collapse types. Finally, the method of deep optimization of the RBN using an ant colony algorithm verifies the effectiveness of the selected feature parameters in identifying early-stage collapse types. While the sample data collected under different working scenarios, equipment, and materials may vary, the overall type prediction method is applicable in all situations. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart of the present invention.

[0057] Figure 2 This is a graph showing the water hammer pressure variation under different conditions as described in this invention.

[0058] Figure 3 The graphs show the variation curves of the third-order characteristic frequency under different conditions described in this invention.

[0059] Figure 4 The original signal and specific frequency band diagram described in this invention

[0060] Figure 5 This is a two-dimensional scatter plot for visualizing the feature parameters described in this invention.

[0061] Figure 6 The flowchart for optimizing the radial basis function neural network using the ant colony algorithm described in this invention is shown.

[0062] Figure 7 The diagram shows the performance of the radial basis function neural network before and after optimization in identifying early collapse types, as described in this invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] like Figure 1 As shown, the present invention includes the following steps:

[0065] S1. Based on the different degrees of perimeter collapse of buried pipelines, defects were uniformly defined and classified. Early collapse of buried pipelines was divided into three situations: bottom collapse, surface exposure, and complete suspension. Then, the vibration data of the pipeline system on the experimental platform was collected using an accelerometer. The collected data was water hammer axial vibration data, where the axial direction is the direction of pipeline extension.

[0066] S2, For the vibration data collected from the pipeline system, perform batch preprocessing to obtain sample data. The preprocessing uses noise reduction and data segmentation.

[0067] S3, Feature Extraction: Extract the time-domain and frequency-domain sensitive features of water hammer vibration signals caused by early collapse of different buried pipelines from the preprocessed sample data to form a feature set; the time-domain sensitive feature information is obtained by establishing a finite element simulation model in COMSOL, simulating the valve closing action to generate water hammer vibration waves in the pipeline by setting a valve closing function, and picking up and analyzing the water hammer vibration wave information through point probes; the time-domain feature parameters and expressions extracted in step S3 are as follows:

[0068]

[0069]

[0070]

[0071] In the formula, x(n) represents the time-domain sequence of the signal, where n = 1, 2, ..., N, and N is the number of sample points. The mean A1 is used to describe the signal, which can detect the energy of the water hammer vibration signal under different conditions. The variance A2 represents the dynamic component of the signal energy, reflecting the degree of dispersion between data, and has good accuracy in model prediction and experimental data description. The effective value A3 describes the energy of the vibration signal.

[0072] The specific formulas for the frequency domain parameters are as follows:

[0073]

[0074] In the formula, S(k) represents the frequency domain sequence of the signal, where k = 1,2,…,K, and K is the number of sample points; frequency domain characteristic statistical parameters are introduced to describe the characteristics of the signal: the spectral mean B1 describes its fluctuation, and the frequency distribution of the pipeline vibration signal is described by the average frequency B2 and the root mean square frequency B3; where the average frequency B2 is the average value of the pipeline vibration frequency, and the root mean square frequency B3 is the arithmetic square root of the mean square frequency, and the mean square frequency is the weighted average of the squares of the signal frequencies.

[0075] The feature set in step S3 includes the time and frequency domain feature sets for the three cases: bottom collapse, surface exposure, and complete suspension.

[0076] S4. Based on the COMSOL simulation model results, the closure of the solenoid valve mainly generates water hammer, and the impact of the solenoid valve's metal ball on the pipeline generates vibration waves. Vibration signal features sensitive to the early collapse type of the pipeline perimeter are extracted in the vicinity of 450HZ. In step S4, when extracting sensitive vibration signal features, wave packet technology is used to decompose and reconstruct the signal to obtain the desired wavelet packet frequency band, and then frequency domain feature parameters are extracted. Specifically, three-layer wavelets are selected, divided into 8 frequency bands, and Haar is the wavelet basis function.

[0077] S5. Construct different basic machine learning models, and train support vector machines, random forests, radial basis function networks and backpropagation neural networks; use cross-validation to verify the accuracy of the models, and then select the most suitable machine learning model for further optimization.

[0078] S6 involves further optimizing the radial basis function (RBF) neural network model, primarily through ant colony optimization of the RBF neural network parameter settings. In step S6, the RBF neural network structure mainly consists of an input layer, radial base layers, and an output layer; the network structure relationships are adjusted primarily based on the training set, followed by determining the center point and width parameters.

[0079] Step S6 specifically includes the following steps:

[0080] S61: Collect and organize pipeline data for training and testing.

[0081] S62: Preprocessing the data and extracting feature values ​​of various types;

[0082] S63: Establish a basic radial basis function neural network, including the network structure, initialization of weights and biases, and radial basis functions. S64: Iteratively optimize the hidden layers and hyperparameters of the radial basis function network using an ant colony optimization algorithm.

[0083] S65: Determine the network structure

[0084] S66: Further optimization of parameters is performed using the backpropagation algorithm.

[0085] Step S64 specifically includes the following steps:

[0086] Initialize the parameters of the ant colony algorithm, including the number of ants, the pheromone volatilization rate, and the initial concentration of pheromones;

[0087] Set appropriate values ​​based on the actual situation for iterative optimization of ants;

[0088] The specific process is as follows:

[0089] S64.1 Initialize ant colony parameters: Set the number of ants and their initial positions, determine the size of the ant colony, and assign an initial position to each ant randomly or according to a specific strategy; the weights and biases of each ant are randomly initialized and used as the initial parameters of the neural network;

[0090] S64.2 Initialize the pheromone matrix: Create a pheromone matrix corresponding to the problem size. The initial values ​​can be set to the same constant or adjusted according to the characteristics of the problem.

[0091] S64.3 Ant Path Selection: Each ant performs forward propagation on the training dataset based on the current neural network parameters, calculates the value of the loss function, and updates the neural network parameters based on pheromone concentration and heuristic function;

[0092] S64.4 Pheromone Update: Update the pheromone concentration matrix based on the path quality of each ant; better paths will increase pheromone concentration, and poorer paths will decrease pheromone concentration.

[0093] S64.5 Repeatedly perform iterative optimization and determine the termination condition: Set the number of iterations or stopping conditions: Set an appropriate number of iterations or stopping conditions based on the actual problem and experience, such as reaching the maximum number of iterations or meeting the convergence condition; determine whether the termination condition is met.

[0094] S64.6 Update Neural Network Parameters: During the iteration process, record and save the optimal solution in each iteration, and update the weights and biases of the neural network according to the neural network parameters corresponding to the optimal path;

[0095] S64.7 returns the neural network model after deep optimization using the ant colony algorithm.

[0096] S7 trains the optimized machine learning model to obtain the parameter set with the highest recognition rate;

[0097] S8 tests the parameter set with the highest recognition rate obtained from S7 to identify early collapses and obtain the early collapse type of the pipeline perimeter.

[0098] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the device on which the computer-readable storage medium is located executes the above-described method for early warning of perimeter collapse of buried transmission pipelines based on water hammer wave characteristic parameter sets.

[0099] An electronic device includes a processor and a memory, the memory storing a program that can run on the processor, the processor executing the program to implement the buried pipeline perimeter collapse early warning method based on water hammer wave feature parameter set machine learning as described above.

[0100] This invention utilizes the ant colony optimization algorithm to deeply optimize radial basis function neural networks (RBNs). The ant colony algorithm optimizes relevant parameters, demonstrating its superior optimization capabilities compared to other mainstream optimization algorithms. Furthermore, cross-validation is employed. First, the number of hidden layer nodes within the range of hyperparameters and empirical formulas is iteratively optimized. Then, the initial weights and biases are iteratively optimized, thereby achieving optimal training performance for the RBN. When using cross-validation to optimize the RBN, the overall optimization problem is decomposed into the following steps: dataset partitioning, parameter selection, model training, model evaluation, and hyperparameter tuning. These steps are repeated until the best-performing parameter combination is found, resulting in a high-performance RBN model. This cross-validation method helps solve the overall optimization problem of RBNs, improving the model's accuracy and generalization ability.

[0101] Furthermore, in order to provide a more detailed explanation of the method of the present invention, the following detailed process description is provided with specific examples.

[0102] S1, using COMSOL simulation, establish a simple pipeline model, such as Figure 2 The simulated valve-closing function for setting the fluid velocity at the pipe outlet uses the sigmoid function, which utilizes fluid inertia to generate water hammer. The transmission characteristics of the water hammer are compared in three scenarios: bottom collapse, surface exposure, and complete suspension. The sensitivity characteristics of its time-domain signal and characteristic frequencies are as follows: Figure 2 As shown. In the time domain, the peak attenuation of water hammer has a significant impact. Based on this geometric model, the fully coupled characteristic frequencies are solved. The third and fourth order characteristic frequencies are particularly sensitive. For example... Figure 3 As shown, based on these patterns, we analyze the sensitive characteristic parameters of valve-closing pipeline vibration.

[0103] The pipeline experimental verification mainly consists of three parts: a water hammer device, an experimental data acquisition system, and surrounding clay. The water hammer device comprises a DN20 pipeline, an automatic solenoid valve, a flow meter, and other system components. It can quickly open and close the solenoid valve to generate water hammer excitation while controlling the water flow velocity to 0.5 m / s. The pipeline system is cantilevered and constrained to a steel frame to ensure that the contact interface between the pipe and the steel frame does not affect the experiment. A rubber box houses a portion of the pipeline as a buried pipeline scenario, simulating different early collapse conditions of buried pipelines. The transmission characteristics of water hammer vibration signals were studied under three conditions: bottom collapse, surface exposure, and complete suspension. Experimental data was collected and analyzed using the data acquisition system. The results are as follows: Figure 4 The upper part is shown, where the accelerometer is used to collect data for analysis.

[0104] This invention, based on the actual conditions of pipelines, uniformly defines and classifies defects by utilizing the different degrees of perimeter collapse of buried pipelines. Early collapse of buried pipelines is categorized into three types: bottom collapse, surface exposure, and complete suspension. A three-dimensional finite element model of the buried pipeline is then established. In the 3D finite element model, each type of early collapse is uniformly distributed within a defined relative position. COMSOL is used to simulate the early collapse of the pipeline, obtaining the corresponding water hammer signals, thereby acquiring water hammer vibration signals under different collapse types.

[0105] S2, Data Preprocessing: The collected pipeline vibration acceleration signals are first subjected to noise reduction. The general steps are as follows:

[0106] Importing data: The pipeline vibration acceleration signal can be imported into the MATLAB environment using the importdata or load function.

[0107] High-pass filtering: If low-frequency noise exists in a signal, it can be filtered out using a high-pass filter. High-pass filters can be designed using the `highpass` or `designfilt` functions from the signal processing tool library, and then applied to the signal using the `filter` function.

[0108] Denoising: The signal processing toolbox provides various noise reduction methods, one of the most commonly used being wavelet transform noise reduction. The following are the basic steps for using wavelet transform noise reduction:

[0109] a. Use the wavedec function to perform wavelet decomposition on the signal to obtain wavelet coefficients.

[0110] b. Threshold the wavelet coefficients, setting the low-amplitude wavelet coefficients to zero and retaining the high-amplitude wavelet coefficients.

[0111] c. Use the waverec function to reconstruct the denoised signal from the processed wavelet coefficients.

[0112] Since the data acquisition system is set to 3000 sampling points, according to the sampling theorem, it can only acquire signals less than 1500Hz, so low-pass filtering is not required.

[0113] The process of extracting time-domain and frequency-domain features of vibration signals using S3 can be achieved by setting up a feature extraction function in MATLAB using the features and expressions in the table below.

[0114] Table 1

[0115]

[0116] In the time domain parameters, the mean is the average value of a set of data. By calculating the mean of pressure or flow data over a period of time, we can understand the overall average level of the system and make a preliminary assessment of water hammer. Variance is a measure of the dispersion of a set of data, describing the fluctuation of the data. In water hammer problems, variance can reflect the instability of the system and the intensity of pressure fluctuations. The RMS value is the square root of the average of the sum of squares of a set of data, also known as the root mean square value. The RMS value can measure the energy magnitude of the water hammer phenomenon, that is, the amplitude of pressure or flow.

[0117] The mean allows us to understand the system's average state and determine whether the overall level is too high or too low. Variance helps us assess the system's stability and fluctuations, and understand the intensity of water hammer. RMS is an indicator of water hammer energy and can be used to assess the severity of water hammer and its impact on the system.

[0118] In the frequency domain parameters, the spectral mean is calculated by performing spectral analysis on pressure data. The spectral mean reflects the distribution of water hammer phenomena in the frequency domain. By observing changes in the spectral mean, the frequency range and main frequency components of water hammer phenomena can be determined. The mean square frequency (MSF) is the frequency point where energy is concentrated, calculated through spectral analysis. The MSF reflects the main vibration frequency of water hammer phenomena in the frequency domain. By monitoring changes in the MSF, the frequency characteristics of water hammer phenomena and possible resonance can be determined. The root mean square frequency (RMS) is the square root of the sum of squares of the frequency distribution calculated through spectral analysis; it is also an indicator reflecting the energy distribution of water hammer phenomena in the frequency domain. The RMS quantifies the energy of water hammer phenomena at different frequencies, helping to understand the frequency characteristics and energy distribution of water hammer.

[0119] By using the spectral mean, we can understand the distribution of valve vibration in a pipeline at different frequencies and find possible frequency components.

[0120] Mean square frequency (MSF) helps us determine the dominant vibrational frequencies and assess the risk of frequency resonance. Root mean square frequency (RMS) quantifies the energy distribution in the frequency domain, helping us understand the system's frequency characteristics and fluctuations. By analyzing these frequency domain characteristic parameters, we can gain a more comprehensive understanding of the frequency domain characteristics, revealing frequency distribution and energy concentration. The invention extracts six identification parameters expressing collapse types: parameters A1, B1, A2, B2, A3, and B3. In the embodiments, wavelet packet technology is first used to decompose and reconstruct the signal to obtain the desired wavelet packet frequency bands before extracting the frequency domain characteristic parameters. Furthermore, the collected feature set is initially trained and tested using the MATLAB deep learning tool library. The advantages and disadvantages of various machine learning methods are analyzed, and the best-performing radial basis function neural network is selected for further optimization and parameter tuning.

[0121] S4. To further improve the training effect, the training parameter set was first optimized. According to the COMSOL simulation model results, in addition to water hammer, the closing of the solenoid valve causes resonance when the solenoid valve's metal ball strikes the pipe. Figure 3 It is known that the third resonant point is sensitive to early collapse types. A new combination is proposed, which involves extracting vibration signals from specific frequency bands to analyze their frequency domain characteristics, such as... Figure 4 The vibration signal from the pipeline is used to extract sensitive frequency bands as frequency domain features. The new feature set is then trained using a radial basis function neural network.

[0122] S5, Support Vector Machine, Random Forest, Radial Basis Functional Network, and Backpropagation Neural Network were used for training. Cross-validation was employed to verify the model's accuracy. The goal was to select the most suitable machine learning model for further optimization. The main steps are as follows:

[0123] Data preparation: Organize the training data into a feature matrix X and a corresponding label vector Y. Ensure that the data has been standardized or preprocessed.

[0124] Support Vector Machine (SVM) Training: a. Create an SVM model object using the `fitcsvm` function. b. Train the model using the `train` method, passing in the feature matrix X and the label vector Y. c. Use the `predict` method to perform classification predictions on new data.

[0125] Random Forest Training: a. Create a random forest object using the TreeBagger function. b. Train the random forest model using the train method, passing in the feature matrix X and label vector Y. c. Use the predict method to perform classification predictions on new data.

[0126] Radial basis function neural network training: a. Create a radial basis function neural network object using the `newgrnn` function. b. Train the network model using the `train` method, passing in the feature matrix X and the label vector Y. c. Use the `sim` method to perform classification prediction on new data.

[0127] Backpropagation (BP) Neural Network Training: a. Create a BP neural network object using the `feedforwardnet` function. b. Set the network structure and training parameters. c. Train the network model using the `train` method, passing in the feature matrix X and label vector Y. d. Use the `sim` method to perform classification and prediction on new data.

[0128] Table 2 shows that the radial basis function neural network outperforms the other three in terms of overall accuracy.

[0129] Table 2

[0130]

[0131] S6, according to claim S5, a method for identifying and optimizing the characteristic parameter set of early perimeter collapse vibration information of buried pipelines based on machine learning, characterized in that, in step S5, during the training process of the radial basis function neural network, data preprocessing includes data cleaning, feature selection, etc., to improve data quality and usability. Neural network construction: determining the network structure, including the number of neurons in the input layer, hidden layer, and output layer. Simultaneously, selecting appropriate activation functions and loss functions. Weight and bias initialization: initializing the weights and biases in the neural network, generally using random initialization methods. Parameter optimization using the backpropagation algorithm: minimizing the loss function by forward propagating the input data, calculating the loss, and using the backpropagation algorithm to update the weights and biases. Iterative training: repeating step 4 until a stopping condition is met, such as reaching the maximum number of iterations or loss function convergence. Model evaluation and optimization: evaluating the trained model using methods such as validation sets or cross-validation, and optimizing the model based on the evaluation results, such as adjusting the network structure and learning rate.

[0132] The above process, which requires manual experience to adjust and optimize, can be automatically optimized using ant colony optimization (ACO) algorithms. This can improve the classification accuracy and convergence speed of neural networks, thereby enhancing the effectiveness of early collapse identification for buried pipelines. The optimization process is as follows: Figure 6 As shown, the specific steps are as follows:

[0133] Initialization parameters: Set the number of ants and initial positions: Determine the size of the ant colony and assign an initial position to each ant randomly or according to a specific strategy.

[0134] Initialize the pheromone matrix: Create a pheromone matrix corresponding to the problem size. The initial values ​​can be set to the same constant or adjusted according to the characteristics of the problem.

[0135] Ants move by selecting their next location based on a certain probability. A roulette wheel selection mechanism can be used, where the probability of selection is determined by pheromone concentration and a heuristic factor.

[0136] Solution construction: Ants construct solutions based on the chosen positions, usually through traversal or path connection.

[0137] Pheromone Update: The pheromone update mechanism updates the pheromone matrix based on the ant's construction solution and the problem's evaluation function. A common update strategy is to update the pheromone matrix using the pheromone increment along the ant's trajectory for each movement.

[0138] Pheromones evaporate: After each update, pheromones will decay at a certain rate to increase the globality of the search and the ability to explore.

[0139] Repeated iteration: Set the number of iterations or stopping conditions: Set an appropriate number of iterations or stopping conditions based on the actual problem and experience, such as reaching the maximum number of iterations or meeting the convergence condition.

[0140] Iterative optimization: In each iteration, the movement of ants and the construction of solutions, as well as the updating of pheromones, are continuously updated through the previous steps until the stopping condition is met.

[0141] Output and save the optimal solution: During the iteration process, record and save the optimal solution in each iteration.

[0142] Output: After the algorithm finishes, the global optimal solution is determined based on the optimal solution and the evaluation criteria.

[0143] For each early collapse type, 1250 samples were used as the training set and 350 samples as the test set. The radial basis function neural network (RBN) had 6 input layer nodes, 3 output layer nodes, and 12 hidden layer nodes after optimization. The output of the RBN was rounded. In this embodiment, the datasets for the three cases—bottom collapse, exposed surface, and completely suspended—were labeled 1, 2, and 3, respectively.

[0144] Combination Figure 7 It can be seen that the optimized radial basis function neural network algorithm model has a significant improvement in the classification effect of sample data, and can effectively identify the early collapse type. The recognition accuracy of the optimized radial basis function neural network is close to 98%. At the same time, the optimization of the radial basis function neural network by ant colony algorithm also verifies the effectiveness of the selected recognition parameters for identifying the early collapse type.

[0145] This invention provides a method for early warning of perimeter collapse of buried pipelines based on machine learning of water hammer wave feature parameter sets. It can accurately identify the early collapse type of buried pipeline perimeter, which has significant engineering significance and good application prospects.

[0146] Furthermore, COMSOL analysis was used to analyze the sensitivity characteristics of water hammer vibration during solenoid valve closure and early perimeter collapse of buried pipelines. The results were then verified using an experimental platform to extract relevant identification feature parameters and frequency domain features of sensitive frequency bands. The effectiveness of different feature parameter sets in inverting the type of early perimeter collapse of buried pipelines was analyzed. This approach can improve convenience and accuracy in identifying pipeline collapse in the field of buried pipeline inspection, and is of great significance for preventing buried pipeline collapse.

[0147] In the description of this invention, it should be noted that directional terms such as "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", and "counterclockwise" indicate the orientation and positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. They should not be construed as limiting the specific protection scope of this invention.

[0148] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0149] Note that the above description is merely a preferred embodiment and application of the technical principles of the present invention. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the specific embodiments described herein, and may include many other effective embodiments without departing from the concept of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A buried fluid-carrying pipeline perimeter collapse early warning method based on a water hammer wave characteristic parameter set, characterized in that: The method comprises the following steps: S1, the defects are uniformly defined and classified by using the different periphery collapse degree information of the buried flow pipeline, and the early collapse of the buried flow pipeline is divided into three cases of bottom collapse, surface exposure and complete suspension; then the vibration data of the pipeline system of the experimental platform are collected by using an acceleration sensor; the collected is the water hammer axial vibration data, wherein the axial direction is the extending direction of the pipeline; S2, for the vibration data collected by the pipeline system, data batch preprocessing is performed to obtain sample data, and the preprocessing adopts denoising and data segmentation; S3, feature extraction: time domain sensitive feature information and frequency domain features of water hammer vibration signals caused by different early collapses of the buried flow pipeline in the sample data after preprocessing are extracted to form a feature set; wherein the time domain sensitive feature information is obtained by establishing a finite element simulation model by COMSOL, setting a valve closing function to simulate the valve closing action to make the buried flow pipeline produce water hammer vibration waves, and picking up and analyzing the water hammer vibration wave information by a point probe; S4, according to the results of the COMSOL simulation model, the water hammer caused by the closing of the electromagnetic valve and the vibration wave caused by the impact of the electromagnetic valve metal ball on the buried flow pipeline, the time domain sensitive feature information and the frequency domain features of the vibration signal sensitive to the early collapse type of the periphery of the buried flow pipeline near 450HZ are extracted in the feature set; S5, different machine learning basic models are constructed, and support vector machines, random forests, radial basis neural networks and BP neural networks are trained; cross-validation is adopted to verify the accuracy of the model, so as to select the most suitable machine learning model for further optimization; S6, the radial basis neural network model is further optimized, and the parameter setting of the radial basis neural network is optimized by using the ant colony algorithm; S7, the optimized machine learning model is trained to obtain the parameter set with the highest recognition rate; S8, the parameter set with the highest recognition rate obtained in S7 is tested by corresponding radial basis neural network to identify the early collapse respectively, and the early collapse type of the periphery of the buried flow pipeline is obtained.

2. The water hammer wave based feature parameter set based perimeter collapse warning method for buried fluid carrying pipeline as claimed in claim 1 wherein, The expression of the time domain sensitive feature information extracted in the step S3 is as follows: ； ； ； In the formula, x(n) represents the time domain sequence of the signal, wherein n = 1, 2, …, N, n is the number of sample points, the mean A1 is used to describe the signal, and the energy of the water hammer vibration signal under different conditions can be detected; the variance A2 represents the dynamic component of the signal energy; and the effective value A3 describes the energy of the vibration signal; The specific frequency domain feature formula is as follows: ； In the formula, S(k) represents the frequency domain sequence of the signal, wherein k = 1, 2, …, K, k is the number of sample points; the frequency domain characteristic statistical parameter is introduced to describe the characteristics of the signal: the spectrum mean B1 describes the fluctuation of the signal, the frequency distribution of the buried flow pipeline vibration signal is described by the average frequency B2 and the root mean square frequency B3; represents the square of the corresponding frequency of the corresponding frequency domain sequence.

3. The water hammer wave based feature parameter set based warning method for perimeter collapse of a buried fluid pipeline of claim 1, wherein, The feature set in the step S3 includes time and frequency domain feature sets of the three cases of bottom collapse, surface exposure and complete suspension.

4. The water hammer wave based feature parameter set based warning method for perimeter collapse of a buried fluid pipeline of claim 1, wherein, In step S4, when the sensitive vibration signal features are extracted, the wave packet technology is used for signal decomposition and reconstruction to obtain the desired wave packet frequency band, and then the frequency domain feature parameters are extracted; the wave packet selects three layers of wavelets, which are divided into 8 frequency bands, and Haar is the wavelet base function.

5. The water hammer wave based feature parameter set based warning method for perimeter collapse of buried fluid carrying pipeline as claimed in claim 1 wherein, In step S6, the radial basis neural network structure includes an input layer, a radial basis layer and an output layer; the network structure relationship is adjusted according to the feature set to determine the center point and the width parameter.

6. The water hammer wave based feature parameter set based warning method for perimeter collapse of a buried fluid pipeline of claim 1, wherein, Step S6 specifically comprises the following steps: S61: Collect and organize the vibration data collected for the pipeline system for training and testing, S62: Preprocessing of data, extracting various types of characteristic values; S63: Establishing a basic radial basis neural network, including the structure of the network, the initialization of weights and biases, and the radial basis function, S64: Iterative optimization of the hidden layer and super parameters of the radial basis network through the ant colony algorithm, S65: Determining the hidden layer and super parameters of the network; S66: Parameter optimization through the backpropagation algorithm.

7. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor, so that the device where the computer readable storage medium is located executes the buried flow pipeline perimeter collapse early warning method based on the water hammer wave characteristic parameter set in any one of claims 1-6.

8. An electronic device, comprising: The device comprises a processor and a memory, and the memory stores a program capable of running on the processor, and the processor implements the buried flow pipeline perimeter collapse early warning method based on the water hammer wave characteristic parameter set machine learning in any one of claims 1-6 when executing the program.

Images