Pipeline leakage prediction method based on variational mode decomposition and time convolution network
By applying HOA-optimized VMD filtering technology and time convolution network in pressured water pipelines, the problem of low leakage signal extraction and prediction accuracy in the prior art is solved, and high-precision leakage prediction and model universality are achieved.
Patent Information
- Application Number
- CN202411620703.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-14
- Publication Date
- 2025-02-28
AI Technical Summary
The prior art is difficult to effectively extract leakage signal characteristics from pressured water pipelines, resulting in low leakage prediction accuracy and difficult to deal with equipment noise and environmental noise interference in the signal.
The leakage signal is filtered by a variational modal decomposition (VMD) method based on hiking optimization algorithm (HOA) optimization, and a leakage prediction model is constructed in combination with a time convolutional network (TCN) to improve the training accuracy of the model.
The equipment noise and environmental noise interference are effectively removed, and the effective characteristics of the leakage signal are extracted, which significantly improves the accuracy of leakage prediction and the universality of the model.
Smart Images
Figure CN119538726A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of pipeline leakage detection, and in particular relates to a water pipeline leakage prediction method based on variational mode decomposition (VMD) optimized by a hiked optimization algorithm (HOA) and a temporal convolutional network. Background Art
[0002] Pipeline transportation has become one of the important means of transporting fluid industrial products such as oil, gas, and water. (Yang L, Zhao Q. ABiLSTM based pipeline leak detection and disturbance assisted localization method [J]. IEEE Sensors Journal, 2021, 22 (1): 611-620.). When a water supply pipeline leaks, it will not only cause a serious waste of water resources, but also seriously affect urban transportation, residents' lives and industrial production (OMAamo, "Leak detection, size estimation and localization in pipe flows," IEEE Trans. Autom. Control, vol. 61, no. 1, pp. 246–251, Jan. 2016.). With the acceleration of urbanization, the demand for various types of water has increased dramatically, and the mileage of newly laid water pipelines has also increased accordingly. However, due to pipeline wear, equipment aging, corrosion, geographical environment and human activities, leakage accidents in water pipelines have become one of the increasingly serious problems (Zhou, S., O'Neill, Z., O'Neill, C., 2018. A review of leakage detection methods for district heating networks. Appl. Therm. Eng. 137, 567–574.). Therefore, it is urgent to develop high-precision pipeline leakage detection technology to improve leakage detection capabilities and reduce pipeline leakage rates.
[0003] In recent years, with the rapid development of deep learning technology and big data technology, emerging pipeline leakage detection methods are gradually replacing or enhancing traditional methods in some aspects, and have achieved remarkable results. However, the leakage signal generated when the pipeline leaks contains interference factors such as pump vibration, fluid in the pipe, and environmental noise. How to extract effective leakage signal features from the measured leakage signal to achieve accurate prediction of leakage in pressurized water pipelines is still a technical problem in the current development of leakage prediction models. Therefore, it is necessary to use a filtering method with strong adaptability and good effect to reduce the noise of the measured leakage signal. And it is necessary to construct a leakage prediction model suitable for pressurized water pipelines based on the characteristics of the leakage signal of the pressurized water pipeline. Summary of the invention
[0004] The purpose of the present invention is to provide a pipeline leakage prediction method based on variational mode decomposition and time convolution network, which can effectively extract the characteristics of water pipeline leakage signals and realize effective prediction of pressurized water pipeline leakage events.
[0005] The technical solution adopted by the present invention is a pipeline leakage prediction method based on variational mode decomposition and time convolution network, which specifically includes the following steps:
[0006] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is conducted by controlling the switch ball valve to simulate pipeline leakage;
[0007] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0008] Step 3: Construct a walking optimization algorithm-optimized variational mode decomposition (HOA-VMD) algorithm, then perform mode decomposition and signal reconstruction on the leakage experimental test data obtained in step 2, and filter the experimental test data;
[0009] Step 4: Construct a temporal convolutional network (TCN) pipeline leakage prediction model and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
[0010] The present invention is also characterized in that:
[0011] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0012] Step 2 is as follows:
[0013] Step 2.1: Turn on the variable frequency water pump;
[0014] Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h;
[0015] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording.
[0016] Step 2.4: After recording a single leak point, apply step 2.3 to new leak points in sequence until all leak points are tested;
[0017] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0018] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0019] In step 2.2, the selection requirements for the specified flow conditions are: 3 / h~8m 3 The interval between each two specified flow conditions in the / h interval is 2m 3 / h, at 8m 3 / h~13m 3 The interval between each two specified flow conditions in the / h interval is 1.5m 3 / h.
[0020] Step 3 is as follows:
[0021] Step 3.1: Initialize the parameters of the walking optimization algorithm (HOA), that is, initialize the position and fitness of the traveler group. The best travel position of the traveler corresponds to the parameter combination [K, α]. Among them, the HOA parameters are set as follows: the number of people in the travel group is 10 to 100, and the maximum number of iterations is 20 to 500; the fitness function selects the envelope entropy, and the calculation formula of the envelope entropy E is shown in formula (1);
[0022]
[0023] Where n represents the number of the discretized envelope signal interval; N is the number of discretized envelope signal intervals; p n is the probability that the envelope signal falls in the nth interval, which is obtained by calculating the ratio of the number of sample points in each interval to the total number of sample points;
[0024] Step 3.2: Use the HOA algorithm to optimize the combination of VMD decomposition parameters [K, α]. The smaller the envelope entropy value, the better the combination of [K, α]. Then, the high-frequency pressure signal data S = [S1, S2, …, S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition. After decomposition, S i Each signal data s in p K IMF components and one residual component will be obtained, and the time domain signals of the decomposed IMF components are named IMF1, IMF2, ..., IMFK in sequence;
[0025] Step 3.3: Use Fourier transform (FFT) to convert the time domain signal of each IMF component into a frequency domain signal, draw the spectrum of each IMF component, and verify whether the spectrum of each IMF component is concentrated in the target frequency interval. The judgment standard is: if the cumulative spectrum energy in the target frequency interval is not less than 80% of the total energy, it means that the spectrum of the IMF component is concentrated in the target frequency interval; then judge whether the spectrum of different IMF components is well separated. The judgment standard is: if the spectrum overlap rate between different IMF components is less than 20%, it is considered to be well separated;
[0026] If the IMF component spectrum is concentrated in the target frequency interval and is well separated, it means that the optimization effect is good; if the spectrum energy overlap of the IMF component exceeds 30% (that is, the frequency interval where the energy is concentrated has significant overlap) or the frequency range covered by the spectrum distribution of a certain IMF component exceeds 500Hz, it means that the parameters of VMD decomposition are improperly set, and the optimization result is discarded, and step 3.2 is repeated until the stop condition of good IMF separation effect is met, then the algorithm process is completed, the optimization is stopped, and the optimal parameter combination [K, α] is output;
[0027] Step 3.4: Use the optimal parameter combination [K, α] obtained by the optimization to test the high-frequency pressure signal data S = [S1, S2, ..., S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition, and the IMF component signals with a similarity of more than 90% with the original signal spectrum characteristics are selected as the filtered signal data Q i =[q1,q2,...,q p ], then the signal data after filtering for all specified flow conditions is obtained. Q=[Q1,Q2,…,Q i ], is the signal data after filtering under the i-th flow condition, and there are i specified flow conditions in total.
[0028] In step 3.2, the parameter optimization range of VMD is: the number of modal decompositions K is 3 to 13, and the penalty factor α is 100 to 3000.
[0029] Step 4 is as follows:
[0030] Step 4.1: Transform the filtered signal data Q i It is divided into three parts: training set, validation set and test set;
[0031] Step 4.2: Construct a leakage prediction model. The structure of the leakage prediction model includes an input layer, a convolutional layer, a Chomp layer, a ReLU activation layer, a dropout layer, a residual connection, a fully connected layer, and an output layer. The Chomp layer is used to prune the boundary effects generated by the convolution operation to ensure that the output has the same shape as the input. The dropout rate of the dropout layer is 10-30% (the dropout rate indicates the proportion of neurons that do not update weights and biases in each training iteration). Training goal: The accuracy of the training set and the validation set both reach the expected accuracy: the prediction accuracy is ≥90%, or the training rounds reach the set maximum iteration rounds: 300-1000.
[0032] Step 4.3: Import the training set and test set data into the input layer of the leakage prediction model for training; the data passes through the input layer, convolution layer, Chomp layer, ReLU activation layer, drop layer, residual connection, full connection layer and output layer to complete a model training; the convolution kernel size is set to 3; the initial value of the learning rate is set to 0.001, and during the training process, the learning rate is dynamically adjusted based on the performance of the validation set; the ReLU function is used as the activation function for iterative training, and the expression of the ReLU function is shown in formula (2); then the prediction result is output at the output layer, and the leakage prediction model uses the root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 As evaluation indicators, RMSE, MAE and R 2 The calculation formulas are shown in formulas (3), (4), and (5). The difference between the evaluation prediction results of the evaluation index and the actual true results is used to confirm the prediction accuracy of the results. If the prediction accuracy reaches ≥90%, the trained leakage prediction model is saved. Otherwise, step 4.3 is continued until the training accuracy requirement is met.
[0033] The expressions of formulas (2)-(5) are as follows:
[0034]
[0035] In formula (1), c is the input vector;
[0036]
[0037]
[0038] In formulas (2), (3), and (4), n is the length of the leakage signal data, j is the time point in the sequence, and y is i is the measured pressure at time j; is the predicted pressure at time j; is the mean value of the measured pressure.
[0039] In step 4.1, the ratio of the training set, validation set and test set is 6:3:1, the training set and validation set account for 90% of the total data set; the test set accounts for 10% of the total data set.
[0040] The beneficial effects of the present invention are:
[0041] (1) The collected signal is a non-stationary and nonlinear pressure wave signal, and the signal is affected by equipment noise, environmental noise, etc., and is often difficult to be directly applied to leakage prediction. The method of the present invention uses the variational mode decomposition (VMD) method to filter the collected signal, which can effectively remove interference signals such as equipment noise and environmental noise and correctly extract leakage signal features, which can effectively improve the accuracy of the leakage prediction model.
[0042] (2) When performing variational mode decomposition (VMD) on pressure signals, it is often necessary to manually determine the number of mode decompositions K and the penalty factor α, which is unstable. If the K value is too small, the signal decomposition will be incomplete and the information contained in the signal cannot be fully revealed. If the K value is too large, too many components will be generated, resulting in signal distortion. Based on this, the method of the present invention uses a walking optimization algorithm (HOA) to optimize the variational mode decomposition algorithm (VMD), that is, constructs an HOA-VMD optimization algorithm to optimize the number of mode decompositions K and the penalty factor α, effectively improving the decomposition effect of VMD on pressure signals.
[0043] (3) When performing VMD decomposition, in order to reasonably select effective IMF components and retain leakage characteristic information, the method of the present invention uses Fourier transform (FFT) to convert the decomposed IMF component signals into frequency domain signals for spectrum analysis. Whether the spectrum of each IMF is concentrated in the target frequency range and whether the spectrum of each IMF component is well separated is used as an evaluation criterion for the VMD decomposition effect, which effectively improves the VMD decomposition effect.
[0044] (4) The pressure wave signal collected by the water pipeline is a typical time series data with complex signal characteristics, and a large amount of data is required for prediction. The temporal convolutional network (TCN) is more suitable for processing sequence data with large receptive fields and strong temporality. It uses the characteristics of dilated convolution to ensure that the model only relies on the data points at the previous time point when generating the output at any time point, which is more suitable for predicting long time series data. Therefore, the method of the present invention constructs a temporal convolutional network (TCN) prediction model.
[0045] (5) The method of the present invention has strong scalability. With the increase of input data volume and dimension and the diversification of data types, the leakage prediction accuracy of the model can be continuously improved and the scope of application can be continuously expanded.
[0046] (6) The method of the present invention can predict the leakage signal of unknown flow rate with a known flow condition leakage pressure signal with high accuracy.
[0047] (7) The method of the present invention has good versatility and can be conveniently applied to pipeline leakage detection in pipelines of various materials, under different working conditions, and under different fluid media. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 is a flow chart of the HOA-VMD algorithm in the method of the present invention;
[0049] Figure 2 is an algorithm flow chart corresponding to the leakage prediction model used in the method of the present invention;
[0050] Figure 3 is a schematic diagram of the experimental circuit in Example 1;
[0051] Figure 4 is a schematic diagram of the original signal collected by the first high-frequency pressure sensor in Example 1;
[0052] Figure 5 is a schematic diagram of the original signal collected by the second high-frequency pressure sensor in Example 1;
[0053] Figure 6 is a schematic diagram of the original signal collected by the third high-frequency pressure sensor in Example 1;
[0054] Figure 7 is a schematic diagram of the original signal collected by the fourth high-frequency pressure sensor in Example 1;
[0055] Figure 8 It is a comparison diagram of the filtered signal and the original signal in Example 1.
[0056] Fig. 9 This is a prediction effect diagram of the first high-frequency pressure sensor of the leakage prediction model in Example 1;
[0057] Fig.10 This is a prediction effect diagram of the second high-frequency pressure sensor of the leakage prediction model in Example 1;
[0058] Fig.11 This is a prediction effect diagram of the third high-frequency pressure sensor of the leakage prediction model in Example 1;
[0059] Fig.12 This is a diagram showing the prediction effect of the fourth high-frequency pressure sensor of the leakage prediction model in Example 1.
[0060] In the figure, 1. water tank, 2. first gate valve, 3. variable frequency water pump, 4. first electromagnetic flowmeter, 5. first high-frequency pressure sensor, 6. first leakage point, 7. second leakage point, 8. third leakage point, 9. fourth leakage point, 10. second high-frequency pressure sensor, 11. fifth leakage point, 12. sixth leakage point, 13. seventh leakage point, 14. third high-frequency pressure sensor, 15. eighth leakage point, 16. ninth leakage point, 17. tenth leakage point, 18. fourth high-frequency pressure sensor, 19. second electromagnetic flowmeter, 20. second gate valve, 21. data acquisition system. DETAILED DESCRIPTION
[0061] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.
[0062] The present invention provides a pipeline leakage prediction method based on variational mode decomposition and time convolutional network, such as Figure 1 As shown, the specific steps include:
[0063] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is performed by controlling the switch ball valve to simulate pipeline leakage.
[0064] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0065] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0066] The method for obtaining experimental test data in step 2 is as follows:
[0067] Step 2.1: Turn on the variable frequency water pump (no flow restriction is imposed at startup);
[0068] Step 2.2: After the pipeline system is running normally, adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition (the specified frequency corresponds to the specified flow condition, that is, the flow condition to be tested), and wait for the electromagnetic flowmeter to show stable readings (that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed); the range of the specified flow condition is 3m 3 / h~13m 3 / h(Selection requirements for specified flow conditions: at 3m 3 / h~8m 3 The interval between each two specified flow conditions in the / h interval is 2m 3 / h, at 8m 3 / h~13m 3 The interval between each two specified flow conditions in the / h interval is 1.5m 3 / h).
[0069] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording.
[0070] Step 2.4: After recording a single leak point, apply step 2.3 to new leak points in sequence until all leak points are tested;
[0071] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0072] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0073] Step 3: Construct a walking optimization algorithm-optimized variational mode decomposition (HOA-VMD) algorithm, then perform mode decomposition and signal reconstruction on the leakage experimental test data obtained in step 2, and filter the experimental test data;
[0074] Step 3 is as follows:
[0075] Step 3.1: Initialize the parameters of the walking optimization algorithm (HOA), that is, initialize the position and fitness of the traveler group. The best travel position of the traveler corresponds to the parameter combination [K, α]. Among them, the HOA parameters are set as follows: the number of people in the travel group is 10 to 100, and the maximum number of iterations is 20 to 500; the fitness function selects the envelope entropy, and the calculation formula of the envelope entropy E is shown in formula (1);
[0076]
[0077] Where n represents the number of the discretized envelope signal interval; N is the number of discretized envelope signal intervals; p n is the probability that the envelope signal falls in the nth interval, which is obtained by calculating the ratio of the number of sample points in each interval to the total number of sample points;
[0078] Step 3.2: Use the HOA algorithm to optimize the combination of VMD decomposition parameters [K, α]. The smaller the envelope entropy value, the better the combination of [K, α]. The parameter optimization range of VMD is: the number of modal decompositions K is 3 to 13, and the penalty factor α is 100 to 3000. Then, the high-frequency pressure signal data S = [S1, S2, …, S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition. After decomposition, S i Each signal data s in p K IMF components and one residual component will be obtained, and the time domain signals of the decomposed IMF components are named IMF1, IMF2, ..., IMFK in sequence;
[0079] Step 3.3: Use Fourier transform (FFT) to convert the time domain signal of each IMF component into a frequency domain signal, draw the spectrum of each IMF component, and verify whether the spectrum of each IMF component is concentrated in the target frequency interval. The judgment standard is: if the cumulative spectrum energy in the target frequency interval is not less than 80% of the total energy, it means that the spectrum of the IMF component is concentrated in the target frequency interval; then judge whether the spectrum of different IMF components is well separated. The judgment standard is: if the spectrum overlap rate between different IMF components is less than 20%, it is considered to be well separated;
[0080] If the IMF component spectrum is concentrated in the target frequency interval and is well separated, it means that the optimization effect is good; if the spectrum energy overlap of the IMF component exceeds 30% (that is, the frequency interval where the energy is concentrated has significant overlap) or the frequency range covered by the spectrum distribution of a certain IMF component exceeds 500Hz, it means that the parameters of VMD decomposition are improperly set, and the optimization result is discarded, and step 3.2 is repeated until the stop condition of good IMF separation effect is met, then the algorithm process is completed, the optimization is stopped, and the optimal parameter combination [K, α] is output;
[0081] Step 3.4: Use the optimal parameter combination [K, α] obtained by the optimization to test the high-frequency pressure signal data S = [S1, S2, ..., S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition, and the IMF component signals with a similarity of more than 90% with the original signal spectrum characteristics are selected as the filtered signal data Q i =[q1,q2,...,q p ], then the signal data after filtering for all specified flow conditions is obtained. Q=[Q1,Q2,…,Q i ], is the signal data after filtering under the i-th flow condition, and there are i specified flow conditions in total.
[0082] Step 4: Build a temporal convolutional network (TCN) pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy;
[0083] like Figure 2 As shown, step 4 is specifically as follows:
[0084] Step 4.1: Transform the filtered signal data Q i It is divided into three parts: training set, validation set and test set. The ratio of training set, validation set and test set is 6:3:1. The training set and validation set account for 90% of the total data set; the test set accounts for 10% of the total data set.
[0085] Step 4.2: Construct a leakage prediction model. The structure of the leakage prediction model includes an input layer, a convolutional layer, a Chomp layer, a ReLU activation layer, a dropout layer, a residual connection, a fully connected layer, and an output layer. The Chomp layer is used to prune the boundary effects generated by the convolution operation to ensure that the output has the same shape as the input. The dropout rate of the dropout layer is 10-30% (the dropout rate indicates the proportion of neurons that do not update weights and biases in each training iteration). Training goal: The accuracy of the training set and the validation set both reach the expected accuracy: the prediction accuracy is ≥90%, or the training rounds reach the set maximum iteration rounds: 300-1000.
[0086] Step 4.3: Import the training set and test set data into the input layer of the leakage prediction model for training; the data passes through the input layer, convolution layer, Chomp layer, ReLU activation layer, drop layer, residual connection, full connection layer and output layer to complete a model training; the convolution kernel size is set to 3; the initial value of the learning rate is set to 0.001, and during the training process, the learning rate is dynamically adjusted based on the performance of the validation set; the ReLU function (expression as shown in formula (2)) is used as the activation function for iterative training; then the prediction results are output at the output layer. The leakage prediction model uses the root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 As evaluation indicators, RMSE, MAE and R 2 The calculation formulas are shown in formulas (3), (4), and (5). The evaluation index is used to evaluate the difference between the predicted result and the actual result to confirm the prediction accuracy of the result. If the prediction accuracy reaches ≥90%, the trained leakage prediction model is saved. Otherwise, step 4.3 is continued until the training accuracy requirement is met.
[0087] The expressions of formulas (2)-(5) are as follows:
[0088]
[0089] In formula (1), c is the input vector;
[0090]
[0091]
[0092] In formulas (2), (3), and (4), n is the length of the leakage signal data, j is the time point in the sequence, and y is i is the measured pressure at moment j; is the predicted pressure at time j; is the mean value of the measured pressure.
[0093] Example 1
[0094] In order to demonstrate the practical application effect of the method of the present invention and more specific practical application steps, the present invention built a pressurized water pipeline leakage detection experimental platform, obtained corresponding experimental test data according to the steps of the present invention, and successfully constructed a leakage prediction model. Based on Example 1, the application of the present invention is demonstrated in detail.
[0095] According to step 1, an experimental test was conducted on a pressurized water transmission leakage detection experimental platform equipped with a leakage test point, and eight flow conditions (3m 3 / h、5m 3 / h、7m 3 / h、8m3 / h、9.5m 3 / h、10.5m 3 / h、11.5m 3 / h、13m 3 / h) under 10 simulated leakage points, a total of 240 sets of data were obtained, each set of data contained 4-dimensional signal data (i.e., 4 high-frequency pressure sensor data). The schematic diagram of the experimental circuit of Example 1 is shown in FIG. Figure 3 As shown, it includes a water tank 1, a first gate valve 2, a variable frequency water pump 3, a first electromagnetic flowmeter 4, a first high-frequency pressure sensor 5, a first leakage point 6, a second leakage point 7, a third leakage point 8, a fourth leakage point 9, a second high-frequency pressure sensor 10, a fifth leakage point 11, a sixth leakage point 12, a seventh leakage point 13, a third high-frequency pressure sensor 14, an eighth leakage point 15, a ninth leakage point 16, a tenth leakage point 17, a fourth high-frequency pressure sensor 18, a second electromagnetic flowmeter 19, and a second gate valve 20, wherein a ball valve is installed at each leakage test point; it also includes a data acquisition system 21, and the first electromagnetic flowmeter 4, the first high-frequency pressure sensor 5, the second high-frequency pressure sensor 10, the third high-frequency pressure sensor 14, the fourth high-frequency pressure sensor 18 and the second electromagnetic flowmeter 19 are all connected to the data acquisition system 21. The experimental loop uses a 36mm inner diameter, 8mm wall thickness three-propylene polypropylene tube (PP-R) to connect various experimental components, the inner diameter of the leakage point is 8mm, and the sampling frequency is 1000Hz.
[0096] The experimental data acquisition method of the embodiment is the same as step 2. The sampling point length of the pressure signal data under different flow conditions is 20000, and the sampling frequency is 1000. The main working condition parameters of the experiment are: pressure 100-500 kPa, flow 3-13 m 3 / h. Figure 4-Figure 7 8m 3 Schematic diagram of the original leakage signal collected by four high-frequency pressure sensors at 1000 rpm / h. When a leak occurs, the pipeline leak will cause the pressure signal to drop instantly, and then the signal will recover to a stable value slightly lower than the level before the leak. The four high-frequency pressure sensors of the pressurized water transmission leak detection experimental platform built by the invention can clearly detect the fluctuation characteristics of the pressure signal when a leak occurs, laying a data foundation for the construction of the leak prediction model.
[0097] According to step 3, the HOA-VMD algorithm is used to adaptively iterate and optimize the key parameters of variational mode decomposition (VMD), "number of modal decompositions K" and "penalty factor α", and the optimal result is K = 7, α = 1547. The experimental test data obtained in step 2 are filtered and reconstructed using the obtained optimal parameters, such as Figure 8As shown in the figure, the filtered signal significantly reduces the random fluctuations in the original signal data, making the pressure curve smoother, making it easier to observe trends and abnormal points, and still effectively capturing significant changes in pressure;
[0098] According to step 4, the leakage prediction model is constructed: divided into three parts: training set, validation set and test set. The ratio of training set and validation set to test set is 6:3:1, and the training set and validation set account for 90% of the total data set; the test set accounts for 10% of the total data set, including 216 groups of data and 24 groups of data respectively; the parameters of the model are set as follows: the discard rate of the discard layer is 20%; the initial value of the expansion rate is set to 2; the maximum number of iterations is 500; the prediction accuracy target is 90%. The prediction effects of the trained leakage prediction model on the 4 high-frequency pressure sensors of the test set data are as follows Fig. 9 , Fig.10 , Fig.11 and Fig.12 As shown (the middle 12,000 points of forecast data are intercepted to facilitate the display of forecast results), the determination coefficient R between the forecast value and the actual value 2 All of them are greater than 0.9, indicating that the leakage prediction model constructed by the present invention has good prediction effect and generalization ability. It can be seen that the method can achieve effective extraction of leakage pressure signal features by adaptively selecting variational mode decomposition (VMD) decomposition parameters using the walking optimization algorithm (HOA); the method uses the pressure signal data after feature extraction to train the constructed time convolution network (TCN) leakage prediction model, and achieves accurate prediction of leakage pressure signal features under different flow rates in pressurized water pipelines.
[0099] Example 2
[0100] The pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network specifically includes the following steps:
[0101] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high-frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is performed by simulating pipeline leakage by controlling the switch ball valve; in step 1, the variable frequency water pump is set at the outlet of the water tank to transport a fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for pipeline system safety protection; the electromagnetic flowmeter is respectively set after the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0102] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0103] Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data;
[0104] Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
[0105] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0106] Example 3
[0107] The pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network specifically includes the following steps:
[0108] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high-frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is performed by simulating pipeline leakage by controlling the switch ball valve; in step 1, the variable frequency water pump is set at the outlet of the water tank to transport a fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for pipeline system safety protection; the electromagnetic flowmeter is respectively set after the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0109] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0110] Step 2 is as follows:
[0111] Step 2.1: Turn on the variable frequency water pump;
[0112] Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h;
[0113] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording.
[0114] Step 2.4: After recording a single leak point, apply step 2.3 to new leak points in sequence until all leak points are tested;
[0115] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0116] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0117] Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data;
[0118] Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
[0119] Example 4
[0120] The pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network specifically includes the following steps:
[0121] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is conducted by controlling the switch ball valve to simulate pipeline leakage;
[0122] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0123] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0124] Step 2 is as follows:
[0125] Step 2.1: Turn on the variable frequency water pump;
[0126] Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h;
[0127] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording.
[0128] Step 2.4: After recording a single leak point, apply step 2.3 to the new leak point in sequence until all leak points are tested;
[0129] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0130] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0131] Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data;
[0132] Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
[0133] Example 5
[0134] The pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network specifically includes the following steps:
[0135] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is conducted by controlling the switch ball valve to simulate pipeline leakage;
[0136] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0137] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0138] Step 2 is as follows:
[0139] Step 2.1: Turn on the variable frequency water pump;
[0140] Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h;
[0141] In step 2.2, the selection requirements for the specified flow conditions are: 3 / h~8m 3 The interval between each two specified flow conditions in the / h interval is 2m 3 / h, at 8m 3 / h~13m 3 The interval between each two specified flow conditions in the / h interval is 1.5m 3 / h.
[0142] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording.
[0143] Step 2.4: After recording a single leak point, apply step 2.3 to new leak points in sequence until all leak points are tested;
[0144] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0145] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0146] Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data;
[0147] Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
[0148] Example 6
[0149] The pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network specifically includes the following steps:
[0150] Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is conducted by controlling the switch ball valve to simulate pipeline leakage;
[0151] In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
[0152] Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions;
[0153] Step 2 is as follows:
[0154] Step 2.1: Turn on the variable frequency water pump;
[0155] Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h;
[0156] In step 2.2, the selection requirements for the specified flow conditions are: 3 / h~8m 3 The interval between each two specified flow conditions in the / h interval is 2m 3 / h, at 8m 3 / h~13m 3 The interval between each two specified flow conditions in the / h interval is 1.5m 3 / h.
[0157] Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system. The signal data acquisition time is 10 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times. Before each recording, the pressure and flow readings are waited to be stable.
[0158] Step 2.4: After recording a single leak point, apply step 2.3 to the new leak point in sequence until all leak points are tested;
[0159] Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total;
[0160] Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
[0161] Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data;
[0162] Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
Claims
1. A pipeline leakage prediction method based on variational mode decomposition and temporal convolutional network, characterized in that: The specific steps include: Step 1: In a pressurized water pipeline system equipped with a variable frequency water pump, a high frequency pressure sensor, an electromagnetic flowmeter, a gate valve, a ball valve and a water tank, a leakage test experiment is conducted by controlling the switch ball valve to simulate pipeline leakage; Step 2: Conduct pipeline leakage test research in the pressurized water pipeline system of step 1 to obtain experimental leakage test data at different leakage locations under different flow conditions; Step 3: Construct a walking optimization algorithm to optimize the variational mode decomposition algorithm, then perform mode decomposition on the leakage experimental test data obtained in step 2 and reconstruct the signal, and filter the experimental test data; Step 4: Build a temporal convolutional network pipeline leakage prediction model, and use the filtered data obtained in step 3 to train the model to improve the model training accuracy.
2. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 1 is characterized in that: In step 1, a variable frequency water pump is set at the outlet of the water tank to transport fluid with a specific flow rate, and the flow rate is adjusted by adjusting the frequency of the variable frequency pump; the installation requirements of the high-frequency pressure sensor and the ball valve are as follows: a high-frequency pressure sensor is installed every 2m to 1000m, and the sampling frequency is 500Hz to 10000Hz; 2 to 10 leakage test points are set between every two high-frequency pressure sensors, and a ball valve is installed at each leakage test point, and the pipeline leakage event is simulated by switching the ball valve; the installation requirements of the gate valve are as follows: gate valves need to be installed at the outlet of the water tank and the end of the pipeline system for the safety protection of the pipeline system; the electromagnetic flowmeter is respectively set behind the variable frequency water pump and at the end of the pipeline system to measure the inlet and outlet fluid flow in the pipeline.
3. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 1 is characterized in that: Step 2 is as follows: Step 2.1: Turn on the variable frequency water pump; Step 2.2: Adjust the frequency of the variable frequency pump to the specified frequency so that the fluid flow in the pipeline reaches the specified flow condition and the electromagnetic flowmeter reading is stable, that is, the change amplitude of the average value of the displayed pressure and flow is less than 1% when no leakage test is performed; the range of the specified flow condition is 3m 3 / h~13m 3 / h; Step 2.3: Conduct a water pipeline leakage test and record a set of high-frequency pressure signal data S under the flow conditions that need to be tested. i , S i =[s1,s2,...,s p ], p is the number of high-frequency pressure sensors in the pressurized water pipeline system, and the signal data acquisition time is 10 seconds to 60 seconds. When the recording time reaches half of the acquisition time, a leak point ball valve is opened to simulate pipeline leakage until the end of the recording. After the recording is completed, the ball valve is closed. One leak point is recorded three times, and the pressure and flow readings are waited for stability before each recording. Step 2.4: After recording a single leak point, apply step 2.3 to the new leak point in sequence until all leak points are tested; Step 2.5: Change the fluid flow rate to the next specified flow condition by adjusting the frequency of the variable frequency water pump, and repeat steps 2.3 and 2.4 until the high-frequency pressure signal data S = [S1, S2, ..., S i ],S i is the experimental test data obtained under the i-th flow condition, and there are i specified flow conditions in total; Step 2.6: After completing the operations in step 2.5 under all specified flow conditions, turn off the variable frequency water pump and end the experiment.
4. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 1 is characterized in that: In step 2.2, the selection requirements for the specified flow conditions are: 3 / h~8m 3 The interval between each two specified flow conditions in the / h interval is 2m 3 / h, at 8m 3 / h~13m 3 The interval between each two specified flow conditions in the / h interval is 1.5m 3 / h.
5. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 1 is characterized in that: Step 3 is as follows: Step 3.1: Initialize the parameters of the walking optimization algorithm, that is, initialize the position and fitness of the traveler group. The best travel position of the traveler corresponds to the parameter combination [K, α]. Among them, the HOA parameters are set as follows: the number of travel groups is 10 to 100, and the maximum number of iterations is 20 to 500. The fitness function selects the envelope entropy. The calculation formula of the envelope entropy E is shown in formula (1). Where n represents the number of the discretized envelope signal interval; N is the number of discretized envelope signal intervals; p n is the probability that the envelope signal falls within the nth interval; Step 3.2: Use the HOA algorithm to optimize the combination of VMD decomposition parameters [K, α]. The smaller the envelope entropy value, the better the combination of [K, α]. Then, the high-frequency pressure signal data S = [S1, S2, …, S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition. After decomposition, S i Each signal data s in p K IMF components and one residual component will be obtained, and the time domain signals of the decomposed IMF components are named IMF1, IMF2, ..., IMFK in sequence; Step 3.3: Use Fourier transform to convert the time domain signal of each IMF component into a frequency domain signal, draw the spectrum of each IMF component, and verify whether the spectrum of each IMF component is concentrated in the target frequency interval. The judgment standard is: if the cumulative spectrum energy in the target frequency interval is not less than 80% of the total energy, it means that the spectrum of the IMF component is concentrated in the target frequency interval; then judge whether the spectrum of different IMF components is well separated. The judgment standard is: if the spectrum overlap rate between different IMF components is less than 20%, it is considered to be well separated; If the IMF component spectrum is concentrated in the target frequency range and is well separated, it means that the optimization effect is good; if the spectral energy overlap of the IMF component exceeds 30% or the frequency range covered by the spectrum distribution of a certain IMF component exceeds 500Hz, it means that the parameters of VMD decomposition are improperly set, and the optimization result is discarded, and step 3.2 is repeated until the stop condition of good IMF separation effect is met, then the algorithm process is completed, the optimization is stopped, and the optimal parameter combination [K, α] is output; Step 3.4: Use the optimal parameter combination [K, α] obtained by the optimization to test the high-frequency pressure signal data S = [S1, S2, ..., S i ] in the pressure signal data S i =[s1,s2,...,s p ] are all subjected to VMD decomposition, and the IMF component signals with a similarity of more than 90% with the original signal spectrum characteristics are selected as the filtered signal data Q i =[q1,q2,...,q p ], then the signal data after filtering for all specified flow conditions is obtained. Q=[Q1,Q2,…,Q i ], is the signal data after filtering under the i-th flow condition, and there are i specified flow conditions in total.
6. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 5 is characterized in that: In step 3.2, the parameter optimization range of VMD is: the number of modal decompositions K is 3 to 13, and the penalty factor α is 100 to 3000.
7. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 1 is characterized in that: Step 4 is as follows: Step 4.1: Transform the filtered signal data Q i It is divided into three parts: training set, validation set and test set; Step 4.2: Construct a leakage prediction model. The structure of the leakage prediction model includes input layer, convolution layer, Chomp layer, ReLU activation layer, drop layer, residual connection, fully connected layer and output layer. The Chomp layer is used to trim the boundary effect generated by the convolution operation to ensure that the output has the same shape as the input. The drop rate of the drop layer is 10-30%. Training goal: The accuracy of the training set and the validation set both reach the expected accuracy: prediction accuracy ≥ 90%, or the training rounds reach the set maximum iteration rounds: 300-1000. Step 4.3: Import the training set and test set data into the input layer of the leakage prediction model for training; the data passes through the input layer, convolution layer, Chomp layer, ReLU activation layer, drop layer, residual connection, full connection layer and output layer to complete a model training; the convolution kernel size is set to 3; the initial value of the learning rate is set to 0.001, and during the training process, the learning rate is dynamically adjusted based on the performance of the validation set; the ReLU function is used as the activation function for iterative training, and the expression of the ReLU function is shown in formula (2); then the prediction result is output at the output layer, and the leakage prediction model uses the root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 As evaluation indicators, RMSE, MAE and R 2 The calculation formulas are shown in formulas (3), (4), and (5). The difference between the evaluation prediction results of the evaluation index and the actual true results is used to confirm the prediction accuracy of the results. If the prediction accuracy reaches ≥90%, the trained leakage prediction model is saved. Otherwise, step 4.3 is continued until the training accuracy requirement is met. The expressions of formulas (2)-(5) are as follows: In formula (1), c is the input vector; In formulas (2), (3), and (4), n is the length of the leakage signal data, j is the time point in the sequence, and y is i is the measured pressure at moment j; is the predicted pressure at time j; is the mean value of the measured pressure.
8. The pipeline leakage prediction method based on variational mode decomposition and time convolution network according to claim 7 is characterized in that: In step 4.1, the ratio of the training set, validation set and test set is 6:3:1, the training set and validation set account for 90% of the total data set; the test set accounts for 10% of the total data set.