Multi-dimensional heat distribution pipeline monitoring system and method based on integrated monitoring unit

Abstract

The invention relates to a multi-dimensional heat distribution pipeline monitoring system and method based on an integrated monitoring unit, and the system comprises an underwater sound sensing module which employs piezoelectric ceramics as a transduction material, deforms under the sound pressure effect to generate an electric signal, converts the electric signal into an output signal through an amplification circuit, and transmits the output signal to a data receiving and transmitting module; the high-frequency pressure module and the high-frequency pressure sensing module adopt an MEMS (Micro Electro Mechanical System) process to directly manufacture a strain resistor on a stainless steel isolation diaphragm as a resistance element, and the isolation diaphragm is stressed to generate an electric signal which is transmitted to the data receiving and transmitting module through a connecting cable after being amplified; the temperature sensing module outputs an electric signal based on the characteristic that a metal resistor changes along with temperature, and the electric signal is amplified and then transmitted to the data receiving and transmitting module through a connecting cable. According to the invention, the daily operation environment and the abnormal environment of the pipeline can be distinguished, and the dependence on the material and environment of the pipeline is reduced, so that the sensing monitoring unit has higher adaptability, and leakage can be better judged.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of pipeline monitoring, and particularly relates to a multi-dimensional heat pipe monitoring system and method based on an integrated monitoring unit. BACKGROUND

[0002] Currently, the sensitivity of the hydrophone is low, which makes it difficult to capture weak acoustic signals. In addition, the adaptability to different material pipes (such as metal and plastic) and complex environments (such as high noise and multi-branch pipes) is poor. If the complete pipe acoustic signal cannot be obtained, it cannot provide support for subsequent leakage judgment and positioning analysis.

[0003] The currently used monitoring system monitors the collected water sound, pressure, temperature and water contact information through conventional experience samples. However, during the operation of the pipeline, different pipe diameters, operating pressures and laying environments will have different effects on the data, and it is impossible to analyze them through general empirical constants.

[0004] At the same time, the currently used pipeline monitoring equipment cannot monitor the operating environment of the equipment, which may cause the monitoring signal to have data shock caused by changes in the water contact environment, and the reason cannot be analyzed, resulting in false alarms.

[0005] In addition, the currently used pipeline monitoring method is mostly single dimension, such as monitoring leakage and pipe burst from a single water sound dimension, monitoring leakage and pipe burst from a single pressure dimension, and monitoring pipeline leakage from a single temperature change dimension. Since the size of the pipeline leakage has no linear relationship with the acoustic signal, the degree of pipeline leakage cannot be judged by the acoustic signal, that is, the differentiation between leakage and pipe burst cannot be realized. Since the pressure wave generated by the leakage does not have the characteristic of persistence, if the leakage is not captured, the subsequent pipeline water pressure will gradually rise and form a steady state. Therefore, it is easy to miss the peak of the pressure characteristics formed instantaneously, resulting in poor positioning accuracy and the possibility of missing reports. The principle of pipeline leakage monitoring by a single temperature sensor is that the loss of high-temperature medium in the pipeline causes the temperature in the pipeline to drop, and when the temperature setting threshold is exceeded, the leakage alarm is triggered. This method cannot avoid the influence of insufficient heating of the pipeline itself, which may cause false alarms.

[0006] And the prior art usually adopts a sound bar and a correlator to manually judge suspected leakage points, and the sound bar and the correlator are used to monitor pipeline leakage, usually obvious leakage is found on the road surface, and the leakage has occurred for a long time, and losses have been caused. In addition, the detection of leakage points by the two instruments requires a high requirement for the inspection workers, and the workers need to have certain leakage detection experience and skilled use of the instruments. Moreover, the water overflow position on the road surface is not necessarily the same as the leakage position, and there may be a distance of hundreds of meters, so the leakage position may be difficult to find in a short time.

[0007] Especially, the heat supply pipeline is usually buried, and the heat supply pipeline is not easy to stop water for maintenance during operation. It is a technical problem to be solved how the pipeline operation management department can ensure the safety of the pipeline and timely find the pipeline leakage and the abnormal pressure in the pipeline. SUMMARY

[0008] The technical problem to be solved by the present application is to provide a multi-dimensional heat pipeline monitoring system and method based on an integrated monitoring unit. The present application can distinguish between the daily operation environment and the abnormal environment of the pipeline by using a high-sensitivity monitoring module and a monitoring model based on a twin neural network, reduce the dependence on the pipeline material and the environment, make the sensing monitoring unit have higher adaptability, and better judge the leakage.

[0009] The technical solution of the present application is to provide a multi-dimensional heat pipeline monitoring system based on an integrated monitoring unit, which comprises

[0010] An underwater acoustic sensing module, which uses piezoelectric ceramic as a transducer material and generates an electric signal under the action of sound pressure and through an amplification circuit to convert the output signal into an output signal transmitted to a data transceiver module;

[0011] A high-frequency pressure module, which uses a MEMS process to directly manufacture a strain resistor on a stainless steel isolation diaphragm as a resistor element, and the isolation diaphragm generates an electric signal under stress and transmits the signal to the data transceiver module after amplification through a connecting cable;

[0012] A temperature sensing module, which outputs an electric signal based on the temperature change characteristic of a metal resistor, and transmits the electric signal to the data transceiver module through a connecting cable after amplification;

[0013] A water contact detection module, which triggers a detection circuit based on the conductivity of a liquid, and transmits the detection circuit result to the data transceiver module through a connecting cable in the form of on / off;

[0014] A data transceiver module, which is used to collect the electric signals of the underwater acoustic module, the high-frequency pressure module, the temperature sensing module and the water contact sensing module;

[0015] The host computer: through the management platform, all monitoring stations on the pipeline are uniformly managed, and the integrated monitoring unit equipment of the monitoring end is uniformly managed, remote parameter configuration and remote monitoring are realized.

[0016] As preferred, the host computer is built-in with a leakage detection model based on a twin neural network, combined with an evaluation index of mel-frequency cepstral coefficient (MFCC) and Pearson correlation coefficient (PCC) and structural similarity (SSIM), and a pipeline anomaly judgment model based on multiple parameters to improve the accuracy of monitoring leakage.

[0017] The application also provides a monitoring method of the multi-dimensional heat pipeline monitoring system based on the integrated monitoring unit, the underwater acoustic sensing module, the high-frequency pressure module, the temperature sensing module and the water contact sensing module collect corresponding underwater acoustic, pressure, temperature and water contact data, and then send the data to the host computer through the data transceiver module, the host computer is built-in with a monitoring model based on a twin neural network, a pipeline anomaly judgment model based on multiple parameters and an evaluation method combined with mel-frequency cepstral coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) to judge the pipeline anomaly and push the corresponding alarm.

[0018] As preferred, the monitoring model based on the twin neural network realizes accurate identification of abnormal acoustic signals by constructing the intrinsic feature space of the normal state, which includes four stages of data preprocessing, feature learning, dynamic detection and model evolution, wherein in the data preprocessing stage, a physical constraint signal enhancement method is used to construct a training sample set, and the specific operation is as follows, after eliminating the pump vibration noise through a 20-500Hz band-pass filter, the underwater acoustic signal is subjected to sliding window standardization processing (window 200ms, step 50ms) to suppress the time-varying drift of the sensor.

[0019] As preferred, to solve the problem of leakage sample scarcity, a simulation enhancement strategy based on an acoustic propagation model is also included: in the normal signal, a leakage pulse waveform with an amplitude of 0.1-0.3 times the fundamental signal and a frequency deviation of ±50Hz is injected, and noise samples containing valve action and water flow impact are superimposed to construct a mixed data set with physical rationality.

[0020] As preferred, the feature learning stage adopts a parameter-shared twin network architecture, and a compact feature space of normal state is established through a contrastive learning mechanism. In the network structure, three layers of depth separable convolution (32 / 64 / 128 3×1 kernels) are deployed to capture the characteristics of the longitudinal propagation of sound waves while reducing the computational complexity; an overlapping max-pooling layer (window 3, step 2) is used to retain local extreme features and suppress random noise; and in the retraining process, the Euclidean distance between normal sample pairs in the 128-dimensional embedding space is constrained to converge to 0.2±0.05, and the distance between normal and abnormal sample pairs is expanded to more than 1.5±0.3, so as to form significant separability.

[0021] As preferred, the dynamic detection stage constructs a pressure-adaptive double-threshold discrimination mechanism, and a sliding reference feature library is established. The moving average features of the normal samples in the last 24 hours are updated every 5 minutes to eliminate the influence of working condition drift. The K-nearest neighbor dynamic time warping distance (K=5, window 50) between the input signal and the reference library is calculated in real time, and the sound intensity-pressure coupling calibration is performed in combination with the pressure sensor reading. The threshold setting adopts a composite strategy of a basic value (μ+3σ) and a dynamic compensation term (0.1×dP / dt), and the alarm is triggered when three consecutive sampling points (interval 0.2 seconds) exceed the threshold, ensuring that the detection accuracy is maintained at more than 95% under pressure transient conditions.

[0022] As preferred, the evaluation method combining Mel frequency cepstral coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) is as follows. After obtaining the audio data through the integrated terminal, the time-frequency spectrum is abstracted by the Mel frequency cepstral coefficient, and the time-frequency spectrum will be presented as a two-dimensional matrix. The linear relationship strength between the two matrices is quantified by the Pearson correlation coefficient, with a value range of -1 to 1, where 1 represents complete correlation. Finally, the structural similarity is used as an index to measure the similarity of two images, with a value range of 0 to 1, where 1 indicates that the two images are completely identical.

[0023] As preferred, the multi-parameter-based pipeline anomaly judgment model combines water sound data, pressure data and temperature data in three dimensions to distinguish between water hammer, leakage, burst and temperature loss events in the pipeline. After the host computer receives all the data, the temperature data, pressure data and water sound data in the pipeline are first extracted, including temperature drop and temperature gradient for temperature data, pressure drop, pressure fluctuation and pressure transient related features for pressure data, and high-frequency acoustic signal, low-frequency acoustic signal and acoustic signal intensity related features for water sound data.

[0024] As preferred, when water hammer occurs in the pipeline, the high-frequency pressure module detects the pressure transient signal, and when the pressure, water sound and temperature data meet the water hammer characteristics, it is determined as a water hammer event.

[0025] When a leak occurs in the pipeline, the high-frequency pressure module detects a pressure drop trend, the underwater acoustic sensing module captures the acoustic characteristics of the leak, and the built-in monitoring model based on the twin neural network and the evaluation method combining the Mel-frequency cepstrum coefficient, the Pearson correlation coefficient, and the structural similarity are used to locate the leak point. When the data of pressure and underwater sound meet the characteristics of the leak, it is determined as a leak event;

[0026] When a pipe burst event occurs in the pipeline, the high-frequency pressure module detects an abnormal pressure, the underwater acoustic sensing module captures the acoustic characteristics of the pipe burst, and the temperature sensor detects an abnormal temperature. When the data of temperature, pressure, and underwater sound all meet the characteristics of the pipe burst, it is determined as a pipe burst event.

[0027] When a temperature loss event occurs in the pipeline, the temperature gradient change is used to determine the temperature loss event in the pipeline.

[0028] After determining the pipe burst, leak, and water hammer event, the touch water module is used to detect whether each sensing module is in a normal operating environment. If the touch water sensing module detects that each module has deviated from the normal detection water level environment, all types of alarms are shielded, and the non-touch water alarm is pushed forward.

[0029] Compared with the prior art, the present application has the following advantages:

[0030] The present application integrates underwater acoustic monitoring, pressure monitoring, temperature monitoring, and water level monitoring into four dimensions, and can couple pressure, underwater sound, and temperature data to realize pipeline monitoring. The multi-dimensional heat pipe monitoring system based on the city monitoring unit can monitor and evaluate leaks, water hammer, temperature loss, and pipe bursts through the monitoring model based on the twin neural network, the multi-parameter pipeline anomaly judgment model, and the evaluation method combining the Mel-frequency cepstrum coefficient (MFCC), the Pearson correlation coefficient (PCC), and the structural similarity (SSIM), realize rapid response and high-precision positioning of leaks, and monitor the contact of each sensing device with water through the built-in touch water sensor. When the gas accumulation in the sensing unit branch of the pipeline is too much, the touch water sensor can automatically alarm, helping to solve the problem of false alarm of the sensor information after the sensor is away from the water. Moreover, it can realize 24h real-time online monitoring, respond quickly to pipeline leakage and temperature anomalies, and seamlessly connect with the on-duty personnel to repair the leakage area in time. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The system block diagram of the present application.

[0032] Figure 2 The monitoring unit layout of the present application.

[0033] Figure 3 A schematic diagram based on a twin neural network model output of an embodiment of the present application.

[0034] Figure 4 A Pearson correlation coefficient of an embodiment of the present application.

[0035] Figure 5 Structural similarity of an embodiment of the present application.

[0036] Figure 6 A flowchart of an evaluation process combining a mel-frequency cepstral coefficient, a Pearson correlation coefficient and structural similarity of an embodiment of the present application.

[0037] Figure 7 A pipeline anomaly judgment model structure diagram based on multiple parameters of an embodiment of the present application. DETAILED DESCRIPTION

[0038] The present application will be further described in the specific embodiments in conjunction with the accompanying drawings:

[0039] The heat supply pipeline is mostly buried and is not easy to stop water for maintenance during operation. For such pipelines, the operation and management department needs to combine manual inspection with pipeline monitoring stations to ensure pipeline safety and timely detect pipeline leakage and pressure anomalies in the pipeline.

[0040] The present application provides a multi-dimensional heat supply pipeline monitoring system based on an integrated monitoring unit, as shown in Figures 1-2 The integrated monitoring unit includes a water sound monitoring module, a high-frequency pressure monitoring module, a water contact monitoring module, a temperature monitoring module and a data transceiver module. The external power supply unit supplies power to the integrated monitoring unit.

[0041] The water sound monitoring module is a hydrophone that uses piezoelectric ceramic as the transducer material. The ceramic tube deforms under the action of sound pressure, and generates a corresponding weak voltage change through its piezoelectric effect. It can respond to omnidirectional acoustic signals in the range of 20Hz-20kHz in the pipeline and form an electric signal. The electric signal will pass through an amplification circuit before entering the data transceiver module. The amplification circuit converts the high impedance of the piezoelectric element into low impedance, greatly reducing the coupling loss caused by long cables and reducing electromagnetic interference during transmission. The weak voltage is amplified by the amplification circuit and converted into an output signal. These electric signals are amplified and transmitted to the data transceiver module through the connecting cable for preliminary processing.

[0042] The high-frequency pressure module collects signals by high frequency. Because the propagation speed of pressure wave in the pipeline is about 1000 m / s, the sampling frequency is 128 Hz to collect the pressure wave signals in the pipeline and convert the pressure wave signals into electrical signals. In this embodiment, the high-frequency pressure module is a pressure sensor. The strain resistance is directly made on the stainless steel isolation diaphragm as a resistance element by MEMS technology. When the isolation diaphragm is forced to produce elastic deformation, the resistance on the surface changes accordingly, forming a linear relationship of electrical signals. These electrical signals are amplified and transmitted to the data transceiver module for preliminary processing through the connecting cable.

[0043] The water-touch monitoring module is a water-touch sensor composed of two electrodes with a certain length. The electrode length is flush with the minimum water line that can ensure the normal work of each sensor in the pipeline. When the water level in the pipeline is lower than the minimum water line, the sensor data will be automatically shielded to ensure the accuracy of the data. Based on the electrical conductivity of the liquid, the water-touch sensor is composed of two electrodes. When the liquid contacts the electrodes, the resistance between the electrodes will decrease significantly, thereby triggering the detection circuit. The detection circuit result is transmitted to the data transceiver module for preliminary processing through the connecting cable in the form of on / off.

[0044] The temperature monitoring module is a conventional platinum resistance sensor, which is used to monitor the temperature change of the pipeline and serves as a parameter into the leakage composite judgment model as an auxiliary; all information is converted into electrical signals, and the output electrical signals change linearly with the resistance. These electrical signals are amplified and transmitted to the data transceiver module for preliminary processing through the connecting cable.

[0045] The data transceiver module is responsible for collecting the electrical signals of the hydrophone module, the high-frequency pressure module, the temperature sensor module and the water-touch sensor module. The collected data is transmitted to the monitoring center through 4G wireless communication or wired optical fiber. The RTU supports remote configuration and monitoring of sensor parameters to improve operation and maintenance efficiency. The RTU has built-in flash memory that can store 30 days of data to ensure that data is not lost. It supports analog and audio signal input and provides sensor power supply. High-precision acquisition: the pressure acquisition precision is greater than or equal to 16 bits, which can provide high-precision data.

[0046] The upper computer unifies the management of all monitoring stations on the pipeline through the management platform, and uniformly manages the integrated monitoring unit equipment of the monitoring end, realizes remote parameter configuration and remote monitoring, etc., and divides different areas and management permissions of users through the platform end to meet the needs of multi-user monitoring and management of the system. And through the leakage detection model based on the twin neural network built in the upper computer software, combined with the evaluation index of mel frequency cepstrum coefficient MFCC and pearson correlation coefficient (PCC) and structural similarity (SSIM) and the pipeline anomaly judgment model based on multiple parameters, the accuracy of monitoring leakage is improved. Through the combination of the above model and evaluation method, the functions of leakage monitoring, water hammer monitoring, temperature monitoring and burst pipe monitoring of underground heating pipeline can be effectively realized. The upper computer software has short delay, can quickly judge various abnormal events in the pipeline, and reduce the economic loss caused by delay.

[0047] The application realizes wireless communication with the upper computer through the data transceiver module, sends data to the upper computer, and the upper computer processes the received information through the pre-trained monitoring model based on the twin neural network and the three-dimensional composite model based on underwater sound, pressure, temperature and water contact information, and then obtains the monitoring result.

[0048] The working method is as follows, the underwater sound sensing module, high-frequency pressure module, temperature sensing module and water contact sensing module collect corresponding underwater sound, pressure, temperature and water contact data, and then send them to the upper computer software end through the data transceiver module, and the upper computer software built-in monitoring model based on the twin neural network, the pipeline anomaly judgment model based on multiple parameters and the evaluation method combining mel frequency cepstrum coefficient (MFCC), pearson correlation coefficient (PCC) and structural similarity (SSIM) are used to judge the pipeline anomaly and push the corresponding alarm.

[0049] Specifically, as shown in Figure 3 The monitoring model based on the twin neural network realizes accurate identification of abnormal acoustic signals by constructing the intrinsic feature space of the normal state. Its technical implementation system includes four core modules of data preprocessing, feature learning, dynamic detection and model evolution, and the specific technical scheme is as follows:

[0050] In the data preprocessing stage, a signal enhancement method with physical constraints is used to construct a training sample set. After eliminating the pump vibration noise by a 20-500Hz band-pass filter, the underwater sound signal is subjected to sliding window standardization processing (window 200ms, step 50ms), which effectively suppresses the time-varying drift of the sensor. In order to solve the problem of lack of leakage samples, a simulation enhancement strategy based on the acoustic propagation model is proposed: in the normal signal, a leakage pulse waveform with an amplitude of 0.1-0.3 times the fundamental signal and a frequency deviation of ±50Hz is injected, and noise samples containing valve action, water flow impact and other actual working conditions are superimposed to construct a mixed data set with physical rationality.

[0051] The feature extraction module adopts a parameter-shared twin network architecture, and a compact feature space of a normal state is established through a contrast learning mechanism. Three layers of depth separable convolution (32 / 64 / 128 3x1 kernels) are arranged in the network structure, the longitudinal propagation characteristics of sound waves are captured while the computational complexity is reduced, and an overlapping maximum pooling layer (window 3, step 2) retains local extreme value features and suppresses random noise. During the training process, the Euclidean distance of the normal sample pair in the 128-dimensional embedding space is constrained to converge to 0.2±0.05 through a contrast loss function, and the distance between the normal and abnormal sample pairs is expanded to more than 1.5±0.3, forming significant separability.

[0052] The dynamic detection module constructs a pressure-adaptive double-threshold discrimination mechanism. A sliding reference feature library is established during the deployment stage, and the moving average features of the normal samples in the last 24 hours are updated every 5 minutes to eliminate the influence of working condition drift. During real-time detection, the K-nearest neighbor dynamic time warping distance (K=5, window 50) of the input signal and the reference library is calculated, and the sound intensity-pressure coupling calibration is performed in combination with the pressure sensor reading. The threshold setting adopts a composite strategy of a basic value (mu+3sigma) and a dynamic compensation term (0.1xdP / dt), and an alarm is triggered when three consecutive sampling points (interval 0.2 seconds) exceed the threshold, ensuring that the detection accuracy is maintained at more than 95% under pressure transient working conditions.

[0053] The model evolution module realizes continuous learning by using an elastic weight solidification algorithm. Important feature weights are frozen during monthly incremental training to protect existing knowledge from being covered; when a new unknown interference mode is detected, an active learning process is started to automatically label suspicious samples, gradually adapting to long-term changes such as pipe network aging. Through actual measurement verification, the F1 value of this method reaches 0.92 with only 5% labeled abnormal data, which is 40% higher than the traditional threshold method, and the false positive rate is stably controlled below 0.5%, meeting the all-weather monitoring needs of urban pipe networks.

[0054] The scheme innovatively combines the twin network architecture with the pipe acoustic characteristics, reduces the model complexity through the parameter sharing mechanism, and enhances the feature separability through contrast learning. The double-branch structure can effectively handle the sample imbalance problem and adapt flexibly to different pipe diameter detection needs by adjusting the network depth.

[0055] Figure 4 and Figure 5 Pearson correlation coefficient and structural similarity involved in the present application, Figure 6For the evaluation flowchart combined with mel frequency cepstral coefficient, Pearson correlation coefficient and structural similarity, after obtaining the audio data through the integrated terminal, the time-frequency spectrum is abstracted through the mel frequency cepstral coefficient, at this time the time-frequency spectrum will be presented as a two-dimensional matrix. The Pearson correlation coefficient is used to quantify the linear relationship strength between the two matrices, and the value range is from-1 to 1, wherein 1 represents complete correlation. Finally, the structural similarity is used as an index for measuring the similarity of two images, and the value range is from 0 to 1, wherein 1 indicates that the two images are completely the same. The method of converting the sensing signal obtained by the integrated monitoring unit into the mel frequency cepstral coefficient (MFCC) can effectively distinguish different event segments.

[0056] Figure 7 For the pipeline abnormality judgment model structure based on multiple parameters involved in the application, after the upper computer receives all the data, first, the data of each sensor module in the pipeline is extracted, the temperature data includes extracting temperature drop, temperature gradient and other characteristics. The pressure data extracts pressure drop, pressure fluctuation, pressure transient and other characteristics. The underwater acoustic data extracts high-frequency acoustic signal, low-frequency acoustic signal, acoustic signal intensity and other characteristics. Because the pressure data is collected at high frequency, the pressure data is used as the judgment starting point.

[0057] When water hammer occurs in the pipeline, the water hammer will cause the pressure in the pipeline to fluctuate violently, and the high-frequency pressure sensor will detect the pressure transient signal. The pressure peak near the water hammer point will be significantly higher than the preset threshold, and at the same time, the underwater acoustic data and the temperature data will not change significantly. When the pressure, underwater acoustic and temperature data meet the water hammer characteristics, it is determined as a water hammer event.

[0058] When a leak occurs in the pipeline, it is usually accompanied by weak pressure changes and relatively obvious changes in underwater acoustic signals. The leak will cause the pressure in the pipeline to slowly decrease, and the high-frequency pressure sensor will detect the pressure decrease trend, but this process is slow and not easy to detect, and will be in equilibrium state after a period of time, and there will be no obvious pressure wave mutation, so the pressure wave change is used as one of the leakage judgment bases; the leak will produce low-frequency acoustic signals, and the underwater acoustic module will capture the acoustic characteristics of the leak, and according to the monitoring model based on the twin neural network and the evaluation method combined with mel frequency cepstral coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) of 4.2.2, the leak point can be located. In addition, the leak may cause local temperature changes, but usually not obvious, so the temperature change is not used as a basis for judging the leak. When the pressure and underwater acoustic data meet the characteristics of the leak, it is determined as a leak event

[0059] When a pipe burst event occurs in the pipeline, the pressure is judged: a pipe burst will cause the pressure in the pipeline to drop rapidly, and the high-frequency pressure sensor will detect a sudden drop in pressure, and the pressure fluctuation near the pipe burst point will increase significantly; a pipe burst will produce high-frequency acoustic signals, and the hydrophone will capture obvious pipe burst acoustic characteristics, and the acoustic signal strength near the pipe burst point will be significantly higher than the normal value, and the monitoring model based on the twin neural network combined with the evaluation method combining Mel frequency cepstrum coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) can locate the leakage point of the feature; a pipe burst will cause the medium in the pipeline to flow out rapidly, and the temperature sensor will detect a sudden drop in temperature, and the temperature gradient near the pipe burst point will be significantly higher than the normal value, so the temperature is one of the bases for re-determination. When the data of temperature, pressure and underwater sound meet the pipe burst characteristics, it is determined that a pipe burst event has occurred.

[0060] When a pipe burst event occurs in the pipeline, the pressure is judged: a pipe burst will cause the pressure in the pipeline to drop rapidly, and the high-frequency pressure sensor will detect a sudden drop in pressure, and the pressure fluctuation near the pipe burst point will increase significantly; a pipe burst will produce high-frequency acoustic signals, and the hydrophone will capture obvious pipe burst acoustic characteristics, and the acoustic signal strength near the pipe burst point will be significantly higher than the normal value, and the monitoring model based on the twin neural network combined with the evaluation method combining Mel frequency cepstrum coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) can locate the leakage point of the feature; a pipe burst will cause the medium in the pipeline to flow out rapidly, and the temperature sensor will detect a sudden drop in temperature, and the temperature gradient near the pipe burst point will be significantly higher than the normal value, so the temperature is one of the bases for re-determination. When the data of temperature, pressure and underwater sound meet the pipe burst characteristics, it is determined that a pipe burst event has occurred.

[0061] When a pipe burst, leakage and water hammer event is determined, the water touch module is used to detect whether each sensing module is in a normal operating environment at this time. If the water touch sensing module detects that each module has deviated from the normal detection water level environment at this time, all types of alarms are shielded, and a non-water touch alarm is pushed forward.

[0062] The present application can distinguish between the daily operation environment and the abnormal environment of the pipeline through the high-sensitivity monitoring module and the monitoring model based on the twin neural network, reduce the dependence on the pipeline material and the environment, and make the sensing and monitoring unit more adaptable, so that the leakage can be better judged.

[0063] The evaluation method combining Mel frequency cepstrum coefficient (MFCC), Pearson correlation coefficient (PCC) and structural similarity (SSIM) based on the monitoring model of the twin neural network is established. The difference between different pipe diameters, pressures and environments can be differentiated through four dimensions and similarity evaluation, and dynamic experience index of a specific position of a specific pipeline is established as an evaluation basis.

[0064] The integrated sensing unit is provided with a water touch sensor, which can be used as a fourth dimension judgment parameter to judge the false alarm of the system due to environmental changes. The pipe abnormality judgment model based on multiple parameters can shield the alarm and push the non-water touch alarm when the integrated monitoring unit is not in contact with water.

[0065] The integrated sensing unit is combined with a multi-parameter pipeline abnormality judgment model to distinguish water hammer, leakage, pipe burst and hypothermia events in the pipeline by combining the three dimensions of hydroacoustic data, pressure data and temperature data, thereby improving the accuracy of judgment, breaking the information island and making judgments on pipeline abnormal events based on three reference quantities.

[0066] The multi-dimensional thermal pipeline monitoring system based on the integrated monitoring unit uses high sensitivity and low response delay as the judgment starting from the sensor end. The response time at the sensor end is in milliseconds. At the same time, the system and its built-in models can respond within seconds, solving the problem of high latency.

[0067] The above description is only for the preferred embodiment of the present invention, which should not be understood as limiting the claims. Any equivalent process changes made using the present invention description are included in the patent protection scope of the present invention.

Claims

1. A multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit, characterized by: include The underwater acoustic sensing module uses piezoelectric ceramics as the transducer material, and generates electrical signals by deformation under the action of sound pressure, which are then converted into output signals through an amplification circuit and transmitted to the data transceiver module; High-frequency pressure module: The high-frequency pressure sensing module uses MEMS technology to directly manufacture strain resistors on a stainless steel isolation diaphragm as a resistance element. The isolation diaphragm generates an electrical signal when it is stressed and is amplified and transmitted to the data transceiver module. The temperature sensing module outputs an electrical signal based on the characteristic that metal resistance changes with temperature. The electrical signal is amplified and transmitted to the data transceiver module; The water contact detection module triggers the detection circuit based on the liquid conductivity, and the detection circuit results are transmitted to the data transceiver module in the form of on / off; A data transceiver module is used to collect electrical signals from the hydrophone module, high-frequency pressure module, temperature sensor module, and water contact sensor module; Host computer: unifies all monitoring stations on the pipeline through the management platform, manages the integrated monitoring unit equipment at the monitoring end, and realizes remote parameter configuration and remote monitoring.

2. The multi-dimensional thermal pipeline monitoring system based on the integrated monitoring unit according to claim 1 is characterized in that: The upper computer has a built-in leakage detection model based on a twin neural network. It combines the evaluation indicators of Mel-frequency cepstral coefficient MFCC, Pearson correlation coefficient and structural similarity, as well as a pipeline anomaly judgment model based on multiple parameters to improve the accuracy of leakage monitoring. By combining the above models and evaluation methods, leakage monitoring, water hammer monitoring, temperature monitoring and pipe burst monitoring of underground heating pipelines can be realized.

3. A monitoring method for a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 1 or 2, characterized in that: The hydroacoustic sensing module, high-frequency pressure module, temperature sensing module and water contact sensing module collect the corresponding hydroacoustic, pressure, temperature and water contact data and send them to the host computer through the data transceiver module. The host computer uses the built-in monitoring model based on the twin neural network, the pipeline anomaly judgment model based on multiple parameters and the evaluation method combining the Mel frequency cepstral coefficient, Pearson correlation coefficient and structural similarity to judge the pipeline anomaly and push the corresponding alarm.

4. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 3 is characterized in that: The monitoring model based on the twin neural network achieves accurate identification of abnormal acoustic signals by constructing the intrinsic feature space of the normal state. It includes four stages: data preprocessing, feature learning, dynamic detection, and model evolution. Among them, in the data preprocessing stage, a physical constraint signal enhancement method is used to construct a training sample set. The specific operation is as follows: after eliminating the pump vibration noise through a 20-500Hz bandpass filter, the underwater acoustic signal is subjected to sliding window normalization processing to suppress the time-varying drift of the sensor.

5. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 4 is characterized in that: It also includes a simulation enhancement strategy based on the acoustic propagation model: a leakage pulse waveform with an amplitude of 0.1-0.3 times the fundamental signal and a frequency deviation of ±50Hz is injected into the normal signal, and noise samples of actual working conditions including valve action and water flow impact are superimposed to construct a physically reasonable hybrid data set.

6. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 3, characterized in that: In the feature learning stage, a parameter-sharing twin network architecture is adopted to establish a compact feature space of the normal state through a contrastive learning mechanism. Three layers of depthwise separable convolution are deployed in the network structure to capture the longitudinal propagation characteristics of sound waves while reducing computational complexity; overlapping maximum pooling layers retain local extreme value features and suppress random noise; and during the retraining process, through the contrastive loss function constraint, the Euclidean distance of normal sample pairs in the 128-dimensional embedding space converges to 0.2±0.05, and the distance between normal and abnormal sample pairs expands to above 1.5±0.3, to form significant separability.

7. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 3, characterized in that: During the dynamic detection phase, a pressure-adaptive dual-threshold discrimination mechanism was constructed, and a sliding benchmark feature library was established. The moving average features of normal samples in the last 24 hours were updated every 5 minutes to eliminate the impact of operating condition drift. During real-time detection, the K-nearest neighbor dynamic time-warped distance between the input signal and the benchmark library was calculated, and the sound intensity-pressure coupling calibration was performed in combination with the pressure sensor readings. The threshold setting adopts a composite strategy of base value and dynamic compensation term, and an alarm is triggered when three consecutive sampling points exceed the threshold.

8. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 3, characterized in that: The evaluation method combining Mel-frequency cepstral coefficients, Pearson correlation coefficients, and structural similarity operates as follows: after obtaining audio data through an integrated terminal, the time-frequency spectrum is abstracted using the Mel-frequency cepstral coefficients, and the time-frequency spectrum is then presented as a two-dimensional matrix; the strength of the linear relationship between the two matrices is quantified using the Pearson correlation coefficient, with a value range from -1 to 1, where 1 represents complete correlation; finally, structural similarity is used as an indicator to measure the similarity between two images, with a value range from 0 to 1, where 1 indicates that the two images are exactly the same.

9. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 3, characterized in that: The multi-parameter pipeline anomaly judgment model combines hydroacoustic data, pressure data and temperature data to distinguish water hammer, leakage, pipe burst and hypothermia events in the pipeline. After the host computer receives all the data, it first extracts the temperature data, pressure data and hydroacoustic data in the pipeline. Among them, the temperature data includes the extraction of temperature drop and temperature gradient; the pressure data includes pressure drop, pressure fluctuation and pressure transient related features; the hydroacoustic data includes high-frequency acoustic signals, low-frequency acoustic signals and acoustic signal intensity related features.

10. The monitoring method of a multi-dimensional thermal pipeline monitoring system based on an integrated monitoring unit according to claim 9, characterized in that: When water hammer occurs in the pipeline, the high-frequency pressure module will detect the pressure transient signal. When the pressure, water sound and temperature data meet the water hammer characteristics, it is determined to be a water hammer event; When a leak occurs in the pipeline, the high-frequency pressure module detects the downward pressure trend, while the hydroacoustic sensing module captures the acoustic characteristics of the leak. The leak point is located using a built-in twin neural network-based monitoring model and an evaluation method that combines Mel-frequency cepstral coefficients, Pearson correlation coefficients, and structural similarity. When both pressure and hydroacoustic data match the leak characteristics, a leak is identified. When a pipe bursts, the high-frequency pressure module detects abnormal pressure, the hydroacoustic sensor module captures the acoustic signature of a burst, and the temperature sensor detects abnormal temperature. A burst is identified when all three data points, including temperature, pressure, and hydroacoustic signature, meet the burst signature. When a temperature drop occurs in the pipeline, the temperature drop is judged by the temperature gradient change; After confirming the pipe burst, leakage and water hammer events, the water contact module is used to detect whether each sensor module is in a normal operating environment at this time. If the water contact sensor module detects that each module has left the normal detection water level environment at this time, all types of alarms will be shielded and the non-water contact alarm will be pushed forward.

Images