An intelligent detection device for natural gas odorant concentration suitable for remote monitoring

Abstract

The present invention discloses a natural gas odorant concentration intelligent detection device suitable for remote monitoring, which belongs to the field of natural gas safety monitoring, and comprises: an odorant concentration sensor array, which detects the original concentration data of the odorant in the natural gas; a cross-validation module, which performs fusion processing on the original concentration data corresponding to a plurality of sensors, and obtains calibrated odorant concentration detection data; a pipeline monitoring module, which obtains pipeline operation data; a communication module, which connects the cross-validation module, the pipeline monitoring module and the remote monitoring module; the remote monitoring module, which determines the odorant concentration prediction value based on the integrated multi-model according to the original concentration data and pipeline operation data corresponding to the plurality of sensors; and obtains the final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value. The present invention realizes high-precision, real-time and reliable remote intelligent monitoring of natural gas odorant concentration.

Description

Technical Field

[0001] The present invention relates to the field of natural gas safety monitoring, and in particular to an intelligent detection device for natural gas odorant concentration suitable for remote monitoring. Background Art

[0002] As an important clean energy, natural gas is widely used in urban residents' lives and industrial production. In order to detect natural gas leaks in time and ensure user safety, odorants play a vital role in the natural gas transportation process. Traditional natural gas odorant concentration detection methods usually rely on manual on-site sampling and laboratory analysis, which has many limitations such as long detection cycle, poor real-time performance, and high cost.

[0003] Most of the existing natural gas odorant detection equipment uses a single sensor to monitor concentration. This method is easily affected by environmental interference and sensor errors, resulting in inaccurate and unreliable detection results. At the same time, these devices usually lack comprehensive consideration of pipeline operation status, and cannot fully evaluate the changing trend and potential risks of odorant concentration, making it difficult to meet the urgent needs of modern natural gas transportation systems for accurate, real-time, and intelligent monitoring.

[0004] Traditional natural gas odorant detection methods have been unable to adapt to the increasingly complex energy monitoring environment. There is an urgent need to develop an intelligent detection device that can monitor the concentration of natural gas odorants in a multi-dimensional, high-precision, and remote real-time manner. This can not only improve the detection accuracy, but also realize dynamic and accurate monitoring of the entire natural gas transportation process, providing strong technical support for ensuring energy security and the safety of users' lives and property. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a natural gas odorant concentration intelligent detection device suitable for remote monitoring to solve the above technical problems.

[0006] In order to achieve the above-mentioned purpose, a natural gas odorant concentration intelligent detection device suitable for remote monitoring is provided, which comprises:

[0007] An odorant concentration sensor array is used to detect raw concentration data of odorants in natural gas through a variety of sensors;

[0008] A cross-validation module is used to fuse the original concentration data corresponding to the multiple sensors to obtain calibrated odorant concentration detection data;

[0009] A pipeline monitoring module is used to obtain pipeline operation data corresponding to the pipeline that transports natural gas;

[0010] A communication module, used to connect the cross-validation module and the remote monitoring module, and to connect the pipeline monitoring module and the remote monitoring module;

[0011] The remote monitoring module is used to determine the odorant concentration prediction value based on the integrated multi-model according to the original concentration data corresponding to the multiple sensors and the pipeline operation data; and obtain the final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0012] The above technical solution has the following beneficial technical effects:

[0013] The intelligent detection device for natural gas odorant concentration of the present invention realizes high-precision calibration of odorant concentration detection data through a multi-sensor array and a cross-validation module, thereby improving the accuracy and reliability of the detection results. The introduction of the pipeline monitoring module, combined with the integrated multi-model analysis, enables the detection device to comprehensively consider the original sensor data and the pipeline operation status, and can not only obtain the odorant concentration in real time and accurately, but also predict the concentration change trend, providing a more comprehensive and intelligent monitoring solution for the safety of natural gas transportation. The detection device has the advantages of remote real-time monitoring, multi-dimensional data fusion, high anti-interference, etc., which can improve the accuracy and reliability of natural gas odorant concentration detection and reduce detection costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0015] Figure 1 It is a functional block diagram of a natural gas odorant concentration intelligent detection device suitable for remote monitoring according to an embodiment of the present invention;

[0016] Figure 2 is a functional block diagram of an odorant concentration sensor array according to an embodiment of the present invention;

[0017] Figure 3 is a functional block diagram of a cross-validation module according to an embodiment of the present invention;

[0018] Figure 4 is a functional block diagram of a pipeline monitoring module according to an embodiment of the present invention;

[0019] Figure 5 is a functional block diagram of a remote monitoring module according to an embodiment of the present invention;

[0020] Figure 6 is a schematic diagram of the overall structure of an odorant concentration sensor array according to an embodiment of the present invention;

[0021] Figure 7 is a functional block diagram of a computer-readable storage medium according to an embodiment of the present invention;

[0022] Figure 8 It is a schematic diagram of the structure of a computer system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0024] like Figure 1 As shown, an embodiment of the present invention provides a natural gas odorant concentration intelligent detection device suitable for remote monitoring, which includes: an odorant concentration sensor array, which is used to detect the original concentration data of the odorant in the natural gas through a variety of sensors; a cross-validation module, which is used to fuse the original concentration data corresponding to the various sensors to obtain calibrated odorant concentration detection data; a pipeline monitoring module, which is used to obtain pipeline operation data corresponding to the pipeline transporting natural gas; a communication module, which is used to connect the cross-validation module and the remote monitoring module, and to connect the pipeline monitoring module and the remote monitoring module; the remote monitoring module is used to determine the odorant concentration prediction value based on the integrated multi-model according to the original concentration data corresponding to the various sensors and the pipeline operation data; and obtain the final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0025] In some embodiments, the odorant concentration sensor array is composed of multiple different types of sensors, such as electrochemical sensors, spectral sensors, and gas sensitive resistor sensors. These sensors are deployed at key monitoring points of the natural gas pipeline and can capture the concentration information of the odorant (such as tetrahydrothiophene) from different dimensions and angles. Each sensor has its own detection principle and advantages. Through the coordinated work of multiple sensors, the accuracy and reliability of detection can be improved.

[0026] In some embodiments, the cross-validation module helps ensure the quality of the detection data. The module uses a data fusion algorithm to perform in-depth analysis and processing of the raw concentration data from different sensors. Specifically, the cross-validation module first evaluates the performance stability, environmental adaptability, and scene compatibility of each sensor, and dynamically calculates the weight of each sensor based on these indicators. For example, in a low temperature environment, some sensors will experience a decrease in sensitivity, and the cross-validation module will adjust its weight accordingly to ensure that the final concentration detection result will not be significantly affected by unstable sensor data.

[0027] In some embodiments, the pipeline monitoring module provides comprehensive operating status information of the natural gas transmission system. The pipeline monitoring module obtains key data such as pipeline vibration, pressure, and temperature. These data provide important contextual information for subsequent concentration prediction and risk assessment. For example, pipeline vibration can reflect the mechanical integrity of the pipeline, pressure fluctuations can indicate potential leakage risks, and temperature gradients are related to natural gas flow characteristics and odorant distribution.

[0028] In some embodiments, the communication module ensures real-time, reliable data transmission between modules of the equipment and with the remote monitoring center. It uses industrial-grade communication protocols, such as Modbus TCP / IP or OPC UA, to support high-reliability, low-latency data exchange. The communication module not only transmits sensor data and pipeline operation data, but also transmits equipment status, diagnostic information, and remote control instructions.

[0029] In some embodiments, the remote monitoring module is an intelligent decision-making center. It is based on an integrated multi-model algorithm, comprehensively considers the sensor raw data and pipeline operation data, and dynamically generates the odorant concentration prediction value. The multi-model integration can include a variety of deep learning models, such as capsule networks, deep belief networks, etc., and improves the prediction accuracy through integrated learning methods. The final odorant concentration confirmation result will take into account both the real-time detection data and the odorant concentration prediction value, and a multiple threshold alarm mechanism can be set. Once an abnormal situation is detected, the early warning and / or risk response process will be immediately triggered.

[0030] In some embodiments, the remote monitoring module is set up on the cloud big data platform, which can not only perform remote monitoring, but also conduct large-scale data analysis to provide technical support for the safety of natural gas transportation. The entire system meets the requirements of harsh industrial environments such as explosion-proof and corrosion-proof, ensuring stable and reliable operation in complex natural gas transportation scenarios.

[0031] The intelligent detection device for natural gas odorant concentration of the embodiment of the present invention has the following advantages: First, through the collaborative work of multiple sensor arrays, it breaks through the limitations of traditional single sensor detection and improves the accuracy and reliability of odorant concentration detection. Secondly, by introducing a cross-validation module, dynamic weight adjustment and intelligent fusion of sensor data are realized, effectively suppressing the errors and drifts existing in a single sensor. Furthermore, the introduction of a pipeline monitoring module is not only limited to concentration detection, but also can comprehensively obtain the pipeline operation status, providing rich contextual information for odorant concentration detection. In addition, based on the integrated multi-model algorithm, it can realize accurate prediction of odorant concentration through deep learning technology, and generate the final confirmation result in combination with real-time detection data, thereby improving the intelligence level of the detection system. Finally, the detection equipment has remote monitoring and abnormal warning functions, can transmit data in real time, and respond to risks quickly, providing all-round and intelligent protection for the safety of natural gas transportation.

[0032] like Figure 2 As shown, the odorant concentration sensor array specifically includes any of the following:

[0033] An electrochemical sensor for oxidizing or reducing odorant molecules through an electrochemical reaction between the working electrode and a reference electrode when the odorant molecules come into contact with the surface of the working electrode of the electrochemical sensor, thereby generating a current signal related to the concentration; determining the concentration of the odorant according to the intensity of the current signal, and identifying the type of the odorant according to the waveform characteristics of the current signal; the waveform characteristics include the oxidation peak position or the reduction peak position;

[0034] A spectral sensor for obtaining the wavelength position and absorption intensity of a spectral absorption peak based on the absorption characteristics or scattering characteristics of the odorant molecules to the target wavelength light when the odorant molecules pass through the optical detection area of ​​the spectral sensor; matching a known odorant molecular characteristic spectrum according to the wavelength position of the spectral absorption peak to determine the type of the odorant, and calculating the concentration of the odorant according to the absorption intensity;

[0035] A gas sensitive resistor sensor is used to cause the resistance of the sensitive material to change when odorant molecules are adsorbed onto the surface of the sensitive material of the gas sensitive resistor sensor; obtain the resistance change curve by real-time detection of the characteristic value of the resistance change of the gas sensitive resistor sensor over time; and detect the concentration of the odorant based on the amplitude and response time of the resistance change curve.

[0036] Specifically, the electrochemical sensor is one of the key sensors for detecting natural gas odorants. Its working principle is based on the current response characteristics of electrochemical reactions. When odorant molecules (such as tetrahydrothiophene) come into contact with the working electrode surface of the electrochemical sensor, an oxidation or reduction reaction occurs. For example, when scanning the potential in the forward direction, the odorant molecules are oxidized and generate a specific current signal. By analyzing the intensity of this current signal, the concentration of the odorant can be accurately quantified. Different types of odorants will present unique waveform characteristics in the electrochemical reaction, such as the position of the oxidation peak or the reduction peak, which provides an important basis for the identification of the type of odorant.

[0037] Specifically, the spectral sensor uses the principle of optical detection to detect the unique absorption or scattering characteristics of the odorant molecules to the target wavelength of light. When the odorant molecules pass through the optical detection area, they absorb or scatter the light of a specific wavelength, thus forming a characteristic spectral absorption peak. By accurately measuring the wavelength position and absorption intensity of these spectral absorption peaks, two key functions can be achieved: first, by matching the known odorant molecular characteristic spectrum, the specific type of odorant can be accurately identified; second, based on the change in absorption intensity, the concentration of the odorant can be accurately calculated. This detection method is highly selective and sensitive, and can accurately detect odorants in complex natural gas environments.

[0038] Specifically, the gas resistor sensor uses the detection principle of changes in the electrical properties of materials. When the odorant molecules are adsorbed onto the surface of the sensitive material of the sensor (composed of semiconductor materials), the electrical properties of the sensitive material will change, which is mainly manifested as a change in resistance. The gas resistor sensor obtains a characteristic resistance change curve by real-time monitoring of the change in resistance over time. The main detection parameters include the amplitude and response time of the resistance change curve. The amplitude of the resistance change directly reflects the concentration of the odorant molecules, while the response time reflects the sensitivity of the gas resistor sensor to the odorant. This detection method has the advantages of fast response and low cost, and is particularly suitable for rapid and continuous odorant concentration monitoring in industrial sites.

[0039] Through the coordinated work of these three sensors, the odorant in natural gas can be comprehensively and accurately detected from three different dimensions: electrochemical, optical and electrical, which improves the reliability and accuracy of the detection system. Each sensor has its own working principle and advantages. They complement and verify each other, and together form a highly intelligent and reliable odorant concentration monitoring solution.

[0040] Key nodes of pipelines are the preferred locations for sensor deployment, such as distribution stations, valve chambers, and metering stations. These locations have the characteristics of sufficient natural gas medium flow and relatively stable pressure and temperature, which can obtain representative test samples, facilitate sensor maintenance and replacement, and can also achieve segmented monitoring and quickly locate abnormal areas.

[0041] The electrochemical sensor is installed on the outside of the straight section of the pipeline, especially at the upstream position where the inner diameter of the pipeline changes, and exchanges gas with the natural gas flow inside the pipeline through a precise sampling system. The sampling system consists of key components such as sampling holes, sampling interfaces, sampling pipelines, flow control valves and pressure regulators: the sampling hole is opened at the upstream position of the outer wall of the pipeline as the initial channel for the introduction of natural gas; the sampling interface is set at the external outlet of the sampling hole, connecting the sampling hole and the sampling pipeline to ensure the airtightness and accuracy of the gas transmission process; the sampling pipeline is used to transmit the gas sample from the sampling interface to the electrochemical sensor detection cavity, and the material is selected from inert materials such as stainless steel or special polymers to prevent gas adsorption and secondary reactions; the flow control valve is installed in the midstream position of the sampling pipeline to accurately adjust the gas flow rate to ensure that the sampling process complies with the isokinetic sampling principle and is representative; the pressure regulator is set at the end of the sampling pipeline, close to the electrochemical sensor detection cavity, to stabilize the gas pressure and eliminate the influence of pressure fluctuations on the detection results. The entire sampling process is based on the principle of isokinetic sampling, which accurately controls the sampling flow and pressure to ensure that the obtained gas samples truly reflect the actual state of the natural gas in the pipeline, maintaining the integrity of the pipeline while ensuring monitoring accuracy.

[0042] The spectral sensor is installed on the outside of the pipeline, and the transparent or translucent section of the pipeline is selected as the installation location. To improve the detection accuracy, the sensor can be equipped with an optical detection cabin, which provides a stable and clean optical environment for spectral detection and effectively isolates external interference factors. The optical detection cabin is installed close to the main body of the spectral sensor and maintains a close optical coupling relationship with the spectral sensor to ensure the continuity and integrity of the optical path transmission. To ensure the stability of optical detection, the sensor system can also be equipped with cleaning mechanisms and anti-pollution mechanisms. These mechanisms can be automatic cleaning devices installed around the optical detection cabin or on the outer casing of the spectral sensor, which can promptly remove dust, water vapor and other impurities on the optical path and maintain the clarity of the optical path, thereby ensuring the accuracy and reliability of spectral detection and effectively improving the long-term monitoring performance of the sensor.

[0043] The gas sensitive resistor sensor is installed on the outside of the pipeline and connected to the pipeline through the sampling interface. The interface adopts a highly sealed sampling valve to ensure that there will be no natural gas leakage during the gas sampling process. The sampling chamber is connected to the sampling interface through a connecting pipeline. A small amount of gas sample is extracted from the inside of the pipeline by a precise exhaust mechanism and transported to the detection area of ​​the gas sensitive resistor sensor. As the front-end signal acquisition unit of the gas sensitive resistor sensor, the sampling chamber plays a key role in the gas transmission and signal conversion process. In order to ensure the stability and accuracy of the sensor, it can also be equipped with a temperature and humidity adjustment module to accurately control the environmental parameters so that the gas sensitive resistor sensor always works under the best detection conditions. In terms of deployment strategy, the principle of multi-point deployment is followed to form a three-dimensional sensor network. The sensor spacing is flexibly determined according to factors such as pipeline length and gas flow characteristics. Specifically, according to the concentration gradient and spatial distribution of natural gas in the pipeline, and according to the potential risk level of different pipeline sections, the sensors are reasonably arranged to achieve comprehensive and accurate monitoring of the pipeline medium. At the same time, redundant backup sensors are reserved and modular installation interfaces are designed to improve the reliability and maintenance convenience of the system, ensure that certain sensors can be quickly replaced when they fail, and maintain the continuity and integrity of the monitoring network.

[0044] To improve the reliability of the detection system, auxiliary monitoring facilities can also be deployed, including temperature sensors, pressure monitoring devices, flow meters, etc., which can provide more comprehensive working condition information. In addition, through the signal conditioning circuit, interference can be effectively suppressed and signal quality can be improved.

[0045] In a further embodiment, Figure 2 As shown, the odorant concentration sensor array may also include any of the following:

[0046] An optical fiber sensor, for causing attenuation of light signal intensity by the absorption or scattering characteristics of target wavelength light when odorant molecules contact the sensitive layer on the optical fiber surface of the optical fiber sensor; detecting the time-varying characteristics of light signal intensity in real time by the optical detection device of the optical fiber sensor, obtaining a light signal intensity variation curve, extracting the attenuation amplitude of light signal intensity from the light signal intensity variation curve; and calculating the concentration of the mercaptan odorant according to the attenuation amplitude of the light signal intensity;

[0047] A micro-resonance cavity sensor, used for, when odorant molecules come into contact with a sensitive layer on the surface of the micro-resonance cavity sensor, causing the resonant frequency of the micro-resonance cavity to shift due to the adsorption of the odorant molecules; obtaining the shift of the resonant frequency by monitoring the change characteristics of the resonant response of the micro-resonance cavity; and detecting the concentration of the odorant according to the amplitude of the shift of the resonant frequency;

[0048] A graphene sensor, for when odorant molecules come into contact with the graphene sensitive layer of the graphene sensor, the odorant molecules are adsorbed to cause the conductivity of the graphene sensitive layer to change; a conductivity change curve is obtained by real-time detection of characteristic values ​​of the conductivity of the graphene sensor changing with time; and the concentration of the odorant is detected according to the amplitude and dynamic response characteristics of the conductivity change determined from the conductivity change curve;

[0049] A surface enhanced Raman spectroscopy sensor is used for enhancing the Raman scattering signal of the odorant molecule due to the surface enhanced Raman scattering effect when the odorant molecule contacts the metal nanostructure surface of the surface enhanced Raman spectroscopy sensor; obtaining the Raman spectral peak by detecting the intensity change characteristics of the target wavelength in the Raman scattering signal; matching the known molecular characteristic Raman spectrum according to the wavelength position of the Raman spectral peak to determine the type of the odorant, and determining the concentration of the odorant according to the intensity of the Raman spectral peak.

[0050] Specifically, the optical fiber sensor uses a sensitive layer that has a highly selective adsorption performance for mercaptan odorants. When the natural gas sample passes through the optical fiber sensor, the odorant molecules come into contact with the sensitive layer on the surface of the optical fiber, causing the absorption or scattering characteristics of the target wavelength light to change. The optical detection device records the change in light signal intensity in real time and draws a light signal intensity change curve. By analyzing the attenuation amplitude, the concentration of mercaptan odorants can be accurately calculated. For example, for ethyl mercaptan odorant, under a light source with a wavelength of 850nm, its concentration shows a good linear relationship with the light signal attenuation amplitude, and the detection range can cover 0.1-10 ppm.

[0051] Specifically, the surface of the micro-resonance cavity sensor is covered with a polymer sensitive layer. When the odorant molecules are adsorbed on the sensitive layer, the mass and elasticity of the micro-resonance cavity will change slightly, resulting in a shift in the resonant frequency. The change in resonant frequency can be monitored in real time through a high-precision frequency detection circuit. When the odorant concentration increases from 0.5 ppm to 5 ppm, the resonant frequency shift shows an obvious linear change, and the relative error is less than ±3%, achieving high-sensitivity detection of the odorant concentration.

[0052] Specifically, in this embodiment, a single layer of graphene can be used as a sensitive layer. When the odorant molecules are adsorbed on the graphene surface, the electron cloud distribution of the graphene will be changed, resulting in a change in conductivity. The conductivity change is monitored in real time by a four-probe measurement method. For the butanethiol odorant, the conductivity change shows a good logarithmic linear relationship with the concentration, the detection limit can reach 0.05ppm, and the response time is less than 3 seconds, which has high sensitivity and fast response characteristics.

[0053] Specifically, for surface-enhanced Raman spectroscopy sensors, gold nanoparticle arrays are used as surface enhancement structures to increase the intensity of Raman scattering signals. When odorant molecules come into contact with the surface of the metal nanostructure, their Raman scattering signals are greatly enhanced due to the surface-enhanced Raman scattering (SERS) effect. By comparing the standard Raman spectrum, the chemical composition of the odorant can be accurately identified. Taking methyl mercaptan as an example, its characteristic Raman peak is located at 2570 cm -1 The peak intensity shows a good linear relationship with the concentration, and the detection range covers 0.2-20 ppm, with high selectivity and quantitative analysis capability.

[0054] like Figure 3 As shown, the cross-validation module specifically includes:

[0055] The stability determination submodule is used to analyze the performance stability of each sensor based on the degree of data dispersion within the monitoring period, and calculate the stability index corresponding to each sensor within the monitoring period;

[0056] The environmental adaptability determination submodule is used to analyze the adaptability of each sensor in the current working environment based on the degree of deviation of the working environment conditions, and calculate the environmental adaptability index corresponding to each sensor;

[0057] The scene compatibility determination submodule is used to analyze the matching degree of each sensor in the current detection scene based on the multi-dimensional characteristics of each sensor, and calculate the scene compatibility index corresponding to each sensor;

[0058] A dynamic weight calculation submodule, used to calculate the dynamic weight of each sensor through the stability index, the environmental adaptability index and the scene compatibility index;

[0059] A fusion submodule is used to perform weighted fusion on the raw concentration data corresponding to the multiple sensors based on the dynamic weight of each sensor to obtain a fused odorant concentration detection value;

[0060] The calibration submodule is used to perform error correction on the fused odorant concentration detection value to obtain calibrated odorant concentration detection data.

[0061] Specifically, the stability determination submodule is used to evaluate the short-term performance stability of each sensor in a continuous detection cycle and calculate the short-term stability index , the calculation formula is: ;in, It represents the standard deviation of the sensor output data in a continuous monitoring period. is the offset constant; The larger the value, the more stable the sensor's short-term performance;

[0062] Specifically, the environmental adaptability determination submodule is used to quantify the adaptability of each sensor under the current environmental conditions and calculate the environmental adaptability index , the calculation formula is:

[0063] ;

[0064] in, is the current actual environment parameter value, is the optimal working environment parameter value of sensor i, is the sensor i's response to the environmental parameter Tolerance range, that is, the ability to adapt to changes in environmental parameters; The closer the value is to 1, the more the sensor adapts to the current environment; exp is the natural exponent, ensuring The output is in the range [0,1], The square of the tolerance range multiplied by a constant 2 is used for normalization so that the formula conforms to the probability density of the normal distribution.

[0065] Specifically, the scene compatibility determination submodule is used to analyze the matching degree of each sensor in the current detection scene based on the multi-dimensional characteristics of each sensor, and calculate the scene compatibility index corresponding to each sensor. , the calculation formula is: ;in, is a nonlinear mapping function, is the sensor sensitivity score, is the sensor detection range matching degree, is the drift characteristic of the sensor; this function maps multidimensional features to standardized scenario fit indicators through nonlinear transformation. The sensor sensitivity score is obtained based on the sensor's responsiveness to the target signal through experimental measurement, manufacturer data, or comparative scoring. The sensor detection range matching is obtained based on the overlap between the sensor's detection range and the detection scenario's required range through coverage range calculation or ratio scoring. If the sensor's detection range completely covers the scenario's required range, the sensor detection range matching is 1. If it partially covers, it is calculated based on the coverage ratio. If it does not cover at all, the sensor detection range matching is 0. The drift characteristics of the sensor are determined by long-term experimental measurement of drift, drift parameters provided by the manufacturer, or observations in actual applications. The nonlinear mapping function can be in the form of, but not limited to, weighted nonlinear combinations to comprehensively consider the contribution of each feature to the fit indicator.

[0066] Specifically, the dynamic weight calculation submodule is used to calculate the dynamic weight of each sensor through the stability index, the environmental adaptability index and the scene compatibility index. , the calculation formula is:

[0067] , i is used to specify a single sensor for the current weight calculation, and j is used to traverse all sensors for normalization.

[0068] Specifically, the fusion submodule is used to perform weighted fusion on multi-sensor data based on dynamic weights to obtain a fused concentration detection value. The calculation formula is: ;in, is the dynamic weight of the ith sensor, is the concentration detection value of the i-th sensor, is the fusion concentration value.

[0069] Specifically, the calibration submodule is used to perform error correction on the fusion concentration value to obtain a calibrated concentration value, and the calculation formula is: ;in, is the time-dependent dynamic calibration coefficient, is the time-dependent dynamic offset correction value, is the current detection cycle, is the calibration concentration value.

[0070] The following are examples:

[0071] The stability index calculation process is as follows:

[0072] Assuming there are three types of sensors, optical fiber sensor, graphene sensor and micro-resonance cavity sensor, the concentration data in a continuous 30-minute monitoring period are as follows:

[0073] The fiber optic sensor data is: [2.1, 2.3, 1.9, 2.2, 2.0] ppm, σ 1 = 0.15, P 1 = 1 / (0.15 + 0.01) = 6.25;

[0074] The graphene sensor data is: [2.3, 2.4, 2.2, 2.5, 2.1] ppm, σ 2 = 0.17, P 2 = 1 / (0.17 + 0.01) = 5.56;

[0075] The data of the micro-resonance cavity sensor are: [2.0, 2.1, 1.8, 2.2, 1.9] ppm, σ 3 = 0.16, P3 = 1 / (0.16 + 0.01) = 5.88.

[0076] The calculation process of environmental adaptability index is as follows:

[0077] Optimum operating temperature opt They are: Fiber Optic Sensor X opt,1 = 25°C, graphene sensor X opt,2 = 30°C, Micro-resonant Cavity Sensor X opt,3 = 28°C;

[0078] Current ambient temperature Xenv = 22°C;

[0079] Temperature tolerance deviation σ X They are: optical fiber sensor σ X,1 = 5°C, graphene sensor σ X,2 = 6°C, micro-resonant cavity sensor σ X,3 = 4°C;

[0080] Computing environment adaptability index: T 1 = exp(-((22 - 25)² / (2 * 5²))) = 0.88, T 2 = exp(-((22 - 30)² / (2 * 6²))) = 0.75, T 3 = exp(-((22 - 28)² / (2 * 4²))) = 0.91.

[0081] The calculation process of the scenario fit index is as follows:

[0082] Evaluating multidimensional features for fiber optic sensors: S sens , 1 = 0.90, R range , 1 = 0.85, D drift , 1 = 0.88, F 1 = 0.4 * 0.90 + 0.3 * 0.85 + 0.3 * 0.88 = 0.89;

[0083] Evaluating multidimensional features for graphene sensors: S sens , 2 = 0.82, R range , 2 = 0.78, D drift , 2 =0.80, F 2= 0.4 * 0.82 + 0.3 * 0.78 + 0.3 * 0.80 = 0.81;

[0084] Evaluating Multidimensional Features for Micro-Resonant Cavity Sensors: S sens , 3 = 0.95, R range , 3 = 0.90, D drift , 3 =0.92, F 3 = 0.4 * 0.95 + 0.3 * 0.90 + 0.3 * 0.92 = 0.93.

[0085] The dynamic weight calculation process is as follows:

[0086] Calculate the dynamic weight of each sensor: W 1 = (P 1 * T 1 * F 1 ) / Σ(P j * T j * F j ) = 0.42, W 2 = (P 2 * T 2 * F 2 ) / Σ(P j * T j * F j ) = 0.28, W 3 = (P 3 * T 3 * F 3 ) / Σ(P j * T j *F j ) = 0.30.

[0087] The fusion and calibration process is as follows:

[0088] Each sensor detects concentration: C 1 (fiber) = 2.1 ppm, C 2 (graphene) = 2.3 ppm, C 3 (microresonant cavity) = 2.0 ppm;

[0089] Fusion concentration: C f = W 1 * C 1 + W 2 * C 2 + W3 * C 3 = 0.42 * 2.1 + 0.28 * 2.3 + 0.30 * 2.0 = 2.11 ppm.

[0090] The calibration module process is as follows: dynamic calibration coefficient k(t) = 0.97, dynamic offset correction value b(t) = 0.06.

[0091] Concentration after calibration: C c = C f * k(t) + b(t) = 2.11 * 0.97 + 0.06 = 2.09 ppm.

[0092] like Figure 4 As shown, the pipeline monitoring module specifically includes at least one of the following:

[0093] The pipeline vibration characteristic monitoring submodule is used to collect pipeline vibration data through a vibration sensor, and perform frequency domain analysis and time domain feature extraction on the pipeline vibration data to determine the frequency, amplitude and vibration mode of the pipeline vibration;

[0094] The pipeline pressure fluctuation monitoring submodule is used to collect the natural gas pressure signal in the pipeline in real time through the pressure sensor, and obtain the amplitude, frequency and pulse characteristics of the pressure fluctuation based on the dynamic change characteristics and frequency characteristic analysis method of the natural gas pressure signal;

[0095] The pipeline temperature gradient monitoring submodule is used to continuously collect temperature data along the pipeline through a temperature sensor, and determine the pipeline surface temperature change and / or longitudinal temperature gradient according to the spatial distribution characteristics and time variation characteristics of the temperature data;

[0096] The pipeline health monitoring submodule is used to determine the physical state of the pipeline, including the degree of corrosion, distribution of leak points and the state of internal sediments, through ultrasonic detectors and leak detectors.

[0097] Specifically, pipeline vibration data is collected in real time through vibration sensors installed on the outer wall of the pipeline. Accelerometers, piezoelectric vibration sensors or fiber optic vibration sensors are mainly used to detect the frequency, amplitude and mode of vibration, and capture the vibration changes caused by mechanical disturbances or pressure fluctuations during the operation of the pipeline. Sensors are set at key locations such as pipe bends, flange connections or reducers in the pipeline, and are arranged at intervals of several meters to cover the monitoring needs of the entire pipeline vibration characteristics. The sensor is installed close to the outer wall of the pipeline using magnetic adsorption or a fixed bracket to ensure that the collected vibration signals are accurate and reliable, providing a reference for analyzing the pipeline operation status and abnormal gas concentration distribution.

[0098] Specifically, the pipeline pressure fluctuation characteristics are monitored in real time through dynamic pressure sensors, which mainly capture the transient changes in the natural gas pressure inside the pipeline, including the amplitude, frequency and pulse characteristics of the pressure fluctuations. Common sensors include piezoelectric pressure sensors, strain gauge pressure sensors and dynamic pressure sensors, which can detect the dynamic trend of short-term pressure changes or long-term pressure fluctuations in the pipeline. The sensor is installed at the pressure interface on the inner wall of the pipeline, or connected to the internal pressure source through the detection port outside the pipeline. Key locations include compressor stations, near valves and pressure regulation points. By reasonably arranging the sensor distribution, the pressure fluctuation data provides an accurate reference for the dynamic fluid characteristics for concentration detection.

[0099] Specifically, the temperature gradient and heat distribution data are obtained through temperature sensors installed on the outer wall of the pipeline. Common sensors include thermocouples, infrared temperature sensors and fiber optic temperature sensors, which are used to monitor the temperature changes on the surface of the pipeline and the gradient distribution along the length. Temperature sensors are evenly distributed at key locations such as the inlet, outlet, vertical section and bend section of the pipeline, and are arranged at intervals of several meters to fully cover the thermal distribution characteristics of the pipeline. Thermocouples and infrared sensors are installed on the outer wall of the pipeline using fixed brackets or adhesives, and fiber optic temperature sensors are laid along the pipeline. This data is used to correct the uneven gas diffusion caused by temperature differences to ensure the accuracy of concentration detection.

[0100] Specifically, pipeline health characteristic data is comprehensively evaluated through a variety of detection methods, mainly including the degree of corrosion, distribution of leak points and the state of internal deposits. Specific technologies include ultrasonic detectors and leak detectors. Ultrasonic detectors are fixedly installed at key points on the outer wall of the pipeline, such as the connection section and valve area, to monitor the change in pipeline wall thickness and stress distribution in real time. Leak detectors identify trace leaks by analyzing the gas composition (such as methane or hydrogen sulfide) around the pipeline.

[0101] In some embodiments, the pipeline vibration characteristic monitoring submodule specifically includes:

[0102] A vibration sensor array is arranged at a pipe bend, flange connection or reducer section of the pipeline to obtain a vibration signal of the pipeline;

[0103] A vibration data acquisition unit, used for acquiring vibration signals of the vibration sensor array in real time;

[0104] The vibration feature extraction unit is used to perform frequency analysis, amplitude calculation and vibration mode recognition on the collected vibration signal to obtain the frequency feature, amplitude feature and vibration mode feature of the vibration signal.

[0105] In some embodiments, the pipeline pressure fluctuation monitoring submodule specifically includes:

[0106] The pressure sensor group is installed at the compressor station, near the valve or at the pressure regulating point of the pipeline to collect the pressure signal of the natural gas in the pipeline in real time;

[0107] A pressure dynamic monitoring unit is connected to the pressure sensor group for communication, receives the pressure signal collected by the pressure sensor group, captures the transient change of the natural gas pressure in the pipeline in real time, and pre-processes the pressure signal to obtain the pre-processed pressure data;

[0108] The pressure fluctuation characteristic analysis unit is used to calculate the amplitude, frequency and pulse characteristics of the pressure fluctuation as pressure fluctuation characteristic parameters based on the preprocessed pressure data.

[0109] In some embodiments, the pipeline temperature gradient monitoring submodule specifically includes:

[0110] A temperature sensor network is distributed at the inlet, outlet, vertical section and bend section of the pipeline;

[0111] A temperature data acquisition unit, used to acquire temperature change data of the pipeline surface from the temperature sensor network in real time;

[0112] The temperature gradient calculation unit is used to analyze and determine the temperature distribution and longitudinal temperature gradient along the pipeline by using numerical interpolation or differential analysis method according to the temperature change data of the pipeline surface.

[0113] In some embodiments, the pipeline health monitoring submodule specifically includes:

[0114] An ultrasonic detector, used to detect the wall thickness change and defect characteristic data of the pipeline through ultrasonic scanning, and transmit the wall thickness change data and defect characteristic data to the pipeline structure monitoring unit;

[0115] A gas leakage detector, used to detect gas leakage in the environment around the pipeline, obtain leakage characteristic data, and transmit the leakage characteristic data to the abnormality detection unit;

[0116] A pipeline structure monitoring unit, used to determine the stress concentration area and corrosion degree of the pipeline based on the wall thickness change data and defect characteristic data provided by the ultrasonic detector in combination with a multi-source data analysis model;

[0117] The abnormality detection unit is used to identify the trace leakage points and internal sediment status in the pipeline according to the leakage characteristic data provided by the gas leakage detector and the stress concentration area and corrosion degree provided by the pipeline structure monitoring unit, and generate a detection report of the abnormal point.

[0118] During the operation of natural gas pipelines, vibration sensor arrays are deployed at the bends, flange connections and reducers of the pipelines, which are the key locations where stress concentration and vibration abnormalities are most likely to occur in the pipeline structure. The selected vibration sensors have high sensitivity and wide-band response characteristics, and can capture tiny vibration signals, including multi-dimensional vibration parameters such as acceleration, velocity and displacement.

[0119] The vibration data acquisition unit uses high-speed analog-to-digital conversion technology to capture the analog signal of the vibration sensor in real time at a sampling frequency of thousands of times per second, and performs real-time filtering, amplification and digital conversion through a digital signal processor. The data acquisition unit also has a built-in data compression algorithm, which can reduce the data transmission bandwidth and storage space requirements while ensuring signal integrity.

[0120] The vibration feature extraction unit is used to perform frequency analysis, amplitude calculation and vibration mode recognition on the collected vibration signals, and extract the frequency features, amplitude features and vibration mode features of the vibration signals. Through signal processing algorithms such as Short-Time Fourier Transform (STFT) and wavelet transform, the frequency distribution, amplitude characteristics and time-frequency characteristics of the vibration signal are accurately extracted, and further combined with machine learning algorithms to achieve vibration mode modeling and anomaly detection. The unit can analyze the dynamic changes of vibration signals in real time, identify abnormal vibration behaviors that deviate from the normal range, and provide a reliable basis for pipeline health status assessment.

[0121] Pressure sensors are placed at the compressor stations, near key valves and pressure regulating points of the pipeline, which can fully reflect the dynamic changes in the pressure of the natural gas pipeline. The selected pressure sensors have high precision and fast response characteristics, and can capture millisecond-level pressure fluctuation signals.

[0122] The pressure dynamic monitoring unit adopts multi-channel synchronous sampling technology to ensure that the pressure sensor data from different locations can be accurately synchronized. The built-in data filtering algorithm can effectively suppress electromagnetic interference and high-frequency noise and extract the essential characteristics of the pressure signal. At the same time, the pressure dynamic monitoring unit also has a real-time abnormal mutation recognition function, which can quickly detect and alarm unexpected and drastic changes in pressure.

[0123] The pressure fluctuation feature analysis unit conducts in-depth mining of pressure data through complex signal processing algorithms. Fast Fourier Transform (FFT) is used to decompose the spectral characteristics of the pressure signal, and power spectrum density analysis can quantify the energy distribution of different frequency components. Wavelet packet decomposition technology can perform multi-scale analysis of pressure fluctuations in both time and frequency domains, and extract key parameters that reflect the fluid dynamics characteristics inside the pipeline.

[0124] The temperature sensor network is systematically arranged at the inlet, outlet, vertical section and bend section of the pipeline to form a dense and uniform temperature monitoring network. The location of each sensor point ensures that the temperature change characteristics of different areas of the pipeline can be accurately reflected. The selected temperature sensor has extremely high temperature measurement accuracy (±0.1℃) and fast response characteristics.

[0125] The temperature data acquisition unit is responsible for acquiring the monitoring data of the temperature sensor network in real time and efficiently. The unit uses high-speed data acquisition technology to perform multiple temperature sampling per second, and eliminates measurement noise and outliers through data preprocessing algorithms. The unit's built-in data communication module can reliably transmit temperature data and supports multiple communication protocols, such as Modbus TCP / IP, CAN bus, etc.

[0126] The temperature gradient calculation unit is the key to converting discrete temperature measurement data into a continuous and analyzable temperature distribution model. This unit uses spatial interpolation algorithms (such as Kriging interpolation) and finite difference methods to reconstruct the continuous temperature field along the pipeline from discrete temperature measurement points. By calculating the spatial derivative of temperature, the temperature gradient of different sections of the pipeline can be accurately quantified and possible thermal anomaly areas can be identified.

[0127] The ultrasonic detector uses high-frequency (1-10MHz) ultrasonic detection technology, which can penetrate the pipe wall material, accurately measure the wall thickness change and detect internal defects. Through multi-angle and multi-frequency ultrasonic scanning, it can draw a three-dimensional structure diagram of the pipe wall and identify potential defects such as corrosion and cracks, with a resolution of up to 0.1mm.

[0128] The gas leak detector uses gas sensing technology, which can respond quickly (<1s) and accurately identify extremely small gas leaks (<10ppm). The detector can not only detect natural gas leaks, but also identify the composition and concentration of the leaked gas, providing key information for subsequent processing. Its high sensitivity and selectivity enable it to accurately locate the leak point in a complex environmental background.

[0129] The pipeline structure monitoring unit builds a pipeline structure health assessment model by integrating ultrasonic detection data and multi-source monitoring information. The pipeline structure monitoring unit uses machine learning and big data analysis technology to identify stress concentration areas, assess the degree of corrosion, and predict the failure risk of pipelines. By establishing a digital twin model of the pipeline structure, dynamic and accurate monitoring of the entire life cycle of the pipeline can be achieved.

[0130] The anomaly detection unit integrates data from each monitoring submodule and uses complex correlation analysis and anomaly recognition algorithms to accurately locate and evaluate the micro-leakage points and internal sediment status of the pipeline. The anomaly detection unit can automatically generate detailed anomaly detection reports and support multi-level early warning mechanisms to provide timely and accurate decision support for pipeline operation and maintenance personnel.

[0131] like Figure 5 As shown, the remote monitoring module specifically includes:

[0132] A multi-model integration submodule, for integrating at least two deep learning models, and obtaining a predicted value of odorant concentration based on the original concentration data corresponding to the multiple sensors and the pipeline operation data; the at least two deep learning models include at least two of the following: a capsule network, a deep belief network, and a sparse autoencoder;

[0133] The concentration confirmation submodule is used to generate a final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0134] In some embodiments, the first deep learning model is a capsule network, which is used to extract basic features of raw concentration data of multiple sensors through a convolutional layer, and associate local correlations of output data of multiple sensors in a capsule conversion layer and a capsule layer through a dynamic routing algorithm, thereby generating a first concentration sub-result.

[0135] In some embodiments, the second deep learning model is a deep belief network, which is used to analyze the raw concentration data and pipeline operation data of multiple sensors through two layers of restricted Boltzmann machine pre-training layers and multiple layers of deep network hidden layers, and use nonlinear activation functions to model the nonlinear relationship between the raw concentration data and the pipeline operation data, extract key features layer by layer, and finally generate a second concentration sub-result.

[0136] In some embodiments, the third deep learning model is a sparse autoencoder, which is used to extract sparse features from raw concentration data and pipeline operation data of multiple sensors, reconstruct the extracted sparse features, and obtain a third concentration sub-result based on the reconstructed sparse features.

[0137] In some embodiments, the first concentration sub-result, the second concentration sub-result and the third concentration sub-result are weightedly fused by a weighted average algorithm to generate a predicted value of odorant concentration.

[0138] In the natural gas pipeline odorant concentration monitoring scenario, this embodiment selects three deep learning models, namely capsule network, deep belief network and sparse autoencoder, for integration. The data sources include the original concentration data of optical fiber sensors, graphene sensors and micro-resonance cavity sensors, as well as operation data such as pipeline temperature and pressure. Through multi-model integration, the unique advantages of different deep learning models in feature extraction and data processing can be fully utilized to improve the overall performance and accuracy of odorant concentration monitoring.

[0139] The capsule network is mainly used to extract the spatial distribution characteristics of the raw concentration data of multiple sensors and associate the local correlation of the output data of multiple sensors. The network design includes an input layer (inputting the raw concentration data of multiple sensors, dimension 3x 30), a capsule conversion layer (using a dynamic routing algorithm), 6 16-dimensional capsule units, and an output layer. The specific network structure is: input layer → convolution layer (32 cores) → capsule conversion layer → capsule layer → fully connected layer → first concentration sub-result. By capturing the spatial distribution characteristics between sensor data, the capsule network can effectively identify the local correlation between different sensor outputs.

[0140] Specifically, in the scenario of the intelligent detection device for natural gas odorant concentration, the local correlation in the capsule network mainly refers to the local correlation and interdependence between the output data of different sensors in the spatial distribution and feature dimensions. This local correlation is reflected in the following aspects: First, different sensors (such as optical fiber sensors, graphene sensors, and micro-resonance cavity sensors) will capture slightly different but interrelated characteristic signals when detecting the concentration of natural gas odorants. When the concentration of odorant molecules changes, the output signals of these sensors show similar trends, but there are subtle differences. Through the dynamic routing algorithm, the capsule network can identify and learn the local correlation patterns of the output signals of these sensors in the spatial distribution and feature dimensions, including the synchronization of signal changes, the strength of correlation, and the similarity of local features. Secondly, the local correlation is also reflected in the spatial topological structure of the sensor data. Each sensor has a different response mechanism to odorant molecules, such as electrochemical sensors through current signals, spectral sensors through spectral absorption characteristics, and graphene sensors through conductivity changes. The capsule network can capture the topological relationship of these different sensor output signals in the local space, identify their relative positions and mapping relationships in the feature space, and thus more accurately understand and associate the local feature information of multiple sensors.

[0141] Specifically, local correlation refers to the microscopic, nonlinear interdependencies between the output data of different sensors. In a multi-sensor system, the data captured by each sensor may contain partially overlapping or complementary information. Through the dynamic routing algorithm of the capsule network, it is possible to identify subtle correlation patterns between these sensor data, such as the coordinated changes or potential interactive effects of certain sensors under specific conditions.

[0142] The dynamic routing algorithm learns the weighted connections between different capsule units in an iterative manner. It adaptively adjusts the correlation strength between sensor outputs based on the intrinsic characteristics of the data. Specifically, the dynamic routing algorithm dynamically enhances or suppresses the information transfer between different capsule units based on the similarity and consistency of the data, thereby capturing the complex local correlation patterns between sensor data.

[0143] The first concentration sub-result refers to the preliminary comprehensive prediction of multi-sensor concentration data. Through the previous network layers (convolutional layer, capsule conversion layer, capsule layer), the capsule network extracts cross-sensor features and related information. The fully connected layer further integrates these features, and the final output layer converts this comprehensive information into a specific concentration prediction value.

[0144] The process of obtaining local correlation is mainly based on the architecture of the capsule network. First, the convolution layer extracts basic features; then, the capsule conversion layer converts these features into capsule representations. In the six 16-dimensional capsule units, the dynamic routing algorithm gradually adjusts the connection weights between each capsule unit through iterative calculation. This process is similar to the attention mechanism in neural networks, which can adaptively capture the complex correlation patterns between sensor data, and finally generate the first concentration sub-result of comprehensive prediction through the fully connected layer and the output layer.

[0145] Specifically, the specific structure of the capsule network is as follows:

[0146] The input layer is the data entry of the capsule network, receiving the raw concentration data from multiple sensors with a data dimension of 3 x30. That is, there are 3 sensors, each with 30 data points. The main function of the input layer is to convert the raw multi-dimensional sensor data into a standardized input format that can be processed by the neural network, preparing data for subsequent feature extraction.

[0147] The convolution layer uses 32 convolution kernels, whose main purpose is to extract low-level spatial features from the raw input data. Through the convolution operation, the capsule network is able to capture local patterns and spatial correlations in the sensor data. 32 convolution kernels indicate that features can be extracted from different angles and scales at the same time, which enhances the capsule network's ability to represent input data. The convolution layer can be seen as a feature extractor that converts raw data into a more abstract and representative feature representation.

[0148] The capsule conversion layer uses a dynamic routing algorithm to convert the features extracted by the convolutional layer into capsule representations. This layer converts the traditional scalar feature representation into a vector representation with direction and magnitude. The dynamic routing algorithm adaptively learns the connection weights between different capsule units in an iterative manner, enabling the capsule network to capture the complex nonlinear relationships between data.

[0149] The capsule layer consists of 6 16-dimensional capsule units, each of which can be regarded as a vector carrying rich information. These capsule units not only store features, but also retain the spatial and hierarchical information of the features. Through the dynamic routing algorithm, these capsule units can collaborate with each other to jointly construct a higher-level representation of sensor data.

[0150] The fully connected layer is a bridge between the capsule layer and the output layer. Its main function is to integrate the complex features learned in the previous layers. Through the fully connected layer, the capsule network can map the scattered capsule features into a compact representation space to prepare for the final concentration prediction. This layer captures the global relationship between capsule features by learning weights, and further extracts the most critical information for concentration prediction.

[0151] The output layer is the final stage of the capsule network and directly generates the first concentration sub-result. It converts the feature representation of the fully connected layer into a specific concentration prediction value. This layer uses a specific activation function (such as linear activation or sigmoid) to ensure that the output is within a reasonable numerical range. As the last link of the capsule network, the output layer combines the features learned by all previous layers to provide the final multi-sensor joint concentration prediction result. In some embodiments, the output layer can be the last layer of the fully connected layer, which maps the features processed by the previous fully connected layer to the final prediction result.

[0152] The deep belief network is used to analyze the characteristic distribution of raw concentration data and pipeline operation data from multiple sensors, and uses the multi-layer nonlinear transformation of the deep network to reveal the nonlinear relationship between sensor data and pipeline operation data. The network design includes an input layer (multiple sensor raw data and pipeline operation data, dimension 4 x 30), a 2-layer restricted Boltzmann Machine (RBM) pre-training layer, a deep network with 5 hidden layers, and an output layer using ReLU as a nonlinear activation function. The specific network structure is: input layer → RBM pre-training → deep network → fully connected layer → second concentration sub-result. The deep belief network can capture the complex nonlinear relationship between sensor data and pipeline operation data.

[0153] Specifically, the specific structure of the deep belief network is as follows:

[0154] The input layer is the data entry of the deep belief network. It consists of raw concentration data from multiple sensors and pipeline operation data, with a dimension of 4 x 30. This means that the network receives 30 data points from 4 different sensors at the same time. The input layer converts the raw multi-dimensional data into numerical representations that can be processed by the network, and is the starting point of the entire deep belief network analysis.

[0155] In the RBM pre-training layer of the deep belief network, two consecutive restricted Boltzmann machines (RBMs) work together to achieve complex unsupervised feature learning. The first layer of RBM extracts preliminary features from the 4 x 30 raw multi-sensor data, maps the input to a hidden layer representation of about 120 neurons, and captures the basic statistical patterns and local correlations of the data; the second layer of RBM takes the output of the first layer of RBM as input, further compresses and abstracts the features, reduces the feature dimension to 60-80 neurons, and extracts a higher level of compressed feature representation. This process is based on the contrastive divergence algorithm. By simulating forward and backward propagation, the weights are continuously adjusted to make the reconstructed data distribution gradually close to the original input data distribution, and finally generate a more concise, abstract and representative compressed feature representation, realizing the gradual transformation from raw multi-dimensional data to compressed feature representation.

[0156] The hidden layer of the deep network contains 5 consecutive hidden layers, which use ReLU as a nonlinear activation function. Each layer presents progressive feature abstraction: the bottom hidden layer (the first 1-2 layers) takes the compressed feature representation of RBM pre-training as input, and captures the low-level local features extracted from these feature representations; the middle hidden layer (the middle 2-3 layers) integrates and reorganizes these low-level local features to form more complex and semantic intermediate features; the high-level hidden layer (the last 1-2 layers) further extracts more abstract and complex deep feature interactions based on the low-level local features and intermediate features extracted by the previous layers, expresses high-level semantic information and global patterns, and obtains high-dimensional abstract features. Through nonlinear transformation, each layer can capture the complex nonlinear relationship between sensor data and pipeline operation data, and realize multi-level and nonlinear mapping of data.

[0157] The fully connected layer is the conversion link after the hidden layer of the deep network. It fully integrates and maps the high-dimensional abstract features extracted by the high-level hidden layers (the last 1-2 layers) in the first 5 hidden layers. Through the linear transformation of the weight matrix, the fully connected layer further compresses and reconstructs these high-dimensional abstract features to generate a compact feature representation, which is a feature vector space with lower dimensions but more concentrated information. This layer plays the role of feature fusion and dimensionality reduction, converting complex high-dimensional features into a concise but information-rich compact feature representation.

[0158] The output layer directly receives this compact feature representation from the fully connected layer and generates the second density sub-result through the final nonlinear transformation (ReLU).

[0159] The sparse autoencoder is used to extract sparse features from the raw concentration data of multiple sensors, obtain the sparse features of the data through sparsity constraints (such as L1 regularization or KL divergence), and obtain the third concentration sub-result based on the extracted sparse features. The network design includes an input layer (raw data from multiple sensors, dimension 3 x 30), an encoder with 3 hidden layers using L1 regularization, a symmetric decoder with 3 hidden layers, and a sparsity constraint based on KL divergence (Kullback-LeiblerDivergence, also known as relative entropy). The specific network structure is: input layer → encoder (including hidden layer) → decoder (including hidden layer) → third concentration sub-result. Through sparse constraints, the autoencoder can learn the key sparse features of the data and effectively reduce the impact of redundant information.

[0160] In the sparse autoencoder for multi-sensor natural gas odorant concentration detection, the key sparse features are the smallest and most representative feature set extracted from the 3×30-dimensional raw sensor data. Through L1 regularization and KL divergence constraints, they can most sensitively capture the key signals of odorant concentration changes.

[0161] Sparse autoencoders are multi-layer neural networks used to extract the most critical and representative sparse features from multi-sensor raw data. This network architecture can effectively capture the essential characteristics of data and reduce the impact of redundant information through complex nonlinear transformations and sparsity constraints.

[0162] The input layer is the starting point of the multi-layer neural network. It consists of raw concentration data from multiple sensors. The specific dimension is 3 × 30, representing three types of sensors and 30 data points for each sensor.

[0163] The encoder consists of three hidden layers with L1 regularization, and its main function is to gradually compress and transform the input data. Through nonlinear transformation and L1 regularization, the encoder can selectively retain the most important sparse features and suppress unnecessary information. Each hidden layer will gradually reduce the data dimension, achieving efficient data compression and sparse feature extraction.

[0164] Sparse feature extraction is achieved through a variety of mechanisms. L1 regularization forces the network to learn sparse representations by adding an L1 norm penalty to the weights; the KL divergence constraint ensures that the number of activated neurons is limited, further enhancing the sparsity of features; nonlinear activation functions (such as ReLU) also help produce sparse feature representations.

[0165] The hidden layer is an integral part of the encoder and decoder, and is also a key stage for sparse feature representation. In this layer, the data is highly concentrated, retaining only the most critical sparse information features. The sparsity constraint based on KL divergence ensures that only the most meaningful features are retained, enabling the network to extract the essential sparse features of the data.

[0166] The decoder is a symmetrical structure of the encoder, consisting of three hidden layers corresponding to the encoder. Its task is to gradually restore the highly compressed sparse feature information and try to reconstruct a data representation close to the original input. Through the symmetrical network structure, the decoder can not only reconstruct the data, but also learn the intrinsic sparse representation of the data.

[0167] As a symmetric mapping of the encoder, the decoder is a sophisticated feature reconstruction network consisting of three hidden layers corresponding to the encoder structure. Its task is to restore the sparse feature information compressed to low dimensions in high-dimensional space layer by layer through nonlinear transformation, and finally reconstruct a data representation that is as close to the original input as possible.

[0168] The hidden layers of the encoder are responsible for data compression and sparse feature extraction, while the hidden layers of the decoder are responsible for sparse feature restoration and data reconstruction, working together to achieve efficient sparse representation of data. This design enables the network to extract the most critical information from complex multi-sensor data.

[0169] The process of generating the third concentration sub-result by the sparse autoencoder is a sophisticated feature extraction and reconstruction process. First, the encoder extracts the most representative sparse features from the original 3×30-dimensional multi-sensor data through multiple layers of nonlinear transformation and L1 regularization; then, these sparse features are used as the input of the decoder, and the decoder gradually reconstructs the data representation through symmetrical hidden layers. Its purpose is not only to verify the effectiveness of feature extraction, but also to provide refined feature information for subsequent concentration prediction. The reconstruction process is a self-supervised learning mechanism that continuously optimizes the feature representation by comparing the reconstructed results with the original input. Finally, based on the sparse features reconstructed by the decoder, the network outputs the third concentration sub-result, which represents the most refined and meaningful odorant concentration information extracted from complex multi-sensor data, providing key feature input for multi-model integration.

[0170] The sub-results of the three deep learning models are weighted and fused, and the initial weights are set as: capsule network weight W1 = 0.4, deep belief network weight W2 = 0.3, sparse autoencoder weight W3 = 0.3. The fusion formula is: concentration prediction value = W1 * first concentration sub-result + W2 * second concentration sub-result + W3 * third concentration sub-result. This weighted average method can comprehensively utilize the prediction results of different models, balance the advantages and disadvantages of each model, and improve the accuracy of the final concentration prediction.

[0171] The concentration confirmation submodule receives the calibrated odorant concentration detection data (2.09 ppm) and the multi-model concentration prediction value (2.11 ppm). The confirmation algorithm first calculates the error between the predicted value and the detected value, and sets the error threshold to ±0.05 ppm. If the error is within the threshold, the prediction result is considered valid. In this case, the error is 0.02 ppm, which is less than the threshold, so the final odorant concentration confirmation result is output as 2.10 ppm. This confirmation mechanism can effectively filter out abnormal or unreliable prediction results and improve the reliability of the monitoring system.

[0172] like Figure 6 As shown, the remote monitoring module further includes: an abnormal warning submodule, which is used to determine the risk level according to the final odorant concentration confirmation result; when the risk level exceeds a preset warning threshold, an abnormal warning is triggered.

[0173] like Figure 6 As shown, further, the remote monitoring module also includes: a dynamic risk response parameter generation submodule, which is used to generate a dynamic risk response parameter adjustment plan according to the risk level, the pipeline operation data and the environmental data obtained in real time by the environmental sensors arranged around the pipeline; the dynamic risk response parameter adjustment plan includes adjusting the operating parameters of the natural gas odorant concentration intelligent detection equipment, adjusting the pipeline pressure parameters or adjusting the odorant concentration control parameters; the dynamic risk response parameter adjustment plan is sent to the corresponding execution unit through remote instructions.

[0174] Specifically, in the natural gas pipeline odorant concentration monitoring system, the abnormal warning submodule adopts a multi-dimensional risk assessment method to divide the risk level into four levels: low risk (green), medium risk (yellow), high risk (orange) and extremely high risk (red). The assessment is based on the odorant concentration confirmation results, historical data and real-time monitoring parameters. For natural gas pipelines with a standard odorant concentration of 2.0 ppm, the warning mechanism will be triggered when the concentration is lower than 1.8 ppm or higher than 2.2 ppm. Different concentration deviations correspond to different risk levels.

[0175] The risk levels are divided into precise numerical ranges and corresponding warning mechanisms. Green (low risk) is defined as odorant concentrations in the range of 1.9-2.1 ppm; yellow (medium risk) is 1.8-1.9 ppm or 2.1-2.2 ppm; orange (high risk) is 1.6-1.8 ppm or 2.2-2.4 ppm; and red (very high risk) is below 1.6 ppm or above 2.4 ppm. Each risk level corresponds to a different warning level and treatment strategy to ensure timely and accurate response to potential risks.

[0176] The dynamic risk response parameter generation submodule is an intelligent decision-making system that integrates multi-source data for risk assessment and parameter adjustment. The data collected by this module include odorant concentration confirmation results, pipeline operation data (such as pipeline vibration data, pipeline pressure fluctuation characteristic data, pipeline temperature gradient and heat distribution data, pipeline health characteristic data) and surrounding environment data (such as temperature, humidity, air pressure). Using machine learning algorithms, the system can quickly and accurately generate targeted risk response plans and achieve refined management of natural gas pipeline odorant concentration.

[0177] The risk response parameter adjustment includes three main dimensions: natural gas odorant concentration intelligent detection equipment parameters, pipeline pressure parameters and odorant concentration control parameters. For the detection equipment, the sensor sampling frequency can be adjusted, the signal processing algorithm can be optimized, and the calibration parameters can be modified; the pipeline pressure parameter adjustment includes fine-tuning the delivery pressure, adjusting the pressure buffer zone, and optimizing the pressure fluctuation control strategy; the odorant concentration control parameter adjustment involves fine-tuning the odorant injection amount, optimizing the odorant mixing algorithm, and adjusting the odorant injection point location.

[0178] The remote command issuing system adopts multiple security verification mechanisms to ensure the reliability and security of the adjustment plan. The complete command issuing process includes risk assessment and plan generation, security verification, multi-level authorization review, encrypted transmission and execution unit feedback confirmation. This strict process design can effectively prevent misoperation and ensure the safety and stability of natural gas pipeline operation.

[0179] When the odorant concentration in a certain section of the pipeline is detected to be in the orange high-risk range (2.2-2.4 ppm), the system will perform a series of intelligent adjustment measures. Specific operations include increasing the sensor sampling frequency to once every 5 seconds, reducing the pipeline pressure by 0.2 MPa, reducing the odorant injection volume by 10%, and sending adjustment instructions to the on-site execution unit through an encrypted channel, while continuously monitoring the adjustment effect to ensure that the system can respond quickly and effectively to potential risks.

[0180] To further enhance the security and reliability of the system, the remote monitoring module has set up multiple security protection mechanisms. These mechanisms include setting up multiple command verification processes, retaining complete operation logs, supporting remote manual intervention, and having automatic rollback functions. Through these multi-level security protection measures, the system operation risks can be minimized to the greatest extent and the safe operation of the natural gas pipeline can be guaranteed.

[0181] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the above-mentioned method embodiment, which will not be repeated here.

[0182] like Figure 7 As shown, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the following intelligent detection method for natural gas odorant concentration suitable for remote monitoring is implemented, which specifically includes the following steps:

[0183] Using an odorant concentration sensor array, a variety of sensors are used to detect raw concentration data of odorants in natural gas;

[0184] Through the cross-validation module, the original concentration data corresponding to various sensors are fused and processed to obtain the calibrated odorant concentration detection data;

[0185] Through the pipeline monitoring module, the pipeline operation data corresponding to the pipeline transporting natural gas is obtained;

[0186] Connecting the cross-validation module and the remote monitoring module, and connecting the pipeline monitoring module and the remote monitoring module through the communication module;

[0187] Through the remote monitoring module, the odorant concentration prediction value is determined based on the integrated multi-model according to the original concentration data and pipeline operation data corresponding to various sensors; the final odorant concentration confirmation result is obtained based on the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0188] In some embodiments, the step of detecting raw concentration data by the odorant concentration sensor array specifically includes any of the following:

[0189] Using an electrochemical sensor, the odorant molecules are oxidized or reduced through an electrochemical reaction between a working electrode and a reference electrode to generate a current signal related to the concentration; the concentration of the odorant is determined based on the intensity of the current signal, and the type of the odorant is identified based on the waveform characteristics of the current signal;

[0190] Using a spectral sensor, based on the absorption characteristics or scattering characteristics of the odorant molecules to the target wavelength light, the wavelength position and absorption intensity of the spectral absorption peak are obtained; the type of the odorant is determined by matching the known odorant molecular characteristic spectrum according to the wavelength position of the spectral absorption peak, and the concentration of the odorant is calculated according to the absorption intensity;

[0191] When the odorant molecules are adsorbed onto the sensitive material surface of the sensor using a gas sensitive resistor sensor, a resistance change curve is obtained by detecting the characteristic value of the resistance change over time; the concentration of the odorant is detected based on the amplitude and response time of the resistance change curve.

[0192] In some embodiments, the step of detecting raw concentration data by the odorant concentration sensor array further includes any of the following:

[0193] Using an optical fiber sensor, by monitoring the characteristics of light signal intensity changing with time, a light signal intensity change curve is obtained, from which the attenuation amplitude of the light signal intensity is extracted; the concentration of the mercaptan odorant is calculated according to the attenuation amplitude of the light signal intensity;

[0194] Using a micro-resonance cavity sensor, the offset of the resonance frequency is obtained by monitoring the resonance response change characteristics of the micro-resonance cavity; the concentration of the odorant is detected according to the amplitude of the offset of the resonance frequency;

[0195] Using a graphene sensor, a conductivity change curve is obtained by real-time detection of a characteristic value of the conductivity of the graphene sensor that changes with time; and the concentration of the odorant is detected according to the amplitude and dynamic response characteristics of the conductivity change determined in the conductivity change curve;

[0196] A surface enhanced Raman spectroscopy sensor is used to obtain a Raman spectral peak by detecting the intensity change characteristics of the target wavelength in the Raman scattering signal; the type of odorant is determined by matching the known molecular characteristic Raman spectrum according to the wavelength position of the Raman spectral peak, and the concentration of the odorant is determined according to the intensity of the Raman spectral peak.

[0197] In some embodiments, the processing steps of the cross-validation module specifically include:

[0198] Based on the data dispersion within the monitoring period, the performance stability of each sensor is analyzed, and the corresponding stability index of each sensor is calculated;

[0199] Based on the degree of deviation of working environment conditions, analyze the adaptability of each sensor in the current working environment, and calculate the environmental adaptability index corresponding to each sensor;

[0200] Based on the multi-dimensional characteristics of each sensor, the matching degree of each sensor in the current detection scene is analyzed, and the scene compatibility index corresponding to each sensor is calculated;

[0201] The dynamic weight of each sensor is calculated through the stability index, environmental adaptability index and scene compatibility index;

[0202] Based on the dynamic weight of each sensor, the raw concentration data corresponding to the multiple sensors are weighted fused to obtain the fused odorant concentration detection value;

[0203] The fused odorant concentration detection values ​​are error corrected to obtain calibrated odorant concentration detection data.

[0204] In some embodiments, the monitoring steps of the pipeline monitoring module specifically include at least one of:

[0205] The vibration sensor is used to collect pipeline vibration data, and the frequency domain analysis and time domain feature extraction are performed on the pipeline vibration data to determine the frequency, amplitude and vibration mode of the pipeline vibration;

[0206] The natural gas pressure signal in the pipeline is collected in real time by the pressure sensor, and the amplitude, frequency and pulse characteristics of the pressure fluctuation are obtained based on the dynamic change characteristics and frequency characteristic analysis method of the natural gas pressure signal;

[0207] Continuously collect temperature data along the pipeline through a temperature sensor, and determine the pipeline surface temperature change and / or longitudinal temperature gradient based on the spatial distribution characteristics and time variation characteristics of the temperature data;

[0208] Ultrasonic detectors and leak detectors are used to determine the physical condition of the pipeline, including the degree of corrosion, distribution of leak points and the state of internal sediments.

[0209] In some embodiments, the processing steps of the remote monitoring module specifically include:

[0210] Integrate at least two deep learning models to obtain odorant concentration prediction values ​​based on original concentration data and / or pipeline operation data corresponding to a plurality of sensors; the at least two deep learning models include at least two of a capsule network, a deep belief network, and a sparse autoencoder;

[0211] The final odorant concentration confirmation result is generated based on the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0212] In some embodiments, the method further comprises:

[0213] The risk level is determined based on the final odorant concentration confirmation result; when the risk level exceeds the preset warning threshold, an abnormal warning is triggered.

[0214] In some embodiments, the method further comprises:

[0215] A dynamic risk response parameter adjustment plan is generated based on the risk level, pipeline operation data and environmental data obtained in real time by environmental sensors installed around the pipeline; the dynamic risk response parameter adjustment plan includes adjusting the operating parameters of the natural gas odorant concentration intelligent detection equipment, adjusting the pipeline pressure parameters or adjusting the odorant concentration control parameters; the dynamic risk response parameter adjustment plan is sent to the corresponding execution unit through remote instructions.

[0216] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. Of course, there are other forms of readable storage media, such as quantum memory, graphene memory, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunication signals.

[0217] The present invention also provides an electronic device. The electronic device of the embodiment of the present invention includes: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement a natural gas odorant concentration intelligent detection method suitable for remote monitoring and executed by a remote monitoring module provided by the present invention, which includes the following steps:

[0218] Receive raw concentration data and pipeline operation data corresponding to various sensors;

[0219] Based on the integrated multi-model, the received raw concentration data and pipeline operation data are analyzed to determine the predicted value of odorant concentration;

[0220] receiving calibrated odorant concentration detection data provided by the cross-validation module;

[0221] A final odorant concentration confirmation result is generated according to the calibrated odorant concentration detection data and the odorant concentration prediction value.

[0222] In some embodiments, the step of determining the predicted value of odorant concentration based on the integrated multi-model specifically includes:

[0223] Integrate at least two deep learning models, including at least two of capsule networks, deep belief networks, and sparse autoencoders;

[0224] Using the deep learning model to perform multi-dimensional feature extraction and pattern recognition on the original concentration data and pipeline operation data;

[0225] Through the multi-model integration algorithm, the prediction results of each model are integrated to generate the predicted value of odorant concentration.

[0226] In some embodiments, the method further comprises: determining a risk level according to a final odorant concentration confirmation result; and triggering an abnormality warning when the risk level exceeds a preset warning threshold.

[0227] In some embodiments, the step of determining the risk level specifically includes:

[0228] Compare the odorant concentration confirmation result with the preset standard concentration range;

[0229] According to the degree of concentration deviation, the risk level is divided into low risk, medium risk, high risk and very high risk;

[0230] Based on the severity of the risk level, determine the corresponding warning level and handling strategy.

[0231] In some embodiments, the method further comprises:

[0232] Generate a dynamic risk response parameter adjustment plan based on the risk level, pipeline operation data and environmental data obtained by environmental sensors; the dynamic risk response parameter adjustment plan includes: adjusting the operating parameters of the natural gas odorant concentration intelligent detection device, adjusting the pipeline pressure parameters or adjusting the odorant concentration control parameters;

[0233] The dynamic risk response parameter adjustment plan is sent to the corresponding execution unit through remote instructions.

[0234] In some embodiments, the step of issuing the adjustment plan specifically includes:

[0235] Perform multiple security verifications on the adjustment plan, conduct a comprehensive risk assessment on the adjustment plan through a risk assessment model or a security check algorithm, and generate a risk assessment report that lists the risk level, risk type, and risk controllability; if the risk assessment result shows that the risk is controllable and within an acceptable range, proceed to the next step;

[0236] After the risk verification is passed, the adjustment plan will be reviewed and authorized step by step through the preset multi-level authority management system;

[0237] Adopting encryption transmission method based on quantum key distribution to ensure information security and communication integrity during instruction transmission;

[0238] Monitor and obtain the execution feedback of the execution unit on the adjustment plan in real time, and record the execution status and results.

[0239] Reference below Figure 8 , which shows a schematic diagram of the structure of a computer system 800 of an electronic device suitable for implementing an embodiment of the present invention. Figure 8 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0240] like Figure 8 As shown, the computer system 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage part 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The CPU 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0241] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed.

[0242] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above.

[0243] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0244] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent detection device for natural gas odorant concentration suitable for remote monitoring, characterized in that: include: An odorant concentration sensor array is used to detect raw concentration data of odorants in natural gas through a variety of sensors; A cross-validation module is used to fuse the original concentration data corresponding to the multiple sensors to obtain calibrated odorant concentration detection data; A pipeline monitoring module is used to obtain pipeline operation data corresponding to the pipeline that transports natural gas; A communication module, used to connect the cross-validation module and the remote monitoring module, and to connect the pipeline monitoring module and the remote monitoring module; A remote monitoring module, for determining an odorant concentration prediction value based on the integrated multi-model according to the original concentration data respectively corresponding to the multiple sensors and the pipeline operation data; and obtaining a final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value; The remote monitoring module specifically includes: A multi-model integration submodule, for integrating at least two deep learning models, and obtaining a predicted value of odorant concentration based on the original concentration data corresponding to the multiple sensors and the pipeline operation data; the at least two deep learning models include at least two of the following: a capsule network, a deep belief network, and a sparse autoencoder; The concentration confirmation submodule is used to generate a final odorant concentration confirmation result according to the calibrated odorant concentration detection data and the odorant concentration prediction value.

2. The natural gas odorant concentration intelligent detection device according to claim 1 is characterized in that: The odorant concentration sensor array specifically includes any of the following: An electrochemical sensor for oxidizing or reducing odorant molecules through an electrochemical reaction between the working electrode and a reference electrode when the odorant molecules come into contact with the surface of the working electrode of the electrochemical sensor, thereby generating a current signal related to the concentration; determining the concentration of the odorant according to the intensity of the current signal, and identifying the type of the odorant according to the waveform characteristics of the current signal; the waveform characteristics include the oxidation peak position or the reduction peak position; A spectral sensor for obtaining the wavelength position and absorption intensity of a spectral absorption peak based on the absorption characteristics or scattering characteristics of the odorant molecules to the target wavelength light when the odorant molecules pass through the optical detection area of ​​the spectral sensor; matching a known odorant molecular characteristic spectrum according to the wavelength position of the spectral absorption peak to determine the type of the odorant, and calculating the concentration of the odorant according to the absorption intensity; A gas-sensitive resistor sensor is used to cause the resistance of the sensitive material to change when the odorant molecules are adsorbed on the surface of the sensitive material of the gas-sensitive resistor sensor; and obtain a resistance change curve by real-time detection of the characteristic value of the resistance of the gas-sensitive resistor sensor changing with time; The concentration of the odorant is detected according to the amplitude and response time of the resistance change curve.

3. The intelligent detection device for natural gas odorant concentration according to claim 2 is characterized in that: The odorant concentration sensor array further includes any of the following: An optical fiber sensor, for causing attenuation of light signal intensity by the absorption or scattering characteristics of target wavelength light when odorant molecules contact the sensitive layer on the optical fiber surface of the optical fiber sensor; detecting the time-varying characteristics of light signal intensity in real time by the optical detection device of the optical fiber sensor, obtaining a light signal intensity variation curve, extracting the attenuation amplitude of light signal intensity from the light signal intensity variation curve; and calculating the concentration of the mercaptan odorant according to the attenuation amplitude of the light signal intensity; A micro-resonance cavity sensor, used for, when odorant molecules come into contact with a sensitive layer on the surface of the micro-resonance cavity sensor, causing the resonant frequency of the micro-resonance cavity to shift due to the adsorption of the odorant molecules; obtaining the shift of the resonant frequency by monitoring the change characteristics of the resonant response of the micro-resonance cavity; and detecting the concentration of the odorant according to the amplitude of the shift of the resonant frequency; A graphene sensor, for when odorant molecules come into contact with the graphene sensitive layer of the graphene sensor, the odorant molecules are adsorbed to cause the conductivity of the graphene sensitive layer to change; a conductivity change curve is obtained by real-time detection of characteristic values ​​of the conductivity of the graphene sensor changing with time; and the concentration of the odorant is detected according to the amplitude and dynamic response characteristics of the conductivity change determined from the conductivity change curve; A surface enhanced Raman spectroscopy sensor, for enhancing the Raman scattering signal of the odorant molecule due to the surface enhanced Raman scattering effect when the odorant molecule contacts the metal nanostructure surface of the surface enhanced Raman spectroscopy sensor; The Raman spectrum peak is obtained by detecting the intensity change characteristics of the target wavelength in the Raman scattering signal; the type of the odorant is determined according to the wavelength position of the Raman spectrum peak and the concentration of the odorant is determined according to the intensity of the Raman spectrum peak.

4. The intelligent detection device for natural gas odorant concentration according to claim 1 is characterized in that: The cross-validation module specifically includes: The stability determination submodule is used to analyze the performance stability of each sensor based on the degree of data dispersion within the monitoring period, and calculate the stability index of each sensor within the monitoring period; The environmental adaptability determination submodule is used to analyze the adaptability of each sensor in the current working environment based on the degree of deviation of the working environment conditions, and calculate the environmental adaptability index corresponding to each sensor; The scene compatibility determination submodule is used to analyze the matching degree of each sensor in the current detection scene based on the multi-dimensional characteristics of each sensor, and calculate the scene compatibility index corresponding to each sensor; A dynamic weight calculation submodule, used to calculate the dynamic weight of each sensor through the stability index, the environmental adaptability index and the scene compatibility index; A fusion submodule is used to perform weighted fusion on the raw concentration data corresponding to the multiple sensors based on the dynamic weight of each sensor to obtain a fused odorant concentration detection value; The calibration submodule is used to perform error correction on the fused odorant concentration detection value to obtain calibrated odorant concentration detection data.

5. The intelligent detection device for natural gas odorant concentration according to claim 1 is characterized in that: The pipeline monitoring module specifically includes at least one of the following: The pipeline vibration characteristic monitoring submodule is used to collect pipeline vibration data through a vibration sensor, and perform frequency domain analysis and time domain feature extraction on the pipeline vibration data to determine the frequency, amplitude and vibration mode of the pipeline vibration; The pipeline pressure fluctuation monitoring submodule is used to collect the natural gas pressure signal in the pipeline in real time through the pressure sensor, and obtain the amplitude, frequency and pulse characteristics of the pressure fluctuation based on the dynamic change characteristics and frequency characteristic analysis method of the natural gas pressure signal; The pipeline temperature gradient monitoring submodule is used to continuously collect temperature data along the pipeline through a temperature sensor, and determine the pipeline surface temperature change and / or longitudinal temperature gradient according to the spatial distribution characteristics and time variation characteristics of the temperature data; The pipeline health monitoring submodule is used to determine the physical state of the pipeline, including the degree of corrosion, distribution of leak points and the state of internal sediments, through ultrasonic detectors and leak detectors.

6. The intelligent detection device for natural gas odorant concentration according to claim 1 is characterized in that: The remote monitoring module further includes: The abnormal warning submodule is used to determine the risk level according to the final odorant concentration confirmation result; when the risk level exceeds a preset warning threshold, trigger an abnormal warning.

7. The intelligent detection device for natural gas odorant concentration according to claim 6 is characterized in that: The remote monitoring module further includes: The dynamic risk response parameter generation submodule is used to generate a dynamic risk response parameter adjustment plan based on the risk level, the pipeline operation data and the environmental data obtained in real time by the environmental sensors arranged around the pipeline; the dynamic risk response parameter adjustment plan includes adjusting the operating parameters of the natural gas odorant concentration intelligent detection equipment, adjusting the pipeline pressure parameters or adjusting the odorant concentration control parameters; the dynamic risk response parameter adjustment plan is sent to the corresponding execution unit through remote instructions.

8. The intelligent detection device for natural gas odorant concentration according to claim 5, characterized in that: The pipeline vibration characteristic monitoring submodule specifically includes: A vibration sensor array is arranged at a pipe bend, flange connection or reducer section of the pipeline to obtain a vibration signal of the pipeline; A vibration data acquisition unit, used for acquiring vibration signals of the vibration sensor array in real time; The vibration feature extraction unit is used to perform frequency analysis, amplitude calculation and vibration mode recognition on the collected vibration signal to obtain the frequency feature, amplitude feature and vibration mode feature of the vibration signal.

9. The intelligent detection device for natural gas odorant concentration according to claim 5, characterized in that: The pipeline pressure fluctuation monitoring submodule specifically includes: The pressure sensor group is installed at the compressor station, near the valve or at the pressure regulating point of the pipeline to collect the pressure signal of the natural gas in the pipeline in real time; A pressure dynamic monitoring unit is connected to the pressure sensor group for communication, receives the pressure signal collected by the pressure sensor group, captures the transient change of the natural gas pressure in the pipeline in real time, and pre-processes the pressure signal to obtain the pre-processed pressure data; The pressure fluctuation characteristic analysis unit is used to calculate the amplitude, frequency and pulse characteristics of the pressure fluctuation as pressure fluctuation characteristic parameters based on the preprocessed pressure data.

Images