Gas pipeline leakage rapid detection method and system

Abstract

The invention discloses a gas pipeline leakage rapid detection method and system, and belongs to the technical field of computers. According to the technical scheme, the problem that the reliability of the sensor is reduced due to environmental interference is solved to a certain extent, and optimal utilization of the multi-modal data is ensured through dynamic weight distribution. The cross-modal feature fusion mechanism significantly improves the feature recognition precision in a complex leakage scene, and the modal collaboration degree parameter effectively eliminates abnormal data interference. The introduction of spatio-temporal evolution characteristics provides diffusion path prediction for emergency disposal, and forms a full-flow solution of coverage detection, positioning and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a gas pipeline leakage rapid detection method and system. BACKGROUND

[0002] As an important infrastructure for urban energy transportation, the safe operation of gas pipelines is directly related to public safety and energy supply stability. The leakage detection technology widely used in the industry currently has the following technical bottlenecks: first, the traditional acoustic detection method is easily disturbed by environmental noise, and has a high false alarm rate under complex working conditions; second, the infrared thermal imaging technology is limited by the detection distance and environmental temperature, and it is difficult to realize continuous monitoring of long-distance pipelines; third, although the laser methane remote sensing technology has high detection accuracy, it cannot provide accurate positioning information of the leakage point. These single-sensor detection methods often show obvious limitations when facing complex pipeline networks and changing environments in actual application scenarios.

[0003] The multi-sensor fusion scheme in the related art mainly adopts a simple data superposition or fixed weight fusion strategy, which fails to effectively solve the following key problems: first, there is a lack of adaptive processing mechanism for dynamic changes in sensor performance, and when the performance of the sensor decreases due to aging or environmental interference, the detection accuracy of the system is significantly reduced; second, a scientific cross-modal feature fusion framework has not been established, and the complementary advantages between different sensing data cannot be fully utilized; third, the modeling of spatiotemporal evolution characteristics of leakage is insufficient, and accurate diffusion trend prediction cannot be provided for emergency decision-making. These problems seriously restrict the real-time performance and reliability of gas pipeline leakage detection, and pose a major risk to the safe operation of pipelines.

[0004] In view of the above problems, the related art needs to be improved. SUMMARY

[0005] The embodiments of the present application provide a gas pipeline leakage rapid detection method and system, which can improve the multi-modal data fusion accuracy, enhance the dynamic adaptability of sensors, and realize the prediction of spatiotemporal evolution of leakage. The technical solutions are as follows: On the one hand, a gas pipeline leakage rapid detection method is provided, which comprises: The multi-source sensing device is arranged at a key monitoring point of the gas pipeline to collect multi-modal monitoring data, and the multi-source sensing system includes a distributed fiber sensor array, a laser methane detection unit, a sound wave sensor group, and an environmental parameter monitoring module; based on performance parameters of each sensing device in the multi-source sensing device and environmental influence factors, a data confidence weight matrix is determined, the data confidence weight matrix is used to represent a contribution degree of each modal monitoring data in the multi-modal monitoring data in the current leakage detection task; based on the multi-modal monitoring data and the data confidence weight matrix, a multi-modal monitoring data comprehensive feature and a modal coordination degree parameter are determined, the comprehensive feature is a high-order feature representation after fusion of the multi-modal monitoring data, and the modal coordination degree parameter is used to quantify a consistency degree of detection results obtained based on different modal monitoring data; based on the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter, a leakage detection result of the gas pipeline is determined, the leakage detection result includes a leakage probability distribution and a spatio-temporal evolution feature, the leakage probability distribution is used to describe a possibility of leakage at each position of the gas pipeline, and the spatio-temporal evolution feature is used to represent a propagation and change rule of a leakage point in a spatio-temporal dimension after leakage at each position.

[0006] Further, the application further provides that based on the multi-modal monitoring data and the data confidence weight matrix, the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter are determined, including: based on the multi-modal monitoring data and the data confidence weight matrix, cross-modal feature extraction and enhancement are performed, a cross-modal feature set and a feature saliency vector are determined, the cross-modal feature set is an enhanced feature obtained after fusion of monitoring data of different modes, and the feature saliency vector is used to quantify a contribution degree of each feature dimension to the leakage detection task; based on the cross-modal feature set and the feature saliency vector, a target feature representation and a modal correlation measure are determined, the target feature representation is obtained by coupling the feature saliency vector and the cross-modal feature set, and the modal correlation measure is used to evaluate a coupling strength of monitoring data of different modes at a feature level; based on the target feature representation and the modal correlation measure, the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter are determined.

[0007] Further, the application also proposes that based on the multi-modal monitoring data and the data confidence weight matrix, cross-modal feature extraction and enhancement are performed to determine a cross-modal feature set and a feature saliency vector, including: performing time domain feature analysis and frequency domain feature conversion on the multi-modal monitoring data to obtain a time domain feature set and a frequency domain feature set, the time domain feature set containing statistical features and waveform features of the monitoring data, and the frequency domain feature set containing spectral features and energy distribution features of the monitoring data; performing feature enhancement and weight optimization on the time domain feature set, the frequency domain feature set, and the data confidence weight matrix to obtain enhanced time-frequency domain features and feature weight distribution, the feature weight distribution reflecting the importance degree of each feature dimension; performing cross-modal feature fusion and contribution evaluation on the enhanced time-frequency domain features and the feature weight distribution to obtain the cross-modal feature set and the feature saliency vector.

[0008] Further, the application also proposes that based on the cross-modal feature set and the feature saliency vector, target feature representation and modality correlation measure are determined, including: performing feature dimension reduction and feature selection on the cross-modal feature set and the feature saliency vector to obtain reduced feature representation and feature retention index, the feature dimension reduction and feature selection being used to retain main feature information, and the feature retention index being used to represent the information loss degree in the dimension reduction process; performing feature fusion and feature optimization on the reduced feature representation and the feature retention index to determine preliminary fusion features and feature quality parameters, the feature quality parameters being used to represent the reliability and stability of the preliminary fusion features; based on the preliminary fusion features and the feature quality parameters, target feature representation and modality correlation measure are determined.

[0009] Further, the application also proposes that based on the target feature representation and the modality correlation measure, multi-modal monitoring data comprehensive features and modality synergy degree parameters are determined, including: performing feature optimization and redundant feature elimination on the target feature representation and the modality correlation to obtain optimized feature representation and feature distribution consistency index, the feature optimization being achieved through feature selection and reconstruction, and the feature distribution consistency index being used to evaluate the consistency degree of the optimized features among different modalities; performing feature standardization and distribution regularization on the optimized feature representation and the feature distribution consistency index to obtain standardized feature representation and feature stability parameters, the standardization processing making features of different dimensions comparable, and the feature stability parameters being used to represent the stability and reliability of the features in the time dimension; performing multi-dimensional feature aggregation and synergy degree calculation on the standardized feature representation and the feature stability parameters to obtain multi-modal monitoring data comprehensive features and modality synergy degree parameters.

[0010] Further, the application also proposes to determine the data confidence weight matrix based on the performance parameters of each sensor device in the multi-source sensing device and the environmental influence factors, including: performing sensor performance and environmental fitness analysis on the performance parameters and the environmental influence factors to obtain an initial weight distribution corresponding to the performance parameters and an environmental fitness coefficient, the environmental fitness coefficient being used to quantify the working performance retention capability of each sensor device in the multi-source sensing device under the current comprehensive environmental conditions; performing dynamic weight fusion and stability evaluation on the initial weight distribution and the environmental fitness coefficient to obtain an optimized weight distribution corresponding to the performance parameters and a weight stability index, the optimized weight distribution being used to reflect the optimal weight configuration under multi-objective optimization, and the weight stability index being used to represent the fluctuation degree and reliability level of the weight distribution in the time sequence; and performing online calibration and optimization on the optimized weight distribution and the weight stability index to obtain the data confidence weight matrix.

[0011] Further, the application also proposes to perform sensor performance and environmental fitness analysis on the performance parameters and the environmental influence factors to obtain an initial weight distribution corresponding to the performance parameters and an environmental fitness coefficient, including: performing multi-dimensional quantitative evaluation on the performance parameters to obtain a performance score vector, the performance parameters including measurement accuracy, response time and stability index; performing interference degree analysis on the environmental influence factors to obtain an environmental interference coefficient, the environmental influence factors including temperature, humidity, wind speed and precipitation intensity; and calculating the initial weight distribution and the environmental fitness coefficient based on the performance score vector and the environmental interference coefficient by using a weight distribution algorithm.

[0012] Further, the application also proposes to dynamically fuse the initial weight distribution and the environmental fitness coefficient and evaluate the stability to obtain the optimized weight distribution corresponding to the performance parameter and the weight stability index, including: using a multi-objective optimization method to fuse and process the initial weight distribution and the environmental fitness coefficient to obtain a preliminary optimized weight; performing time series analysis on the preliminary optimized weight to obtain a weight fluctuation feature of the preliminary optimized weight, the weight fluctuation feature being used to reflect the change of the preliminary optimized weight over time; based on the weight fluctuation feature, using a stability evaluation algorithm to generate a weight stability index, the stability evaluation algorithm quantifying the stability level by calculating the variance and the coefficient of variation of the weight distribution; based on the preliminary optimized weight and the weight stability index, using a constrained optimization method to generate an optimized weight distribution, the constrained optimization method being used to constrain the weight distribution to meet the performance optimization and stability requirements at the same time; performing online calibration and optimization on the optimized weight distribution and the weight stability index to obtain a data confidence weight matrix, including: based on the optimized weight distribution and the weight stability index, using a sliding window statistical method to analyze the multi-modal monitoring data to obtain a weight adjustment amount and a performance deviation index; based on the weight adjustment amount and the weight stability index, dynamically adjusting the weight parameter through an incremental learning algorithm to obtain a calibrated weight distribution; performing regularization processing and stability verification on the calibrated weight distribution to obtain the data confidence weight matrix.

[0013] Further, the application also proposes to determine the leakage detection result of the gas pipeline based on the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter, including: based on the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter, performing leakage state recognition through a deep learning classification model to obtain a preliminary leakage probability distribution and a classification confidence; based on the preliminary leakage probability distribution and the classification confidence, performing leakage diffusion simulation through a time-space evolution analysis algorithm to obtain a time-space evolution feature and a diffusion trend prediction, the time-space evolution analysis algorithm combining a fluid mechanics model and a pipeline topology structure of the gas pipeline to simulate the propagation process of the leaked gas in the gas pipeline; based on the preliminary leakage probability distribution, the classification confidence, the time-space evolution feature and the diffusion trend prediction, generating the leakage detection result.

[0014] Further, the application also proposes to generate the leakage detection result based on the preliminary leakage probability distribution, the classification confidence, the time-space evolution feature and the diffusion trend prediction, including: performing confidence weighted fusion processing on the preliminary leakage probability distribution and the classification confidence to obtain a weighted leakage probability distribution and a fusion confidence index, the confidence weighted fusion processing dynamically adjusting the weight distribution of the regional leakage probability according to the classification confidence; based on the weighted leakage probability distribution and the time-space evolution feature, performing leakage risk quantification through a risk level evaluation algorithm to obtain a risk level distribution and an influence range evaluation; based on the risk level distribution, the influence range evaluation and the diffusion trend prediction, generating the leakage detection result.

[0015] In one aspect, a gas pipeline leakage rapid detection system is provided, the system comprising: an acquisition module configured to collect multi-modal monitoring data through a multi-source sensing device arranged at a key monitoring point of a gas pipeline, the multi-source sensing system comprising a distributed fiber sensor array, a laser methane detection unit, a sound wave sensor group, and an environmental parameter monitoring module; a first determination module configured to determine a data confidence weight matrix based on performance parameters of each sensing device in the multi-source sensing device and environmental influence factors, the data confidence weight matrix being used to represent a contribution degree of each modal monitoring data in the multi-modal monitoring data in the current leakage detection task; a second determination module configured to determine a multi-modal monitoring data comprehensive feature and a modal coordination degree parameter based on the multi-modal monitoring data and the data confidence weight matrix, the comprehensive feature being a high-order feature representation after fusion of the multi-modal monitoring data, and the modal coordination degree parameter being used to quantify a consistency degree of detection results based on different modal monitoring data; a third determination module configured to determine a leakage detection result of the gas pipeline based on the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter, the leakage detection result comprising a leakage probability distribution and a spatio-temporal evolution feature, the leakage probability distribution being used to describe a possibility of leakage at each position of the gas pipeline, and the spatio-temporal evolution feature being used to represent a propagation and change rule of a leakage point in a spatio-temporal dimension after leakage at each position.

[0016] In one aspect, a computer device is provided, the computer device comprising one or more processors and one or more memories, the one or more memories having stored therein at least one computer program, the computer program being loaded and executed by the one or more processors to implement the gas pipeline leakage rapid detection method.

[0017] In one aspect, a computer readable storage medium is provided, the computer readable storage medium having stored therein at least one computer program, the computer program being loaded and executed by a processor to implement the gas pipeline leakage rapid detection method.

[0018] In one aspect, a computer program product or computer program is provided, the computer program product or computer program comprising program code stored in a computer readable storage medium, the program code being read by a processor of a computer device from the computer readable storage medium, the processor executing the program code to cause the computer device to perform the above-mentioned gas pipeline leakage rapid detection method.

[0019] From the above, the application provides a kind of gas pipeline leak rapid detection method and system, by multi-source sensing device collection multi-modal monitoring data and constructs data confidence weight matrix, combined with cross-modal feature fusion and space-time evolution analysis, effectively solve the technical problems of traditional detection method to a certain extent, environmental interference is large, multi-modal data fusion is insufficient, with high detection precision, strong dynamic adaptability, leakage trend prediction accurate advantage. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0021] Figure 1 is a schematic diagram of an implementation environment of a gas pipeline leak rapid detection method provided by the embodiments of the present application; Figure 2 is a flowchart of a gas pipeline leak rapid detection method provided by the embodiments of the present application; Figure 3 is a partial flowchart of a gas pipeline leak rapid detection method provided by the embodiments of the present application; Figure 4 is a partial flowchart of another gas pipeline leak rapid detection method provided by the embodiments of the present application; Figure 5 is a partial flowchart of another gas pipeline leak rapid detection method provided by the embodiments of the present application; Figure 6 is a structural schematic diagram of a gas pipeline leak rapid detection system provided by the embodiments of the present application; Figure 7 is a structural schematic diagram of a server provided by the embodiments of the present application. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present application more clear, the embodiments of the present application will be further described in detail below.

[0023] In the present application, the terms "first", "second", etc. are used to distinguish the same or similar items with basically the same function, and it should be understood that there is no logical or time sequence dependence between "first", "second", "nth", and the number and execution order are not limited.

[0024] Gas pipeline: A closed pressure pipeline system for transporting flammable gases such as natural gas, liquefied petroleum gas, etc., including pipe body, valves, connectors and auxiliary facilities, is an important infrastructure for energy transportation.

[0025] Optical fiber sensor array: A monitoring system composed of multiple optical fiber sensing nodes in a specific topology, which detects changes in characteristics such as light intensity, phase, wavelength, etc. caused by external vibration, temperature and other parameter changes when light signals are transmitted in optical fibers to perceive pipeline status.

[0026] Laser methane detection unit: A detection device based on tunable diode laser absorption spectroscopy (TDLAS) technology, which emits a specific wavelength of laser light and detects the absorption intensity of methane gas to achieve high-precision and high-sensitivity measurement of methane concentration.

[0027] Acoustic sensor group: A collection of multiple acoustic pressure sensors or accelerometers used to collect sound waves or vibration signals generated by pipeline leaks, with a working frequency band usually covering infrasound to ultrasonic waves (0.01Hz~20kHz).

[0028] Environmental parameter monitoring module: A functional unit integrating temperature, humidity, air pressure, wind speed and other environmental sensors to monitor real-time environmental conditions around the pipeline, providing reference for data correction and interference elimination.

[0029] Cross-modal feature extraction: Extracting feature information with relevance and complementarity from different sensing modalities (such as sound, light, electricity, chemistry, etc.) to establish a feature mapping relationship between multiple sources of data.

[0030] Feature enhancement: Improve feature quality through signal processing algorithms (such as filtering, amplification, transformation, etc.) to improve signal-to-noise ratio and distinguishability, and enhance the expressiveness of useful information.

[0031] Time domain feature analysis: Statistical analysis of signals in the time dimension to extract mean, variance, peak value, waveform factor and other characteristic parameters representing the time variation of signals.

[0032] Frequency domain feature conversion: Convert signals from time domain to frequency domain through Fourier transform, wavelet transform and other methods to obtain frequency distribution features such as spectral energy, dominant frequency and band power.

[0033] Feature dimension reduction: Use principal component analysis (PCA), linear discriminant analysis (LDA) and other methods to reduce the number of features, eliminate redundant information and retain the most discriminant feature components.

[0034] Feature selection: Select the most relevant and most contributory feature subset from the original feature set to improve model efficiency and generalization ability.

[0035] Feature Optimization: Improve feature representation through methods such as feature transformation, combination, reconstruction, etc., to make features more suitable for subsequent processing and analysis.

[0036] Redundant Feature Elimination: Identify and remove features with high correlation and overlapping information in the feature set, reduce feature dimensionality, and improve processing efficiency.

[0037] Environmental Fitness Analysis: Evaluate the ability of the sensor system to maintain performance stability under different environmental conditions (temperature, humidity, air pressure, etc.), and quantify the degree of influence of environmental factors on detection effectiveness.

[0038] Dynamic Weight Fusion: Adjust the weight coefficients of each sensor data in the fusion process according to the real-time performance of the sensor and environmental changes, to achieve optimal fusion effect.

[0039] Stability Evaluation: Quantify the fluctuation degree of the system or parameters in time series through statistical analysis (such as variance calculation, coefficient of variation analysis, etc.), to evaluate its reliability.

[0040] Multi-dimensional Quantitative Evaluation: Establish a comprehensive index system from multiple evaluation dimensions (such as accuracy, stability, response speed, etc.) to comprehensively quantify and evaluate system performance.

[0041] Interference Degree Analysis: Quantitatively evaluate the influence intensity of environmental interference factors (such as wind and rain, mechanical vibration, electromagnetic interference, etc.) on the detection system, to provide basis for interference suppression.

[0042] Sliding Window Statistical Method: Use a fixed length time window to slide along the time axis, and perform real-time statistical analysis on the data within the window, to realize dynamic data processing.

[0043] Spacetime Evolution Analysis Algorithm: Combine time series analysis and spatial distribution characteristics to simulate and predict the dynamic change process of physical phenomena (such as gas leakage and diffusion) in time and space dimensions.

[0044] Artificial Intelligence (AI) is the use of digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain better results.

[0045] Machine Learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge sub-models to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence.

[0046] Normalization: mapping a number series with different value ranges to the interval (0, 1) for easy data processing. In some cases, the normalized value can be directly implemented as a probability.

[0047] Embedded coding (Embedded Coding): Embedded coding represents a corresponding relationship in mathematics, that is, mapping data on X space to Y space through a function F, where the function F is a single function, and the mapping result is structure preserving. The single function means that the mapped data is uniquely corresponding to the pre-mapped data, and the structure preserving means that the size relationship of the pre-mapped data is the same as that of the post-mapped data, for example, there are data X1 and X2 before mapping, and Y1 corresponding to X1 and Y2 corresponding to X2 after mapping. If the data X1 > X2 before mapping, then the data Y1 > Y2 after mapping. For words, it is to map words to another space for subsequent machine learning and processing.

[0048] Attention weight: can represent the importance of certain data in the training or prediction process. Importance represents the size of the influence of input data on output data. The data with high importance has a higher value of the corresponding attention weight, and the data with low importance has a lower value of the corresponding attention weight. In different scenarios, the importance of data is not the same, and the process of training attention weight of the model is also the process of determining the importance of data.

[0049] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data for analysis, stored data, displayed data, etc.) and signals involved in the present application are authorized by the user or fully authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0050] Figure 1 is an implementation environment schematic diagram of a gas pipeline leakage rapid detection method provided by an embodiment of the present application, referring to Figure 1 The implementation environment can include a node 110 and a server 140.

[0051] The node 110 is connected to the server 140 through a wireless network or a wired network. Optionally, the node 110 is a smart phone, a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto. The node 110 is installed and runs an application program supporting rapid detection of gas pipeline leakage.

[0052] The server 140 is a stand-alone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content delivery network (CDN), and big data and artificial intelligence platform. The server 140 can provide background services for the application program running on the node 110.

[0053] In the related art, gas pipeline leakage detection mainly relies on single sensing technologies such as acoustic detection, infrared thermal imaging, or laser methane remote sensing, which is difficult to meet the multi-dimensional detection needs in complex environments. The traditional multi-sensor fusion method adopts a fixed weight or a simple weighted average strategy, which cannot adapt to the dynamic changes of sensor performance and the influence of environmental interference factors, resulting in reduced detection reliability. The related technology lacks sufficient collaborative use of multi-modal data and lacks an effective cross-modal feature fusion mechanism, making it difficult to accurately identify leakage characteristics and predict diffusion trends.

[0054] To solve the above problems, the inventors found that the existing methods have obvious defects in sensor weight distribution and feature fusion. Through analysis, it is found that the reliability of different sensors in complex environments is significantly different, and the fixed weight strategy cannot reflect the real-time working condition changes. Further research found that the correlation between multi-modal data has not been fully explored, resulting in insufficient consistency of detection results. Based on this, the inventors propose to establish a dynamic weight distribution mechanism to optimize sensor contribution, construct a cross-modal feature fusion framework to improve data collaboration, and introduce a spatiotemporal evolution model to enhance the leakage trend prediction capability.

[0055] Therefore, the present application proposes a rapid detection method for gas pipeline leakage, as shown in Figure 2 Taking the server as an execution subject, the method includes the following steps: 201. Collecting multi-modal monitoring data through a multi-source sensing device arranged at a key monitoring point of a gas pipeline, the multi-source sensing system including a distributed fiber sensor array, a laser methane detection unit, a sound wave sensor group, and an environmental parameter monitoring module; 202. Determining a data confidence weight matrix based on performance parameters of each sensing device in the multi-source sensing device and environmental influence factors; 203. determine a multi-modal monitoring data comprehensive feature and a modality coordination degree parameter based on the multi-modal monitoring data and the data confidence weight matrix; 204. determine a gas pipeline leakage detection result based on the multi-modal monitoring data comprehensive feature and the modality coordination degree parameter, the leakage detection result including a leakage probability distribution and a spatio-temporal evolution feature.

[0056] Wherein, the data confidence weight matrix is used to represent the contribution degree of each modality monitoring data in the multi-modal monitoring data in this leakage detection task, the comprehensive feature is a high-order feature representation after fusion of the multi-modal monitoring data, the modality coordination degree parameter is used to quantify the consistency degree of the detection results obtained based on different modality monitoring data, the leakage probability distribution is used to describe the possibility of leakage at each position of the gas pipeline, and the spatio-temporal evolution feature is used to represent the propagation and change rule of the leakage point in the spatio-temporal dimension after leakage at each position. The multi-source sensing device refers to a composite detection equipment integrating multiple sensing technologies, which can specifically monitor the strain distribution of the pipeline by using a distributed optical fiber sensor, capture the gas concentration change by using a laser methane detection unit, collect the leakage soundprint signal by using a sound wave sensor group, and obtain the working condition data such as temperature and humidity by using an environmental parameter monitoring module. The data confidence weight matrix refers to a dynamic parameter matrix reflecting the real-time reliability of each sensor, which can specifically be generated by using a multi-objective optimization algorithm based on the historical performance data of the sensor and the current environmental interference coefficient. The comprehensive feature refers to a high-order feature vector after fusion of the multi-modal data, which can specifically extract the associated features of the acoustic signal and the gas concentration data by using a cross-modal attention mechanism. The modality coordination degree parameter refers to an evaluation index quantifying the consistency of the multi-source data, which can specifically be realized by using a feature space cosine similarity calculation method. The leakage probability distribution refers to a possibility heat map of leakage at each position of the pipeline, which can specifically be generated by using a deep learning classification model combined with Bayesian inference. The spatio-temporal evolution feature refers to a dynamic simulation result of the leakage diffusion process, which can specifically be realized by using a computational fluid dynamics model combined with pipeline topology structure analysis.

[0057] Specifically, the multi-source sensing device synchronously collects multi-dimensional data such as sound waves, gas concentration, and environmental parameters at key nodes of the pipeline. The data confidence weight matrix dynamically adjusts the weights of each modality data according to the real-time performance of the sensor and the degree of environmental interference, for example, reducing the confidence weight of the sound wave sensor in heavy rain. The cross-modal feature fusion process extracts waveform features and spectral features through time-frequency domain analysis, and performs feature enhancement combined with the weight matrix. The modality coordination degree parameter calculates the consistency degree of the detection results of different sensors, and reduces the confidence of the region when the sound wave signal and the gas concentration data appear contradictory. The leakage probability distribution is generated by fusing the weighted multi-modal features, and the spatio-temporal evolution feature simulates the leakage gas diffusion path combined with the fluid mechanics model. The whole process forms a closed-loop detection system from data acquisition, dynamic weighting, feature fusion to trend prediction.

[0058] Compared with the related art, the traditional method adopts a fixed weight distribution strategy and is difficult to adapt to sensor performance fluctuations, while the present scheme dynamically adjusts the weight matrix through environmental parameters to improve detection stability under complex working conditions. The related art only performs simple superposition at the data level, and the present scheme mines the deep correlation between sound signals and gas concentration through cross-modal feature fusion to enhance the leakage feature representation capability. The conventional method is limited to leakage point positioning, and the present scheme generates spatiotemporal evolution features in combination with a fluid mechanics model to realize dynamic prediction of the leakage diffusion trend.

[0059] Through the above technical solutions, the present application to some extent overcomes the problem of sensor reliability decline caused by environmental interference, and ensures the optimal use of multi-modal data through dynamic weight distribution. The cross-modal feature fusion mechanism significantly improves the feature recognition accuracy in complex leakage scenarios, and the modality coordination degree parameter effectively excludes abnormal data interference. The introduction of spatiotemporal evolution features provides diffusion path prediction for emergency disposal, forming a full-process solution covering detection, positioning and prediction.

[0060] The present application further proposes the following technical solutions, see Figure 3 , comprising the following steps.

[0061] 301. Cross-modal feature extraction and enhancement based on multi-modal monitoring data and data confidence weight matrix, to determine a cross-modal feature set and a feature saliency vector; 302. Determining a target feature representation and a modality correlation measure based on the cross-modal feature set and the feature saliency vector; 303. Determining a multi-modal monitoring data comprehensive feature and a modality coordination degree parameter based on the target feature representation and the modality correlation measure.

[0062] The cross-modal feature set is an enhanced feature obtained by fusing monitoring data of different modalities, and the feature saliency vector is used to quantify the contribution of each feature dimension to the leakage detection task. The cross-modal feature set refers to an enhanced feature obtained by fusing monitoring data of different modalities, which can be implemented by combining time domain feature analysis and frequency domain feature conversion with feature enhancement algorithms to eliminate noise interference in single modality data and enhance the expression ability of effective information. The feature saliency vector is a parameter that quantifies the contribution of each feature dimension to the leakage detection task, which can be generated by combining a feature weight optimization algorithm with a contribution evaluation model, and is used to dynamically adjust the weight distribution of key features under different environmental conditions. The target feature representation is the feature expression after coupling the feature saliency vector and the cross-modal feature set, which can be generated by combining feature dimension reduction algorithms and feature selection strategies, and is used to retain high-contribution features and eliminate redundant information. The modality correlation measure is an index for evaluating the coupling strength of different modal monitoring data at the feature level, which can be calculated by combining covariance analysis with feature distribution consistency algorithms, and is used to quantify the cooperative detection capability of multi-source data.

[0063] Specifically, first, the multi-modal monitoring data is analyzed in time-frequency domain, time domain statistical features and frequency domain energy distribution features are extracted, feature weight distribution is optimized by combining feature enhancement algorithm with data confidence weight matrix, and enhanced cross-modal feature set is generated. Subsequently, a contribution evaluation model is used to dynamically allocate weights to feature dimensions, and a feature saliency vector is generated to reflect the importance difference of different features in leakage detection. Further, high-dimensional features are compressed by a feature dimension reduction algorithm, and key features are selected to form target feature representation by combining feature retention indicators, and the correlation strength between different modal features is analyzed by using a covariance matrix to generate a modal correlation measurement parameter. Finally, multi-dimensional feature aggregation is performed based on the optimized feature representation and modal correlation parameter to generate a comprehensive feature and calculate a modal synergy parameter, providing a high-consistency feature input for subsequent leakage detection.

[0064] Compared with related technologies, the traditional method uses fixed weights to superimpose multi-sensor data, which cannot adapt to the feature effectiveness changes caused by sensor performance fluctuations and environmental interference. The scheme realizes adaptive fusion of multi-modal data through dynamic feature enhancement and weight optimization mechanism, effectively solving the problem of insufficient feature extraction under environmental interference to a certain extent. At the same time, the introduction of the modal correlation measurement parameter overcomes the defect of ignoring the multi-source data collaborative detection capability of the traditional method, and significantly improves the reliability of feature expression under complex working conditions.

[0065] Through the above technical solutions, the feature redundancy problem caused by insufficient multi-modal data fusion is effectively solved to a certain extent, and the dynamic weight distribution mechanism is adapted to the performance difference of different sensors, improving the pertinence of feature extraction. Cross-modal feature enhancement processing enhances the effective signal recognition ability in a noisy environment, and the modal correlation measurement parameter provides a quantitative evaluation basis for multi-source data collaborative detection, thereby significantly improving the accuracy of leakage detection results and the system anti-interference ability.

[0066] The application further proposes to perform time domain feature analysis and frequency domain feature conversion on the multi-modal monitoring data to obtain a time domain feature set and a frequency domain feature set, the time domain feature set containing statistical features and waveform features of the monitoring data, and the frequency domain feature set containing frequency spectrum features and energy distribution features of the monitoring data; feature enhancement and weight optimization are performed on the time domain feature set, the frequency domain feature set and the data confidence weight matrix to obtain enhanced time-frequency domain features and feature weight distribution, the feature weight distribution reflecting the importance of each feature dimension; cross-modal feature fusion and contribution evaluation are performed on the enhanced time-frequency domain features and feature weight distribution to obtain a cross-modal feature set and a feature saliency vector.

[0067] The time domain feature analysis refers to extracting statistical features and waveform features of a signal from the time dimension, and can be specifically implemented by mean, variance, and peak factor calculation, and is used to capture the change law of the monitoring data in the time sequence. The frequency domain feature conversion refers to converting the time domain signal into a frequency domain representation, and can be specifically implemented by fast Fourier transform or wavelet transform, and is used to reveal the energy distribution characteristics of different frequency components in the signal. The feature enhancement and weight optimization refers to dynamically adjusting the time-frequency domain features according to the data confidence weight matrix, and can be specifically implemented by weighted average or adaptive filtering algorithm, and is used to eliminate the interference of environmental noise on feature extraction. The cross-modal feature fusion and contribution evaluation refers to jointly analyzing the features of different modalities, and can be specifically implemented by principal component analysis or a neural network model, and is used to quantify the contribution of each feature dimension to the leakage detection task.

[0068] Specifically, the multi-modal monitoring data is first decomposed into a time domain feature set and a frequency domain feature set. The time domain feature set captures the instantaneous change characteristics of the pipeline pressure or vibration signal by calculating statistical parameters and waveform parameters of the signal, such as root mean square value and kurtosis coefficient. The frequency domain feature set identifies specific frequency components related to leakage by extracting energy concentrated frequency bands and spectral peaks through spectral analysis. Subsequently, the data confidence weight matrix is applied to the weighted fusion of the time-frequency domain features, for example, features corresponding to high-confidence sensor data are given higher weights, thereby suppressing abnormal features caused by environmental interference. Finally, the optimized time-frequency domain features are fused across modalities through a feature space mapping method to generate a cross-modal feature set containing multi-dimensional information, and the saliency scores of each feature are calculated through a contribution evaluation algorithm.

[0069] Compared with related technologies, the traditional method usually only uses single domain feature analysis and the weight distribution is fixed, which cannot adapt to feature fluctuations in complex environments. The present scheme realizes complementary feature extraction through joint analysis of time and frequency domains, effectively suppresses environmental interference through a dynamic weight optimization mechanism, and improves the integrity of feature representation through a cross-modal fusion algorithm. For example, static weight distribution in related technologies will cause feature distortion when sensor performance fluctuates, while the present scheme can real-time correct the deviation in the feature extraction process through dynamic adjustment of the confidence weight matrix.

[0070] Through the above technical solutions, the present application solves the problem of insufficient multi-modal monitoring data feature extraction to some extent, obtains a complementary feature set through joint analysis of time and frequency domains, eliminates the influence of environmental interference on feature quality through dynamic weight optimization, and constructs a cross-modal fusion mechanism to realize collaborative use of multi-source data. The present scheme enhances the robustness of feature representation and improves the adaptability of the leakage detection model to complex working conditions, providing high-discrimination feature input for subsequent leakage probability calculation.

[0071] The application further proposes feature dimension reduction and feature selection on the cross-modal feature set and the feature saliency vector to obtain a reduced feature representation and a feature retention index. The feature dimension reduction and feature selection are used to retain main feature information, and the feature retention index is used to represent the information loss degree in the dimension reduction process. The reduced feature representation and the feature retention index are subjected to feature fusion and feature optimization to determine a preliminary fusion feature and a feature quality parameter. The feature quality parameter is used to represent the reliability and stability of the preliminary fusion feature. Based on the preliminary fusion feature and the feature quality parameter, a target feature representation and a modal correlation measure are determined.

[0072] wherein the feature dimension reduction refers to a process of mapping high-dimensional features to a low-dimensional space through mathematical transformation, which can be specifically implemented by principal component analysis or t-SNE algorithm, and is used to eliminate data redundancy and improve computational efficiency. The feature retention index refers to a quantitative parameter of information retention degree before and after dimension reduction, which can be specifically implemented by calculating feature variance retention rate or reconstruction error, and is used to dynamically adjust the dimension reduction threshold to avoid loss of key information. The feature quality parameter refers to an evaluation index of stability and consistency of the fusion feature, which can be specifically implemented by covariance matrix analysis or time series stability test, and is used to select effective features with strong anti-interference ability. The modal correlation measure refers to an evaluation value of coupling strength of different modal data at the feature level, which can be specifically implemented by calculating cross-modal feature correlation coefficient or mutual information entropy, and is used to guide the weight allocation of multi-modal data.

[0073] Specifically, the cross-modal feature set is first subjected to dimension reduction processing by principal component analysis, and the variance retention rate after dimension reduction is calculated as the feature retention index. If the variance retention rate is lower than a preset threshold, the t-SNE algorithm is automatically switched for nonlinear dimension reduction to ensure that key information is not lost. The reduced features and the retention index are jointly input into a feature fusion module to generate a preliminary fusion feature by weighted fusion, and a stability index of the inter-feature covariance matrix is calculated as a feature quality parameter. When the feature quality parameter is lower than a set standard, a feature optimization mechanism is triggered to reselect a feature subset. Finally, based on the optimized feature set, the correlation strength between different modal features is calculated by mutual information entropy to generate a modal correlation measure value.

[0074] Compared with related technologies, the traditional method uses a fixed threshold for feature dimension reduction, which is easy to cause loss of effective information and cannot quantify the loss degree. The feature fusion process in related technologies lacks a quality evaluation mechanism, making it difficult to identify feature degradation problems caused by noise interference. Related technologies usually ignore the dynamic correlation between different modal features, resulting in low multi-modal data collaboration efficiency. The present scheme realizes adaptive control of the dimension reduction process through the feature retention index, constructs a closed-loop optimization mechanism in combination with the feature quality parameter, and introduces the modal correlation measure to guide the dynamic proportioning of multi-modal data.

[0075] By the technical solution, the application effectively reduces the calculation complexity in multi-modal data fusion, and to some extent, solves the problem of low detection efficiency caused by high-dimensional features. By quantifying the feature retention degree and fusion quality, the stability and anti-interference ability of the detection result are significantly improved. The dynamic adjustment mechanism based on modal correlation measurement enhances the cooperative detection performance between different sensing modalities, so that the system can still maintain high-precision detection under complex environmental conditions.

[0076] The application further proposes a method for determining multi-modal monitoring data comprehensive features and modal synergy degree parameters based on target feature representation and modal correlation measurement, including feature optimization and redundant feature elimination of target feature representation and modal correlation to obtain optimized feature representation and feature distribution consistency index, feature standardization and distribution regularization of the optimized feature representation and feature distribution consistency index to obtain standardized feature representation and feature stability parameter, multi-dimensional feature aggregation and synergy degree calculation of the standardized feature representation and feature stability parameter to obtain multi-modal monitoring data comprehensive features and modal synergy degree parameters.

[0077] Among them, feature optimization and redundant feature elimination refer to filtering and reconstructing the feature space by algorithm to remove redundant or low-contribution features, which can be realized by principal component analysis combined with recursive feature elimination method, and is used to reduce the data dimension and retain key information. The feature distribution consistency index refers to a numerical parameter that quantifies the distribution difference between different sensor modal features, which can be calculated by KL divergence algorithm to calculate the probability distribution difference between feature vectors, and is used to evaluate the matching degree of cross-modal feature fusion. Standardization refers to converting feature data of different dimensions to a unified standard scale, which can be realized by Z-score standardization method, and is used to eliminate the influence of dimension difference on the fusion result. The feature stability parameter refers to an index that measures the fluctuation degree of features over time, which can be calculated by sliding window statistical method to calculate the variance coefficient of feature value, and is used to identify feature drift caused by environmental interference. Multi-dimensional feature aggregation refers to integrating features according to spatial distribution, time sequence and modal type, which can realize multi-dimensional data fusion by tensor decomposition method, and is used to build a comprehensive feature expression system. Synergy degree calculation refers to an evaluation parameter that quantifies the complementary effect between different modal features, which can be calculated by mutual information entropy combined with Pearson correlation coefficient to calculate the correlation strength between features, and is used to guide the optimization combination of multi-modal data.

[0078] Specifically, in the gas pipeline leakage detection process, first, the feature selection algorithm is used to screen out the feature subset with strong correlation with leakage from the target feature representation, and the automatic encoder is used to perform nonlinear reconstruction on the feature space to eliminate redundant features and generate an optimized feature representation. At this time, by calculating the probability distribution difference between different modal features, a feature distribution consistency index is generated to evaluate the effectiveness of cross-modal data fusion. Then, the optimized features are standardized to convert the heterogeneous data such as pressure fluctuations and sound wave frequencies collected by different sensors into a unified dimension, and the stability parameters of the feature values are calculated through time series analysis to identify the sensor drift phenomenon caused by temperature changes. Finally, a three-dimensional tensor model is used to aggregate the standardized spatial distribution features, time evolution features and modal type features, analyze the complementary relationship between different modal features through mutual information entropy, generate a comprehensive feature representation that can fully represent the leakage state, and quantify the optimization degree of multi-modal data fusion through the synergy parameter.

[0079] Compared with related technologies, the traditional method uses a fixed threshold for feature selection, which results in insufficient adaptability to dynamic environments, and does not consider the distribution difference between different modal features. The present scheme automatically removes redundant features through a dynamic feature optimization mechanism, and combines a distribution consistency index to evaluate the matching degree of cross-modal data in real time, effectively solving the interference problem caused by feature space overlap. The simple normalization processing in related technologies cannot eliminate the dimension difference of heterogeneous data. The present scheme uses a dual mechanism of standardization and stability evaluation to identify feature drift while unifying the dimension, ensuring the reliability of the fused data. The existing method uses linear weighted fusion of multi-modal data, which fails to mine the nonlinear correlation between features. The present scheme realizes deep coupling and collaborative optimization of cross-modal features through multi-dimensional tensor aggregation combined with mutual information analysis.

[0080] Through the above technical solutions, the present application effectively solves the feature redundancy problem in multi-modal data fusion to a certain extent, reduces invalid information interference through dynamic feature selection and reconstruction, significantly improves the consistency level between different modal features, enhances data comparability through distribution difference quantification and standardization processing, establishes a multi-dimensional feature collaboration mechanism, and realizes deep complementarity of cross-modal features through tensor aggregation and mutual information analysis, thereby improving the accuracy and false alarm suppression ability of leakage detection in complex environmental conditions.

[0081] Referring to Figure 4 The present application further proposes a method for determining a data confidence weight matrix based on the performance parameters of each sensor device in a multi-source sensing device and environmental influence factors, including the following steps.

[0082] 401. Perform sensor performance and environmental adaptability analysis on the performance parameters and environmental influence factors to obtain an initial weight distribution and an environmental adaptability coefficient; 402. Dynamic weight fusion and stability evaluation are performed on the initial weight distribution and environmental fitness coefficient to obtain an optimized weight distribution and a weight stability index; 403. Online calibration and optimization are performed on the optimized weight distribution and the weight stability index to obtain a data confidence weight matrix.

[0083] The environmental fitness coefficient is used to quantify the performance retention capability of each sensor in the multi-source sensing device under the current comprehensive environmental conditions, the optimized weight distribution is used to reflect the optimal weight configuration under multi-objective optimization, the weight stability index is used to represent the fluctuation degree and reliability level of the weight distribution in the time series, the performance parameter refers to a quantitative index reflecting the measurement capability of the sensor, which can be specifically implemented by using measurement accuracy, response time and stability index, and a performance score vector is generated by multi-dimensional quantitative evaluation for objectively measuring the inherent performance difference of different sensors. The environmental influence factor refers to the interference factor of external conditions on the working state of the sensor, which can be specifically implemented by using temperature, humidity, wind speed and precipitation intensity parameters, and an environmental interference coefficient is generated by interference degree analysis for quantifying the influence of dynamic environmental changes on the effectiveness of the sensor. The initial weight distribution refers to the benchmark weight configuration based on the inherent performance of the sensor, which can be specifically implemented by using the analytic hierarchy process combined with the entropy weight method, and is used to establish the quantitative relationship between the sensor performance and the weight distribution. The environmental fitness coefficient refers to the performance retention capability of the sensor under a specific environment, which can be specifically implemented by calculating the matching degree of the environmental interference coefficient and the anti-interference capability of the sensor, and is used to dynamically correct the weight of the sensor. Dynamic weight fusion refers to the optimization process of comprehensive performance and environmental factors, which can be specifically implemented by using a multi-objective optimization algorithm combined with time series analysis, and is used to generate a weight configuration that meets the performance optimization and stability requirements at the same time. Online calibration and optimization refer to the process of adjusting the weight parameters in real time, which can be specifically implemented by using sliding window statistics combined with incremental learning algorithm, and is used to ensure that the weight matrix adapts to real-time monitoring requirements.

[0084] Specifically, in the gas pipeline leakage detection scenario, first, the performance parameters such as measurement accuracy and response speed of the distributed optical fiber sensor array, laser methane detection unit and other equipment are quantitatively evaluated, for example, the measurement accuracy is converted into a score value in the 0-1 interval through standardization to form a performance score vector. At the same time, environmental parameters such as temperature and humidity are collected, for example, when the wind speed is detected to exceed 5 meters per second, the environmental interference coefficient of the laser methane detection unit is calculated through the interference model. The performance score and the environmental interference coefficient are input into the weight distribution algorithm, for example, the fuzzy logic system is used to calculate the initial weight distribution and the environmental fitness coefficient. Then, the initial weight is dynamically adjusted using a multi-objective optimization algorithm, for example, the NSGA-II algorithm is used to minimize the weight fluctuation while ensuring detection accuracy, and the weight stability index is generated by analyzing the variation coefficient of historical weight data. Finally, based on the sliding window statistical monitoring of real-time data, for example, the sensor data deviation is statistically monitored using a 10-minute window, the weight parameters are dynamically adjusted through incremental learning, and the final data confidence weight matrix is generated after L2 regularization processing to eliminate abnormal fluctuations.

[0085] Compared with related technologies, the traditional method usually adopts fixed weight distribution or simple weighted average, for example, a fixed weight ratio of 0.6:0.4 is preset in the combined application of acoustic wave sensors and infrared sensors. This method cannot cope with sensor performance degradation problems, for example, when the response time of the acoustic wave sensor is prolonged by 30% due to aging, the fixed weight will cause the detection error to increase. However, the present scheme automatically reduces the weight distribution ratio of the sensor when its performance score decreases, for example, when the laser methane detection unit has a 20% measurement deviation in a heavy rain environment, the system dynamically adjusts its weight from 0.5 to 0.3, while increasing the weight of the distributed optical fiber sensor to 0.6, thereby maintaining the overall detection accuracy. In addition, related technologies lack stability evaluation mechanisms, and false judgments are prone to occur when the weight is frequently adjusted, while the present scheme calculates the weight variation coefficient through time series analysis, and when the weight of a certain sensor fluctuates by more than a threshold value, the weight freezing mechanism is triggered to avoid false adjustment.

[0086] Through the above technical solutions, the present application effectively solves the problem of detection error caused by unreasonable sensor weight distribution in complex environments. Through the cooperative analysis of performance parameters and environmental fitness, the dynamic matching of sensor weight and real-time working conditions is realized, for example, the weight of the ultrasonic sensor which is easily affected by humidity is automatically reduced in high temperature and high humidity environments. Through the dual constraints of multi-objective optimization and stability evaluation, the detection accuracy is improved while avoiding excessive weight fluctuations, for example, the weight adjustment frequency is controlled to not more than 3 times per minute. Through the online calibration mechanism, the weight matrix continuously adapts to the monitoring requirements, for example, in the event of a sudden sensor failure, the system can complete weight redistribution within 10 seconds, maintaining the continuous and reliable operation of the detection system.

[0087] The application further proposes to perform multi-dimensional quantitative evaluation on performance parameters to obtain a performance score vector, wherein the performance parameters include measurement accuracy, response time and stability indicators; to perform interference degree analysis on environmental impact factors to obtain environmental interference coefficients, wherein the environmental impact factors include temperature, humidity, wind speed and precipitation intensity; and to calculate initial weight distribution and environmental fitness coefficients based on the performance score vector and the environmental interference coefficients using a weight distribution algorithm.

[0088] The multi-dimensional quantitative evaluation refers to establishing an evaluation system including measurement accuracy error rate, response time standard deviation, and stability indicator fluctuation value, and specifically, an analytic hierarchy process can be used to construct a judgment matrix to realize quantitative scoring, which is used to comprehensively reflect the comprehensive performance level of the sensor under complex working conditions. The environmental interference coefficient refers to establishing a temperature sensor drift model, a humidity influence transfer function, a wind speed noise correlation matrix and a precipitation intensity attenuation curve, and specifically, a multiple regression analysis method can be used to calculate the interference degree of each environmental factor on the sensor data, which is used to quantify the influence effect of the external environment on the detection system. The weight distribution algorithm refers to nonlinearly coupling the performance score vector and the environmental interference coefficient, and specifically, an entropy weight method combined with fuzzy comprehensive evaluation can be used to realize dynamic weight calculation, which is used to generate an initial weight distribution that takes into account the sensor performance and environmental fitness.

[0089] Specifically, the measurement accuracy error rate is calculated by comparing the sensor output value with the standard calibration value, the response time standard deviation is obtained by statistically analyzing the dispersion degree of the sensor response delay time, and the stability indicator fluctuation value is determined by analyzing the variance of the long-term running data of the sensor. The temperature sensor drift model is constructed by experimentally calibrating the sensor zero drift at different temperatures, the humidity influence transfer function is derived by testing the sensor sensitivity change under different humidity conditions, the wind speed noise correlation matrix is established by measuring the sensor noise level under different wind speeds through wind tunnel experiment, and the precipitation intensity attenuation curve is drawn by simulating the sensor signal attenuation law under rainfall environment. The weight distribution algorithm inputs the performance score vector and the environmental interference coefficient into a multi-layer perceptron network, and outputs the initial weight distribution after nonlinear transformation, while calculating the fitness coefficients of each sensor under the current environmental conditions.

[0090] Compared with the related art, the conventional method only adopts fixed weights or weights allocated according to a single performance indicator, without considering environmental interference factors such as sensor drift caused by temperature changes and signal attenuation caused by humidity changes. The present scheme accurately quantifies the degree of influence of environmental factors on the sensor by establishing a temperature-drift model and a humidity-sensitivity function, and combines a multi-dimensional performance evaluation system to make the weight allocation reflect both the performance status of the sensor itself and the external environmental interference level. The weight allocation method based on simple weighted average in the related art cannot adapt to the dynamic attenuation of sensor performance, while the present scheme realizes continuous monitoring and evaluation of the real-time working state of the sensor by introducing the response time standard deviation and the stability indicator fluctuation value.

[0091] Through the above technical scheme, the present application effectively solves the problem of unreasonable weight allocation caused by sensor performance differences and environmental interference in gas pipeline leakage detection. By quantitatively evaluating the sensor measurement accuracy, response time and stability indicators, the real-time performance status of each sensor is accurately reflected; by analyzing the influence of temperature, humidity, wind speed and precipitation intensity on sensor data, the measurement error caused by environmental interference is eliminated; by using a dynamic weight allocation algorithm to integrate the performance score and the environmental interference coefficient, the data confidence is adaptively adjusted. The scheme can dynamically optimize the fusion weight of multi-source data according to the actual working state of the sensor and the change of environmental conditions, and improve the robustness and detection accuracy of the leakage detection system in complex environments.

[0092] The present application further proposes to use a multi-objective optimization method to fuse the initial weight distribution and the environmental fitness coefficient to obtain a preliminary optimized weight; to perform time series analysis on the preliminary optimized weight to obtain the weight fluctuation characteristics of the preliminary optimized weight; to generate a weight stability indicator based on the weight fluctuation characteristics using a stability evaluation algorithm; to generate an optimized weight distribution based on the preliminary optimized weight and the weight stability indicator using a constraint optimization method; to analyze multi-modal monitoring data using a sliding window statistical method based on the optimized weight distribution and the weight stability indicator to obtain a weight adjustment amount and a performance deviation indicator; to dynamically adjust the weight parameters based on the weight adjustment amount and the weight stability indicator through an incremental learning algorithm to obtain a calibrated weight distribution; and to perform regularization processing and stability verification on the calibrated weight distribution to obtain a data confidence weight matrix.

[0093] Among them, the multi-objective optimization method refers to an optimization algorithm that simultaneously considers sensor performance optimization and environmental adaptability balance. Specifically, it can be implemented by using the Pareto frontier solution or weighted summation method to avoid the weight distribution vulnerability caused by a single optimization target. Time series analysis refers to statistical modeling of the trend of weight parameters within a continuous time window. Specifically, it can be implemented by using an autoregressive model or moving average method to capture abnormal patterns in weight fluctuations. The stability evaluation algorithm refers to quantifying the reliability of the weight distribution through mathematical indicators. Specifically, it can be implemented by using variance calculation combined with coefficient of variation analysis to identify unstable factors in the weight parameters. The sliding window statistical method refers to local feature extraction based on fixed length data segments. Specifically, it can be implemented by using overlapping window division and statistical quantity calculation to monitor the impact of environmental changes on weights in real time. The incremental learning algorithm refers to a machine learning method that updates model parameters based on newly arrived data. Specifically, it can be implemented by using online gradient descent or incremental training of random forests to dynamically adjust the weight parameters. Regularization processing refers to mathematical constraints on weight parameters to prevent overfitting. Specifically, it can be implemented by using L2 regularization or weight truncation methods to ensure the reasonableness of the weight distribution.

[0094] Specifically, the initial weight distribution and environmental adaptability coefficient are fused by a multi-objective optimization method to generate a preliminary optimized weight that meets both sensor performance requirements and environmental conditions. This weight parameter is then input into the time series analysis module to identify its fluctuation characteristics over time through statistical modeling. Based on the calculated variance and coefficient of variation, the stability evaluation algorithm generates an index parameter that quantifies the stability of the weight. The constrained optimization method uses the stability index as a hard constraint condition to perform secondary optimization on the preliminary optimized weight, ensuring that the optimal weight distribution meets both accuracy requirements and anti-interference ability. In the online calibration stage, the sliding window statistical method analyzes the monitoring data stream in real time to detect weight deviations caused by environmental mutations or sensor abnormalities. The incremental learning algorithm dynamically adjusts the weight parameters based on the detected deviation, enabling the system to adapt to gradual or sudden changes in sensor performance. After eliminating the influence of extreme values through regularization processing, the calibrated weight distribution is evaluated again through the stability verification link to form a closed-loop optimization mechanism.

[0095] Compared with related technologies, the traditional method uses a fixed weight distribution mode, which cannot cope with dynamic environmental changes and sensor performance fluctuations. This scheme realizes the dynamic balance configuration of weight parameters through a double-layer optimization mechanism combining multi-objective optimization and constrained optimization. Related technologies lack effective online calibration methods, and this scheme introduces sliding window statistics and incremental learning algorithms to establish a real-time update mechanism for weight parameters. Compared with the simple weighted average method, this scheme effectively improves the reliability and anti-interference ability of the weight distribution through stability evaluation and regularization processing.

[0096] Through the technical solution, the application realizes dynamic optimization configuration of sensor weights in gas pipeline leakage detection, and to some extent, solves the problem of insufficient weight stability of traditional methods in complex environments. Through the online calibration mechanism, the weight deviation caused by environmental mutations is corrected in time, improving the adaptability of the detection system to sensor performance degradation. Combined with the double protection mechanism of regularization processing and stability verification, the weight parameter overfitting or extreme value interference is effectively prevented, ensuring the reliability of the detection result.

[0097] The application further proposes the following technical solution, see Figure 5 , comprising the following steps.

[0098] 501. Based on the multi-modal monitoring data comprehensive feature and the mode coordination degree parameter, the leakage state is recognized through a deep learning classification model to obtain a preliminary leakage probability distribution and a classification confidence; 502. Based on the preliminary leakage probability distribution and the classification confidence, a leakage diffusion simulation is performed through a spatio-temporal evolution analysis algorithm to obtain spatio-temporal evolution features and diffusion trend prediction. The spatio-temporal evolution analysis algorithm combines a fluid mechanics model and the pipe topology structure of the gas pipeline to simulate the propagation process of the leaked gas in the gas pipeline; 503. Based on the preliminary leakage probability distribution, the classification confidence, the spatio-temporal evolution features, and the diffusion trend prediction, a leakage detection result is generated.

[0099] The multi-modal monitoring data comprehensive feature refers to a high-order feature representation formed by fusing different sensor data through a cross-modal feature fusion algorithm. Specifically, a convolutional neural network combined with an attention mechanism can be used to enhance the expression ability of the leakage feature. The mode coordination degree parameter refers to a quantitative indicator generated by calculating the consistency degree of the detection results of different sensor data. Specifically, a combination of the Pearson correlation coefficient and the dynamic time warping algorithm can be used to constrain the conflicts between multi-source data. The deep learning classification model refers to a leakage state recognition model based on a multi-layer neural network. Specifically, a hybrid architecture of a residual network and a long short-term memory network can be used to extract leakage patterns from comprehensive features. The spatio-temporal evolution analysis algorithm refers to a dynamic simulation algorithm that combines fluid mechanics equations and pipe structure parameters. Specifically, a combination of the finite volume method and the graph neural network can be used to predict the propagation path of the leaked gas.

[0100] Specifically, the multi-modal monitoring data comprehensive feature integrates multi-source data such as distributed optical fiber sensors and laser methane detection units through a cross-modal feature fusion algorithm to form a high-order feature representation with spatiotemporal correlation. The modal coordination degree parameter filters out data sources with high reliability by calculating the dynamic consistency of detection results of different sensors. The deep learning classification model performs pattern recognition on the comprehensive features and outputs the leakage probability distribution and corresponding confidence of each pipeline location. The spatiotemporal evolution analysis algorithm simulates the gas diffusion process based on a fluid mechanics model, calculates the propagation path and speed of the leakage point in combination with the pipeline topology structure, and generates spatiotemporal evolution features and trend prediction results.

[0101] In some embodiments, the confidence weighted fusion processing can adopt an adaptive weighting algorithm to dynamically adjust the probability weights of different regions according to the classification confidence. The risk level evaluation algorithm can establish a risk evaluation matrix based on the leakage probability distribution and diffusion speed to quantify the risk level of different regions.

[0102] Compared with related technologies, the existing method relies on a single sensor or fixed weight fusion and cannot handle misjudgment problems caused by sensor performance fluctuations and environmental interference. The present solution dynamically evaluates data reliability through the modal coordination degree parameter and realizes dynamic propagation simulation of the leakage by combining a fluid mechanics model, which to some extent solves the defects of traditional methods that the detection results are static and cannot predict the diffusion trend.

[0103] Through the above technical solutions, the present application realizes dynamic fusion of multi-modal data and accurate prediction of leakage diffusion trend, effectively improving the reliability of gas pipeline leakage detection in complex environments. Through the combination of spatiotemporal evolution features and risk level evaluation, dynamic decision-making basis is provided for leakage emergency response, which to some extent solves the problem in related technologies that the detection results are disconnected from the physical propagation law.

[0104] The present application further proposes a technical solution for generating a leakage detection result based on a preliminary leakage probability distribution, classification confidence, spatiotemporal evolution features, and diffusion trend prediction, including confidence weighted fusion processing of the preliminary leakage probability distribution and classification confidence to obtain a weighted leakage probability distribution and a fusion confidence index, generating a risk level distribution and impact range evaluation based on the weighted results and spatiotemporal evolution features through a risk level evaluation algorithm, and finally generating a leakage detection result in combination with the risk level distribution, impact range evaluation, and diffusion trend prediction.

[0105] The confidence weighted fusion processing refers to a data fusion method of dynamically adjusting the leakage probability weight of each region according to the classification confidence, which can be specifically realized by adopting a weighted manner of multiplying the region confidence and the probability value, and by giving a higher weight to the leakage probability of a high-confidence region, the credibility of the detection result is enhanced. The risk level evaluation algorithm refers to a calculation method of quantifying risk by combining the leakage probability distribution and the space-time propagation law, which can be specifically realized by adopting the product of the leakage probability and the diffusion speed as the risk index, and by combining the static probability distribution and the dynamic propagation characteristics, the dynamic evaluation of the risk level is realized. The diffusion trend prediction refers to a technology of simulating the gas diffusion path based on a fluid mechanics model, which can be specifically realized by adopting a finite element method to simulate a three-dimensional flow field in combination with the pipeline topological structure, and by predicting the propagation path of the leaked gas in the pipeline, a basis is provided for the impact range evaluation.

[0106] Specifically, the scheme first optimizes the preliminary detection result through a dynamic weight distribution mechanism. In the confidence weighted fusion stage, the confidence of each region output by the classification model is converted into a weight coefficient, for example, when the classification confidence of a certain region reaches 0.9, the leakage probability thereof will be magnified to 1.5 times the original value, and the probability value of a region with a confidence lower than 0.6 is suppressed. This dynamic adjustment effectively avoids the misjudgment problem of low-confidence regions caused by fixed weights. Subsequently, the risk level evaluation algorithm combines the weighted probability distribution with the space-time evolution characteristics, for example, the diffusion speed parameter of the region upstream of the leakage point is input into the evaluation model, and a risk level change curve over time is calculated. Finally, the diffusion trend prediction module constructs a fluid mechanics model according to the pipeline pressure and the pipe diameter parameters, simulates the gas diffusion range in the next ten minutes, and generates a three-dimensional impact region thermodynamic map containing the time dimension.

[0107] Compared with the related art, in the traditional method, the fixed weight fusion is adopted, and the credibility difference of the detection results of different regions is not considered, for example, the same weight is still given to the sensor interference area under strong wind, resulting in an increased false positive rate. The present scheme automatically reduces the weight proportion of the low-confidence region when the sensor performance fluctuates through dynamic confidence adjustment. The existing risk evaluation method is mostly based on static probability distribution, for example, only the current leakage probability is used to divide the risk level, and the present scheme can predict the expansion range of the high-risk region in the next five minutes in combination with the gas diffusion speed parameter. In addition, the traditional diffusion simulation does not consider the influence of the pipeline branch structure, and the present scheme can simulate the propagation path of the gas in the complex pipe network by introducing the pipeline topological data.

[0108] All the optional technical solutions described above can be combined to form optional embodiments of the present application, which will not be described here.

[0109] By the technical solution, the application solves the problem of misjudgment of low reliability area caused by fixed weight fusion to some extent, and improves the reliability of the detection result through dynamic weight distribution. Meanwhile, the spatial and temporal dynamic evaluation of the risk level is realized, which can early warn the leakage diffusion trend and provide accurate time window prediction for emergency disposal. The finally generated detection result integrates three-dimensional information of instant state, spatial influence range and future trend, forming a complete leakage event description system.

[0110] Figure 6 is a structural schematic diagram of a gas pipeline leakage rapid detection system provided by an embodiment of the application, referring to Figure 6 , the system comprises: The acquisition module 601 is configured to collect multi-modal monitoring data through multi-source sensing devices arranged at key monitoring points of the gas pipeline. The multi-source sensing system comprises a distributed fiber sensor array, a laser methane detection unit, a sound wave sensor group and an environmental parameter monitoring module. The first determination module 602 is configured to determine a data confidence weight matrix based on performance parameters of each sensing device in the multi-source sensing device and environmental influence factors. The data confidence weight matrix is used to represent the contribution degree of each modal monitoring data in the multi-modal monitoring data in the current leakage detection task. The second determination module 603 is configured to determine a multi-modal monitoring data comprehensive feature and a modal coordination degree parameter based on the multi-modal monitoring data and the data confidence weight matrix. The comprehensive feature is a high-order feature representation after fusion of the multi-modal monitoring data, and the modal coordination degree parameter is used to quantify the consistency degree of the detection results obtained based on different modal monitoring data. The third determination module 604 is configured to determine a leakage detection result of the gas pipeline based on the multi-modal monitoring data comprehensive feature and the modal coordination degree parameter. The leakage detection result comprises a leakage probability distribution and a spatio-temporal evolution feature. The leakage probability distribution is used to describe the possibility of leakage at each position of the gas pipeline, and the spatio-temporal evolution feature is used to represent the propagation and change rule of the leakage point in the spatio-temporal dimension after leakage at each position.

[0111] It should be noted that the gas pipeline leakage rapid detection system provided in the above embodiments is only exemplified by the division of the above functional modules during leakage detection. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the functions described above. In addition, the gas pipeline leakage rapid detection system and the gas pipeline leakage rapid detection method provided in the above embodiments belong to the same concept, and the specific implementation process is described in the method embodiment, which will not be repeated here.

[0112] Figure 7Fig. 7 is a structural schematic diagram of a server provided by an embodiment of the present application. The server 700 can have great differences due to different configurations or performances, and can include one or more processors (Central Processing Units, CPUs) 701 and one or more memories 702. The one or more memories 702 store at least one computer program, which is loaded and executed by the one or more processors 701 to implement the method provided by the above-mentioned various method embodiments. Of course, the server 700 can also have a wired or wireless network interface, a keyboard, an input and output interface, and other components for implementing device functions, which are not described here.

[0113] In an example embodiment, a computer-readable storage medium, such as a memory including a computer program, is also provided. The computer program can be executed by a processor to complete the gas pipeline leakage rapid detection method in the above-mentioned embodiments. For example, the computer-readable storage medium can be a Read-Only Memory (ROM), a Random Access Memory (RAM), a Compact Disc Read-Only Memory (CD-ROM), a magnetic tape, a floppy disk, and an optical data storage device, etc.

[0114] In an example embodiment, a computer program product or computer program is also provided. The computer program product or computer program includes program code stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium, and executes the program code to make the computer device execute the above-mentioned gas pipeline leakage rapid detection method.

[0115] In some embodiments, the computer program related to the embodiments of the present application can be deployed to execute on one computer device, or on multiple computer devices located in one place, or on multiple computer devices distributed in multiple places and interconnected through a communication network, which can constitute a blockchain system.

[0116] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned embodiments can be completed by hardware, or by a program instructing relevant hardware, which can be stored in a computer-readable storage medium. The storage medium mentioned above can be a Read-Only Memory, a magnetic disk or an optical disk, etc.

[0117] The above merely is the optional embodiment of the present application, and does not use to limit the present application, any modification, equivalent replacement, improvement and so on, which are made in the spirit and principle of the present application, should be included in the protection scope of the present application.

Claims

1. A rapid detection method for gas pipeline leaks, characterized in that, The method includes: Multi-modal monitoring data is collected by multi-source sensing devices deployed at key monitoring points of gas pipelines. The multi-source sensing system includes a distributed fiber optic sensor array, a laser methane detection unit, an acoustic sensor group, and an environmental parameter monitoring module. Based on the performance parameters and environmental influencing factors of each sensor in the multi-source sensing device, a data confidence weight matrix is ​​determined. The data confidence weight matrix is ​​used to represent the contribution of each modal monitoring data in the multi-modal monitoring data to this leak detection task. Based on the multimodal monitoring data and the data confidence weight matrix, the comprehensive features and modal synergy parameters of the multimodal monitoring data are determined. The comprehensive features are high-order feature representations after the fusion of the multimodal monitoring data, and the modal synergy parameters are used to quantify the consistency of detection results obtained based on different modal monitoring data. Based on the comprehensive characteristics of the multimodal monitoring data and the modal synergy parameter, the leakage detection result of the gas pipeline is determined. The leakage detection result includes leakage probability distribution and spatiotemporal evolution characteristics. The leakage probability distribution is used to describe the probability of leakage at each location of the gas pipeline, and the spatiotemporal evolution characteristics are used to characterize the propagation and change law of the leakage point in the spatiotemporal dimension after leakage occurs at each location.

2. The method according to claim 1, characterized in that, The determination of the comprehensive characteristics and modal synergy parameters of the multimodal monitoring data based on the multimodal monitoring data and the data confidence weight matrix includes: Based on the multimodal monitoring data and the data confidence weight matrix, cross-modal feature extraction and enhancement are performed to determine the cross-modal feature set and feature saliency vector. The cross-modal feature set is the enhanced feature obtained by fusing monitoring data from different modalities, and the feature saliency vector is used to quantify the contribution of each feature dimension to the leak detection task. Based on the cross-modal feature set and the feature saliency vector, a target feature representation and a modal correlation metric are determined. The target feature representation is obtained by coupling the feature saliency vector with the cross-modal feature set, and the modal correlation metric is used to evaluate the coupling strength of monitoring data of different modalities at the feature level. Based on the target feature representation and the modal correlation metric, the comprehensive features of the multimodal monitoring data and the modal synergy parameter are determined.

3. The method according to claim 2, characterized in that, The step of performing cross-modal feature extraction and enhancement based on the multimodal monitoring data and the data confidence weight matrix, and determining the cross-modal feature set and feature saliency vector, includes: The multimodal monitoring data is subjected to time-domain feature analysis and frequency-domain feature transformation to obtain a time-domain feature set and a frequency-domain feature set. The time-domain feature set includes the statistical features and waveform features of the monitoring data, and the frequency-domain feature set includes the spectral features and energy distribution features of the monitoring data. The time-domain feature set, the frequency-domain feature set, and the data confidence weight matrix are subjected to feature enhancement and weight optimization to obtain enhanced time-frequency domain features and feature weight distribution, wherein the feature weight distribution reflects the importance of each feature dimension. Cross-modal feature fusion and contribution evaluation are performed on the enhanced time-frequency domain features and the feature weight distribution to obtain the cross-modal feature set and feature saliency vector.

4. The method according to claim 2, characterized in that, The step of determining the target feature representation and modality correlation metric based on the cross-modal feature set and the feature saliency vector includes: The cross-modal feature set and the feature saliency vector are subjected to feature dimensionality reduction and feature selection to obtain the dimensionality-reduced feature representation and feature retention index. The feature dimensionality reduction and feature selection are used to retain the main feature information, and the feature retention index is used to represent the degree of information loss during the dimensionality reduction process. The reduced-dimensional feature representation and feature retention index are subjected to feature fusion and feature optimization to determine preliminary fused features and feature quality parameters. The feature quality parameters are used to represent the reliability and stability of the preliminary fused features. Based on the preliminary fusion features and the feature quality parameters, the target feature representation and the modal correlation metric are determined.

5. The method according to claim 2, characterized in that, The determination of the comprehensive features of the multimodal monitoring data and the modal synergy parameter based on the target feature representation and the modal correlation metric includes: The target feature representation and the modal correlation degree are optimized and redundant features are eliminated to obtain the optimized feature representation and feature distribution consistency index. The feature optimization process is achieved through feature selection and reconstruction. The feature distribution consistency index is used to evaluate the degree of consistency of the optimized features among different modalities. The optimized feature representation and feature distribution consistency index are subjected to feature standardization and distribution normalization to obtain standardized feature representation and feature stability parameter. The standardization process makes features of different dimensions comparable, and the feature stability parameter is used to characterize the stability and reliability of the feature in the time dimension. Multi-dimensional feature aggregation and synergy calculation are performed on the standardized feature representation and feature stability parameters to obtain the comprehensive features of the multimodal monitoring data and the modal synergy parameters.

6. The method according to claim 1, characterized in that, The determination of the data confidence weight matrix based on the performance parameters of each sensor in the multi-source sensing device and environmental influencing factors includes: Sensor performance and environmental adaptability analysis is performed on the performance parameters and environmental influencing factors to obtain the initial weight distribution and environmental adaptability coefficient corresponding to the performance parameters. The environmental adaptability coefficient is used to quantify the ability of each sensor in the multi-source sensing device to maintain its working performance under the current comprehensive environmental conditions. Dynamic weight fusion and stability evaluation are performed on the initial weight distribution and the environmental fitness coefficient to obtain the optimized weight distribution and weight stability index corresponding to the performance parameters. The optimized weight distribution is used to reflect the optimal weight configuration under multi-objective optimization, and the weight stability index is used to characterize the degree of fluctuation and reliability of the weight distribution in the time series. The optimized weight distribution and weight stability index are calibrated and optimized online to obtain the data confidence weight matrix.

7. The method according to claim 6, characterized in that, The step of performing sensor performance and environmental adaptability analysis on the performance parameters and environmental influencing factors to obtain the initial weight distribution and environmental adaptability coefficients corresponding to the performance parameters includes: The performance parameters are evaluated in a multi-dimensional quantitative manner to obtain a performance score vector, wherein the performance parameters include measurement accuracy, response time and stability indicators; An environmental interference coefficient is obtained by analyzing the degree of interference of the environmental factors, which include temperature, humidity, wind speed and precipitation intensity. Based on the performance score vector and the environmental interference coefficient, the initial weight distribution and the environmental fitness coefficient are calculated using a weight allocation algorithm.

8. The method according to claim 6, characterized in that, The step of dynamically fusioning and evaluating the initial weight distribution and the environmental fitness coefficient to obtain the optimized weight distribution and weight stability index corresponding to the performance parameters includes: A multi-objective optimization method is used to fuse the initial weight distribution and the environmental fitness coefficient to obtain preliminary optimized weights. Time series analysis is performed on the preliminary optimized weights to obtain their weight fluctuation characteristics, which reflect the changes in the preliminary optimized weights over time. Based on these characteristics, a stability assessment algorithm is used to generate a weight stability index, which quantifies the stability level by calculating the variance and coefficient of variation of the weight distribution. Finally, based on the preliminary optimized weights and the weight stability index, a constrained optimization method is used to generate the optimized weight distribution, which constrains the weight distribution to simultaneously meet optimal performance and stability requirements. The online calibration and optimization of the optimized weight distribution and weight stability index to obtain the data confidence weight matrix includes: Based on the optimized weight distribution and the weight stability index, the multimodal monitoring data is analyzed using a sliding window statistical method to obtain the weight adjustment amount and performance deviation index; based on the weight adjustment amount and the weight stability index, the weight parameters are dynamically adjusted through an incremental learning algorithm to obtain the calibrated weight distribution; the calibrated weight distribution is then regularized and its stability is verified to obtain the data confidence weight matrix.

9. The method according to claim 1, characterized in that, The process of determining the leak detection result of the gas pipeline based on the comprehensive characteristics of the multimodal monitoring data and the modal synergy parameter includes: Based on the comprehensive features of the multimodal monitoring data and the modal synergy parameters, a deep learning classification model is used to identify the leakage status, and a preliminary leakage probability distribution and classification confidence level are obtained. Based on the preliminary leakage probability distribution and the classification confidence level, leakage diffusion simulation is performed using a spatiotemporal evolution analysis algorithm to obtain spatiotemporal evolution characteristics and diffusion trend prediction. The spatiotemporal evolution analysis algorithm combines a fluid dynamics model and the pipeline topology of the gas pipeline to simulate the propagation process of leaked gas in the gas pipeline. The leakage detection result is generated based on the preliminary leakage probability distribution, the classification confidence level, the spatiotemporal evolution characteristics, and the diffusion trend prediction.

10. A rapid detection system for gas pipeline leaks, characterized in that, include: The acquisition module is used to collect multimodal monitoring data through multi-source sensing devices deployed at key monitoring points of gas pipelines. The multi-source sensing system includes a distributed fiber optic sensor array, a laser methane detection unit, an acoustic sensor group, and an environmental parameter monitoring module. The first determining module is used to determine a data confidence weight matrix based on the performance parameters and environmental influencing factors of each sensor in the multi-source sensing device. The data confidence weight matrix is ​​used to represent the contribution of each modal monitoring data in the multi-modal monitoring data to this leakage detection task. The second determining module is used to determine the comprehensive features and modal synergy parameters of the multimodal monitoring data based on the multimodal monitoring data and the data confidence weight matrix. The comprehensive features are high-order feature representations after the fusion of the multimodal monitoring data, and the modal synergy parameters are used to quantify the consistency of detection results obtained based on different modal monitoring data. The third determining module is used to determine the leakage detection result of the gas pipeline based on the comprehensive characteristics of the multimodal monitoring data and the modal synergy parameter. The leakage detection result includes leakage probability distribution and spatiotemporal evolution characteristics. The leakage probability distribution is used to describe the probability of leakage at each location of the gas pipeline, and the spatiotemporal evolution characteristics are used to characterize the propagation and change law of the leakage point in the spatiotemporal dimension after leakage occurs at each location.

Images