Digital intelligent gas leakage monitoring method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of intelligent inspection, and provides a digital intelligent gas leakage monitoring method and device, electronic equipment and a storage medium. The method comprises the following steps: filtering and denoising an original sensing data stream, performing signal decomposition and trend analysis on a purified data set to obtain an intrinsic mode component set and a trend baseline signal, and performing multi-dimensional feature fusion on the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set; performing classification evaluation on the high-dimensional feature set according to the leakage state type to obtain leakage state probability distribution data, and performing leakage source positioning and leakage intensity speculation according to the leakage state probability distribution data to obtain a pipeline coordinate of a leakage point and a corresponding leakage intensity estimation value; and carrying out risk assessment and decision according to the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimation value to generate an inspection instruction and a corresponding early warning level. Through guided wave positioning and fuzzy reasoning decision making, leakage positioning and quantitative risk response are realized.

Description

Technical Field

[0001] This application relates to the field of intelligent inspection technology, and in particular to a digital intelligent gas leak monitoring method, device, electronic equipment, and storage medium. Background Technology

[0002] The role of gas pipeline inspection is to promptly detect and locate potential leaks in the pipeline network, assess the degree of danger, and trigger appropriate measures, thereby ensuring the safety of personnel and facilities, reducing property damage, and maintaining supply reliability. To achieve these goals, the inspection system must meet the following requirements: high sensitivity and high positioning accuracy, low false alarm rate and low false alarm rate, real-time or near real-time response capability, and robustness to complex environments; at the same time, the results must be quantifiable and traceable to facilitate decision-making and compliance audits, while also taking into account economy, maintainability, and interoperability with existing operation and maintenance systems.

[0003] Existing gas inspection methods mainly include manual inspection (visual and portable gas detector inspection), periodic or online pressure / flow monitoring and threshold alarms, local inspection using portable acoustic or ultrasonic detectors, spot checks using infrared or gas imaging technology, and SCADA-based abnormal condition identification and fixed-point pressure testing (such as pressure testing, flaw detection, or pipeline cleaning). These methods typically rely on a single sensor type or human experience judgment, combined with periodic maintenance records and historical data for risk assessment, and respond to alarms manually or through established procedures.

[0004] Single sensor or threshold alarms often lack sufficient sensitivity or location capability, thus missing minute transient leaks. To avoid missing leaks, a large number of false alarms are generated, increasing the maintenance burden. Manual inspections are time-consuming and have long cycles. Complex inspection methods are costly and difficult to cover the entire network, making it impossible to provide quantitative confidence and immediate, interpretable location and risk assessment for emergency decision-making. Summary of the Invention

[0005] In view of this, this application provides a digital gas leak monitoring method, device, electronic device, and storage medium to solve the problems of insufficient accuracy and timeliness in leak detection.

[0006] The first aspect of this application provides a digital intelligent gas leak detection method, the method comprising: The raw sensor data stream acquired in real time is filtered and denoised to obtain a purification sensor dataset. The purification sensor dataset is then subjected to signal decomposition and trend analysis to obtain the intrinsic mode component set and trend baseline signal. Multidimensional feature fusion is performed on the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set; The high-dimensional feature set is classified and evaluated according to the preset leakage state type to obtain leakage state probability distribution data; Based on the probability distribution data of the leakage state, the leakage source is located and the leakage intensity is inferred, and the pipeline coordinates of the leakage point and the corresponding leakage intensity estimate are obtained. Risk assessment and response decisions are made based on the leakage state probability distribution data, the pipeline coordinates, and the leakage intensity estimate, generating inspection instructions and corresponding early warning levels.

[0007] In an optional implementation, the step of filtering and denoising the raw sensor data stream acquired in real time to obtain a purified sensor dataset includes: The raw sensor data stream acquired in real time is processed in the time domain to obtain a synchronized sensor dataset. Based on the preset sensor deployment data, the synchronous sensing dataset is spatially mapped using a preset spatial coordinate system to obtain a spatiotemporally aligned dataset. The pressure sensor dataset in the spatiotemporally aligned dataset is state estimated using a preset unscented Kalman filter algorithm to generate purification pressure data. Frequency band thresholds are defined based on the spectral characteristics of the ultrasonic sensing dataset in the spatiotemporal aligned dataset. Adaptive wavelet packet denoising is then performed on the ultrasonic sensing dataset based on the frequency band thresholds to generate purified ultrasonic data.

[0008] In an optional implementation, the step of performing signal decomposition and trend analysis on the purification sensor dataset to obtain the intrinsic mode component set and trend baseline signal includes: The distribution of dominant frequency components is analyzed on a pre-defined historical sensor dataset to determine the optimal combination of decomposition parameters for each dataset. Based on the optimal decomposition parameter combination, variational mode decomposition is performed on the purification pressure data and the purification ultrasonic data to obtain a subset of pressure intrinsic mode components and a subset of ultrasonic intrinsic mode components. According to a preset frequency threshold, components with frequencies lower than the frequency threshold are selected from the subset of pressure intrinsic mode components as a low-frequency mode component set. The low-frequency modal component set is fitted into a corresponding trend baseline signal using a preset regularization fitting algorithm.

[0009] In an optional implementation, the step of multidimensional feature fusion of the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set includes: Time-frequency feature extraction is performed on the intrinsic mode component set to obtain the corresponding time-frequency feature set; The trend baseline signal is dynamically changed to obtain the corresponding trend change feature set. The intrinsic modal component set is subjected to temporal correlation change features between different modes to obtain a correlation feature set; Principal component analysis is performed on the time-frequency feature set, the trend change feature set, and the correlation feature set to obtain each principal component feature set; The importance of leakage identification is evaluated on the principal component feature set to obtain the corresponding identification association score, and the identification association score is converted into the corresponding weight coefficient according to the preset weight mapping rule; The principal component feature set is weighted and fused according to the weight coefficients to generate a high-dimensional feature set.

[0010] In an optional implementation, the step of classifying and evaluating the high-dimensional feature set according to a preset leakage state type to obtain leakage state probability distribution data includes: Multi-scale local feature extraction is performed on the high-dimensional feature set to obtain a deep feature set; The importance of the deep feature set is evaluated through a preset attention mechanism to calculate the importance weight of each deep feature. Based on the preset leakage state type and the importance weight, the deep feature set is classified and a leakage state probability distribution data is obtained.

[0011] In an optional implementation, the step of locating the leak source and inferring the leak intensity based on the leak state probability distribution data to obtain the pipe coordinates of the leak point and the corresponding estimated leak intensity includes: Using a preset generalized cross-correlation algorithm, the time difference of arrival of the purified ultrasonic data is calculated based on the sensor deployment data to obtain the set of time differences of arrival of the signals between each sensor. Using a preset particle swarm optimization algorithm, the leak source is located based on the leak state probability distribution data and the signal arrival time difference set, and the pipe coordinates of the leak point are obtained. Based on the pipeline coordinates and the signal arrival time difference set, ultrasonic energy attenuation inversion is performed to obtain the corresponding leakage intensity estimate.

[0012] In an optional implementation, the step of performing risk assessment and response decision based on the leakage state probability distribution data, the pipeline coordinates, and the leakage intensity estimate, and generating inspection instructions and corresponding early warning levels, includes: By using a preset coordinate feature database, the corresponding environmental sensitivity data is queried based on the pipeline coordinates; By using preset multi-criteria fuzzy inference rules, the leakage state probability distribution data, the environmental sensitivity data, and the leakage intensity estimate are mapped to a set of fuzzy linguistic variables; The fuzzy language variable set is converted into a risk level fuzzy set by a preset safety procedure library, and the risk level fuzzy set is defuzzified to obtain a comprehensive risk score. The comprehensive risk score is converted into a corresponding warning level according to the preset scoring and level mapping rules, and the warning level, the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimate are encapsulated into an inspection command through a preset command generation method.

[0013] A second aspect of this application provides a digital gas leak monitoring device, the device comprising: The filtering and denoising module is used to filter and denoise the raw sensor data stream acquired in real time to obtain a purified sensor dataset. The signal trend module is used to perform signal decomposition and trend analysis on the purification sensor dataset to obtain the intrinsic mode component set and trend baseline signal. The feature fusion module is used to perform multi-dimensional feature fusion on the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set. The classification and evaluation module is used to classify and evaluate the high-dimensional feature set according to the preset leakage state type to obtain leakage state probability distribution data. The location estimation module is used to locate the leak source and estimate the leak intensity based on the leak state probability distribution data, and obtain the pipeline coordinates of the leak point and the corresponding estimated leak intensity value. The risk decision module is used to perform risk assessment and response decisions based on the leakage state probability distribution data, the pipeline coordinates, and the leakage intensity estimate, and to generate inspection instructions and corresponding early warning levels.

[0014] A third aspect of this application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the digital gas leak monitoring method described above.

[0015] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the digital gas leak monitoring method described above.

[0016] In summary, this application includes at least the following beneficial technical effects: 1. By using multi-modal and multi-rate synchronous acquisition, unscented Kalman and adaptive wavelet packet denoising, VMD decomposition and multi-scale feature fusion, as well as time difference localization and optimization solution based on guided wave dispersion, the sensitivity and localization accuracy of weak transient leakage signals are significantly improved, and false alarms and missed alarms are reduced.

[0017] 2. Output the leakage probability distribution, feature importance, and location confidence interval, and incorporate these quantitative indicators into fuzzy inference risk assessment to make the detection results interpretable, traceable, and easy to determine for operation and maintenance.

[0018] 3. Based on location and intensity estimation, perform multi-criteria fuzzy decision-making and map it to hierarchical early warning and actions, thereby shortening response time, optimizing resource allocation, and ensuring compliance with safety procedures and audit requirements. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a digital gas leak monitoring method provided in an embodiment of this application; Figure 2 This is a functional block diagram of a digital gas leak monitoring device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] like Figure 1 The diagram shown is a flowchart of the intelligent gas leak monitoring method provided in this application embodiment. The intelligent gas leak monitoring method provided in this application embodiment includes the following steps.

[0023] Step S1: Filter and denoise the raw sensor data stream acquired in real time to obtain a purification sensor dataset, and perform signal decomposition and trend analysis on the purification sensor dataset to obtain the intrinsic mode component set and trend baseline signal.

[0024] It should be understood that the raw sensor data stream originates from a multimodal sensor network specifically built for gas pipeline inspection scenarios. This multimodal sensor network consists of two complementary types of sensors: one is an ultrasonic sensor array based on microelectromechanical systems (MEMS), with a frequency response range covering 90 Hz to 80 kHz. Sensing units are deployed along the four directions (up, down, left, and right) at each pipeline monitoring node, forming a spatial capture capability for high-frequency acoustic signals from leaks; the other is a high-precision strain gauge pressure sensor with a range set from 0 to 10 kPa, specifically designed to detect minute pressure fluctuations in low-pressure gas pipelines caused by leaks or normal gas usage. All sensors are connected to a distributed data acquisition unit via double-shielded cables, and each unit is equipped with a local temperature-controlled crystal oscillator clock module to maintain stable acquisition timing. The time series containing ultrasonic waveforms and pressure readings generated in real time by the multimodal sensor network constitutes the raw sensor data stream. This raw sensor data stream provides the most original and comprehensive physical field observation evidence for subsequent analysis.

[0025] Because the ultrasonic signal sampling rate used in this embodiment is as high as 200 kHz while the pressure signal sampling rate is 10 Hz, and there is slight clock drift at each acquisition node, direct fusion would lead to signal misalignment. Therefore, after obtaining the original sensor data stream, time-domain synchronization processing is required first. This embodiment adopts a mechanism combining a precise time protocol and local clock redundancy. The master clock sends synchronization pulses to all nodes via industrial Ethernet. Each node compensates for network jitter based on this pulse and the stability of its local crystal oscillator, and assigns a unified timestamp to each data sample, thereby precisely aligning the data streams with heterogeneous sampling rates on the time axis and generating a synchronized sensor dataset. For example, if an ultrasonic sensor node's acquisition time is 1 millisecond ahead due to clock drift, its recorded leakage ultrasonic signal will not match the pressure drop event recorded by the pressure sensor. This time-domain synchronization ensures that all sensors' records of the same event are strictly corresponding in time.

[0026] To establish a precise correlation between the synchronized sensor data and the pipeline location in the physical world, spatial coordinate mapping of the synchronized sensor dataset is required. The spatial mapping process used in this embodiment relies on two key preset inputs: first, a preset spatial coordinate system, a three-dimensional Cartesian coordinate system built upon the pipeline system. Its origin is typically set at the starting point of the pipeline network or a key hub station. The Z-axis aligns with the direction of gravity, and the XY plane is parallel to the horizontal plane of the earth. This coordinate system provides a unified three-dimensional coordinate description for any point within the pipeline network. Second, sensor deployment data, a set of structured data recording the precise installation parameters of each sensor on the pipeline, sourced from the pipeline geographic information system. Specifically, this includes: the sensor's precise station number in the pipeline mileage marker system, its circumferential installation angle relative to the pipeline axis (e.g., located at the top, bottom, or side of the pipeline), the sensor's own three-dimensional coordinates in the preset spatial coordinate system, the attributes of the pipe section it belongs to (including pipe diameter, wall thickness, and material), and important environmental parameters for buried pipe sections (such as burial depth, backfill soil type, and anti-corrosion layer condition). Specifically, based on the unique sensor identifier attached to each data packet in the synchronous sensing dataset, the sensor deployment database is queried to obtain the spatial attributes of the sensor. Then, coordinate transformation is used to map the local observation data of the sensor to a unified spatial coordinate system. For example, an ultrasonic sensor installed at station K23+150, at the top of the pipeline and a depth of 1.2 meters, will have its collected sound pressure data, after mapping, not only include physical readings but also be assigned precise coordinates in this spatial coordinate system (e.g., X=102345.67, Y=305678.90, Z=-1.20). After all sensor data undergoes this mapping, a spatiotemporally aligned dataset is generated. This spatial mapping operation provides a spatial reference for leak source location. When the system detects an abnormal signal, it can immediately project the signal source back to the specific pipeline location. For example, by analyzing the temporal and intensity differences of the same leak signal received by sensors at different spatial locations, it is possible to accurately infer that the leak point occurs at the top of the pipeline between K23+100 and K23+200, rather than in other sections or at the bottom.

[0027] Based on a spatiotemporally aligned dataset, targeted filtering and denoising were applied to both pressure and ultrasonic data. The pressure sensing data underwent state estimation using an unscented Kalman filter algorithm. This algorithm approximates the nonlinear variation of the pressure signal through a carefully designed set of Sigma sampling points, continuously optimizing the state estimate during the prediction and update phases. It effectively smooths random fluctuations and suppresses gross errors caused by sudden pressure changes or electromagnetic interference. Its advantage lies in not requiring the assumption of a linear system, making it particularly suitable for handling complex dynamic pressure changes in gas pipelines caused by valve actions and user gas consumption. After this processing, the output is purified pressure data that clearly reflects the true trend of pipeline pressure changes.

[0028] Meanwhile, adaptive wavelet packet denoising is employed for the ultrasonic sensing data. First, its spectral characteristics are analyzed. Since the ultrasonic energy generated by gas leaks is mainly concentrated in the 20 kHz to 60 kHz range, while environmental mechanical vibration noise is mostly distributed in other frequency bands, a frequency band threshold containing leak characteristics is determined. The wavelet packet analysis used in this embodiment has advantages over traditional wavelet analysis, as it can perform more refined secondary division of high-frequency bands. The denoising process first decomposes the signal into multiple sub-bands, and for the wavelet packet coefficients of each sub-band, an adaptive threshold is calculated based on its statistical characteristics. This threshold... From the formula It is confirmed that, among them, The noise standard deviation is estimated based on the median of the subband coefficients. Let be the number of coefficients, and j be the subband index. Coefficients with absolute values ​​below a threshold are considered noise and set to zero, while those above the threshold are contracted and retained. Finally, the subbands are reconstructed to obtain the purified ultrasonic data. This method can effectively remove high-frequency electrical noise and random interference while preserving the transient characteristics of the leaked ultrasonic signal to the maximum extent.

[0029] After data purification is completed, the obtained purification sensor dataset (including purification pressure data and purification ultrasonic data) needs to be processed into the signal decomposition and trend analysis stage shown below.

[0030] First, this embodiment utilizes a particle swarm optimization algorithm to learn from a preset historical sensor dataset to determine the optimal combination of decomposition parameters for variational mode decomposition. The optimal combination of decomposition parameters includes the number of modes K and a penalty parameter α. The optimization objective of this embodiment is to minimize the envelope entropy of the decomposition result and maximize its correlation coefficient with the original signal, ensuring that the decomposed modes are both pure and physically meaningful. For example, for a complex historical signal containing periodic noise and transient leakage pressure from a diaphragm gas meter, the optimization process automatically finds a set of K and α, enabling VMD to clearly separate the meter's mechanical noise, leakage signal, and background disturbances into different modes.

[0031] Subsequently, using this optimal combination of decomposition parameters, variational mode decomposition was performed on the purification pressure data and purification ultrasonic data, respectively. The decomposition process adaptively decomposes the signal into a series of intrinsic mode components with specific center frequencies by solving a constrained variational problem, thus obtaining subsets of pressure intrinsic mode components and subsets of ultrasonic intrinsic mode components. For pressure signals, key features containing leakage information are often hidden in low-frequency trends. Therefore, based on a preset frequency threshold (e.g., 0.5 Hz), all components with frequencies below this threshold are selected from the pressure intrinsic mode component subset to form a low-frequency mode component set. The selection of this frequency threshold is based on the fact that the pressure recovery process caused by small leaks in gas pipelines is usually slow, and its dynamic characteristics are concentrated in the extremely low frequency range.

[0032] Finally, a regularized fitting algorithm is used to fit this low-frequency modal component set into a smooth trend baseline signal. The fitting process employs Tikhonov regularization to control the smoothness of the fitted curve, and cross-validation is used to determine the regularization parameters to avoid overfitting to noise and ensure that the extracted trend accurately reflects the overall recovery trend of pipeline pressure after a leak. For example, in the event of a minor leak, the pressure may exhibit a slow recovery or remain at a low level after an initial sharp drop, while the pressure usually recovers rapidly after normal gas supply is shut off. This trend baseline signal is the core feature for quantifying and identifying this difference. Thus, step S1 completes the entire preprocessing process from raw multimodal data to a purified dataset, and then to the refined decomposition of signals and key trend features, laying a reliable data foundation for subsequent feature fusion and intelligent diagnosis.

[0033] Step S2: Perform multi-dimensional feature fusion on the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set.

[0034] First, time-domain and frequency-domain features are extracted from the intrinsic modal component set. Specifically, the time-domain feature extraction operation used in this embodiment focuses on the waveform statistical characteristics of each modal component, calculating its mean, variance, skewness, and kurtosis. The mean reflects the DC offset or average energy level of the signal; the variance characterizes the signal's fluctuation intensity; the skewness describes the asymmetry of the signal amplitude distribution; and the kurtosis quantifies the prominence of extreme events or impulse components in the signal. The frequency-domain feature extraction operation used in this embodiment converts the modal components to the frequency domain using a Fast Fourier Transform (FFT), calculating their centroid frequency, frequency standard deviation, and spectral entropy. The centroid frequency indicates the concentrated frequency band of the signal energy; the frequency standard deviation describes the degree of spectral diffusion; and the spectral entropy measures the disorder or randomness of the frequency components. The combination of the obtained time-domain and frequency-domain features constitutes a time-frequency feature set, aiming to characterize the local and global patterns of the leaked signal from two orthogonal perspectives: statistical and spectral analysis. For example, a specific mode of ultrasonic component caused by a leak may have a variance significantly higher than the background noise, while its spectral entropy may be lower than the ambient noise because the energy is concentrated in the leak's characteristic frequency band.

[0035] Meanwhile, since the trend baseline signal carries information about the macroscopic recovery process of pipeline pressure after a leak, it is necessary to calculate the dynamic change characteristics of the trend baseline signal to focus on extracting its morphological evolution features. The calculations used in this embodiment include the initial slope during the pressure drop phase, the curvature change during the pressure recovery phase, the horizontal offset after reaching steady state, and the head-to-tail pressure difference within the entire observation window. The initial slope can distinguish between the rapid pressure collapse caused by a leak and the slow decline after normal gas usage ends; the curvature change can capture whether the pressure recovery is exponential, linear, or other complex patterns; small leaks often lead to weak recovery or even a continuous slow decline; the steady-state offset directly reflects the permanent pressure loss caused by the leak. These trend change features quantify the macroscopic pattern in pressure dynamics that best distinguishes between leaks and normal operating conditions. For example, a continuous small-flow leak will cause the trend baseline to exhibit a unique shape with a negative head-to-tail difference and a near-zero recovery curvature, which contrasts sharply with the trend of rapid pressure recovery and stabilization after the stove is normally turned off.

[0036] To uncover the synergistic and causal relationships between pressure and ultrasonic signals in leakage events, it is necessary to extract temporal correlation features between different modes to obtain a correlation feature set. Specifically, the cross-correlation function between subsets of pressure intrinsic modal components and subsets of ultrasonic intrinsic modal components is calculated, and the maximum cross-correlation coefficient and its corresponding time delay are sought. This time delay may reveal the sequence between the pressure drop event and the ultrasonic emission during leakage. Furthermore, the coherence coefficients of the two modal sets in specific frequency bands are calculated to assess their linear dependence in the frequency domain. Leakage events typically induce high coherence in specific frequency bands (such as the 20-60kHz ultrasonic band and the extremely low-frequency pressure fluctuation band). In addition, phase synchronization indices are calculated to analyze whether the phases of different modal signals exhibit locking or synchronization phenomena during leakage. The correlation feature set obtained in this embodiment enhances the reliability of detection by utilizing the coupling relationship of multi-physics field signals; it is difficult to find consistent correlation evidence in the signals of another physical field for false alarms from a single sensor.

[0037] After obtaining three sets of heterogeneous features—time-frequency feature set, trend change feature set, and correlation feature set—redundant information needs to be removed from these features to reduce data processing complexity. The principal component analysis (PCA) used in this embodiment is a linear dimensionality reduction and noise reduction technique. It transforms potentially correlated original features into a set of linearly uncorrelated principal components through orthogonal transformation. These principal components are arranged in descending order of variance contribution rate. In practice, the merged original high-dimensional feature matrix is ​​standardized, and then the eigenvalues ​​and eigenvectors of its covariance matrix are calculated. The top k principal components with a cumulative variance contribution rate exceeding 95% are selected to form each principal component feature set. Through the above PCA processing, the data size is significantly compressed while retaining most of the original information, and collinearity between features is eliminated, providing numerically stable input for subsequent classification processing.

[0038] Subsequently, the importance of leakage identification is evaluated on the principal component feature set. This embodiment uses a random forest algorithm as the evaluation tool, training a random forest model containing a large number of decision trees on a pre-labeled historical dataset. After training, based on out-of-bag data or through a permutation importance method, the average reduction in impurity brought by each denoising feature when used to split nodes across all decision trees in the forest is calculated. This average value is the corresponding identification association score. This score quantitatively reflects the contribution of a single feature to correctly classifying leaked and non-leaked samples. For example, the head-to-tail difference feature of a trend baseline signal and the energy feature of a specific frequency band of ultrasound often have significantly higher identification association scores than some general statistical features in practice. Next, the identification association score is converted into corresponding weight coefficients. This embodiment uses a weight mapping rule to realize the conversion between the identification association score and the weight coefficients. This weight mapping rule is a Softmax function, which converts a set of real scores into a set of probability distributions, significantly amplifying the weights of important features, while the sum of the weights of all features is 1. Specifically, given the identification association score S of the i-th feature... i Its weighting coefficient W i From the formula The calculation shows that the denominator is the sum of all feature scores after taking the exponent. This operation ensures that features with high discriminative power are given greater weight in the feature fusion process.

[0039] Finally, the principal component feature sets are weighted and fused based on the calculated weight coefficients to generate the final high-dimensional feature set. The fusion process is not a simple vector concatenation, but rather a weighted combination of each feature vector in the principal component feature set. Specifically, there are m denoised feature vectors (i.e., the principal component feature set), each vector having dimension k (i.e., the number of principal components), and corresponding m weight coefficients. The high-dimensional feature set generated after weighted fusion represents each sample's feature as a weighted sum of the values ​​of that sample across all m denoised feature vectors, forming a unified feature vector that condenses multi-dimensional information (i.e., the high-dimensional feature set). This high-dimensional feature set is the final mathematical representation input to the subsequent classification and evaluation model. Thus, step S2 completes the complex transformation from the original signal decomposition results to the optimized feature representation, preparing the key input for the final intelligent diagnosis of the leakage state.

[0040] Step S3: Classify and evaluate the high-dimensional feature set according to the preset leakage state type to obtain leakage state probability distribution data.

[0041] It should be understood that the leakage state types set in the embodiments of this application generally include three categories: "normal state", "small leakage state" and "large leakage state". Step S3 of this application maps this mathematical representation that integrates multi-dimensional information to the preset leakage state type and outputs a quantified probability distribution, thereby completing the key transformation from data to diagnostic decision.

[0042] First, multi-scale local feature extraction is required from the high-dimensional feature set. In this embodiment, the multi-scale local feature extraction is performed by a one-dimensional convolutional neural network. The input layer of this network receives the high-dimensional feature vector generated in step S2 and treats it as a one-dimensional sequence. This network structure contains multiple parallel one-dimensional convolutional layers, which use convolutional kernels of different scales (e.g., kernels with widths of 3, 5, and 7) to scan the input feature sequence in parallel. Smaller-scale convolutional kernels focus on capturing local fine patterns formed by adjacent elements in the feature sequence, such as the short-range coupling relationship between energy in a specific ultrasonic frequency band and the slope of a pressure trend; larger-scale convolutional kernels are responsible for perceiving larger-scale interaction and dependency patterns in the feature sequence, such as the macroscopic synergistic situation formed between time-frequency features, trend features, and correlation features. Each convolutional layer is followed by a non-linear activation function (e.g., ReLU) and pooling operation, ultimately concatenating the output feature maps of all parallel convolutional paths along the channel dimension to form a deep feature set capable of simultaneously representing local details and global context. Through the above-described operation, the execution system of this application can automatically learn and combine complex patterns in high-dimensional features that are crucial for leak identification, rather than relying on fixed rules designed manually.

[0043] After obtaining the deep feature set, its importance is evaluated through a pre-defined attention mechanism. The attention mechanism used in this embodiment typically employs a self-attention or scaled dot product attention model. Its working principle is to treat each feature vector in the deep feature set as a query vector, key vector, and value vector. By calculating the dot product of the query vector and all key vectors and applying the Softmax function for normalization, a set of attention weights is obtained, which are the importance weights corresponding to each deep feature. The obtained importance weights quantify the relative contribution of each feature dimension in the deep feature set to the final classification decision when evaluating the current input sample. For example, for a sample suspected of having a small leak, the attention mechanism might give higher weights to the "recovery curvature" feature from the trend baseline signal and the "spectral entropy" feature from the low-frequency ultrasonic mode, while reducing the weights of some general statistical features. This indicates that the model will focus on the two key pieces of evidence: weak pressure recovery and changes in the ultrasonic spectrum.

[0044] Subsequently, the deep feature set is weighted and fused according to the calculated importance weights to generate a deep feature set containing importance weights. The fusion operation used in this embodiment is not a simple summation, but rather a positionally weighted linear combination of each feature vector in the deep feature set. Specifically, there are m deep feature vectors, each with dimension d, and corresponding m importance weights. The deep feature set generated after weighted fusion, containing importance weights, has a feature representation for each sample that is the weighted sum of the values ​​of that sample across all m deep feature vectors, forming a unified feature vector that condenses attention information. The above operation is essentially a feature selection and information condensation process. It strengthens the feature signals most relevant to the current leakage scenario while suppressing irrelevant or redundant information, enabling the subsequent classifier to make decisions based on a purer and more discriminative feature representation.

[0045] Finally, based on the preset leakage state types, a classification decision is made on the weighted fusion deep feature set containing importance weights to obtain leakage state probability distribution data. The embodiments of this application typically consist of a fully connected layer followed by a Softmax function. The fully connected layer maps the deep feature set containing importance weights to an unnormalized score vector equal to the number of leakage state types, with each score corresponding to original evidence for one state type. Subsequently, the Softmax function transforms these original scores into a probability distribution. Given a score vector Z=[z1,z2,z3], corresponding to normal, small leakage, and large leakage respectively, the probability p of its i-th class is... i From the formula The calculation yields a probability distribution that ensures the sum of probabilities for all states equals 1, thus outputting an intuitive probability distribution of leakage states, such as [Normal: 0.05, Small Leak: 0.85, Large Leak: 0.10]. This probability distribution not only provides the most probable leakage state but also reflects the confidence level of the model's judgment and the probability differences between different states, offering richer and more flexible information for subsequent risk assessment and decision-making than a single-category label. For example, a sample with a probability distribution of [Normal: 0.45, Small Leak: 0.50, Large Leak: 0.05] indicates that the system has detected weak signs of leakage with high uncertainty. In such cases, a more cautious review instruction might be triggered rather than an immediate alarm.

[0046] Step S4: Based on the probability distribution data of the leakage state, locate the leakage source and infer the leakage intensity to obtain the pipeline coordinates of the leakage point and the corresponding estimated leakage intensity value.

[0047] The core objective of step S4 in this application is to determine the precise physical location of the leak event and assess its severity. This process is accomplished by integrating acoustic localization principles with optimization algorithms.

[0048] First, the time difference of arrival (TDOA) of the ultrasonic data is calculated using a pre-defined generalized cross-correlation algorithm. It should be understood that the generalized cross-correlation algorithm is an improved version of the traditional cross-correlation algorithm, enhancing robustness against noise and reverberation through weighted processing in the frequency domain. Specifically, the ultrasonic signal segments recorded by each sensor node are extracted from the ultrasonic data obtained in step S1 above. Based on the spatial location information recorded in the sensor deployment data, all possible sensor pair combinations are selected. For each sensor pair (i,j), the received signals are transformed to the frequency domain using a Fast Fourier Transform (FFT). In the frequency domain, the cross-power spectrum of the two signals is multiplied by a weighting function. A commonly used weighting function is PHAT (Phase Transform), which has the form: .in, This is the cross-power spectrum of the two signals. This weighting method essentially whitens the cross-power spectrum, preserving phase information while weakening the amplitude effect, making the time delay estimation more immune to environmental reverberation. The weighted cross-power spectrum is then subjected to an inverse Fourier transform to obtain the generalized cross-correlation function. The peak position of this function corresponds to the time difference of arrival between the two sensor signals. By systematically calculating the time differences between all sensor pairs, a complete set of signal arrival time differences is finally obtained. The above operations transform the physical phenomenon of sound wave propagation caused by leakage into a computable mathematical quantity. For example, when the leak point is closer to sensor i and farther from sensor j, the signal arrives at sensor i significantly earlier than at sensor j. The calculated time difference Δt in this case... ij It will be a significantly negative value.

[0049] After obtaining a reliable set of signal arrival time differences, the leak source is precisely located spatially using a pre-defined particle swarm optimization algorithm. Considering the dispersion effect of sound wave propagation in gas pipelines—that is, sound wave components of different frequencies propagate at different speeds—a sound wave propagation time delay model incorporating dispersion correction needs to be established. The sound wave propagation time delay model used in this embodiment can be expressed as: L i and L j Let i and j represent the propagation path lengths from the leak point to sensors i and j, respectively. It is the group velocity corresponding to frequency f. It is the signal bandwidth. Based on this physical model and the measured time difference set, the positioning objective function is constructed as follows: .in It measures time difference. The time difference is calculated theoretically based on the coordinates (x, y, z) of the candidate leak point. These are weighting coefficients (values ​​derived from the leakage state probability distribution data output in step S3). Specifically, when the overall confidence level of the "leakage" category in the leakage state probability distribution is high, the weighting coefficient increases accordingly, and vice versa. This allows the localization process to fully utilize the judgment confidence of the front-end classifier. The particle swarm optimization algorithm performs a global search within the parameter space of this objective function (i.e., the three-dimensional space of the pipeline). The algorithm initializes a set of random particles, each representing a candidate leak location. The particles continuously update their velocity and position based on their individual historical best position and the collective historical best position. Through iterative search, the algorithm eventually converges to the optimal coordinate point that minimizes the objective function; this point is the pipeline coordinate of the leak point. The advantage of this optimization method is that it avoids getting trapped in local optima, which is particularly important when there are multiple possible leak points.

[0050] After obtaining the pipe coordinates of the leak point, ultrasonic energy attenuation is inverted and inferred based on these coordinates and the signal arrival time difference set to estimate the leak intensity. In this embodiment, the leak intensity estimation is based on a sound wave propagation attenuation model in the pipe medium, which considers geometric diffusion attenuation, medium absorption attenuation, and scattering attenuation caused by the pipe structure. Specifically, the implementation process is as follows: first, based on the known leak point coordinates and the coordinates of each sensor, the theoretical path length of the sound wave propagating from the leak point to each sensor is calculated. Then, an inversion model is established based on the physical laws of ultrasonic wave attenuation in the pipe. ,in It is the signal strength received by the i-th sensor (extracted from the purification ultrasonic data). The original strength at the leakage source (to be solved), and α is the attenuation coefficient of the medium. This is the propagation path length. The leakage source strength is obtained by solving this overdetermined system of equations using the least squares method. The optimal estimate is determined by the sensor's sensitivity. In practice, individual differences in sensor sensitivity and additional attenuation caused by pipe connections must also be considered; these factors are compensated for through prior calibration experiments. The final output leak intensity estimate is a normalized dimensionless index, whose magnitude directly reflects the severity of the leak. For example, the ultrasonic source intensity generated by a small hole leak might be estimated at 0.15, while the source intensity of a larger crack leak might reach 0.75. This quantitative index provides an objective basis for subsequent risk assessment.

[0051] The operation of step S4 above completes the entire process from probability judgment to physical location and quantitative assessment, transforming the abstract leakage probability into specific spatial location and intensity indicators, providing key spatial information and severity parameters for the final response decision.

[0052] Step S5: Based on the leakage state probability distribution data, the pipeline coordinates, and the leakage intensity estimate, perform risk assessment and response decision-making, and generate inspection instructions and corresponding early warning levels.

[0053] To transform the detection, location, and quantitative information generated in the above steps into executable safety management instructions, the environmental sensitivity data corresponding to the pipeline coordinates determined in step S4 is first queried using a pre-defined coordinate feature database. This coordinate feature database is a structured database linked to the pipeline geographic information system. Key fields recorded include the pipeline section's mileage range, surrounding population density level (e.g., classified as uninhabited areas, low-density residential areas, high-density residential areas, commercial areas, schools, hospitals, etc.), underground space complexity (e.g., whether it intersects with other pipelines, whether there are enclosed spaces), surface building structure type, and emergency resource accessibility. When a specific pipeline coordinate (e.g., K23+150) is input, the database query engine matches spatial relationships and returns the environmental sensitivity data for the pipeline section to which that coordinate point belongs. Environmental sensitivity data is typically quantified as a comprehensive sensitivity score or classification level. For example, an underground pipeline traversing a downtown commercial area would be marked as "extremely high" in environmental sensitivity, while a pipeline located in suburban farmland would be marked as "low."

[0054] After obtaining environmental sensitivity data, the three heterogeneous input parameters—leakage state probability distribution data, environmental sensitivity data, and leakage intensity estimates—are mapped into a unified set of fuzzy linguistic variables using pre-defined multi-criteria fuzzy inference rules. Fuzzy inference is an effective tool for handling such uncertainties and subjective judgments. Specifically, each input parameter is first defined with its fuzzy set and membership function. For example, the leakage state probability, taking the probability value of the "leakage" category, is mapped to three fuzzy linguistic variables: "low," "medium," and "high." Its membership function is defined using a trapezoidal or triangular function; for instance, when the probability value is below 0.3, the membership degree for "low" is 1, and between 0.3 and 0.7, the membership degree for "medium" increases linearly. Environmental sensitivity data is mapped to four levels: "low," "medium," "high," and "extremely high." Leakage intensity estimates are mapped to three levels: "weak," "moderate," and "severe." This mapping process is fuzzification, which converts precise input values ​​into membership degrees on various fuzzy linguistic variables. For example, a specific input combination (probability value = 0.85, environmental sensitivity data = "draft", leakage intensity estimate = 0.6), after fuzzification, may have a membership degree of 0.9 for "high" leakage probability and 0.1 for "medium" leakage probability; a membership degree of 1.0 for "high" environmental sensitivity; and a membership degree of 0.7 for "medium" leakage intensity and 0.3 for "severe" leakage intensity.

[0055] Next, the fuzzy linguistic variable set is converted into a risk level fuzzy set using a pre-defined safety procedure library. In this embodiment, the safety procedure library is composed of fuzzy rules in the form of "IF-THEN," which are based on industry safety standards, historical accident analysis, and expert experience. A typical rule form is: "IF Leakage Probability IS High AND Environmental Sensitivity IS High AND Leakage Intensity IS Medium THEN Risk Level IS High." The inference engine receives all fuzzified input premises and, through fuzzy logic operations (typically using the Mamdani inference model, employing "smaller" as the AND operator and "larger" as the OR operator), aggregates the consequents (i.e., risk level conclusions) of all triggered rules to form a comprehensive risk level fuzzy set. The obtained risk level fuzzy set represents all possible output risk levels and their corresponding membership degrees.

[0056] Then, the fuzzy risk level set is defuzzified to obtain an accurate comprehensive risk score. Defuzzification is the process of transforming fuzzy reasoning conclusions into clear values, and the most commonly used method is the centroid method. This method calculates the abscissa value corresponding to the centroid of the area enclosed by the membership function curve of the fuzzy risk level set and the abscissa axis. Through this calculation, the fuzzy risk description (such as "between high risk and extremely high risk") is transformed into a specific, comparable numerical value, such as 85 points. This comprehensive risk score quantitatively integrates information from three dimensions: leakage probability, severity of consequences, and environmental vulnerability. Subsequently, the comprehensive risk score is converted into the corresponding warning level according to a preset scoring and level mapping rule. The scoring and level mapping rule used in this embodiment is a piecewise function that maps continuous scoring intervals to discrete warning levels. For example, a score of 0-30 corresponds to "Blue Level IV Warning" (low risk), 31-60 corresponds to "Yellow Level III Warning" (medium risk), 61-80 corresponds to "Orange Level II Warning" (high risk), and 81-100 corresponds to "Red Level I Warning" (extremely high risk). Each warning level is not just a color and a name, but also associated with a set of pre-set emergency response plans and handling authorities.

[0057] Finally, using a preset instruction generation method, the warning level, leakage status probability distribution data, pipeline coordinates, and leakage intensity estimate are encapsulated into a structured inspection instruction. This embodiment employs an instruction generator (i.e., a logic module), which calls the corresponding template from the instruction template library based on the final warning level. A complete inspection instruction is a standardized data packet whose fields include at least: a unique event identifier, timestamp, warning level, pipeline coordinates of the leak point (including mileage and 3D coordinates), leakage status classification and its probability, leakage intensity estimate, recommended handling measures (such as "record and file," "dispatch personnel for on-site verification," "strengthen monitoring by the dispatch center," "recommend pressure adjustment," "immediately remotely close segment valves," etc.), response time limit requirements, and information on the responsible unit or personnel. For example, a red level 1 warning instruction would explicitly instruct the automatic closure of valves in a specific section within a few minutes and notify the emergency response team, while a yellow level 3 warning might only require inspection personnel to conduct on-site confirmation within a few hours. After the instruction is encapsulated, it is distributed to the monitoring center, mobile inspection terminals, and automatic control equipment via a secure communication protocol, driving the entire emergency response system.

[0058] This application is applied to the field of intelligent inspection technology. It obtains a purified sensor dataset by filtering and denoising the original sensor data stream. Signal decomposition and trend analysis are then performed on the purified sensor dataset to obtain intrinsic mode components (IMCs) and trend baseline signals. Multi-dimensional feature fusion is then performed on the IMCs and trend baseline signals to generate a high-dimensional feature set. The high-dimensional feature set is classified and evaluated according to the leakage state type to obtain leakage state probability distribution data. Based on the leakage state probability distribution data, the leakage source is located and the leakage intensity is inferred to obtain the pipeline coordinates of the leakage point and the corresponding leakage intensity estimate. Based on the leakage state probability distribution data, pipeline coordinates, and leakage intensity estimate, risk assessment and decision-making are performed to generate inspection instructions and corresponding warning levels. This application achieves high-precision leakage location and quantified risk response by simultaneously acquiring and purifying signals at multiple rates, followed by VMD decomposition and multi-scale feature fusion, and then classification using an attention-based one-dimensional convolutional network and decision-making based on guided wave localization and fuzzy inference.

[0059] like Figure 2 The diagram shown is a functional block diagram of a digital gas leak monitoring device provided in an embodiment of this application.

[0060] In some embodiments, the intelligent gas leak monitoring device 2 may include multiple functional modules composed of computer program segments. The computer programs for each program segment in the intelligent gas leak monitoring device 2 may be stored in the server's memory and executed by at least one processor to perform (see details). Figure 1 (Description) Functions of the digital gas leak monitoring method.

[0061] In this embodiment, the intelligent gas leak monitoring device 2 can be divided into multiple functional modules according to its functions. These functional modules may include: a filtering and noise reduction module 21, a signal trend module 22, a feature fusion module 23, a classification and evaluation module 24, a location estimation module 25, and a risk decision-making module 26. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, stored in memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0062] The filtering and denoising module 21 is used to filter and denoise the raw sensor data stream acquired in real time to obtain a purified sensor dataset. Signal trend module 22 is used to perform signal decomposition and trend analysis on the purification sensor dataset to obtain the intrinsic mode component set and trend baseline signal. Feature fusion module 23 is used to perform multi-dimensional feature fusion on the intrinsic mode component set and the trend baseline signal to generate a high-dimensional feature set; The classification and evaluation module 24 is used to classify and evaluate the high-dimensional feature set according to the preset leakage state type to obtain leakage state probability distribution data. The location estimation module 25 is used to locate the leak source and estimate the leak intensity based on the leak state probability distribution data, and obtain the pipeline coordinates of the leak point and the corresponding estimated leak intensity value. The risk decision module 26 is used to perform risk assessment and response decision based on the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimate, and generate inspection instructions and corresponding early warning levels.

[0063] In an optional implementation, the filtering and denoising module 21 is used for: The raw sensor data stream acquired in real time is processed in the time domain to obtain a synchronized sensor dataset. Based on the preset sensor deployment data, the synchronous sensing dataset is spatially mapped using a preset spatial coordinate system to obtain a spatiotemporally aligned dataset. The pressure sensor dataset in the spatiotemporally aligned dataset is state estimated using a preset unscented Kalman filter algorithm to generate purification pressure data. Frequency band thresholds are defined based on the spectral characteristics of the ultrasonic sensing dataset in the spatiotemporal aligned dataset. Adaptive wavelet packet denoising is then performed on the ultrasonic sensing dataset based on the frequency band thresholds to generate purified ultrasonic data.

[0064] In an optional implementation, the signal trend module 22 is used for: The distribution of dominant frequency components is analyzed on a pre-defined historical sensor dataset to determine the optimal combination of decomposition parameters for each dataset. Based on the optimal decomposition parameter combination, variational mode decomposition is performed on the purification pressure data and the purification ultrasonic data to obtain a subset of pressure intrinsic mode components and a subset of ultrasonic intrinsic mode components. According to a preset frequency threshold, components with frequencies lower than the frequency threshold are selected from the subset of pressure intrinsic mode components as a low-frequency mode component set. The low-frequency modal component set is fitted into a corresponding trend baseline signal using a preset regularization fitting algorithm.

[0065] In an optional implementation, the feature fusion module 23 is used to: Time-frequency feature extraction is performed on the intrinsic mode component set to obtain the corresponding time-frequency feature set; The trend baseline signal is dynamically changed to obtain the corresponding trend change feature set. The intrinsic modal component set is subjected to temporal correlation change features between different modes to obtain a correlation feature set; Principal component analysis is performed on the time-frequency feature set, the trend change feature set, and the correlation feature set to obtain each principal component feature set; The importance of leakage identification is evaluated on the principal component feature set to obtain the corresponding identification association score, and the identification association score is converted into the corresponding weight coefficient according to the preset weight mapping rule; The principal component feature set is weighted and fused according to the weight coefficients to generate a high-dimensional feature set.

[0066] In an optional implementation, the classification evaluation module 24 is used to: Multi-scale local feature extraction is performed on the high-dimensional feature set to obtain a deep feature set; The importance of the deep feature set is evaluated through a preset attention mechanism to calculate the importance weight of each deep feature. Based on the preset leakage state type and the importance weight, the deep feature set is classified and a leakage state probability distribution data is obtained.

[0067] In an optional implementation, the positioning estimation module 25 is used for: Using a preset generalized cross-correlation algorithm, the time difference of arrival of the purified ultrasonic data is calculated based on the sensor deployment data to obtain the set of time differences of arrival of the signals between each sensor. Using a preset particle swarm optimization algorithm, the leak source is located based on the leak state probability distribution data and the signal arrival time difference set, and the pipe coordinates of the leak point are obtained. Based on the pipeline coordinates and the signal arrival time difference set, ultrasonic energy attenuation inversion is performed to obtain the corresponding leakage intensity estimate.

[0068] In an optional implementation, the risk decision module 26 is used to: By using a preset coordinate feature database, the corresponding environmental sensitivity data is queried based on the pipeline coordinates; By using preset multi-criteria fuzzy inference rules, the leakage state probability distribution data, the environmental sensitivity data, and the leakage intensity estimate are mapped to a set of fuzzy linguistic variables; The fuzzy language variable set is converted into a risk level fuzzy set by a preset safety procedure library, and the risk level fuzzy set is defuzzified to obtain a comprehensive risk score. The comprehensive risk score is converted into a corresponding warning level according to the preset scoring and level mapping rules, and the warning level, the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimate are encapsulated into an inspection command through a preset command generation method.

[0069] It should be understood that the various variations and specific embodiments of the methods provided in the above embodiments are also applicable to the digital intelligent gas leak monitoring device of this embodiment. Through the foregoing detailed description of the digital intelligent gas leak monitoring method, those skilled in the art can clearly understand the implementation method of the digital intelligent gas leak monitoring device in this embodiment. For the sake of brevity, it will not be described in detail here.

[0070] like Figure 3 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application.

[0071] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to, a memory 31, at least one processor 32, and at least one communication bus 33.

[0072] Those skilled in the art should understand that Figure 3 The structure of the electronic device 3 shown does not constitute a limitation of the embodiments of the present invention. The electronic device 3 may also include more or fewer other hardware or software than shown, or different component arrangements.

[0073] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, and embedded devices.

[0074] It should be noted that the electronic device 3 is merely an example. Other existing or future electronic products that are suitable for this application should also be included within the scope of protection of this application and are incorporated herein by reference.

[0075] In some embodiments, the memory 31 stores a computer program that, when executed by the at least one processor 32, implements all or part of the steps in the digital intelligent gas leak monitoring method described above. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store an operating system, at least one application program required for a function, etc.

[0076] In some embodiments, the at least one processor 32 is the control unit of the electronic device 3, connecting various components of the electronic device 3 via various interfaces and lines. It executes programs or modules stored in the memory 31 and calls data stored in the memory 31 to perform various functions and process data. For example, when the at least one processor 32 executes the computer program stored in the memory 31, it implements all or part of the steps of the intelligent gas leak monitoring method described in this application embodiment; or it implements all or part of the functions of the intelligent gas leak monitoring device. The at least one processor 32 may be composed of integrated circuits, such as a single-packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0077] In some embodiments, the at least one communication bus 33 is configured to enable communication between the memory 31 and the at least one processor 32, etc. Although not shown, the electronic device 3 may also include a power supply (e.g., a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 32 via a power management device, thereby enabling functions such as charging, discharging, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device 3 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0078] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause an electronic device (which may be a personal computer, electronic device, or network device, etc.) or processor to execute portions of the methods described in the various embodiments of this application.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0080] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0081] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A digital and intelligent gas leakage monitoring method, characterized in that, The method comprises: filtering and denoising the real-time collected original sensor data stream to obtain a purified sensor data set, performing signal decomposition and trend analysis on the purified sensor data set to obtain an intrinsic modal component set and a trend baseline signal; performing multi-dimensional feature fusion on the intrinsic modal component set and the trend baseline signal to generate a high-dimensional feature set; classifying and evaluating the high-dimensional feature set according to a preset leakage state type to obtain leakage state probability distribution data; locating a leakage source and inferring a leakage intensity according to the leakage state probability distribution data to obtain a pipeline coordinate of a leakage point and a corresponding leakage intensity estimate value; performing risk assessment and response decision-making according to the leakage state probability distribution data, the pipeline coordinate and the leakage intensity estimate value to generate an inspection instruction and a corresponding warning level.

2. The smart gas leak monitoring method as claimed in claim 1, wherein the purged sensor data set comprises purged pressure data and purged ultrasonic data. The filtering and denoising of the real-time collected original sensor data stream to obtain a purified sensor data set comprises: performing time domain synchronization processing on the real-time collected original sensor data stream to obtain a synchronized sensor data set; mapping the synchronized sensor data set to a spatial coordinate system according to preset sensor layout data to obtain a space-time aligned data set; performing state estimation on the pressure sensor data set in the space-time aligned data set by a preset unscented Kalman filter algorithm to generate purified pressure data; dividing frequency band thresholds according to the frequency spectrum characteristics of the ultrasonic sensor data set in the space-time aligned data set, and performing adaptive wavelet packet denoising on the ultrasonic sensor data set according to the frequency band thresholds to generate purified ultrasonic wave data.

3. The cognitive gas leakage monitoring method of claim 2, the set of intrinsic modal components comprising a subset of pressure intrinsic modal components and a subset of ultrasonic intrinsic modal components, characterized in that, The signal decomposition and trend analysis on the purified sensor data set to obtain an intrinsic modal component set and a trend baseline signal comprises: analyzing the dominant frequency component distribution of a preset historical sensor data set to determine the optimal decomposition parameter combination corresponding to each data; performing variational modal decomposition on the purified pressure data and the purified ultrasonic wave data according to the optimal decomposition parameter combination to obtain a pressure intrinsic modal component subset and an ultrasonic wave intrinsic modal component subset; selecting components with frequencies lower than a preset frequency threshold from the pressure intrinsic modal component subset as a low-frequency modal component set; fitting the low-frequency modal component set into a corresponding trend baseline signal by a preset regularization fitting algorithm.

4. The smart gas leak monitoring method as claimed in claim 1, wherein, The multi-dimensional feature fusion on the intrinsic modal component set and the trend baseline signal to generate a high-dimensional feature set comprises: performing time domain-frequency domain feature extraction on the intrinsic modal component set to obtain a corresponding time-frequency feature set; performing dynamic change feature calculation on the trend baseline signal to obtain a corresponding trend change feature set; performing time sequence correlation change feature extraction between different modal components of the intrinsic modal component set to obtain a correlation feature set; performing principal component analysis on the time-frequency feature set, the trend change feature set and the correlation feature set to obtain a principal component feature set; The importance evaluation of the leakage identification on the principal component feature set obtains corresponding identification correlation scores, and the identification correlation scores are converted into corresponding weight coefficients according to a preset weight mapping rule; The principal component feature set is weighted and fused according to the weight coefficients, and a high-dimensional feature set is generated.

5. The smart gas leak monitoring method as claimed in claim 1, wherein, The classification evaluation of the high-dimensional feature set according to the preset leakage state type obtains leakage state probability distribution data, which includes: Multi-scale local feature extraction is performed on the high-dimensional feature set to obtain a deep feature set; The importance of the deep feature set is evaluated through a preset attention mechanism to calculate the importance weight corresponding to each deep feature; The deep feature set is classified and decided according to the preset leakage state type and the importance weight, and leakage state probability distribution data is obtained.

6. The smart gas leak monitoring method as claimed in claim 2, wherein, The positioning of the leakage source and the speculation of the leakage intensity according to the leakage state probability distribution data obtain the pipeline coordinates of the leakage point and the corresponding leakage intensity estimate value, which includes: The signal arrival time difference set between sensors is obtained by calculating the signal arrival time difference of the purified ultrasonic wave data according to the sensor layout data through a preset generalized cross-correlation algorithm; The pipeline coordinates of the leakage point are obtained by positioning the leakage source according to the leakage state probability distribution data and the signal arrival time difference set through a preset particle swarm optimization algorithm; The leakage intensity estimate value is obtained by ultrasonic energy attenuation inversion speculation according to the pipeline coordinates and the signal arrival time difference set.

7. The smart gas leak monitoring method as claimed in claim 1, wherein, The risk assessment and response decision according to the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimate value generate the inspection instruction and the corresponding warning level, which includes: The corresponding environmental sensitivity data is queried according to the pipeline coordinates through a preset coordinate feature database; The leakage state probability distribution data, the environmental sensitivity data and the leakage intensity estimate value are mapped into a fuzzy language variable set through a preset multi-criteria fuzzy reasoning rule; The risk level fuzzy set is converted from the fuzzy language variable set through a preset safety procedure library, and the risk level fuzzy set is de-fuzzified to obtain a comprehensive risk score; The comprehensive risk score is converted into a corresponding warning level according to a preset score and level mapping rule, and the warning level, the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimate value are packaged into an inspection instruction through a preset instruction generation method.

8. A digital and intelligent gas leakage monitoring device applied to the digital and intelligent gas leakage monitoring method of claim 1, characterized in that, The device includes: A filtering and denoising module for filtering and denoising the real-time collected original sensor data stream to obtain a purified sensor data set; A signal trend module for signal decomposition and trend analysis on the purified sensor data set to obtain a set of intrinsic mode components and a trend baseline signal; A feature fusion module for multi-dimensional feature fusion on the set of intrinsic mode components and the trend baseline signal to generate a high-dimensional feature set; A classification evaluation module for classification evaluation of the high-dimensional feature set according to a preset leakage state type to obtain leakage state probability distribution data; a positioning and estimation module configured to perform positioning of the leakage source and estimation of the leakage intensity according to the leakage state probability distribution data, and obtain pipeline coordinates of the leakage point and a corresponding leakage intensity estimation value; a risk decision module configured to perform risk assessment and response decision according to the leakage state probability distribution data, the pipeline coordinates and the leakage intensity estimation value, and generate an inspection instruction and a corresponding warning level.

9. An electronic device, comprising: The electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the digital gas leakage monitoring method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the digital gas leakage monitoring method according to any one of claims 1 to 7.

Images