A DAS-based chemical gas leakage safety detection method and system

Abstract

The application discloses a chemical gas leakage safety detection method and system based on DAS, and relates to the technical field of leakage detection; global signal data of a target area is acquired, safety detection is carried out according to the global signal data to obtain an abnormal detection result; a sub-area is determined according to the abnormal detection result, and local signal data of the sub-area is acquired; the local signal data is detected to obtain a local detection result, and safety early warning is carried out according to the local detection result. Global signals are collected through DAS, distributed characteristics are utilized to quickly identify abnormal pipeline areas, the detection workload is reduced, and the response efficiency is improved; DAS and infrasonic wave sensors are coordinated, the former provides an abnormal direction, the latter focuses on local details, and micro-leakage characteristics are accurately captured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of leakage detection, and particularly relates to a chemical gas leakage safety detection method and system based on DAS. BACKGROUND

[0002] With the rapid development of the chemical industry, the chemical gas detection technology has undergone a major change from simple qualitative to precise quantitative. From the early chemical reagent test paper detection to the current high-precision sensor and intelligent analysis system, the detection efficiency and accuracy have been greatly improved, providing a solid guarantee for chemical safety and environmental protection.

[0003] Patent No. CN119915446A discloses a gas pipeline leakage acoustic detection device and method based on CPO-VMD and multi-feature extraction; acoustic signals of gas pipeline leakage are collected; the original signals are adaptively decomposed by combining the CPO algorithm and the variational mode decomposition; the key modal components are reconstructed and denoised by the correlation coefficient; the multi-dimensional features of the reconstructed signals are extracted, the signal features are selected and optimally selected, and the detection and classification of the gas pipeline leakage are performed by the classification detector.

[0004] The existing chemical pipeline gas leakage detection method has problems such as low detection efficiency, large workload, inability to quickly locate abnormal areas, and difficulty in accurately capturing micro-leakage characteristics, which cannot meet the efficient and accurate leakage detection requirements. SUMMARY

[0005] The purpose of the present application is to solve the problems of the existing chemical pipeline gas leakage detection method, such as low detection efficiency, large workload, inability to quickly locate abnormal areas, and difficulty in accurately capturing micro-leakage characteristics, which cannot meet the efficient and accurate leakage detection requirements, and to propose a chemical gas leakage safety detection method and system based on DAS.

[0006] In the first aspect of the present application, a chemical gas leakage safety detection method based on DAS is first proposed, which comprises:

[0007] Obtaining global signal data of a target area, and performing safety detection according to the global signal data to obtain an abnormal detection result;

[0008] Determining a sub-area according to the abnormal detection result, and obtaining local signal data of the sub-area; the target area contains multiple sub-areas;

[0009] Performing detection on the local signal data to obtain a local detection result, and performing safety warning according to the local detection result.

[0010] Optionally, before the safety detection according to the global signal data to obtain the abnormal detection result, the method further comprises:

[0011] obtain historical data, pre-process the historical data to obtain target historical data, determine the length of a sliding window, and segment the target historical data according to the sliding window to obtain a plurality of signal segments; the historical data is a leakage sample signal under different pressure and leakage aperture conditions;

[0012] extract each signal segment to obtain a plurality of upper envelope lines, perform quadratic Gaussian smoothing and Lagrange downsampling processing on each upper envelope line to obtain a plurality of target upper envelope lines, and construct a sample library according to each target upper envelope line; the signal segment and the upper envelope line correspond one-to-one.

[0013] Optionally, the safety detection is performed according to the global signal data to obtain an anomaly detection result, which includes:

[0014] Step one, obtain a plurality of intrinsic mode functions by variational mode decomposition of the global signal data, determine the frequency of each intrinsic mode function, and reconstruct the intrinsic mode function with a frequency less than a preset frequency to obtain a noise reduction signal;

[0015] Step two, segment the noise reduction signal according to a sliding window to obtain a plurality of segmented signals, and perform a target operation on each segmented signal to obtain a real-time signal; the target operation includes upper envelope extraction, quadratic Gaussian smoothing and Lagrange downsampling operation, and the real-time signal includes x1 and x2;

[0016] Step three, match the real-time signal according to the sample library to obtain minimum cumulative distance matrices D1 and D2, calculate the total cumulative distance matrix according to D1 and D2, and determine the matching signal according to the total cumulative distance matrix;

[0017] Step four, determine the step length of the sliding window, repeatedly perform steps two and three, and update the real-time signal and the matching signal to obtain a signal sequence; the signal sequence contains a plurality of matching signals;

[0018] Step five, set a leakage warning threshold, if the matching signals in the signal sequence are all less than the leakage warning threshold, add an anomaly label to the global signal data to obtain an anomaly detection result.

[0019] Optionally, the local signal data is detected to obtain a local detection result, which includes:

[0020] convert the local signal data into a local frequency spectrum by short-time Fourier transform, and extract features of the local frequency spectrum by a target YOLOv5 model to obtain a local detection result;

[0021] The improvement of the target YOLOv5 model includes:

[0022] replacing a C3 module in a YOLOv5 model with a target C3 module and replacing a neck structure of the YOLOv5 model with a target neck structure to obtain a target YOLOv5 model;

[0023] The working principle of the target C3 module includes:

[0024] obtaining an initial feature map, inputting the initial feature map into a CBS module to obtain a first feature map, inputting the first feature map into a PConv module, a CBS module and a CBS module in sequence to obtain a second feature map, and adding the first feature map and the second feature map to obtain a third feature map;

[0025] inputting the initial feature map into a CBS module to obtain a fourth feature map, adding the third feature map and the fourth feature map to obtain a fifth feature map, inputting the fifth feature map into a CBS module to obtain a sixth feature map, and inputting the sixth feature map into four SPAB modules in sequence to obtain a seventh feature map.

[0026] Optionally, the working principle of the target neck structure includes:

[0027] obtaining a feature tensor set of a backbone network, inputting a first feature tensor into a DBS module to obtain a fourth feature tensor, upsampling the fourth tensor to obtain a target fourth tensor, and splicing the fourth feature tensor, the target fourth feature tensor and a second feature tensor to obtain a fifth feature tensor; the feature tensor set includes the first feature tensor, the second feature tensor and a third feature tensor;

[0028] inputting the fifth feature tensor into a target C3 module and a DBS module in sequence to obtain a sixth feature tensor, upsampling the sixth feature tensor to obtain a target sixth feature tensor, and splicing the third feature tensor and the target sixth feature tensor to obtain a seventh feature tensor;

[0029] inputting the seventh feature tensor into a target CSPStage module to obtain an eighth feature tensor, inputting the eighth feature tensor into a CBS module to obtain a ninth feature tensor, splicing the ninth feature tensor and the sixth feature tensor to obtain a tenth feature tensor, inputting the tenth feature tensor into the target CSPStage module to obtain a first target output, and taking the eighth feature tensor as a second target output;

[0030] The working process of the target CSPStage module includes:

[0031] An initial target feature map is acquired, the initial target feature map is input into a 1x1Conv module to obtain a first target feature map, the first target feature map is input into a target Conv module to obtain a second target feature map, the second target feature map is sequentially input into a Biformer module and a 3x3Conv module to obtain a third target feature map, the first target feature map and the third target feature map are added to obtain a fourth target feature map, the fourth target feature map and the first target feature map are spliced to obtain a fifth target feature map, the fifth target feature map is input into a 1x1Conv module to obtain a sixth target feature map, and the sixth target feature map is taken as the output of a target CSPStage module.

[0032] In a second aspect of the embodiment of the present application, a DAS-based chemical gas leakage safety detection system is provided, comprising a global detection module, a local detection module and a safety warning module.

[0033] The global detection module is configured to acquire global signal data of a target area, and perform safety detection based on the global signal data to obtain an abnormal detection result.

[0034] The local detection module is configured to determine a sub-area based on the abnormal detection result, and acquire local signal data of the sub-area; the target area comprises a plurality of sub-areas.

[0035] The safety warning module is configured to detect the local signal data to obtain a local detection result, and perform safety warning based on the local detection result.

[0036] Optionally, the system further comprises a data acquisition module and a sample library construction module.

[0037] The data acquisition module is configured to acquire historical data, pre-process the historical data to obtain target historical data, determine the length of a sliding window, and segment the target historical data into a plurality of signal segments based on the sliding window; the historical data is a leakage sample signal under different pressure and leakage aperture conditions.

[0038] The sample library construction module is configured to extract a plurality of upper envelope lines from the signal segments, perform quadratic Gaussian smoothing and Lagrange decimation processing on the upper envelope lines to obtain a plurality of target upper envelope lines, and construct a sample library based on the target upper envelope lines; the signal segments and the upper envelope lines are in one-to-one correspondence.

[0039] Optionally, the global detection module comprises a first operation module, a second operation module, a third operation module, a fourth operation module and a fifth operation module.

[0040] The first operation module is configured to obtain multiple intrinsic mode functions by variational modal decomposition on the global signal data, determine frequencies of the intrinsic mode functions, and reconstruct intrinsic mode functions with frequencies less than a preset frequency to obtain a denoised signal.

[0041] The second operation module is configured to segment the denoised signal into multiple segmented signals according to a sliding window, and perform a target operation on each segmented signal to obtain a real-time signal; the target operation includes envelope extraction, secondary Gaussian smoothing, and Lagrange downsampling operation; and the real-time signal includes x1 and x2.

[0042] The third operation module is configured to match the real-time signal with the sample library to obtain minimum cumulative distance matrices D1 and D2, calculate a total cumulative distance matrix according to D1 and D2, and determine a matching signal according to the total cumulative distance matrix.

[0043] The fourth operation module is configured to determine a step size of the sliding window, repeatedly perform the second operation module and the third operation module, and update the real-time signal and the matching signal to obtain a signal sequence; the signal sequence includes multiple matching signals.

[0044] The fifth operation module is configured to set a leakage warning threshold, and if all the matching signals in the signal sequence are less than the leakage warning threshold, add an anomaly label to the global signal data to obtain an anomaly detection result.

[0045] Optionally, the local detection module is further configured to convert local signal data into a local frequency spectrum by short-time Fourier transform, and extract features of the local frequency spectrum by a target YOLOv5 model to obtain a local detection result.

[0046] Improvements of the target YOLOv5 model include:

[0047] The C3 module in the YOLOv5 model is replaced by a target C3 module, and a neck structure of the YOLOv5 model is replaced by a target neck structure to obtain the target YOLOv5 model.

[0048] The working principle of the target C3 module includes:

[0049] An initial feature map is obtained, the initial feature map is input into a CBS module to obtain a first feature map, the first feature map is sequentially input into a PConv module, the CBS module, and the CBS module to obtain a second feature map, and the first feature map and the second feature map are added to obtain a third feature map.

[0050] Input the initial feature map into a CBS module to obtain a fourth feature map, add the third feature map and the fourth feature map to obtain a fifth feature map, input the fifth feature map into a CBS module to obtain a sixth feature map, and input the sixth feature map into four SPAB modules in sequence to obtain a seventh feature map.

[0051] Optionally, the working principle of the target neck structure comprises:

[0052] Obtain a feature tensor set of the backbone network, input a first feature tensor into a DBS module to obtain a fourth feature tensor, up-sample the fourth tensor to obtain a target fourth tensor, and splice the fourth feature tensor, the target fourth feature tensor and a second feature tensor to obtain a fifth feature tensor; the feature tensor set comprises: the first feature tensor, the second feature tensor and a third feature tensor;

[0053] Input the fifth feature tensor into a target C3 module and a DBS module in sequence to obtain a sixth feature tensor, up-sample the sixth feature tensor to obtain a target sixth feature tensor, and splice the third feature tensor and the target sixth feature tensor to obtain a seventh feature tensor;

[0054] Input the seventh feature tensor into a target CSPStage module to obtain an eighth feature tensor, input the eighth feature tensor into a CBS module to obtain a ninth feature tensor, splice the ninth feature tensor and the sixth feature tensor to obtain a tenth feature tensor, input the tenth feature tensor into the target CSPStage module to obtain a first target output, and take the eighth feature tensor as a second target output;

[0055] The working process of the target CSPStage module comprises:

[0056] Obtain an initial target feature map, input the initial target feature map into a 1x1Conv module to obtain a first target feature map, input the first target feature map into a target Conv module to obtain a second target feature map, input the second target feature map into a Biformer module and a 3x3Conv module in sequence to obtain a third target feature map, add the first target feature map and the third target feature map to obtain a fourth target feature map, splice the fourth target feature map and the first target feature map to obtain a fifth target feature map, input the fifth target feature map into a 1x1Conv module to obtain a sixth target feature map, and take the sixth target feature map as the output of the target CSPStage module.

[0057] The beneficial effects of the present application are:

[0058] The application provides a chemical gas leakage safety detection method based on DAS, which comprises the following steps: acquiring global signal data of a target area, performing safety detection on the global signal data to obtain an abnormality detection result, determining a sub-area according to the abnormality detection result, acquiring local signal data of the sub-area, detecting the local signal data to obtain a local detection result, and performing safety early warning according to the local detection result. BRIEF DESCRIPTION OF DRAWINGS

[0059] The application will be further described below with reference to the drawings.

[0060] Figure 1 A flowchart of a chemical gas leakage safety detection method based on DAS provided by the embodiment of the application is shown in the figure.

[0061] Figure 2 A target C3 module structure schematic diagram of another chemical gas leakage safety detection method based on DAS provided by the embodiment of the application is shown in the figure.

[0062] Figure 3 A target neck structure schematic diagram of another chemical gas leakage safety detection method based on DAS provided by the embodiment of the application is shown in the figure.

[0063] Figure 4 A framework diagram of a chemical gas leakage safety detection system based on DAS provided by the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, but not all the embodiments of the application. In this document, the term "and / or" is only used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, A and B can mean that A exists alone, A and B exist together, and B exists alone. In addition, in the application, the description of "first", "second", etc. is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it, and when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope required by the application.

[0065] All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative labor fall within the scope of the present application.

[0066] The embodiments of the present application provide a DAS-based chemical gas leakage safety detection method. Figure 1 Figure 1 A flowchart of the DAS-based chemical gas leakage safety detection method is provided in the embodiments of the present application. The method comprises the following steps:

[0067] In S101, global signal data of a target area is acquired, and safety detection is performed according to the global signal data to obtain an abnormality detection result.

[0068] In S102, a sub-area is determined according to the abnormality detection result, and local signal data of the sub-area is acquired.

[0069] In S103, local detection is performed on the local signal data to obtain a local detection result, and safety early warning is performed according to the local detection result.

[0070] The target area comprises a plurality of sub-areas.

[0071] The DAS-based chemical gas leakage safety detection method provided in the embodiments of the present application collects global signals by DAS, quickly identifies abnormal pipeline areas by using the distributed characteristics, reduces the detection workload, and improves the response efficiency; DAS and infrasonic wave sensors are coordinated, the former provides the abnormal direction, the latter focuses on local details, and the micro-leakage characteristics are accurately captured.

[0072] In an implementation manner, the global signal is collected by DAS, and the local signal is collected by the infrasonic wave sensor; the DAS is deployed on all pipelines, and the infrasonic wave sensor is deployed on the pipeline at a preset distance (for example, an interval of five meters).

[0073] In an implementation manner, the global signal data of the target area is collected by DAS, and the distributed optical fiber sensing characteristics (covering the whole pipeline, resisting electromagnetic interference, and long-distance real-time monitoring) are used to perform blind area-free scanning on the overall operation state of the pipeline. The existence area of the abnormal signal (such as vibration / temperature abnormality caused by leakage) is quickly identified, the monitoring range is focused from “full-line investigation” to “suspected sub-area”, the workload of subsequent local detection is greatly reduced, and the system response efficiency is improved. For example, DAS can locate an abnormal interval of 10 meters in a pipeline of several kilometers, and the blindness of manual inspection is avoided.

[0074] ​In one implementation, the global detection filters noise: the distributed nature of DAS can eliminate environmental interference (such as accidental vibration) through the law of signal space-time distribution, and only trigger sub-area detection for persistent abnormal signals; the local detection focuses on details: infrasound sensors are deployed at high density in sub-areas (such as 5-meter intervals), which makes up for the insufficient resolution of DAS in detecting local high-frequency signals (such as low-frequency vibrations with micro-leakage), and avoids missed detection caused by signal attenuation or noise of a single technology.

[0075] In one implementation, the advantages of DAS: long-distance, non-contact monitoring, suitable for identifying high-frequency vibrations or temperature changes caused by high-pressure leakage, and quickly locking abnormal sections; the advantages of infrasound sensors: single-point high sensitivity (especially for <20Hz infrasound), local signal high-density sampling under dense deployment, suitable for capturing low-frequency characteristics of low-pressure micro-leakage or air-borne infrasound signals; the synergistic value: global signals provide abnormal "azimuth angle", and local signals provide "precise coordinates".

[0076] In one implementation, based on the local detection result (high-precision signal analysis of infrasound sensors), a safety warning is triggered, and the warning information contains "specific sub-area where the leakage occurs + detailed signal characteristics" (such as leakage aperture, pressure level); according to the severity of the local detection result (such as signal amplitude, duration), the warning level is dynamically adjusted (such as first-level warning for micro-leakage and second-level warning for high-pressure jet leakage).

[0077] In one embodiment, before obtaining the abnormal detection result according to the global signal data, the method further comprises:

[0078] Obtaining historical data, pre-processing the historical data to obtain target historical data, determining the length of a sliding window, and segmenting the target historical data according to the sliding window to obtain a plurality of signal segments; the historical data are leakage sample signals under different pressure and leakage aperture conditions;

[0079] Extracting each signal segment to obtain a plurality of upper envelope lines, performing quadratic Gaussian smoothing and Lagrange decimation processing on each upper envelope line to obtain a plurality of target upper envelope lines, and constructing a sample library according to each target upper envelope line; the signal segments and the upper envelope lines are one-to-one corresponding.

[0080] In one implementation, the sliding window: W = L * f / v, where L represents the total length of the pipeline, v represents the propagation speed of the leakage signal, and f represents the sampling frequency of the signal; all pipeline leakage signals are obtained by DAS periodically (i.e. sampling frequency).

[0081] In one implementation, preprocessing (such as noise reduction, filtering, etc.) of historical data can eliminate noise and interference components (such as environmental vibration, equipment operation noise) in the original signal, ensuring that the target historical data only retains effective leakage characteristics. By focusing on samples under different pressure and leakage aperture conditions, a variety of leakage scenarios can be covered, making the subsequently constructed sample library have stronger generalization ability and improving the adaptability of the leakage detection model to complex working conditions.

[0082] In one implementation, the sample library is constructed based on leakage signals under different pressure, leakage aperture, etc. conditions, and contains waveform characteristics of a variety of typical leakage scenarios. This enables the system to recognize leakage signals of different severity and in different environments by matching the diversified templates in the sample library during real-time monitoring, improving the generalization ability of the detection model and avoiding missed detection or false detection due to single working condition. For example, the signal characteristics of high-pressure small-aperture leakage and low-pressure large-aperture leakage can be effectively distinguished by corresponding templates in the sample library.

[0083] In one embodiment, the anomaly detection result is obtained according to the safety detection based on the global signal data, including:

[0084] Step one, the global signal data is decomposed into a plurality of intrinsic mode functions by variational mode decomposition, the frequency of each intrinsic mode function is determined, and the intrinsic mode functions with a frequency less than a preset frequency are reconstructed to obtain a denoised signal;

[0085] Step two, the denoised signal is segmented into a plurality of segmented signals according to a sliding window, and a target operation is performed on each segmented signal to obtain a real-time signal; the target operation includes envelope extraction, quadratic Gaussian smoothing and Lagrange downsampling operation, and the real-time signal includes x1 and x2;

[0086] Step three, the real-time signal is matched according to the sample library to obtain minimum cumulative distance matrices D1 and D2, the total cumulative distance matrix is calculated according to D1 and D2, and the matching signal is determined according to the total cumulative distance matrix;

[0087] Step four, the step length of the sliding window is determined, steps two and three are repeatedly executed, and the real-time signal and the matching signal are updated to obtain a signal sequence; the signal sequence contains a plurality of matching signals;

[0088] Step five, a leakage warning threshold is set, if all the matching signals in the signal sequence are less than the leakage warning threshold, an anomaly label is added to the global signal data to obtain an anomaly detection result.

[0089] In one implementation, the variational mode decomposition (VMD) decomposes a global signal into multiple intrinsic mode functions (IMFs) by constructing and solving a variational model, avoiding the mode mixing problem of traditional empirical mode decomposition (EMD). For the infrasonic wave component (usually below 20 Hz) in the leakage signal, the IMFs with a dominant frequency less than a preset frequency (e.g., 15 Hz) are selected for reconstruction, which can effectively separate the low-frequency leakage characteristics from high-frequency noise (e.g., pipeline vibration, environmental interference), and improve the signal-to-noise ratio. The signal in the infrasonic frequency band directly related to the leakage is retained, and the irrelevant high-frequency noise is suppressed, providing a pure feature input for subsequent analysis.

[0090] In one implementation, the noise-reduced signal is segmented by a sliding window to ensure that each segmented signal contains complete leakage characteristics (such as bidirectional stress waves). The upper envelope extraction can highlight the amplitude variation trend of the signal, and the quadratic Gaussian smoothing can further filter out local noise. The Lagrange downsampling reduces the data dimension while maintaining the waveform profile (e.g., reducing the high-frequency sampling signal to the characteristic frequency band). The original time-domain signal is converted into a feature vector with noise resistance (e.g., x1, x2), reducing the subsequent matching calculation amount and improving the real-time performance.

[0091] In one implementation, the leakage signals of different working conditions (pressure, aperture) in the sample library are used as templates, and the minimum cumulative distance between the real-time signal and the sample is calculated by DTW, which can solve the time axis stretching problem caused by the propagation distance and medium difference of the leakage signal (such as the time delay difference of the signal arriving at different sensors). This implementation achieves accurate matching of non-aligned time series and improves the recognition accuracy of leakage signals under complex working conditions.

[0092] In one implementation, the sliding interval of the window is controlled by setting the step size ΔW, ensuring signal continuity (such as overlapping segmentation to avoid missed detection), and achieving real-time scanning of the entire pipeline. The real-time signal and the matching result are continuously updated during the iteration process, which can capture the dynamic changes of the leakage signal (such as waveform evolution caused by pressure decay). This implementation balances the monitoring real-time performance and calculation efficiency, ensuring the coverage and signal integrity of long-distance pipeline monitoring.

[0093] In one embodiment, the local signal data is detected to obtain a local detection result, including:

[0094] The local signal data is converted into a local frequency spectrum by short-time Fourier transform, and the local frequency spectrum is subjected to feature extraction by a target YOLOv5 model to obtain a local detection result;

[0095] The improvement of the target YOLOv5 model includes:

[0096] The C3 module in the YOLOv5 model is replaced by the target C3 module, and the neck structure of the YOLOv5 model is replaced by the target neck structure to obtain the target YOLOv5 model;

[0097] The working principle of the target C3 module comprises:

[0098] The initial feature map is obtained, the initial feature map is input into the CBS module to obtain the first feature map, the first feature map is sequentially input into the PConv module, the CBS module and the CBS module to obtain the second feature map, and the first feature map and the second feature map are added to obtain the third feature map.

[0099] The initial feature map is input into the CBS module to obtain the fourth feature map, the third feature map and the fourth feature map are added to obtain the fifth feature map, the fifth feature map is input into the CBS module to obtain the sixth feature map, and the sixth feature map is sequentially input into the four SPAB modules to obtain the seventh feature map. Figure 2 , Figure 2 The target C3 module structure diagram of another DAS-based chemical gas leakage safety detection method provided by the embodiment is shown in the figure, wherein "⊕" represents addition.

[0100] In an implementation manner, the original signal is converted into a frequency spectrum graph through short-time Fourier transform, and the frequency spectrum graph is subjected to feature extraction through the YOLOv5 model; the original signal is subjected to feature capture through the YOLOv5 model, and the accuracy of YOLOv7 and YOLOv8 is higher than that of YOLOv5, but the larger model size, higher calculation complexity and resource utilization rate make it difficult to meet the real-time requirement.

[0101] In an implementation manner, the local signal data is converted into a local frequency spectrum graph through short-time Fourier transform (STFT), and the frequency spectrum graph is subjected to feature extraction through the target YOLOv5 model to obtain a local detection result, which can improve the accuracy and robustness of gas leakage detection. Short-time Fourier transform is a time-frequency analysis method, which decomposes the signal into frequency spectrum components in multiple short time, and can capture the dynamic changes of the signal in time and frequency dimensions at the same time. This conversion mode makes the originally one-dimensional signal data present rich two-dimensional features in the frequency spectrum graph, so that the YOLOv5 model can quickly and accurately extract key features from the frequency spectrum graph and identify the characteristic mode of gas leakage, improve the classification accuracy, and meet the real-time detection requirement.

[0102] In an implementation manner, by introducing the PConv module and replacing the traditional Bottleneck module, the calculation complexity and memory occupation can be reduced, and the CBS module is combined to effectively reduce the memory consumption while maintaining feature diversity, realizing lightweight feature extraction and improving the real-time performance of the model.

[0103] In an implementation, in the feature extraction process, the diversity of the features is preserved by adding the feature map processed by the PConv and CBS modules to the initial feature map. This residual connection method ensures the integrity of the feature information and avoids information loss due to lightweight processing, thereby maintaining the detection performance of the model while reducing the computational cost.

[0104] In an implementation, the SPAB structure is introduced at the end of feature extraction, and four consecutive SPAB modules are further used to enhance the spatial expression ability of the features. The SPAB module uses an activation function and a residual connection to significantly improve the quality of feature information without additional parameters, while suppressing the interference of redundant noise and improving the capture and expression ability of the model for gas leakage signal features.

[0105] In an implementation, referring to Figure 3 , Figure 3 Another DAS-based safety detection method for chemical gas leakage provided by an embodiment of the present application is provided. The DBS module is composed of a DWConv module, a batch normalization layer (Batch Normalization), and an activation function (such as ReLU or SiLU). The CBS module is composed of a convolutional layer (Convolutional Layer), a batch normalization layer (Batch Normalization), and an activation function (such as ReLU or SiLU). The PConv module is a partial convolution (Partial Convolution) structure, and its core idea is to only calculate the effective part of the input feature map in the convolution operation, thereby significantly reducing the computational load and memory occupation. The SPAB module is a structure designed based on the fast parameter-free self-attention mechanism, which aims to enhance the spatial expression ability of the features through the attention mechanism while avoiding the introduction of additional parameters. The C3 module is composed of a CBS module and a plurality of stacked bottleneck modules. Concat represents splicing, and Upsample represents up-sampling.

[0106] In one embodiment, the working principle of the target neck structure includes:

[0107] The feature tensor set of the backbone network is obtained, the first feature tensor is input into the DBS module to obtain the fourth feature tensor, the fourth tensor is up-sampled to obtain the target fourth tensor, and the fourth feature tensor, the target fourth feature tensor, and the second feature tensor are spliced to obtain the fifth feature tensor. The feature tensor set includes: the first feature tensor, the second feature tensor, and the third feature tensor.

[0108] The fifth feature tensor is sequentially input into a target C3 module and a DBS module to obtain a sixth feature tensor, the sixth feature tensor is up-sampled to obtain a target sixth feature tensor, the third feature tensor and the target sixth feature tensor are spliced to obtain a seventh feature tensor;

[0109] The seventh feature tensor is input into a target CSPStage module to obtain an eighth feature tensor, the eighth feature tensor is input into a CBS module to obtain a ninth feature tensor, the ninth feature tensor and the sixth feature tensor are spliced to obtain a tenth feature tensor, and the tenth feature tensor is input into the target CSPStage module to obtain a first target output, and the eighth feature tensor is taken as a second target output.

[0110] The working process of the target CSPStage module includes:

[0111] An initial target feature map is obtained, the initial target feature map is input into a 1x1Conv module to obtain a first target feature map, the first target feature map is input into a target Conv module to obtain a second target feature map, the second target feature map is sequentially input into a Biformer module and a 3x3Conv module to obtain a third target feature map, the first target feature map and the third target feature map are added to obtain a fourth target feature map, the fourth target feature map and the first target feature map are spliced to obtain a fifth target feature map, and the fifth target feature map is input into the 1x1Conv module to obtain a sixth target feature map, and the sixth target feature map is taken as an output of the target CSPStage module.

[0112] In an implementation manner, the fusion of multi-scale features is realized by splicing feature tensors of different scales. Specifically, the up-sampled fourth feature tensor is spliced with the second feature tensor, and the up-sampled sixth feature tensor is spliced with the third feature tensor. This operation can integrate feature information of different levels and enhance the perception ability of the model to multi-scale features, thereby improving the capture effect of narrowband information features in the time-frequency spectrum graph of the gas leakage signal.

[0113] In an implementation manner, the target CSPStage module constructed by introducing the double-layer routing attention mechanism can capture cross-sectional feature information. The attention mechanism focuses on important areas in the feature map, enhances the feature representation, and enables the network to more effectively learn and extract narrowband information features. This mechanism, combined with the spatial feature extraction capability of the convolution layer, improves the performance and generalization ability of the model. By adding the output feature tensor to the original input feature tensor, residual connection is realized. This design helps to solve the problem of gradient vanishing in deep networks, allows the network to learn the residual mapping between the input and the output, thereby improving the training stability and convergence speed of the model and enhancing the learning ability of the model to complex signal features.

[0114] In an implementation manner, the target Conv module is used to preliminarily convert the features, the number of calculation parameters is reduced, the nonlinear expression capability of the network is enhanced, the lightweight design expands the feature space without significantly increasing the calculation burden, provides a richer information basis for subsequent feature extraction, and further improves the efficiency and performance of the model. The target Conv module is composed of two branches, the first branch is composed of a 1x1 Conv module and a Batch Normalization layer, the second branch is composed of a 1x1 Conv module, a 1x3 Conv module, a 3x1 Conv module and a Batch Normalization layer, and the output of the first branch and the output of the second branch are added to serve as the output of the target Conv module; the same receptive field is maintained through 1x1 convolution, 1x3 convolution and 3x1 convolution, the nonlinear expression capability of the network is enhanced, and the number of parameter calculations is reduced.

[0115] Based on the same inventive concept, the embodiment of the present application also provides a DAS-based chemical gas leakage safety detection system. Referring to Figure 4 , Figure 4 A framework diagram of a DAS-based chemical gas leakage safety detection system provided by the embodiment of the present application comprises a global detection module, a local detection module and a safety warning module:

[0116] The global detection module is used to acquire global signal data of a target area, and perform safety detection according to the global signal data to obtain an abnormal detection result;

[0117] The local detection module is used to determine a sub-area according to the abnormal detection result, and acquire local signal data of the sub-area; the target area comprises a plurality of sub-areas;

[0118] The safety warning module is used to detect the local signal data to obtain a local detection result, and perform safety warning according to the local detection result.

[0119] The DAS-based chemical gas leakage safety detection system provided by the embodiment of the present application acquires global signals through DAS, quickly identifies abnormal pipeline areas by using the distributed characteristics of DAS, reduces the detection workload, and improves the response efficiency; DAS cooperates with infrasound sensors, the former provides an abnormal direction, and the latter focuses on local details to accurately capture micro-leakage characteristics.

[0120] In one embodiment, the system further comprises a data acquisition module and a sample library construction module:

[0121] The data acquisition module is configured to acquire historical data, pre-process the historical data to obtain target historical data, determine a length of a sliding window, and segment the target historical data according to the sliding window to obtain a plurality of signal segments; the historical data are leakage sample signals under different pressure and leakage aperture conditions.

[0122] The sample library construction module is configured to extract the plurality of signal segments to obtain a plurality of upper envelope lines, perform quadratic Gaussian smoothing and Lagrange downsampling processing on the upper envelope lines to obtain a plurality of target upper envelope lines, and construct the sample library according to the target upper envelope lines; the signal segments and the upper envelope lines are in one-to-one correspondence.

[0123] In one embodiment, the global detection module includes a first operation module, a second operation module, a third operation module, a fourth operation module, and a fifth operation module.

[0124] The first operation module is configured to decompose the global signal data into a plurality of intrinsic mode functions by variational mode decomposition, determine frequencies of the intrinsic mode functions, and reconstruct intrinsic mode functions with frequencies less than a preset frequency to obtain a noise-reduced signal.

[0125] The second operation module is configured to segment the noise-reduced signal into a plurality of segmented signals according to the sliding window, and perform a target operation on each segmented signal to obtain a real-time signal; the target operation includes upper envelope extraction, quadratic Gaussian smoothing, and Lagrange downsampling operation; the real-time signal includes x1 and x2.

[0126] The third operation module is configured to match the real-time signal with the sample library to obtain minimum cumulative distance matrices D1 and D2, calculate a total cumulative distance matrix according to D1 and D2, and determine a matching signal according to the total cumulative distance matrix.

[0127] The fourth operation module is configured to determine a step length of the sliding window, repeatedly execute the second operation module and the third operation module, and update the real-time signal and the matching signal to obtain a signal sequence; the signal sequence includes a plurality of matching signals.

[0128] The fifth operation module is configured to set a leakage warning threshold, and if all the matching signals in the signal sequence are less than the leakage warning threshold, add an anomaly label to the global signal data to obtain an anomaly detection result.

[0129] In one embodiment, the local detection module is further configured to convert the local signal data into a local frequency spectrum by short-time Fourier transform, and extract features of the local frequency spectrum by a target YOLOv5 model to obtain a local detection result.

[0130] Improvements of the target YOLOv5 model include:

[0131] replacing the C3 module in the YOLOv5 model with the target C3 module and replacing the neck structure of the YOLOv5 model with the target neck structure to obtain the target YOLOv5 model;

[0132] The working principle of the target C3 module includes:

[0133] The initial feature map is obtained, the initial feature map is input into the CBS module to obtain the first feature map, the first feature map is sequentially input into the PConv module, the CBS module and the CBS module to obtain the second feature map, and the first feature map and the second feature map are added to obtain the third feature map.

[0134] The initial feature map is input into the CBS module to obtain the fourth feature map, the third feature map and the fourth feature map are added to obtain the fifth feature map, the fifth feature map is input into the CBS module to obtain the sixth feature map, and the sixth feature map is sequentially input into the four SPAB modules to obtain the seventh feature map.

[0135] In one embodiment, the working principle of the target neck structure includes:

[0136] The feature tensor set of the backbone network is obtained, the first feature tensor is input into the DBS module to obtain the fourth feature tensor, the fourth tensor is up-sampled to obtain the target fourth tensor, and the fourth feature tensor, the target fourth feature tensor and the second feature tensor are spliced to obtain the fifth feature tensor; the feature tensor set includes: the first feature tensor, the second feature tensor and the third feature tensor.

[0137] The fifth feature tensor is sequentially input into the target C3 module and the DBS module to obtain the sixth feature tensor, the sixth feature tensor is up-sampled to obtain the target sixth feature tensor, and the third feature tensor and the target sixth feature tensor are spliced to obtain the seventh feature tensor.

[0138] The seventh feature tensor is input into the target CSPStage module to obtain the eighth feature tensor, the eighth feature tensor is input into the CBS module to obtain the ninth feature tensor, the ninth feature tensor and the sixth feature tensor are spliced to obtain the tenth feature tensor, the tenth feature tensor is input into the target CSPStage module to obtain the first target output, and the eighth feature tensor is taken as the second target output.

[0139] The working process of the target CSPStage module includes:

[0140] The initial target feature map is obtained, the initial target feature map is input into a 1*1Conv module to obtain a first target feature map, the first target feature map is input into a target Conv module to obtain a second target feature map, the second target feature map is sequentially input into a Biformer module and a 3*3Conv module to obtain a third target feature map, the first target feature map and the third target feature map are added to obtain a fourth target feature map, the fourth target feature map and the first target feature map are spliced to obtain a fifth target feature map, the fifth target feature map is input into a 1*1Conv module to obtain a sixth target feature map, and the sixth target feature map is taken as the output of the target CSPStage module.

[0141] The above describes one embodiment of the present application in detail, but the content is only the preferred embodiment of the present application, and cannot be considered as used for limiting the implementation range of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the patent coverage range of the present application.

Claims

1. A DAS-based safety detection method for chemical gas leakage, characterized in that, The method comprises: obtaining global signal data of a target region, and performing safety detection on the global signal data to obtain an anomaly detection result; determining a sub-region according to the anomaly detection result, and obtaining local signal data of the sub-region; the target region comprises a plurality of sub-regions; performing detection on the local signal data to obtain a local detection result, and performing safety warning according to the local detection result; the detection on the local signal data comprises: converting the local signal data into a local frequency spectrum graph through short-time Fourier transform, and extracting features of the local frequency spectrum graph through a target YOLOv5 model to obtain a local detection result; the improvement of the target YOLOv5 model comprises: replacing a C3 module in a YOLOv5 model with a target C3 module, and replacing a neck structure of the YOLOv5 model with a target neck structure to obtain a target YOLOv5 model; the working principle of the target C3 module comprises: obtaining an initial feature map, inputting the initial feature map into a CBS module to obtain a first feature map, inputting the first feature map into a PConv module, a CBS module and a CBS module in sequence to obtain a second feature map, and adding the first feature map and the second feature map to obtain a third feature map; inputting the initial feature map into the CBS module to obtain a fourth feature map, adding the third feature map and the fourth feature map to obtain a fifth feature map, inputting the fifth feature map into the CBS module to obtain a sixth feature map, and inputting the sixth feature map into four SPAB modules in sequence to obtain a seventh feature map; the working principle of the target neck structure comprises: obtaining a feature tensor set of a backbone network, inputting a first feature tensor into a DBS module to obtain a fourth feature tensor, up-sampling the fourth feature tensor to obtain a target fourth tensor, and splicing the fourth feature tensor, the target fourth tensor and a second feature tensor to obtain a fifth feature tensor; the feature tensor set comprises the first feature tensor, the second feature tensor and a third feature tensor; inputting the fifth feature tensor into the target C3 module and the DBS module in sequence to obtain a sixth feature tensor, up-sampling the sixth feature tensor to obtain a target sixth feature tensor, and splicing the third feature tensor and the target sixth feature tensor to obtain a seventh feature tensor; inputting the seventh feature tensor into a target CSPStage module to obtain an eighth feature tensor, inputting the eighth feature tensor into a CBS module to obtain a ninth feature tensor, splicing the ninth feature tensor and the sixth feature tensor to obtain a tenth feature tensor, inputting the tenth feature tensor into the target CSPStage module to obtain a first target output, and taking the eighth feature tensor as a second target output; the working process of the target CSPStage module comprises: The initial target feature map is obtained, the initial target feature map is input into a 1*1Conv module to obtain a first target feature map, the first target feature map is input into a target Conv module to obtain a second target feature map, the second target feature map is sequentially input into a Biformer module and a 3*3Conv module to obtain a third target feature map, the first target feature map and the third target feature map are added to obtain a fourth target feature map, the fourth target feature map and the first target feature map are spliced to obtain a fifth target feature map, the fifth target feature map is input into a 1*1Conv module to obtain a sixth target feature map, and the sixth target feature map is taken as an output of a target CSPStage module.

2. The DAS-based chemical gas leakage safety detection method according to claim 1, characterized in that, Before the safety detection is performed on the global signal data to obtain the anomaly detection result, the method further comprises: obtaining historical data, pre-processing the historical data to obtain target historical data, determining the length of a sliding window, and segmenting the target historical data according to the sliding window to obtain a plurality of signal segments; the historical data are leakage sample signals under different pressure and leakage aperture conditions; extracting each signal segment to obtain a plurality of upper envelope lines, performing quadratic Gaussian smoothing and Lagrange downsampling processing on each upper envelope line to obtain a plurality of target upper envelope lines, and constructing a sample library according to each target upper envelope line; the signal segment and the upper envelope line are in one-to-one correspondence.

3. The DAS-based chemical gas leak safety detection method of claim 2, wherein, The safety detection is performed on the global signal data to obtain the anomaly detection result, comprising: Step one, the global signal data is decomposed into a plurality of intrinsic mode functions by variational modal decomposition, the frequency of each intrinsic mode function is determined, and the intrinsic mode function less than the preset frequency is reconstructed to obtain a denoising signal; Step two, the denoising signal is segmented into a plurality of segmented signals according to a sliding window, and a target operation is performed on each segmented signal to obtain a real-time signal; the target operation includes upper envelope extraction, quadratic Gaussian smoothing and Lagrange downsampling operation, and the real-time signal includes x1 and x2; Step three, the sample library is matched with the real-time signal to obtain minimum cumulative distance matrices D1 and D2, a total cumulative distance matrix is calculated according to D1 and D2, and a matching signal is determined according to the total cumulative distance matrix; Step four, the step length of the sliding window is determined, steps two and three are repeatedly executed, and the real-time signal and the matching signal are updated to obtain a signal sequence; the signal sequence contains a plurality of matching signals; Step five, a leakage warning threshold is set, if the matching signals in the signal sequence are all less than the leakage warning threshold, an anomaly label is added to the global signal data to obtain an anomaly detection result.

4. A DAS-based chemical gas leak safety detection system, characterized by, The system comprises a global detection module, a local detection module and a safety warning module: The global detection module is configured to obtain global signal data of a target region, and perform safety detection on the global signal data to obtain an anomaly detection result; The local detection module is configured to determine a sub-region according to the anomaly detection result, and obtain local signal data of the sub-region; the target region comprises a plurality of sub-regions; The safety warning module is configured to detect the local signal data to obtain a local detection result, and perform safety warning according to the local detection result. The local detection module is further configured to convert the local signal data into a local frequency spectrum graph through short-time Fourier transform, and extract features of the local frequency spectrum graph through the target YOLOv5 model to obtain the local detection result. The improvement of the target YOLOv5 model comprises: The C3 module in the YOLOv5 model is replaced by the target C3 module, and the neck structure of the YOLOv5 model is replaced by the target neck structure to obtain the target YOLOv5 model; The working principle of the target C3 module comprises: An initial feature map is obtained, the initial feature map is input into the CBS module to obtain a first feature map, the first feature map is sequentially input into the PConv module, the CBS module and the CBS module to obtain a second feature map, and the first feature map and the second feature map are added to obtain a third feature map; The initial feature map is input into the CBS module to obtain a fourth feature map, the third feature map and the fourth feature map are added to obtain a fifth feature map, the fifth feature map is input into the CBS module to obtain a sixth feature map, and the sixth feature map is sequentially input into four SPAB modules to obtain a seventh feature map; The working principle of the target neck structure comprises: A feature tensor set of a backbone network is obtained, a first feature tensor is input into the DBS module to obtain a fourth feature tensor, the fourth feature tensor is up-sampled to obtain a target fourth tensor, and the fourth feature tensor, the target fourth tensor and a second feature tensor are spliced to obtain a fifth feature tensor; the feature tensor set comprises the first feature tensor, the second feature tensor and a third feature tensor; The fifth feature tensor is sequentially input into the target C3 module and the DBS module to obtain a sixth feature tensor, the sixth feature tensor is up-sampled to obtain a target sixth feature tensor, and the third feature tensor and the target sixth feature tensor are spliced to obtain a seventh feature tensor; The seventh feature tensor is input into the target CSPStage module to obtain an eighth feature tensor, the eighth feature tensor is input into the CBS module to obtain a ninth feature tensor, the ninth feature tensor and the sixth feature tensor are spliced to obtain a tenth feature tensor, the tenth feature tensor is input into the target CSPStage module to obtain a first target output, and the eighth feature tensor is taken as a second target output; The working process of the target CSPStage module comprises: An initial target feature map is obtained, the initial target feature map is input into a 1x1 Conv module to obtain a first target feature map, the first target feature map is input into a target Conv module to obtain a second target feature map, the second target feature map is sequentially input into a Biformer module and a 3x3 Conv module to obtain a third target feature map, the first target feature map and the third target feature map are added to obtain a fourth target feature map, the fourth target feature map and the first target feature map are spliced to obtain a fifth target feature map, the fifth target feature map is input into a 1x1 Conv module to obtain a sixth target feature map, and the sixth target feature map is taken as an output of a target CSPStage module.

5. The DAS-based chemical gas leak safety detection system of claim 4, wherein, The system further comprises a data acquisition module and a sample library construction module. The data acquisition module is configured to acquire historical data, pre-process the historical data to obtain target historical data, determine the length of a sliding window, and segment the target historical data into a plurality of signal segments according to the sliding window; the historical data are leakage sample signals under different pressure and leakage aperture conditions. The sample library construction module is configured to extract a plurality of upper envelope lines from the signal segments, perform quadratic Gaussian smoothing and Lagrange downsampling processing on the upper envelope lines to obtain a plurality of target upper envelope lines, and construct a sample library according to the target upper envelope lines; the signal segments and the upper envelope lines are in one-to-one correspondence.

6. The DAS-based chemical gas leak safety detection system of claim 5, wherein, The global detection module comprises a first operation module, a second operation module, a third operation module, a fourth operation module and a fifth operation module. The first operation module is configured to decompose the global signal data into a plurality of intrinsic mode functions through variational mode decomposition, determine the frequency of each intrinsic mode function, and reconstruct the intrinsic mode functions with a frequency less than a preset frequency to obtain a denoised signal. The second operation module is configured to segment the denoised signal into a plurality of segmented signals according to a sliding window, and perform a target operation on each segmented signal to obtain a real-time signal; the target operation comprises upper envelope extraction, quadratic Gaussian smoothing and Lagrange downsampling operation, and the real-time signal comprises x1 and x2. The third operation module is configured to match the real-time signal according to the sample library to obtain minimum cumulative distance matrices D1 and D2, calculate a total cumulative distance matrix according to D1 and D2, and determine a matching signal according to the total cumulative distance matrix. The fourth operation module is configured to determine the step length of the sliding window, repeatedly execute the second operation module and the third operation module, and update the real-time signal and the matching signal to obtain a signal sequence; the signal sequence comprises a plurality of matching signals. The fifth operation module is configured to set a leakage warning threshold, and if all the matching signals in the signal sequence are less than the leakage warning threshold, add an abnormal label to the global signal data to obtain an abnormal detection result.

Images