CN117053124A - Method and device for detecting leakage of oil-gas branch pipeline - Google Patents

Abstract

The invention discloses a method and device for detecting branch pipeline leakage, which relates to the technical field of detection. The main purpose is to solve the existing problem of poor detection accuracy of branch pipeline leakage. The method includes: collecting acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, including at least one branch pipe between the two ends of the oil and gas pipeline; performing short-time frame timing on the acoustic emission signals and the pressure signal according to a preset time interval Align to obtain the sound pressure time series signal; classify and detect the sound pressure time series signal based on the pipeline leakage classification model that has completed model training, and obtain the pipeline detection results. The pipeline leakage classification model includes feature information based on the cross attention mechanism. Fusion coding layer.

Description

Detection methods and devices for oil and gas branch pipeline leakage

Technical field

The present invention relates to the field of detection technology, and in particular to a method and device for detecting branch pipeline leakage.

Background technique

Pipeline transportation is widely used in oil and gas transportation due to its safety, reliability, economy and practicality. Among them, due to the complex pipeline environment and pipelines with branches are more common, leakage detection of branch pipelines has always been a research hotspot.

At present, the existing detection of leakage in oil and gas branch pipelines usually collects a single signal and combines it with model algorithms such as neural networks and deep learning for classification prediction. However, due to the complexity and diversity of the pipeline site, it is difficult to ensure the integrity of the extracted leakage signal using one signal alone, and there is a commonality in all of them. Only the straight pipe section is considered, and leakage in branch pipelines is not considered. Therefore, there is an urgent need for a A method for detecting leakage in branch pipelines to solve the above problems.

Contents of the invention

In view of this, the present invention provides a branch pipeline leakage detection method and device, with the main purpose of solving the existing problem of poor detection accuracy of branch pipeline leakage.

According to one aspect of the present invention, a branch pipeline leakage detection method is provided, including:

Collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends;

Perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal;

Based on the pipeline leakage classification model that has completed model training, the acoustic emission signal and the pressure signal are classified and detected to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism. .

Further, before collecting the acoustic emission signals and pressure signals at both ends of the oil and gas pipeline, the method also includes:

Obtain the pipeline leakage sample data set, and build an initial pipeline leakage classification model based on the encoder, which includes a position coding layer carrying pressure tokens and sound tokens, a cross-fusion coding layer, and a fusion decision-making layer;

Model training is performed on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain a pipeline leakage classification model.

Further, performing model training on the initial pipeline leakage classification model based on the pipeline leakage sample data set, and obtaining the pipeline leakage classification model includes:

Based on the denoised sound time series signals and pressure time series signals in the pipeline leakage sample data set, local features are extracted through the fully connected layer in the initial pipeline leakage classification model, and pressure tokens and sound tokens are added, and The local features are position-coded in the position coding layer to obtain a pressure packet token sequence and a sound packet token sequence;

Acquire sound timing features based on the voice encoder in the cross-fusion coding layer, and obtain the pressure timing features based on the pressure encoder in the cross-fusion coding layer;

Perform feature exchange between the pressure pack token sequence and the sound timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused sound features;

Perform feature exchange between the sound packet token sequence and the pressure timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused pressure features;

In the fusion decision-making layer, the pressure feature and the sound feature are fused and converted into leakage category probabilities, and when the probability category matching model training requirements are obtained, it is determined to complete the model training of the pipeline leakage classification model.

Further, performing modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal includes:

Perform complementary ensemble empirical mode decomposition on the acoustic emission signal and the pressure signal to obtain an acoustic component and a pressure component, and determine a high correlation component based on the acoustic component and the pressure component;

Screen the sound components and pressure components that match the classification type, and perform filtering and noise reduction on the sound components and the pressure components to obtain the noise-reduced sound components and the pressure components;

Signal reconstruction is performed on the high correlation component and the noise-reduced acoustic component and the pressure component to obtain the noise-reduced acoustic emission signal and the pressure signal.

Further, the acoustic emission signal and the pressure signal are classified and detected based on the pipeline leakage classification model that has completed model training. After obtaining the pipeline detection result, the method further includes:

If the pipeline detection result is a pipeline leak, the attenuation trend of the acoustic emission signal and the pressure signal is determined, and the leakage positioning interval corresponding to the attenuation trend is determined according to the branch pipeline distribution mapping relationship.

According to another aspect of the present invention, a branch pipeline leakage detection device is provided, including:

An acquisition module, used to collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends;

A noise reduction module, used to perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal;

A detection module, configured to classify and detect the acoustic emission signal and the pressure signal based on a pipeline leakage classification model that has completed model training, and obtain pipeline detection results. The pipeline leakage classification model includes features based on a cross-attention mechanism. Coding layer of information fusion.

Further, the device also includes:

The acquisition module is used to obtain the pipeline leakage sample data set and build an initial pipeline leakage classification model based on the encoder. The encoder includes a position encoding layer carrying pressure tokens and sound tokens, and a cross-fusion encoding layer. , integrated decision-making layer;

A training module, configured to perform model training on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain a pipeline leakage classification model.

Further, the training module is specifically used to extract local features based on the denoised sound time series signals and pressure time series signals in the pipeline leakage sample data set through the fully connected layer in the initial pipeline leakage classification model and then add pressure. class tokens and sound class tokens, and perform position coding on the local features in the position coding layer to obtain the pressure packet token sequence and the sound packet token sequence; obtain the sound based on the voice encoder in the cross-fusion coding layer Temporal features, and the pressure encoder in the cross-fusion coding layer obtains pressure timing features; perform feature exchange between the pressure packet token sequence and the sound timing features, and perform feature interactive fusion through the cross-attention mechanism to obtain The fused sound features; feature exchange of the sound packet token sequence and the pressure timing features, and interactive fusion of features through a cross-attention mechanism to obtain the fused pressure features; in the fusion decision-making layer for all The pressure characteristics and the sound characteristics are fused and converted into leakage category probabilities, and when the probability category matching model training requirements are obtained, it is determined to complete the model training of the pipeline leakage classification model.

Further, the noise reduction module is specifically configured to perform complementary set empirical mode decomposition on the acoustic emission signal and the pressure signal to obtain the acoustic component and the pressure component, and determine based on the acoustic component and the pressure component. Highly correlated components; filter the sound components and pressure components that match the classification type, and perform filtering and noise reduction on the sound components and the pressure components to obtain the noise-reduced sound components and the pressure components; filter the high-correlation components The correlation component and the noise-reduced acoustic component and the pressure component are subjected to signal reconstruction to obtain the noise-reduced acoustic emission signal and the pressure signal.

Further, the device also includes:

A determination module, configured to determine the attenuation trend of the acoustic emission signal and the pressure signal if the pipeline detection result is a pipeline leakage, and determine the leakage positioning interval corresponding to the attenuation trend according to the branch pipeline distribution mapping relationship.

According to another aspect of the present invention, a storage medium is provided, and at least one executable instruction is stored in the storage medium. The executable instruction causes the processor to perform operations corresponding to the above-mentioned method for detecting branch pipe leakage.

According to yet another aspect of the present invention, a terminal is provided, including: a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface complete communication with each other through the communication bus. ;

The memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the above-mentioned method for detecting branch pipeline leakage.

Through the above technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages:

The present invention provides a method and device for detecting branch pipeline leakage. Compared with the existing technology, the embodiment of the present invention collects acoustic emission signals and pressure signals located at both ends of an oil and gas pipeline. The two ends of the oil and gas pipeline include at least A branch pipeline; perform short-time frame timing alignment of the acoustic emission signal and the pressure signal according to a preset time interval to obtain a sound pressure timing signal; based on the pipeline leakage classification model that has completed model training, the sound pressure timing signal is Classification detection is performed to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism, which greatly improves the accuracy of branch pipeline leakage inspection and achieves effective detection of branch pipeline leakage.

The above description is only an overview of the technical solution of the present invention. In order to have a clearer understanding of the technical means of the present invention, it can be implemented according to the content of the description, and in order to make the above and other objects, features and advantages of the present invention more obvious and understandable. , the specific embodiments of the present invention are listed below.

Description of the drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for the purpose of illustrating preferred embodiments only and are not to be construed as limiting the invention. Also throughout the drawings, the same reference characters are used to designate the same components. In the attached picture:

Figure 1 shows a flow chart of a branch pipeline leakage detection method provided by an embodiment of the present invention;

Figure 2 shows a schematic structural diagram of a branch pipeline provided by an embodiment of the present invention;

Figure 3 shows a schematic diagram of the overall model of a sound-pressure signal fusion Transformer provided by an embodiment of the present invention;

Figure 4 shows a schematic structural diagram of a Transformer encoder provided by an embodiment of the present invention;

Figure 5 shows a schematic diagram of a self-attention mechanism provided by an embodiment of the present invention;

Figure 6 shows a CEEMD-LMS noise reduction flow chart provided by an embodiment of the present invention;

Figure 7 shows a branch pipeline tiling diagram provided by an embodiment of the present invention;

Figure 8 shows a block diagram of a device for detecting branch pipeline leakage provided by an embodiment of the present invention;

Figure 9 shows a schematic structural diagram of a terminal provided by an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a thorough understanding of the disclosure, and to fully convey the scope of the disclosure to those skilled in the art.

An embodiment of the present invention provides a method for detecting branch pipeline leakage, as shown in Figure 1. The method includes:

101. Collect acoustic emission signals and pressure signals at both ends of the oil and gas pipeline.

In the embodiment of the present invention, the current execution end is a cloud server or a terminal server that performs pipeline leakage detection. Specifically, the oil and gas pipeline includes at least one branch pipeline between two ends. As shown in Figure 2, when a leak occurs in the branch pipeline, Due to the deformation of the pipeline, an acoustic emission signal will be generated. At the same time, the gas at the leakage point will be discharged outward through the aperture under the action of pressure, resulting in a negative pressure wave. Therefore, the current execution end is equipped with two acoustic emission sensors and two pressure sensors at the beginning and end of the pipeline to detect the acoustic emission signal and negative pressure wave signal generated by pipeline leakage.

102. Perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and the pressure signal.

In the embodiment of the present invention, in order to avoid the presence of more distorted signals and useless signals in the acoustic emission signal and pressure signal, the current execution end uses a modal decomposition method to reduce noise on the signal. Among them, the CEEMD mode decomposition method can be used, that is, CEEMD is a decomposition method further optimized from the EEMD decomposition. It mainly optimizes the residual noise of the EEMD decomposition. The main principle is to use a pair of positive and negative white noises that are opposite to each other as auxiliary noise. It is added to the source signal to eliminate the excess auxiliary white noise remaining in the reconstructed signal after decomposition by the original EEMD method, while reducing the number of iterations required during decomposition and reducing the computational cost.

103. Classify and detect the acoustic emission signal and the pressure signal based on the pipeline leakage classification model that has completed model training, and obtain pipeline detection results.

In the embodiment of the present invention, after obtaining the acoustic emission signal and the pressure signal, the sound timing signal and the pressure timing signal are classified based on the pipeline leakage classification model that has completed model training, and the pipeline detection results are obtained. Among them, the pipeline leakage classification model includes a coding layer for feature information fusion based on the cross-attention mechanism. At this time, the coding layer as the encoder carries a position coding layer for pressure tokens and sound tokens, and cross fusion The coding layer and fusion decision-making layer can classify and detect pressure signals and acoustic emission signals to obtain pipeline detection results, such as leakage or no leakage.

In another embodiment of the present invention, for further definition and explanation, before the step of collecting acoustic emission signals and pressure signals at both ends of the oil and gas pipeline, the method further includes:

Obtain a pipeline leakage sample data set and build an initial pipeline leakage classification model based on the encoder;

Model training is performed on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain a pipeline leakage classification model.

In order to implement classification detection based on the pipeline leakage classification model, the current execution end first obtains the pipeline leakage sample data set and builds an initial pipeline leakage classification model based on the encoder. Among them, the encoder includes a position coding layer carrying pressure tokens and sound tokens, as well as a cross-fusion coding layer and a fusion decision-making layer, as shown in Figure 3, that is, the input is the CEEMD collected in the same period. -LMS reduces noise and improves acoustic emission signals and pressure signals, and outputs different leak types, enabling leakage detection of pipeline acoustic-pressure signal fusion. Among them, the pipeline leakage sample data set contains noise-reduced acoustic emission signal samples and pressure signal samples marked with different leakage results, which are not specifically limited in the embodiment of the present invention.

It should be noted that the initial pipeline leakage classification model constructed is a Transformer encoder, as shown in Figure 4, which is mainly divided into 3 parts: (1) Position encoding. Since the self-attention mechanism does not consider the positional relationship of elements in the time series, the input sequence X needs to be positionally encoded to add positional information, that is, P _E (X) = position matrix. (2) Residual connection. The residual connection module is used in Transformer to enhance information flow, improve model performance, and combine with layer normalization operations to optimize the training process, R _C = L _N [X + S _A (X)]; where R _C is the residual connection operation, L _N is the layer normalization operation. (3) Feedforward network. The feedforward network in Transformer consists of 2 linear transformation layers and 1 nonlinear activation layer, N _FFN (X) = W ₂ ·f (W ₁ +X); where: N _FFN (X) is the feedforward network , W ₁ and W ₂ are two linear layer parameters, and f (W ₁ +X) is a nonlinear activation function. in addition,

In another embodiment of the present invention, in order to further define and illustrate, the step of performing model training on the initial pipeline leakage classification model based on the pipeline leakage sample data set, and obtaining the pipeline leakage classification model includes:

Based on the denoised sound time series signals and pressure time series signals in the pipeline leakage sample data set, local features are extracted through the fully connected layer in the initial pipeline leakage classification model, and pressure tokens and sound tokens are added, and The local features are position-coded in the position coding layer to obtain a pressure packet token sequence and a sound packet token sequence;

Acquire sound timing features based on the voice encoder in the cross-fusion coding layer, and obtain the pressure timing features based on the pressure encoder in the cross-fusion coding layer;

Perform feature exchange between the pressure pack token sequence and the sound timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused sound features;

Perform feature exchange between the sound packet token sequence and the pressure timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused pressure features;

In the fusion decision-making layer, the pressure feature and the sound feature are fused and converted into leakage category probabilities, and when the probability category matching model training requirements are obtained, it is determined to complete the model training of the pipeline leakage classification model.

In order to improve the classification detection accuracy of the leak detection model, the current execution end samples the short-term token generation method during the training process to encode the signal in the position encoding layer. Specifically, the current execution end uses short-time Fourier transform to analyze the original time domain signal (the noise-reduced sound time series signal and pressure time series signal in the pipeline leakage sample data set), that is, the time domain signal is divided into frames and then quickly Fourier transform is analyzed from two dimensions: time and short-term local frequency. In a specific implementation scenario, a sound signal with a length of L _s is collected in one sampling The pressure signal V∈R ^Lv×1 with length L _v , (S is the acoustic signal and V is the pressure signal, V1 is the local feature of the first frame pressure signal, S1 is the local feature of the first frame acoustic signal, ds is The fully connected layer output dimension, n represents the number of segments). Take the pressure signal as an example to illustrate. First, divide V into n small segments of length l _v to obtain the framed signal/> The framed signal passes through a fully connected layer to extract the local features of each frame/> Furthermore, insert the pressure category token before the first element of V ¹ At this time/> The purpose of the stress class token is to store the leaked class information in the current sequence and randomly initialize it before training. Finally, V ¹ is position-coded according to the formula P _E (X) = X + Z. The position-encoded V ¹ is called a pressure package token sequence, which includes 1 pressure category token/> and n pressure short-term tokens/> In the same way, there is a sound packet token sequence S ¹ , including 1 sound category token/> and n sound short-term tokens/> After the above method, the number of frames of S ¹ and V ¹ is the same and corresponds one to one, thus achieving the time alignment of acoustic emission signals and pressure signals with different sampling rates.

In addition, for the feature fusion in the cross-fusion coding layer, the current execution end is to realize the fusion of the acoustic emission signal and the pressure signal at the feature level, and the fusion of the two feature extraction processes is carried out through the designed acoustic-pressure cross-fusion Transformer module. In a specific implementation scenario, the sound timing feature S ² of the serially connected M-layer sound Transformer encoder S ¹ is passed, and the pressure timing feature V ² is passed through the serially connected N-layer pressure Transformer encoder V ¹ . Each contains new category tokens in S ² and V ² and/> Jinre's, exchanges the category tokens in the sound temporal characteristics and the pressure temporal characteristics, and carries out the interactive fusion of the two feature levels through the cross-attention mechanism of the respective series L layers. Finally, the cross-fusion Transformer encoder composed of 2 basic M-layer, N-layer Transformer encoders and L-layer serial cross-attention modules can be connected in series K times to obtain a deeper network structure.

It should be noted that the category tokens added in the initial pipeline leak classification model initially consist of randomly initialized parameters and do not contain any information about leak detection. As the sound-pressure data flows in the network, due to the characteristics of the self-attention mechanism, information about the leakage type is added during the feature extraction process by calculating the correlation score with each short-term token of the current input into the category token. Therefore, in the detection stage, only the information of sound-pressure category tokens is used for fusion decision-making to reduce the information redundancy caused by using all packet token sequences for detection. Among them, the sound category tokens and detection category tokens output by the cross-fusion Transformer encoder are input into their respective multi-layer perception heads (MLPHeader). For example, when the cross-fusion Transformer encoder has only 1 layer, the output is with/> but In the formula,/> and/> are two linear transformations, ξ _v and ξ _s are the output logical values corresponding to the result of leakage, and n _c is the number of leakage types. Fusion ξ _v and ξ _s and convert them into probabilities corresponding to n _c leakage categories, that is/> Among them,/> Add element by element.

It should be noted that the attention mechanism used in the embodiment of the present invention can be regarded as a non-local filtering operation. The response of each position in the sequence is calculated by estimating the attention scores of all positions and collecting corresponding inputs based on the scores. , as shown in Figure 5, for the input X=[x _1, x ₂ ,...,x _n ]∈R ^n×d , the final output sequence is Y=[y ₁ , y ₂ ,..., y _n ]∈R ^n×d , the calculation process is as follows: In the formula: R is a matrix composed of n×d, n is the sequence length, d is the number of dimensions, Q, K, V are three intermediate matrices, and the three matrices belong to the same sequence W _q , W _k , W _v are three different linear transformation matrices, S _A (Q, K, V) is the self-attention calculation function; A is the self-attention matrix, each element of which represents the attention score between two elements in X.

In another embodiment of the present invention, for further definition and explanation, the step of performing modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal includes:

Perform complementary ensemble empirical mode decomposition on the acoustic emission signal and the pressure signal to obtain an acoustic component and a pressure component, and determine a high correlation component based on the acoustic component and the pressure component;

Screen the sound components and pressure components that match the classification type, and perform filtering and noise reduction on the sound components and the pressure components to obtain the noise-reduced sound components and the pressure components;

Signal reconstruction is performed on the high correlation component and the noise-reduced acoustic component and the pressure component to obtain the noise-reduced acoustic emission signal and the pressure signal.

In order to de-noise the signal and thereby improve the accuracy of pipeline leak detection, the current execution end first uses CEEMD decomposition combined with LMS de-noising. Specifically, the specific steps of CEEMD decomposition are: (1) Set the number of processing times of the original signal; (2) Add a set of noise signals with opposite signs to the original signal, with the same amplitude for each iteration; In the formula: x(t) represents the original signal,/> and/> Represents positive and negative noise. (3)Yes/> and/> Perform EMD decomposition to obtain two groups of decomposed IMF components (IMF+ and IMF-), and then average the two groups of IMF components to obtain the corresponding mean IMF component. The calculation formula is as follows:/> In the formula: δ _IMF1 and δ _IMF2 are the average values of δ _IMF+ and δ _IMF- respectively, and N represents the number of decomposition layers. The final decomposition result is the integrated average of the four groups of mean IMF components, and steps (2) (3) are repeated continuously. The iteration stops when the final residual component is a monotonic function or a constant. Specifically, as shown in Figure 6, the steps for combining CEEMD-LMS noise reduction are: Step 1: Perform CEEMD decomposition of the noisy signal to obtain a set of IMF components; Step 2: Calculate the correlation coefficient of the IMF component and determine the sum of high correlation components. Low correlation components; Step 3: Use the SE-Hurst index to evaluate the low correlation components and screen out 2 types of IMF components; Step 4: Perform LMS filtering and denoising on the noisy components to obtain the denoised components; Step 5: Filter the high-correlation components The relevant components and the denoised components are reconstructed to obtain a pure signal.

In another embodiment of the present invention, in order to further define and illustrate, the step is to classify and detect the sound pressure time series signal based on the pipeline leakage classification model that has completed model training. After obtaining the pipeline detection result, the method further includes:

If the pipeline detection result is a pipeline leak, the attenuation trend of the acoustic emission signal and the pressure signal is determined, and the leakage positioning interval corresponding to the attenuation trend is determined according to the branch pipeline distribution mapping relationship.

In order to improve the accuracy of detecting branch pipelines, the current execution end determines that the pipeline detection result is pipeline leakage to ensure detection accuracy and determine the attenuation trend of the acoustic emission signal and pressure signal. At this time, the attenuation trend is after the liquid leakage in the pipeline. , the attenuation caused by sound and pressure, therefore, the attenuation trend of the acoustic emission signal and the pressure signal can be determined according to the changes in the signal. After the attenuation trend is determined, the leakage location interval corresponding to the attenuation trend is determined according to the branch pipeline distribution mapping relationship. Among them, the branch pipe distribution mapping relationship is the mapping relationship between the branch pipe positioning interval and the attenuation trend configured based on the leakage experiments of different branch pipes. Therefore, the leakage positioning interval corresponding to the attenuation trend can be determined according to the branch pipe distribution mapping relationship. , the embodiments of the present invention are not specifically limited.

In a specific embodiment, the current execution end uses acoustic-pressure data to collect on the branch pipeline detection platform, installs acoustic emission sensors and pressure transmitters at the first and last stations of the pipeline, uses valves to simulate the size of the leakage, and uses the data The collector collects the acoustic emission signals and pressure signals when leakage occurs and sends them to the PC for data analysis. Among them, the pipe material is steel pipe, the diameter of the straight pipe section is 50mm, and the diameter of the branch pipe is 25mm. There is gas in the pipe. The gas is injected through the air compressor. The working pressure of the air compressor is 0.8MPA. The acoustic emission sensor model is RS-5A, the width is 18.8mm, the height is 15mm, the base material is ceramic, and the sampling frequency is 50-800HZ; the pressure sensor model is ICP 106B pressure sensor, the working pressure range is 8.3psi, and it is made of stainless steel. Measuring media are liquids and gases. As shown in Figure 7, it is a tile diagram of the branch pipeline, marking the first station, the last station of the pipeline and the location of the simulated leakage of the pipeline. The total length of the branch pipeline is 2500m. Then based on the branch pipeline distribution mapping relationship, it is determined that the distance between the leakage point and the first station is 1000m, and the distance between the leakage point and the first station is 1500m.

Embodiments of the present invention provide a method for detecting leakage in branch pipelines. Compared with the existing technology, embodiments of the present invention collect acoustic emission signals and pressure signals located at both ends of an oil and gas pipeline. The two ends of the oil and gas pipeline include at least A branch pipeline; perform short-time frame timing alignment of the acoustic emission signal and the pressure signal according to a preset time interval to obtain a sound pressure timing signal; based on the pipeline leakage classification model that has completed model training, the sound pressure timing signal is Classification detection is performed to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism, which greatly improves the accuracy of branch pipeline leakage inspection and achieves effective detection of branch pipeline leakage.

Further, as an implementation of the method shown in Figure 1 above, an embodiment of the present invention provides a device for detecting branch pipeline leakage. As shown in Figure 8, the device includes:

The acquisition module 21 is used to collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends;

The noise reduction module 22 is used to perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal;

The detection module 23 is used to classify and detect the acoustic emission signal and the pressure signal based on the pipeline leakage classification model that has completed model training to obtain pipeline detection results. The pipeline leakage classification model includes detection based on a cross-attention mechanism. Coding layer for feature information fusion.

Further, the device also includes:

The acquisition module is used to obtain the pipeline leakage sample data set and build an initial pipeline leakage classification model based on the encoder. The encoder includes a position encoding layer carrying pressure tokens and sound tokens, and a cross-fusion encoding layer. , integrated decision-making layer;

A training module, configured to perform model training on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain a pipeline leakage classification model.

Further, the training module is specifically used to extract local features based on the denoised sound time series signals and pressure time series signals in the pipeline leakage sample data set through the fully connected layer in the initial pipeline leakage classification model and then add pressure. class tokens and sound class tokens, and perform position coding on the local features in the position coding layer to obtain the pressure packet token sequence and the sound packet token sequence; obtain the sound based on the voice encoder in the cross-fusion coding layer Temporal features, and the pressure encoder in the cross-fusion coding layer obtains pressure timing features; perform feature exchange between the pressure packet token sequence and the sound timing features, and perform feature interactive fusion through the cross-attention mechanism to obtain The fused sound features; feature exchange of the sound packet token sequence and the pressure timing features, and interactive fusion of features through a cross-attention mechanism to obtain the fused pressure features; in the fusion decision-making layer for all The pressure characteristics and the sound characteristics are fused and converted into leakage category probabilities, and when the probability category matching model training requirements are obtained, it is determined to complete the model training of the pipeline leakage classification model.

Further, the noise reduction module is specifically configured to perform complementary set empirical mode decomposition on the acoustic emission signal and the pressure signal to obtain the acoustic component and the pressure component, and determine based on the acoustic component and the pressure component. Highly correlated components; filter the sound components and pressure components that match the classification type, and perform filtering and noise reduction on the sound components and the pressure components to obtain the noise-reduced sound components and the pressure components; filter the high-correlation components The correlation component and the noise-reduced acoustic component and the pressure component are subjected to signal reconstruction to obtain the noise-reduced acoustic emission signal and the pressure signal.

Further, the device also includes:

A determination module, configured to determine the attenuation trend of the acoustic emission signal and the pressure signal if the pipeline detection result is a pipeline leakage, and determine the leakage positioning interval corresponding to the attenuation trend according to the branch pipeline distribution mapping relationship.

Embodiments of the present invention provide a device for detecting branch pipeline leakage. Compared with the prior art, embodiments of the present invention collect acoustic emission signals and pressure signals located at both ends of an oil and gas pipeline. The two ends of the oil and gas pipeline include at least A branch pipeline; perform short-time frame timing alignment of the acoustic emission signal and the pressure signal according to a preset time interval to obtain a sound pressure timing signal; based on the pipeline leakage classification model that has completed model training, the sound pressure timing signal is Classification detection is performed to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism, which greatly improves the accuracy of branch pipeline leakage inspection and achieves effective detection of branch pipeline leakage.

According to an embodiment of the present invention, a storage medium is provided. The storage medium stores at least one executable instruction. The computer executable instruction can execute the branch pipeline leak detection method in any of the above method embodiments.

Figure 9 shows a schematic structural diagram of a terminal provided according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the terminal.

As shown in Figure 9, the terminal may include: a processor (processor) 302, a communications interface (Communications Interface) 304, a memory (memory) 306, and a communication bus 308.

Among them: the processor 302, the communication interface 304, and the memory 306 complete communication with each other through the communication bus 308.

The communication interface 304 is used to communicate with network elements of other devices such as clients or other servers.

The processor 302 is configured to execute the program 310. Specifically, it can execute the relevant steps in the embodiment of the method for detecting branch pipeline leakage.

Specifically, program 310 may include program code including computer operating instructions.

The processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the terminal may be the same type of processor, such as one or more CPUs; or they may be different types of processors, such as one or more CPUs and one or more ASICs.

Memory 306 is used to store program 310. The memory 306 may include high-speed RAM memory, and may also include non-volatile memory (non-volatile memory), such as at least one disk memory.

The program 310 may be specifically used to cause the processor 302 to perform the following operations:

Collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends;

Perform short-time frame timing alignment on the acoustic emission signal and the pressure signal according to a preset time interval to obtain a sound pressure timing signal;

Based on the pipeline leakage classification model that has completed model training, the sound pressure time series signals are classified and detected to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism.

Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented using general-purpose computing devices. They can be concentrated on a single computing device, or distributed across a network composed of multiple computing devices. , optionally, they may be implemented in program code executable by a computing device, such that they may be stored in a storage device for execution by the computing device, and in some cases, may be in a sequence different from that herein. The steps shown or described are performed either individually as individual integrated circuit modules, or as multiple modules or steps among them as a single integrated circuit module. As such, the invention is not limited to any specific combination of hardware and software.

The above descriptions are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims (8)

1. A method for detecting branch pipeline leakage, which is characterized by including: Collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends; Perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal; Based on the pipeline leakage classification model that has completed model training, the acoustic emission signal and the pressure signal are classified and detected to obtain pipeline detection results. The pipeline leakage classification model includes a coding layer for feature information fusion based on a cross-attention mechanism. . 2. The method according to claim 1, characterized in that before collecting the acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, the method further includes: Obtain the pipeline leakage sample data set, and build an initial pipeline leakage classification model based on the encoder, which includes a position coding layer carrying pressure tokens and sound tokens, a cross-fusion coding layer, and a fusion decision-making layer; Model training is performed on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain a pipeline leakage classification model. 3. The method according to claim 2, characterized in that said performing model training on the initial pipeline leakage classification model based on the pipeline leakage sample data set to obtain the pipeline leakage classification model includes: Based on the denoised sound time series signals and pressure time series signals in the pipeline leakage sample data set, local features are extracted through the fully connected layer in the initial pipeline leakage classification model, and pressure tokens and sound tokens are added, and The local features are position-coded in the position coding layer to obtain a pressure packet token sequence and a sound packet token sequence; Acquire sound timing features based on the voice encoder in the cross-fusion coding layer, and obtain the pressure timing features based on the pressure encoder in the cross-fusion coding layer; Perform feature exchange between the pressure pack token sequence and the sound timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused sound features; Perform feature exchange between the sound packet token sequence and the pressure timing features, and perform feature interactive fusion through a cross-attention mechanism to obtain the fused pressure features; In the fusion decision-making layer, the pressure feature and the sound feature are fused and converted into leakage category probabilities, and when the probability category matching model training requirements are obtained, it is determined to complete the model training of the pipeline leakage classification model. 4. The method of claim 3, wherein performing modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal includes: Perform complementary ensemble empirical mode decomposition on the acoustic emission signal and the pressure signal to obtain an acoustic component and a pressure component, and determine a high correlation component based on the acoustic component and the pressure component; Screen the sound components and pressure components that match the classification type, and perform filtering and noise reduction on the sound components and the pressure components to obtain the noise-reduced sound components and the pressure components; Signal reconstruction is performed on the high correlation component and the noise-reduced acoustic component and the pressure component to obtain the noise-reduced acoustic emission signal and the pressure signal. 5. The method according to any one of claims 1 to 4, characterized in that the pipeline leakage classification model based on completed model training performs classification and detection on the acoustic emission signal and the pressure signal to obtain pipeline detection. After the results, the method also includes: If the pipeline detection result is a pipeline leak, the attenuation trend of the acoustic emission signal and the pressure signal is determined, and the leakage positioning interval corresponding to the attenuation trend is determined according to the branch pipeline distribution mapping relationship. 6. A branch pipeline leakage detection device, characterized in that it includes: An acquisition module, used to collect acoustic emission signals and pressure signals located at both ends of the oil and gas pipeline, which includes at least one branch pipe between the two ends; A noise reduction module, used to perform modal decomposition on the acoustic emission signal and the pressure signal to obtain the noise-reduced acoustic emission signal and pressure signal; A detection module, configured to classify and detect the acoustic emission signal and the pressure signal based on a pipeline leakage classification model that has completed model training, and obtain pipeline detection results. The pipeline leakage classification model includes features based on a cross-attention mechanism. Coding layer of information fusion. 7. A storage medium, at least one executable instruction is stored in the storage medium, and the executable instruction causes the processor to execute the branch pipeline leakage detection method corresponding to any one of claims 1-5. operate. 8. A terminal, including: a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface completing mutual communication through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the method for detecting branch pipeline leakage according to any one of claims 1 to 5.