Ship pipeline leakage detection network training method, ship pipeline leakage detection network application method and pipeline detector

Abstract

The invention provides a ship pipeline leakage detection network training method, an application method and a pipeline detector, and belongs to the field of industrial equipment detection. The training method comprises the following steps: performing physical modeling analysis on acquired ship pipeline sensing data to obtain physical derived data, and fusing the physical derived data and the ship pipeline sensing data to obtain fused training data; performing multilayer graph convolution on the fusion training data to obtain spatial information features, performing spatio-temporal information attention enhancement to obtain spatio-temporal information features, fusing the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, predicting the spatio-temporal fusion features to obtain pipeline leakage prediction output, and determining prediction loss according to the pipeline leakage prediction output. And iteratively training to obtain a ship pipeline leakage detection network. According to the ship pipeline leakage detection method, the data dimension and the data volume are enriched through the physical derived data, the space-time relevance in the mined data is enhanced through multilayer graph convolution and space-time information attention, and the accuracy of ship pipeline leakage detection is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment detection, and particularly to a method for training a ship pipeline leakage detection network, an application method, and a pipeline detector. Background Art

[0002] The pipeline system in the ship engine room plays a crucial role during ship operation. It conveys various liquid substances required during ship operation, such as fuel oil, lubricating oil, and cooling water. When there is a leakage in the ship engine room pipeline system, it will seriously affect the safety of ship navigation. Therefore, timely and accurate detection of ship pipeline leakage is the basis for ensuring the safety of ship navigation.

[0003] Currently, neural network models are widely used in the detection of many industrial equipment. In the detection of ship pipeline leakage, compared with judging based on the abnormalities in relevant instrument data after pipeline leakage occurs, the neural network model can detect abnormalities at an early stage of pipeline leakage, which can effectively ensure the safety of ship navigation. However, due to the complex engine room environment, especially during ship navigation, the ship pipeline leakage sensing data that can be obtained is limited. In the case of limited sample data, it will significantly affect the accuracy of the ship pipeline leakage detection network. At the same time, there are often spatio-temporal connections between the pipeline sensing data during ship operation. For example, the detection data of the cooling water pipeline around the engine is usually significantly affected by the operating power of the ship engine. Existing neural network models often do not consider these spatio-temporal correlations between pipeline sensing data, which also reduces the detection accuracy to a certain extent.

[0004] Therefore, the existing technology has the problem that in the detection of ship pipeline leakage, due to the small amount of data and the lack of consideration of the spatio-temporal correlations between sensing data, the detection accuracy of the ship pipeline leakage detection network is relatively low, and improvement is needed. Summary of the Invention

[0005] In view of this, it is necessary to provide a method for training a ship pipeline leakage detection network, an application method, and a pipeline detector to solve the problem in the existing technology that due to the small amount of data and the lack of consideration of the spatio-temporal correlations between sensing data, the detection accuracy of the ship pipeline leakage detection network is relatively low.

[0006] To solve the above problems, on the one hand, the present invention provides a method for training a ship pipeline leakage detection network, including: Performing physical modeling analysis on the obtained ship pipeline sensing data to obtain physical derivative data, and fusing the physical derivative data and the ship pipeline sensing data to obtain fusion training data; Input the fused training data into the initial ship pipeline leakage detection network, perform multi-layer graph convolution on the fused training data to obtain spatial information features, perform spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features, fuse the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, perform prediction on the spatio-temporal fusion features to obtain the pipeline leakage prediction output, determine the prediction loss according to the pipeline leakage prediction output, and iteratively train according to the prediction loss to obtain a trained and complete ship pipeline leakage detection network.

[0007] In a possible implementation, perform physical modeling analysis on the obtained ship pipeline sensing data to obtain physical derivative data, including: Perform physical modeling analysis on the obtained ship pipeline sensing data based on the Hazen-Williams equation and the basic principle of heat transfer to obtain physical derivative data; Among them, the physical derivative data includes temperature difference value and pressure drop value.

[0008] In a possible implementation, performing multi-layer graph convolution on the fused training data to obtain spatial information features includes: Perform normalization processing and non-linear operation on the fused training data to obtain preprocessing features; Perform graph convolution and pooling operations on the preprocessing features to obtain the first graph convolution feature, and perform graph convolution and pooling operations on the first graph convolution feature to obtain the second graph convolution feature; After fusing the first graph convolution feature and the second graph convolution feature, perform batch normalization, non-linear operation and Dropout operation to obtain spatial information features.

[0009] In a possible implementation, performing spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features includes: Perform spatio-temporal position embedding on the spatial information features to obtain an initial embedding vector; Perform multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector, and perform multi-head attention enhanced feature decoding on the encoded feature vector to obtain a decoded feature vector; Perform linear processing on the decoded feature vector to obtain spatio-temporal information features.

[0010] In a possible implementation, performing multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector includes: Perform multi-head attention feature extraction, residual connection of the initial embedding vector and layer normalization on the initial embedding vector to obtain the first encoded feature; Perform feed-forward neural network feature extraction, residual connection of the first encoded feature and layer normalization on the first encoded feature to obtain an encoded feature vector.

[0011] In a possible implementation, performing multi-head attention enhancement feature decoding on the encoded feature vector to obtain a decoded feature vector, including: Performing masked multi-head attention feature extraction, residual connection of the encoded feature vector, and layer normalization on the encoded feature vector to obtain a first decoded feature; Performing multi-head attention feature extraction, residual connection of the first decoded feature, and layer normalization on the first decoded feature to obtain a second decoded feature; Performing feed-forward neural network feature extraction, residual connection of the second decoded feature, and layer normalization on the second decoded feature to obtain a decoded feature vector.

[0012] In a possible implementation, predicting the spatio-temporal fusion feature to obtain a pipeline leakage prediction output, and determining a prediction loss according to the pipeline leakage prediction output, including: Performing a fully connected layer mapping on the spatio-temporal fusion feature to obtain a pipeline leakage prediction output; Determining a prediction loss between the pipeline leakage prediction output and the corresponding true leakage result based on the cross-entropy loss function.

[0013] On the other hand, the present invention provides a method for applying a ship pipeline leakage detection network, including: Obtaining real-time sensing data of a ship pipeline to be detected, performing physical modeling analysis on the real-time sensing data to obtain real-time derived data, and fusing the real-time sensing data and the real-time derived data to obtain real-time input data; Inputting the real-time input data into a trained ship pipeline leakage detection network to obtain a pipeline leakage prediction result; Wherein, the trained ship pipeline leakage detection network is determined according to the above-mentioned ship pipeline leakage detection network training method.

[0014] On the other hand, the present invention provides a pipeline detector, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above-mentioned ship pipeline leakage detection network training method and / or the above-mentioned ship pipeline leakage detection network application method are implemented.

[0015] On the other hand, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned ship pipeline leakage detection network training method and / or the above-mentioned ship pipeline leakage detection network application method are implemented.

[0016] The beneficial effects of the present invention are as follows: In the method for training a ship pipeline leakage detection network provided by the present invention, first, physical modeling analysis is performed on the obtained ship pipeline sensing data to obtain physically derived data, and the physically derived data and the ship pipeline sensing data are fused to obtain fused training data; then, the fused training data is input into the initial ship pipeline leakage detection network, multi-layer graph convolution is performed on the fused training data to obtain spatial information features, spatio-temporal information attention enhancement is performed on the spatial information features to obtain spatio-temporal information features, the spatial information features and the spatio-temporal information features are fused to obtain spatio-temporal fusion features, prediction is performed on the spatio-temporal fusion features to obtain a pipeline leakage prediction output, a prediction loss is determined according to the pipeline leakage prediction output, and an iteratively trained ship pipeline leakage detection network is obtained according to the prediction loss. By generating physically derived data through physical modeling analysis, the data dimension and data volume can be enriched. Spatial information in the data is captured through multi-layer graph convolution, and long and short time and spatial relationships in the data are captured through spatio-temporal information attention enhancement, so that the spatio-temporal correlation in the data can be fully mined, and the accuracy of ship pipeline leakage detection is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those skilled in the art can obtain other drawings without creative efforts based on these drawings.

[0018] Figure 1 It is a schematic flowchart of an embodiment of the method for training a ship pipeline leakage detection network provided by the present invention; Figure 2 It is a schematic flowchart of the multi-layer graph convolution in the embodiment of the present invention; Figure 3 It is a schematic flowchart of the spatio-temporal information attention enhancement in the embodiment of the present invention; Figure 4 It is a schematic flowchart of the multi-head attention enhancement feature encoding in the embodiment of the present invention; Figure 5 It is a schematic flowchart of the multi-head attention enhancement feature decoding in the embodiment of the present invention; Figure 6 It is a schematic flowchart of calculating the prediction loss in the embodiment of the present invention; Figure 7 It is a schematic flowchart of an embodiment of the method for applying a ship pipeline leakage detection network provided by the present invention; Figure 8 It is a schematic structural diagram of an embodiment of the pipeline detector provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0020] In the description of the embodiments of the present invention, unless otherwise specified, the meaning of "a plurality" is two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0021] The descriptions such as "first" and "second" involved in the embodiments of the present invention are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly specifying the quantity of the indicated technical features. Therefore, the technical features defined with "first" and "second" may explicitly or implicitly include at least one such feature.

[0022] Referring to "embodiments" in this article means that the specific features, structures, or characteristics described in connection with the embodiments may be included in at least one embodiment of the present invention. The phrase appears in various positions in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0023] The present invention provides a method for training a ship pipeline leakage detection network, an application method, and a pipeline detector, which will be described separately below.

[0024] It should be noted that the ship pipeline leakage detection method and application method provided in this embodiment can be applied to the pipeline leakage detection systems of various types of liquid transportation pipelines such as fuel pipelines, lubricating oil pipelines, and cooling water pipelines in the ship system. The pipeline leakage detection system can be a software system running on a terminal device, and the terminal device can be a server, a tablet computer, a laptop computer, a shipborne control center, or a mobile phone, etc. The specific type of the terminal device is not limited in the embodiments of the present application. Figure 1 As shown in Figure 1 the following, for an embodiment of the method for training a ship pipeline leakage detection network provided by the present invention, the ship pipeline leakage detection network training method includes: S101. Perform physical modeling analysis on the obtained ship pipeline sensing data to obtain physically derived data, and fuse the physically derived data and the ship pipeline sensing data to obtain fused training data; S102. Input the fused training data into the initial ship pipeline leakage detection network, perform multi-layer graph convolution on the fused training data to obtain spatial information features, perform spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features, fuse the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, perform prediction on the spatio-temporal fusion features to obtain pipeline leakage prediction outputs, determine the prediction loss according to the pipeline leakage prediction outputs, and iteratively train according to the prediction loss to obtain a well-trained ship pipeline leakage detection network.

[0025] Among them, step S101 is specifically as follows: Install sensors such as flow rate, pressure, and vibration at the inlet and outlet of the ship pipeline and at key pipeline positions, and set a predetermined acquisition frequency to collect the mass flow rate, inlet and outlet pressure, and vibration data at the inlet and outlet of the pipeline during normal operation of the ship, and generate physically derived data through physical modeling analysis. The collected data and the physically derived data are fused and then used as training data to be input into the initial ship pipeline leakage detection network for training.

[0026] Compared with the prior art, the ship pipeline leakage detection network training method provided by the present invention first performs physical modeling analysis on the obtained ship pipeline sensing data to obtain physically derived data, and fuses the physically derived data and the ship pipeline sensing data to obtain fused training data; then inputs the fused training data into the initial ship pipeline leakage detection network, performs multi-layer graph convolution on the fused training data to obtain spatial information features, performs spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features, fuses the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, performs prediction on the spatio-temporal fusion features to obtain pipeline leakage prediction outputs, determines the prediction loss according to the pipeline leakage prediction outputs, and iteratively trains according to the prediction loss to obtain a well-trained ship pipeline leakage detection network. The present invention generates physically derived data through physical modeling analysis, which can enrich the data dimension and data volume of the data. Capturing the spatial information in the data through multi-layer graph convolution and capturing the long and short time and spatial relationships in the data through spatio-temporal information attention enhancement can fully exploit the spatio-temporal correlation in the data and effectively improve the accuracy of ship pipeline leakage detection.

[0027] In some embodiments of the present invention, performing physical modeling analysis on the obtained ship pipeline sensing data to obtain physically derived data includes: Performing physical modeling analysis on the obtained ship pipeline sensing data based on the Hazen-Williams equation and the basic principle of heat transfer to obtain physically derived data; Among them, the physically derived data includes temperature difference values and pressure drop values.

[0028] Specifically, during the physical modeling analysis process, considering that when a pipeline leak occurs, the two physical quantities of the temperature and pressure of the liquid transported in the pipeline often change accordingly, the temperature difference of the liquid in the ship pipeline is used in the embodiment and the pressure drop value as physical derivative data and added to the input data. In the embodiment, the Hazen-Williams equation and the basic principle of heat transfer are used to calculate the temperature difference and pressure drop at each data node position, and they are used as new data nodes to enhance the input of the ship pipeline leak detection network. These newly added data nodes carry prior information based on physical principles, effectively expanding and enriching the data dimension and data volume of the input data.

[0029] During the calculation of the temperature difference and the pressure drop value process, for the temperature difference , according to the physical principle of heat transfer, the embodiment uses the formula to calculate the temperature difference through the mass flow rate, specific heat capacity, and heat transfer rate, that is:

[0030] where, represents the heat transfer rate, represents the mass flow rate, represents the specific heat capacity.

[0031] For the pressure drop value, the pressure drop value is calculated according to the Hazen-Williams equation through the pipeline length, volume flow rate, pipeline diameter, and Hazen-Williams roughness coefficient, that is:

[0032] where, represents the pipeline length, represents the volume flow rate, represents the Hazen-Williams roughness coefficient, represents the pipeline diameter.

[0033] Finally, the physical derivative data and the sensing data are uniformly preprocessed so that they can be applied to subsequent calculations and model inputs, and based on a preset fusion strategy, the data is fused and used as the fusion training data of the ship pipeline leak detection network, enabling the physical derivative data to be embedded and represented in the same way as other data nodes in subsequent model calculations and to be represented and calculated in the same vector space.

[0034] In some embodiments of the present invention, Figure 2 is a schematic flow diagram of the multi-layer graph convolution of the embodiment of the present invention, as shown in Figure 2As shown in the figure, multi-layer graph convolution is performed on the fused training data to obtain spatial information features, including: S201. Perform normalization processing and non-linear operation on the fused training data to obtain preprocessed features; S202. Perform graph convolution and pooling operations on the preprocessed features to obtain the first graph convolution features, and perform graph convolution and pooling operations on the first graph convolution features to obtain the second graph convolution features; S203. After fusing the first graph convolution features and the second graph convolution features, perform batch normalization, non-linear operation and Dropout operation to obtain spatial information features.

[0035] Specifically, in order to fully capture the spatio-temporal correlation in the data, the embodiment extracts the spatial information in the data through multi-layer graph convolution, and extracts the complex relationships of long and short time and space in the data through a Transformer network including a multi-head attention mechanism, so as to effectively utilize the spatio-temporal correlation of the input data and improve the accuracy of pipeline leakage detection.

[0036] In the multi-layer graph convolution feature extraction, the embodiment first performs data normalization processing and non-linear operation on the fused input data to obtain preprocessed features. The embodiment sets two consecutive graph convolution layers, and each layer includes a graph convolution operation and a pooling operation. In each graph convolution layer, the data node updates its own feature representation according to the information of its neighboring nodes, gradually fuses local and global spatial features, and uses the output of the first graph convolution layer as the input of the next graph convolution layer. Finally, considering the scarcity of leakage data, in order to better retain the original features, the graph convolution features obtained by two graph convolutions are fused through skip connection, and the fused information is processed by batch normalization and non-linear operation, and a Dropout operation is added to prevent overfitting.

[0037] In some embodiments of the present invention, Figure 3 is a schematic flow chart of spatio-temporal information attention enhancement according to an embodiment of the present invention. As Figure 3 shown, spatio-temporal information attention enhancement is performed on the spatial information features to obtain spatio-temporal information features, including: S301. Perform spatio-temporal position embedding on the spatial information features to obtain an initial embedding vector; S302. Perform multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector, and perform multi-head attention enhanced feature decoding on the encoded feature vector to obtain a decoded feature vector; S303. Perform linear processing on the decoded feature vector to obtain spatio-temporal information features.

[0038] Specifically, to capture the spatio-temporal correlation between sensing data, the embodiment adopts a Transformer network structure to extract spatio-temporal information. In the Transformer network, first, a spatio-temporal position embedding operation is performed on the spatial information features to generate a continuous vector according to the temporal and position relationships in the feature information, obtaining an initial embedding vector. Then, the initial embedding vector is input into a feature encoding module and a feature decoding module that contain a multi-head attention mechanism for encoding and decoding operations. Finally, the decoded feature vector after decoding is passed through a linear process and a Softmax activation function to obtain spatio-temporal information features. In feature encoding and feature decoding, by using the multi-head attention mechanism, it simultaneously focuses on different position information in the sequence, captures the complex relationships of data in long and short time and space, and through self-attention calculation, assigns different weights to each position, highlighting the features that have an important impact on fault diagnosis and further optimizing the feature representation.

[0039] In some embodiments of the present invention, Figure 4 is a schematic flow diagram of multi-head attention enhanced feature encoding according to an embodiment of the present invention. As Figure 4 shown, performing multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector includes: S401. Perform multi-head attention feature extraction on the initial embedding vector, perform residual connection on the initial embedding vector and layer normalization to obtain a first encoded feature; S402. Perform feed-forward neural network feature extraction on the first encoded feature, perform residual connection on the first encoded feature and layer normalization to obtain an encoded feature vector.

[0040] Specifically, the feature encoding module contains two levels. For the input initial embedding vector, in the first level, the multi-head attention mechanism is first used to simultaneously focus on different position information in the sequence, capture the complex relationships of data in long and short time and space, extract the more critical features among them, and perform residual connection fusion and layer normalization on the features obtained by the multi-head attention mechanism and the initial embedding vector input to the multi-head attention to obtain a first encoded feature, which is used as the input for the next level. The next level performs feature extraction through a feed-forward neural network, and similarly performs residual connection fusion and layer normalization on the features output by the feed-forward neural network and the first encoded feature input to the feed-forward neural network to obtain the encoded feature vector output by the feature encoding module. In addition, a Dropout mechanism is added to each level of the feature encoding module in the embodiment to prevent overfitting during training.

[0041] In some embodiments of the present invention, Figure 5 is a schematic flow diagram of multi-head attention enhanced feature decoding according to an embodiment of the present invention. As Figure 5As shown in the figure, the encoded feature vector is subjected to multi-head attention enhanced feature decoding to obtain a decoded feature vector, including: S501. Perform masked multi-head attention feature extraction on the encoded feature vector, perform residual connection on the encoded feature vector, and perform layer normalization to obtain a first decoded feature; S502. Perform multi-head attention feature extraction on the first decoded feature, perform residual connection on the first decoded feature, and perform layer normalization to obtain a second decoded feature; S503. Perform feed-forward neural network feature extraction on the second decoded feature, perform residual connection on the second decoded feature, and perform layer normalization to obtain a decoded feature vector.

[0042] Specifically, the feature decoding module includes three levels. The first level is a masked multi-head self-attention mechanism, and the second and third levels are similar multi-head attention mechanisms and feed-forward neural networks as in the feature encoding module. The masked multi-head self-attention mechanism is to assign different weights to each position to increase the weights of features that are more important for fault detection, and then perform the multi-head attention mechanism again to further strengthen the information of different positions in the sequence. In the feature decoding module, for each level, similar to the feature encoding module, the features input to each module and the features obtained after processing and output are fused through residual connection, and are used as the input of the next level after layer normalization.

[0043] In some embodiments of the present invention, Figure 6 is a schematic flowchart of calculating the prediction loss of the embodiments of the present invention. As Figure 6 shown, the spatio-temporal fusion feature is predicted to obtain a pipeline leakage prediction output, and the prediction loss is determined according to the pipeline leakage prediction output, including: S601. Perform a fully connected layer mapping on the spatio-temporal fusion feature to obtain a pipeline leakage prediction output; S602. Determine the prediction loss between the pipeline leakage prediction output and the corresponding true leakage result based on the cross-entropy loss function.

[0044] Specifically, considering that the pipeline leakage prediction is essentially a classification prediction problem, the embodiments perform some post-processing operations on the spatio-temporal fusion feature that combines spatio-temporal correlation, map it to class probabilities through a fully connected layer, and then obtain the leakage state of the predicted pipeline according to the threshold judgment.

[0045] When calculating the prediction loss of the ship pipeline leakage detection network during the training process, the embodiments adopt the cross-entropy loss function, which is more conducive to the training of the model. The calculation of the prediction loss can be expressed by the formula:

[0046] Wherein, represents the number of classes, Denotes the true label, which represents the probability of the model's predicted output.

[0047] In summary, in order to obtain a ship pipeline leakage detection network with higher prediction accuracy, the present invention first performs physical modeling analysis on the acquired ship pipeline sensing data to obtain physical derivative data, and fuses the physical derivative data and the ship pipeline sensing data to obtain fused training data; then inputs the fused training data into the initial ship pipeline leakage detection network, performs multi-layer graph convolution on the fused training data to obtain spatial information features, performs spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features, fuses the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, performs prediction on the spatio-temporal fusion features to obtain pipeline leakage prediction output, determines the prediction loss according to the pipeline leakage prediction output, and iteratively trains according to the prediction loss to obtain a trained and complete ship pipeline leakage detection network. By generating physical derivative data through physical modeling analysis, the present invention can enrich the data dimension and data volume of the data. By using multi-layer graph convolution to capture the spatial information in the data and spatio-temporal information attention enhancement to capture the long and short time and spatial relationships in the data, the spatio-temporal correlation in the data can be fully mined, effectively improving the accuracy of ship pipeline leakage detection.

[0048] In addition, the present invention also provides a method for applying a ship pipeline leakage detection network. Combining Figure 7 to look, Figure 7 is a schematic flowchart of an embodiment of the method for applying a ship pipeline leakage detection network provided by the present invention. As Figure 7 shown, the method for applying a ship pipeline leakage detection network includes: S701. Obtain the real-time sensing data of the ship pipeline to be detected, perform physical modeling analysis on the real-time sensing data to obtain real-time derivative data, and fuse the real-time sensing data and the real-time derivative data to obtain real-time input data; S702. Input the real-time input data into the trained and complete ship pipeline leakage detection network to obtain a pipeline leakage prediction result; Among them, the trained and complete ship pipeline leakage detection network is determined according to the above-mentioned ship pipeline leakage detection network training method.

[0049] In the embodiment of the present invention, first, effectively obtain the real-time sensing data of the ship pipeline to be detected, then perform physical modeling analysis for data derivation and data fusion to obtain real-time input data, and then use the above-mentioned trained and complete ship pipeline leakage detection network to effectively predict the pipeline leakage situation, and the pipeline leakage prediction result can be output.

[0050] Such as Figure 8As shown, the present invention also correspondingly provides a pipeline detector 800, which includes a sensor 801, a processor 802, a memory 803, and a display 804. Figure 8 Only some components of the pipeline detector 800 are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0051] In some embodiments, the sensor 801 can be a mass flow sensor, a pressure sensor, a vibration sensor, etc. disposed at the inlet and outlet of the pipeline to be measured and at key positions, for collecting sensing data and sending it to the processor 802 and the memory 803, and storing the collected sensing data and the data processed by the processor 802 in the memory 803.

[0052] In some embodiments, the processor 802 can be a central processing unit (CPU), a microprocessor, or other data processing chips, for running the program code stored in the memory 803 or processing data, such as implementing the pipeline leakage detection program of the ship pipeline leakage detection network training method and / or the ship pipeline leakage detection network application method in the present invention.

[0053] In some embodiments, the memory 803 can be an internal storage unit of the pipeline detector 800, such as the hard disk or memory of the pipeline detector 800. In some other embodiments, the memory 803 can also be an external storage device of the pipeline detector 800, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the pipeline detector 800.

[0054] In some embodiments, the display 804 can be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (organic light-emitting diode) toucher, etc. The display 804 is used to display information of the pipeline detector 800 and to display a visual user interface. In some embodiments, the display 804 can also include a sound module or a vibration module. When the processor 802 detects a pipeline leakage, it transmits an alarm control signal to the display 804, and after receiving the alarm control signal, the display 804 emits an alarm signal through but not limited to sound, light, and vibration. The components 801 - 804 of the pipeline detector 800 communicate with each other through a system bus.

[0055] In one embodiment, when the processor 802 executes the pipeline leakage detection program in the memory 803, the following steps can be implemented: Perform physical modeling analysis on the acquired ship pipeline sensing data to obtain physically derived data, and fuse the physically derived data and the ship pipeline sensing data to obtain fused training data; Input the fused training data into the initial ship pipeline leakage detection network, perform multi-layer graph convolution on the fused training data to obtain spatial information features, perform spatio-temporal information attention enhancement on the spatial information features to obtain spatio-temporal information features, fuse the spatial information features and the spatio-temporal information features to obtain spatio-temporal fusion features, perform prediction on the spatio-temporal fusion features to obtain pipeline leakage prediction output, determine the prediction loss according to the pipeline leakage prediction output, and iteratively train according to the prediction loss to obtain a trained complete ship pipeline leakage detection network.

[0056] and / or implement the following steps: Perform a fully connected layer mapping on the spatio-temporal fusion features to obtain pipeline leakage prediction output; Determine the prediction loss between the pipeline leakage prediction output and the corresponding true leakage result based on the cross-entropy loss function.

[0057] It should be understood that when the processor 802 executes the pipeline leakage detection program in the memory 803, in addition to the above functions, other functions can also be implemented. For details, please refer to the description of the corresponding method embodiments above.

[0058] Correspondingly, an embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium is used to store computer-readable programs or instructions. When the programs or instructions are executed by a processor, the steps or functions in the ship pipeline leakage detection network training method and / or the ship pipeline leakage detection network application method provided by the above method embodiments can be implemented.

[0059] Those skilled in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The computer program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a disk, an optical disc, a read-only memory, or a random access memory, etc.

[0060] The above has introduced in detail the ship pipeline leakage detection network training method, application method and pipeline detector provided by the present invention. Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A ship pipeline leakage detection network training method, characterized in that: include: Performing physical modeling analysis on the acquired ship pipeline sensor data to obtain physical derived data, and fusing the physical derived data with the ship pipeline sensor data to obtain fused training data; The fused training data is input into an initial ship pipeline leakage detection network, multi-layer graph convolution is performed on the fused training data to obtain spatial information features, spatiotemporal information attention enhancement is performed on the spatial information features to obtain spatiotemporal information features, the spatial information features and the spatiotemporal information features are fused to obtain spatiotemporal fusion features, the spatiotemporal fusion features are predicted to obtain pipeline leakage prediction output, prediction loss is determined according to the pipeline leakage prediction output, and iterative training is performed according to the prediction loss to obtain a fully trained ship pipeline leakage detection network.

2. The ship pipeline leakage detection network training method according to claim 1 is characterized in that: The physical modeling analysis of the acquired ship pipeline sensor data to obtain physical derived data includes: Based on the Hasson-Williams equation and the basic principle of heat transfer, physical modeling and analysis are performed on the acquired ship pipeline sensor data to obtain physical derived data; The physically derived data include temperature difference and pressure drop.

3. The ship pipeline leakage detection network training method according to claim 1 is characterized in that: The step of performing multi-layer graph convolution on the fused training data to obtain spatial information features includes: Performing normalization and nonlinear operation on the fused training data to obtain preprocessing features; Performing graph convolution and pooling operations on the preprocessed features to obtain a first graph convolution feature, and performing graph convolution and pooling operations on the first graph convolution feature to obtain a second graph convolution feature; After fusing the first graph convolutional features and the second graph convolutional features, batch normalization, nonlinear operation and Dropout operation are performed to obtain spatial information features.

4. The ship pipeline leakage detection network training method according to claim 1 is characterized in that: The step of performing spatiotemporal information attention enhancement on the spatial information feature to obtain the spatiotemporal information feature includes: Performing spatiotemporal position embedding on the spatial information feature to obtain an initial embedding vector; Performing multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector, and performing multi-head attention enhanced feature decoding on the encoded feature vector to obtain a decoded feature vector; The decoded feature vector is linearly processed to obtain spatiotemporal information features.

5. The ship pipeline leakage detection network training method according to claim 4 is characterized in that: The performing multi-head attention enhanced feature encoding on the initial embedding vector to obtain an encoded feature vector includes: Performing multi-head attention feature extraction on the initial embedding vector, residual connection of the initial embedding vector and layer normalization to obtain a first encoding feature; Perform feedforward neural network feature extraction on the first coding feature, residually connect the first coding feature and layer normalization to obtain a coding feature vector.

6. The ship pipeline leakage detection network training method according to claim 4 is characterized in that: The performing multi-head attention enhanced feature decoding on the encoded feature vector to obtain a decoded feature vector includes: Performing masked multi-head attention feature extraction on the encoded feature vector, residually connecting the encoded feature vector and layer normalization to obtain a first decoding feature; Performing multi-head attention feature extraction on the first decoding feature, residual connection of the first decoding feature and layer normalization to obtain a second decoding feature; The second decoding feature is subjected to feedforward neural network feature extraction, residual connection of the second decoding feature and layer normalization to obtain a decoding feature vector.

7. The ship pipeline leakage detection network training method according to claim 1 is characterized in that: The step of predicting the spatiotemporal fusion features to obtain a pipeline leakage prediction output, and determining a prediction loss according to the pipeline leakage prediction output, comprises: Performing full connection layer mapping on the spatiotemporal fusion features to obtain pipeline leakage prediction output; The prediction loss of the pipeline leakage prediction output and the corresponding actual leakage result is determined based on the cross entropy loss function.

8. A network application method for ship pipeline leakage detection, characterized in that: include: Acquire real-time sensor data of the ship pipeline to be inspected, perform physical modeling analysis on the real-time sensor data to obtain real-time derived data, and fuse the real-time sensor data and the real-time derived data to obtain real-time input data; Inputting the real-time input data into a well-trained ship pipeline leakage detection network to obtain a pipeline leakage prediction result; Wherein, the fully trained ship pipeline leakage detection network is determined according to the ship pipeline leakage detection network training method according to any one of claims 1 to 7.

9. A pipeline detector, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the ship pipeline leakage detection network training method according to any one of claims 1 to 7 and / or the ship pipeline leakage detection network application method according to claim 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the ship pipeline leakage detection network training method according to any one of claims 1 to 7 and / or the ship pipeline leakage detection network application method according to claim 8 are implemented.

Images