CN116447528A - Pipeline oil gas leakage detection method based on graphic neural network and LSTM network - Google Patents
Pipeline oil gas leakage detection method based on graphic neural network and LSTM network Download PDFInfo
- Publication number
- CN116447528A CN116447528A CN202310355850.2A CN202310355850A CN116447528A CN 116447528 A CN116447528 A CN 116447528A CN 202310355850 A CN202310355850 A CN 202310355850A CN 116447528 A CN116447528 A CN 116447528A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- neural network
- sample
- optical fiber
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 53
- 238000001514 detection method Methods 0.000 title claims abstract description 26
- 239000013307 optical fiber Substances 0.000 claims abstract description 46
- 230000004927 fusion Effects 0.000 claims abstract description 26
- 238000000547 structure data Methods 0.000 claims abstract description 20
- 238000000034 method Methods 0.000 claims abstract description 17
- 230000002159 abnormal effect Effects 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 239000000835 fiber Substances 0.000 claims description 17
- 230000002776 aggregation Effects 0.000 claims description 16
- 238000004220 aggregation Methods 0.000 claims description 16
- 238000010606 normalization Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000007477 logistic regression Methods 0.000 claims description 3
- 239000007789 gas Substances 0.000 description 21
- 239000003921 oil Substances 0.000 description 17
- 230000006870 function Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 239000000446 fuel Substances 0.000 description 4
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 238000001069 Raman spectroscopy Methods 0.000 description 3
- 238000009529 body temperature measurement Methods 0.000 description 3
- 210000004027 cell Anatomy 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 235000002020 sage Nutrition 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 239000000149 chemical water pollutant Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000000253 optical time-domain reflectometry Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000004931 aggregating effect Effects 0.000 description 1
- 239000003570 air Substances 0.000 description 1
- 239000012080 ambient air Substances 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 239000002737 fuel gas Substances 0.000 description 1
- 239000000295 fuel oil Substances 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000013450 outlier detection Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000004861 thermometry Methods 0.000 description 1
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F17—STORING OR DISTRIBUTING GASES OR LIQUIDS
- F17D—PIPE-LINE SYSTEMS; PIPE-LINES
- F17D5/00—Protection or supervision of installations
- F17D5/02—Preventing, monitoring, or locating loss
- F17D5/06—Preventing, monitoring, or locating loss using electric or acoustic means
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M3/00—Investigating fluid-tightness of structures
- G01M3/002—Investigating fluid-tightness of structures by using thermal means
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Examining Or Testing Airtightness (AREA)
Abstract
The invention discloses a pipeline oil gas leakage detection method based on a graph neural network and an LSTM network, which comprises the steps of collecting optical fiber temperature data and preprocessing, marking abnormal temperature of the data, and cutting into sample sections; converting the sample segment into graph structure data; and constructing a fusion network formed by the graphic neural network and the LSTM network, and training by using graphic structure data to obtain a trained graphic neural network and LSTM network fusion model which is used for detecting leakage of the region to be detected. Compared with the prior art, the method comprehensively considers the spatial correlation and the time sequence of the signals, can accurately detect and position the temperature abnormal event of the urban underground pipe gallery, and has the advantages of high detection precision, strong practicability and high speed.
Description
Technical Field
The invention relates to a pipeline leakage detection method, in particular to a pipeline oil gas leakage detection method based on a graphic neural network and an LSTM network.
Background
The safe operation of utility tunnel is very important to optimizing perfect urban function and promote urban civilization construction, and urban utility tunnel has held urban engineering pipeline including electric power, communication, gas, feedwater, drainage, reclaimed water, traffic safety, urban illumination, landfill leachate pipeline etc. wherein gas, fuel, feedwater, drainage etc. pipeline all have the risk of leaking. If leakage occurs, the utility tunnel is affected, thereby affecting the safety of other lines such as power, communication, illumination, etc.
We have found that a leak of gas or fuel occurs somewhere in the pipe, which causes a temperature fluctuation that is independent of the ambient air pressure. Therefore, whether the pipeline is leaked with fuel gas or fuel oil can be judged according to the fluctuation condition of the temperature of the optical fiber.
Outlier detection techniques play an important role in various application fields. The distributed optical fiber temperature sensing technology is widely applied to the scenes of pipeline leakage detection, power cable monitoring, nuclear facility anomaly monitoring and the like, and a large number of signals at different positions can be acquired each time. In the time sequence signal, algorithms such as Principal Component Analysis (PCA), a classification support vector machine (OC-SVM), local anomaly factor (LOF), histogram-based outlier score (HBOS), isolated forest (IsolationForest) and the like have good effect on detecting outliers, but the detection method only considers the correlation of signals in the time domain and breaks the spatial relationship of the signals. The pipeline oil gas leakage detection method based on the graph neural network and the LSTM network simultaneously considers the time domain correlation and the spatial relation between signals, and improves the detection precision of abnormal temperature.
Noun interpretation:
a distributed optical fiber temperature measurement system, which is called DTS for short and is called as Distributed Temperature Sensing in English. Also known as fiber thermometry, temperature monitoring is achieved based on the Optical Time Domain Reflectometry (OTDR) principle and the sensitivity of Raman (Raman) scattering effects to temperature.
A graph is a data structure with strong expressive power that models a set of objects, here referred to as nodes, and their relationships, referred to as edges. As a unique non-euclidean data structure, graph analysis mainly focuses on the tasks of node classification, link prediction, and clustering. The graph neural network, english Graph Neural Networks, is called GNN for short, is a deep learning-based method, runs on a graph domain, and captures the dependency of a graph through message passing among graph nodes.
The Long-Short-Term Memory neural network, english Short-Term Memory, is abbreviated as LSTM, is a recursion unit of a special model, can solve the problems of Long-Term dependence, gradient disappearance, gradient explosion and the like in the RNN training process, and captures the time sequence characteristics of signals.
Disclosure of Invention
The invention aims to provide the pipeline oil gas leakage detection method based on the graph neural network and the LSTM network, which can solve the problems, simultaneously consider the time domain correlation and the spatial relation between signals and improve the detection accuracy of abnormal temperature.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: a pipeline oil gas leakage detection method based on a graph neural network and an LSTM network comprises the following steps of;
(1) Data acquisition and pretreatment;
(11) Collecting optical fiber temperature data on different oil and gas pipelines, wherein each oil and gas pipeline at least comprises a leakage position, the optical fiber temperature data comprises data of the leakage position, the length of each optical fiber temperature data is L data points, each data point comprises temperature data, and the temperature data are one stokes data and one an-stokes data;
(12) Preprocessing optical fiber temperature data, classifying each data point, if the data point is positioned at a leakage position, marking the data point as a positive sample, marking the data point as a negative sample, marking the data point as a 0, marking the data point as a label of the data point, cutting the labeled optical fiber temperature data into equal-length sample sections, and forming a data set by all the sample sections;
(2) Converting the sample segment into graph structure data, including steps (21) - (22);
(21) For a sample segment, taking each data point as a node, forming an edge by the node and the connecting lines of four adjacent data points, and converting the sample segment into graph structure data G= (V, E), wherein V is a set of all nodes, E is a set of all edges, and for each node, the data comprise temperature data and labels of the node;
(22) Converting all sample segments into graph structure data according to the step (21);
(3) Constructing a fusion network, wherein the fusion network comprises a graph neural network and an LSTM network which are sequentially connected;
(4) Sequentially sending the graph structure data into a fusion network for training until the network converges to obtain a trained graph neural network and LSTM network fusion model, wherein one graph structure data process comprises the steps (41) - (43);
(41) Sending the graph structure data into a graph neural network, taking each node as a central node, and outputting an aggregation feature after feature aggregation by an aggregator, wherein each node corresponds to L aggregation features;
(42) Each aggregation feature is respectively sent into an LSTM network to obtain an output, and the output is subjected to binary division and then used as the prediction probability that the node belongs to a positive sample;
(43) Taking the label of the node as a desired output, and adjusting the network weights of the graph neural network and the LSTM network;
(5) Detecting leakage of a region to be detected;
(51) Collecting optical fiber temperature data on the outer wall of an oil gas pipeline to be tested, preprocessing, and cutting into a sample section to be tested, wherein the length of the sample section is the same as that of the sample section;
(52) And sending the sample segment to be detected into a fusion model of the graph neural network and the LSTM network, outputting the prediction probability that each data point belongs to a positive sample on the sample segment to be detected, presetting a threshold value, and if the prediction probability is larger than the threshold value, determining that the data point is abnormal temperature, otherwise, determining that the data point is normal temperature.
As preferable: in the step (11), the optical fiber temperature data are collected on different oil and gas pipelines specifically, the optical fiber temperature data are collected through a distributed optical fiber temperature measuring system, the distributed optical fiber temperature measuring system comprises sensing optical fibers, and the sensing optical fibers are distributed along the length direction of the pipelines and pass through the leakage positions on the oil and gas pipelines.
As preferable: and (12) preprocessing the fiber temperature data, namely, the fiber temperature data comprise L data points, L pieces of stokes data and L pieces of an-stokes data, a stokes data vector and an-stokes data vector are respectively formed, and for each vector, normalization processing is respectively carried out on each element in the vector.
As preferable: in the step (42), a logistic regression method Sigmoid is adopted to carry out binary division, specifically; the output end of the fusion network is connected with a Sigmoid layer;
the Sigmoid layer converts the output result of the LSTM network into the prediction probability that the node belongs to a positive sample, calculates the prediction probability and the loss value of the label of the positive sample through the cross entropy loss function BECLoss, executes back propagation and updating weight and iterates training, and enables the loss value not to be reduced until convergence, so that a trained fusion model of the graphic neural network and the LSTM network is obtained.
As preferable: the graph neural network is GraphSAGE, PNA, GCN, GAT or GAE.
Compared with the prior art, the invention has the advantages that:
the invention provides a novel method for positioning abnormal temperature, which collects temperature data of a real-time oil pipe and an air pipe through a distributed optical fiber temperature measuring system, analyzes whether abnormal fluctuation occurs in the temperature in a pipeline to judge whether leakage exists at the position, and can accurately position the leakage position of the pipeline of the underground comprehensive pipe gallery so as to ensure safe operation of the underground comprehensive pipe gallery.
The invention provides a novel abnormal temperature positioning method based on the combination of a graph neural network and a long-term memory neural network, which is provided by the invention under the condition that original signals are not converted, the spatial relationship existing between the signals is explicitly established in a graph mode, specifically, the spatial relationship of each signal node is established by adopting the graph neural network, the spatial characteristics of the signals are captured, the time domain characteristics of the signals are captured by adopting an LSTM network, and the abnormal condition of the signals is judged by a Sigmoid function according to the captured spatial characteristics and the time domain characteristics, so that the effective positioning of the abnormal temperature event is obtained.
Compared with the independent graph neural network detection method and the abnormality detection algorithm which only considers time domain, the detection accuracy of the abnormal temperature can be improved after the processing of the method.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a diagram of spatial relationships of nodes in the structured data of FIG. 2;
FIG. 3 is a block diagram of a converged network;
FIG. 4 is a block diagram of an LSTM network;
FIG. 5 is a graph of detection accuracy of the fusion model of the neural network and the LSTM network.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
Example 1: referring to fig. 1 to 4, a pipeline oil gas leakage detection method based on a graph neural network and an LSTM network includes the following steps;
(1) Data acquisition and pretreatment;
(11) Collecting optical fiber temperature data on different oil and gas pipelines, wherein each oil and gas pipeline at least comprises a leakage position, the optical fiber temperature data comprises data of the leakage position, the length of each optical fiber temperature data is L data points, each data point comprises temperature data, and the temperature data are one stokes data and one an-stokes data;
(12) Preprocessing optical fiber temperature data, classifying each data point, if the data point is positioned at a leakage position, marking the data point as a positive sample, marking the data point as a negative sample, marking the data point as a 0, marking the data point as a label of the data point, cutting the labeled optical fiber temperature data into equal-length sample sections, and forming a data set by all the sample sections;
(2) Converting the sample segment into graph structure data, including steps (21) - (22);
(21) For a sample segment, taking each data point as a node, forming an edge by the node and the connecting lines of four adjacent data points, and converting the sample segment into graph structure data G= (V, E), wherein V is a set of all nodes, E is a set of all edges, and for each node, the data comprise temperature data and labels of the node;
(22) Converting all sample segments into graph structure data according to the step (21);
(3) Constructing a fusion network, wherein the fusion network comprises a graph neural network and an LSTM network which are sequentially connected;
(4) Sequentially sending the graph structure data into a fusion network for training until the network converges to obtain a trained graph neural network and LSTM network fusion model, wherein one graph structure data process comprises the steps (41) - (43);
(41) Sending the graph structure data into a graph neural network, taking each node as a central node, and outputting an aggregation feature after feature aggregation by an aggregator, wherein each node corresponds to L aggregation features;
(42) Each aggregation feature is respectively sent into an LSTM network to obtain an output, and the output is subjected to binary division and then used as the prediction probability that the node belongs to a positive sample;
(43) Taking the label of the node as a desired output, and adjusting the network weights of the graph neural network and the LSTM network;
(5) Detecting leakage of a region to be detected;
(51) Collecting optical fiber temperature data on the outer wall of an oil gas pipeline to be tested, preprocessing, and cutting into a sample section to be tested, wherein the length of the sample section is the same as that of the sample section;
(52) And sending the sample segment to be detected into a fusion model of the graph neural network and the LSTM network, outputting the prediction probability that each data point belongs to a positive sample on the sample segment to be detected, presetting a threshold value, and if the prediction probability is larger than the threshold value, determining that the data point is abnormal temperature, otherwise, determining that the data point is normal temperature.
In this embodiment, in step (11), the optical fiber temperature data is collected on different oil and gas pipelines specifically through a distributed optical fiber temperature measurement system, where the distributed optical fiber temperature measurement system includes a sensing optical fiber, and the sensing optical fiber is distributed along the length direction of the pipeline and passes through the leakage position on the oil and gas pipeline.
And (12) preprocessing the fiber temperature data, namely, the fiber temperature data comprise L data points, L pieces of stokes data and L pieces of an-stokes data, a stokes data vector and an-stokes data vector are respectively formed, and for each vector, normalization processing is respectively carried out on each element in the vector.
In the step (42), a logistic regression method Sigmoid is adopted to carry out binary division, specifically; the output end of the fusion network is connected with a Sigmoid layer;
the Sigmoid layer converts the output result of the LSTM network into the prediction probability that the node belongs to a positive sample, calculates the prediction probability and the loss value of the label of the positive sample through the cross entropy loss function BECLoss, executes back propagation and updating weight and iterates training, and enables the loss value not to be reduced until convergence, so that a trained fusion model of the graphic neural network and the LSTM network is obtained.
The graphic neural network is GraphSAGE, PNA, GCN, GAT or GAE.
Before the pretreatment of the fiber temperature data in step (12) in this embodiment, we can also filter and clean the fiber temperature data obtained in step (11), remove abnormal and erroneous data, and supplement the length.
In this embodiment, the step (43) adjusts the criteria of the network weights of the neural network and the LSTM network as follows: and for each node, taking the label of the node as a desired output, and adjusting the network weights in the graph neural network and the LSTM network with the aim of reducing the difference between the prediction probability and the desired output.
Example 2: referring to fig. 1 to 5, on the basis of embodiment 1, we give a specific scenario.
Regarding step (11), when data acquisition is performed, we need to acquire a large amount of fiber temperature data in different scenes and different environments. For example, one of the scenarios is an area utility tunnel that contains city engineering lines including electricity, communications, gas, water supply, water drainage, reclaimed water, traffic safety, city lighting, landfill leachate lines, etc., where we collect the collection fiber temperature data on the gas pipeline. In this embodiment, each fiber temperature data length is l=2000 data points; and cleaning the fiber temperature data, and screening out abnormal temperature of the leakage position.
The preprocessing of the fiber temperature data in step (12) is referred to as normalization, because different characteristics of the data have different dimensions and dimension units, which affect the result of data analysis, and data normalization is required to eliminate the dimensional influence between indexes. The present embodiment scales each feature value using the MinMaxScale () function in the sklearn library, normalizing the data between [0,1 ].
Regarding step (12), the labeled fiber temperature data is cut into equal length sample segments, where each fiber temperature data is cut into 20 sample segments, each sample segment contains 100 data points, if the fiber temperature data has 1000, the total number of sample segments is 200000, all sample segments form a dataset, and are proportionally divided into a training set and a test set, and during subsequent model training, the sample segments in the training set are used for training, and the sample segments in the test set are used for testing.
In the step (2), after the optical fiber is affected by temperature, the heat of the heated center of the optical fiber diffuses to the periphery and affects the nearby points, so that the intensity of the spontaneous raman scattered optical signal in the optical fiber is changed, and the temperature of the peripheral node is changed. Due to the spatial continuity of the optical fiber, there is some relationship between the intensity of the optical signal. The concept of a graph can help us understand this relationship by defining each data point as a node that forms an edge with the links between four adjacent data points, as shown in fig. 2, so that one of the sample segments can be converted into a graph containing spatial information, structural data g= (V, E), which has a spatial relationship with four adjacent data points for one node, and effectively retains structural information between node signals on the optical fiber for the graph structural data, which has not only a temporal correlation but also a spatial relationship.
Regarding the fusion module in the step (3), the idea of the graph neural network, specifically GraphSAGE, graphSAGE algorithm, is that the central node continuously aggregates the information of the neighbor nodes, and as the iteration number increases, the information aggregated by each node is almost global, so that the graph neural network constructs the spatial relationship between the signals on the optical fiber. In the network, an average aggregator is adopted by an aggregator, k aggregators are adopted, k layers of aggregation are carried out in total, and the aggregation characteristic corresponding to each node is represented by the ebedding of the k layerWhere v is the current central node. After the K-layer graph SAGE features of the graph neural network are aggregated, one email can be obtained for each central node, the information in the K-hop neighborhood of the central node is captured, and the spatial features are extracted. Then, the ebedding representation of all the nodes is sent to the LSTM module for mining the time domain characteristics of the signals.
And (3) an Embedding: in the graph neural network, the nodes of the graph are represented as a low-dimensional vector space through aggregation neighbor information and normalization, and meanwhile, the topological structure and the node information of the network are reserved. And continuously aggregating neighbor information and iteratively updating through the Embedding, and finally enabling each node to almost aggregate global information. Herein we refer to the low-dimensional vector representation of the nodes that are ultimately obtained by means of Embedding as an aggregated feature.
As shown in fig. 3, in the structure diagram of the graph neural network in this embodiment, the graph neural network includes three graph SAGE layers and an LSTM layer, and each graph SAGE layer includes an SAG Conv layer, an AGG layer and a RELU layer, which are respectively a sampling neighbor convolution, a neighbor aggregation and a nonlinear activation function.
Fig. 4 is a block diagram of an LSTM network, which is intended for mining the time domain characteristics of signals. As shown in FIG. 4, the LSTM model structure uses the input information x at time t t The output h at the moment is obtained through the combined action of the forgetting gate, the memory gate, the output gate and the cell state t And updated cell state c t . Each time new information x is input t LSTM is based on x at the current time t And the output h of the neuron at the previous time t-1 And calculating the information to be forgotten by forgetting gate selection and outputting the information to be memorized by gate selection.
The following formula may be used to describe the forgetting door f t Memory gate and output gate m t Cell state c t Is calculated by (1):
f t ＝σ(W f ·[h t-1 ，x t ]+b f )；
C′ t ＝tanh(W c ·[h t-1 ，x t ]+b c )
i t ＝σ(W i ·[h t-1 ，x t ]+b i )
m t ＝σ(W m ·[h t-1 ，x t ]+b m )；
c t ＝f t *c t-1+ i t *C′ t ；
h t ＝m t *tanh(c t )
wherein σ is a Sigmoid function, tanh is a tanh function, and W, b is weight and bias respectively; c' t Is effective information extracted from the current input, i t Effective information is screened.
FIG. 5 is a graph of detection accuracy of the fusion model of the neural network and the LSTM network, reflecting the change in detection accuracy during training. As shown in FIG. 5, the model accuracy fluctuates greatly at the first 43 epochs, and tends to be optimal at 44-50 epochs, reaching 99.44%.
Example 3: referring to fig. 1 to 5, based on the embodiment 2, when data is collected in the step (1), the temperature on the fuel pipeline in the comprehensive gallery is collected, and when leakage detection is performed in the region to be detected in the step (5), the temperature is also distributed on the fuel pipeline in the region to be detected.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, and alternatives falling within the spirit and principles of the invention.
Claims (5)
1. A pipeline oil gas leakage detection method based on a graph neural network and an LSTM network is characterized in that: comprises the following steps of;
(1) Data acquisition and pretreatment;
(11) Collecting optical fiber temperature data on different oil and gas pipelines, wherein each oil and gas pipeline at least comprises a leakage position, the optical fiber temperature data comprises data of the leakage position, the length of each optical fiber temperature data is L data points, each data point comprises temperature data, and the temperature data are one stokes data and one an-stokes data;
(12) Preprocessing optical fiber temperature data, classifying each data point, if the data point is positioned at a leakage position, marking the data point as a positive sample, marking the data point as a negative sample, marking the data point as a 0, marking the data point as a label of the data point, cutting the labeled optical fiber temperature data into equal-length sample sections, and forming a data set by all the sample sections;
(2) Converting the sample segment into graph structure data, including steps (21) - (22);
(21) For a sample segment, taking each data point as a node, forming an edge by the node and the connecting lines of four adjacent data points, and converting the sample segment into graph structure data G= (V, E), wherein V is a set of all nodes, E is a set of all edges, and for each node, the data comprise temperature data and labels of the node;
(22) Converting all sample segments into graph structure data according to the step (21);
(3) Constructing a fusion network, wherein the fusion network comprises a graph neural network and an LSTM network which are sequentially connected;
(4) Sequentially sending the graph structure data into a fusion network for training until the network converges to obtain a trained graph neural network and LSTM network fusion model, wherein one graph structure data process comprises the steps (41) - (43);
(41) Sending the graph structure data into a graph neural network, taking each node as a central node, and outputting an aggregation feature after feature aggregation by an aggregator, wherein each node corresponds to L aggregation features;
(42) Each aggregation feature is respectively sent into an LSTM network to obtain an output, and the output is subjected to binary division and then used as the prediction probability that the node belongs to a positive sample;
(43) Taking the label of the node as a desired output, and adjusting the network weights of the graph neural network and the LSTM network;
(5) Detecting leakage of a region to be detected;
(51) Collecting optical fiber temperature data on the outer wall of an oil gas pipeline to be tested, preprocessing, and cutting into a sample section to be tested, wherein the length of the sample section is the same as that of the sample section;
(52) And sending the sample segment to be detected into a fusion model of the graph neural network and the LSTM network, outputting the prediction probability that each data point belongs to a positive sample on the sample segment to be detected, presetting a threshold value, and if the prediction probability is larger than the threshold value, determining that the data point is abnormal temperature, otherwise, determining that the data point is normal temperature.
2. The pipeline oil gas leakage detection method based on the graphic neural network and the LSTM network as set forth in claim 1, wherein the method comprises the following steps: in the step (11), the optical fiber temperature data are collected on different oil and gas pipelines specifically, the optical fiber temperature data are collected through a distributed optical fiber temperature measuring system, the distributed optical fiber temperature measuring system comprises sensing optical fibers, and the sensing optical fibers are distributed along the length direction of the pipelines and pass through the leakage positions on the oil and gas pipelines.
3. The pipeline oil gas leakage detection method based on the graphic neural network and the LSTM network as set forth in claim 1, wherein the method comprises the following steps: and (12) preprocessing the fiber temperature data, namely, the fiber temperature data comprise L data points, L pieces of stokes data and L pieces of an-stokes data, a stokes data vector and an-stokes data vector are respectively formed, and for each vector, normalization processing is respectively carried out on each element in the vector.
4. The pipeline oil gas leakage detection method based on the graphic neural network and the LSTM network as set forth in claim 1, wherein the method comprises the following steps: in the step (42), a logistic regression method Sigmoid is adopted to carry out binary division, specifically; the output end of the fusion network is connected with a Sigmoid layer;
the Sigmoid layer converts the output result of the LSTM network into the prediction probability that the node belongs to a positive sample, calculates the prediction probability and the loss value of the label of the positive sample through the cross entropy loss function BECLoss, executes back propagation and updating weight and iterates training, and enables the loss value not to be reduced until convergence, so that a trained fusion model of the graphic neural network and the LSTM network is obtained.
5. The pipeline oil gas leakage detection method based on the graphic neural network and the LSTM network as set forth in claim 1, wherein the method comprises the following steps: the graph neural network is GraphSAGE, PNA, GCN, GAT or GAE.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310355850.2A CN116447528A (en) | 2023-04-06 | 2023-04-06 | Pipeline oil gas leakage detection method based on graphic neural network and LSTM network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310355850.2A CN116447528A (en) | 2023-04-06 | 2023-04-06 | Pipeline oil gas leakage detection method based on graphic neural network and LSTM network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116447528A true CN116447528A (en) | 2023-07-18 |
Family
ID=87126537
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310355850.2A Pending CN116447528A (en) | 2023-04-06 | 2023-04-06 | Pipeline oil gas leakage detection method based on graphic neural network and LSTM network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116447528A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150911A (en) * | 2023-09-04 | 2023-12-01 | 吉林建筑大学 | Coal rock instability fracture prediction method and system based on graph neural network |
-
2023
- 2023-04-06 CN CN202310355850.2A patent/CN116447528A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117150911A (en) * | 2023-09-04 | 2023-12-01 | 吉林建筑大学 | Coal rock instability fracture prediction method and system based on graph neural network |
| CN117150911B (en) * | 2023-09-04 | 2024-04-26 | 吉林建筑大学 | Coal rock instability fracture prediction method and system based on graph neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Graph neural network-based fault diagnosis: a review | |
| Zanfei et al. | Novel approach for burst detection in water distribution systems based on graph neural networks | |
| Mustafa et al. | Fault identification for photovoltaic systems using a multi-output deep learning approach | |
| Sun et al. | Data-driven modeling and analysis based on complex network for multimode recognition of industrial processes | |
| CN116447528A (en) | Pipeline oil gas leakage detection method based on graphic neural network and LSTM network | |
| CN116557787B (en) | Intelligent evaluation system and method for pipe network state | |
| Jiang et al. | A multisensor cycle-supervised convolutional neural network for anomaly detection on magnetic flux leakage signals | |
| Zhang et al. | Remaining Useful Life Prediction of Rolling Bearings Using Electrostatic Monitoring Based on Two‐Stage Information Fusion Stochastic Filtering | |
| CN115654381A (en) | Water supply pipeline leakage detection method based on graph neural network | |
| Priyanka et al. | Multi-type feature extraction and classification of leakage in oil pipeline network using digital twin technology | |
| CN116244443A (en) | Ultrasonic gas meter node fault diagnosis method based on knowledge graph | |
| CN117743909A (en) | Heating system fault analysis method and device based on artificial intelligence | |
| Lučin et al. | Detailed leak localization in water distribution networks using random forest classifier and pipe segmentation | |
| Zhang et al. | Recognition of oil & gas pipelines operational states using graph network structural features | |
| Ye et al. | A novel self-supervised learning-based anomalous node detection method based on an autoencoder for wireless sensor networks | |
| Gong et al. | Photovoltaic arrays fault diagnosis based on an improved dilated convolutional neural network with feature-enhancement | |
| Zhang et al. | An unsupervised spatiotemporal fusion network augmented with random mask and time-relative information modulation for anomaly detection of machines with multiple measuring points | |
| CN117521800A (en) | Intelligent power transmission line inspection method based on knowledge graph | |
| CN115469585B (en) | Hydropower unit state monitoring method and system based on big data | |
| CN116484219A (en) | Water supply network water quality abnormal pollution source identification method based on gate control graph neural network | |
| Li et al. | An anomaly detection method for multiple time series based on similarity measurement and louvain algorithm | |
| Guo et al. | RobustFlow: An unsupervised paradigm toward real-world wear detection and segmentation with normalizing flow | |
| CN115310693A (en) | Comprehensive energy data cleaning method, system, equipment and medium | |
| CN115577364A (en) | Vulnerability mining method for result fusion of multiple static analysis tools | |
| CN115081741A (en) | Natural gas metrological verification intelligent prediction method based on neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |