Online tracking and positioning system and equipment for detector in oil and gas pipeline

Abstract

The invention discloses an online tracking and positioning system and device for a detector in an oil and gas pipeline, and the system comprises the following modules: a disturbance data collection module which is used for constructing a disturbance observation tensor; the reflection disturbance manifold modeling module is used for establishing a reflection disturbance manifold model and generating a disturbance reflection track data set; the neural radiation field modeling module is used for constructing a neural radiation field model based on the disturbance reflection trajectory data set; the graph structure regularization module is used for introducing disturbance trajectory graph structure constraints in the neural radiation field model training process; the real-time positioning reasoning module is used for collecting a disturbance observation tensor in real time and inputting the disturbance observation tensor into the neural radiation field model for real-time reasoning to generate a spatial disturbance response predicted value; and the reverse optimization module is used for constructing a disturbance response difference loss function and obtaining a minimum error space position corresponding to the current disturbance characteristic. According to the method, the reflection disturbance manifold model and the neural radiation field model are combined, and high-precision independent positioning of the detector in the pipeline is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of oil and gas pipeline detection, and particularly relates to an online tracking and positioning system and equipment for an oil and gas pipeline internal detector. BACKGROUND

[0002] With the increasing demand for intelligent detection of long-distance oil and gas pipelines, high-precision, long-distance, and GNSS-independent real-time positioning technology for the internal space of the pipeline has become an important research direction for the development of intelligent detectors. Existing oil and gas pipeline internal detection devices generally use gyroscopic inertial navigation, pipe wall feature matching, magnetic positioning, or acoustic positioning methods to estimate the position of the detector. However, in actual applications, the following problems generally exist: Inertial navigation systems have serious cumulative errors during long-time operation, especially in complex and winding pipe sections, which significantly drift and cannot meet the high-precision online positioning requirements. Positioning methods based on magnetic signals or sound waves are easily disturbed or blocked in complex terrains and closed environments, resulting in unstable positioning and poor reliability. Some feature matching-based methods rely on pre-set pipeline geometry or structure templates and have poor adaptability to actual pipeline deformation or working condition disturbances. Existing methods do not fully utilize the spatial propagation characteristic information carried by fluid disturbances during detector operation, resulting in weak interpretation of disturbance signals and fuzzy response to spatial positions.

[0003] Therefore, how to provide an online tracking and positioning system and equipment for an oil and gas pipeline internal detector is a problem that needs to be solved by those skilled in the art. SUMMARY

[0004] One object of the present application is to provide an online tracking and positioning system and equipment for an oil and gas pipeline internal detector. The present application combines neural radiation field modeling and disturbance graph structure regularization methods, establishes a reflection disturbance manifold and a disturbance trajectory graph based on disturbance response characteristics, realizes disturbance field learning and spatial response prediction, and introduces a reverse optimization algorithm to realize three-dimensional positioning without external positioning sources. The present application has the advantages of not relying on GPS, high precision, strong adaptability, and good robustness, and is suitable for online tracking and positioning of complex oil and gas pipeline internal detectors, effectively improving the automation and intelligence level of pipeline inspection.

[0005] According to an online tracking and positioning system for an oil and gas pipeline internal detector according to an embodiment of the present application, the following modules are included: A disturbance data acquisition module is arranged on the body of the oil and gas pipeline internal detector and includes a plurality of distributed sensor arrays for real-time acquisition of flow velocity disturbance, pressure fluctuation, and temperature information generated during the movement of the detector, and constructs a disturbance observation tensor in combination with the layout coordinates and timestamp information of each sensor; a reflected perturbation manifold modeling module configured to establish a reflected perturbation manifold model based on the perturbation observation tensor and generate a perturbation reflected trajectory dataset; a neural radiance field modeling module configured to construct a neural radiance field model based on the perturbation reflected trajectory dataset, the neural radiance field model including an input encoding layer, a densely connected hidden layer, and a perturbation response output layer; a graph structure regularization module configured to introduce a perturbation trajectory graph structure constraint in a neural radiance field model training process to limit a perturbation response difference of adjacent perturbation points in space; a real-time positioning inference module configured to collect a perturbation observation tensor in real time during detector operation and input the neural radiance field model for real-time inference to generate a spatial perturbation response prediction value; a reverse optimization module configured to construct a perturbation response difference loss function to obtain a minimum error spatial position corresponding to a current perturbation feature as a current three-dimensional absolute coordinate of the detector in the oil and gas pipeline.

[0006] An oil and gas pipeline detector online tracking positioning method according to an embodiment of the present application includes the following steps: Step one: a sensor array arranged on the detector in the oil and gas pipeline is used to collect flow rate perturbations, pressure fluctuations, and temperatures generated during the movement of the detector in real time, and a perturbation observation tensor is constructed by combining the arrangement coordinates and time stamp information of each sensor; Step two: a reflected perturbation manifold model is established based on the perturbation observation tensor, the reflected perturbation manifold model forms a perturbation reflected trajectory dataset through reflection and diffusion effects of the perturbation on the pipe wall; Step three: a neural radiance field model is constructed, the neural radiance field model includes an input encoding layer, a densely connected hidden layer, and a perturbation response output layer; Step four: the neural radiance field model is trained using the perturbation reflected manifold dataset, and a perturbation trajectory graph structure is constructed, and a perturbation trajectory graph structure regularization constraint is introduced in the training process; Step five: the trained neural radiance field model is deployed to a detector real-time positioning inference module, a perturbation observation tensor is collected in real time during detector operation and input to the neural radiance field model for real-time inference to generate a spatial perturbation response prediction value; Step six: a perturbation response difference loss function is constructed, and a reverse optimization algorithm is used to iteratively solve the perturbation response difference loss function to obtain a minimum error spatial position corresponding to a current perturbation feature; Step seven: the minimum error spatial position is taken as a real-time positioning output result of the detector in the oil and gas pipeline to realize online tracking positioning of the detector in the pipeline.

[0007] Optionally, the perturbation observation tensor is constructed in the following manner: calculating a rate of change of velocity per unit time for the collected flow rate disturbance data, calculating a pressure gradient for the pressure fluctuation data, and calculating a spatial temperature partial derivative for the temperature change data; constructing a disturbance feature vector based on the rate of change of velocity, the pressure gradient, and the spatial temperature partial derivative, and binding the disturbance feature vector with layout coordinates and timestamp information of each sensor; filling the disturbance feature vector in time sequence into a four-dimensional tensor with layout coordinates and time as dimensions to form a disturbance observation tensor.

[0008] Optionally, the step two is specifically: According to the rate of change of velocity, the pressure gradient, and the spatial temperature partial derivative recorded in the disturbance observation tensor, identifying a plurality of spatial position points reaching local extreme values within a unit time; Based on the time concentration and spatial distribution of the spatial position points, calculating a geometric center of mass as a detector disturbance source position; Taking the disturbance source position as a starting point, back-propagating the disturbance path to the pipe inner wall direction to identify the position and corresponding time of the reflection of the disturbance on the pipe wall, and receiving the vibration by the optical fiber coated outside the pipe; Taking the reflection position as a new disturbance starting point, simulating the diffusion process of the disturbance in three-dimensional space, extracting continuous disturbance feature points along the propagation path, and integrating each disturbance path and reflection response to form a disturbance reflection trajectory dataset; Organizing all the disturbance feature points in time sequence and layout coordinates to establish a reflection disturbance manifold model, including: A four-dimensional point set structure for representing the spatial position and time coordinates of each disturbance feature point in the disturbance trajectory; A disturbance feature vector set, each disturbance feature vector corresponding to a disturbance feature point; A reflection association structure for marking the propagation direction, reflection relationship, and multiple bounce paths between the disturbance feature points; A disturbance energy attenuation identifier for representing the amplitude decreasing process of the disturbance feature during propagation.

[0009] Optionally, the constructing a neural radiation field model is specifically: Taking the three-dimensional coordinates, time parameters, and disturbance feature vectors of each disturbance feature point in the reflection disturbance manifold model as inputs, inputting them into the neural radiation field model, the neural radiation field model sequentially including an input encoding layer, a densely connected hidden layer, and a disturbance response output layer; The input encoding layer performs position encoding processing on the three-dimensional coordinates and the time parameters respectively, encodes each dimension of the coordinates using a sine and cosine function with a set frequency, concatenates the encoded results with the original disturbance feature vectors to generate a high-dimensional expression tensor; The dense connection hidden layer comprises at least four full connection neural network layers arranged in sequence, the input of each hidden layer is the splicing result of the output of the previous hidden layer and the initial high-dimensional expression vector, the dense connection structure is adopted between the hidden layers, and the output tensor dimensions are consistent between the hidden layers; The perturbation response output layer comprises two independent output channels, the first channel is a single neuron node for outputting a perturbation density value, and the second channel is a multi-dimensional neuron output node for outputing a perturbation residual vector, and the two output channels both receive the output tensor of the last hidden layer as input.

[0010] Optionally, the perturbation trajectory graph structure is constructed, and specifically: The perturbation feature points in the reflection perturbation manifold model are arranged in time sequence, and a perturbation trajectory graph structure is constructed based on the layout coordinates and timestamp information of each perturbation feature point, the perturbation feature points are taken as graph nodes, and the spatial connectivity between any two time-continuous perturbation feature points is taken as a graph edge; Each graph edge is associated with a weight value, the weight value is obtained by weighted calculation according to the perturbation feature similarity and spatial distance between two nodes, the perturbation feature similarity is measured by the Euclidean distance between the velocity change rate, the pressure gradient and the temperature gradient, and the spatial distance is determined by the Euclidean distance between three-dimensional coordinates.

[0011] Optionally, the perturbation trajectory graph structure regularization constraint is introduced in the training process, and specifically: The perturbation trajectory graph structure is taken as a graph structure regularization constraint in the training process of the neural radiation field model, for any pair of adjacent nodes connected by an edge in the perturbation trajectory graph structure, the perturbation density value and the perturbation residual vector output by the two nodes in the neural radiation field model are obtained; For the perturbation density value, the square of the difference between the perturbation density values of adjacent nodes is calculated as a density constraint term; For the perturbation residual vector, the square of the Euclidean norm of the difference between the three-dimensional perturbation residual vectors corresponding to adjacent nodes is calculated as a residual constraint term; The density constraint term and the residual constraint term corresponding to all pairs of nodes connected by edges in the graph are weighted and accumulated, as a graph structure regularization term, to limit the change amplitude of adjacent perturbation feature points in the output result of the neural radiation field model.

[0012] Optionally, the disturbance response prediction value is an intermediate prediction result output by the neural radiation field model after receiving the disturbance observation tensor input during the detector operation, including a disturbance density prediction value and a disturbance residual vector prediction value, the disturbance density prediction value representing the disturbance response intensity at the corresponding position of the disturbance feature point in the neural radiation field, and the disturbance residual vector prediction value being in the form of a vector with x, y and z direction components, used to depict the position deviation of the disturbance feature point in space relative to the disturbance center.

[0013] Optionally, the step six is specifically: The disturbance density value in the disturbance observation tensor collected at the current time and the disturbance residual vector are taken as the target response value, the three-dimensional space coordinates are taken as the to-be-optimized variables, and the neural radiation field model is input together with the disturbance observation tensor to obtain the disturbance density prediction value and the disturbance residual vector prediction value under each candidate coordinate. The square error between the disturbance density prediction value and the target disturbance density value is taken as the density error term, the square of the Euclidean distance between the disturbance residual vector prediction value and the target disturbance residual vector is taken as the residual error term, and a weighted sum is constructed to build a disturbance response difference loss function. A gradient descent-based back optimization algorithm is adopted, the three-dimensional space coordinates are taken as the iterative variables, and iterative updating is performed within the disturbance response space range defined by the neural radiation field model until the disturbance response difference loss function converges to a preset threshold, so that the minimum error space position corresponding to the current disturbance feature is obtained, and the minimum error space position is the current three-dimensional absolute coordinate of the detector in the oil and gas pipeline.

[0014] An online tracking and positioning device for a detector in an oil and gas pipeline, comprising a hardware processing module and a storage module, the hardware processing module being configured as a method of an online tracking and positioning system for a detector in an oil and gas pipeline.

[0015] The present application has the following advantages: The application realizes the real-time output of the three-dimensional absolute coordinates of the detector in the oil and gas pipeline by the synergistic fusion of the disturbance observation tensor construction and the reflection disturbance manifold modeling, has the advantages of no need for external communication, strong robustness to complex fluid disturbance, high positioning accuracy and good model deployability. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, illustrate embodiments of the application, and are used to explain the application together with the embodiments of the application, and do not constitute a limitation on the application. In the drawings: Figure 1 A structural schematic diagram of an online tracking and positioning system of an oil and gas pipeline detector according to the application; Figure 2 A flowchart of an online tracking and positioning method of an oil and gas pipeline detector according to the application. DETAILED DESCRIPTION

[0017] The application will now be further described in detail in conjunction with the drawings. These drawings are all simplified schematic diagrams, and only illustrate the basic structure of the application in a schematic manner, and therefore only show the components related to the application.

[0018] Reference Figure 1 An online tracking and positioning system of an oil and gas pipeline detector, comprising the following modules: A disturbance data acquisition module, arranged on the body of the oil and gas pipeline detector, comprising a plurality of distributed sensor arrays, for acquiring the flow rate disturbance, pressure fluctuation and temperature information generated in the movement process of the detector in real time, and constructing a disturbance observation tensor in combination with the layout coordinates and time stamp information of each sensor; A reflection disturbance manifold modeling module, for establishing a reflection disturbance manifold model based on the disturbance observation tensor, and generating a disturbance reflection trajectory data set; a neural radiance field modeling module configured to construct a neural radiance field model based on the set of perturbation reflection trajectory data, the neural radiance field model comprising an input encoding layer, a densely connected hidden layer, and a perturbation response output layer; a graph structure regularization module configured to introduce a perturbation trajectory graph structure constraint in the training process of the neural radiance field model to limit the difference in perturbation response of spatially adjacent perturbation points; a real-time positioning inference module configured to collect perturbation observation tensors in real time during the operation of the detector and input the neural radiance field model for real-time inference to generate spatial perturbation response prediction values; a reverse optimization module configured to construct a perturbation response difference loss function to obtain a minimum error spatial position corresponding to a current perturbation feature as a current three-dimensional absolute coordinate of the detector in the oil and gas pipeline.

[0019] Reference Figure 2 An online tracking and positioning method for a detector in an oil and gas pipeline, comprising the following steps: Step one: through a sensor array arranged on the detector in the oil and gas pipeline, real-time collection of flow velocity perturbation, pressure fluctuation and temperature generated during the movement of the detector, and combination of the arrangement coordinates and time stamp information of each sensor to construct a perturbation observation tensor; Step two: establishing a reflection perturbation manifold model based on the perturbation observation tensor, the reflection perturbation manifold model forming a set of perturbation reflection trajectory data through reflection and diffusion effects of the perturbation on the pipe wall; Step three: constructing a neural radiance field model, the neural radiance field model comprising an input encoding layer, a densely connected hidden layer, and a perturbation response output layer; Step four: training the neural radiance field model using the set of perturbation reflection manifold data, and constructing a perturbation trajectory graph structure, introducing a perturbation trajectory graph structure regularization constraint in the training process; Step five: deploying the trained neural radiance field model to a real-time positioning inference module of the detector, collecting perturbation observation tensors in real time during the operation of the detector and inputting the neural radiance field model for real-time inference to generate spatial perturbation response prediction values; Step six: constructing a perturbation response difference loss function, and iteratively solving the perturbation response difference loss function using a reverse optimization algorithm to obtain a minimum error spatial position corresponding to a current perturbation feature; Step seven: taking the minimum error spatial position as a real-time positioning output result of the detector in the oil and gas pipeline to realize online tracking and positioning of the detector in the pipeline.

[0020] In the embodiment, the construction of the perturbation observation tensor specifically comprises: calculating the rate of change of velocity per unit time for the collected flow rate disturbance data, calculating the pressure gradient for the pressure fluctuation data, and calculating the spatial temperature partial derivative for the temperature change data; constructing a disturbance feature vector based on the rate of change of velocity, the pressure gradient, and the spatial temperature partial derivative, and binding the disturbance feature vector with the layout coordinates and timestamp information of each sensor; filling the disturbance feature vector in time sequence into a four-dimensional tensor with layout coordinates and time as dimensions to form a disturbance observation tensor.

[0021] In this embodiment, step two is specifically: According to the rate of change of velocity, the pressure gradient, and the spatial temperature partial derivative recorded in the disturbance observation tensor, a plurality of spatial position points reaching local extreme values per unit time are identified; Based on the time concentration and spatial distribution of the spatial position points, the geometric center of mass is calculated as the detector disturbance source position; Starting from the disturbance source position, the disturbance path is back-propagated towards the inner wall of the pipeline to identify the position and corresponding time of the reflection of the disturbance on the pipe wall; Taking the reflection position as a new disturbance starting point, the diffusion process of the disturbance in three-dimensional space is simulated, the continuous disturbance feature points along the propagation path are extracted, and the disturbance reflection trajectory dataset is formed by integrating each disturbance path and reflection response; Organize all disturbance feature points in time sequence and layout coordinates to establish a reflection disturbance manifold model, including: A four-dimensional point set structure is used to represent the spatial position and time coordinates of each disturbance feature point in the disturbance trajectory. A set of disturbance feature vectors, each corresponding to a disturbance feature point. A reflection association structure is used to mark the propagation direction, reflection relationship, and multiple bounce paths between disturbance feature points. A disturbance energy attenuation identifier is used to represent the amplitude reduction process of the disturbance feature during propagation.

[0022] In this embodiment, the construction of the neural radiation field model is specifically: The three-dimensional spatial coordinates, time parameters, and disturbance feature vectors corresponding to each disturbance feature point in the reflection disturbance manifold model are input into the neural radiation field model, which sequentially includes an input encoding layer, a dense connection hidden layer, and a disturbance response output layer. The reflection disturbance manifold model is a model established based on the propagation trajectory of the disturbance in space. Its essence is to restore the model by adding the calculated disturbance source and reflection point in the oil and gas pipeline space to the relative position of the sensor. Therefore, each disturbance feature point in the model has a clear spatial position.

[0023] The input encoding layer respectively performs position encoding processing on three-dimensional space coordinates and time parameters, encodes each coordinate using a sine and cosine function with a set frequency, splices the encoded results with the original perturbation feature vector to generate a high-dimensional expression tensor; The dense connection hidden layer includes at least four fully connected neural network layers arranged in series, the input of each hidden layer is the splicing result of the output of the previous hidden layer and the initial high-dimensional expression vector, the dense connection structure is used between the hidden layers, and the output tensor dimensions are kept consistent between the hidden layers; The perturbation response output layer includes two independent output channels, the first channel is a single neuron node for outputting a perturbation density value, and the second channel is a multi-dimensional neuron output node for outputting a perturbation residual vector, both output channels receive the output tensor of the last hidden layer as input; The perturbation response output layer includes two independent output channels, the first channel is a single neuron node, the input of which is the output tensor of the last hidden layer, which is linearly transformed and then activated by a sigmoid function, and outputs a normalized perturbation density value representing the corresponding perturbation response intensity of the current input point in the neural radiation field; The second channel is a multi-dimensional neuron output node, the input of which is also the output tensor of the last hidden layer, which is linearly transformed and outputs a perturbation residual vector, the perturbation residual vector includes three components corresponding to the perturbation response offsets of the input point in the x, y and z directions, respectively, for representing the spatial offset features of the current input point relative to the modeling perturbation center trajectory, the modeling perturbation center is a spatial reference point determined by the neural radiation field model during the training process according to the perturbation reflection trajectory dataset, the spatial reference point is obtained by calculating the extreme points of multiple perturbation reflection paths and the geometric center of gravity, and is used as a spatial reference for the output of the perturbation residual vector, to measure the offset relationship between the input perturbation point and the perturbation main path.

[0024] In the embodiment, the perturbation trajectory graph structure is constructed, specifically: The perturbation feature points in the reflection perturbation manifold model are arranged in time sequence, and a perturbation trajectory graph structure is constructed based on the layout coordinates and timestamp information of each perturbation feature point, with the perturbation feature points as graph nodes and the spatial connectivity between any two time-continuous perturbation feature points as graph edges; Each graph edge is associated with a weight value, which is obtained by weighted calculation according to the perturbation feature similarity and spatial distance between the two nodes, the perturbation feature similarity is measured by the Euclidean distance between the velocity change rate, pressure gradient and temperature gradient, and the spatial distance is determined by the Euclidean distance between the three-dimensional space coordinates.

[0025] In the embodiment, the disturbance trajectory graph structure is introduced as a graph structure regularization constraint in the training process of the neural radiance field model, and specifically: The disturbance trajectory graph structure is taken as a graph structure regularization constraint in the training process of the neural radiance field model, and for any pair of adjacent nodes connected by an edge in the disturbance trajectory graph structure, disturbance density values and disturbance residual vectors output by the two nodes in the neural radiance field model are obtained. For the disturbance density values, the square of the difference in disturbance density values between adjacent nodes is taken as a density constraint term. For the disturbance residual vectors, the square of the Euclidean norm of the difference in three-dimensional disturbance residual vectors corresponding to adjacent nodes is taken as a residual constraint term. The density constraint term and the residual constraint term corresponding to all pairs of nodes connected by edges in the graph are weighted and accumulated as a graph structure regularization term, which is used to limit the variation amplitude of adjacent disturbance feature points in space in the output results of the neural radiance field model.

[0026] Let the disturbance trajectory graph structure be graph G = (V, E), where V is a set of nodes, and each node represents a disturbance feature point. Wherein: V is a set of nodes, and each node represents a disturbance feature point. E is a set of edges, and each edge connects two spatially adjacent disturbance feature points.

[0027] For any edge e = (u, v) in E, define: and : are the disturbance density values of nodes u and v in the neural radiance field model, respectively. and : are the disturbance residual vectors of nodes u and v in the neural radiance field model, respectively, with a dimension of ; : is the weight value of edge e. Then the formula of the graph structure regularization term is: ; Wherein, represents the square of the Euclidean norm. ;

[0028] ​​​​​In this embodiment, the disturbance response prediction value is an intermediate prediction result output by the neural radiation field model after receiving the disturbance observation tensor input during the operation of the detector, including a disturbance density prediction value and a disturbance residual vector prediction value, the disturbance density prediction value representing the disturbance response intensity at the corresponding position of the disturbance feature point in the neural radiation field, and the disturbance residual vector prediction value being in the form of a vector with x, y and z direction components, and used to depict the position deviation of the disturbance feature point relative to the disturbance center in space.

[0029] In this embodiment, the step six is specifically: The disturbance density value in the disturbance observation tensor collected at the current time and the disturbance residual vector are taken as the target response value, the three-dimensional space coordinates are taken as the to-be-optimized variables, and the neural radiation field model is input with the three-dimensional space coordinates and the disturbance observation tensor to obtain the disturbance density prediction value and the disturbance residual vector prediction value under each candidate coordinate. The square error between the disturbance density prediction value and the target disturbance density value is taken as the density error term, the square of the Euclidean distance between the disturbance residual vector prediction value and the target disturbance residual vector is taken as the residual error term, and the disturbance response difference loss function is constructed by weighted summation. A gradient descent-based back optimization algorithm is adopted to iteratively update the three-dimensional space coordinates within the disturbance response space defined by the neural radiation field model until the disturbance response difference loss function converges to a preset threshold, and the minimum error space position corresponding to the current disturbance feature is obtained. The minimum error space position is the current three-dimensional absolute coordinate of the detector in the oil and gas pipeline.

[0030] An online tracking and positioning device for a detector in an oil and gas pipeline includes a hardware processing module and a storage module, and the hardware processing module is configured as a method of an online tracking and positioning system for a detector in an oil and gas pipeline.

[0031] Embodiment 1 In order to verify the feasibility of the application in implementation, the application is applied to a gas transmission main pipeline detection operation in a certain province, the pipeline has a total length of 86 kilometers, is buried in a complex terrain and city pipe network intersection area, has frequent changes in pipe diameter in some sections, and has dense branch interfaces. The pipeline detector based on traditional inertial navigation and magnetic field positioning often has problems such as large positioning error, data loss and drift accumulation in such working conditions, which seriously affects the accuracy of subsequent leakage analysis and structure health assessment. Therefore, the research team selected three types of structure regions, i.e., a typical straight pipe section, a bend pipe section and a complex node section, in the pipeline, with a total length of about 18 kilometers, and carried out online tracking and positioning system measurement verification of the detector in the oil and gas pipeline.

[0032] In this embodiment, the intelligent detector with the function of the present application is used to carry out two rounds of complete detection on the target pipe section. The detector integrates 12 groups of distributed disturbance sensing units. Each group of sensors collects three types of data: velocity disturbance, pressure fluctuation and temperature change, and has a high-precision clock synchronization module for tensor construction. During the detection process, the disturbance response data is written into the local cache module in real time and uploaded to the edge computing terminal through the relay transmission node. The trained neural radiation field model is deployed in the terminal, and the reverse optimization engine is operated in parallel for positioning and reasoning.

[0033] Firstly, the system collects baseline data and preliminary training in a static environment. The detector collects the trajectory and sensor response data excited by each disturbance source when it passes through the pipeline center smoothly. Then in the dynamic actual detection, the detector passes through the typical structure section at an average speed of 0.2 m / s. The system collects disturbance data in real time and reasons and locates the output at the edge terminal. The positioning frequency is controlled at 1 Hz, and the error performance is evaluated by comparing with the high-precision odometer laid by artificial. The results show that the system has strong robustness and high positioning accuracy during the whole section operation. Especially in the complex section with bend reflection and branch disturbance coupling, it can still achieve sub-meter positioning performance. The following table shows the positioning error statistical data of the disturbance observation characteristic data samples and the reasoning output of the neural radiation field model under typical working conditions.

[0034] During the test, the disturbance data collected from each distributed sensor forms an observation tensor, and the corresponding spatial position prediction is obtained by the reasoning of the system. The following Table 1 shows the original disturbance values (velocity change rate, pressure gradient, temperature partial derivative) and spatial coordinate information corresponding to some sampling points.

[0035] Table 1: Disturbance observation characteristic data example table under typical working conditions Sample number Time stamp (ms) Rate of change of velocity (m / s²) Pressure gradient (Pa / m) Temperature gradient (°C / m) Sensor coordinates (x, y, z) (m) 001 104952 -0.143 85.2 0.012 (12.8,0.0,-0.3) 002 104962 0.155 76.5 0.017 (13.0,0.0,-0.3) 003 104972 -0.134 80.4 0.014 (13.2,0.0,-0.3) 004 104982 -0.125 92.1 0.011 (13.4,0.0,-0.3) 005 104992 0.162 78.3 0.016 (13.6,0.0,-0.3) Table 1 shows the positioning performance of the present application under three typical operating conditions in oil and gas pipelines, namely normal flow state, local high pressure and variable temperature and speed conditions. For each working condition, the disturbance characteristic data collected on site is recorded, including flow velocity disturbance amplitude, pressure fluctuation range and temperature change gradient. Further, the disturbance density prediction value and three-dimensional disturbance residual vector are generated by the neural radiation field model. On this basis, the system uses the reverse optimization module to iteratively calculate the spatial positioning coordinates of the detector, and compares them with the actual reference coordinates at that time to calculate the positioning error.

[0036] To further verify the actual performance of the system in the online tracking and positioning of the oil and gas pipeline detector, the positioning accuracy in different structure sections is counted, including straight pipe sections, elbow pipe sections, branch pipe sections, and reduced diameter sections, and the like typical complex structure areas. A plurality of reference positioning points are arranged in a pipeline experimental section with a total length of 1200 meters, and the disturbance response and system output coordinates of the detector at these positions are recorded synchronously. The spatial deviation between the reference positions and the system estimated positions is compared, and the average positioning error, the maximum error, and the error range of the 95% confidence interval in different structure sections are counted.

[0037] Table 2: Positioning accuracy statistics table of each typical structure section in the detection area Section type Section length (m) Average disturbance response strength Mean squared error (MSE, m²) Maximum error (m) Average error (m) Straight pipe section 8200 0.87 0.014 0.28 0.09 Elbow section 6100 0.91 0.022 0.41 0.13 Node section 3700 0.93 0.029 0.52 0.17 As can be seen from the data in Table 2 above, the online tracking and positioning system of the oil and gas pipeline detector of the present application shows excellent positioning performance in different structure sections, which reflects the good versatility and adaptability of the system.

[0038] In the straight pipe section area, the total length is 8200 meters, the average value of the disturbance response strength is 0.87, the system realizes the lowest mean square error (MSE) of 0.014 m², the maximum error is 0.28 meters, and the average error is only 0.09 meters. This shows that in the straight pipe area where the fluid state is relatively stable and the disturbance propagation path is regular, the neural radiation field model can highly restore the spatial distribution of the disturbance field, realize high-precision positioning, and the error control effect is remarkable.

[0039] In the elbow pipe section area, the total length is 6100 meters, the average disturbance response strength is slightly higher to 0.91, the mean square error rises to 0.022 m², the maximum error is 0.41 meters, and the average error is 0.13 meters. Although the fluid is prone to form backflow and turbulent disturbance at the elbow, thanks to the precise modeling of the disturbance reflection characteristics of the system and the ability of the graph structure regularization to suppress local abnormal response, the system still maintains high positioning stability in the elbow structure.

[0040] In the node section with complex structure (such as tee, branch intersection), the total length is 3700 meters, and the average disturbance response strength further rises to 0.93. Due to the significant disturbance reflection and interference at the node, the positioning difficulty is the greatest, but the system of the present application still controls the mean square error to 0.029 m², the maximum error is 0.52 meters, and the average error is 0.17 meters, which shows good robustness of the model in the nonlinear flow field environment.

[0041] Through the field deployment and verification of the detector online tracking and positioning system proposed in the application in a typical oil and gas pipeline structure scene, the results show that the system can stably extract multi-source disturbance feature information in a complex disturbance environment, and realize high-precision real-time positioning of the detector position by combining reflection disturbance manifold modeling and neural radiation field reasoning. The average positioning error can be controlled within 0.2 meters, and the maximum error is not more than 0.52 meters. The system has strong adaptability to pipeline structures and high robustness, and can effectively solve the technical bottlenecks of the traditional inertial navigation system, such as easy drift and precision easily affected by environmental interference, and has good engineering practicability and application value.

[0042] The above is only a preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can make equivalent replacement or change according to the technical scheme and the inventive concept of the application within the technical range disclosed by the application, which should be covered within the protection scope of the application.

Claims

1. An oil and gas pipeline internal detector on-line tracking and positioning system, characterized by, The method comprises the following modules: A disturbance data acquisition module is arranged on the detector body in the oil and gas pipeline, comprising a plurality of distributed sensor arrays, for real-time acquisition of flow velocity disturbance, pressure fluctuation and temperature information generated during detector movement, and construction of a disturbance observation tensor combined with the layout coordinates and timestamp information of each sensor; A reflected disturbance manifold modeling module is used to establish a reflected disturbance manifold model based on the disturbance observation tensor, and generate a disturbance reflection trajectory data set; A neural radiation field modeling module is used to construct a neural radiation field model based on the disturbance reflection trajectory data set, the neural radiation field model comprising an input encoding layer, a densely connected hidden layer and a disturbance response output layer; A graph structure regularization module is used to introduce a disturbance trajectory graph structure constraint in the neural radiation field model training process, to limit the disturbance response difference of adjacent disturbance points in space; A real-time positioning inference module is used to real-time acquire disturbance observation tensors during detector operation and input the neural radiation field model for real-time inference, to generate spatial disturbance response prediction values; A reverse optimization module is used to construct a disturbance response difference loss function, to obtain the minimum error spatial position corresponding to the current disturbance feature as the current three-dimensional absolute coordinates of the detector in the oil and gas pipeline.

2. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 1, wherein, The modules are realized through the following methods: Step one: through the sensor array arranged on the detector in the oil and gas pipeline, real-time acquisition of flow velocity disturbance, pressure fluctuation and temperature generated during detector movement, and construction of a disturbance observation tensor combined with the layout coordinates and timestamp information of each sensor; Step two: based on the disturbance observation tensor, a reflected disturbance manifold model is established, the reflected disturbance manifold model forms a disturbance reflection trajectory data set through reflection and diffusion effects of disturbance on the pipe wall; Step three: a neural radiation field model is constructed, the neural radiation field model comprising an input encoding layer, a densely connected hidden layer and a disturbance response output layer; Step four: the neural radiation field model is trained using the disturbance reflection manifold data set, and a disturbance trajectory graph structure is constructed, and a disturbance trajectory graph structure regularization constraint is introduced in the training process; Step five: the trained neural radiation field model is deployed to the detector real-time positioning inference module, real-time acquisition of disturbance observation tensors during detector operation and input of the neural radiation field model for real-time inference, to generate spatial disturbance response prediction values; Step six: a disturbance response difference loss function is constructed, and the disturbance response difference loss function is iteratively solved using a reverse optimization algorithm, to obtain the minimum error spatial position corresponding to the current disturbance feature; Step seven: the minimum error spatial position is taken as the real-time positioning output result of the detector in the oil and gas pipeline, to realize online tracking and positioning of the detector in the pipeline.

3. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The construction of the disturbance observation tensor is specifically: The velocity change rate per unit time is calculated for the acquired flow velocity disturbance data, the pressure gradient is calculated for the pressure fluctuation data, and the spatial temperature partial derivative is calculated for the temperature change data; A disturbance feature vector is constructed based on the velocity change rate, pressure gradient and spatial temperature partial derivative, and is bound combined with the layout coordinates and timestamp information of each sensor; The disturbance eigenvectors are sequentially filled into a four-dimensional tensor with layout coordinates and time as dimensions to form a disturbance observation tensor.

4. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The step two specifically includes: According to the velocity change rate, pressure gradient and spatial temperature partial derivative recorded in the disturbance observation tensor, a plurality of spatial position points reaching local extreme values in a unit time are identified; Based on the time concentration and spatial distribution of the spatial position points, a geometric barycenter is calculated as a detector disturbance source position; Starting from the disturbance source position, the disturbance path is backtracked towards the pipe inner wall to identify the position and corresponding time of the disturbance reflection on the pipe wall; Taking the reflection position as a new disturbance starting point, the diffusion process of the disturbance in the three-dimensional space is simulated, the continuous disturbance feature points along the propagation path are extracted, and the disturbance path and reflection response are integrated to form a disturbance reflection trajectory dataset; All disturbance feature points are organized according to time sequence and layout coordinates to establish a reflection disturbance manifold model, including: A four-dimensional point set structure for representing the spatial position and time coordinates of each disturbance feature point in the disturbance trajectory; A disturbance eigenvector set, each disturbance eigenvector corresponding to a disturbance feature point; A reflection association structure for marking the propagation direction, reflection relationship and multiple bounce paths between the disturbance feature points; A disturbance energy attenuation identifier for representing the amplitude reduction process of the disturbance feature during propagation.

5. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The neural radiation field model is constructed, specifically: The three-dimensional coordinates, time parameters and disturbance eigenvectors corresponding to each disturbance feature point in the reflection disturbance manifold model are input into the neural radiation field model, which sequentially includes an input encoding layer, a densely connected hidden layer and a disturbance response output layer; The input encoding layer performs position encoding processing on the three-dimensional coordinates and time parameters respectively, encodes each dimension coordinate using a sine function with a set frequency, concatenates the encoded results with the original disturbance eigenvectors to generate a high-dimensional expression tensor; The densely connected hidden layer includes at least four layers of fully connected neural network layers arranged in series, the input of each hidden layer is the concatenation result of the output of the previous hidden layer and the initial high-dimensional expression vector, the densely connected structure is used between the hidden layers, and the output tensor dimensions are kept consistent between the hidden layers; The disturbance response output layer includes two independent output channels, the first channel is a single neuron node for outputting the disturbance density value, and the second channel is a multi-dimensional neuron output node for outputting the disturbance residual vector, both output channels receive the output tensor of the last hidden layer as input.

6. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The disturbance trajectory graph structure is constructed, specifically: The disturbance feature points in the reflection disturbance manifold model are arranged in time sequence, and the disturbance trajectory graph structure is constructed based on the layout coordinates and time stamp information of each disturbance feature point, taking the disturbance feature points as graph nodes and the spatial connectivity between any two time-continuous disturbance feature points as graph edges. Each edge of the graph is associated with a weight value, which is obtained by weighting calculation according to the disturbance feature similarity and the spatial distance between two nodes, the disturbance feature similarity is measured by the Euclidean distance between the three disturbance features of velocity change rate, pressure gradient and temperature gradient, and the spatial distance is determined by the Euclidean distance between three-dimensional coordinates.

7. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The disturbance trajectory graph structure is introduced as a regularization constraint in the training process, specifically: The disturbance trajectory graph structure is used as a graph structure regularization constraint in the training process of the neural radiation field model, for any pair of adjacent nodes connected by an edge in the disturbance trajectory graph structure, the disturbance density value and the disturbance residual vector output by the two nodes in the neural radiation field model are obtained; For the disturbance density value, the square of the difference between the disturbance density values of adjacent nodes is calculated as a density constraint term; For the disturbance residual vector, the square of the Euclidean norm of the difference between the three-dimensional disturbance residual vectors corresponding to adjacent nodes is calculated as a residual constraint term; The density constraint term and the residual constraint term corresponding to all node pairs connected by edges in the graph are weighted and accumulated as a graph structure regularization term, which is used to limit the variation amplitude of adjacent disturbance feature points in space in the output results of the neural radiation field model.

8. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The disturbance response prediction value is the intermediate prediction result output by the neural radiation field model after receiving the disturbance observation tensor input during the operation of the detector, including the disturbance density prediction value and the disturbance residual vector prediction value, the disturbance density prediction value represents the disturbance response intensity at the corresponding position of the current input disturbance feature point in the neural radiation field, and the disturbance residual vector prediction value is in the form of a vector with x, y and z direction components, which is used to describe the position deviation of the disturbance feature point relative to the disturbance center in space.

9. An inline tracking and positioning system for an oil and gas pipeline inspection tool according to claim 2, wherein, The step six is specifically: The disturbance density value and the disturbance residual vector in the disturbance observation tensor collected at the current time are taken as the target response value, the three-dimensional space coordinates are taken as the optimization variables, and the neural radiation field model is input with the disturbance observation tensor to obtain the disturbance density prediction value and the disturbance residual vector prediction value under each candidate coordinate; The square error between the disturbance density prediction value and the target disturbance density value is taken as the density error term, the Euclidean distance square between the disturbance residual vector prediction value and the target disturbance residual vector is taken as the residual error term, and the disturbance response difference loss function is constructed by weighted summation; A gradient descent-based back optimization algorithm is adopted, the three-dimensional space coordinates are taken as the iteration variables, and the iteration is updated within the disturbance response space range defined by the neural radiation field model until the disturbance response difference loss function converges to a preset threshold, the minimum error space position corresponding to the current disturbance feature is obtained, and the minimum error space position is the current three-dimensional absolute coordinate of the detector in the oil and gas pipeline.

10. An inline tracking and positioning apparatus for an oil and gas pipeline inspection tool, comprising a hardware processing module and a storage module, wherein, The hardware processing module is configured to execute the method of the online tracking and positioning system for the detector in the oil and gas pipeline according to any one of claims 1 to 9.

Images