Parameterized oil and gas pipeline leakage diffusion prediction self-correction method and system

Abstract

The invention relates to the technical field of oil and gas pipeline leakage, and provides a parameterized oil and gas pipeline leakage diffusion prediction self-correction method and system. In the method, a three-dimensional, uniform and discrete concentration field during oil and gas pipeline leakage is reduced into a first concentration field hidden space vector with a fixed length through parameterization of an oil and gas pipeline leakage diffusion flow field, and results at the first # imgabs0 # moment are used as input of subsequent prediction through autoregression prediction. And full-time-sequence prediction during oil and gas pipeline leakage is completed in an iteration mode. Meanwhile, in order to eliminate deviation of a prediction result from a true value caused by an accumulative error in a prediction process, concentration field actual measurement data and concentration field prediction data are fused through judgment of a relative error, and the prediction data are corrected by using correction data of a hidden space dynamic prediction model, so that the accumulative error in an autoregression prediction process is effectively eliminated, and the prediction accuracy is improved. And the prediction precision and accuracy are improved.

Description

Technical Field

[0001] The present application relates to the technical field of emergency response for oil and gas pipeline leakage, and in particular to a parameterized oil and gas pipeline leakage diffusion prediction self-correction method and system. Background Art

[0002] Oil and gas pipeline transportation is an efficient and safe way to transport oil and natural gas over long distances. However, since pipelines may face problems such as aging and corrosion during long-term operation, the risk of leakage is relatively high. In addition, leakage is usually hidden, and the transported oil and natural gas are flammable and explosive, and once leaked, they may cause serious environmental pollution and safety accidents. Therefore, the safety hazards of oil and gas pipelines are more prominent, especially in emergency plugging work after leakage, which often faces difficult and time-consuming challenges. This makes it crucial to grasp the diffusion situation after the leak in a timely and accurate manner. Through real-time situation prediction, the diffusion path and impact range of the leak can be effectively predicted, thereby providing valuable time and decision-making basis for rapid disposal and plugging.

[0003] The oil and gas pipeline leakage and diffusion situation prediction system can quickly identify the location of the leakage, predict the diffusion path and impact range of the leaked material, and propose prevention and control measures according to the severity and impact range of the leakage, thereby providing strong technical support for the safe operation of the oil and gas pipeline. However, the current pipeline leakage situation prediction system has problems such as poor timeliness, strong dependence on field data, and poor prediction accuracy.

[0004] Therefore, there is an urgent need to provide a technical solution to the above-mentioned deficiencies in the prior art. Summary of the invention

[0005] The purpose of the present application is to provide a parameterized oil and gas pipeline leakage diffusion prediction self-correction method and system to solve or alleviate the problems existing in the above-mentioned prior art.

[0006] In order to achieve the above objectives, this application provides the following technical solutions: The present application provides a parameterized oil and gas pipeline leakage diffusion prediction self-correction method, comprising: step S101, based on A leakage parameter characterization model of a fully connected neural network is used to generate the leakage parameters of the oil and gas pipeline according to the leakage parameters of the oil and gas pipeline. of the moment The first concentration field hidden space vector; among them, is a positive integer; Step S102: according to the oil and gas pipeline leakage The first concentration field latent space vector at a moment, based on a pre-built autoregressive prediction model, performs time series prediction on the concentration field when the oil and gas pipeline leaks, and obtains the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; Step S103: In response to the first predicted time series concentration field The concentration field prediction data at the moment is compared with the leakage diffusion simulation of the oil and gas pipeline The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time are merged to obtain the mixed concentration field data when the oil and gas pipeline leaks; Step S104: input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline into the latent space dynamic prediction model based on TimesNet, and use the output of the oil and gas pipeline leakage time The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0007] Preferably, in step S101, the leakage parameters of the oil and gas pipeline when it leaks , input the leakage parameter characterization model The fully connected neural network structure , generating The first concentration field implicit space vector at the moment ; in,

[0008] For the The fully connected neural network structure a weight matrix in the leakage parameter standard model; For the The fully connected neural network structure The bias vector of is the activation function; .

[0009] Preferably, step S102 includes: training a concentration field dimensionality reduction model constructed based on a variational autoencoder according to concentration field sample data obtained by simulating the leakage and diffusion of the oil and gas pipeline; converting the concentration field sample data into concentration field latent space sample vectors based on the trained concentration field dimensionality reduction model to generate a dimensionality reduction sample data set of the natural gas concentration field when the oil and gas pipeline leaks; and training the constructed autoregressive prediction model based on TimesNet according to the dimensionality reduction sample data set.

[0010] Preferably, the concentration field dimensionality reduction model based on the training is used to convert the concentration field sample data into a concentration field latent space sample vector to generate a dimensionality reduction sample data set of the natural gas concentration field when the oil and gas pipeline leaks. Specifically, a plurality of the concentration field sample data are respectively input into an encoder of a variational autoencoder based on a convolutional neural network, according to the formula:

[0011] conduct The convolution operation is performed on the first Layer convolution operation The resulting convolution vector Input the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector The mean ; and, Layer convolution operation The resulting convolution vector Input the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector Variance ; According to the formula:

[0012] Correspondingly generate multiple concentration field hidden space sample vectors , to form the dimension-reduced sample data set; in, is a nonlinear activation function, Indicates that the concentration field sample data is processed Layer convolution operation; is the bias vector of the convolutional neural network in the encoder of the variational autoencoder; are all integers, ; For the The concentration field sample data; , is a positive integer; is a noise vector that follows a standard normal distribution.

[0013] Preferably, in step S102, based on the trained autoregressive prediction model, the oil and gas pipeline leak is The first concentration field latent space vector at each moment is used to perform time series prediction on the concentration field when the oil and gas pipeline leaks, so as to obtain the first predicted concentration field latent space vector when the oil and gas pipeline leaks; based on the concentration field dimensionality reduction model, according to the first predicted concentration field latent space vector when the oil and gas pipeline leaks, the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks is generated.

[0014] Preferably, the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks is generated based on the concentration field dimensionality reduction model according to the first predicted concentration field latent space vector when the oil and gas pipeline leaks, specifically: a decoder of a variational autoencoder based on a convolutional neural network, according to the formula:

[0015] The first predicted concentration field implicit space vector of the oil and gas pipeline leakage Reconstruction is performed to generate a first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; wherein, is a nonlinear activation function, Expressing the Layer input vector Conduct the Layer deconvolution operation; is the bias vector of the convolutional neural network in the decoder of the variational autoencoder; For the Layer deconvolution operation The output, For the Layer deconvolution operation Output: is an integer.

[0016] Preferably, in step S103, the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time instant are fused to obtain the mixed concentration field data when the oil and gas pipeline leaks, specifically: the concentration field measured data when the oil and gas pipeline leaks is repaired to generate the uniformly discrete repaired concentration field measured data when the oil and gas pipeline leaks ; Based on the conditional adversarial network, according to the formula:

[0017] The measured data of the repair concentration field and The concentration field prediction data at time Fusion is performed to obtain the mixed concentration field data when the oil and gas pipeline leaks ; In the formula, is a nonlinear activation function, represents the deconvolution operation, represents the weighted summation of data features. Represents the measured data of the repair concentration field Perform convolution operation, Represents the concentration field prediction data Perform convolution operation.

[0018] Preferably, in step S104, an encoder based on a concentration field dimensionality reduction model of a variational autoencoder performs dimensionality reduction on the mixed concentration field data to obtain the mixed concentration field latent space vector.

[0019] Preferably, in step S104, The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to generate a second predicted concentration field latent space vector when the oil and gas pipeline leaks; the decoder of the concentration field dimensionality reduction model based on the variational autoencoder reconstructs and restores the second predicted concentration field latent space vector to generate a second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0020] The embodiment of the present application also provides a parameterized oil and gas pipeline leakage diffusion prediction self-correction system, including: a latent space generation unit, configured to be based on A leakage parameter characterization model of a fully connected neural network is used to generate the leakage parameters of the oil and gas pipeline according to the leakage parameters of the oil and gas pipeline. The first concentration field latent space vector at the moment; among them, is a natural number; The autoregressive prediction unit is configured to predict the oil and gas pipeline according to the previous The first concentration field latent space vector at a moment, based on a pre-built autoregressive prediction model, performs time series prediction on the concentration field when the oil and gas pipeline leaks, and obtains the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; A data fusion unit is configured to respond to the first predicted time series concentration field The concentration field prediction data at the moment is compared with the leakage diffusion simulation of the oil and gas pipeline The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time are merged to obtain the mixed concentration field data when the oil and gas pipeline leaks; The prediction and correction unit is configured to input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline when leaking into the latent space dynamic prediction model based on TimesNet, and use the output of the oil and gas pipeline leakage time The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0021] Beneficial effects:

[0022] In the parameterized oil and gas pipeline leakage diffusion prediction self-correction method provided in the embodiment of the present application, the leakage parameter of the oil and gas pipeline leakage is input based on A fully connected neural network leakage parameter characterization model is used to generate the oil and gas pipeline leakage front of the moment The first concentration field latent space vector; and in the pre-built autoregressive prediction model, the generated of the moment The first concentration field hidden space vector is used to perform time series prediction on the concentration field when the oil and gas pipeline leaks, and the first predicted concentration field hidden space vector is obtained when the oil and gas pipeline leaks, and then the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks is generated; when the first predicted time series concentration field The concentration field prediction data at the time is compared with the data obtained by leak diffusion simulation of oil and gas pipelines. When the relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, the concentration field measured data at the time of oil and gas pipeline leakage and The concentration field prediction data at each moment is fused, and the mixed concentration field data obtained by fusion when the oil and gas pipeline leaks is reduced in dimension. The mixed concentration field latent space vector obtained by dimensionality reduction and the leakage parameters when the oil and gas pipeline leaks are input into the latent space dynamic prediction model based on TimesNet. The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0023] By means of A leakage parameter characterization model of a fully connected neural network is used to parameterize the diffusion flow field of oil and gas pipeline leakage, and the three-dimensional, uniform, discrete concentration field of oil and gas pipeline leakage is reduced to a fixed-length first concentration field latent space vector. The results at each moment are used as the input for subsequent predictions, and the full time series prediction of oil and gas pipeline leakage is completed in an iterative manner.

[0024] At the same time, in order to eliminate the deviation of the prediction results from the true value caused by the cumulative error in the prediction process, The concentration field prediction data at the moment The relative error of the concentration field sample data at the time of the oil and gas pipeline leakage is judged by comparing the concentration field measured data and The concentration field prediction data at each moment are fused and reduced in dimension, and the combined concentration field latent space vector and the leakage parameters when the oil and gas pipeline leaks are input into the latent space dynamic prediction model based on TimesNet. The correction data of the latent space dynamic prediction model is used to correct the corresponding data in the latent space of the first predicted concentration field, effectively eliminating the accumulated error in the autoregressive prediction process, realizing the effective correction of the autoregressive prediction, and improving the prediction precision and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. Among them: Figure 1 A schematic flow chart of a parameterized oil and gas pipeline leakage diffusion prediction self-correction method provided according to some embodiments of the present application; Figure 2 A logical schematic diagram of a parameterized oil and gas pipeline leakage diffusion prediction self-correction method provided according to some embodiments of the present application; Figure 3 A schematic diagram of the structure of a concentration field dimensionality reduction model provided according to some embodiments of the present application; Figure 4 A schematic diagram of slices of concentration field sample data at different leakage times provided according to some embodiments of the present application; Figure 5 A schematic diagram of slices of concentration field prediction data at different leakage times provided according to some embodiments of the present application; Figure 6 A schematic diagram of the error between concentration field prediction data and concentration field sample data at different leakage times provided according to some embodiments of the present application; Figure 7 A schematic structural diagram of a parameterized oil and gas pipeline leakage diffusion prediction and correction system provided according to some embodiments of the present application. DETAILED DESCRIPTION

[0026] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations may be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as a part of an embodiment may be used in another embodiment to produce yet another embodiment. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the embodiments of the present invention should belong to the scope of protection of the embodiments of the present invention.

[0027] Oil and gas pipeline transportation is an efficient and safe way to transport oil and natural gas over long distances. However, due to the hidden nature of pipeline leakage and the flammable and explosive characteristics of the transported medium, there are major safety hazards in the process of oil and gas pipeline transportation. In addition, in the emergency plugging work after the leakage, it is necessary to grasp the diffusion trend after the leakage in a timely and accurate manner. However, the existing prediction of oil and gas pipeline leakage has problems such as poor timeliness, inaccurate prediction, and poor system integration. The main manifestations are: first, the calculation performance of the model used is poor and the real-time prediction effect cannot be achieved; second, the model prediction ability is limited. The existing prediction model relies too much on field data collection, and there are large errors when there are problems with the field data; third, the calibration process of the prediction model is not in-depth enough and the prediction accuracy of the model cannot be fully optimized.

[0028] Based on this, the present application embodiment proposes a parameterized oil and gas pipeline leakage diffusion prediction self-correction method, such as Figures 1 to 6 As shown, the prediction method includes: Step S101: Based on A leakage parameter characterization model of a fully connected neural network is used to generate the previous prediction of the oil and gas pipeline leakage according to the leakage parameters of the oil and gas pipeline. of the moment The first concentration field latent space vector.

[0029] In this application, the parameters of oil and gas pipeline leakage are detected by ground unmanned vehicle-mounted radar or drone-mounted mid-infrared spectral camera, and the leakage parameters such as pipeline leakage pressure, leakage direction, soil type, ambient wind speed, etc. are input into the A leakage parameter characterization model of a fully connected neural network architecture is proposed. In the leakage parameter characterization model, the bias vector and weight matrix of each fully connected neural network are different. Therefore, when the same set of leakage parameters are input into fully connected neural networks with different bias vectors and weight matrices, the output of each fully connected neural network is different.

[0030] Specifically, the leakage parameters of oil and gas pipeline leakage , input leakage parameter characterization model A fully connected neural network structure , generating The first concentration field implicit space vector at the moment Specifically,

[0031] in, For the A fully connected neural network structure Weight matrix in the standard model of leakage parameters; For the A fully connected neural network structure The bias vector of is the activation function; ;in, Is a positive integer.

[0032] The leakage parameter characterization model uses a multi-layer fully connected neural network as its core structure, which transforms the operating parameters (leakage parameters ) is mapped to the corresponding front In this process, the leakage position, leakage direction, leakage pressure, leakage medium, leakage aperture, soil type, and ambient wind speed under different working conditions are used as model data, and the previous The latent space vector at the moment is used as the target output, so that the model can generate the previous The concentration field hidden space vector at the moment (i.e., the first concentration field hidden space vector).

[0033] Step S102: according to the oil and gas pipeline leakage The first concentration field latent space vector at each moment is obtained, and based on the pre-built autoregressive prediction model, the concentration field when the oil and gas pipeline leaks is predicted in time series to obtain the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0034] Before generating the model by characterizing the leakage parameters The concentration field hidden space vector at the moment (i.e., the first concentration field hidden space vector) is The first concentration field latent space vector at the moment is used as the input of the autoregressive prediction model based on TimesNet, so that the autoregressive prediction model is based on the previous The first concentration field latent space vector at each moment is used to perform autoregressive prediction on the gas pipeline leakage, and the obtained gas pipeline leakage is the time series concentration field of all subsequent time series (i.e., the first predicted time series concentration field).

[0035] In this process, first, an autoregressive prediction model is constructed based on TimesNet, and the autoregressive prediction model is trained through sample training data when the oil and gas pipeline leaks. Specifically, a concentration field dimensionality reduction model based on a variational autoencoder is constructed, and the concentration field dimensionality reduction model is trained through the concentration field sample data obtained by simulating the leakage and diffusion of the oil and gas pipeline. The concentration field sample data is converted into a concentration field latent space sample vector through the trained concentration field dimensionality reduction data to generate a reduced dimensionality sample data set of the natural gas concentration field when the oil and gas pipeline leaks to train the autoregressive prediction model. Thus, the concentration field dimensionality reduction data based on the variational autoencoder is self-supervised trained through the simulated data of the leakage and diffusion of the oil and gas pipeline, so as to reduce the dimensionality of the three-dimensional, uniform, and discrete concentration field into a set of latent space vectors (concentration field latent space sample vectors) of fixed length, so as to train the autoregressive prediction model.

[0036] Among them, when converting the concentration field sample data into the concentration field latent space sample vector, multiple concentration field sample data are respectively input into the encoder of the variational autoencoder based on the convolutional neural network, and multi-layer convolution operations are performed, and the convolution vector obtained by the last convolution operation is input into two fully connected layers, and the concentration field latent space sample vector is obtained from the output of the two fully connected layers. Specifically, according to the formula:

[0037] conduct The convolution operation is performed on the first Layer convolution operation The resulting convolution vector Input to the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector The mean ; and, Layer convolution operation The resulting convolution vector Input to the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector Variance . Then, according to the formula:

[0038] Generate multiple concentration field latent space sample vectors , to form a dimensionality reduction sample data set; where, is a nonlinear activation function, Indicates the first Layer convolution operation; is the bias vector of the convolutional neural network in the encoder of the variational autoencoder; are all integers, ; For the Concentration field sample data; , is a positive integer; is a noise vector that follows a standard normal distribution.

[0039] The concentration field sample database is constructed by simulating the leakage and diffusion of oil and gas pipelines to obtain the concentration field sample data when the oil and gas pipeline leaks. The generation of the concentration field latent space sample vector is carried out by reducing the dimension of all the concentration field sample data under each working condition and at each moment in the simulated concentration field sample database. The sample vectors of the latent space of the concentration field constitute a dimensionality reduction sample data set, and the autoregressive prediction model based on TimesNet is trained. The concentration field of the oil and gas pipeline leakage is predicted in time series through the trained autoregressive prediction model.

[0040] Specifically, firstly, based on the trained autoregressive prediction model, the The first concentration field latent space vector at each moment is used to perform time series prediction on the concentration field when the oil and gas pipeline leaks, and the first predicted concentration field latent space vector when the oil and gas pipeline leaks is obtained; then, based on the concentration field dimensionality reduction model, the first predicted concentration field latent space vector when the oil and gas pipeline leaks is restored and reconstructed. It should be noted that the concentration field dimensionality reduction model uses a variational autoencoder based on a convolutional neural network, which includes an encoder and a decoder. The decoder can reconstruct the concentration field latent space vector obtained by dimensionality reduction and restore it to a time series concentration field.

[0041] That is to say, the first predicted concentration field latent space vector obtained when the oil and gas pipeline leaks is input into the decoder of the variational autoencoder based on the convolutional neural network, and according to the formula:

[0042] The first prediction concentration field implicit space vector of oil and gas pipeline leakage Reconstruction is performed to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks. is a nonlinear activation function, Expressing the Layer input vector Conduct the Layer deconvolution operation; is the bias vector of the convolutional neural network in the decoder of the variational autoencoder; For the Layer deconvolution operation The output, For the Layer deconvolution operation Output: is an integer.

[0043] Therefore, through parameterized concentration field dimensionality reduction and time series autoregressive prediction, the input parameters such as leakage position, leakage direction, leakage pressure, leakage medium, leakage aperture, soil type, ambient wind speed, etc. during gas pipeline leakage are subjected to autoregressive prediction to obtain the predicted time series concentration field in any future time period when the gas pipeline leaks.

[0044] Step S103: In response to the first predicted time series concentration field The concentration field prediction data at the time is compared with the data obtained by leak diffusion simulation of oil and gas pipelines. The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data at the time of oil and gas pipeline leakage and The concentration field prediction data at each moment are fused to obtain the mixed concentration field data when the oil and gas pipeline leaks.

[0045] In the autoregressive prediction, the TimesNet-based autoregressive prediction model performs autoregressive prediction of the concentration field according to the n sets of vectors mapped by the parameters. In this process, the autoregressive prediction model uses the prediction results of each step as the input of the next step, and completes the full time series prediction in an iterative manner. In other words, the autoregressive prediction model uses the prediction results of the previous step as the input of the next step. However, due to the inevitable errors in the prediction itself, the deviations of each step will be superimposed on the prediction of the next step. These errors are continuously transmitted and amplified in the subsequent prediction steps, forming error accumulation, causing the prediction results to deviate from the true value, which may eventually make the prediction inaccurate or invalid.

[0046] In this regard, in this application, in the autoregressive prediction process, the field data (measured concentration field data) of oil and gas pipeline leakage is introduced at a certain step length to correct the error latent space vector in the autoregressive prediction process. Specifically, the first predicted time series concentration field obtained by the autoregressive prediction is The concentration field prediction data at the time is obtained by leak diffusion simulation of oil and gas pipelines The concentration field sample data at the time is calculated according to the formula:

[0047] Calculate the relative error between the predicted concentration field data and the simulated concentration field sample data at the corresponding time. Here, the average relative error is used to calculate the relative error of the predicted concentration field when the oil and gas pipeline leaks, and the relative error threshold is set to 10% to evaluate the cumulative error in the autoregressive prediction process. The concentration field prediction data at the time is compared with the data obtained by leak diffusion simulation of oil and gas pipelines. When the relative error of the concentration field sample data at the time is greater than or equal to 10%, the first predicted time series concentration field is repaired and corrected by using the concentration field measured data collected on site.

[0048] Specifically, first, the concentration field measured data when the oil and gas pipeline leaks is repaired. During the data repair process, the missing parts of the concentration field measured data are filled with 0 to generate uniform and discrete repaired concentration field measured data when the oil and gas pipeline leaks. Then, the measured concentration field data will be repaired and The concentration field prediction data at each moment is fused through a conditional adversarial network.

[0049] The conditional adversarial network consists of two parts: the generator and the discriminator. The generator is used to fuse data and generate a prediction structure that is close to the actual concentration field distribution. The discriminator continuously judges the difference between the generated data and the real data during the training process to optimize the generator parameters. and The predicted concentration field data at the moment is input into the generator, which uses a three-dimensional convolution layer to reduce the dimension of the two data to obtain a new feature dimension, and then restores the measured concentration field data. and The low-dimensional data of the concentration field prediction data at the time is weighted summed in the new feature dimension and deconvolved to restore the concentration field at the time of oil and gas pipeline leakage. Mixed concentration field data at time .

[0050] Specifically, based on the conditional adversarial network, according to the formula:

[0051] Repair concentration field measured data and Concentration field prediction data at time Fusion is performed to obtain the mixed concentration field data when the oil and gas pipeline leaks ; In the formula, is a nonlinear activation function, represents the deconvolution operation, represents the weighted summation of data features. Indicates the measured data of the repair concentration field Perform convolution operation, Represents the concentration field prediction data Perform convolution operation.

[0052] By means of The concentration field prediction data at the moment The relative error of the concentration field sample data at the time of the oil and gas pipeline leakage is judged by comparing the concentration field measured data and The concentration field prediction data at each moment are fused and reduced in dimension, and mixed concentration field data containing the characteristics of field measured data of oil and gas pipeline leakage are output to eliminate the accumulated error of autoregressive prediction.

[0053] In this application, when simulating the leakage and diffusion of oil and gas pipelines, the simulation operating data takes into account parameters such as pipeline leakage pressure, leakage direction, soil type, ambient wind speed, etc., and generates simulation data (concentration field sample data) for different types of pipeline leakage. For the oil pipeline scenario, the smooth particle dynamics (SPH) and discrete element method (DEM) coupling technology are used for leakage simulation, which can accurately simulate the diffusion process of liquid petroleum under complex terrain conditions. For the gas pipeline leakage scenario, a program based on the finite volume method (FVM) and component transport equations is used to simulate gas diffusion to describe the diffusion and transmission behavior of the gas after leakage.

[0054] Step S104: Input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline into the latent space dynamic prediction model based on TimesNet, and use the output of the oil and gas pipeline leakage time The corrected concentration field latent space vector replaces the first predicted concentration field latent space vector in the time period The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0055] After obtaining the mixed concentration field data when the oil and gas pipeline leaks, the encoder of the concentration field dimensionality reduction model based on the variational autoencoder is used to reduce the dimension of the mixed concentration field data to obtain the mixed concentration field latent space vector; then, the obtained mixed concentration field latent space vector and the leakage position, leakage direction, leakage pressure, leakage medium, leakage aperture, soil type, ambient wind speed and other parameters of the oil and gas pipeline leakage are input into the latent space dynamic prediction module to obtain the correction time ( ) before and after the latent space vector. That is, the latent space dynamic prediction model outputs the oil and gas pipeline leakage Implicit space vector of the corrected concentration field over the time period.

[0056] Next, replace the latent space vectors of several moments before and after the correction moment in the autoregressive prediction process The latent space vector at the time of the oil and gas pipeline leakage is calculated to achieve the correction and elimination of the accumulated error in the autoregressive prediction. The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to generate the second predicted concentration field latent space vector when the oil and gas pipeline leaks.

[0057] Finally, the decoder of the concentration field dimensionality reduction model reconstructs the first predicted concentration field latent space vector after data replacement and restores it to the three-dimensional concentration field when the gas pipeline leaks. That is, the decoder of the concentration field dimensionality reduction model based on the variational autoencoder reconstructs and restores the second predicted concentration field latent space vector to generate the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0058] Here, it should be noted that the concentration field dimensionality reduction model is constructed based on the variational autoencoder, and the variational autoencoder adopts an encoder and decoder architecture based on a convolutional neural network. The encoder and decoder both contain 5 convolution layers, the convolution kernel is 3-dimensional, the size is 3, and the stride is 2. Each layer in the architecture uses a convolution layer and a ReLU activation function. The fully connected layer is connected after the encoder to obtain the mean and variance of the latent space vector for time series prediction. The mean and variance are used to generate the latent space vector for time series prediction. The decoder is used to reconstruct the concentration field latent space vector obtained by dimensionality reduction and restore it to a time series concentration field. In this way, the corresponding data in the latent space of the first predicted concentration field is corrected using the correction data of the latent space dynamic prediction model, effectively eliminating the accumulated error in the autoregressive prediction process, achieving effective correction of the autoregressive prediction, and improving the prediction precision and accuracy.

[0059] At the same time, the prediction results can be displayed in three dimensions through visualization operations to intuitively present the diffusion path and range of gas pipeline leakage, identify dangerous areas and risk levels, help decision-makers to quickly evaluate and respond to sudden leakage accidents, and effectively improve the timeliness and accuracy of emergency response. Specifically, a visualization unit is built based on the open source PyVista library to combine complex prediction data with actual physical scenes to achieve a visual display of volume cloud maps. In addition, the visualization unit can also be used to fuse multi-source sensor data, such as environmental factors such as temperature, pressure, and wind speed, to further improve the accuracy of leakage prediction and the integrity of the scene. Furthermore, based on the displayed time series field data, the location of the leakage source, leakage volume, diffusion speed, etc. are determined, and the explosion limit, explosion overpressure injury range and other indicators can be combined to intuitively present the diffusion path and range of the leakage through three-dimensional display, identify dangerous areas and risk levels, so as to help decision makers quickly evaluate and respond to sudden leakage accidents and support on-site emergency decision-making.

[0060] like Figure 7 As shown, the embodiment of the present application also provides a parameterized oil and gas pipeline leakage diffusion prediction and correction system, including: Latent space generation unit 701 is configured based on A leakage parameter characterization model of a fully connected neural network is used to generate the previous prediction of the oil and gas pipeline leakage according to the leakage parameters of the oil and gas pipeline. The first concentration field latent space vector at the moment; among them, is a natural number; The autoregressive prediction unit 702 is configured to predict the oil and gas pipeline according to the previous The first concentration field latent space vector at the moment is used to perform time series prediction on the concentration field when the oil and gas pipeline leaks based on the pre-built autoregressive prediction model, and obtain the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; The data fusion unit 703 is configured to respond to the first predicted time series concentration field The concentration field prediction data at the time is compared with the data obtained by leak diffusion simulation of oil and gas pipelines. The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data at the time of oil and gas pipeline leakage and The concentration field prediction data at each moment are fused to obtain the mixed concentration field data when the oil and gas pipeline leaks; The prediction correction unit 304 is configured to input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline into the latent space dynamic prediction model based on TimesNet, and use the output oil and gas pipeline leakage time The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

[0061] The parameterized oil and gas pipeline leakage diffusion prediction self-correction system provided in the embodiments of the present application can implement the steps and processes of the parameterized oil and gas pipeline leakage diffusion prediction self-correction method of any of the above embodiments and achieve the same technical effects, which will not be repeated here.

[0062] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0063] In the present invention, the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0064] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A parameterized oil and gas pipeline leakage diffusion prediction self-correction method, characterized in that: include: Step S101: Based on A leakage parameter characterization model of a fully connected neural network is used to generate the leakage parameters of the oil and gas pipeline according to the leakage parameters of the oil and gas pipeline. of the moment The first concentration field hidden space vector; among them, is a positive integer; Step S102: according to the oil and gas pipeline leakage The first concentration field latent space vector at a moment, based on a pre-built autoregressive prediction model, performs time series prediction on the concentration field when the oil and gas pipeline leaks, and obtains the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; Step S103: In response to the first predicted time series concentration field The concentration field prediction data at the moment is compared with the leakage diffusion simulation of the oil and gas pipeline The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time are merged to obtain the mixed concentration field data when the oil and gas pipeline leaks; Step S104: input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline into the latent space dynamic prediction model based on TimesNet, and use the output of the oil and gas pipeline leakage time The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

2. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 1 is characterized in that: In step S101, The leakage parameters of the oil and gas pipeline when it leaks , input the leakage parameter characterization model The fully connected neural network structure , generating The first concentration field implicit space vector at the moment ; in, ； For the The fully connected neural network structure a weight matrix in the leakage parameter standard model; For the The fully connected neural network structure The bias vector of is the activation function; .

3. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 1 is characterized in that: Step S102 includes: According to the concentration field sample data obtained by leakage and diffusion simulation of the oil and gas pipeline, the constructed concentration field dimensionality reduction model based on the variational autoencoder is trained; Based on the trained concentration field dimensionality reduction model, the concentration field sample data is converted into a concentration field latent space sample vector to generate a dimensionality reduction sample data set of the natural gas concentration field when the oil and gas pipeline leaks; According to the reduced-dimensional sample data set, the constructed autoregressive prediction model based on TimesNet is trained.

4. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 3 is characterized in that: The concentration field dimensionality reduction model based on the training is used to convert the concentration field sample data into a concentration field latent space sample vector to generate a dimensionality reduction sample data set of the natural gas concentration field when the oil and gas pipeline leaks, specifically: The plurality of concentration field sample data are respectively input into the encoder of the variational autoencoder based on the convolutional neural network, according to the formula: ； conduct The convolution operation is performed on the first Layer convolution operation The resulting convolution vector Input the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector The mean ; and, Layer convolution operation The resulting convolution vector Input the fully connected layer of the encoder of the variational autoencoder , and obtain the concentration field hidden space sample vector Variance ; According to the formula: ； Correspondingly generate multiple concentration field hidden space sample vectors , to form the dimension-reduced sample data set; in, is a nonlinear activation function, Indicates that the concentration field sample data is processed Layer convolution operation; is the bias vector of the convolutional neural network in the encoder of the variational autoencoder; are integers, ; For the The concentration field sample data; , is a positive integer; is a noise vector that follows a standard normal distribution.

5. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 3 is characterized in that: In step S102, Based on the trained autoregressive prediction model, according to the oil and gas pipeline leakage a first concentration field latent space vector at a moment, performing time series prediction on the concentration field when the oil and gas pipeline leaks, and obtaining a first predicted concentration field latent space vector when the oil and gas pipeline leaks; Based on the concentration field dimensionality reduction model, a first predicted time series concentration field in a future time period when the oil and gas pipeline leaks is generated according to the first predicted concentration field latent space vector when the oil and gas pipeline leaks.

6. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 5 is characterized in that: The first predicted time series concentration field in the future time period when the oil and gas pipeline leaks is generated based on the concentration field dimensionality reduction model according to the first predicted concentration field latent space vector when the oil and gas pipeline leaks, specifically: The decoder of the variational autoencoder based on the convolutional neural network is as follows: ； The first predicted concentration field implicit space vector of the oil and gas pipeline leakage Reconstructing to generate a first predicted time series concentration field in a future time period when the oil and gas pipeline leaks; in, is a nonlinear activation function, Expressing the Layer input vector Conduct the Layer deconvolution operation; is the bias vector of the convolutional neural network in the decoder of the variational autoencoder; For the Layer deconvolution operation The output, For the Layer deconvolution operation Output: is an integer.

7. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 1 is characterized in that: In step S103, the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time is fused to obtain the mixed concentration field data when the oil and gas pipeline leaks, specifically: Repair the concentration field measured data when the oil and gas pipeline leaks to generate uniformly discrete repaired concentration field measured data when the oil and gas pipeline leaks ; Based on the conditional adversarial network, according to the formula: ； The measured data of the repair concentration field and The concentration field prediction data at time Fusion is performed to obtain the mixed concentration field data when the oil and gas pipeline leaks ; In the formula, is a nonlinear activation function, represents the deconvolution operation, represents the weighted summation of data features. Represents the measured data of the repair concentration field Perform convolution operation, Represents the concentration field prediction data Perform convolution operation.

8. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 1 is characterized in that: In step S104, An encoder based on a concentration field dimension reduction model of a variational autoencoder performs dimension reduction on the mixed concentration field data to obtain a latent space vector of the mixed concentration field.

9. The parameterized oil and gas pipeline leakage diffusion prediction self-correction method according to claim 1 is characterized in that: In step S104, use The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to generate the second predicted concentration field latent space vector when the oil and gas pipeline leaks; The decoder of the concentration field dimensionality reduction model based on the variational autoencoder reconstructs and restores the second predicted concentration field latent space vector to generate the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

10. A parameterized oil and gas pipeline leakage diffusion prediction and correction system, characterized in that: include: Latent space generation unit, configured based on A leakage parameter characterization model of a fully connected neural network is used to generate the leakage parameters of the oil and gas pipeline according to the leakage parameters of the oil and gas pipeline. The first concentration field latent space vector at the moment; among them, is a natural number; The autoregressive prediction unit is configured to predict the oil and gas pipeline according to the previous The first concentration field latent space vector at a moment, based on a pre-built autoregressive prediction model, performs time series prediction on the concentration field when the oil and gas pipeline leaks, and obtains the first predicted concentration field latent space vector when the oil and gas pipeline leaks, so as to generate the first predicted time series concentration field in the future time period when the oil and gas pipeline leaks; A data fusion unit is configured to respond to the first predicted time series concentration field The concentration field prediction data at the moment is compared with the leakage diffusion simulation of the oil and gas pipeline The relative error of the concentration field sample data at the time is greater than or equal to the preset error threshold, and the concentration field measured data when the oil and gas pipeline leaks and The concentration field prediction data at the time are merged to obtain the mixed concentration field data when the oil and gas pipeline leaks; The prediction and correction unit is configured to input the latent space vector of the mixed concentration field obtained by reducing the dimension of the mixed concentration field data and the leakage parameters of the oil and gas pipeline when leaking into the latent space dynamic prediction model based on TimesNet, and use the output of the oil and gas pipeline leakage time The corrected concentration field latent space vector in the time period replaces the first predicted concentration field latent space vector The latent space vector of the time period is used to obtain the second predicted time series concentration field in the future time period when the oil and gas pipeline leaks.

Images