CN116030897A - Separation method and system for oilfield associated gas - Google Patents

Abstract

The invention discloses a separation method and a separation system of oilfield associated gas, which are used for acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and monitoring videos of atomization states in the preset time period; the method comprises the steps of adopting an artificial intelligence technology based on deep learning to mine mode characteristics of working power distribution in different time spans in working power of an ultrasonic atomizer, and processing an atomization state monitoring video to obtain image state characteristics containing granular hydrate so as to adaptively adjust the working power of the ultrasonic atomizer based on the state of the granular hydrate. In this way, methane separation efficiency and sufficiency can be improved.

Description

Separation method and system for oilfield associated gas

Technical Field

The application relates to the technical field of intelligent separation, and more particularly relates to a separation method and a separation system of oilfield associated gas.

Background

The output of the oilfield associated gas is huge, but because of the distribution and dispersion of well sites and stations on land oilfield, the associated gas is basically discharged to the atmosphere at early stage of oilfield development or is treated by torch burning, and according to statistics, only small oilfield dispersed annually in China burns 10×10 ⁸ m ³ Not only does this result in waste of associated gas resources, but also increases the emission of hundreds of millions of greenhouse gases.

Associated gas mainly comprises two greenhouse gases of methane and carbon dioxide, and along with the increasing demands of national development on the environment, two common practices in oil fields currently exist: firstly, the associated gas is connected to a boiler of a well station through a pipeline to supply heat, but in summer with higher air temperature, the associated gas is mainly treated in a combustion mode, so that the utilization rate of the associated gas is very low. The other is in the form of a manifold, and the associated gas is transported to a light hydrocarbon plant for further processing by the light hydrocarbon plant.

From the composition point of view, the associated gas contains a large amount of greenhouse gases, mainly methane and a certain amount of carbon dioxide and nitrogen, so the associated gas is also a part of oil field yield, is a main yield source of a light hydrocarbon plant on site, is limited by the limit of economic yield and is complicated in process treatment, and the recovery technology cost of the associated gas is higher.

Thus, an optimized separation scheme for oilfield associated gas is desired.

Disclosure of Invention

The present application has been made in order to solve the above technical problems. The embodiment of the application provides a separation method and a separation system of oilfield associated gas, which are used for acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and monitoring videos of atomization states in the preset time period; the method comprises the steps of adopting an artificial intelligence technology based on deep learning to mine mode characteristics of working power distribution in different time spans in working power of an ultrasonic atomizer, and processing an atomization state monitoring video to obtain image state characteristics containing granular hydrate so as to adaptively adjust the working power of the ultrasonic atomizer based on the state of the granular hydrate. In this way, methane separation efficiency and sufficiency can be improved.

According to one aspect of the present application, there is provided a method of separating oilfield associated gas comprising:

acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period;

extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video;

the atomization state monitoring key frames at a plurality of preset time points are processed through a first convolution neural network model of a time attention mechanism to obtain atomization time sequence state feature vectors;

the working power of the ultrasonic atomizers at a plurality of preset time points is arranged into a working power input vector according to a time dimension, and then the working power input vector is transmitted through a time sequence encoder to obtain a working power characteristic vector;

calculating the responsiveness estimation of the atomization time sequence state feature vector relative to the working power feature vector based on a Gaussian density diagram to obtain a classification feature matrix; and

and passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

In the above method for separating oilfield associated gas, the step of obtaining the characteristic vector of the atomization time sequence state by using the first convolutional neural network model of the time attention mechanism to the atomization state monitoring key frames at the plurality of preset time points includes: passing the atomized state monitoring key frames at a plurality of preset time points through the first convolution neural network model using a time attention mechanism to obtain an atomized time sequence state characteristic diagram; and carrying out global averaging treatment on each feature matrix of the atomizing time sequence state feature diagram along the channel dimension to obtain the atomizing time sequence state feature vector.

In the above method for separating oilfield associated gas, the step of passing the atomized state monitoring key frames at the plurality of predetermined time points through the first convolutional neural network model using the time attention mechanism to obtain an atomized time sequence state feature map includes: extracting adjacent first and second frames from the atomized state monitoring key frames at the plurality of predetermined time points; passing the first frame and the second frame through a first convolution layer and a second convolution layer of the first convolution neural network model respectively to obtain a first frame feature map corresponding to the first frame and a second frame feature map corresponding to the second frame; performing position point multiplication on the first frame characteristic diagram and the second frame characteristic diagram, and then activating a function through Softmax to obtain a time attention diagram; passing the first frame through a third convolutional layer of the first convolutional neural network model to obtain a local feature map; and multiplying the local feature map and the time attention map by position points to obtain the atomization time sequence state feature map.

In the above separation method of oilfield associated gas, the time sequence encoder comprises a first convolution layer and a second convolution layer which are parallel, and a multi-scale fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use one-dimensional convolution kernels with different scales.

In the above separation method of oilfield associated gas, the step of arranging the working powers of the ultrasonic atomizers at the plurality of predetermined time points into working power input vectors according to a time dimension and then obtaining the working power feature vectors through a time sequence encoder includes: and performing full-connection coding on the working power input vector by using a full-connection layer of the time sequence coder to extract high-dimensional implicit characteristics of characteristic values of all positions in the working power input vector, wherein the formula is as follows:

wherein X is the operating power input vector, Y is the operating power output vector, W is a weight matrix, B is a bias vector,>

representing a matrix multiplication; and performing one-dimensional convolution coding on the working power input vector by using a one-dimensional convolution layer of the time sequence encoder to extract high-dimensional implicit correlation features among feature values of each position in the working power input vector, wherein the formula is as follows:

wherein a is the width of the convolution kernel in the X direction, F (a) is a convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the convolution kernel, and X represents the working power input vector.

In the above separation method of oilfield associated gas, the calculating, based on a gaussian density map, a response estimate of the atomization timing state feature vector relative to the working power feature vector to obtain a classification feature matrix includes: constructing a Gaussian density map of the atomization time sequence state characteristic vector and the working power characteristic vector to obtain a first Gaussian density map and a second Gaussian density map; calculating a responsiveness estimate of the first gaussian density map relative to the second gaussian density map to obtain a responsive gaussian density map; and performing Gaussian discretization on the responsive Gaussian density map to obtain the classification feature matrix.

In the above separation method of oilfield associated gas, the step of passing the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the working power of the ultrasonic atomizer at the current time point should be increased or decreased, includes: the classification feature matrix is subjected to matrix expansion to obtain a classification feature vector; carrying out local structure optimization on the classification feature vector to obtain an optimized classification feature vector; and passing the optimized classification feature vector through the classifier to obtain the classification result.

In the above separation method of oilfield associated gas, the performing local structural optimization on the classification feature vector to obtain an optimized classification feature vector includes: carrying out local structural optimization on the classification characteristic vector by the following formula to obtain an optimized classification characteristic vector; wherein, the formula is:

wherein V represents the pre-correction classification feature vector, V' represents the post-correction classification feature vector, V ^T A transpose vector representing the pre-correction classification feature vector,

representing the square of the two norms of the classification feature vector before correction, V _order Is an ordered vector in which the feature values of the classification feature vectors before correction are arranged in order of magnitude, and the classification feature vector V before correction is in the form of a column vector.

In the above separation method of oilfield associated gas, the step of passing the optimized classification feature vector through the classifier to obtain the classification result includes: performing full-connection coding on the optimized classification feature vector by using a plurality of full-connection layers of the classifier to obtain a coding classification feature vector; and passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

According to another aspect of the present application, there is provided a separation system for oilfield associated gas, comprising:

the data acquisition module is used for acquiring the working power of the ultrasonic atomizer at a plurality of preset time points in a preset time period and the atomization state monitoring video of the preset time period;

the key frame extraction module is used for extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video;

the feature extraction module is used for enabling the atomization state monitoring key frames at a plurality of preset time points to obtain atomization time sequence state feature vectors through a first convolution neural network model using a time attention mechanism;

the time sequence coding module is used for arranging the working powers of the ultrasonic atomizers at a plurality of preset time points into working power input vectors according to time dimension and then obtaining working power feature vectors through the time sequence coder;

the responsiveness estimation calculation module is used for calculating responsiveness estimation of the atomization time sequence state characteristic vector relative to the working power characteristic vector based on a Gaussian density chart so as to obtain a classification characteristic matrix; and

and the working power control module is used for passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the working power of the ultrasonic atomizer at the current time point should be increased or decreased.

Compared with the prior art, the method and the system for separating the oilfield associated gas acquire the working power of the ultrasonic atomizer at a plurality of preset time points in the preset time period, and monitor videos of the atomization state in the preset time period; the method comprises the steps of adopting an artificial intelligence technology based on deep learning to mine mode characteristics of working power distribution in different time spans in working power of an ultrasonic atomizer, and processing an atomization state monitoring video to obtain image state characteristics containing granular hydrate so as to adaptively adjust the working power of the ultrasonic atomizer based on the state of the granular hydrate. In this way, methane separation efficiency and sufficiency can be improved.

Drawings

The foregoing and other objects, features and advantages of the present application will become more apparent from the following more particular description of embodiments of the present application, as illustrated in the accompanying drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

Fig. 1 is a schematic view of a scenario of a separation method of oilfield associated gas according to an embodiment of the present application.

Fig. 2 is a flow chart of a method of separation of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 3 is a schematic diagram of an architecture of a separation method of oilfield associated gas according to an embodiment of the application.

Fig. 4 is a flow chart of the sub-steps of step S130 in a separation method of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 5 is a flow chart of the sub-steps of step S210 in a separation method of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 6 is a flow chart of the sub-steps of step S150 in a separation method of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 7 is a flow chart of the sub-steps of step S160 in a separation method of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 8 is a flow chart of the sub-steps of step S530 in a separation method of oilfield associated gas in accordance with an embodiment of the present application.

Fig. 9 is a block diagram of a separation system for oilfield associated gas in accordance with an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application and not all of the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

Scene overview

Aiming at the separation of the associated gas of the oil field, china patent publication No. CN111876211A discloses a device and a method for separating and recovering greenhouse gases in the associated gas of the oil field, which fully utilizes the phase state to realize the separation of carbon dioxide in the associated gas, realizes carbon capture and has good greenhouse gas emission reduction benefit, and utilizes the spray design to ensure that the contact surface of methane in the associated gas in a reaction chamber is increased, the reaction rate is improved, and the two-stage reaction chamber is adopted to ensure that the generation and decomposition processes of methane hydrate can be circularly carried out.

The scheme comprises the following steps:

step 1: the associated gas in the oil well is produced from the oil sleeve annulus through a sleeve valve gate, and enters a gas-liquid separator through a gas-liquid classifier inlet valve gate through a first well site main machine valve gate, a second well site main machine valve gate and a third well site main machine valve gate to separate water vapor and liquid drops contained in the associated gas;

step 2: the water separated by the gas-liquid separator passes through a liquid discharge valve gate to reach a liquid buried tank, the gas separated by the gas-liquid separator passes through a one-way valve, is pressurized into a carbon dioxide separator by a booster pump through an air compressor, and the pressure of associated gas exceeds the value by controlling the phase equilibrium pressure of carbon dioxide at the temperature by a temperature regulating chamber, so that the carbon dioxide becomes liquid and is separated from the associated gas;

Step 3: the associated gas which is obtained from the carbon dioxide separator and does not contain carbon dioxide passes through a one-way valve, passes through a primary associated gas inlet valve and a secondary associated gas inlet valve, respectively enters a primary reaction chamber and a secondary reaction chamber, and under the action of an ultrasonic atomizer, methane gas fully contacts with water mist and rapidly reacts to generate granular hydrate, so that methane is separated from the associated gas;

step 4: and (3) the decomposed methane enters a depressurization decomposing tank for storage, and when the methane reaches a certain amount, an outlet valve of the depressurization decomposing tank is opened, and a methane output pipeline is opened for further treatment.

In the implementation of the above scheme, it is found that in step 3, the operation power control of the ultrasonic atomizer greatly affects the separation efficiency and effect of methane. However, in existing solutions, the ultrasonic atomizer is operated at a fixed power, which clearly does not meet the need for efficient separation of methane. Therefore, in the technical scheme of the application, the separation scheme of the oilfield associated gas is further optimized.

Accordingly, in the technical scheme of the application, a power self-adaptive control scheme is configured for the ultrasonic atomizer to optimize the methane separation efficiency and sufficiency, specifically, the working power of the ultrasonic atomizer is adaptively adjusted based on the state of the granular hydrate so that methane gas and water mist can be fully contacted but not excessively contacted or insufficiently contacted, and in such a way, the methane separation efficiency and sufficiency are improved. The construction of the scheme comprises the following keys: 1. expressing the state characteristics of the particulate hydrate in a suitable manner; 2. and constructing a correlation model between the state characteristics of the granular hydrate and the working power of the ultrasonic atomizer.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks have also shown levels approaching and even exceeding humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like. Deep learning and the development of neural networks provide new solutions and solutions for the above-mentioned key breakthroughs and implementations.

Specifically, first, the working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period are obtained. That is, the state characteristics of the particulate hydrate are represented by the atomized state monitoring video for the predetermined period of time. Specifically, the atomization state monitoring video of the preset time period is passed through a first convolution neural network model using a time attention mechanism to obtain an atomization time sequence state characteristic vector. More specifically, the atomized state monitoring video is processed using a deep learning based convolutional neural network model having excellent performance in the field of image feature extraction as a feature extractor to obtain the atomized time-series state feature vector containing image state features of particulate hydrate.

Particularly, in the technical scheme of the application, in the working power control of the ultrasonic atomizer, if the change rule of the state characteristics of the granular hydrate in the time sequence dimension can be focused more, the accuracy of the power self-adaptive control can be improved. Therefore, in the technical scheme of the application, a time attention mechanism is introduced into the convolutional neural network model so that the mode characteristics of the state characteristics of the granular hydrate in the time sequence dimension have stronger characteristic significance.

Aiming at the working powers of the ultrasonic atomizers at a plurality of preset time points, the working powers of the ultrasonic atomizers at the preset time points are arranged into working power input vectors according to the time dimension and then pass through a time sequence encoder to obtain working power characteristic vectors. In particular, in the technical solution of the present application, the timing encoder includes a first convolution layer and a second convolution layer that are parallel, and a multi-scale fusion layer connected to the first convolution layer and the second convolution layer, where the first convolution layer and the second convolution layer use one-dimensional convolution kernels having different scales. In the encoding process of the time sequence encoder, one-dimensional convolution encoding based on one-dimensional convolution kernels of different scales is carried out on the working power input vector so as to capture mode characteristics of working power distribution in different time spans in the working power input vector, and the working power characteristic vector is obtained.

In the logic level, the power of the ultrasonic atomizer is the cause, and the state characteristic change of the granular hydrate is the effect, so in the technical scheme of the application, the responsiveness estimation of the atomizing time sequence state characteristic vector relative to the working power characteristic vector is calculated further based on a Gaussian density chart so as to obtain a classification characteristic matrix. And further, the classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for representing that the working power of the ultrasonic atomizer at the current time point is increased or decreased. That is, the class probability label to which the classification feature matrix belongs is determined by the classifier, wherein the class probability label includes that the operation power of the ultrasonic atomizer at the current time point should be increased (first label) and that the operation power of the ultrasonic atomizer at the current time point should be decreased (second label). It should be noted that the first tag and the second tag are operation power control tags of an ultrasonic nebulizer, and thus, after the classification result is obtained, a power adaptive control strategy of the ultrasonic nebulizer may be obtained based on the classification result.

Particularly, in the technical scheme of the application, when the classification feature matrix is obtained by calculating the response estimation of the atomization time sequence state feature vector relative to the working power feature vector, partial randomness is introduced in the response estimation process using a Gaussian density chart, so that the feature distribution of the classification feature matrix may have local structural ambiguity, the expression certainty of the classification feature matrix is reduced, and the accuracy of a classification result obtained by the classification feature matrix through a classifier is influenced.

Based on this, the applicant of the present application performs vector ordered hilbert completion on the classification feature vector V obtained after the classification feature matrix is developed, which is expressed as:

v and V' are classification feature vectors before and after correction respectively,

representing the square of the two norms of the classification feature vector, i.e. the inner product of the classification feature vector itself, V _order Is an ordered vector in which the feature values of the classification feature vectors are arranged in order of magnitude, and the classification feature vector V is a column vector form.

Here, by mapping the ordered vectors into the hilbert space defined by the self-inner product of the vectors, a meaningful measure of the numerical relation of the feature set in the consistency space can be realized, based on which, a feature space with an orthorhombic structure is built by embedding the relative positions of the feature vectors, and the structure in the feature space is completed for the high-dimensional manifold of the feature vectors based on vector query, so that the reduction of the expression certainty of the feature vectors due to the fuzzification structure can be avoided, and the accuracy of the classification result obtained by the classifier of the classification feature matrix can be improved.

Based on this, the present application provides a separation method of oilfield associated gas, which includes: acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period; extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video; the atomization state monitoring key frames at a plurality of preset time points are processed through a first convolution neural network model of a time attention mechanism to obtain atomization time sequence state feature vectors; the working power of the ultrasonic atomizers at a plurality of preset time points is arranged into a working power input vector according to a time dimension, and then the working power input vector is transmitted through a time sequence encoder to obtain a working power characteristic vector; calculating the responsiveness estimation of the atomization time sequence state feature vector relative to the working power feature vector based on a Gaussian density diagram to obtain a classification feature matrix; and passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for representing that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

Fig. 1 is a schematic view of a scenario of a separation method of oilfield associated gas according to an embodiment of the present application. As shown in fig. 1, in this application scenario, first, the operating power of an ultrasonic atomizer (e.g., C1 as illustrated in fig. 1) at a plurality of predetermined time points within a predetermined period of time and an atomization state monitoring video (e.g., C2 as illustrated in fig. 1) of the predetermined period of time are acquired; the acquired operating power and nebulization status monitoring video of the ultrasonic nebulizer is then input into a server (e.g., S as illustrated in fig. 1) deployed with a separation algorithm of oilfield associated gas, wherein the server is capable of processing the operating power of the ultrasonic nebulizer and the nebulization status monitoring video based on the separation algorithm of oilfield associated gas to generate a classification result indicating whether the operating power of the ultrasonic nebulizer should be increased or decreased at the current point in time.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.

Exemplary method

Fig. 2 is a flow chart of a method of separation of oilfield associated gas in accordance with an embodiment of the present application. As shown in fig. 2, the method for separating oilfield associated gas according to the embodiment of the application includes the steps of: s110, acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period; s120, extracting atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video; s130, enabling the atomization state monitoring key frames at a plurality of preset time points to pass through a first convolution neural network model using a time attention mechanism to obtain an atomization time sequence state characteristic vector; s140, arranging the working powers of the ultrasonic atomizers at a plurality of preset time points into working power input vectors according to time dimension, and then obtaining working power feature vectors through a time sequence encoder; s150, calculating the response estimation of the characteristic vector of the atomization time sequence state relative to the characteristic vector of the working power based on a Gaussian density diagram to obtain a classification characteristic matrix; and S160, passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

Fig. 3 is a schematic diagram of an architecture of a separation method of oilfield associated gas according to an embodiment of the application. As shown in fig. 3, in the network architecture, first, working power of an ultrasonic atomizer at a plurality of predetermined time points in a predetermined time period and an atomization state monitoring video of the predetermined time period are acquired; then extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video; s130, enabling the atomization state monitoring key frames at a plurality of preset time points to pass through a first convolution neural network model using a time attention mechanism to obtain an atomization time sequence state characteristic vector; then, the working power of the ultrasonic atomizers at a plurality of preset time points is arranged into a working power input vector according to a time dimension, and then the working power input vector is obtained through a time sequence encoder; then, based on a Gaussian density map, calculating the responsiveness estimation of the atomization time sequence state feature vector relative to the working power feature vector to obtain a classification feature matrix; and finally, the classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for representing that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

Specifically, in step S110, the operating power of the ultrasonic atomizer at a plurality of predetermined time points within a predetermined period of time and the atomization state monitoring video of the predetermined period of time are acquired. Aiming at the separation of the associated gas of the oil field, china patent publication No. CN111876211A discloses a device and a method for separating and recovering greenhouse gases in the associated gas of the oil field, which fully utilizes the phase state to realize the separation of carbon dioxide in the associated gas, realizes carbon capture and has good greenhouse gas emission reduction benefit, and utilizes the spray design to ensure that the contact surface of methane in the associated gas in a reaction chamber is increased, the reaction rate is improved, and the two-stage reaction chamber is adopted to ensure that the generation and decomposition processes of methane hydrate can be circularly carried out.

The scheme comprises the following steps:

step 1: the associated gas in the oil well is produced from the oil sleeve annulus through a sleeve valve gate, and enters a gas-liquid separator through a gas-liquid classifier inlet valve gate through a first well site main machine valve gate, a second well site main machine valve gate and a third well site main machine valve gate to separate water vapor and liquid drops contained in the associated gas;

step 2: the water separated by the gas-liquid separator passes through a liquid discharge valve gate to reach a liquid buried tank, the gas separated by the gas-liquid separator passes through a one-way valve, is pressurized into a carbon dioxide separator by a booster pump through an air compressor, and the pressure of associated gas exceeds the value by controlling the phase equilibrium pressure of carbon dioxide at the temperature by a temperature regulating chamber, so that the carbon dioxide becomes liquid and is separated from the associated gas;

Step 3: the associated gas which is obtained from the carbon dioxide separator and does not contain carbon dioxide passes through a one-way valve, passes through a primary associated gas inlet valve and a secondary associated gas inlet valve, respectively enters a primary reaction chamber and a secondary reaction chamber, and under the action of an ultrasonic atomizer, methane gas fully contacts with water mist and rapidly reacts to generate granular hydrate, so that methane is separated from the associated gas;

step 4: and (3) the decomposed methane enters a depressurization decomposing tank for storage, and when the methane reaches a certain amount, an outlet valve of the depressurization decomposing tank is opened, and a methane output pipeline is opened for further treatment.

In the implementation of the above scheme, it is found that in step 3, the operation power control of the ultrasonic atomizer greatly affects the separation efficiency and effect of methane. However, in existing solutions, the ultrasonic atomizer is operated at a fixed power, which clearly does not meet the need for efficient separation of methane. Therefore, in the technical scheme of the application, the separation scheme of the oilfield associated gas is further optimized.

Accordingly, in the technical scheme of the application, a power self-adaptive control scheme is configured for the ultrasonic atomizer to optimize the methane separation efficiency and sufficiency, specifically, the working power of the ultrasonic atomizer is adaptively adjusted based on the state of the granular hydrate so that methane gas and water mist can be fully contacted but not excessively contacted or insufficiently contacted, and in such a way, the methane separation efficiency and sufficiency are improved. The construction of the scheme comprises the following keys: 1. expressing the state characteristics of the particulate hydrate in a suitable manner; 2. and constructing a correlation model between the state characteristics of the granular hydrate and the working power of the ultrasonic atomizer.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks have also shown levels approaching and even exceeding humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like. Deep learning and the development of neural networks provide new solutions and solutions for the above-mentioned key breakthroughs and implementations.

Specifically, first, the working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period are obtained. That is, the state characteristics of the particulate hydrate are represented by the atomized state monitoring video for the predetermined period of time.

Specifically, in step S120, the atomized state monitoring key frames at the plurality of predetermined time points are extracted from the atomized state monitoring video. Next, considering that in the atomization state monitoring video, in order to improve the accuracy of the association relationship between the working power of the ultrasonic atomizer and the atomization state of the atomization state monitoring video, which is extracted later, in the technical scheme of the application, the atomization state monitoring key frames of the plurality of preset time points are extracted from the atomization state monitoring video.

Specifically, in step S130, the atomized state monitoring key frames at the plurality of predetermined time points are passed through a first convolutional neural network model using a time-focusing mechanism to obtain an atomized time sequence state feature vector. Specifically, the atomization state monitoring video of the preset time period is passed through a first convolution neural network model using a time attention mechanism to obtain an atomization time sequence state characteristic vector. More specifically, the atomized state monitoring video is processed using a deep learning based convolutional neural network model having excellent performance in the field of image feature extraction as a feature extractor to obtain the atomized time-series state feature vector containing image state features of particulate hydrate.

Particularly, in the technical scheme of the application, in the working power control of the ultrasonic atomizer, if the change rule of the state characteristics of the granular hydrate in the time sequence dimension can be focused more, the accuracy of the power self-adaptive control can be improved. Therefore, in the technical scheme of the application, a time attention mechanism is introduced into the convolutional neural network model so that the mode characteristics of the state characteristics of the granular hydrate in the time sequence dimension have stronger characteristic significance.

In an embodiment of the present application, fig. 4 is a flowchart of a sub-step of step S130 in a separation method of oilfield associated gas according to an embodiment of the present application, and as shown in fig. 4, the step of passing the atomized state monitoring key frames at the plurality of predetermined time points through a first convolutional neural network model using a time attention mechanism to obtain an atomized time sequence state feature vector includes: s210, enabling the atomization state monitoring key frames at a plurality of preset time points to pass through a first convolution neural network model using a time attention mechanism to obtain an atomization time sequence state characteristic diagram; and S220, carrying out global average pooling processing on each feature matrix of the atomization time sequence state feature diagram along the channel dimension to obtain the atomization time sequence state feature vector.

Fig. 5 is a flowchart of a sub-step of step S210 in the oilfield associated gas separation method according to an embodiment of the present application, as shown in fig. 5, the step of passing the atomized state monitoring key frames at the plurality of predetermined time points through the first convolutional neural network model using the time attention mechanism to obtain an atomized time sequence state feature map includes: s310, extracting adjacent first frames and second frames from the atomization state monitoring key frames at the plurality of preset time points; s320, enabling the first frame and the second frame to pass through a first convolution layer and a second convolution layer of the first convolution neural network model respectively so as to obtain a first frame characteristic diagram corresponding to the first frame and a second frame characteristic diagram corresponding to the second frame; s330, multiplying the first frame characteristic diagram and the second frame characteristic diagram according to position points, and then activating a function through Softmax to obtain a time attention diagram; s340, the first frame passes through a third convolution layer of the first convolution neural network model to obtain a local feature map; and S350, multiplying the local characteristic diagram and the time attention diagram by position points to obtain the atomization time sequence state characteristic diagram.

It should be appreciated that the mechanism of attention (Attention Mechanism) stems from research on human vision. In cognitive sciences, due to bottlenecks in information processing, humans may selectively focus on a portion of all information while ignoring other visible information, a mechanism commonly referred to as an attention mechanism. Time attention can be seen as a dynamic time selection mechanism that decides when to pay attention and is therefore commonly used for video processing.

Specifically, in step S140, the working powers of the ultrasonic atomizers at the plurality of predetermined time points are arranged into working power input vectors according to a time dimension, and then the working power input vectors are obtained through a time sequence encoder. Aiming at the working powers of the ultrasonic atomizers at a plurality of preset time points, the working powers of the ultrasonic atomizers at the preset time points are arranged into working power input vectors according to the time dimension and then pass through a time sequence encoder to obtain working power characteristic vectors.

In particular, in the technical solution of the present application, the timing encoder includes a first convolution layer and a second convolution layer that are parallel, and a multi-scale fusion layer connected to the first convolution layer and the second convolution layer, where the first convolution layer and the second convolution layer use one-dimensional convolution kernels having different scales. In the encoding process of the time sequence encoder, one-dimensional convolution encoding based on one-dimensional convolution kernels of different scales is carried out on the working power input vector so as to capture mode characteristics of working power distribution in different time spans in the working power input vector, and the working power characteristic vector is obtained.

Further, the step of arranging the working powers of the ultrasonic atomizers at the plurality of preset time points into working power input vectors according to a time dimension and then obtaining the working power feature vectors through a time sequence encoder includes the steps of: and performing full-connection coding on the working power input vector by using a full-connection layer of the time sequence coder to extract high-dimensional implicit characteristics of characteristic values of all positions in the working power input vector, wherein the formula is as follows:

wherein X is the operating power input vector, Y is the operating power output vector, W is a weight matrix, B is a bias vector,

representing a matrix multiplication; and performing one-dimensional convolution coding on the working power input vector by using a one-dimensional convolution layer of the time sequence encoder to extract high-dimensional implicit correlation features among feature values of each position in the working power input vector, wherein the formula is as follows:

wherein a is the width of the convolution kernel in the X direction, F (a) is a convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the convolution kernel, and X represents the working power input vector.

Specifically, in step S150, based on the gaussian density map, a responsiveness estimate of the atomizing timing state feature vector with respect to the operating power feature vector is calculated to obtain a classification feature matrix. In the logic level, the power of the ultrasonic atomizer is the cause, and the state characteristic change of the granular hydrate is the effect, so in the technical scheme of the application, the responsiveness estimation of the atomizing time sequence state characteristic vector relative to the working power characteristic vector is calculated further based on a Gaussian density chart so as to obtain a classification characteristic matrix.

Fig. 6 is a flowchart of a sub-step of step S150 in the separation method of oilfield associated gas according to an embodiment of the present application, and as shown in fig. 6, the calculating, based on the gaussian density map, a response estimate of the atomization timing state feature vector relative to the working power feature vector to obtain a classification feature matrix includes: s410, constructing a Gaussian density map of the atomization time sequence state characteristic vector and the working power characteristic vector to obtain a first Gaussian density map and a second Gaussian density map; s420, calculating the response estimation of the first Gaussian density map relative to the second Gaussian density map to obtain a response Gaussian density map; and S430, performing Gaussian discretization on the response Gaussian density map to obtain the classification feature matrix.

Specifically, in step S160, the classification feature matrix is passed through a classifier to obtain a classification result, which indicates that the operating power of the ultrasonic atomizer at the current time point should be increased or decreased. And further, the classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for representing that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

That is, the class probability label to which the classification feature matrix belongs is determined by the classifier, wherein the class probability label includes that the operation power of the ultrasonic atomizer at the current time point should be increased (first label) and that the operation power of the ultrasonic atomizer at the current time point should be decreased (second label). It should be noted that the first tag and the second tag are operation power control tags of an ultrasonic nebulizer, and thus, after the classification result is obtained, a power adaptive control strategy of the ultrasonic nebulizer may be obtained based on the classification result.

Fig. 7 is a flowchart of a sub-step of step S160 in the separation method of oilfield associated gas according to an embodiment of the present application, as shown in fig. 7, where the classification feature matrix is passed through a classifier to obtain a classification result, where the classification result is used to indicate that the working power of the ultrasonic atomizer at the current time point should be increased or decreased, and includes: s510, performing matrix expansion on the classification characteristic matrix to obtain a classification characteristic vector; s520, carrying out local structure optimization on the classification feature vector to obtain an optimized classification feature vector; and S530, passing the optimized classification feature vector through the classifier to obtain the classification result.

Particularly, in the technical scheme of the application, when the classification feature matrix is obtained by calculating the response estimation of the atomization time sequence state feature vector relative to the working power feature vector, partial randomness is introduced in the response estimation process using a Gaussian density chart, so that the feature distribution of the classification feature matrix may have local structural ambiguity, the expression certainty of the classification feature matrix is reduced, and the accuracy of a classification result obtained by the classification feature matrix through a classifier is influenced.

Based on this, the applicant of the present application performs vector ordered hilbert completion on the classification feature vector V obtained after the classification feature matrix is developed, that is, performs local structural optimization on the classification feature vector to obtain an optimized classification feature vector, and includes: carrying out local structural optimization on the classification characteristic vector by the following formula to obtain an optimized classification characteristic vector; wherein, the formula is:

wherein V represents the pre-correction classification feature vector, V' represents the post-correction classification feature vector, V ^T A transpose vector representing the pre-correction classification feature vector,

Representing the square of the two norms of the classification feature vector before correction, V _order Is an ordered vector in which the feature values of the classification feature vectors before correction are arranged in order of magnitude, and the classification feature vector V before correction is in the form of a column vector.

Here, by mapping the ordered vectors into the hilbert space defined by the self-inner product of the vectors, a meaningful measure of the numerical relation of the feature set in the consistency space can be realized, based on which, a feature space with an orthorhombic structure is built by embedding the relative positions of the feature vectors, and the structure in the feature space is completed for the high-dimensional manifold of the feature vectors based on vector query, so that the reduction of the expression certainty of the feature vectors due to the fuzzification structure can be avoided, and the accuracy of the classification result obtained by the classifier of the classification feature matrix can be improved.

Fig. 8 is a flowchart of a sub-step of step S530 in the separation method of oilfield associated gas according to an embodiment of the present application, as shown in fig. 8, where the step of passing the optimized classification feature vector through the classifier to obtain the classification result includes: s610, performing full-connection coding on the optimized classification feature vector by using a plurality of full-connection layers of the classifier to obtain a coded classification feature vector; and S620, passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

In a specific example of the application, the classifier is used to process the optimized classification feature vector in the following formula to obtain the classification result; wherein, the formula is:

the formula is: o=softmax { (W) _n ,B _n ):…:(W ₁ ,B ₁ ) X, where X represents the optimized classification characteristicSign vector, W ₁ To W _n Is a weight matrix, B ₁ To B _n Representing the bias vector.

In summary, according to the method for separating oilfield associated gas in the embodiment of the application, working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period are obtained; the method comprises the steps of adopting an artificial intelligence technology based on deep learning to mine mode characteristics of working power distribution in different time spans in working power of an ultrasonic atomizer, and processing an atomization state monitoring video to obtain image state characteristics containing granular hydrate so as to adaptively adjust the working power of the ultrasonic atomizer based on the state of the granular hydrate. In this way, methane separation efficiency and sufficiency can be improved.

Exemplary System

Fig. 9 is a block diagram of a separation system for oilfield associated gas in accordance with an embodiment of the present application. As shown in fig. 9, a separation system 100 for oilfield associated gas according to an embodiment of the present application includes: a data acquisition module 110, configured to acquire working powers of an ultrasonic atomizer at a plurality of predetermined time points within a predetermined time period and an atomization state monitoring video of the predetermined time period; a key frame extracting module 120, configured to extract the atomized state monitoring key frames at the plurality of predetermined time points from the atomized state monitoring video; the feature extraction module 130 is configured to obtain an atomized time sequence state feature vector by using a first convolutional neural network model of a time attention mechanism through the atomized state monitoring key frames at the plurality of predetermined time points; the time sequence encoding module 140 is configured to arrange the working powers of the ultrasonic atomizers at the plurality of predetermined time points into working power input vectors according to a time dimension, and then obtain working power feature vectors through the time sequence encoder; a responsiveness estimation calculation module 150, configured to calculate a responsiveness estimation of the atomization timing state feature vector relative to the working power feature vector based on a gaussian density map to obtain a classification feature matrix; and an operating power control module 160, configured to pass the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the operating power of the ultrasonic atomizer at the current time point should be increased or decreased.

In one example, in the separation system 100 of oilfield associated gas described above, the feature extraction module includes: the time attention unit is used for enabling the atomization state monitoring key frames at a plurality of preset time points to pass through the first convolution neural network model using the time attention mechanism so as to obtain an atomization time sequence state characteristic diagram; and the global averaging unit is used for carrying out global averaging processing on each feature matrix of the atomizing time sequence state feature diagram along the channel dimension so as to obtain the atomizing time sequence state feature vector.

In one example, in the above-described separation system 100 for oilfield associated gas, the time attention unit includes: an adjacent frame extraction subunit, configured to extract adjacent first frames and second frames from the atomized state monitoring key frames at the plurality of predetermined time points; a first feature extraction subunit, configured to pass the first frame and the second frame through a first convolution layer and a second convolution layer of the first convolutional neural network model, respectively, to obtain a first frame feature map corresponding to the first frame and a second frame feature map corresponding to the second frame; the activating subunit is used for multiplying the first frame characteristic diagram and the second frame characteristic diagram according to position points and then obtaining a time attention diagram through a Softmax activating function; a second feature extraction subunit, configured to pass the first frame through a third convolutional layer of the first convolutional neural network model to obtain a local feature map; and a calculating subunit, configured to multiply the local feature map and the time attention map by a location point to obtain the atomization time sequence state feature map.

In one example, in the above-described oilfield associated gas separation system 100, the timing encoder includes first and second parallel convolution layers and a multi-scale fusion layer coupled to the first and second convolution layers, wherein the first and second convolution layers use one-dimensional convolution kernels having different scales.

In one example, in the above-described oilfield associated gas separation system 100, the timingAn encoding module, comprising: and the full-connection unit is used for carrying out full-connection coding on the working power input vector by using a full-connection layer of the time sequence encoder according to the following formula to extract high-dimensional implicit characteristics of characteristic values of all positions in the working power input vector, wherein the formula is as follows:

wherein X is the operating power input vector, Y is the operating power output vector, W is a weight matrix, B is a bias vector,>

representing a matrix multiplication; the one-dimensional convolution unit is used for carrying out one-dimensional convolution coding on the working power input vector by using a one-dimensional convolution layer of the time sequence encoder to extract high-dimensional implicit correlation features among feature values of all positions in the working power input vector, wherein the formula is as follows:

Wherein a is the width of the convolution kernel in the X direction, F (a) is a convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the convolution kernel, and X represents the working power input vector.

In one example, in the above-described separation system 100 for oilfield associated gas, the responsiveness estimation calculation module includes: the Gaussian density map construction unit is used for constructing the Gaussian density map of the atomization time sequence state characteristic vector and the working power characteristic vector to obtain a first Gaussian density map and a second Gaussian density map; a computing unit configured to compute a responsiveness estimate of the first gaussian density map relative to the second gaussian density map to obtain a responsiveness gaussian density map; and the Gaussian discretization unit is used for performing Gaussian discretization on the responsive Gaussian density map to obtain the classification characteristic matrix.

In one example, in the above separation system 100 for oilfield associated gas, the working power control module includes: the matrix unfolding unit is used for conducting matrix unfolding on the classification characteristic matrix to obtain a classification characteristic vector; the local structure optimization unit is used for carrying out local structure optimization on the classification characteristic vector so as to obtain an optimized classification characteristic vector; and the classification result generating unit is used for enabling the optimized classification feature vector to pass through the classifier to obtain the classification result.

In one example, in the separation system 100 of oilfield associated gas described above, the local structure optimization unit is configured to: carrying out local structural optimization on the classification characteristic vector by the following formula to obtain an optimized classification characteristic vector; wherein, the formula is:

wherein V represents the pre-correction classification feature vector, V' represents the post-correction classification feature vector, V ^T A transpose vector representing the pre-correction classification feature vector,

representing the square of the two norms of the classification feature vector before correction, V _order Is an ordered vector in which the feature values of the classification feature vectors before correction are arranged in order of magnitude, and the classification feature vector V before correction is in the form of a column vector.

In one example, in the separation system 100 of oilfield associated gas, the classification result generating unit includes: the full-connection coding subunit is used for carrying out full-connection coding on the optimized classification feature vector by using a plurality of full-connection layers of the classifier so as to obtain a coding classification feature vector; and a classification result subunit, configured to pass the encoded classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

Here, it will be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described oilfield associated gas separation system 100 have been described in detail in the above description of the oilfield associated gas separation method with reference to fig. 1 to 8, and thus, repetitive descriptions thereof will be omitted.

The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not limiting, and these advantages, benefits, effects, etc. are not to be considered as necessarily possessed by the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not intended to be limited to the details disclosed herein as such.

The block diagrams of the devices, apparatuses, devices, systems referred to in this application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.

It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent to the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims (10)

1. A method of separating oilfield associated gas comprising:

acquiring working power of an ultrasonic atomizer at a plurality of preset time points in a preset time period and an atomization state monitoring video of the preset time period;

Extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video;

the atomization state monitoring key frames at a plurality of preset time points are processed through a first convolution neural network model of a time attention mechanism to obtain atomization time sequence state feature vectors;

the working power of the ultrasonic atomizers at a plurality of preset time points is arranged into a working power input vector according to a time dimension, and then the working power input vector is transmitted through a time sequence encoder to obtain a working power characteristic vector;

calculating the responsiveness estimation of the atomization time sequence state feature vector relative to the working power feature vector based on a Gaussian density diagram to obtain a classification feature matrix; and

and passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the working power of the ultrasonic atomizer at the current time point is increased or decreased.

2. The method of claim 1, wherein said passing the atomized status monitoring key frames at the plurality of predetermined time points through a first convolutional neural network model using a time attention mechanism to obtain an atomized time sequence status feature vector comprises:

Passing the atomized state monitoring key frames at a plurality of preset time points through the first convolution neural network model using a time attention mechanism to obtain an atomized time sequence state characteristic diagram; and

and carrying out global averaging pooling treatment on each feature matrix of the atomizing time sequence state feature diagram along the channel dimension to obtain the atomizing time sequence state feature vector.

3. The method of claim 2, wherein passing the plurality of predetermined time points of the fog state monitoring key frames through the first convolutional neural network model using the time attention mechanism to obtain the fog time sequence state feature map comprises:

extracting adjacent first and second frames from the atomized state monitoring key frames at the plurality of predetermined time points;

passing the first frame and the second frame through a first convolution layer and a second convolution layer of the first convolution neural network model respectively to obtain a first frame feature map corresponding to the first frame and a second frame feature map corresponding to the second frame;

performing position point multiplication on the first frame characteristic diagram and the second frame characteristic diagram, and then activating a function through Softmax to obtain a time attention diagram;

Passing the first frame through a third convolutional layer of the first convolutional neural network model to obtain a local feature map; and

multiplying the local feature map by the time attention map by a position point to obtain the atomization time sequence state feature map.

4. A method of separating oilfield associated gas as defined in claim 3, wherein the timing encoder comprises first and second parallel convolution layers and a multi-scale fusion layer coupled to the first and second convolution layers, wherein the first and second convolution layers use one-dimensional convolution kernels having different scales.

5. The method for separating oilfield associated gas according to claim 4, wherein the step of arranging the operating powers of the ultrasonic atomizers at the plurality of predetermined time points into the operating power input vector according to the time dimension and then passing the operating power input vector through the time sequence encoder to obtain the operating power feature vector comprises the steps of:

and performing full-connection coding on the working power input vector by using a full-connection layer of the time sequence coder to extract high-dimensional implicit characteristics of characteristic values of all positions in the working power input vector, wherein the formula is as follows:

Wherein X is the operating power input vector, Y is the operating power output vector, W is a weight matrix, B is a bias vector,>

representing a matrix multiplication;

and performing one-dimensional convolution coding on the working power input vector by using a one-dimensional convolution layer of the time sequence encoder to extract high-dimensional implicit correlation features among feature values of each position in the working power input vector, wherein the formula is as follows:

wherein a is the width of the convolution kernel in the X direction, F (a) is a convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the convolution kernel, and X represents the working power input vector.

6. The method of claim 5, wherein calculating a responsiveness estimate of the atomization timing state eigenvector relative to the operating power eigenvector based on a gaussian density map to obtain a classification eigenvector comprises:

constructing a Gaussian density map of the atomization time sequence state characteristic vector and the working power characteristic vector to obtain a first Gaussian density map and a second Gaussian density map;

calculating a responsiveness estimate of the first gaussian density map relative to the second gaussian density map to obtain a responsive gaussian density map; and

And performing Gaussian discretization on the response Gaussian density map to obtain the classification characteristic matrix.

7. The method of claim 6, wherein the step of passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the operating power of the ultrasonic atomizer at the current time point should be increased or decreased, comprises:

the classification feature matrix is subjected to matrix expansion to obtain a classification feature vector;

carrying out local structure optimization on the classification feature vector to obtain an optimized classification feature vector; and

and passing the optimized classification feature vector through the classifier to obtain the classification result.

8. The method of claim 7, wherein the locally structurally optimizing the classification feature vector to obtain an optimized classification feature vector comprises: carrying out local structural optimization on the classification characteristic vector by the following formula to obtain an optimized classification characteristic vector;

wherein, the formula is:

wherein V represents the pre-correction classification feature vector, V' represents the post-correction classification feature vector, V ^T A transpose vector representing the pre-correction classification feature vector,

Representing the square of the two norms of the classification feature vector before correction, V _order Is an ordered vector in which the feature values of the classification feature vectors before correction are arranged in order of magnitude, and the classification feature vector V before correction is in the form of a column vector.

9. The method of claim 8, wherein said passing the optimized classification feature vector through the classifier to obtain the classification result comprises:

performing full-connection coding on the optimized classification feature vector by using a plurality of full-connection layers of the classifier to obtain a coding classification feature vector; and

and the coding classification feature vector is passed through a Softmax classification function of the classifier to obtain the classification result.

10. A separation system for oilfield associated gas, comprising:

the data acquisition module is used for acquiring the working power of the ultrasonic atomizer at a plurality of preset time points in a preset time period and the atomization state monitoring video of the preset time period;

the key frame extraction module is used for extracting the atomization state monitoring key frames of the plurality of preset time points from the atomization state monitoring video;

the feature extraction module is used for enabling the atomization state monitoring key frames at a plurality of preset time points to obtain atomization time sequence state feature vectors through a first convolution neural network model using a time attention mechanism;

The time sequence coding module is used for arranging the working powers of the ultrasonic atomizers at a plurality of preset time points into working power input vectors according to time dimension and then obtaining working power feature vectors through the time sequence coder;

the responsiveness estimation calculation module is used for calculating responsiveness estimation of the atomization time sequence state characteristic vector relative to the working power characteristic vector based on a Gaussian density chart so as to obtain a classification characteristic matrix; and

and the working power control module is used for passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the working power of the ultrasonic atomizer at the current time point should be increased or decreased.

Images