Method for purifying high-purity helium from lean-helium natural gas and co-producing LNG

Abstract

The application discloses a method for purifying high-purity helium from helium-lean natural gas and co-producing LNG, which comprises the following steps: firstly, pretreating the helium-lean natural gas to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas; then, preliminarily screening the helium-lean natural gas by using the pressure of the raw gas natural gas through a membrane separation technology; further concentrating helium in the preliminarily screened gas by cryogenic rectification to obtain helium-rich gas; meanwhile, fully utilizing the cold energy in the cryogenic liquefaction process to co-produce high-value-added LNG products, sharing the raw gas consumption and energy consumption in the helium cryogenic rectification concentration process; finally, purifying the helium-rich gas to obtain high-purity helium through a combined process of membrane separation, catalytic oxidation, chemical adsorption and physical adsorption, so as to realize low-cost extraction of high-quality helium products from the helium-lean natural gas.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of raw material processing, and more particularly, to a method for purifying high-purity helium from helium-lean natural gas and co-producing LNG. TECHNICAL BACKGROUND

[0002] Helium has unique properties such as low boiling point, good thermal conductivity, good permeability, stable properties, and non-solidification at absolute zero, and is therefore widely used in aerospace, military, submarines, cutting-edge scientific research, high-end manufacturing, medical treatment, and other fields, and is an indispensable strategic material for the development of national defense, military industry, and high-tech industries.

[0003] Currently, global helium is mainly extracted from helium-containing natural gas. According to the amount of helium in natural gas, helium-containing natural gas is divided into helium-lean natural gas (He≤500ppm), helium-rich natural gas (500≤He≤2000ppm), and extremely helium-rich natural gas (He≥3000ppm). Of the global proven helium reserves, the United States, Qatar, Algeria, and Russia account for more than 85%. China is extremely short of helium resources, and most of the helium-containing natural gas is helium-lean natural gas. The cost of extracting helium from natural gas is high, and it is not economically viable. Therefore, more than 95% of the helium in China is currently dependent on imports from abroad.

[0004] Some existing preparation schemes for high-purity helium, for example, first treating helium-rich natural gas by BOG membrane separation, and then using catalytic dehydrogenation and adsorption purification to prepare high-purity helium. However, this preparation scheme is suitable for helium-rich natural gas, and for helium-lean natural gas, the energy consumption is high, the operating cost is high, and the economic efficiency is poor. In addition, using oxygen-catalyzed dehydrogenation, the residual hydrogen and oxygen after the reaction is high, and without deep purification, the subsequent adsorption purification cannot remove the trace amounts of residual hydrogen, oxygen, and other impurities, the purity of the helium product is low, and it cannot meet the standard of high-purity helium.

[0005] Therefore, an optimized preparation scheme for high-purity helium is expected. SUMMARY

[0006] To solve the above technical problems, the present application is proposed. Embodiments of the present application provide a method for purifying high-purity helium gas from helium-lean natural gas and co-producing LNG, which first pretreats the helium-lean natural gas to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas; then uses the pressure of the raw gas natural gas to realize preliminary screening of the helium-lean natural gas through membrane separation technology; further concentrates helium in the preliminarily screened gas to obtain helium-rich gas through cryogenic rectification technology, while fully utilizing the cold energy in the cryogenic liquefaction process to co-produce high-value-added LNG products, sharing the raw gas consumption and energy consumption of the helium cryogenic rectification concentration process, and finally purifying the helium-rich gas to obtain high-purity helium gas through a combination process of membrane separation, catalytic oxidation, chemical adsorption, and physical adsorption, thereby realizing low-cost extraction of high-quality helium gas products from helium-lean natural gas. In particular, the present application uses reaction temperature and oxygen content control to improve impurity removal efficiency.

[0007] According to one aspect of the present application, a method for purifying high-purity helium gas from helium-lean natural gas and co-producing LNG is provided, which comprises:

[0008] Pretreating the helium-lean natural gas to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas to obtain pretreated gas;

[0009] Preliminarily screening the pretreated gas using a membrane separation unit to obtain high-pressure non-permeable gas as product gas for export and low-pressure permeable gas;

[0010] After the low-pressure permeable gas is pressurized using a permeable gas compressor, the pressurized gas is subjected to MDEA decarburization treatment and dehydration and demercuration treatment to obtain purified gas;

[0011] Cryogenic liquefaction and rectification treatment of the purified gas to obtain LNG products and helium-rich gas;

[0012] Concentrating the helium-rich gas using a two-stage membrane assembly to obtain crude helium gas with a concentration greater than or equal to 90%; and

[0013] Directionally removing impurities and adsorbing and purifying the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999.

[0014] In the method for purifying high-purity helium from helium-lean natural gas and co-producing LNG, the targeted impurity removal and adsorption purification of the crude helium to obtain high-purity helium with a purity of greater than or equal to 99.999% include: obtaining the heated temperature of the crude helium, the oxygen content value at multiple predetermined time points within a predetermined time period, and the gas chromatogram of the post-reaction product; arranging the heated temperature of the crude helium and the oxygen content value at the multiple predetermined time points into a temperature input vector and an oxygen content input vector, respectively, and calculating the product between the transpose vector of the temperature input vector and the oxygen content input vector to obtain a synergistic input matrix; passing the synergistic input matrix through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector; passing the gas chromatogram of the post-reaction product at multiple predetermined time points within the predetermined time period through a second convolutional neural network model using a time attention mechanism to obtain a product feature map; performing global mean pooling on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector; performing channel-recursive squeezing-activation optimization on the product feature vector to obtain an optimized product feature vector; calculating the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector to obtain a classification feature matrix; and passing the classification feature matrix through a classifier to obtain a classification result, which is used to indicate whether the power of the heater at the current time point should be increased or decreased.

[0015] In the method for purifying high-purity helium from helium-lean natural gas and co-producing LNG, the passing of the synergistic input matrix through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector includes: using each layer of the first convolutional neural network as a filter to perform convolution processing, mean pooling processing based on a feature matrix, and nonlinear activation processing on input data in the forward transmission of the layer to output the synergistic reaction feature vector from the last layer of the first convolutional neural network as a filter, wherein the input of the first layer of the first convolutional neural network as a filter is the synergistic input matrix.

[0016] In the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas, the gas chromatogram of the post-reaction product at each of the plurality of predetermined time points within the predetermined time period is obtained by using a second convolutional neural network model with a time attention mechanism, comprising: extracting a first gas chromatogram and a second gas chromatogram of adjacent time points from the gas chromatogram of the post-reaction product at each of the plurality of predetermined time points within the predetermined time period; passing the first gas chromatogram and the second gas chromatogram through a first convolutional layer and a second convolutional layer of the second convolutional neural network model, respectively, to obtain a first gas chromatogram feature map corresponding to the first gas chromatogram and a second gas chromatogram feature map corresponding to the second gas chromatogram; multiplying the first gas chromatogram feature map and the second gas chromatogram feature map by position points and then passing through a Softmax activation function to obtain a time attention map; passing the first gas chromatogram through a third convolutional layer of the second 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 product feature map.

[0017] In the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas, the channel recurrent squeeze-and-excitation optimization of the product feature vector to obtain an optimized product feature vector comprises: performing channel recurrent squeeze-and-excitation optimization of the product feature vector to obtain an optimized product feature vector according to the following formula:

[0018]

[0019] wherein v i is the eigenvalue of the product feature vector V, μ and σ are the mean and variance of the eigenvalue set v i ∈V, ReLU(·) represents the ReLU activation function, exp(·) represents the exponential operation of the negative variance, and the exponential operation of the negative variance represents the calculation of the natural exponential function value with the negative variance as the power.

[0020] In the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas, the calculation of the responsiveness estimate of the synergistic reaction feature vector with respect to the optimized product feature vector to obtain a classification feature matrix comprises: calculating the responsiveness estimate of the synergistic reaction feature vector with respect to the optimized product feature vector to obtain a classification feature matrix according to the following formula:

[0021]

[0022] wherein V1 represents the synergistic reaction feature vector, V2 represents the optimized product feature vector, M represents the classification feature matrix, represents the multiplication of a matrix and a vector.

[0023] In the method for purifying high-purity helium gas from helium-lean natural gas and co-producing LNG, the step of obtaining a classification result by using a classifier on the classification feature matrix comprises: expanding the classification feature matrix into a classification feature vector according to a row vector or a column vector; performing full connection coding on the classification feature vector by using a full connection layer of the classifier to obtain an encoded classification feature vector; and inputting the encoded classification feature vector into a Softmax classification function of the classifier to obtain the classification result.

[0024] According to another aspect of the present application, a system for purifying high-purity helium gas from helium-lean natural gas and co-producing LNG is provided, comprising:

[0025] a pretreatment module configured to pretreat the helium-lean natural gas to remove sulfides and liquid hydrocarbons in the helium-lean natural gas to obtain pretreated gas;

[0026] a screening module configured to use a membrane separation unit to preliminarily screen the pretreated gas to obtain high-pressure non-permeated gas as product gas for export and low-pressure permeated gas;

[0027] a purified gas obtaining module configured to use a permeated gas compressor to pressurize the low-pressure permeated gas, and perform MDEA decarburization treatment and dehydration and demercuration treatment on the pressurized gas to obtain purified gas;

[0028] a cryogenic rectification module configured to perform cryogenic liquefaction and rectification treatment on the purified gas to obtain LNG product and helium-rich gas;

[0029] a concentration module configured to use a secondary membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and

[0030] a directional impurity removal and adsorption purification module configured to perform directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999%.

[0031] In the system for purifying high-purity helium gas from lean-helium natural gas and co-producing LNG, the directional impurity removal and adsorption purification module comprises: a data acquisition unit configured to acquire the temperature of the crude helium gas after heating, the oxygen content value, and the gas chromatogram of the product after reaction at a plurality of predetermined time points within a predetermined time period; a synergy unit configured to arrange the temperature of the crude helium gas after heating and the oxygen content value at the plurality of predetermined time points into a temperature input vector and an oxygen content input vector respectively, and calculate the product between the transposed vector of the temperature input vector and the oxygen content input vector to obtain a synergy input matrix; a synergy reaction feature vector generation unit configured to pass the synergy input matrix through a first convolutional neural network as a filter to obtain a synergy reaction feature vector; a product feature map extraction unit configured to pass the gas chromatogram of the product after reaction at a plurality of predetermined time points within a predetermined time period through a second convolutional neural network model using a time attention mechanism to obtain a product feature map; a global mean pooling unit configured to perform global mean pooling on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector; an optimization unit configured to perform channel-recursive squeezing-boosting optimization on the product feature vector to obtain an optimized product feature vector; a responsiveness estimation unit configured to calculate the responsiveness estimation of the synergy reaction feature vector with respect to the optimized product feature vector to obtain a classification feature matrix; and a power adjustment result generation unit configured to pass the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the power of the heater at the current time point should be increased or decreased.

[0032] Compared with the prior art, the method for purifying high-purity helium gas from lean-helium natural gas and co-producing LNG provided by the present application first pretreats the lean-helium natural gas to filter out sulfides and liquid hydrocarbons in the lean-helium natural gas; then, by using the membrane separation technology, the pressure of the raw material gas natural gas can be used to realize the preliminary screening of the lean-helium natural gas; then, by using the cryogenic rectification technology, the helium gas is further concentrated to obtain the rich-helium gas, and at the same time, the cold energy in the cryogenic liquefaction process is fully utilized to co-produce the LNG product with high added value, so as to share the raw gas consumption and energy consumption in the helium cryogenic rectification concentration process, and finally, by using the combined process of membrane separation, catalytic oxidation, chemical adsorption, and physical adsorption, the high-purity helium gas is purified from the rich-helium gas, so as to realize the low-cost extraction of the high-quality helium gas product from the lean-helium natural gas. In particular, the present application uses the control of the reaction temperature and the oxygen content to improve the impurity removal efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0033] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description thereof taken in conjunction with the accompanying drawings, in which like reference characters designate like elements in the several views. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and are incorporated into and constitute a part of this specification, illustrate embodiments of the present application and serve to explain the present application, and are not intended to limit the present application in any manner. In the drawings:

[0034] Figure 1 FIG. 1 illustrates a flow diagram of a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0035] Figure 2 FIG. 2 illustrates a membrane separation process flow diagram in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0036] Figure 3 FIG. 3 illustrates a cryogenic rectification process flow diagram in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0037] Figure 4 FIG. 4 illustrates a membrane concentration process flow diagram in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0038] Figure 5 FIG. 5 illustrates a targeted impurity removal process flow diagram in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0039] Figure 6 FIG. 6 illustrates an adsorption purification process flow diagram in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0040] Figure 7 FIG. 7 illustrates an application scenario diagram of targeted impurity removal and adsorption purification of the crude helium gas to obtain high purity helium gas with a purity greater than or equal to 99.999% in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0041] Figure 8 FIG. 8 illustrates an architectural diagram of targeted impurity removal and adsorption purification of the crude helium gas to obtain high purity helium gas with a purity greater than or equal to 99.999% in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0042] Figure 9 FIG. 9 illustrates a flow diagram of targeted impurity removal and adsorption purification of the crude helium gas to obtain high purity helium gas with a purity greater than or equal to 99.999% in a process for purifying high purity helium gas and co-producing LNG from lean helium natural gas according to embodiments of the present application.

[0043] Figure 10 FIG. 10 is a flow chart illustrating a process of obtaining a product feature map by using a second convolutional neural network model with a time attention mechanism from a gas chromatogram of a post-reaction product at a plurality of predetermined time points within a predetermined time period in a method of purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application.

[0044] Figure 11 FIG. 11 is a flow chart illustrating a process of obtaining a classification result by using a classifier from a classification feature matrix in a method of purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application.

[0045] Figure 12 FIG. 12 is a block diagram illustrating a system of purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application.

[0046] Figure 13 FIG. 13 is a block diagram illustrating a directional impurity removal and adsorption purification module in a system of purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application. DETAILED DESCRIPTION

[0047] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application, and the present application can be implemented in many different ways. Therefore, the attached drawings should not be used to limit and define the present application, and the present application should cover all changes falling within the scope of the present application through the equivalent replacement and modification of technical features.

[0048] Summary of the application

[0049] As described above, there are some existing preparation schemes for high-purity helium gas, for example, a helium-rich natural gas is first treated by BOG membrane separation, and then catalytic dehydrogenation and adsorption purification are used to prepare high-purity helium gas, but this preparation scheme is suitable for helium-rich natural gas, and for helium-lean natural gas, the energy consumption is high, the operation cost is high, and the economy is poor. Moreover, oxygen is added for catalytic dehydrogenation, and the residual hydrogen and oxygen after the reaction is high, without deep purification, the subsequent adsorption purification cannot remove the trace amounts of residual hydrogen, oxygen and other impurities, the purity of helium gas product is low, and the high-purity helium gas standard cannot be reached. Therefore, an optimized preparation scheme for high-purity helium gas is expected.

[0050] To solve the above technical problems, the application provides a solution for extracting high-purity helium from helium-poor natural gas. First, the helium-poor natural gas is pretreated to filter out sulfides and liquid hydrocarbons in the helium-poor natural gas to obtain pretreated gas. Then, the preliminary screening of the helium-poor natural gas is realized by using the pressure of the raw gas natural gas through membrane separation technology. Further helium concentration of the preliminarily screened gas is realized to obtain helium-rich gas through cryogenic rectification technology. At the same time, the cold energy in the cryogenic liquefaction process is fully utilized to co-produce LNG products with high added value. The raw gas consumption and energy consumption in the helium cryogenic rectification concentration process are shared. Finally, high-purity helium is obtained by purifying the helium-rich gas through a combination process of membrane separation, catalytic oxidation, chemical adsorption, and physical adsorption, so as to realize the low-cost extraction of high-quality helium products from helium-poor natural gas.

[0051] More specifically, the application provides a method for purifying high-purity helium from helium-poor natural gas and co-producing LNG, which comprises the following steps: S110: pretreating the helium-poor natural gas to filter out sulfides and liquid hydrocarbons in the helium-poor natural gas to obtain pretreated gas; S120: using a membrane separation unit to preliminarily screen the pretreated gas to obtain high-pressure non-permeable gas as product gas for external delivery and low-pressure permeable gas; S130: using a permeable gas compressor to pressurize the low-pressure permeable gas, and then performing MDEA decarburization treatment and dehydration and demercuration treatment on the pressurized gas to obtain purified gas; S140: performing cryogenic liquefied rectification treatment on the purified gas to obtain LNG products and helium-rich gas; S150: using a secondary membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and S160: performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999.

[0052] In particular, in step S160, the crude helium gas is first added with an oxidizing agent in a certain proportion, heated to 150-200℃ by a heater, and then enters a catalytic dehydrogenation reactor for catalytic oxidation reaction under the action of a double-mechanism noble metal catalyst filled in the reactor. The addition amount of the oxidizing agent is controlled by the oxygen content at the outlet of the catalytic dehydrogenation reactor (the outlet oxygen content is ≤50 ppm). After the reaction, the crude helium gas is cooled and separated by a cooler and a separator, and then enters a chemical adsorption tower for dehydrogenation and deoxidation. The hydrogen + oxygen content is removed to ≤1 ppm(V) by a nickel-based chemical adsorbent filled in the tower. The water content in the crude helium gas is removed to ≤1 ppm(V) by a molecular sieve adsorber. It can be understood that the control of the reaction temperature and the oxygen content can improve the efficiency of impurity removal.

[0053] In the process of improving the impurity removal efficiency by controlling the reaction temperature and oxygen content, on the one hand, different reaction stages have different requirements for the optimal reaction temperature, and on the other hand, the reaction temperature, the reactants and the oxygen content need to be adapted to optimize the impurity removal efficiency. The above requirements make it difficult for traditional control algorithms for chemical reactions to adapt.

[0054] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have shown a level close to or even beyond human level in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks provides new solutions and schemes for the above parameter control.

[0055] Specifically, in the technical solution of the present application, the crude helium gas heating temperature, oxygen content value and gas chromatogram of the reaction product at a plurality of predetermined time points in a predetermined time period are first obtained. Then, the crude helium gas heating temperature and oxygen content value at the plurality of predetermined time points are arranged into a temperature input vector and an oxygen content input vector respectively, and the product between the transpose vector of the temperature input vector and the oxygen content input vector is calculated to obtain a collaborative input matrix. That is, the correlation between the reaction temperature and the oxygen content in the time dimension is represented at the data level. It should be understood that the value at each position in the collaborative input matrix is the correlation information between the reaction temperature and the oxygen content at the corresponding two time points.

[0056] Then, the collaborative input matrix is passed through a first convolutional neural network as a filter to obtain a collaborative reaction feature vector. That is, a convolutional neural network model with excellent performance in local feature extraction is used as a feature extractor to capture high-dimensional local implicit correlation information in the collaborative input matrix, i.e., the high-dimensional implicit feature representation of the correlation between the reaction temperature and the oxygen content in different time windows, to obtain the collaborative reaction feature vector.

[0057] In the technical solution of the present application, the synergistic effect of the reaction temperature and the oxygen content can be represented by the gas chromatogram of the reaction product at a plurality of predetermined time points in a predetermined time period. From the perspective of chemical reaction, the collaborative reaction feature vector is a high-dimensional feature representation of the reaction condition, and the feature representation of the gas chromatogram of the reaction product at the plurality of predetermined time points is a high-dimensional feature representation of the reaction result.

[0058] Specifically, in the technical solution of the present application, the gas chromatogram of the reaction product at a plurality of predetermined time points in the predetermined time period is encoded by using a second convolutional neural network model with a time attention mechanism to obtain a product feature map. Those skilled in the art should know that the gas chromatogram of the reaction product can represent the component type and composition of the reaction product. Accordingly, the gas chromatogram is essentially an image data, so in the technical solution of the present application, a convolutional neural network model is also used as a feature extractor to capture high-dimensional local features of the reaction product. In particular, in order to make the convolutional neural network model pay more attention to the change characteristics of the reaction product in the time dimension when extracting features, the time attention mechanism is integrated into the convolutional neural network model.

[0059] Then, the response estimation of the synergistic reaction feature vector with respect to the product feature map can be calculated to obtain a feature representation for representing the influence of the reaction condition on the reaction result. However, considering that the synergistic reaction feature vector is a one-dimensional feature vector and the product feature map is a three-dimensional feature tensor, they are not aligned in the feature dimension, so before the response estimation calculation is performed, the dimension is unified.

[0060] Specifically, in the technical solution of the present application, first, the global mean pooling is performed on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector; here, when the global mean pooling is performed on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector, since the global pooling of each feature matrix along the channel dimension of the product feature map reduces the distribution correlation between the feature values of the product feature vector, in order to improve the expression consistency of the feature values of each position of the product feature vector with respect to the overall feature distribution of the product feature map, the product feature vector is subjected to channel recursive squeezing-activation optimization, specifically:

[0061]

[0062] μ and σ are the mean and variance of the feature set v i ∈ V, where v i is the feature value of the product feature vector V.

[0063] That is, based on the statistical characteristics of the feature set of the product feature vector along the channel dimension of the product feature map, the channel recursion of the feature distribution of the product feature vector is activated, so as to infer the channel dimension distribution of the feature value of each position of the product feature vector at each channel sampling position of the product feature map, and by using the squeezing-activation mechanism composed of the ReLU-Sigmoid function, the confidence value of attention enhancement in the channel direction of the product feature map is obtained to enhance the distribution correlation of the product feature vector in its distribution direction with the channel direction of the product feature map, so that the product feature vector with high expression consistency relative to the overall feature distribution of the product feature map is obtained. That is, the squeezing-activation optimization of the channel recursion of the product feature vector is performed to obtain an optimized product feature vector.

[0064] Further, the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector is calculated to obtain a classification feature matrix. For example, in one technical solution of the present application, the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector is represented by a transfer matrix of the synergistic reaction feature vector relative to the optimized product feature vector. Then, the classification feature matrix is passed through a classifier to obtain a classification result, which is used to represent whether the power of the heater at the current time point should be increased or decreased. In this way, an artificial intelligence technology based on deep learning and deep neural network is used to construct a parameter control scheme for the removal of impurities in crude helium gas, so that the reaction parameters can be adapted to the reaction requirements to improve the reaction efficiency and effect.

[0065] Based on this, the present application proposes a method for purifying high-purity helium gas from helium-poor natural gas and co-producing LNG, which comprises: pretreating the helium-poor natural gas to filter out sulfides and liquid hydrocarbons in the helium-poor natural gas to obtain pretreated gas; using a membrane separation unit to preliminarily screen the pretreated gas to obtain high-pressure non-permeable gas as product gas for export and low-pressure permeable gas; using a permeable gas compressor to pressurize the low-pressure permeable gas, and then performing MDEA decarburization treatment and dehydration and demercuration treatment on the pressurized gas to obtain purified gas; performing cryogenic liquefaction and rectification treatment on the purified gas to obtain LNG product and helium-rich gas; using a two-stage membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999.

[0066] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be specifically introduced with reference to the accompanying drawings.

[0067] Exemplary method

[0068] Figure 1 Figure 1 illustrates a flow chart of a method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application. As shown in Figure 1, the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application comprises the following steps: S110, pre-treating the helium-lean natural gas to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas to obtain a pre-treated gas; S120, using a membrane separation unit to preliminarily screen the pre-treated gas to obtain high-pressure non-permeated gas as product gas for export and low-pressure permeated gas; S130, using a permeated gas compressor to pressurize the low-pressure permeated gas, and then performing MDEA decarburization treatment and dehydration and demercuration treatment on the pressurized gas to obtain a purified gas; S140, performing cryogenic liquefaction and rectification treatment on the purified gas to obtain LNG product and helium-rich gas; S150, using a secondary membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and S160, performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999%. Figure 1

[0069] In step S110, the helium-lean natural gas is pre-treated to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas to obtain a pre-treated gas. As described above, some existing preparation schemes for high-purity helium gas, for example, first treating helium-rich natural gas by BOG membrane separation, and then using catalytic dehydrogenation and adsorption purification to prepare high-purity helium gas, but this preparation scheme is suitable for helium-rich natural gas, and has high energy consumption and high operating cost for helium-lean natural gas, and poor economic efficiency. Moreover, oxygen is added for catalytic dehydrogenation, and the residual hydrogen and oxygen after the reaction is high, and no deep purification is performed, so that the subsequent adsorption purification cannot remove the trace amounts of residual hydrogen, oxygen and other impurities, the purity of the helium product is low, and the high-purity helium gas standard cannot be met. In view of the above technical problems, the present application provides a solution for extracting high-purity helium gas from helium-lean natural gas, which first pre-treats the helium-lean natural gas to filter out sulfides and liquid hydrocarbons in the helium-lean natural gas to obtain a pre-treated gas. According to the component characteristics of the raw gas, the pre-treatment can use gas-liquid separation, desulfurization, dehydrocarbon and other processes to preliminarily purify the raw gas, so as to avoid damage to the subsequent membrane assembly caused by sulfides and liquid hydrocarbons.

[0070] In step S120, the membrane separation unit is used to preliminarily screen the pre-treated gas to obtain high-pressure non-permeated gas as product gas for export and low-pressure permeated gas. Then, the preliminary screening of the helium-lean natural gas can be realized by using the pressure of the raw gas through the membrane separation technology.

[0071] Figure 2 Figure 2 illustrates a flow chart of a membrane separation process in the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application. As shown in Figure 2, the membrane separation process in the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application comprises the following steps: S121, using a membrane separation unit to preliminarily screen the pre-treated gas to obtain high-pressure non-permeated gas as product gas for export and low-pressure permeated gas; S122, using a permeated gas compressor to pressurize the low-pressure permeated gas; S123, performing MDEA decarburization treatment on the pressurized gas; S124, performing dehydration and demercuration treatment on the pressurized gas; S125, performing cryogenic liquefaction and rectification treatment on the purified gas to obtain LNG product and helium-rich gas; S126, using a secondary membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and S127, performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999%. Figure 2 ​As shown, the pretreated feed gas first passes through a coalescing filter to remove any remaining free water and heavy hydrocarbons, is heated to ~50°C by a heater (to prevent water and hydrocarbons from accumulating and damaging the membrane module), enters a high-pressure hollow fiber membrane module, and most of the helium and part of the nitrogen and methane permeate through the membrane interface. After cooling and separation, the low-pressure permeate gas with a helium concentration increased by more than 10 times is obtained, and the high-pressure non-permeate gas that does not permeate through the membrane interface is used as product natural gas. The unit fully utilizes the pressure energy of the feed gas to preliminarily screen and concentrate the helium-poor natural gas, with a helium recovery rate of ≥90%, and the pressure loss of the natural gas (non-permeate gas) is very small, which does not affect the export of natural gas and does not need to increase the export pressurization equipment.

[0072] In step S130, the low-pressure permeate gas is pressurized using a permeate gas compressor, and the pressurized gas is subjected to MDEA decarburization treatment and dehydration and mercury removal treatment to obtain purified gas. In one specific embodiment, the low-pressure permeate gas is pressurized to 4.5-5.0 MPa.G by the permeate gas compressor. In the MDEA decarburization process unit, according to the CO2 content of the feed gas, a one-stage absorption and one-stage regeneration MDEA decarburization process is selected to remove the CO2 content in the gas to below 50 ppm(V) to meet the requirements of the low-temperature cold box. In the dehydration and mercury removal process unit, mature molecular sieve dehydration and sulfur-loaded activated carbon mercury removal processes are used to remove the water content in the gas to below 1 ppm(V) and the mercury content to below 10 ng / Nm3 to meet the requirements of the low-temperature cold box.

[0073] In step S140, the purified gas is subjected to cryogenic liquefaction and rectification treatment to obtain LNG products and helium-rich gas. Through cryogenic rectification technology, the preliminarily screened gas is further concentrated to obtain helium-rich gas, and the cold energy in the cryogenic liquefaction process is fully utilized to co-produce LNG products with high added value, and the feed gas consumption and energy consumption of the helium cryogenic rectification enrichment process are shared.

[0074] Figure 3 Fig. 1 illustrates a cryogenic rectification process flow diagram in a method for purifying high-purity helium from helium-poor natural gas and co-producing LNG according to an embodiment of the present application. As shown in Fig. 1, the process includes the following steps: Figure 3As shown, the purified gas after MDEA decarburization, dehydration and mercury removal enters the low-temperature cold box, is first cooled to -60°C by heat exchange with counterflow refrigerant through the upper-stage plate-fin heat exchanger, enters the reboiler at the bottom of the rectification tower to provide heat and continues to be cooled, then enters the lower-stage plate-fin heat exchanger to continue to be cooled to -110°C by heat exchange with counterflow refrigerant, enters the middle part of the rectification tower, and the liquid phase temperature at the tower bottom is controlled to be ~ 105.2 by flash distillation rectification, the nitrogen content is ≤1.0%, and after liquid level adjustment, the gas is returned to the lower-stage plate-fin heat exchanger to continue to be cooled to -160°C to obtain the LNG product; the gas phase at the tower top (rich helium gas) is adjusted to 2.0-2.2 MPa.G by pressure adjustment, then is sequentially cooled to room temperature by heat exchange with the lower-stage plate-fin heat exchanger and the upper-stage plate-fin heat exchanger to recover cold energy, and is then sent to the subsequent process unit. The cold energy of the cryogenic rectification unit is provided by mixed refrigerant circulation (MRC), and the refrigerant is composed of five components of nitrogen, methane, ethylene, propane and isopentane. The refrigerant is separated into gas-phase refrigerant and liquid-phase refrigerant after being pressurized and cooled by a refrigerant compressor. The gas-phase refrigerant is sequentially cooled to -160°C by heat exchange with counterflow refrigerant through the upper-stage plate-fin heat exchanger and the lower-stage plate-fin heat exchanger, is depressurized and cooled by a J-T valve, enters the condenser at the top of the rectification tower to provide cold energy for the condenser, is then returned to the lower-stage plate-fin heat exchanger to release cold energy, is combined with the liquid-phase refrigerant after being depressurized by throttling, is returned to the upper-stage plate-fin heat exchanger to continue to release cold energy and is warmed to room temperature, and is then returned to the inlet of the refrigerant compressor. The liquid-phase refrigerant is cooled to -60°C by heat exchange with counterflow refrigerant through the upper-stage plate-fin heat exchanger, is combined with the gas-phase refrigerant after being depressurized by a J-T valve, is returned to the upper-stage plate-fin heat exchanger to release cold energy and is warmed to room temperature, and is then returned to the inlet of the refrigerant compressor.

[0075] In step S150, the helium-rich gas is concentrated using a two-stage membrane assembly to obtain crude helium gas with a concentration greater than or equal to 90%.

[0076] Figure 4 A process flow diagram of the membrane concentration process in the method of purifying high-purity helium gas and co-producing LNG from helium-lean natural gas according to an embodiment of the present application is shown. Figure 4 As shown, the helium-rich gas is separated and concentrated using a two-stage medium-pressure membrane assembly, and the helium in the non-permeated gas is fully recovered, with a helium recovery rate ≥95% and a helium content ≥90%. The helium-rich gas first enters the first-stage one-segment membrane assembly to complete the preliminary concentration of helium, is pressurized to ~2.2 MPa by a stage compressor, and then enters the second-stage membrane assembly to further concentrate the helium, to obtain crude helium gas with a pressure of ~0.8 MPa.G and a helium purity ≥90%. The non-permeated gas of the first-stage one-segment membrane enters the first-stage two-segment membrane for separation and recovery of helium, and the non-permeated gas (rich in nitrogen) is discharged to a venting system. The low-pressure permeated gas rich in helium is pressurized to 2.0-2.2 MPa by a helium recovery compressor, is combined with the non-permeated gas of the second-stage membrane, and is returned to the inlet of the first-stage one-segment membrane for recovery of helium, to improve the helium recovery rate of the device.

[0077] In step S160, the crude helium gas is subjected to directional impurity removal and adsorption purification to obtain high-purity helium gas with a purity greater than or equal to 99.999%.

[0078] Figure 5 The illustration shows a flow chart of the directional impurity removal process in a method for purifying high-purity helium from lean helium natural gas and co-producing LNG according to an embodiment of this application. Figure 5 As shown, crude helium gas with a certain proportion of oxidant is heated to 150-200℃ by a heater, and then enters a catalytic dehydrogenation reactor. Under the action of a dual-mechanism noble metal catalyst packed inside the reactor, a catalytic oxidation reaction occurs. The amount of oxidant added is controlled by the oxygen content at the outlet of the catalytic dehydrogenation reactor (outlet oxygen content ≤50ppm). After the reaction, the crude helium gas is cooled and separated by a cooler and separator, and then enters a chemical adsorption tower for dehydrogenation and deoxygenation. The nickel-based chemical adsorbent packed inside the tower removes the hydrogen + oxygen content to ≤1ppm(V). Then, it passes through a molecular sieve adsorbent to remove the water content from the crude helium gas to ≤1ppm(V). The two chemical adsorption towers operate alternately. After the adsorbent in the tower is saturated, N2 + H2 is used as the regeneration gas to heat and regenerate the adsorbent at 300-350℃. After regeneration, the adsorbent bed is cooled with N2, and the adsorbent regains its adsorption capacity after cooling. Two molecular sieve adsorbers operate alternately. Once one adsorbent becomes saturated, the other adsorbent is switched on, and the adsorbent in the saturated adsorber is replaced.

[0079] Figure 6 The illustration shows a flow chart of the adsorption purification process in a method for purifying high-purity helium from lean helium natural gas and co-producing LNG according to an embodiment of this application. Figure 6 As shown, the purified crude helium gas enters the adsorption tower, where it passes through a molecular sieve adsorbent to remove impurities other than helium, resulting in high-purity helium with a purity ≥ 99.999%. The adsorption purification process employs pressure swing adsorption (PSA), with each adsorption tower operating alternately. A suitable PSA process flow and molecular sieve adsorbent combination are selected based on the impurity content of the crude helium gas and the processing scale. The desorbed gas from the adsorption purification is returned to the inlet of the interstage compressor in the membrane concentration unit for recycling, thereby improving the helium recovery rate.

[0080] In particular, in step S160, first, the crude helium gas with the oxidant is added in a certain proportion, heated to 150-200°C by the heater, and then enters the catalytic dehydrogenation reactor for catalytic oxidation reaction under the action of the dual-mechanism noble metal catalyst filled in the reactor. The addition amount of the oxidant is controlled by the oxygen content at the outlet of the catalytic dehydrogenation reactor (the outlet oxygen content is ≤50 ppm), after the reaction, the crude helium gas is cooled and separated by the cooler and the separator, and then enters the chemical adsorption tower for dehydrogenation and deoxidation. The hydrogen + oxygen content is removed to ≤1 ppm (V) by the nickel-based chemical adsorbent filled in the tower, and the water content in the crude helium gas is removed to ≤1 ppm (V) by the molecular sieve adsorber. It can be understood that the control of the reaction temperature and the oxygen content can improve the efficiency of impurity removal.

[0081] In the process of improving the impurity removal efficiency by controlling the reaction temperature and the oxygen content, on the one hand, different reaction stages have different requirements for the optimal reaction temperature, and on the other hand, the reaction temperature, the reactant and the oxygen content need to be adapted to optimize the impurity removal efficiency. The above requirements make it difficult for the traditional control algorithm for chemical reactions to adapt.

[0082] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have shown a level close to or even surpassing that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks provides new solutions and schemes for the above parameter control.

[0083] Figure 7 application embodiments. As shown in the figure, in this application scenario, in the process of improving the impurity removal efficiency by controlling the reaction temperature and the oxygen content, first, the heated temperature of the crude helium gas (for example, He as shown in Figure 7 application embodiments. As shown in the figure, in this application scenario, in the process of improving the impurity removal efficiency by controlling the reaction temperature and the oxygen content, first, the heated temperature of the crude helium gas (for example, He as shown in Figure 7 application embodiments. As shown in the figure, in this application scenario, in the process of improving the impurity removal efficiency by controlling the reaction temperature and the oxygen content, first, the heated temperature of the crude helium gas (for example, He as shown in Figure 7 application embodiments. As shown in the figure, in this application scenario, in the process of improving the impurity removal efficiency by controlling the reaction temperature and the oxygen content, first, the heated temperature of the crude helium gas (for example, He as shown in Figure 7In the illustrated S), the server can process the heated temperature, oxygen content value and gas chromatogram of the reaction product of the crude helium gas at the plurality of predetermined time points based on the algorithm for purifying high-purity helium gas from lean helium natural gas and co-producing LNG to obtain a classification result indicating that the power of the heater at the current time point should be increased or decreased.

[0084] Figure 8 The architecture diagram of the directional impurity removal and adsorption purification of the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999 in the method of purifying high-purity helium gas from lean helium natural gas and co-producing LNG according to the embodiments of the present application is illustrated. As shown in Figure 8 As shown, the heated temperature, oxygen content value and gas chromatogram of the reaction product of the crude helium gas at a plurality of predetermined time points in a predetermined time period are first obtained. Then, the heated temperature and oxygen content value of the crude helium gas at the plurality of predetermined time points are arranged into a temperature input vector and an oxygen content input vector, respectively, and the product of the transpose vector of the temperature input vector and the oxygen content input vector is calculated to obtain a collaborative input matrix. Then, the collaborative input matrix is passed through a first convolutional neural network as a filter to obtain a collaborative reaction feature vector. Further, the gas chromatogram of the reaction product of the crude helium gas at a plurality of predetermined time points in the predetermined time period is passed through a second convolutional neural network model using a time attention mechanism to obtain a product feature map. Then, each feature matrix of the product feature map along the channel dimension is globally mean-pooled to obtain a product feature vector. Then, the product feature vector is channel-recurrently squeezed and excited to obtain an optimized product feature vector. Further, the responsiveness of the collaborative reaction feature vector to the optimized product feature vector is estimated to obtain a classification feature matrix. Then, the classification feature matrix is passed through a classifier to obtain a classification result indicating that the power of the heater at the current time point should be increased or decreased.

[0085] Figure 9 The flowchart of the directional impurity removal and adsorption purification of the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999 in the method of purifying high-purity helium gas from lean helium natural gas and co-producing LNG according to the embodiments of the present application is illustrated. As shown in Figure 9As shown, in the method for purifying high-purity helium gas from the lean helium natural gas and co-producing LNG, the directional impurity removal and adsorption purification of the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999% includes: S210, obtaining the crude helium gas heating temperature, oxygen content value and gas chromatogram of the reaction product at multiple predetermined time points in a predetermined time period; S220, arranging the crude helium gas heating temperature and oxygen content value at the multiple predetermined time points into a temperature input vector and an oxygen content input vector, respectively, and calculating the product between the transpose vector of the temperature input vector and the oxygen content input vector to obtain a synergistic input matrix; S230, passing the synergistic input matrix through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector; S240, passing the gas chromatogram of the reaction product at multiple predetermined time points in the predetermined time period through a second convolutional neural network model using a time attention mechanism to obtain a product feature map; S250, performing global mean pooling on each feature matrix of the product feature map along the channel dimension to obtain a product feature vector; S260, performing channel recursive squeezing-excitation optimization on the product feature vector to obtain an optimized product feature vector; S270, calculating the responsiveness estimation of the synergistic reaction feature vector relative to the optimized product feature vector to obtain a classification feature matrix; and S280, passing the classification feature matrix through a classifier to obtain a classification result, which is used to indicate whether the power of the heater at the current time point should be increased or decreased.

[0086] In step S210, the crude helium gas heating temperature, oxygen content value and gas chromatogram of the reaction product at multiple predetermined time points in a predetermined time period are obtained. Specifically, in the technical solution of the present application, the crude helium gas heating temperature, oxygen content value and gas chromatogram of the reaction product at multiple predetermined time points in a predetermined time period are first obtained.

[0087] In step S220, the crude helium gas heating temperature and oxygen content value at the multiple predetermined time points are arranged into a temperature input vector and an oxygen content input vector, respectively, and the product between the transpose vector of the temperature input vector and the oxygen content input vector is calculated to obtain a synergistic input matrix. That is, the correlation between the reaction temperature and the oxygen content in the time dimension is represented at the data level. It should be understood that the value at each position in the synergistic input matrix is the correlation information between the reaction temperature and the oxygen content at the corresponding two time points.

[0088] In step S230, the synergistic input matrix is passed through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector. That is, a convolutional neural network model with excellent performance in local feature extraction is used as a feature extractor to capture high-dimensional local implicit correlation information in the synergistic input matrix, i.e., high-dimensional implicit feature representation of the correlation between reaction temperature and oxygen content in different time windows, to obtain the synergistic reaction feature vector.

[0089] Specifically, in one example, in the method of purifying high-purity helium gas from lean helium natural gas and co-producing LNG described above, the passing of the synergistic input matrix through the first convolutional neural network as a filter to obtain a synergistic reaction feature vector comprises: using each layer of the first convolutional neural network as a filter to perform convolution processing, mean pooling processing based on a feature matrix, and nonlinear activation processing on input data in the forward transmission of the layer, respectively, to output the synergistic reaction feature vector from the last layer of the first convolutional neural network as a filter, wherein the input of the first layer of the first convolutional neural network as a filter is the synergistic input matrix.

[0090] In step S240, the gas chromatogram of the reaction product at a plurality of predetermined time points in the predetermined time period is passed through a second convolutional neural network model using a time attention mechanism to obtain a product feature map. In the technical solution of the present application, the synergistic effect of reaction temperature and oxygen content can be represented by the gas chromatogram of the reaction product at a plurality of predetermined time points in the predetermined time period. From the perspective of chemical reaction, the synergistic reaction feature vector is a high-dimensional feature representation of the reaction condition, and the feature representation of the gas chromatogram of the reaction product at a plurality of predetermined time points is a high-dimensional feature representation of the reaction result.

[0091] Specifically, in the technical solution of the present application, the gas chromatogram of the reaction product at a plurality of predetermined time points in the predetermined time period is encoded using a second convolutional neural network model using a time attention mechanism to obtain a product feature map. As known by those skilled in the art, the gas chromatogram of the reaction product can represent the component type and composition of the reaction product. Accordingly, the gas chromatogram is essentially an image data, and therefore, in the technical solution of the present application, a convolutional neural network model is also used as a feature extractor to capture high-dimensional local features of the reaction product. In particular, in order to enable the convolutional neural network model to pay more attention to the change characteristics of the reaction product in the time dimension when extracting features, a time attention mechanism is integrated into the convolutional neural network model.

[0092] Figure 10Fig. 3 illustrates a flowchart of a method for obtaining a product feature map of reaction products at multiple predetermined time points within a predetermined time period in a method for purifying high-purity helium and co-producing LNG from helium-lean natural gas according to an embodiment of the present application by using a second convolutional neural network model with a time attention mechanism. As shown in Fig. 3, specifically, in the above-mentioned method for purifying high-purity helium and co-producing LNG from helium-lean natural gas, the step of obtaining a product feature map of reaction products at multiple predetermined time points within a predetermined time period by using a second convolutional neural network model with a time attention mechanism includes: S310, extracting a first gas chromatogram and a second gas chromatogram of adjacent time points from the gas chromatograms of reaction products at multiple predetermined time points within the predetermined time period; S320, passing the first gas chromatogram and the second gas chromatogram through a first convolutional layer and a second convolutional layer of the second convolutional neural network model, respectively, to obtain a first gas chromatogram feature map corresponding to the first gas chromatogram and a second gas chromatogram feature map corresponding to the second gas chromatogram; S330, multiplying the first gas chromatogram feature map and the second gas chromatogram feature map by position points and then passing the result through a Softmax activation function to obtain a time attention map; S340, passing the first gas chromatogram through a third convolutional layer of the second convolutional neural network model to obtain a local feature map; and S350, multiplying the local feature map and the time attention map by position points to obtain the product feature map. Figure 10

[0093] In step S250, each feature matrix along the channel dimension of the product feature map is globally averaged and pooled to obtain a product feature vector. Considering that the collaborative reaction feature vector is a one-dimensional feature vector and the product feature map is a three-dimensional feature tensor, both of which are not aligned in the feature dimension, therefore, before performing the response estimation calculation, the dimensions are first unified. Specifically, in the technical solution of the present application, first, each feature matrix along the channel dimension of the product feature map is globally averaged and pooled to obtain a product feature vector; here, when each feature matrix along the channel dimension of the product feature map is globally averaged and pooled to obtain a product feature vector, since the global pooling of each feature matrix along the channel dimension of the product feature map reduces the distribution correlation between the feature values of the product feature vector, in order to improve the expression consistency of the feature values of each position of the product feature vector with respect to the overall feature distribution of the product feature map, the product feature vector is subjected to channel recursive squeezing-activation optimization.

[0094] ​In step S260, channel recursive squeeze-activation optimization is performed on the product feature vector to obtain an optimized product feature vector. That is, the channel recursion of the feature distribution of the product feature vector is activated based on the statistical characteristics of the feature set of the product feature vector along the channel dimension of the product feature map, so as to infer the channel dimension distribution of the feature value of each position of the product feature vector at each channel sampling position of the product feature map, and by using the squeeze-activation mechanism composed of the ReLU-Sigmoid function, the confidence value of attention enhancement in the channel direction of the product feature map is obtained to enhance the distribution correlation of the product feature vector in its distribution direction with the channel direction of the product feature map, so that the product feature vector with high expression consistency relative to the overall feature distribution of the product feature map is obtained. That is, the channel recursive squeeze-activation optimization is performed on the product feature vector to obtain an optimized product feature vector.

[0095] Specifically, in one example, in the method for purifying high-purity helium gas and co-producing LNG from helium-lean natural gas described above, the channel recursive squeeze-activation optimization performed on the product feature vector to obtain an optimized product feature vector includes: performing channel recursive squeeze-activation optimization on the product feature vector to obtain an optimized product feature vector according to the following formula; wherein the formula is:

[0096]

[0097] where v i is the feature value of the product feature vector V, μ and σ are the mean and variance of the feature set v i ∈V, ReLU(·) represents the ReLU activation function, exp(·) represents the exponential operation of the negative number of the variance, and the exponential operation of the negative number of the variance represents the calculation of the natural exponential function value with the negative number of the variance as the power.

[0098] In step S270, the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector is calculated to obtain a classification feature matrix. Further, the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector is calculated to obtain a classification feature matrix. For example, in one technical solution of the present application, the responsiveness estimate of the synergistic reaction feature vector relative to the optimized product feature vector is represented by the transition matrix of the synergistic reaction feature vector relative to the optimized product feature vector.

[0099] Specifically, in one example, in the method for purifying high-purity helium gas and co-producing LNG from lean-helium natural gas, the calculating the responsiveness estimate of the synergistic reaction feature vector to the optimized product feature vector to obtain a classification feature matrix comprises: calculating the responsiveness estimate of the synergistic reaction feature vector to the optimized product feature vector to obtain a classification feature matrix according to the following formula: wherein V1 represents the synergistic reaction feature vector, V2 represents the optimized product feature vector, and M represents the classification feature matrix.

[0100]

[0101] wherein V1 represents the synergistic reaction feature vector, V2 represents the optimized product feature vector, and M represents the classification feature matrix, represents matrix and vector multiplication.

[0102] In step S280, the classification feature matrix is passed through a classifier to obtain a classification result, which is used to indicate whether the power of the heater at the current time point should be increased or decreased. In this way, an artificial intelligence technology based on deep learning and deep neural networks is used to construct a parameter control scheme for the removal of impurities in crude helium gas, so that the reaction parameters can be adapted to the reaction requirements to improve the reaction efficiency and effect.

[0103] Figure 11 Fig. 1 illustrates a flowchart of passing the classification feature matrix through a classifier to obtain a classification result in the method for purifying high-purity helium gas and co-producing LNG from lean-helium natural gas according to an embodiment of the present application. As shown in Figure 11 Specifically, in the method for purifying high-purity helium gas and co-producing LNG from lean-helium natural gas, the passing the classification feature matrix through a classifier to obtain a classification result comprises: S410, expanding the classification feature matrix into a classification feature vector according to a row vector or a column vector; S420, using a fully connected layer of the classifier to perform fully connected coding on the classification feature vector to obtain a coded classification feature vector; and S430, inputting the coded classification feature vector into a Softmax classification function of the classifier to obtain the classification result.

[0104] In summary, the method for purifying high-purity helium and co-producing LNG from lean helium natural gas, based on the embodiments of this application, is explained. First, the lean helium natural gas is pretreated to remove sulfides and liquid hydrocarbons. Next, membrane separation technology is used to achieve preliminary screening of the lean helium natural gas using the pressure energy of the feedstock natural gas. Then, cryogenic distillation technology is used to further concentrate the pre-screened gas to obtain helium-rich gas. Simultaneously, the cooling capacity of the cryogenic liquefaction process is fully utilized to co-produce high-value-added LNG products, thus mitigating the feedstock gas consumption and energy consumption of the cryogenic distillation enrichment process. Finally, a combination of processes including membrane separation, catalytic oxidation, chemical adsorption, and physical adsorption is used to purify the helium-rich gas to obtain high-purity helium, thereby achieving low-cost extraction of high-quality helium products from lean helium natural gas. In particular, this application utilizes control of reaction temperature and oxygen content to improve impurity removal efficiency.

[0105] Exemplary system

[0106] Figure 12 The diagram illustrates a block diagram of a system for purifying high-purity helium from lean helium natural gas and co-producing LNG according to an embodiment of this application. Figure 12 As shown, a system 100 for purifying high-purity helium and co-producing LNG from lean helium natural gas according to an embodiment of this application includes: a pretreatment module 110 for pretreating lean helium natural gas to filter out sulfides and liquid hydrocarbons to obtain pretreated gas; a screening module 120 for performing preliminary screening of the pretreated gas using a membrane separation unit to obtain high-pressure non-permeable gas for product gas export and low-pressure permeable gas; and a purified gas acquisition module 130 for using a permeable gas compressor to purify the low-pressure permeable gas. After compression, the pressurized gas undergoes MDEA decarbonization and dehydration / mercury removal treatment to obtain purified gas; a cryogenic distillation module 140 is used to cryogenically liquefy and distill the purified gas to obtain LNG products and helium-rich gas; a concentration module 150 is used to concentrate the helium-rich gas using a secondary membrane module to obtain crude helium gas with a concentration greater than or equal to 90%; and a directional impurity removal and adsorption purification module 160 is used to perform directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999%.

[0107] Figure 13 The diagram illustrates a block diagram of a directional impurity removal and adsorption purification module in a system for purifying high-purity helium from lean helium natural gas and co-producing LNG according to an embodiment of this application. Figure 13As shown, in the system 100 for purifying high-purity helium gas from lean-helium natural gas and co-producing LNG, the directional impurity removal and adsorption purification module 160 comprises: a data acquisition unit 161 configured to acquire the temperature of the heated crude helium gas, the oxygen content value, and the gas chromatogram of the product after reaction at a plurality of predetermined time points within a predetermined time period; a synergy unit 162 configured to arrange the temperature of the heated crude helium gas and the oxygen content value at the plurality of predetermined time points into a temperature input vector and an oxygen content input vector, respectively, and calculate the product between the transpose vector of the temperature input vector and the oxygen content input vector to obtain a synergy input matrix; a synergy reaction feature vector generation unit 163 configured to pass the synergy input matrix through a first convolutional neural network as a filter to obtain a synergy reaction feature vector; a product feature map extraction unit 164 configured to pass the gas chromatogram of the product after reaction at a plurality of predetermined time points within a predetermined time period through a second convolutional neural network model using a time attention mechanism to obtain a product feature map; a global mean pooling unit 165 configured to perform global mean pooling on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector; an optimization unit 166 configured to perform channel-recursive squeezing- excitation optimization on the product feature vector to obtain an optimized product feature vector; a responsiveness estimation unit 167 configured to calculate the responsiveness estimation of the synergy reaction feature vector with respect to the optimized product feature vector to obtain a classification feature matrix; and a power adjustment result generation unit 168 configured to pass the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used to indicate whether the power of the heater at the current time point should be increased or decreased.

[0108] Here, those skilled in the art can understand that the specific functions and operations of each unit and module in the system 100 for purifying high-purity helium gas from lean-helium natural gas and co-producing LNG have been described in detail above with reference to the method for purifying high-purity helium gas from lean-helium natural gas and co-producing LNG, and therefore, the repeated description thereof will be omitted. Figures 1 to 11

[0109] ​As described above, the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG according to the embodiments of the present application can be implemented in various terminal devices, such as a server for purifying high-purity helium from helium-lean natural gas and co-producing LNG, and the like. In one example, the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG according to the embodiments of the present application can be integrated into a terminal device as a software module and / or a hardware module. For example, the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG can be a software module in an operating system of the terminal device, or can be an application program developed for the terminal device; of course, the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG can also be one of many hardware modules of the terminal device.

[0110] Alternatively, in another example, the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG and the terminal device can also be separate devices, and the system 100 for purifying high-purity helium from helium-lean natural gas and co-producing LNG can be connected to the terminal device through a wired and / or wireless network, and transmit interactive information in an agreed data format.

Claims

1. A process for the purification of high purity helium from a helium-lean natural gas and co-production of LNG, characterized in that, The method comprises the following steps: preprocessing a helium-poor natural gas to filter out sulfides and liquid hydrocarbons in the helium-poor natural gas to obtain a pretreated gas; using a membrane separation unit to preliminarily screen the pretreated gas to obtain high-pressure non-permeated gas as product gas for export and low-pressure non-permeated gas; using a permeated gas compressor to increase the pressure of the low-pressure non-permeated gas, and then performing MDEA decarburization treatment and dehydration decarburization treatment on the gas after pressure increase to obtain purified gas; performing cryogenic rectification treatment on the purified gas to obtain LNG product and helium-rich gas; using a secondary membrane assembly to concentrate the helium-rich gas to obtain crude helium gas with a concentration greater than or equal to 90%; and performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999%; the step of performing directional impurity removal and adsorption purification on the crude helium gas to obtain high-purity helium gas with a purity greater than or equal to 99.999% comprises the following steps: arranging the heated temperature of the crude helium gas at the plurality of predetermined time points and the oxygen content value at the plurality of predetermined time points in a predetermined time period into a temperature input vector and an oxygen content input vector respectively, and calculating the product between the transpose vector of the temperature input vector and the oxygen content input vector to obtain a synergistic input matrix; passing the synergistic input matrix through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector; passing the gas chromatogram of the reaction product at the plurality of predetermined time points in the predetermined time period through a second convolutional neural network model using a time attention mechanism to obtain a product feature map; performing global mean pooling on each feature matrix along the channel dimension of the product feature map to obtain a product feature vector; performing channel recursive squeezing-activation optimization on the product feature vector to obtain an optimized product feature vector; calculating the responsiveness estimate of the synergistic reaction feature vector with respect to the optimized product feature vector to obtain a classification feature matrix; and passing the classification feature matrix through a classifier to obtain a classification result, which is used to indicate whether the power of the heater at the current time point should be increased or decreased.

2. The process for the purification of high purity helium and co-production of LNG from a lean helium natural gas according to claim 1, characterized in that, The step of passing the synergistic input matrix through a first convolutional neural network as a filter to obtain a synergistic reaction feature vector comprises the following steps: using each layer of the first convolutional neural network as a filter to perform convolution processing, mean pooling processing based on a feature matrix, and nonlinear activation processing on input data in the forward transmission of the layer to output the synergistic reaction feature vector from the last layer of the first convolutional neural network as a filter, wherein the input of the first layer of the first convolutional neural network as a filter is the synergistic input matrix.

3. The process for the purification of high purity helium and co-production of LNG from a lean helium natural gas according to claim 2, characterized in that, The gas chromatogram of the post-reaction product at each of the plurality of predetermined time points in the predetermined time period is obtained by using a second convolutional neural network model with a time attention mechanism, including: extracting a first gas chromatogram and a second gas chromatogram of adjacent time points from the gas chromatograms of the post-reaction product at the plurality of predetermined time points in the predetermined time period; passing the first gas chromatogram and the second gas chromatogram through a first convolutional layer and a second convolutional layer of the second convolutional neural network model, respectively, to obtain a first gas chromatogram feature map corresponding to the first gas chromatogram and a second gas chromatogram feature map corresponding to the second gas chromatogram; performing position point multiplication on the first gas chromatogram feature map and the second gas chromatogram feature map, and then passing the result through a Softmax activation function to obtain a time attention map; passing the first gas chromatogram through a third convolutional layer of the second convolutional neural network model to obtain a local feature map; and performing position point multiplication on the local feature map and the time attention map to obtain the product feature map.

4. The process for the purification of high purity helium and co-production of LNG from lean helium natural gas according to claim 3, characterized in that, The channel recursive squeeze-excitation optimization on the product feature vector to obtain an optimized product feature vector comprises: performing channel recursive squeeze-excitation optimization on the product feature vector to obtain an optimized product feature vector according to the following formula; wherein the formula is: wherein is an eigenvalue of the product feature vector , and are a mean and a variance of a feature set , denotes an activation function, denotes an exponential operation of the variance negative number, and the exponential operation of the variance negative number represents calculation of a natural exponential function value with the negative number of the variance as the power.

5. The process for the purification of high purity helium and co-production of LNG from a lean helium natural gas according to claim 4, characterized in that, The calculating the response estimation of the synergistic response feature vector relative to the optimized product feature vector to obtain a classification feature matrix comprises: calculating the response estimation of the synergistic response feature vector relative to the optimized product feature vector to obtain a classification feature matrix according to the following formula; wherein the formula is: = wherein represents the synergistic response feature vector, represents the optimized product feature vector, represents the classification feature matrix, represents matrix and vector multiplication.

6. The process of purifying high purity helium and co-producing LNG from lean helium natural gas according to claim 5, characterized in that, The classification feature matrix is passed through a classifier to obtain a classification result, including: expanding the classification feature matrix into a classification feature vector according to a row vector or a column vector; performing fully connected coding on the classification feature vector using a fully connected layer of the classifier to obtain a coded classification feature vector; and inputting the coded classification feature vector into a Softmax classification function of the classifier to obtain the classification result.

Images