A method and system for predicting total sulfur content of gas for a desulfurization system

Abstract

The application discloses a kind of gas total sulfur content prediction method and system for desulfurization system, comprising: obtaining the total sulfur content of the gas output end of the desulfurization system to be predicted stable time history process parameters, and according to the first parameter of history process parameters obtained indicating system internal gas-liquid temperature difference and the second parameter indicating the content proportion of hydrogen sulfide in the total sulfur content of gas input end;Determine the first type of related parameters related to the change of amine liquid concentration in historical process parameters, construct the first relationship model for predicting the amine liquid concentration of the current according to the first type of related parameters of previous period, determine the second type of related parameters related to the change of total sulfur content, construct the second relationship model for predicting the total sulfur content of current period according to the amine liquid concentration of current period, the first parameter of previous period, the second parameter, the second type of related parameters, utilize relationship model, total sulfur content is dynamically predicted.The application realizes the prediction of the sulfur content of gas in gas system.

Description

Technical Field

[0001] This invention belongs to the field of pollutant treatment, and in particular relates to a method and system for predicting the total sulfur content of methane gas in desulfurization systems. Background Technology

[0002] The flue gas from heating furnaces in the refining and chemical industry is currently subject to the Emission Standard of Pollutants for Petroleum Refining Industry (GB 31570-2015), with the sulfur dioxide emission limit in key areas being 50 mg / m³. 3 However, according to existing online monitoring data of pollutants in flue gas from refining and chemical industry heating furnaces, there are still many instances of sulfur dioxide exceeding the standard. Analysis of the causes of sulfur dioxide exceeding the standard reveals that the main factor contributing to the excessive sulfur dioxide emissions from heating furnace flue gas is the high and fluctuating total sulfur content in the gas, which makes it difficult for the gas desulfurization system to operate stably and efficiently.

[0003] The fuel gas system in the refining and chemical industry is mainly divided into a high-pressure gas pipeline network and a low-pressure gas pipeline network. The low-pressure gas pipeline network is connected to the low-pressure venting systems of each unit. Low-pressure gas generated by the production units is directly discharged into the low-pressure gas pipeline network. Some low-pressure gas is pressurized by a C3 compressor, desulfurized in a desulfurization unit, and then enters the high-pressure gas pipeline network. The remaining low-pressure gas enters the gas holder at the gas station and is then pressurized before being connected to the high-pressure gas pipeline network. The main sources of gas include catalytic dry gas, coking dry gas, and other production units. The gas contains combustible gases such as methane and ethane, but also contains high levels of hydrogen sulfide and organic sulfur. In addition, with the gradual upgrading of gasoline and diesel product quality, the requirements for hydrogen purity in hydrogenation units are becoming increasingly stringent. Some waste hydrogen with high hydrogen sulfide content is discharged into the fuel gas system, which can also lead to large fluctuations in the total sulfur content of the fuel gas system. If the gas is burned directly as fuel without desulfurization, the sulfur dioxide concentration in the flue gas will exceed the standard. Currently, most gas is desulfurized by gas desulfurization equipment and then enters the high-pressure gas pipeline network to be used as fuel for the process heating furnaces of various units. However, the desulfurized gas still has a high sulfur content due to the untimely operation and adjustment of the desulfurization facilities.

[0004] Currently, most enterprises reduce hydrogen sulfide concentration by increasing the amount of amine solution used. Research on dynamically adjusting desulfurization system operating parameters based on the sulfur content in the gas is limited. Therefore, existing technologies cannot adjust the gas desulfurization system in a timely and reasonable manner based on the actual operating status of the unit and the control requirements for the total sulfur content of the outlet gas. Furthermore, operators of gas desulfurization units in heating furnaces currently rely mainly on feedback from the sulfur dioxide emission concentration in the furnace flue gas and the experience of technical personnel to adjust the unit's operating parameters, rather than conducting real-time data analysis and judgment from the perspective of process parameters. When the sulfur content in the gas is high and fluctuates significantly, the inability to adjust the desulfurization system operating parameters in a timely manner based on the total sulfur content of the fuel gas leads to a significant risk of sulfur dioxide exceeding the standard in the furnace flue gas.

[0005] There is an urgent need to develop an operation optimization technology for gas desulfurization systems in heating furnaces. This technology should intelligently adjust the operating process parameters of the gas desulfurization system in a timely manner based on the total sulfur content of the gas in the system. This will ensure that sulfur dioxide generation is reduced at the source and help achieve stable and compliant gas emissions at the outlet of the desulfurization system. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a method for predicting the total sulfur content of methane gas in a desulfurization system. The method includes: acquiring historical process parameters of the desulfurization system under stable total sulfur content output, and obtaining a first parameter representing the gas-liquid temperature difference within the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the methane input based on the historical process parameters; determining a first type of correlation parameter related to the amine concentration change in the historical process parameters, and constructing a first relationship model for predicting the current amine concentration based on the first type of correlation parameter from the previous period; determining a second type of correlation parameter related to the total sulfur content change, and constructing a second relationship model for predicting the total sulfur content in the current period based on the current amine concentration, and the first, second, and second type of correlation parameters from the previous period; acquiring real-time process parameters of the desulfurization system under prediction, and dynamically predicting the total sulfur content using the first and second relationship models.

[0007] Preferably, the step of obtaining historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable further includes: preprocessing the historical process parameters to remove abnormal parameters and data generated by the desulfurization system to be predicted in standby mode, and using the preprocessed historical process parameters as the current historical process parameters, wherein the abnormal parameters are obtained based on the abnormal fluctuations of the historical process parameters.

[0008] Preferably, the step of constructing the first relational model includes: constructing a first preset model; using the first type of related parameters from the previous time period as input data for model training and the amine concentration from the current time period as output data for model training, training the first preset model to obtain the first relational model for predicting the amine concentration, wherein the first type of related parameters include the gas flow rate at the gas input end, the pressure difference of the absorption tower, the hydrogen sulfide concentration of the lean liquid, the content of thermally stable salts, and the amine addition rate.

[0009] Preferably, the step of constructing the second relational model includes: constructing a second preset model; using the predicted amine concentration of the current time period, the first parameter of the previous time period, the second parameter of the previous time period, and the second type of related parameters of the previous time period as input data for model training, and using the total sulfur content of the current time period as output data for model training, training the second preset model to obtain the second relational model for predicting the total sulfur content, wherein the second type of related parameters include the gas flow rate at the gas input end, the pressure difference of the absorption tower, the amine circulation volume, and the temperature of the absorption system.

[0010] Preferably, the time interval between each period is matched with the duration from the start of gas entering the desulfurization system from the gas input end to the output of desulfurized gas from the gas output end.

[0011] Preferably, the first preset model and the second preset model are constructed based on a neural network machine learning algorithm framework.

[0012] Preferably, the process of constructing the second preset model includes: determining the input vector of the current second preset model based on the predicted amine concentration value of the current time period, and the first parameter, second parameter, and second type of related parameter of the previous time period; determining the output vector of the current preset second model based on the expected total sulfur content value of the current time period; obtaining the number of input layer neurons, the number of hidden layer neurons, and the number of output layer neurons of the second preset model; and obtaining the initial values ​​of the feature parameters of the second preset model; based on the initial values ​​of the feature parameters, calculating the output value of each neuron in the hidden layer and the output value of each neuron in the output layer, thereby obtaining the initial adjustment weight between each neuron in the output layer and each neuron in the hidden layer, wherein the feature parameters include the connection weight from the hidden layer to the output layer, the center parameter of each neuron in the hidden layer, and the width vector of each neuron in the hidden layer.

[0013] Preferably, the step of obtaining the initial values ​​of the feature parameters of the second preset model includes: using the initial values ​​of the connection weights from the hidden layer to the output layer and the initial values ​​of the center parameters of each neuron in the hidden layer from the initial values ​​of the feature parameters, calculating the initial values ​​of the center parameters of the second preset model, and based on this, calculating the initial value of the width vector of each neuron in the hidden layer.

[0014] Preferably, after obtaining the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer, the method further includes: using a model convergence algorithm to perform multiple rounds of iterative calculations on each feature parameter of the second preset model to obtain the optimal parameters of the second preset model, thereby obtaining the second relational model.

[0015] Preferably, the step of obtaining the optimal parameters of the second preset model includes: setting a root mean square error threshold representing the output error of the second preset model, and comparing the calculation result with the preset root mean square error threshold after each round of iteration calculation to obtain the optimal calculation result, thereby obtaining the optimal parameters of the second preset model.

[0016] On the other hand, the present invention also provides a system for predicting the total sulfur content of gas in a desulfurization system. The system includes the following modules: a process parameter acquisition module, which is used to acquire historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable, and obtain a first parameter representing the gas-liquid temperature difference inside the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end based on the historical process parameters; a model construction module, which is used to determine a first type of correlation parameter related to the change in amine concentration in the historical process parameters, and construct a first relationship model for predicting the current amine concentration based on the first type of correlation parameter in the previous period, and determine a second type of correlation parameter related to the change in total sulfur content, and construct a second relationship model for predicting the total sulfur content in the current period based on the current amine concentration, and the first, second, and second type of correlation parameters in the previous period; and a total sulfur content prediction module, which is used to acquire the real-time process parameters of the desulfurization system to be predicted, and dynamically predict the total sulfur content using the first relationship model and the second relationship model.

[0017] Preferably, the system further includes a duration matching module, which is used to match the time interval between each time period with the duration from the start of gas entering the desulfurization system to be predicted from the gas input end to the output of desulfurized gas at the gas output end.

[0018] Compared with the prior art, one or more embodiments of the above solutions may have the following advantages or beneficial effects:

[0019] This invention proposes a method and system for predicting the total sulfur content of gas in a desulfurization system. The method first obtains historical process parameters at the gas output end of the desulfurization system to be predicted, and based on these historical parameters, obtains a first parameter representing the gas-liquid temperature difference within the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end. Next, it determines a first type of correlation parameter related to the change in amine concentration in the historical process parameters and constructs a first relationship model for predicting the amine concentration. It also determines a second type of correlation parameter related to the change in total sulfur content and constructs a second relationship model for predicting the total sulfur content. Finally, it obtains the real-time process parameters of the desulfurization system to be predicted and uses the first and second relationship models to dynamically predict the total sulfur content. This invention enables early prediction of sulfur dioxide concentration exceeding standards in flue gas from refining and chemical furnaces and dynamically recommends adjustment schemes for the operating parameters of the gas desulfurization system, solving the problem of lag in parameter adjustments by operators and ensuring the stability of the total sulfur content in the fuel gas. This invention features intelligence and rapid response, enabling the reduction of sulfur dioxide generation and emissions at the source without the need to add or replace pollution control facilities.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0022] Figure 1 This is a flowchart illustrating the steps of a method for predicting the total sulfur content of methane in a desulfurization system, as described in an embodiment of this application.

[0023] Figure 2 This is a block diagram of a gas total sulfur content prediction system for a desulfurization system according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and examples, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that, as long as there is no conflict, the various embodiments and features in the various embodiments of the present invention can be combined with each other, and the resulting technical solutions are all within the protection scope of the present invention.

[0025] Furthermore, the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0026] The flue gas from heating furnaces in the refining and chemical industry is currently subject to the Emission Standard of Pollutants for Petroleum Refining Industry (GB 31570-2015), with the sulfur dioxide emission limit in key areas being 50 mg / m³. 3 However, according to existing online monitoring data of pollutants in flue gas from refining and chemical industry heating furnaces, there are still many instances of sulfur dioxide exceeding the standard. Analysis of the causes of sulfur dioxide exceeding the standard reveals that the main factor contributing to the excessive sulfur dioxide emissions from heating furnace flue gas is the high and fluctuating total sulfur content in the gas, which makes it difficult for the gas desulfurization system to operate stably and efficiently.

[0027] The fuel gas system in the refining and chemical industry is mainly divided into a high-pressure gas pipeline network and a low-pressure gas pipeline network. The low-pressure gas pipeline network is connected to the low-pressure venting systems of each unit. Low-pressure gas generated by the production units is directly discharged into the low-pressure gas pipeline network. Some low-pressure gas is pressurized by a C3 compressor, desulfurized in a desulfurization unit, and then enters the high-pressure gas pipeline network. The remaining low-pressure gas enters the gas holder at the gas station and is then pressurized before being connected to the high-pressure gas pipeline network. The main sources of gas include catalytic dry gas, coking dry gas, and other production units. The gas contains combustible gases such as methane and ethane, but also contains high levels of hydrogen sulfide and organic sulfur. In addition, with the gradual upgrading of gasoline and diesel product quality, the requirements for hydrogen purity in hydrogenation units are becoming increasingly stringent. Some waste hydrogen with high hydrogen sulfide content is discharged into the fuel gas system, which can also lead to large fluctuations in the total sulfur content of the fuel gas system. If the gas is burned directly as fuel without desulfurization, the sulfur dioxide concentration in the flue gas will exceed the standard. Currently, most gas is desulfurized by gas desulfurization equipment and then enters the high-pressure gas pipeline network to be used as fuel for the process heating furnaces of various units. However, the desulfurized gas still has a high sulfur content due to the untimely operation and adjustment of the desulfurization facilities.

[0028] Currently, most enterprises reduce hydrogen sulfide concentration by increasing the amount of amine solution used. Research on dynamically adjusting desulfurization system operating parameters based on the sulfur content in the gas is limited. Therefore, existing technologies cannot adjust the gas desulfurization system in a timely and reasonable manner based on the actual operating status of the unit and the control requirements for the total sulfur content of the outlet gas. Furthermore, operators of gas desulfurization units in heating furnaces currently rely mainly on feedback from the sulfur dioxide emission concentration in the furnace flue gas and the experience of technical personnel to adjust the unit's operating parameters, rather than conducting real-time data analysis and judgment from the perspective of process parameters. When the sulfur content in the gas is high and fluctuates significantly, the inability to adjust the desulfurization system operating parameters in a timely manner based on the total sulfur content of the fuel gas leads to a significant risk of sulfur dioxide exceeding the standard in the furnace flue gas.

[0029] There is an urgent need to develop an operation optimization technology for gas desulfurization systems in heating furnaces. This technology should intelligently adjust the operating process parameters of the gas desulfurization system in a timely manner based on the total sulfur content of the gas in the system. This will ensure that sulfur dioxide generation is reduced at the source and help achieve stable and compliant gas emissions at the outlet of the desulfurization system.

[0030] Therefore, to address the aforementioned problems, this invention proposes a method and system for predicting the total sulfur content of gas in a desulfurization system. The method first obtains historical process parameters at the gas output end of the desulfurization system to be predicted, and based on these historical parameters, obtains a first parameter representing the gas-liquid temperature difference within the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end. Next, it determines a first type of correlation parameter related to the change in amine concentration in the historical process parameters and constructs a first relationship model for predicting the amine concentration. It also determines a second type of correlation parameter related to the change in total sulfur content and constructs a second relationship model for predicting the total sulfur content. Finally, it obtains the real-time process parameters of the desulfurization system to be predicted and uses the first and second relationship models to dynamically predict the total sulfur content. This invention enables advance prediction of sulfur dioxide concentration exceeding standards in the flue gas of refining furnaces and dynamically recommends adjustment schemes for the operating parameters of the gas desulfurization system in the refining industry. This solves the problem of lag in parameter adjustments by operators to the gas desulfurization system and ensures the stability of the total sulfur content of the fuel gas. This invention features intelligence and rapid response, enabling the reduction of sulfur dioxide generation and emissions at the source without the need to add or replace pollution control facilities.

[0031] Example One

[0032] Figure 1 This is a flowchart illustrating the steps of a method for predicting the total sulfur content of methane in a desulfurization system according to an embodiment of this application. See below for reference. Figure 1 This will explain each step of the method.

[0033] like Figure 1As shown, in step S110, historical process parameters are obtained when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable. Based on these historical process parameters, a first parameter representing the gas-liquid temperature difference within the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end are obtained. Specifically, historical data of key process parameters of the current gas desulfurization system to be predicted (i.e., process parameters related to changes in the total sulfur content at the gas output end) are collected, and the historical data when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable are used as the historical process parameters in this embodiment. The key process parameters described in this embodiment include, but are not limited to: lean liquid hydrogen sulfide concentration, thermally stable salt content, amine liquid addition rate, amine liquid concentration, amine liquid circulation rate, absorption system temperature, absorber tower top pressure, absorber tower bottom pressure, and the gas temperature, pressure, flow rate, and composition parameters at the gas input end of the desulfurization system to be predicted.

[0034] The first parameter represents the temperature difference between the gas phase and liquid phase inside the desulfurization system under test during operation. In obtaining this first parameter, representing the gas-liquid temperature difference within the system, based on historical process parameters, the first parameter is calculated using Pearson correlation. The correlation between this first parameter and the total sulfur content of the gas at the output end of the desulfurization system under test is significantly higher than the individual correlations between gas phase temperature and liquid phase temperature and the total sulfur content of the gas at the output end. The second parameter represents the ratio of the sulfur content in hydrogen sulfide to the total sulfur content in the gas at the output end of the desulfurization system under test; that is, it characterizes the proportion of hydrogen sulfide in the total sulfur. In obtaining this second parameter, representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end, based on historical process parameters, the second parameter is calculated using Pearson correlation. Among them, the correlation between the second parameter and the total sulfur content of the gas at the gas output end of the desulfurization system to be predicted is much higher than the correlation between the hydrogen sulfide content and the total sulfur content of the gas and the total sulfur content of the gas at the gas output end, respectively. Therefore, this embodiment utilizes the first and second parameters to construct a corresponding relationship model, which plays an important role in accurately predicting the total sulfur content of the gas at the gas output end.

[0035] Next, in the step of obtaining historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable, the historical process parameters are preprocessed to remove abnormal parameters and data generated by the desulfurization system to be predicted in standby mode. The preprocessed historical process parameters are then used as the current historical process parameters. Abnormal parameters are obtained based on abnormal fluctuations in the historical process parameters. Specifically, after obtaining the historical process parameters, this embodiment performs data preprocessing on the current historical process parameters to remove abnormal parameters from abnormal operating periods and data generated by the desulfurization system to be predicted in standby mode. The preprocessed historical process parameters from normal operating periods are then used as the current historical process parameters. The data generated by the desulfurization system to be predicted in standby mode is data from a shutdown state. This embodiment pre-sets a threshold for filtering normal parameters, thereby filtering out abnormal parameters from the original historical process parameters by selecting historical process parameters that do not meet the current threshold. In this embodiment, the threshold for filtering normal parameters is obtained based on the maximum and minimum values ​​of each process parameter. The normal deviation between the maximum and minimum values ​​is used as the threshold for filtering normal parameters. The deviation between the maximum and minimum values ​​is used to determine the data fluctuation. Then, the current operating condition is determined based on the data fluctuation, and the historical process parameters corresponding to the abnormal operating conditions are removed.

[0036] Further, in step S120, a first type of relevant parameter related to the change in amine concentration in historical process parameters is determined, and a first relationship model for predicting the current amine concentration based on the first type of relevant parameter in the previous period is constructed. Similarly, a second type of relevant parameter related to the change in total sulfur content is determined, and a second relationship model for predicting the total sulfur content in the current period is constructed based on the current amine concentration, and the first, second, and second type of relevant parameters from the previous period. In this embodiment, the first type of relevant parameter related to the change in amine concentration in historical process parameters is determined, and a first relationship model for predicting the current amine concentration is constructed based on the correlation between the first type of parameter in the previous period and the amine concentration in the current period. The first type of relevant parameter includes, but is not limited to: the gas flow rate and composition parameters at the gas input end of the desulfurization system to be predicted, as well as the absorber pressure difference, lean liquid hydrogen sulfide concentration, thermally stable salt content, and amine addition rate. After obtaining the predicted amine concentration for the current period, a second type of correlation parameter related to the change in total sulfur content is determined. Based on the correlation between the current amine concentration, the first parameter, the second parameter, the second type parameter from the previous period, and the total sulfur content for the current period, a second relationship model is constructed to predict the total sulfur content for the current period. The second type of parameter includes, but is not limited to: gas flow rate at the gas input end, absorber pressure differential, amine circulation rate, and absorber system temperature.

[0037] In the step of constructing the first relational model, firstly, a first preset model is constructed; then, the first type of relevant parameters from the previous time period are used as input data for model training, and the amine concentration of the current time period is used as output data for model training to train the first preset model, thereby obtaining the first relational model used to predict the amine concentration. Specifically, the first preset model is constructed based on machine learning technology. Then, the first type of relevant parameters from the previous time period are used as input information for the preset machine learning model, and training begins. After training, the first relational model is obtained. The first relational model can calculate the amine concentration of the current time period, generate the amine concentration of the current time period, and obtain the predicted value of the amine concentration of the current time period. In a specific embodiment of this application, the first type of relevant parameters include the gas flow rate at the gas input end, the absorption tower pressure difference, the lean liquid hydrogen sulfide concentration, the thermally stable salt content, and the amine addition rate.

[0038] In the step of constructing the second relational model, firstly, a second preset model is constructed; then, the predicted amine concentration of the current time period, the first parameter of the previous time period, the second parameter of the previous time period, and the second type of correlation parameters of the previous time period are used as input data for model training, and the total sulfur content of the current time period is used as output data for model training. The second preset model is trained to obtain a second relational model for predicting the total sulfur content. Specifically, the second preset model is constructed based on machine learning technology. Then, the predicted amine concentration of the current time period, the first parameter of the previous time period, the second parameter of the previous time period, and the second type of correlation parameters of the previous time period are used as input information for the preset machine learning model, the second preset model, and training begins. After training, the second relational model is obtained. The second relational model can calculate the total sulfur content of the current time period and generate a predicted value for the total sulfur content of the current time period. In a specific embodiment of this application, the second type of correlation parameters include the gas flow rate at the gas input end, the pressure difference of the absorption tower, the amine circulation volume, and the temperature of the absorption system.

[0039] Furthermore, the time interval between each period is matched with the duration from the start of gas entering the desulfurization system from the gas input terminal to the output of desulfurized gas. In this embodiment, the time interval from the start of gas entering the desulfurization system from the gas input terminal to the output of desulfurized gas is used as the time interval between each period, ensuring that the time interval between each period matches the duration from the start of gas entering the desulfurization system from the gas input terminal to the output of desulfurized gas. Accordingly, this invention considers the time difference from the start of the reaction to the end of the reaction in the desulfurization system, thus obtaining an accurate prediction result of the total sulfur content.

[0040] In the embodiments of this application, both the first preset model and the second preset model are constructed based on the neural network machine learning algorithm framework.

[0041] In the process of constructing the second preset model, the input vector of the current second preset model is determined based on the predicted value of amine concentration in the current time period, as well as the first parameter, second parameter, and second type of related parameter in the previous time period. The output vector of the current preset second model is determined based on the expected value of total sulfur content in the current time period. The number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer of the second preset model are obtained, and the initial values ​​of the feature parameters of the second preset model are obtained.

[0042] In one specific embodiment of this application, the neural network machine learning algorithm employs an RBF neural network. The purpose of training the second preset model is to determine the weights between the input layer and the hidden layer in the RBF neural network, and to determine the weights between the hidden layer and the output layer. First, the input vector X used to train the RBF neural network is determined, and the number of elements in the input vector (the number of process parameters) is used as the number of neurons in the input layer. The input vector X is represented by the following expression:

[0043] X = [x1, x2, ..., x] q ] T (1)

[0044] Where q represents the number of neurons in the input layer, x represents various process parameters, and X represents the input vector.

[0045] Next, the output vector Y and the desired output vector O for training the RBF neural network are determined, where the output vector Y and the desired output vector O are represented by the following expressions:

[0046] Y = [y1, y2, ..., y] q ] T (2)

[0047] O = [o1, o2, ..., o] q ] T (3)

[0048] Where q represents the number of neurons in the output layer, y represents the total sulfur content of the gas at the gas output end of the desulfurization system to be predicted, Y represents the output vector, O represents the desired output vector, and o represents the desired output of the total sulfur content of the gas at the gas output end of the desulfurization system to be predicted.

[0049] In this embodiment of the application, there is only one output vector, namely: the total sulfur content of the gas at the gas output end of the desulfurization system to be predicted, which is q=1.

[0050] Then, the reference center initialization method provides a method for initializing the weights from the hidden layer to the output layer, and uses the following expression to initialize the connection weights from the hidden layer to the output layer:

[0051] W k =[w k1 ,w k2 ,....w kp ] T ,k=1,2....,q (4)

[0052]

[0053] Where p represents the number of neurons in the hidden layer, W k w represents the connection weights from the hidden layer to the output layer. kp W represents the connection weights corresponding to the k-th output neuron, mink represents the minimum expected output of the k-th output neuron in the training set, maxk represents the maximum expected output of the k-th output neuron in the training set, and W represents the maximum expected output of the k-th output neuron in the training set. kj This represents the adjustment weight between the k-th neuron in the output layer and the j-th neuron in the hidden layer.

[0054] In this embodiment, the number of hidden layer neurons p = 2.

[0055] Next, the central parameters of each neuron in the hidden layer are initialized using the following expression:

[0056] C j =[c j1 ,c j2 ,....c jn ] T (6)

[0057] Among them, C j c represents the center vector of the j-th neuron in the hidden layer. jn This indicates that the j-th neuron of the n-th hidden layer unit corresponds to the central component of all neurons in the input layer.

[0058] In the step of obtaining the initial values ​​of the feature parameters of the second preset model, the initial values ​​of the connection weights from the hidden layer to the output layer and the initial values ​​of the center parameters of each neuron in the hidden layer are used to calculate the initial values ​​of the center parameters of the second preset model. Based on this, the initial value of the width vector of each neuron in the hidden layer is calculated.

[0059] The initial values ​​of the feature parameters of the second preset model, i.e., the initial values ​​of the center parameters of the RBF neural network, are calculated using the following expression:

[0060]

[0061] Where mini represents the minimum value of all input information for the i-th feature in the training set, maxi represents the maximum value of all input information for the i-th feature in the training set, and C ji This represents the initial values ​​of the feature parameters of the second preset model.

[0062] In this embodiment, a width adjustment coefficient is introduced to make each hidden layer neuron more receptive to local information, thereby improving the local response capability of the RBF neural network. The width vector of the second preset model is initialized using the following expression to obtain the initial value of the width vector of each hidden layer neuron:

[0063] D j =[d j1 ,d j2 ,....d jn (8)

[0064]

[0065] Where, d f D represents the width adjustment coefficient, which takes a value less than 1. j c represents the initial width vector of the j-th neuron in the hidden layer. ji d represents the initial value of the center parameter of the RBF neural network. ji Indicates the relationship with the central parameter c ji The corresponding width, x i This represents the input parameter for the i-th feature.

[0066] Next, based on the initial values ​​of the feature parameters, the output values ​​of each neuron in the hidden layer and each neuron in the output layer are calculated, thereby obtaining the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer. The feature parameters include the connection weights from the hidden layer to the output layer, the center parameters of each neuron in the hidden layer, and the width vector of each neuron in the hidden layer.

[0067] Specifically, the output value z of the j-th neuron in the hidden layer is calculated using the following expression. j :

[0068]

[0069] Among them, z j C represents the output value of the j-th neuron in the hidden layer. j D represents the center vector of the j-th neuron in the hidden layer, which is composed of the center components of all neurons in the input layer corresponding to the j-th neuron in the hidden layer. j Let ||.|| represent the width vector of the j-th neuron in the hidden layer, and ||.|| represent the Euclidean norm.

[0070] In the embodiments of this application, D j This represents the width vector of the j-th neuron in the hidden layer, and is related to C. j Correspondingly. And, D j The larger the layer size, the greater the influence of the hidden layer on the input vector, and the better the smoothness between neurons.

[0071] The output value of the output layer neuron is calculated using the following expression:

[0072]

[0073] Among them, w kj y represents the adjustment weight between the k-th neuron in the output layer and the j-th neuron in the hidden layer. k This represents the output value of the k-th neuron in the output layer.

[0074] In this embodiment, the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer can be obtained based on the output values ​​of the neurons in the output layer when the input data is first input into the second preset model.

[0075] Furthermore, after obtaining the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer, a model convergence algorithm is used to perform multiple rounds of iterative calculations on each feature parameter of the second preset model to obtain the optimal parameters of the second preset model, thereby obtaining the second relational model.

[0076] In one specific embodiment of this application, gradient descent is used as the model convergence algorithm to train the adjustment weights of the RBF neural network. All feature parameters of the second preset model are adaptively adjusted to their optimal values ​​through learning. The optimal parameters of the second preset model are calculated using the following expressions:

[0077]

[0078]

[0079]

[0080] Among them, w kj (t) represents the adjustment weight between the k-th output neuron and the j-th hidden layer neuron during the t-th iteration calculation, c ji (t) represents the central component of the j-th hidden layer neuron in relation to the i-th input neuron during the t-th iteration, d ji (t) represents the relationship with center c. ji (t) represents the width, η represents the learning factor, E represents the RBF neural network evaluation function, and η and α are coefficients.

[0081] Furthermore, the evaluation function of the RBF neural network is calculated using the following expression:

[0082]

[0083] Among them, O lk Let y represent the expected output value of the k-th output neuron when the l-th sample is input. lk This represents the network output value of the k-th output neuron when the l-th sample is input, and N represents the number of samples.

[0084] Furthermore, in the step of obtaining the optimal parameters of the second preset model, a root mean square error (RMS) threshold representing the output error of the second preset model is preset. After each iteration, the calculation result is compared with the preset RMS threshold to obtain the optimal calculation result, thereby obtaining the optimal parameters of the second preset model. Specifically, in this embodiment, the weight adjustment accuracy is optimized according to the target of the second preset model, and the values ​​of η and α and the iteration termination accuracy ε (i.e., the RMS threshold) are preset. After initializing the corresponding parameters in the RBF neural network, the RMS value of the network output is used to determine whether to terminate the iteration, and the current calculation result is used as the optimal parameters of the second preset model. If RMS ≤ ε, the training ends and the predicted value of total sulfur content in the gas is obtained; otherwise, the iteration calculation of each parameter of the second preset model continues. The RMS value of the network output is calculated using the following expression:

[0085]

[0086] In step S130, real-time process parameters of the desulfurization system to be predicted are obtained, and the total sulfur content is dynamically predicted using a first relational model and a second relational model. Specifically, the real-time process parameters of the desulfurization system to be predicted are obtained, and the first type of related parameters in the real-time process parameters are used as input data for the first relational model to predict the current amine concentration. Then, the amine concentration of the current time period, as well as the first parameter, second parameter, and second type of related parameters of the previous time period, are used as input data for the second relational model, thereby realizing the dynamic prediction of the total sulfur content at different time periods.

[0087] In this embodiment of the application, different desulfurization system operating parameter adjustment schemes are preset. After obtaining the predicted total sulfur content, the actual sulfur content of the gas in the current period is compared and analyzed with the predicted total sulfur content generated for the current period. Based on the deviation between the two, the optimal operating parameter adjustment scheme applicable to the desulfurization system to be predicted in the current period is invoked. Accordingly, using the currently invoked optimal operating parameter adjustment scheme, the hydrogen sulfide content at the gas output end of the desulfurization system to be predicted is controlled until the hydrogen sulfide content is maintained within a preset reasonable range, thereby ensuring the stability of the total sulfur content of the gas at the output end.

[0088] In the process of calling the optimal operating parameter adjustment scheme applicable to the current period, this embodiment also monitors the excess state of the total sulfur content of the gas at the gas output end of the desulfurization system to be predicted in real time, and generates the operating parameters applicable to the desulfurization system to be predicted in the current period when the excess state is detected, so as to restore the excess state to the normal state.

[0089] Furthermore, when the total sulfur content at the gas output end of the desulfurization system to be predicted cannot be adjusted immediately, this embodiment uses a corresponding alarm device to issue an alarm for the current operating condition and sends the generated operating parameters applicable to the desulfurization system to be predicted for the current period to the corresponding location (e.g., the mobile phone or email of relevant personnel) in real time, so as to remind relevant personnel of the current operating condition and the operating parameter adjustment plan given for the current operating condition.

[0090] Example Two

[0091] Based on the method for predicting the total sulfur content of methane gas in a desulfurization system described in Embodiment 1 above, this invention also provides a system for predicting the total sulfur content of methane gas in a desulfurization system (hereinafter referred to as the "total sulfur content prediction system"). Figure 2 This is a block diagram of a gas total sulfur content prediction system for a desulfurization system according to an embodiment of this application.

[0092] like Figure 2 As shown, the total sulfur content prediction system for gas in this embodiment of the invention includes: a process parameter acquisition module 21, a model construction module 22, and a total sulfur content prediction module 23. Specifically, the process parameter acquisition module 21 is implemented according to the method described in step S110 above, configured to acquire historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable, and obtain a first parameter representing the gas-liquid temperature difference inside the system and a second parameter representing the proportion of hydrogen sulfide content in the total sulfur content at the gas input end based on the historical process parameters; the model construction module 22 is implemented according to the method described in step S120 above, configured to determine a first type of related parameter related to the change in amine concentration in the historical process parameters acquired by the process parameter acquisition module 21, and construct a first relationship model for predicting the current amine concentration based on the first type of related parameter in the previous period, and determine a second type of related parameter related to the change in total sulfur content, and construct a second relationship model for predicting the total sulfur content in the current period based on the amine concentration in the current period, and the first, second, and second type of related parameters in the previous period; the total sulfur content prediction module 23 is implemented according to the method described in step S130 above, configured to acquire real-time process parameters of the desulfurization system to be predicted, and dynamically predict the total sulfur content using the first and second relationship models constructed by the model construction module 22.

[0093] Furthermore, the total sulfur content prediction system for gas in this embodiment of the invention also includes a time-matching module, which is used to match the time interval between each time period with the time between the start of gas entering the desulfurization system to be predicted from the gas input end and the output of desulfurized gas at the gas output end.

[0094] Example Three

[0095] In one specific embodiment of this application, firstly, 34,800 pieces of raw data from a gas desulfurization system of a heating furnace between May 10, 2020 and July 20, 2020 are collected as historical process parameters for data analysis and correlation analysis. A total of twelve process parameters are retained: gas flow rate at the gas input end, gas temperature, total sulfur content of gas, hydrogen sulfide content of gas, hydrogen sulfide concentration of lean liquid, thermally stable salt content, amine addition rate, amine concentration, amine circulation volume, absorption system temperature, absorption tower top pressure, and absorption tower bottom pressure.

[0096] Next, abnormal parameters and data generated during the standby process of the desulfurization system were removed from each historical process parameter to obtain the current historical process parameters. Then, 30,000 data points from May 10, 2020 to July 10, 2020 were used as model training data, and 4,800 data points from July 11, 2020 to July 20, 2020 were used as model testing data.

[0097] Subsequently, the aforementioned 12 process parameters were used as inputs to the RBF neural network model. This meant setting the model to have 12 neurons in the input layer, 8 neurons in the hidden layer, and 1 neuron in the output layer (used to output the total sulfur content of the gas at the gas output end of the furnace gas desulfurization system) (i.e., a 12-8-1 model structure). Simultaneously, the aforementioned five process parameters—the predicted amine concentration, the gas flow rate at the gas input end, the amine circulation rate, the absorption system temperature, and the absorption tower pressure difference—along with the calculated first and second parameters, totaling seven parameters, were used as inputs to the RBF neural network model. This meant setting the model to have 7 neurons in the input layer, 5 neurons in the hidden layer, and 1 neuron in the output layer (used to output the total sulfur content of the gas at the gas output end of the furnace gas desulfurization system) (i.e., a 7-5-1 model structure).

[0098] Finally, the total sulfur content of the gas at the gas output end of the gas desulfurization system of the heating furnace was predicted using models with structures of 12-8-1 and 7-5-1, respectively. The prediction results showed that the RBF neural network model with 7 input parameters (i.e., including the first and second parameters) had a prediction accuracy of 93.7%, while the original model with 12 input parameters (excluding the first and second parameters) had a prediction accuracy of only 81.5%. Therefore, this invention, by using the first and second parameters, achieves higher prediction accuracy for the total sulfur content of the gas at the gas output end of the gas desulfurization system of the heating furnace.

[0099] This invention proposes a method and system for predicting the total sulfur content of methane gas in a desulfurization system. The method first obtains historical process parameters at the methane output end of the desulfurization system to be predicted, and based on these historical parameters, obtains a first parameter representing the gas-liquid temperature difference within the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the methane input end. Next, it determines a first type of correlation parameter related to the change in amine concentration in the historical process parameters and constructs a first relationship model for predicting the amine concentration. It also determines a second type of correlation parameter related to the change in total sulfur content and constructs a second relationship model for predicting the total sulfur content. Finally, it obtains the real-time process parameters of the desulfurization system to be predicted and uses the first and second relationship models to dynamically predict the total sulfur content. This invention enables advance prediction of sulfur dioxide concentration exceeding standards in flue gas from refining and chemical furnaces, and dynamically recommends adjustment schemes for the operating parameters of the furnace gas desulfurization system. It facilitates real-time monitoring, intelligent sensing, intelligent decision-making, and operational control of the furnace gas desulfurization system, solving the problem of continuous sulfur dioxide emissions exceeding standards due to the lag in operator parameter adjustments to the gas desulfurization system, and ensuring stable total sulfur content in fuel gas. This invention is characterized by intelligence and rapid response, reducing sulfur dioxide generation and emissions at the source without adding or replacing pollution control facilities.

[0100] The above description is merely a specific implementation example of the present invention, and the scope of protection of the present invention is not limited thereto. Any modifications or substitutions made to the present invention by those skilled in the art within the technical specifications described herein should be within the scope of protection of the present invention.

[0101] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the claims of the present invention.

[0102] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0103] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for predicting the total sulfur content of methane gas in a desulfurization system, characterized in that, include: Historical process parameters were obtained when the total sulfur content at the gas output end of the desulfurization system to be predicted was stable. Based on the historical process parameters, a first parameter representing the gas-liquid temperature difference inside the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end were obtained. A first type of related parameter is determined that is related to the change in amine concentration in the historical process parameters, and a first relationship model is constructed to predict the current amine concentration based on the first type of related parameter in the previous period. A second type of related parameter is determined that is related to the change in total sulfur content, and a second relationship model is constructed to predict the total sulfur content in the current period based on the amine concentration in the current period, as well as the first parameter, second parameter and second type of related parameter in the previous period. The real-time process parameters of the desulfurization system to be predicted are obtained, and the total sulfur content is dynamically predicted using the first relational model and the second relational model. The steps of constructing the first relational model include: Construct the first preset model; The first type of relevant parameters from the previous time period are used as input data for model training, and the amine concentration from the current time period is used as output data for model training. The first preset model is trained to obtain the first relationship model for predicting the amine concentration. The first type of relevant parameters include the gas flow rate at the gas input end, the pressure difference in the absorption tower, the hydrogen sulfide concentration in the lean solution, the content of thermally stable salts, and the amine addition rate. The steps in constructing the second relational model include: Construct a second pre-defined model; The predicted amine concentration for the current period, the first parameter for the previous period, the second parameter for the previous period, and the second type of related parameters for the previous period are used as input data for model training, and the total sulfur content for the current period is used as output data for model training. The second preset model is trained to obtain the second relationship model for predicting the total sulfur content. The second type of related parameters include the gas flow rate at the gas input end, the pressure difference of the absorption tower, the amine circulation volume, and the temperature of the absorption system.

2. The method for predicting the total sulfur content of methane gas according to claim 1, characterized in that, The steps for obtaining historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable also include: By preprocessing the historical process parameters, abnormal parameters and data generated by the desulfurization system to be predicted in standby mode are removed from the historical process parameters, and the preprocessed historical process parameters are used as the current historical process parameters. The abnormal parameters are obtained based on the abnormal fluctuations of the historical process parameters.

3. The method for predicting the total sulfur content of methane gas according to claim 1 or 2, characterized in that, The time interval between each period is matched with the duration from the start of gas entering the desulfurization system from the gas input end to the output of desulfurized gas from the gas output end.

4. The method for predicting the total sulfur content of methane gas according to claim 3, characterized in that, The first preset model and the second preset model are constructed based on a neural network machine learning algorithm framework.

5. The method for predicting the total sulfur content of methane gas according to claim 4, characterized in that, The process of constructing the second preset model includes: The input vector of the current second preset model is determined based on the predicted value of amine concentration in the current period, as well as the first parameter, second parameter, and second type of related parameter in the previous period. The output vector of the current second preset model is determined based on the expected value of total sulfur content in the current period. The number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer of the second preset model are obtained, and the initial values ​​of the feature parameters of the second preset model are obtained. Based on the initial values ​​of the feature parameters, the output value of each neuron in the hidden layer and the output value of each neuron in the output layer are calculated, thereby obtaining the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer. The feature parameters include the connection weights from the hidden layer to the output layer, the center parameters of each neuron in the hidden layer, and the width vector of each neuron in the hidden layer.

6. The method for predicting the total sulfur content of methane gas according to claim 5, characterized in that, The step of obtaining the initial values ​​of the feature parameters of the second preset model includes: Using the initial values ​​of the connection weights from the hidden layer to the output layer and the initial values ​​of the center parameters of each neuron in the hidden layer, the initial values ​​of the center parameters of the second preset model are calculated. Based on this, the initial values ​​of the width vector of each neuron in the hidden layer are calculated.

7. The method for predicting the total sulfur content of methane gas according to claim 5, characterized in that, After obtaining the initial adjustment weights between each neuron in the output layer and each neuron in the hidden layer, the following steps are also included: The model convergence algorithm is used to perform multiple rounds of iterative calculations on each feature parameter of the second preset model to obtain the optimal parameters of the second preset model, thereby obtaining the second relation model.

8. The method for predicting the total sulfur content of methane gas according to claim 7, characterized in that, The step of obtaining the optimal parameters of the second preset model includes: The preset represents the root mean square error threshold of the output error of the second preset model. After each round of iteration, the calculation result is compared with the preset root mean square error threshold to obtain the best calculation result, thereby obtaining the optimal parameters of the second preset model.

9. A system for predicting the total sulfur content of methane gas in a desulfurization system, the system comprising the following modules: The process parameter acquisition module is used to acquire historical process parameters when the total sulfur content at the gas output end of the desulfurization system to be predicted is stable, and to obtain a first parameter representing the gas-liquid temperature difference inside the system and a second parameter representing the proportion of hydrogen sulfide in the total sulfur content at the gas input end based on the historical process parameters. The model building module is used to determine the first type of related parameters related to the change in amine concentration in the historical process parameters, and to build a first relationship model for predicting the current amine concentration based on the first type of related parameters of the previous period. It also determines the second type of related parameters related to the change in total sulfur content, and to build a second relationship model for predicting the total sulfur content of the current period based on the current amine concentration, and the first, second and second type of related parameters of the previous period. The total sulfur content prediction module is used to acquire the real-time process parameters of the desulfurization system to be predicted, and to dynamically predict the total sulfur content using the first relational model and the second relational model. The model building module is also used for: When constructing the first relational model, a first preset model is first constructed. Then, the first type of related parameters from the previous time period are used as input data for model training, and the amine concentration from the current time period is used as output data for model training. The first preset model is then trained to obtain the first relational model used to predict the amine concentration. The first type of related parameters include the gas flow rate at the gas input end, the pressure difference in the absorption tower, the hydrogen sulfide concentration in the lean solution, the content of thermally stable salts, and the amine addition rate. When constructing the second relationship model, a second preset model is first constructed. Then, the predicted amine concentration of the current time period, the first parameter of the previous time period, the second parameter of the previous time period, and the second type of related parameters of the previous time period are used as input data for model training, and the total sulfur content of the current time period is used as output data for model training. The second preset model is trained to obtain the second relationship model for predicting the total sulfur content. The second type of related parameters include the gas flow rate at the gas input end, the pressure difference of the absorption tower, the amine circulation volume, and the temperature of the absorption system.

10. The gas total sulfur content prediction system according to claim 9, characterized in that, The total sulfur content prediction system for methane gas also includes: The duration matching module is used to match the time interval between each time period with the duration from the start of gas entering the desulfurization system to be predicted at the gas input end to the output of desulfurized gas at the gas output end.

Images