Method and device for predicting emission concentration of nitrogen oxide in ethylene cracking furnace and medium

Abstract

The invention provides an ethylene cracking furnace nitrogen oxide emission concentration prediction method and device and a medium, and belongs to the field of petrochemical engineering. The method comprises the following steps: collecting real-time operation data of the ethylene cracking furnace; the real-time operation data are preprocessed, emission concentration key parameters are determined, and cracking furnace thermal load composite variables are constructed; performing parameter optimization on the emission concentration key parameters and the composite variables based on an SEM structure equation model to obtain optimized parameters; and inputting the optimization parameters into a pre-constructed nitrogen oxide emission concentration prediction model for calculation to obtain prediction data. According to the method provided by the invention, the temperature factor variables influencing the emission concentration are combined into the composite variable, and the composite variable can reflect the influence of the fuel and the raw material on the nitrogen oxide, so that the significance of the influence of the temperature influence factors on the thermal nitrogen oxide is improved, and finally, the fitting degree of the model is favorably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical industry, and in particular to a method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace, a device for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace, a computer device and a computer-readable storage medium. Background Art

[0002] With the increase in the intensity of air pollution control, sulfur dioxide emissions have dropped significantly, but nitrogen oxide emissions are still not ideal. Refining companies can basically control NOx emissions in ethylene cracking furnace flue gas to 100mg / m 3 In order to meet the needs of regional economic and social development and environmental protection, and to further prevent and control the environmental damage caused by industrial air pollution emissions, local emission standards for air pollutants have been formulated. Currently, it is stipulated that the NOx emissions of industrial gas boilers shall not exceed 80mg / m 3 , nitrogen oxide emissions are difficult to meet the further increased 50mg / m 3 Even stricter emission standards.

[0003] The output and production technology level of ethylene are important indicators of the development level of a petrochemical industry. The NOx generation mechanism of ethylene cracking furnaces is complex and difficult to control. NOx has become a very difficult pollutant to control in the flue gas of cracking furnaces. Due to the high complexity of the combustion process and the diversity of influencing factors, the research on CFD simulation of NOx emissions from ethylene cracking furnaces is not extensive. Some models lack model verification, and the prediction error of some models is as high as 20% to 30%. Only a small number of models can achieve relatively accurate predictions. Due to the complexity of the chemical reaction process, raw materials, fuels, and operating conditions will affect the reaction process. It is still difficult to use mechanism models to predict NOx emissions from ethylene cracking furnaces.

[0004] With the development of artificial intelligence and computer technology, data-driven modeling methods have attracted widespread attention. This method mainly establishes the relationship between NOx emissions and their auxiliary variables. Currently, support vector product, back propagation neural network, extreme learning machine, deep learning method, long short-term memory neural network, etc. are used to predict NOx emissions from coal-fired boilers, diesel engines, etc., and the absolute error of some models is less than 1%. The prediction of NOx emissions from the cracking furnace of the ethylene plant is basically blank, and the prediction of NOx emissions from the cracking furnace of the ethylene plant still needs to be studied in order to effectively control pollutant emissions during the combustion stage. Summary of the invention

[0005] In view of the technical problem of difficulty in predicting NOx emissions from ethylene cracking furnaces in the prior art, the present invention provides a method, device and medium for predicting the concentration of nitrogen oxide emissions from ethylene cracking furnaces. The method can achieve comprehensive and accurate monitoring and timely emission reduction of nitrogen oxides from ethylene cracking furnaces.

[0006] To achieve the above-mentioned purpose, the present invention provides a method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace, comprising the following steps: collecting real-time operation data of the ethylene cracking furnace; preprocessing the real-time operation data, determining key parameters of the emission concentration, and constructing a composite variable of the thermal load of the cracking furnace; optimizing the key parameters of the emission concentration and the composite variable of the thermal load of the cracking furnace based on a SEM structural equation model to obtain optimized parameters; and inputting the optimized parameters into a pre-constructed nitrogen oxide emission concentration prediction model for calculation to obtain predicted data.

[0007] In an exemplary embodiment of the present invention, the method may also include: constructing a nitrogen oxide emission concentration prediction model, which may include: collecting historical operation data and emission concentration data of an ethylene cracking furnace; preprocessing the historical operation data, determining key parameters of emission concentration, and constructing a cracking furnace heat load composite variable; optimizing the key parameters of emission concentration and the cracking furnace heat load composite variable based on a SEM structural equation model to obtain optimized parameters; inputting the optimized parameters and emission concentration data into an LSTM model for training to obtain a nitrogen oxide emission concentration prediction model.

[0008] In an exemplary embodiment of the present invention, the preprocessing of real-time operation data may include: checking and eliminating outliers in the data; filling missing data based on the Lagrange interpolation method; resampling the filled data; using a single-sample KS test to determine whether the data conforms to the normal distribution; if so, standardizing the data; if not, logarithmically transforming the data that does not conform to the normal distribution, and standardizing the data that conforms to the normal distribution after the transformation.

[0009] In an exemplary embodiment of the present invention, the determination of key parameters of emission concentration may include: performing an indicator importance analysis based on expert experience rules and hierarchical analysis method, scoring parameter importance according to an expert diagnostic parameter list of nitrogen oxide emissions from ethylene plants, and determining key parameters of emission concentration according to the parameter importance level.

[0010] In an exemplary embodiment of the present invention, the parameter optimization of the emission concentration key parameter and the cracking furnace heat load composite variable based on the SEM structural equation model to obtain the optimized parameters may include: determining latent variables and corresponding observed variables based on the emission concentration key parameter and the cracking furnace heat load composite variable for constructing the SEM structural equation model; wherein the SEM structural equation model includes a measurement model for calculating the relationship between the observed variables and the latent variables and a structural model for calculating the relationship between the latent variables; inputting historical operation data into the SEM structural equation model, calculating the standardized load coefficient value, and screening out the SEM structural equation model whose standardized load coefficient value meets the set coefficient range; by analyzing the chi-square degree of freedom ratio, goodness of fit index, root mean square error of approximation and Akaike information criterion of the SEM structural equation model whose standardized load coefficient value meets the set coefficient range, screening and determining the optimal model, and obtaining the optimized parameters of the optimal model.

[0011] In an exemplary embodiment of the present invention, constructing a cracking furnace heat load composite variable may include: obtaining temperature factor variables affecting emission concentration based on real-time operation data; and combining the temperature factor variables affecting emission concentration to construct a cracking furnace heat load composite variable.

[0012] In an exemplary embodiment of the present invention, the cracking furnace heat load composite variable can be obtained by formula (1);

[0013] Q hj =K R (F R1 +F R2 +……+F Rn )+K Y (F Y1 +F Y2 +……+F Ym ) (1);

[0014] In formula (1), Q hj K is the heat load equivalent of ethylene cracking furnace; R is the fuel combustion conversion coefficient; K Y is the cracking exothermic conversion coefficient; F Rn is the flow rate of each fuel in the furnace; F Ym is the flow rate of each raw material.

[0015] In an exemplary embodiment of the present invention, the method may further include: determining whether the predicted data exceeds a safety threshold; and issuing an early warning if the predicted data exceeds the safety threshold.

[0016] In an exemplary embodiment of the present invention, the method may further include: if it is determined that the predicted data exceeds a safety threshold, optimizing a parameter adjustment scheme by calculating a nitrogen oxide emission concentration prediction model.

[0017] Another aspect of the present invention provides a device for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace, including: a data acquisition module, used to collect real-time operating data of the ethylene cracking furnace; a data processing module, used to preprocess the real-time operating data, determine the key parameters of the emission concentration, and construct a composite variable of the thermal load of the cracking furnace; a parameter optimization module, used to optimize the key parameters and composite variables of the emission concentration based on the SEM structural equation model to obtain optimized parameters; a data prediction module, used to input the optimized parameters into a pre-constructed nitrogen oxide emission concentration prediction model for calculation to obtain predicted data.

[0018] Another aspect of the present invention provides a computer device, comprising: at least one processor; a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions include instructions for executing the above-mentioned ethylene cracking furnace nitrogen oxide emission concentration prediction method.

[0019] Another aspect of the present invention provides a computer-readable storage medium having computer program instructions stored thereon, which implement the above-mentioned method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace when the computer program instructions are executed by a processor.

[0020] Through the technical solution provided by the present invention, the present invention has at least the following technical effects:

[0021] 1) The present invention combines the temperature factor variables that affect the emission concentration into a composite variable, which can reflect the influence of fuel and raw materials on nitrogen oxides, thereby improving the significance of the influence of temperature factors on thermal nitrogen oxides, and ultimately helping to improve the fitting degree of the model;

[0022] 2) Based on big data analysis and modeling, the present invention makes full use of expert experience rules and hierarchical analysis method to obtain key parameters of emission concentration, thereby helping to improve the accuracy of modeling;

[0023] 3) The present invention first determines the latent variables and corresponding observed variables used to construct the structural equation model for the key parameters of emission concentration and the composite variables of cracking furnace heat load, and then performs exploratory factor analysis, confirmatory factor analysis, path analysis and model optimization on the structural equation model to obtain the optimal model and its corresponding characteristic parameters, and finally further introduces the clear characteristic variables into the neural network model. The above process not only ensures that the key variables are not missing, but also removes the redundant variables;

[0024] 4) Based on historical data, the present invention uses characteristic latent variables as input variables of the LSTM model to obtain a nitrogen oxide emission concentration prediction model, and then generates prediction data through the nitrogen oxide emission concentration prediction model. The above provides favorable technical support for timely warning of nitrogen oxide concentration, improves the accuracy of nitrogen oxide concentration prediction, and can timely propose adjustment plans for relevant process parameters according to the nitrogen oxide concentration.

[0025] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0027] Figure 1 A flow chart of a method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace provided in Example 1 of the present invention;

[0028] Figure 2 This is a process schematic diagram of a method for predicting nitrogen oxide emission concentration in an ethylene cracking furnace according to Example 1 of the present invention;

[0029] Figure 3 This is a schematic diagram of the structural equation model for screening the nitrogen oxide concentration parameters at the outlet of the ethylene cracking furnace in Example 1 of the present invention;

[0030] Figure 4 This is a diagram for verifying the prediction accuracy of the nitrogen oxide concentration model at the outlet of the ethylene cracking furnace of Example 1 of the present invention. DETAILED DESCRIPTION

[0031] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0032] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0033] In the present invention, unless otherwise specified, directional words such as "up, down, top, bottom" are usually used with reference to the directions shown in the drawings or are used to describe the relative positional relationships of components in the vertical, perpendicular or gravity direction.

[0034] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0035] First Exemplary Embodiment

[0036] Please refer to Figure 1 The first exemplary embodiment of the present invention provides a method for predicting the concentration of nitrogen oxide emissions from an ethylene cracking furnace, the method comprising the following steps:

[0037] S101. Collect real-time operation data of the ethylene cracking furnace.

[0038] Optionally, the real-time operation data of the ethylene cracking furnace is collected from the cracking furnace area of ​​the 800,000 tons / year steam cracking unit of a petrochemical enterprise.

[0039] S102, pre-processing the real-time operation data, determining the key parameters of the emission concentration, and constructing the cracking furnace heat load composite variable.

[0040] In this example, the preprocessing of the real-time operation data may include: checking and eliminating outliers in the data; filling missing data based on the Lagrange interpolation method; resampling the filled data; using a single-sample KS test to determine whether the data conforms to the normal distribution; if so, standardizing the data; if not, logarithmically transforming the data that does not conform to the normal distribution, and standardizing the transformed data that conforms to the normal distribution.

[0041] In this example, the determination of the key parameters of emission concentration may include: performing an index importance analysis based on expert experience rules and hierarchical analysis, scoring the importance of parameters according to the expert diagnostic parameter list of nitrogen oxide emissions of ethylene units, and determining the key parameters of emission concentration according to the parameter importance level. If the result of determining the parameter importance score is greater than the first threshold, it is determined to be the first importance level; if the result of determining the parameter importance score is less than the first threshold and greater than the second threshold, it is determined to be the second importance level; if the result of determining the parameter importance score is less than the second threshold and greater than the third threshold, it is determined to be the third importance level; wherein the first threshold is greater than the second threshold, the second threshold is greater than the third threshold, and the key parameters of emission concentration include parameters of the first importance level, parameters of the second importance level, and parameters of the third importance level. For example, the key parameters of emission concentration include furnace temperature, furnace flue gas temperature, flue gas crossover temperature, ethylene cracking furnace outlet temperature, COT (temperature of the furnace tube outlet of the cracking furnace radiation section), furnace oxygen content, dilution steam flow, fan speed, and furnace negative pressure. For example, the first threshold may be 80 points, the second threshold may be 70 points, and the third threshold may be 60 points.

[0042] In this example, the construction of the cracking furnace heat load composite variable may include: obtaining temperature factor variables that affect emission concentration based on real-time operation data; combining the temperature factor variables that affect emission concentration to construct the cracking furnace heat load composite variable through the following formula;

[0043] Qhj =K R (F R1 +F R2 +……+F Rn )+K Y (F Y1 +F Y2 +……+F Ym );

[0044] In the formula, Q hj is the heat load equivalent of ethylene cracking furnace, kJ / hr; K R is the fuel combustion conversion coefficient, kJ / kg; K Y is the cracking exothermic conversion coefficient, kJ / kg; F Rn is the flow rate of each fuel in the furnace, kg / hr; F Ym is the flow rate of each raw material, kg / hr.

[0045] S103. Optimize the key parameters of emission concentration and the composite variables of cracking furnace heat load based on the SEM structural equation model to obtain optimized parameters.

[0046] In this example, the parameter optimization of the emission concentration key parameter and the cracking furnace heat load composite variable based on the SEM structural equation model to obtain the optimized parameters may include: determining the latent variables and the corresponding observed variables based on the emission concentration key parameter and the cracking furnace heat load composite variable for constructing the SEM structural equation model; wherein the SEM structural equation model includes a measurement model for calculating the relationship between the observed variables and the latent variables and a structural model for calculating the relationship between the latent variables; inputting the real-time operation data into the SEM structural equation model, calculating the standardized load coefficient value, and screening out the SEM structural equation model whose standardized load coefficient value meets the set coefficient range; by analyzing the chi-square degree of freedom ratio, goodness of fit index, root mean square error of approximation and Akaike information criterion of the SEM structural equation model whose standardized load coefficient value meets the set coefficient range, screening and determining the optimal model, and obtaining the optimized parameters of the optimal model.

[0047] S104, inputting the optimized parameters into a pre-built nitrogen oxide emission concentration prediction model for calculation to obtain prediction data.

[0048] In this example, the method may also include: constructing a nitrogen oxide emission concentration prediction model, which may include: collecting historical operating data and emission concentration data of the ethylene cracking furnace; preprocessing the historical operating data, determining key parameters of the emission concentration, and constructing a cracking furnace heat load composite variable; optimizing the key parameters of the emission concentration and the cracking furnace heat load composite variable based on the SEM structural equation model to obtain optimized parameters; inputting the optimized parameters and emission concentration data into the LSTM model for training to obtain a nitrogen oxide emission concentration prediction model.

[0049] Among them, the purpose of preprocessing the historical operation data is to test and remove outliers in the data; based on the Lagrange interpolation method, missing data is filled; the filled data is resampled; the single sample KS test is used to determine whether the data conforms to the normal distribution; if it conforms, the data is standardized; if it does not conform, the data that does not conform to the normal distribution is logarithmically transformed, and the data that conforms to the normal distribution after the transformation is standardized. The key parameters of emission concentration are determined to analyze the importance of indicators based on expert experience rules and hierarchical analysis method, and the parameter importance is scored according to the expert diagnosis parameter list of nitrogen oxide emissions of ethylene units, and the key parameters of emission concentration are determined according to the level of parameter importance. Here, if the result of determining the parameter importance score is greater than the first threshold, it is determined to be of the first importance; if the result of determining the parameter importance score is less than the first threshold and greater than the second threshold, it is determined to be of the second importance; if the result of determining the parameter importance score is less than the second threshold and greater than the third threshold, it is determined to be of the third importance; wherein, the first threshold is greater than the second threshold, the second threshold is greater than the third threshold, and the key parameters of the emission concentration include parameters of the first importance, parameters of the second importance, and parameters of the third importance. For example, the first threshold may be 80 points, the second threshold may be 70 points, and the third threshold may be 60 points. For example, the key parameters of the emission concentration include furnace temperature, furnace flue gas temperature, flue gas crossover temperature, ethylene cracking furnace outlet temperature, furnace oxygen content, dilution steam flow, fan speed, and furnace negative pressure.

[0050] Among them, the key parameters of emission concentration include primary characteristic parameters (also known as parameters of the first importance), secondary characteristic parameters (also known as parameters of the second importance) and control characteristic parameters (also known as parameters of the third importance), forming a characteristic parameter pre-selection scheme to provide reliable theoretical support for structural equation models and prediction models based on artificial intelligence algorithms. For example, key parameters can be subjected to Pearson correlation tests, and parameters with high correlation can be used as substitute parameters.

[0051] Specifically, the historical operation data of the relevant process parameters of the ethylene cracking furnace are collected, and the process parameters in the mechanism modeling process are referred to to form an expert diagnosis parameter list. According to the expert experience, the hierarchical analysis method is used to analyze the importance of indicators, and the parameter importance is scored according to the expert diagnosis parameter list of nitrogen oxide emissions of the ethylene unit, and the parameter importance list is formed using expert experience. Here, the hierarchical analysis method decomposes the problem into different components according to the nature of the problem and the overall goal to be achieved, and the factors are clustered and combined at different levels according to the mutual correlation and affiliation between the factors to form a multi-level analysis structure model, so that the problem is ultimately reduced to the determination of the relative importance weight of the lowest level (for example, the plan and measures for decision-making) relative to the highest level (for example, the overall goal) or the ranking of the relative order of superiority and inferiority.

[0052] The method of constructing the cracking furnace heat load composite variable is based on historical operation data to obtain temperature factor variables that affect emission concentration; the temperature factor variables that affect emission concentration are combined to construct the cracking furnace heat load composite variable through the following formula;

[0053] Q hj =K R (F R1 +F R2 +……+F Rn )+K Y (F Y1 +F Y2 +……+F Ym );

[0054] In the formula, Q hj is the heat load equivalent of ethylene cracking furnace, kJ / hr; K R is the fuel combustion conversion coefficient, kJ / kg; K Y is the cracking exothermic conversion coefficient, kJ / kg; F Rn is the flow rate of each fuel in the furnace, kg / hr; F Ym is the flow rate of each raw material, kg / hr.

[0055] Among them, considering that thermal nitrogen oxides are the main source of nitrogen oxide emissions from ethylene cracking furnaces, the factors that affect temperature are combined, and one variable can be used to show the impact of fuel and raw materials on nitrogen oxides. Among them, the cracking furnace heat load (HL) is a composite variable, which is calculated by combining the fuel flow rate of each furnace and the feed load of each furnace.

[0056] Specifically, the time series data of furnace temperature, furnace flue gas temperature, flue gas crossover temperature, cracking furnace outlet temperature, heat load (HL), furnace oxygen content, dilution steam flow, fan speed and furnace negative pressure that affect nitrogen oxide emissions are extracted from historical operation data. In the subsequent process, the above process parameters are screened for abnormal values ​​and difference processing, and data resampling is performed. The frequency of data acquisition is consistent, with a frequency of one number per minute. The data during normal working conditions is retained and the data during shutdown and furnace shutdown is eliminated. In this way, the correctness of the data is guaranteed and the data cleaning step is completed.

[0057] Here, the parameter optimization of the key parameters of emission concentration based on the SEM structural equation model to obtain the optimized parameters may include: based on the key parameters of emission concentration and the composite variable of cracking furnace heat load, determining the latent variables and the corresponding observed variables for constructing the SEM structural equation model; wherein the SEM structural equation model includes a measurement model for calculating the relationship between the observed variables and the latent variables and a structural model for calculating the relationship between the latent variables; inputting the historical operation data into the SEM structural equation model, calculating the standardized load coefficient value, and screening out the SEM structural equation model whose standardized load coefficient value meets the set coefficient range; by analyzing the chi-square degree of freedom ratio, goodness of fit index, root mean square error of approximation and Akaike information criterion of the SEM structural equation model whose standardized load coefficient value meets the set coefficient range, screening and determining the optimal model, and obtaining the optimized parameters of the optimal model.

[0058] Here, factor analysis needs to be satisfied in the SEM structural equation model; wherein, factor analysis includes data with a KMO value greater than 0.7 and a Bartlett P value less than 0.01. Here, factor analysis can further verify the data and better input the structural equation model for subsequent data. Before constructing the SEM structural equation model analysis, it is necessary to ensure that the measurement relationship has good quality, and to ensure the high quality of the measurement relationship through factor analysis. For example, the model sample takes 2000 data in the stable operation stage as the analysis data set to ensure the parameter estimation and successful convergence statistical analysis requirements of the structural equation model (SEM). In the present invention, the sample is a large sample and it is assumed that the observed data obeys the multivariate normal distribution. The chi-square test can be reasonably used, and the ML maximum likelihood estimation method is used as the parameter estimation method.

[0059] Specifically, the method also includes: inputting the data corresponding to the primary characteristic parameter into the structural equation model, then inputting the data corresponding to the secondary characteristic parameter into the structural equation model, and finally inputting the data corresponding to the control characteristic parameter into the structural equation model. Here, gradually incorporating the primary characteristic parameter, the secondary characteristic parameter and the control characteristic parameter can ensure that the important data that may exist is not lost, and the important data that may exist can also be added to the structural equation model to ensure that the characteristic parameters of the optimal model are obtained. Here, all the primary characteristic parameters determined by the early key parameter division are included in the path analysis model, and the secondary characteristic parameters and the control parameters are gradually included in the model in order, and multiple test models containing different secondary characteristic parameters and control parameters are established. Among them, the test model may include the detection of the structural model and the measurement model, such as checking the load coefficient value, whether the standardized load coefficient value is low (for example, less than 0.7), or whether there is a collinearity problem (in this case, the standardized load coefficient value will be greater than 1), and the unreasonable items are eliminated. For example, using the structural equation model (SEM) that combines factor analysis and path analysis multivariate statistical techniques, on the basis of sufficient theoretical support, the process parameter measurement variables related to furnace temperature, furnace flue gas temperature, cracking furnace outlet temperature and dilution steam flow rate are summarized as key latent variables, mainly divided into four latent variables: temperature, oxygen content, load and nitrogen oxide concentration, and the corresponding measurement variables are determined. Specifically, the composite variables of furnace temperature, furnace flue gas temperature, flue gas crossover temperature, ethylene cracking furnace outlet temperature and cracking furnace heat load are summarized as temperature, furnace negative pressure is summarized as pressure, and furnace oxygen content, dilution steam flow rate and fan speed are summarized as oxygen content.

[0060] Alternatively, the structural equation model overcomes the defect that the measured variables in the simultaneous equation model cannot have errors, and allows errors to exist in both the independent variables and the dependent variables. The measurement model calculates the relationship between the measured variables and the latent variables, and the structural model calculates the relationship between the latent variables.

[0061] The expression equation is as follows:

[0062] (1)X=Λ_xξ+δ;

[0063] In the formula, X is the observed variable; Λ_x is the relationship between the exogenous indicator and the exogenous latent variable; ξ is the exogenous latent variable; δ is the measurement error.

[0064] (2) Y = Λ_yη + ε;

[0065] In the formula, Y is the observed variable; Λ_y is the relationship between the endogenous indicator and the endogenous latent variable; η is the endogenous latent variable; ε is the measurement error.

[0066] The structural equation is as follows:

[0067] (3)η=Bη+Γξ+ζ

[0068] In the formula, η is the endogenous latent variable; B is the relationship between endogenous latent variables; Γ is the impact of exogenous latent variables on endogenous latent variables; ξ is the exogenous latent variable; ζ is the residual of the equation structure.

[0069] Here, the test equation is the equation of the relationship between the observed variables X, Y and the latent variables η, ξ.

[0070] In this example, the fit of each test model is analyzed and compared, for example, the chi-square degree of freedom ratio (Chi-square / df), the goodness of fit index (GFI), the root mean square error of approximation (RMSEA), and the Akaike information criterion (AIC). The comparison criteria for each test model are Chi-square / df, RMSEA, and AIC values. The smaller the better, while the larger the GFI value, the better. Here, the optimal model that meets the test requirements is selected from the established models, the construction of the structural equation model is completed, the relationship between the variables is clarified, and the factor load coefficient of the characteristic parameters for nitrogen oxide emissions is quantified. The clear characteristic latent variables are further introduced into the neural network model to ensure that the key variables are not missing while removing redundant variables.

[0071] Specifically, the neural network algorithm is mainly based on the Long Short-Term Memory (LSTM) neural network, which is a neural network improved on the basis of the recurrent neural network, making up for the defect that the recurrent neural network (RNN) cannot process long sequence data. The LSTM neural network controls the transmission of information by adding memory gates, learning gates, and forgetting gates. According to its network structure, its internal calculation formula is as follows:

[0072] (1)ft=σ(Wf·[ht-1,xt]+bf);

[0073] Where ft is the forget gate; Wf is the forget weight; ht-1 is the output of the unit at the previous moment; bf is the forget bias term.

[0074] (2) it = σ(Wi · [ht-1, xt] + bi)

[0075] In the formula, it is the input gate; Wi is the input weight; ht-1 is the output of the unit at the previous moment; bi is the input bias term.

[0076] (3)Ot=σ(WO·[ht-1,xt]+bO)

[0077] Where Ot is the output gate; WO is the current state weight; ht-1 is the output of the unit at the previous moment; bO is the current state bias term.

[0078] (4)Qt=tanh(WC·[ht-1,xt]+bC)

[0079] Where Qt is the state of the cell at the previous moment; ht-1 is the output of the cell at the previous moment; WC is the output weight; and bC is the output bias term.

[0080] (5) Ct = ft × Ct-1 + it × Qt

[0081] In the formula, ft is the forget gate, it is the input gate; Qt is the state of the cell at the previous moment, and Ct is the state of the cell at the current moment.

[0082] (6)ht=Ot×tanh(Ct)

[0083] Where Ot is the output gate; ht is the output of the current unit, and Ct is the state of the current cell.

[0084] Here, the characteristic latent variables of the optimization model obtained by SEM model debugging and optimization are input, and the output is the outlet NOx concentration. The training process mainly includes two processes: pre-training and fine-tuning. At the same time, it is judged whether the model error meets the accuracy requirements. If it meets the accuracy requirements, an automatic alarm will be issued when the model is about to exceed the standard.

[0085] Optionally, an initial prediction model is established according to the process. In the later operation process, the LSTM model needs to be continuously trained according to the operation data, so that the obtained model has the characteristics of self-learning and self-adaptation. The predicted value of nitrogen oxide concentration is calculated based on the trained neural network model according to the operation data and verified using historical data.

[0086] In this example, the method may further include: determining whether the predicted data exceeds a safety threshold; if so, issuing an early warning. Here, the safety threshold may be 65 mg / m 3 , 70mg / m 3 , 75mg / m 3 Specifically, the scope of change of key characteristic parameters during the stable operation of the device is clarified, the safety threshold of the parameters is clarified, and an early warning is issued when the characteristic parameters exceed the safety threshold, providing the first guarantee for the compliance of nitrogen oxide emissions in the ethylene cracking furnace.

[0087] In this example, the method may also include: if it is determined that the predicted data exceeds the safety threshold, the optimization parameter adjustment plan is calculated by the nitrogen oxide emission concentration prediction model. Specifically, the prediction model can automatically alarm when the nitrogen oxide concentration in the ethylene cracking furnace is about to exceed the standard, and automatically calculate a set of relevant characteristic parameter adjustment plans that control the concentration within the limit, and provide this adjustment plan to the on-site operators to remind the relevant personnel that the nitrogen oxide concentration is about to exceed the standard and the process parameters that should be adjusted. Here, the control plan can achieve all-round accurate monitoring and timely emission reduction based on the optimal operation adjustment parameters obtained by the real-time cracking furnace device operation intelligent computing system. Pollutant control is completed based on the prediction of the nitrogen oxide concentration in the ethylene cracking furnace until the nitrogen oxide content is maintained within a reasonable range and the flue gas of the ethylene cracking furnace is guaranteed to meet the emission standards and be stable.

[0088] Second Exemplary Embodiment

[0089] The second exemplary embodiment of the present invention provides a device for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace, and the device comprises: a data acquisition module for collecting real-time operation data of the ethylene cracking furnace; a data processing module for preprocessing the real-time operation data, determining key parameters of the emission concentration, and constructing a composite variable of the thermal load of the cracking furnace; a parameter optimization module for performing parameter optimization on the key parameters and composite variables of the emission concentration based on the SEM structural equation model to obtain optimized parameters; and a data prediction module for inputting the optimized parameters into a pre-constructed nitrogen oxide emission concentration prediction model for calculation to obtain predicted data.

[0090] Third Exemplary Embodiment

[0091] A third exemplary embodiment of the present invention provides a computer device, comprising: at least one processor; a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions include instructions for executing the above-mentioned method for predicting nitrogen oxide emission concentration of an ethylene cracking furnace.

[0092] Fourth Exemplary Embodiment

[0093] A fourth exemplary embodiment of the present invention provides a computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement the above-mentioned method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace.

[0094] The computer-readable storage medium may be any data storage device that stores data that can be read by a computer system. For example, examples of computer-readable storage media may include: read-only memory, random access memory, read-only optical disk, magnetic tape, floppy disk, optical data storage device, and carrier wave (such as data transmission via the Internet via a wired or wireless transmission path).

[0095] In order to better understand the present invention, the content of the present invention is further explained below with reference to the accompanying drawings and examples, but the content of the present invention is not limited to the following embodiments.

[0096] Example 1

[0097] This example is a method for predicting the NOx concentration at the cracking furnace outlet using the ethylene cracking furnace flue gas compliance control technology. Figure 2 As shown, the outlet flue gas NOx concentration under the current working conditions is predicted. The specific steps are as follows:

[0098] Step 101: Take the relevant parameters of the mechanism model as a reference, summarize them according to the actual situation of the selected refining and petrochemical ethylene unit, and form an expert diagnosis parameter list. According to the expert experience, the hierarchical analysis method is used to analyze the importance of indicators, and 22 variables (i.e., key parameters of emission concentration) are selected as the factors affecting NO X Model input for concentration.

[0099] Step 102: Data collection is from the cracking furnace area of ​​a petrochemical enterprise's 800,000 tons / year steam cracking unit, which uses a low-nitrogen burner and the raw materials mainly include naphtha, light naphtha, circulating ethane and liquefied gas. Data from the distributed control system (DCS) of the unit for one month of normal operation is selected for analysis, and the data sampling period is 60 seconds.

[0100] Step 103: The module collects key process parameter data related to the nitrogen oxide concentration of the ethylene cracking furnace within a certain period of time (for example, one month, two months, three months); it should be noted that the process parameters here refer to process parameters related to the prediction of amine solution concentration, which may include ethylene cracking furnace outlet temperature COT, furnace oxygen content (2 groups), furnace negative pressure (2 groups), fan speed, dilution steam (8 groups), furnace bottom negative pressure (2 groups), cross section temperature (2 groups), exhaust temperature (2 groups), cracking furnace heat load, nitrogen reduction steam flow (2 groups). The above 22 process parameters are highly correlated with the nitrogen oxide concentration at the flue gas outlet of the ethylene cracking furnace.

[0101] Step 104: The data processing module pre-processes the collected parameter data, including screening outliers, specifically extracting data from normal operating conditions, eliminating data from shutdown states, and calculating thresholds for all process parameters.

[0102] Step 105: preprocess the parameter data, including difference processing. Specifically, there are a large number of null values ​​in the fuel gas, furnace negative pressure, furnace oxygen content and fan speed data, which will affect the results during data analysis. The missing data values ​​need to be eliminated or supplemented. When there are a large number of null values ​​in the fuel gas, furnace negative pressure, furnace oxygen content and fan speed data, elimination processing is adopted.

[0103] Step 106: There are a few missing values ​​in the dilution steam and feed amount. The Lagrange interpolation method is used, and the previous and next numbers are used as the difference basis. Generally, if the function values ​​at n+1 different points are known (that is, the function passes through these n+1 points), it is possible to consider constructing a polynomial that passes through these n+1 points and has a degree of no more than n, so that it satisfies:

[0104] P n (x k )=y k ,k=0,1,…,n;

[0105] To estimate any point ξ, ξ≠xi, i=0,1,2,...,n, then the value of Pn(ξ) can be used as an approximation of the exact value f(ξ).

[0106] Step 107: Preprocessing the parameter data also includes resampling, specifically using downsampling to reduce the data transmission rate or data size. According to the current data frequency and prediction accuracy requirements of nitrogen oxide concentration and its related parameters, the frequency of the downsampled data is unified to 1 data per minute. According to data analysis, it is found that the data fluctuations per minute are not obvious. By reducing the data sampling rate and setting a data retention strategy that gradually extends over time, it can ensure that the real-time data has highly operational and high-precision characteristics, while also ensuring long-term data trends and analysis. Here, steps 104 to 107 are data cleaning steps.

[0107] Step 108: Establish a cracking furnace heat load (HL) composite variable, calculate the cracking furnace heat load as a composite variable, and combine the fuel flow rate of each furnace and the feeding load of each furnace to calculate the calculation formula as follows:

[0108] Q hj =K R (F R1 +F R2 +……+F Rn )+K Y (F Y1 +F Y2 +……+F Ym )

[0109] In the formula, Q hj is the heat load equivalent of the cracking furnace, K Rand K Y are the heat conversion coefficients for fuel combustion and cracking, respectively, and F Rn is the flow rate of each fuel in the furnace, F Ym is the flow rate of each raw material.

[0110] Step 109: Use the KS test to check whether the data conforms to the normal distribution, and perform logarithmic transformation on the data that does not conform to the normal distribution, analyze the normal distribution of the data sample again, and use the Z-Score standardization method to standardize all parameters to eliminate the impact of the dimension and range of each influencing factor in the ethylene unit system.

[0111] Step 110: Incorporate historical data into the SEM model for modeling analysis. In the present invention, the latent variables are temperature, pressure, and oxygen content. The observed variable X is 38 groups of variables, including ethylene cracking furnace outlet temperature COT, furnace oxygen content (2 groups of characteristic parameters), furnace negative pressure (3 groups), fan speed, dilution steam (8 groups), furnace bottom negative pressure (2 groups), cross section temperature (2 groups), exhaust temperature (2 groups), cracking furnace heat load, nitrogen reduction steam flow (2 groups), air door opening, cross section temperature (2 groups), cross section flue gas negative pressure (3 groups), etc. The observed variable Y is the cracking furnace outlet NO X By observing the size and significance of the path coefficient of the structural model, we can know the influence path of each factor on the production of nitrogen oxides in the cracking furnace.

[0112] Step 111: First, we need to determine whether the data meets the requirements of factor analysis. The results show that the KMO value of the survey data is 0.87>0.7 (standard), the Bartlett sphericity test value is 3785.538, which is relatively large, and the Bartlett P value is 0.000<0.01, indicating that the data is suitable for factor analysis. Figure 3 By building models with different structures and comparing reliability and validity tests (i.e., the size and significance of the path coefficient), the rationality and reliability of the selected observation variable indicators are verified, and 12 groups of parameters (also known as preferred variables) are selected to be included in the next step of LSTM prediction modeling. Table 1 shows the 12 groups of preferred parameters as follows:

[0113]

[0114]

[0115] Table 1 Optimized parameters

[0116] AMOS output index chi-square ratio to degrees of freedom χ 2 / df=2.674<3, indicating that the model fits well; Goodness of Fit Index (GFI)=0.918, Basic Fit Index (NFI)=0.923, and Comparative Fit Index (CFI)=0.943. The closer the value is to 1, the better the model fit is. Here, steps 110 to 111 are for optimizing parameters in the structural equation model.

[0117] Step 112: Select Ethylene Cracking Furnace NO X The concentration was taken as the prediction object, and the selected influencing factor sequence included 13 parameters (i.e., optimal parameters) such as cracking furnace heat load, fan speed, and furnace oxygen content that were optimized by the SEM structural equation model and were incorporated into the prediction model. X The sequence is merged with the selected influencing factor sequence, and the training set and test set are divided into 8:2 ratios. The data is input into the LSTM model for training. The model time step is 30, the input is a 14-dimensional vector, the number of hidden layers is 2, and the hyperparameter tuning result is {'batch_size':8,'epochs':10,'optimizer':'adam'}. Here, step 112 is to establish the LSTM model.

[0118] Step 113: Ethylene cracking furnace NO optimization by SEM parameters X The predicted concentration had an average RMSE of 0.338 and an average R 2 is 0.897, such as Figure 4 As shown, Figure 4 The darker colors are predicted values, and the lighter colors are true values. Since the input of the model parameter optimization contains more accurate information, it is helpful to better reconstruct the NO of the cracking furnace. X Generate complex systems with improved predictability.

[0119] In summary, the LSTM artificial neural network model proposed in this patent based on the cracking furnace heat load and SEM parameter optimization is used to predict the NO X The concentration prediction accuracy is higher.

[0120] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.

[0121] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

[0122] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A method for predicting the concentration of nitrogen oxide emissions from an ethylene cracking furnace, characterized in that: The method comprises: Collect real-time operation data of ethylene cracking furnace; Preprocess the real-time operation data, determine the key parameters of emission concentration, and construct the composite variable of cracking furnace heat load; Based on the SEM structural equation model, the key parameters of emission concentration and the composite variables of cracking furnace heat load were optimized to obtain the optimized parameters; The optimized parameters are input into the pre-built nitrogen oxide emission concentration prediction model for calculation to obtain the prediction data.

2. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The method further includes: constructing a nitrogen oxide emission concentration prediction model, including: Collect historical operation data and emission concentration data of ethylene cracking furnace; Preprocess historical operation data, determine key parameters of emission concentration, and construct composite variables of cracking furnace heat load; Based on the SEM structural equation model, the key parameters of emission concentration and the composite variables of cracking furnace heat load were optimized to obtain the optimized parameters; The optimization parameters and emission concentration data are input into the LSTM model for training to obtain the nitrogen oxide emission concentration prediction model.

3. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The preprocessing of the real-time operation data includes: Detect and remove outliers in the data; Fill missing data based on Lagrange interpolation method; Resample the padded data; The one-sample KS test was used to determine whether the data conformed to the normal distribution; If it meets the requirements, the data will be standardized; If not, the data that do not conform to the normal distribution are logarithmically transformed, and the data that conform to the normal distribution after transformation are standardized.

4. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The key parameters for determining emission concentration include: The importance of indicators is analyzed based on expert experience rules and the hierarchical analysis method. The importance of parameters is scored according to the expert diagnosis parameter list of nitrogen oxide emissions from ethylene plants. The key parameters of emission concentration are determined according to the level of parameter importance.

5. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The SEM structural equation model is used to optimize the key parameters of emission concentration and the composite variable of cracking furnace heat load to obtain the optimized parameters, including: Based on the key parameters of emission concentration and the composite variable of cracking furnace heat load, latent variables and corresponding observed variables are determined to construct a SEM structural equation model; wherein the SEM structural equation model includes a measurement model for calculating the relationship between the observed variables and the latent variables and a structural model for calculating the relationship between the latent variables; Input the historical operation data into the SEM structural equation model, calculate the standardized load coefficient value, and select the SEM structural equation model whose standardized load coefficient value meets the set coefficient range; By analyzing the chi-square freedom ratio, goodness of fit index, root mean square error of approximation and Akaike information criterion of the SEM structural equation model whose standardized load coefficient value meets the set coefficient range, the optimal model is screened and determined, and the optimization parameters of the optimal model are obtained.

6. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The method of constructing a cracking furnace heat load composite variable includes: Based on real-time operation data, obtain temperature factor variables that affect emission concentration; The temperature factor variables that affect the emission concentration are combined to construct a composite variable of the cracking furnace heat load.

7. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 6, characterized in that: The cracking furnace heat load composite variable is obtained by formula (1); Q hj =K R (F R1 +F R2 +……+F Rn )+K Y (F Y1 +F Y2 +……+F Ym ) (1); In formula (1), Q hj K is the heat load equivalent of ethylene cracking furnace; R is the fuel combustion conversion coefficient; K Y is the cracking exothermic conversion coefficient; F Rn is the flow rate of each fuel in the furnace; F Ym is the flow rate of each raw material.

8. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 1, characterized in that: The method further comprises: Determine whether the predicted data exceeds the safety threshold; if it exceeds the safety threshold, issue an early warning.

9. The method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to claim 8, characterized in that: The method further comprises: If it is determined that the predicted data exceeds the safety threshold, the optimization parameter adjustment plan is calculated through the nitrogen oxide emission concentration prediction model.

10. A device for predicting the concentration of nitrogen oxide emissions from an ethylene cracking furnace, characterized in that: The device for predicting the concentration of nitrogen oxides emitted by an ethylene cracking furnace comprises: Data acquisition module, used to collect real-time operation data of ethylene cracking furnace; The data processing module is used to pre-process the real-time operation data, determine the key parameters of emission concentration, and construct the composite variables of cracking furnace heat load; Parameter optimization module, used to optimize the key parameters and composite variables of emission concentration based on the SEM structural equation model to obtain optimized parameters; The data prediction module is used to input the optimization parameters into the pre-built nitrogen oxide emission concentration prediction model for calculation to obtain prediction data.

11. A computer device, characterized in that: include: at least one processor; A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, and the program instructions include instructions for executing the method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to any one of claims 1-9.

12. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method for predicting the nitrogen oxide emission concentration of an ethylene cracking furnace according to any one of claims 1 to 9 is implemented.

Images