Method, system and equipment for analyzing influence factors of preparation of titanium tetrachloride by boiling chlorination

Abstract

The invention discloses an influence factor analysis method, system and equipment for preparation of titanium tetrachloride through boiling chlorination, and the method comprises the steps: obtaining the concentration of titanium tetrachloride, the concentration of byproducts and the time series data of alternative influence factors, and carrying out the preprocessing to obtain a variable data set and a variable derivative matrix of the preparation of titanium tetrachloride through boiling chlorination; selecting a sparse function base of the sparse identification model, adding exponential function terms to cooperatively form a composite sparse function base, and calculating a coefficient matrix in the sparse identification model by adopting a sequential threshold least square method; non-zero items in the coefficient matrix are extracted, and the influence relation between the titanium tetrachloride concentration and the influence factors and / or the influence relation between the by-product concentration and the influence factors are / is determined through the mapping relation. According to the method, the rate control mechanism and the impurity influence item in the high-temperature reaction system can be described more accurately, the ability of describing the nonlinear change trend is enhanced, and the distinguishing performance of the dominant reaction path and the interference mechanism is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of chlorination method titanium dioxide preparation, and particularly relates to a boiling chlorination method for preparing titanium tetrachloride, a system and equipment for analyzing influencing factors. BACKGROUND

[0002] Titanium dioxide is the most important white inorganic pigment in the world, and the preparation process of titanium tetrachloride, the core raw material of titanium dioxide, directly affects the product quality and production cost. The boiling chlorination method has become the mainstream process for industrial production of titanium tetrachloride due to its high production efficiency, low energy consumption, and environmental friendliness. However, the boiling chlorination reaction process involves high temperature, multiphase flow, complex mass and heat transfer, and multi-component chemical reactions (such as the formation of titanium tetrachloride, carbon monoxide, carbon dioxide, and ferrous chloride byproducts), and its dynamic behavior has strong nonlinearity and multivariate coupling characteristics. Under this background, the existing reaction analysis methods for the boiling chlorination production process still have deficiencies: a) the influencing factors of the products in the boiling chlorination production reaction process are not clear; b) the influencing factors of the formation of impurities (such as ferrous chloride) in the boiling chlorination production reaction process are not clear. Traditional modeling techniques mainly start from the gas-solid phase in the boiling chlorination process to describe the changes of various substances in the reaction process, consider the solid particles as pseudo-fluid, use the continuity equation and momentum equation of the gas phase and pseudo-fluid phase to describe the relevant motion parameters in the gas-solid two-phase flow process, and finally use the interaction force between the gas and solid phases to couple and close the equations. In recent years, data-driven modeling techniques have shown significant advantages in chemical process optimization, but the existing machine learning methods have the characteristics of "black box" and the risk of overfitting of high-dimensional data, making it difficult to achieve interpretable variable selection and mechanism correlation in industrial scenarios.

[0003] The "optimal influencing factor evaluation method and system for preparing C4 olefins coupled with ethanol" disclosed in Chinese patent literature, with publication number CN114187973A and publication date of 20220315, includes obtaining all influencing factor data for preparing C4 olefins coupled with ethanol; performing correlation analysis on each influencing factor and C4 olefin yield, and selecting a preset number of influencing factors as regression fitting factors according to the number of correlation from large to small; and obtaining the corresponding catalytic combination and temperature at which the C4 olefin yield is highest under temperature constraints based on the optimal multiple quadratic regression model obtained based on the regression fitting factors. This technology only analyzes the process of preparing C4 olefins coupled with ethanol, and is not applicable to the reaction process of preparing titanium tetrachloride by boiling chlorination, and still lacks an identification analysis method that can find influencing factors related to the yield of titanium tetrachloride and the formation of impurities in actual production processes. SUMMARY

[0004] The present application is to overcome the problem that the influencing factors of product and impurity formation in the process of producing titanium tetrachloride by boiling chlorination are not clear in the prior art, and provides a boiling chlorination titanium tetrachloride influencing factor analysis method, system and equipment.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: A boiling chlorination titanium tetrachloride influencing factor analysis method, comprising: Obtain the time series data of titanium tetrachloride concentration, by-product concentration and alternative influencing factors, and obtain the variable data set and variable derivative matrix of boiling chlorination titanium tetrachloride preparation by preprocessing; Select the sparse function base of the sparse identification model, and add the exponential function item to cooperatively constitute the composite sparse function base, and calculate the coefficient matrix in the sparse identification model by using the sequential threshold least square method; Extract the non-zero items in the coefficient matrix, and determine the influence relationship between the titanium tetrachloride concentration and the influencing factors and / or the influence relationship between the by-product concentration and the influencing factors through the mapping relationship.

[0006] In the present application, the relevant data of the production process in different time periods are obtained as the boiling chlorination titanium tetrachloride data set for sparse identification; the sparse function base is constructed in combination with the boiling chlorination model knowledge; then the sparse model is solved by using the sequential threshold least square method; then the non-zero items in the coefficient matrix are extracted, and the dominant influence relationship between the current variable and other variables is determined through the mapping correlation, and the nonlinear coupling relationship and dynamic response characteristics between the variables are analyzed according to the function base item corresponding to the non-zero coefficient; finally, the model accuracy is determined by using the cross-validation method; thereby the rate control mechanism and impurity influence item in the high-temperature reaction system are more accurately described, the description ability of the nonlinear change trend is enhanced, and the resolution performance of the dominant reaction path and interference mechanism is improved.

[0007] As a preferred, the alternative influencing factors at least include the temperature field distribution of the boiling chlorination reactor, the chlorine flow, the raw material proportioning and the pressure fluctuation; The preprocessing of the time series data includes: using the linear interpolation method to complete the missing data items, and ensuring that all variables have consistent time steps.

[0008] As a preferred, the preprocessing to obtain the variable derivative matrix includes: The variable data set is standardized and combined to obtain the variable matrix; the number of columns in the variable matrix is the number of variables, the number of rows in the variable matrix is the number of elements in the column vector corresponding to each variable, which represents the number of time series data collected by the variable; for each variable in the variable matrix, the time derivative is calculated by using the finite difference method to obtain the variable derivative matrix.

[0009] Preferably, the process of constructing the composite sparse function base and calculating the coefficient matrix comprises: The composite sparse function base is constructed by adding exponential function terms to a polynomial function base, the polynomial function base including at least first-order function terms of each variable and cross terms of multiplication of each variable with each other variable; The variable derivative matrix obtained by preprocessing is introduced into the sparse identification model as an input term, and the coefficient matrix in the sparse identification model is calculated by using sequential threshold least squares.

[0010] Preferably, the sparse identification model is an equation, the left side of the equation being the variable derivative matrix, and the right side of the equation being a matrix product between the composite sparse function base in the form of a matrix and the coefficient matrix.

[0011] Preferably, the process of calculating the coefficient matrix comprises: A sparse threshold λ is preset, an initial coefficient matrix is solved based on least squares, coefficients with absolute values less than λ in the current coefficient matrix are set to zero, non-zero terms of the coefficients are retained, and an updated coefficient matrix is solved by least squares again, and iterative calculation is repeatedly performed until a final coefficient matrix is obtained by satisfying an iteration stopping condition.

[0012] Preferably, the iteration stopping condition comprises: The update amplitude of the coefficient matrix is less than a preset convergence threshold, or the number of iterations reaches a preset maximum iteration threshold.

[0013] Preferably, the numerical distribution characteristics of each column vector in the coefficient matrix are analyzed, based on the mathematical expression form of the composite sparse function base, according to the function base terms corresponding to the non-zero coefficients, the dominant influence relationship between the titanium tetrachloride concentration variable and the influence factor variable and / or the dominant influence relationship between the byproduct concentration variable and the influence factor variable are determined by mapping association.

[0014] A system for analyzing influence factors of titanium tetrachloride produced by boiling chlorination, comprising: A data processing module acquires data of each variable in the process of producing titanium tetrachloride by boiling chlorination, and preprocesses the data to obtain a variable data set and a variable derivative matrix; A function base construction module selects a sparse function base of a sparse identification model, and adds exponential function terms to cooperatively form a composite sparse function base; An influence factor analysis module solves a coefficient matrix in the sparse identification model, and determines a dominant influence relationship between variables based on non-zero terms in the coefficient matrix.

[0015] An electronic device comprising a memory for storing program data and a processor for executing the program data to implement the above-mentioned method for analyzing influence factors of titanium tetrachloride produced by boiling chlorination.

[0016] The present application has the following beneficial effects: obtaining relevant data of the production process in different time periods as a sparse recognition boiling chlorination titanium tetrachloride dataset; combining boiling chlorination model knowledge to construct a sparse function base; then solving the sparse model by sequential threshold least squares method; then extracting the non-zero items in the coefficient matrix, determining the dominant influence relationship between the current variables and other variables through the mapping relationship, analyzing the nonlinear coupling relationship and dynamic response characteristics between the variables according to the function base items corresponding to the non-zero coefficients; finally, the cross-validation method is used to determine the accuracy of the model; thereby more accurately depicting the rate control mechanism and impurity influence term in the high-temperature reaction system, enhancing the ability to depict the nonlinear change trend, and improving the resolution performance of the dominant reaction path and interference mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flowchart of the boiling chlorination titanium tetrachloride influencing factor analysis method in the present application.

[0018] Figure 2 is a schematic diagram of the boiling chlorination titanium tetrachloride influencing factor analysis system in the present application. DETAILED DESCRIPTION

[0019] The present application will be further described below in conjunction with the drawings and specific embodiments.

[0020] As shown in Figure 1 , a boiling chlorination titanium tetrachloride influencing factor analysis method, comprising: obtaining time series data of titanium tetrachloride concentration, byproduct concentration and alternative influencing factors, and pre-processing to obtain a variable dataset and a variable derivative matrix of boiling chlorination titanium tetrachloride; selecting a sparse function base of the sparse recognition model, and adding an exponential function item to cooperatively constitute a composite sparse function base, and using a sequential threshold least squares method to calculate the coefficient matrix in the sparse recognition model; extracting the non-zero items in the coefficient matrix, and determining the influence relationship between the titanium tetrachloride concentration and the influencing factors and / or the influence relationship between the byproduct concentration and the influencing factors through the mapping relationship.

[0021] It should be noted that the related data of the production process in different time periods in the present application is obtained as a sparse identification data set of the boiling chlorination preparation of titanium tetrachloride; the sparse function base is constructed in combination with the boiling chlorination model knowledge; then the sparse model is solved by using the sequential threshold least square method; then the non-zero items in the coefficient matrix are extracted, the dominant influence relationship between the current variable and other variables is determined through mapping association, the nonlinear coupling relationship and dynamic response characteristics between the variables are analyzed according to the function base items corresponding to the non-zero coefficients; finally, the model accuracy is determined by using the cross-validation method; so as to more accurately depict the rate control mechanism and impurity influence term in the high-temperature reaction system, enhance the description ability of the nonlinear change trend, and improve the resolution performance of the dominant reaction path and interference mechanism, and find the decisive physical quantity in the reaction process.

[0022] Specifically, the following steps are included: S1: real-time acquisition of the temperature field distribution, chlorine flow, raw material ratio, and pressure fluctuation time series data in the boiling chlorination reactor by industrial sensors, collection of titanium tetrachloride concentration, byproduct generation amount (such as carbon monoxide, carbon dioxide and ferrous chloride concentration) as a boiling chlorination preparation of titanium tetrachloride data set, for the missing items of the collected data set, time series linear interpolation is used for data completion to ensure uniform time step; S2: selecting the sparse function base required for sparse identification, adding exponential function items to the polynomial function base according to the expert experience of the boiling chlorination process, so as to meet the sudden rise and sudden drop of variables in the time period before and after the chemical reaction equilibrium, inputting the data obtained in step S1 into the sparse identification model, and using the sequential threshold least square method to calculate the coefficient matrix in the sparse identification model; S3: extracting the non-zero items in the coefficient matrix obtained in step S2, determining the dominant influence relationship between the product variable (titanium tetrachloride or other byproducts) and the influence factor variable through mapping association, and verifying the accuracy of the model by using the cross-validation method.

[0023] It is worth noting that the sparse identification method can automatically select the dominant action item from complex high-dimensional observation data by constructing a function item set containing process mechanism prior (i.e. sparse function base) and combining sparse regression algorithm (such as threshold least square method), so as to extract mathematical expressions with physical meaning and simple structure. In the traditional technology, the sparse function base is usually constructed by low-order polynomials and cross terms. However, in the preparation process of titanium tetrachloride with gas-solid phase reaction as the core, there are obvious nonlinear, complex rate control mechanism, activation energy sensitivity and other kinetic characteristics, and the above-mentioned traditional sparse function base is difficult to cover the intrinsic behavior of the reaction.

[0024] To this end, the present application is directed to the characteristics of temperature sensitivity and exponential rate response of the boiling chlorination reaction in the preparation of titanium tetrachloride, innovatively modifies the sparse function base structure under the framework of sparse identification, adds an exponential function term to more accurately depict the rate control mechanism and impurity influence term in the high-temperature reaction system. The composite sparse function base composed of the exponential function term and the traditional term enhances the ability to depict the nonlinear change trend and improves the resolution performance of the dominant reaction path and interference mechanism.

[0025] In addition to providing a boiling chlorination titanium tetrachloride influencing factor analysis method, the present application also provides a boiling chlorination titanium tetrachloride influencing factor analysis system as shown in Figure 2 The system comprises: A data processing module acquires time series data of various variables in the boiling chlorination titanium tetrachloride preparation process and pre-processes to obtain corresponding variable data sets and variable derivative matrices; A function base construction module selects a sparse function base of the sparse identification model and adds an exponential function term to form a composite sparse function base; An influencing factor analysis module solves the coefficient matrix in the sparse identification model and determines the dominant influence relationship between variables based on the non-zero terms in the coefficient matrix.

[0026] It should be noted that the data processing module includes several sensor units, each of which is used for collecting variable data in the boiling chlorination titanium tetrachloride preparation process, and each sensor unit corresponds to the collection of a variable. The data processing module also includes a preprocessing unit that differentially completes and integrates the collected time series data to obtain variable data sets, and further performs time derivation processing on the variable data sets to obtain variable derivative matrices.

[0027] The function base construction module includes a model construction unit of the sparse identification model to determine the sparse identification model for the boiling chlorination titanium tetrachloride reaction process of the present application. It also includes a function base composite unit that selects a sparse function base in the sparse identification model and adds an exponential function term to form a composite sparse function base.

[0028] The influencing factor analysis module includes a coefficient matrix solving unit and an influence relationship analysis unit. The coefficient matrix solving unit inputs the obtained variable derivative matrix into the sparse identification model based on the obtained variable derivative matrix, and calculates the coefficient matrix in the sparse identification model using the sequential threshold least squares method. The influence relationship analysis unit extracts the non-zero coefficient terms from the relationship matrix obtained by solving, and determines the influence relationship between the titanium tetrachloride concentration and the influencing factors and / or the influence relationship between the byproduct concentration and the influencing factors through the non-zero coefficient terms and the corresponding function base expression.

[0029] Further, the present application also provides an electronic device comprising a memory for storing program data and a processor for executing the program data to realize the above-mentioned boiling chlorination titanium tetrachloride influencing factor analysis method. The electronic device of the present application can have the functions and corresponding hardware of the existing conventional electronic device in addition to the processor, the memory and the data interface, and therefore no detailed description is given.

[0030] As a specific embodiment, the alternative influencing factors at least include the temperature field distribution of the boiling chlorination reactor, the chlorine flow rate, the raw material ratio and the pressure fluctuation; The preprocessing of the time series data includes: using linear interpolation method to complete the missing data, ensuring that all variables have consistent time steps.

[0031] Specifically, the time series data of the temperature field distribution of the boiling chlorination reactor, the chlorine flow rate, the raw material ratio and the pressure fluctuation are obtained in real time by industrial sensors, and the titanium tetrachloride concentration, the byproduct concentration (carbon monoxide concentration, carbon dioxide concentration and ferrous chloride concentration) are collected as the boiling chlorination titanium tetrachloride dataset. For the missing items of the collected dataset, linear interpolation of time series is used for data completion to ensure uniform time steps.

[0032] In this embodiment, first, real-time data of the boiling chlorination titanium tetrachloride reactor are collected by various sensor units. The required collected variable modes include: measuring the temperature field in the reactor by thermocouple or infrared sensor and recording its change over time to obtain time series data of the temperature field distribution; using mass flow meters and other devices to collect real-time chlorine flow data to obtain time series data of the chlorine flow rate; recording the inflow ratio of raw materials by flow meters or mass sensors to obtain time series data of the raw material ratio; using pressure sensors to record the pressure fluctuation time series data in the reactor; and using chemical analysis instruments or sampling and analysis techniques to measure the concentration data of the generated titanium tetrachloride and byproducts (carbon monoxide, carbon dioxide and ferrous chloride) in real time. Considering the data missing problem in actual industrial processes, the present application uses linear interpolation method to complete the missing data, ensuring that all variables have consistent time steps for subsequent mathematical analysis.

[0033] Further, the preprocessing obtains the variable derivative matrix, which includes: The variable dataset is standardized and combined to obtain a variable matrix. The number of columns in the variable matrix is the number of variables, and the number of rows in the variable matrix is the number of elements in the column vector corresponding to each variable, indicating the number of time series data collected by the variable. For each variable in the variable matrix, the time derivative is calculated by the finite difference method to obtain the variable derivative matrix.

[0034] Specifically, the variable data set after data completion includes the concentration of titanium tetrachloride, the concentration of by-products (carbon monoxide, carbon dioxide and ferrous chloride), the temperature field distribution, the chlorine flow, the raw material ratio and the time sequence data of pressure fluctuation, and after standardization processing, the corresponding variable matrix [X1, X2, …, X i , …, X m ] is obtained by combining each variable data. i Each X i1 in the matrix corresponds to a column vector [x i2 , x ij , …, x in ]. X ij represents the jth time sequence data of the ith variable, m represents the number of variables, and n represents the number of time sequence data collected by the variable.

[0035] Taking the concentration of titanium tetrachloride, the concentration of carbon monoxide, the concentration of carbon dioxide, the concentration of ferrous chloride, the temperature field distribution, the chlorine flow, the raw material ratio and the pressure fluctuation in the embodiment as an example, an 8-column n-row variable matrix can be obtained. The time derivative is calculated by the finite difference method which guarantees the dimension unchanged (for the same variable, the difference between the next time sequence data and the previous time sequence data is divided by the sampling interval time of the time sequence data), and then a variable derivative matrix for modeling analysis is formed to describe the change rate of each variable over time.

[0036] As a specific embodiment, the process of constructing a composite sparse function base and calculating a coefficient matrix includes: selecting a sparse function base of a sparse identification model, adding an exponential function term to form a composite sparse function base, and calculating a coefficient matrix in the sparse identification model by using a sequential threshold least squares method.

[0037] Specifically, the composite sparse function base is formed by adding an exponential function term based on a polynomial function base, and the polynomial function base at least includes a first function term of each variable and a cross term of multiplication between each variable; The variable derivative matrix obtained by preprocessing is introduced into the sparse identification model as an input term, and a coefficient matrix in the sparse identification model is calculated by using a sequential threshold least squares method.

[0038] It should be noted that in the process of constructing the function base, the composite sparse function base based on the polynomial function and supplemented with the exponential function term is selected to enhance the fitting ability of the model to the nonlinear relationship; on the basis of the polynomial function base, the exponential function term is supplemented according to the expert experience of the boiling chlorination process, so as to conform to the sudden rise and sudden drop of the variable in the time period before and after the chemical reaction equilibrium, and enhance the nonlinear fitting ability of the model.

[0039] In addition to the first order function term of each variable and the cross term of the multiplication between any two variables, the polynomial function base can also include various low-order polynomials such as the second order function term of each variable. The added exponential function term refers to the exponential term of X i to the power of e, because X i is actually a column vector containing n elements, so the exponential term of X i to the power of e is also a column vector with n elements, and the jth element is the exponential term of X ij to the power of e.

[0040] The finally constructed composite sparse function base contains the first order function term, the second order function term, the cross term and the exponential function term of the variable, which can be expressed as where the second order function term may be a column vector obtained by squaring the same position elements in the column vector of the variable X i , and the cross term can be a column vector obtained by multiplying the same position elements in the column vectors of any two variables.

[0041] It is worth noting that in order to solve the problem that the traditional sparse identification technology cannot reflect the temperature sensitivity and exponential rate response of the boiling chlorination reaction in the preparation process of titanium tetrachloride, the structure of the sparse function base is reconstructed. Specifically, by innovatively introducing an exponential function term such as an exponential expression with reaction temperature, chlorine flow and the like as variables on the basis of the traditional polynomial sparse function base, a composite sparse function base is constructed to characterize the nonlinear trend of the rate change in the gas-solid phase reaction. The synergistic effect of the exponential function term and the polynomial term not only enhances the expression ability of the model to complex nonlinear dynamics, but also improves the identification accuracy of the dominant reaction path and key interference factors.

[0042] Further, the sparse identification model is equal to, where the left side of the equation is the variable derivative matrix, and the right side of the equation is the matrix product between the composite sparse function base and the coefficient matrix in the form of a matrix.

[0043] The process of calculating the coefficient matrix by using the sequential threshold least squares method includes: inputting the variable derivative matrix obtained by preprocessing the collected time series data into the sparse identification model for calculation; a preset sparsification threshold λ is set, an initial coefficient matrix is solved based on the least squares method, coefficients with absolute values less than λ in the current coefficient matrix are set to zero, non-zero terms of the coefficients are retained, and an updated coefficient matrix is solved again by using the least squares method, and iterative calculation is repeatedly performed until the final coefficient matrix is obtained, which is used to represent the mathematical relationship between the key variables such as the concentration of titanium tetrachloride and the concentration of by-products.

[0044] The iteration stopping condition includes: the update amplitude of the coefficient matrix is less than a preset convergence threshold; or the iteration calculation number reaches a preset maximum iteration number threshold. The update amplitude of the coefficient matrix can be obtained from the norm of the difference between the coefficient matrices obtained by two adjacent calculations.

[0045] It should be noted that the sparse identification model includes a variable derivative matrix expression, a composite sparse function base expression and a coefficient matrix, wherein the composition of the composite sparse function base is formed by various function terms and exponential function terms of each variable, and the composite sparse function base is a known term after collecting the time series data of each variable; similarly, the variable derivative matrix is also a known term after collecting the time series data of each variable, so the coefficient matrix as an unknown term can be solved by the two known terms, and the coefficient matrix can represent the relationship between each variable.

[0046] The sequential threshold least squares method is used to solve the coefficient matrix. This method introduces a sparsification threshold to force some unimportant coefficients in the model to be zero, thereby realizing the selection of variables. In particular, during the initialization of the algorithm, the traditional least squares method is used to preliminarily estimate the coefficient matrix; by setting the sparsification threshold, each coefficient is sorted, and important non-zero coefficients are gradually selected, and unimportant coefficients are set to zero; repeat the process until all coefficients are stable or reach a preset iteration number. At this time, the final coefficient matrix is obtained, which contains the variables and coefficients that have an important influence on the sparse identification model.

[0047] As a specific embodiment, the non-zero terms in the coefficient matrix are extracted, and the influence relationship between the titanium tetrachloride concentration and the influencing factors and / or the influence relationship between the byproduct concentration and the influencing factors is determined through the mapping relationship, including: The numerical distribution characteristics of each column vector in the coefficient matrix are analyzed, and based on the mathematical expression form of the composite sparse function base, the dominant influence relationship between the titanium tetrachloride concentration variable and the influencing factor variable and / or the dominant influence relationship between the byproduct concentration variable and the influencing factor variable is determined through the mapping association according to the function base term corresponding to the non-zero coefficient.

[0048] Specifically, the numerical characteristics of the obtained coefficient matrix are analyzed, and the function structure with significant non-zero coefficients is extracted to identify the dominant influence relationship and mathematical expression form between variables, and a cross-validation method is used to evaluate the prediction performance of the constructed model to verify the generalization ability and accuracy of the sparse identification model under different data sets. Further, the distribution of each column vector in the coefficient matrix is analyzed to identify each influencing factor variable and the target variable titanium tetrachloride concentration or byproduct concentration (carbon monoxide, carbon dioxide and ferrous chloride) that has a significant correlation.

[0049] For example, for the column vector representing the derivative of titanium tetrachloride concentration, if the coefficients of the function terms related to chlorine flow, temperature field distribution, and raw material ratio are significant, it is preliminarily considered that the above-mentioned variables have a direct driving effect on the yield of titanium tetrachloride. Determine which variables are related to by-product concentrations (such as carbon monoxide, carbon dioxide, ferrous chloride); observe the trend chart of the influencing factors and the target product and by-products to obtain the optimal production scenario; for each non-zero coefficient, combined with the composition of the composite sparse function base, the physical or chemical relationship between the corresponding variable and the target variable is inferred. Specifically, for the titanium tetrachloride concentration variable, the coefficients of the linear function term and the exponential function term of chlorine flow, the linear function term and the quadratic function term of the temperature field distribution, and the linear function term and the exponential function term of the raw material ratio are all significant non-zero, it can be considered that the titanium tetrachloride concentration can be represented by the combination of these function terms to reflect the dominant influence relationship between variables.

[0050] After constructing the dilution recognition model, in order to ensure its stability and accuracy, the cross-validation method is used to verify the model. Specifically, after preprocessing all the collected data, it is divided into training set and validation set (training set and validation set are expressed in the form of variable data set and variable derivative matrix) in the ratio of 7:3, the sparse recognition model is fitted on the training set, and the coefficient matrix is obtained; the validation set is input into the constructed sparse recognition model, the value of the target variable is predicted, and the predicted result is compared with the actual measured value. The specific comparison and evaluation method is the prior art, so it is not described in detail.

[0051] The sparse recognition model constructed by the present application is mainly used for key variable analysis of dynamic reaction process in the process of boiling chlorination of titanium tetrachloride. Specifically, the model identifies the dominant influence relationship of temperature field distribution, chlorine flow, pressure fluctuation and other variables on the yield of titanium tetrachloride and the generation of by-products by fusing process mechanism and high-dimensional time series data. By solving the differential equation of the model, a real-time prediction function is constructed to guide the adjustment of chlorine flow, temperature field uniformity control and other operations, avoid excessive generation of impurities and optimize reaction efficiency.

[0052] The main improvement of the present application is to utilize the data-driven idea, and utilize the sparse identification technology to complete the process modeling, which is different from the traditional modeling technology. The traditional modeling technology is mainly based on the physical mechanism, the continuity equation and momentum equation of the gas phase and the pseudo-fluid phase are established by regarding the solid particles as pseudo-fluid, and the coupling and closed loop of the equation set are realized by relying on the interaction force between the gas and solid phases. This method depends on the in-depth understanding of the complex flow, heat transfer and reaction mechanism inside the reactor, and the model establishment process is complex, the parameters are numerous, and is easily affected by the simplification assumption, which leads to the limited precision and applicability. In contrast, the present application is based on the data-driven idea, the time series data in the actual production process is analyzed, the sparse function item which plays a leading role in the system dynamic behavior is automatically selected from the candidate function library, and a dynamic model with simple mathematical structure and physical interpretability is constructed. The present application does not need to preset the detailed mechanism of the gas-solid two-phase flow, avoids the complex fluid mechanics and reaction kinetics coupling process in the traditional model, and greatly reduces the modeling difficulty.

[0053] The above examples are further elaboration and illustration of the present application, so as to facilitate understanding, and are not any limitation of the present application, any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing the influencing factors of boiling chlorination in the production of titanium tetrachloride, characterized in that, include: Time series data of titanium tetrachloride concentration, by-product concentration, and alternative influencing factors were obtained and preprocessed to obtain the variable dataset and variable derivative matrix for the preparation of titanium tetrachloride by boiling chlorination. Select the sparse function basis of the sparse identification model and add an exponential function term to jointly form a composite sparse function basis. Use the sequential threshold least squares method to calculate the coefficient matrix in the sparse identification model. Extract the non-zero terms from the coefficient matrix, and determine the influence relationship between titanium tetrachloride concentration and influencing factors and / or the influence relationship between by-product concentration and influencing factors through mapping relationships.

2. The method for analyzing the influencing factors of boiling chlorination for titanium tetrachloride according to claim 1, characterized in that, The alternative influencing factors include at least the temperature field distribution, chlorine flow rate, feed formulation, and pressure fluctuations of the fluidized chlorination reactor. Preprocessing the time series data includes: using linear interpolation to fill in missing data items to ensure that all variables have a consistent time step.

3. The method for analyzing the influencing factors of boiling chlorination for titanium tetrachloride according to claim 1 or 2, characterized in that, The preprocessing to obtain the variable derivative matrix includes: The variable dataset is standardized and combined to obtain a variable matrix; the number of columns in the variable matrix is ​​the number of variables, and the number of rows in the variable matrix is ​​the number of elements in the column vector corresponding to each variable, representing the number of time-series data collected for the variable. For each variable in the variable matrix, the time derivative is obtained by using the finite difference method to obtain the variable derivative matrix.

4. The method for analyzing the influencing factors of boiling chlorination for titanium tetrachloride according to claim 1 or 2, characterized in that, The process of constructing the composite sparse function basis and calculating the coefficient matrix includes: A composite sparse function basis is constructed by adding exponential function terms to a polynomial function basis. The polynomial function basis includes at least linear function terms of each variable and cross terms of pairwise multiplication between each variable. The preprocessed variable derivative matrix is ​​used as an input to the sparse identification model, and the coefficient matrix in the sparse identification model is calculated using the sequential threshold least squares method.

5. The method for analyzing the influencing factors of boiling chlorination for producing titanium tetrachloride according to claim 4, characterized in that, The sparse identification model is an equation, where the left side of the equation is the variable derivative matrix, and the right side of the equation is the matrix product between the composite sparse function basis in matrix form and the coefficient matrix.

6. The method for analyzing the influencing factors of boiling chlorination for producing titanium tetrachloride according to claim 4, characterized in that, The process of calculating the coefficient matrix includes: A pre-defined sparsity threshold λ is used to solve the initial coefficient matrix using the least squares method. Coefficients in the current coefficient matrix whose absolute value is less than λ are zeroed out, and non-zero coefficients are retained. The updated coefficient matrix is ​​then solved again using the least squares method. The iterative calculation is repeated until the iteration stopping condition is met to obtain the final coefficient matrix.

7. The method for analyzing the influencing factors of boiling chlorination for producing titanium tetrachloride according to claim 6, characterized in that, The iteration stopping conditions include: The update magnitude of the coefficient matrix is ​​less than the preset convergence threshold; or the number of iterations reaches the preset maximum number of iterations threshold.

8. A method for analyzing the influencing factors of boiling chlorination for producing titanium tetrachloride according to claim 1, 2, 5, 6, or 7, characterized in that, Analyze the numerical distribution characteristics of each column vector in the coefficient matrix, and based on the mathematical expression of the composite sparse function basis, determine the dominant influence relationship between the titanium tetrachloride concentration variable and the influencing factor variable and / or the dominant influence relationship between the by-product concentration variable and the influencing factor variable through mapping correlation according to the function basis terms corresponding to the non-zero coefficients.

9. A system for analyzing influencing factors in the boiling chlorination process of titanium tetrachloride, applicable to the method for analyzing influencing factors in the boiling chlorination process of titanium tetrachloride as described in any one of claims 1-8, characterized in that, include: The data processing module collects data on various variables during the boiling chlorination process to prepare titanium tetrachloride, and performs preprocessing to obtain the variable dataset and variable derivative matrix; The function basis construction module selects the sparse function basis of the sparse identification model and adds exponential function terms to jointly form a composite sparse function basis; The influencing factor analysis module solves the coefficient matrix in the sparse identification model and determines the dominant influence relationship between variables based on the non-zero terms in the coefficient matrix.

10. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory is used to store program data and the processor is used to execute the program data to implement the method for analyzing the influencing factors of boiling chlorination of titanium tetrachloride as described in any one of claims 1-8.

Images