Chlorination process titanium dioxide production process causal modeling optimization method based on neural network agent model, electronic equipment and storage medium

Abstract

The invention discloses a causal modeling optimization method for a chlorination process titanium dioxide production process based on a neural network proxy model, electronic equipment and a storage medium. The method comprises the following steps: acquiring chlorination process titanium dioxide production process data and carrying out normalization preprocessing; eliminating the processed data by using recursive features based on cross validation to obtain confounding factor features strongly related to a target output result; training a proxy model by using the obtained feature data to predict potential results of each operation variable to a target output result; calculating a causal benefit through a potential result output by the agent model; according to the method, the calculation cost of causal modeling in the complex chlorination process titanium dioxide production process is reduced, the causal benefits of the operation variables can be efficiently evaluated under the condition that a system model is unclear or system parameters are insufficient, and a scientific basis is provided for auxiliary decision making and fault diagnosis in the chlorination process titanium dioxide production process.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and particularly relates to a causal modeling optimization method for the production process of titanium dioxide by the chlorination method based on a neural network surrogate model, an electronic device, and a storage medium. Background Art

[0002] In the production process of titanium dioxide by the chlorination method, the production process usually involves complex reactions, heat transfer, mass transfer, and fluid flow, with high nonlinearity and strong coupling. Traditional chemical process modeling methods rely on physical models and empirical formulas. However, in practical applications, due to system complexity, data noise, and experimental cost limitations, relying solely on physical models often fails to fully understand and optimize the process. In recent years, causal inference has gradually been applied to chemical process modeling. By determining causal relationships rather than simple correlations, it is possible to more accurately predict the system's response under different intervention measures and evaluate the effects of different operating conditions. This method is particularly suitable for solving problems such as key variable selection, parameter tuning, and uncertainty analysis.

[0003] Traditional causal inference methods mainly include randomized controlled trials, matching methods, regression analysis, instrumental variable methods, difference methods, and propensity score matching. These methods have their own advantages. For example, randomized controlled trials have high internal validity, instrumental variable methods can solve endogeneity problems, and difference methods are suitable for observational data, etc. However, they also have many limitations, such as high costs, ethical issues, overly strict model assumptions, insufficient control of confounding factors, poor performance on high-dimensional data, etc. In addition, these methods usually rely on strong assumptions (such as linear relationships, independence, etc.), are difficult to handle the complexity in the real world, and are sensitive to unobserved confounding factors or external shocks, resulting in possible biases or insufficient generalizability in causal inference results. Summary of the Invention

[0004] Based on the above background, the present invention proposes a method for constructing a causal model for the production process of titanium dioxide by the chlorination method based on a neural network surrogate model. Based on the theory of causal inference, combined with the potential outcome framework and the surrogate model, an efficient and accurate causal modeling method for chemical processes is constructed. While improving the model accuracy, this method also takes into account the computational efficiency and operability in the production application of titanium dioxide by the chlorination method, and has high practical value.

[0005] To achieve the above effects, the technical solutions adopted by the present invention are as follows:

[0006] The first aspect of the present invention provides a causal modeling optimization method for the production process of titanium dioxide by the chlorination method based on a neural network surrogate model, including the following sub-steps:

[0007] S1: Obtain the production process data of chloride process titanium dioxide. Select the product output variable in the production process data of chloride process titanium dioxide as the target variable, and use the recursive feature elimination method based on cross-validation to eliminate features from the production process data of chloride process titanium dioxide, obtaining confounding factor features related to the target variable;

[0008] S2: After obtaining the confounding factor features, conduct causal modeling on the target variable and the confounding factor features to estimate the causal benefit of the target variable. At the same time, establish a four-layer multi-layer perceptron (MLP) neural network surrogate model based on rule constraints to replace the causal modeling process;

[0009] S3: Use the confounding factor features and operating variables obtained in S1 as the input of the surrogate model, and the target variable as the output of the surrogate model. Train the surrogate model with a loss function added with rule constraints until the loss function converges and reaches the number of iterations, obtaining a trained surrogate model;

[0010] S4: Duplicate the two trained surrogate models in S3, connect the outputs of the two surrogate models with a fully connected layer. The two surrogate models respectively input the same confounding factor variables and different operating variables, and the fully connected layer can output the causal benefit of the target variable caused by the change of the operating variable.

[0011] Further preferably, a method for constructing a causal model of the chloride process titanium dioxide production process based on a neural network surrogate model, the method specifically includes the following sub-steps:

[0012] S1: Obtain the process data of the chloride process titanium dioxide production process, including 19 process measurement variables and 14 process operating variable data. Select the product output variable in the process flow as the target output result for research, and use recursive feature elimination based on cross-validation to obtain confounding factor features strongly related to the target output result;

[0013] S2: Use the potential outcome framework for causal modeling and introduce a surrogate model for optimization; the surrogate model directly proxies the causality to obtain the potential outcomes of each operating variable on the target output for evaluating the causal benefit;

[0014] S3: The surrogate model is modeled using a four-layer multi-layer perceptron (MLP) neural network with rule constraints, and is trained through the mapping relationship between the input confounding factor features and the output target results. During the model construction process, rule constraints and physical information are introduced to enhance the reliability and generalization ability of the model. During training, the input is the confounding factor features obtained in step S1, the output is the corresponding target output result, and the target output result is obtained using predictive regression.

[0015] S4: Designed the MLP proxy model under different operating variables. By integrating rule constraints and combining the fully connected layer to fuse the outputs of the two models, the target causal benefit is calculated, thereby achieving accurate modeling and evaluation of the causal relationship in the chloride process titanium dioxide production process.

[0016] In step S1, obtain the real data of the chloride process titanium dioxide production process and classify the labels according to the operating variables: the normal label is 0, and different operating variables are classified with labels 1, 2, 3, 4, and 5. The total number of samples is 10,000.

[0017] Select the TiO 2 particle size distribution as the target variable for evaluating the causal benefit, and use the genetic algorithm and greedy algorithm in the recursive feature elimination algorithm to obtain the importance degree of each feature with respect to the target variable, and then eliminate the features with relatively low importance degree with respect to the target variable. Finally, obtain variables such as raw material conveying flow rate, titanium ore purity, coke particle size distribution, chlorine flow rate, reactor temperature, titanium tetrachloride (TiCl 4 ) generation rate, TiCl 4 purity, hydrochloric acid (HCl) recovery rate, feeder speed, raw material preheating temperature, chlorine feed temperature, coke addition amount, reactor stirring speed, etc., and use them as confounding factor features.

[0018] Train a four-layer multi-layer perceptron (MLP) neural network proxy model based on rule constraints; during the model training process, use the different operating variables in step 2 and the confounding factor features obtained in step 3 as the input of the proxy model, and the target variable selected in step 3 as the output of the proxy model. Select the root mean square error combined with the prediction confidence and the physical information penalty term fusion as the loss function, and select the R 2 coefficient of determination as the performance metric of the proxy model.

[0019] The prediction confidence uses a Bayesian neural network to estimate the uncertainty of the model prediction. Samples with high uncertainty have low confidence, and samples with low uncertainty have high confidence. The confidence level is represented by α. The physical information penalty term is the mathematical expression of the chloride process titanium dioxide production process model, which contains the real physical laws of the chloride process titanium dioxide production process. Use the gradient descent method to minimize the total loss function that combines the root mean square error and the physical information penalty term with the prediction confidence; the total loss function is expressed as:

[0020]

[0021] where y represents the output data of the real TiO 2 particle size distribution; represents the predicted TiO 2Output data of particle size distribution; α is the confidence level; β is the weight coefficient of physical loss; is the root mean square error term; is the physical information rule penalty term.

[0022] Use the trained surrogate model to output the estimation of causal benefit. The method lies in designing 2 surrogate models under different operating variables; one surrogate model with the input operating variable being normal, and the other with the operating variable being faulty; train a fully connected layer, whose input is the outputs of the two surrogate models and the output is the difference between the outputs of the two surrogate models; the difference between the outputs of the two surrogate models is the causal benefit generated for the target variable when a fault occurs, expressed as:

[0023] CE = Y intervention - Y no_intervention

[0024] where, CE is the causal benefit; Y intervention is the output when faulty; Y no_intervention is the output when normal.

[0025] The second aspect of the present invention provides an electronic device, including a memory and a processor, the memory is coupled to the processor; wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned method for constructing a causal model of the chlorination titanium dioxide production process based on a neural network surrogate model.

[0026] The third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the above-mentioned method for constructing a causal model of the chlorination titanium dioxide production process based on a neural network surrogate model.

[0027] Compared with the prior art, the beneficial effects of the present invention are mainly reflected in:

[0028] Through the way of combining causal inference with a surrogate model, accurate modeling of complex causal relationships in the technological process of chlorination titanium dioxide production is realized, overcoming the defects of traditional causal inference methods in the technological process, enabling quick causal benefit analysis between variables under different operating variables. By optimizing the potential outcome framework, the surrogate model can generate corresponding causal benefit estimations for different fault types, providing important references for technological process optimization, fault detection and fault diagnosis, and having wide practical application value. Description of the Drawings

[0029] Figure 1 It is a schematic diagram of the distribution of normal data and faulty data;

[0030] Figure 2Schematic diagram of the feature elimination process of the method of the present invention;

[0031] Figure 3 It is a schematic diagram of some processes in the production of titanium dioxide by the chlorination method replaced by the surrogate model;

[0032] Figure 4 It is the performance of the potential result framework modeling optimization method based on the neural network surrogate model;

[0033] Figure 5 Schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0034] To make the objectives, design concepts, and technical solutions of the embodiments of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings.

[0035] Refer to Figures 1 to 4 , the present invention proposes a method for constructing a causal model in the production process of titanium dioxide by the chlorination method based on a neural network surrogate model, and the method includes the following steps:

[0036] Step S1: Obtain the process data of the production of titanium dioxide by the chlorination method as Figure 1 shown, including normal sample data, fault 1-5 sample data, label the normal data with the normal type label, and label the fault data with the fault type label, and the operating variables are different fault types;

[0037] The data set is derived from the process flow of the production of titanium dioxide by the chlorination method. This process mainly includes five core units: a chlorination reactor, a condenser, a gas-liquid separator, a compressor, and an oxidation reactor. In addition, it also includes a tail gas absorption tower and a supporting circulation system. The whole process uses titanium ore (such as rutile or high-titanium slag) and chlorine as the main raw materials, supplemented by coke as a reducing agent. Multiple intermediate materials and by-products are also involved in the process.

[0038] The main reactants include solid titanium ore and gaseous chlorine, and the reaction produces titanium tetrachloride (TiCl 4 ), as well as by-product ferric chloride (FeCl 3 ) and silicon tetrachloride (SiCl 4 ). Titanium tetrachloride is converted into titanium dioxide (TiO 2 ) in the downstream oxidation section, and at the same time, by-product hydrogen chloride gas is produced, which is absorbed to form hydrochloric acid solution after absorption. The feed also contains a small amount of inert substances, such as nitrogen and unreacted carbon dioxide, etc.

[0039] The core reaction equations of the whole process are as follows:

[0040] TiO 2 (s)+2C(s)+2Cl 2 (g)→TiCl4 (g) + 2CO(g) (1)

[0041] TiCl 4 (g) + O 2 (g) → TiO 2 (s) + 2Cl 2 (g) (2)

[0042] The main process of the chloride process for titanium dioxide production includes the following steps: First, in the chlorination reaction stage, titanium ore, coke, and chlorine react at high temperature in a chlorination reactor to produce TiCl 4 and by-products; subsequently, the mixed gas generated after the reaction is cooled by a condenser, and the liquid TiCl 4 is separated in a gas-liquid separator, and the remaining gas is recycled by a compressor or partially discharged as by-products. In the refining and oxidation stage of TiCl 4 , after the liquid TiCl 4 is refined to remove impurities by distillation, it is sent to an oxidation reactor to react with oxygen to produce TiO 2 powder, and at the same time, the by-product chlorine is recycled to the chlorination process. Finally, hydrogen chloride in the tail gas reacts with water in an absorption tower to produce by-product hydrochloric acid, which is collected and stored for sale. In this process dataset, the sample data includes measurement variables and operating variables. The measurement variables include flow rate, temperature, pressure, component concentration, etc., such as chlorine flow rate, reactor temperature, and titanium tetrachloride liquid level, etc.; the operating variables include raw material feeding rate, condenser cooling water flow rate, etc. In addition, there are several typical fault types in the process, for example, too high temperature in the chlorination reactor may lead to an increase in by-products or a decrease in the efficiency of the main product; blockage of the gas-liquid separator causes system pressure fluctuations; a decrease in the cooling efficiency of the condenser results in incomplete condensation of high-temperature gas, etc. Each fault type and its occurrence time are recorded in detail in the sample dataset for analyzing and optimizing the chloride process for titanium dioxide production.

[0043] The dataset is obtained from the actual chemical process, with a total of 33-dimensional data, including abnormal situations of 5 types of faults.

[0044] 1. For the 5 types of fault situations, select 1000 normal and fault data for each situation.

[0045] 2. Save all the data and perform label classification: The label for normal is 0, the label for fault 1 is 1, and so on. The total number of samples is 10,000.

[0046] 3. Preprocess the data by performing normalization. The operation is as follows:

[0047]

[0048] Among them, x newis the normalized data; x is the data before normalization; x max is the maximum value in the data; x min is the minimum value in the data.

[0049] 4. Select the particle size distribution of TiO 2 as the target output result of the research. Use recursive feature elimination based on cross-validation to obtain confounding factor features that are strongly correlated with the target output result. The existence of redundant and low-correlation features will increase the complexity of causal inference and reduce the accuracy of causality. Features that have a greater impact on the target output result need to be selected. On the premise of ensuring accurate causal inference, feature selection can reduce the dimensionality of the dataset and reduce complexity. Recursive feature elimination based on cross-validation (RFECV) is a greedy algorithm designed to find the best feature subset, and the process is as Figure 2 shown. First, retrieve all features in the sample set. By constructing a training classifier and calculating the importance of features, eliminate features with lower importance. Iteratively retrieve the remaining features until all features are traversed, and retain the required number of features to obtain a feature subset. Select two models, the genetic algorithm and the greedy algorithm, for feature selection. The judgment of feature importance is related to the dataset. When some features are eliminated, the model will recalculate the importance of each feature based on the new dataset. Therefore, compared with training a single model and selecting based on the feature importance of the result, the RFECV process will be more accurate. This accuracy is reflected in comparing the remaining features in the new environment every time a feature is eliminated. Through this recursive feature elimination algorithm, select the raw material conveying flow rate, titanium ore purity, coke particle size distribution, chlorine flow rate, reactor temperature, titanium tetrachloride production rate, TiCl 4 purity, HCl recovery rate, feeder rotation speed, raw material preheating temperature, chlorine feed temperature, coke addition amount, and reactor stirring speed as confounding factor features.

[0050] Step S2: Use the potential outcome framework for causal modeling. The potential outcome framework is a causal inference method used to estimate the causal benefits of a certain intervention or operation on a system. In the chlorination process for titanium dioxide production, set the type of failure as the operating variable and define the potential outcomes in two cases: intervention and non-intervention.

[0051] The modeling method is as follows:

[0052] S201. Determine the target output result: Select the variable that has the most significant impact on the system performance as the target output result (such as titanium ore purity, chlorine flow rate, etc.).

[0053] S202. Define the intervention strategy: By setting different types of controlling failures, simulate the system behavior under intervention and non-intervention.

[0054] S203. Collect data: Based on the simulation model or actual operation data, record the potential outputs of the system in the intervened and non-intervened states.

[0055] S204. Calculate the difference between the potential outputs in the intervened and non-intervened states to obtain the final causal benefit. The formula is:

[0056] CE = Y intervention - Y no_intervention (4)

[0057] where CE is the causal benefit; Y intervention is the state after intervention; Y no_intervention is the non-intervened state.

[0058] Due to the high complexity of the actual process, directly modeling the potential outcomes may have problems such as insufficient data or high computational overhead. Therefore, a surrogate model is introduced as an approximation tool to quickly predict the potential outcomes. As Figure 3 shown, this process is a flow chart for producing titanium dioxide from by-product hydrochloric acid in the chlorination process of titanium dioxide. When a fault is triggered, we can use the surrogate model to replace part of the process to directly obtain the potential outcomes of the titanium dioxide product after the fault occurs.

[0059] Use the MLP neural network as the surrogate model. Because of its powerful ability in non-linear mapping, it is suitable for modeling complex chemical processes.

[0060] Step 3: The surrogate model adopts a four-layer MLP neural network: The input layer receives the relevant feature variables obtained in S1. Each of the two hidden layers contains an appropriate number of neurons (128 and 64), and the ReLU activation function is used. The output layer predicts the value of the particle size distribution of TiO 2 particles, and the ReLU activation function is used. The data is divided into a training set and a validation set, and the K-fold cross-validation method is used to train the model.

[0061] K-fold cross-validation is a commonly used method to evaluate the performance of a model. First, it divides the data into K mutually exclusive subsets (folds), enabling each subset to be used as the validation set in one iteration, and the other K subsets as the training set, thus making full use of the data and improving the stability and reliability of the evaluation results. Second, K-fold cross-validation can effectively reduce the evaluation bias caused by uneven data division and provide a more accurate estimate of the model performance.

[0062] Customize the loss function. The goal is to comprehensively consider data error, physical consistency constraints, and confidence evaluation in a weighted manner. Select the coefficient of determination (R-square) as the model performance metric, and construct a composite loss function in the following form:

[0063]

[0064]

[0065] Among them, y represents the true TiO 2 output data of particle size distribution; represents the TiO predicted by the model 2 output data of particle size distribution; n represents the number of output data; y i , respectively represent the i-th true TiO 2 output data of particle size distribution and the TiO predicted by the model 2 output data of particle size distribution; α is the confidence level; β is the weight coefficient of physical loss; is the root mean square error term; is the physical information rule penalty term.

[0066] Introducing physical information and confidence level into the loss function can significantly improve the prediction performance and practical application value of the neural network model. First, the physical information constraint ensures that the prediction results of the model conform to scientific laws (such as conservation laws, kinetic equations, etc.), effectively avoiding results that violate physical principles, and improving the reliability and accuracy of the model. Second, guided by physical information, the model can efficiently learn key relationships in the case of scarce or incomplete data, thereby reducing the dependence on large-scale high-quality data. At the same time, the confidence level evaluation introduces the uncertainty weight of the prediction results, enabling the model to impose more stringent constraints on low-confidence prediction values, thereby reducing the impact of uncertainty on the overall results. The confidence level uses a Bayesian neural network to estimate the uncertainty of the model prediction. Samples with high uncertainty have low confidence levels, and samples with low uncertainty have high confidence levels.. In addition, the combination of physical information and confidence level significantly enhances the generalization ability of the model, enabling it to still provide stable prediction results under unknown working conditions or new scenarios. This design also optimizes the training process, improving the convergence speed of the model by balancing data error, physical constraints, and confidence level. The closer the loss function is to 0, the closer the model prediction value is to the true value. In the present invention, R 2 The coefficient of determination and MAE (mean absolute error) are used to characterize the accuracy of the model. The formulas are as follows:

[0067]

[0068] Among them, y represents the true TiO 2 output data of particle size distribution; represents the TiO predicted by the model 2 output data of particle size distribution; n represents the number of true output data; y i , respectively represent the i-th true TiO 2Output data of particle size distribution and the predicted TiO 2 Output data of particle size distribution; represents the mean of the true output data.

[0069] Coefficient of determination (R 2 ) is defined based on the sum of squared residuals and can quantify the explanatory power of the model for the target variable. Its value is usually between 0 and 1, which is convenient for intuitive understanding and comparison. It weights the errors in a squared form and is more sensitive to large errors, thus helping the model to focus on extreme prediction problems. In addition, the coefficient of determination is directly compatible with the least squares method, can reflect the proportion of the change in the target output result explained by the model, and has good interpretability. At the same time, as a standardized index, R 2 is insensitive to the units and scales of the data and is applicable to datasets of different scales. Its formula can also decompose the errors into the part explained by the model and the unexplained part, which is convenient for diagnosing model problems.

[0070]

[0071] where MAE represents the mean absolute error; n represents the number of true output data; y i , respectively represent the i-th true output data of particle size distribution of TiO 2 and the predicted output data of particle size distribution of TiO 2 particle size distribution.

[0072] As an evaluation index for the prediction model, MAE has the following advantages: its result unit is consistent with the target variable and is easy to interpret; it is insensitive to outliers and does not magnify the influence of large errors; it is applicable to any data distribution and is more robust to skewed data; it can directly measure the average error of the model prediction and fairly reflect the overall prediction performance.

[0073] Step 4: Designed MLP surrogate models for different operating variables, namely:

[0074] S401, normal output surrogate model: The input is the confounding factor characteristics under normal conditions, and the output is the target output result under normal conditions.

[0075] S402: Fault output surrogate model: The input is the confounding factor characteristics under fault conditions, and the output is the target output result under fault conditions.

[0076] Integrate the outputs of the trained normal output surrogate model and the fault output surrogate model through a fully connected layer, and calculate the target causal benefit according to the normal and fault model outputs.

[0077] This invention uses an example of causal modeling in the production process of chloride process titanium dioxide to illustrate the effectiveness of the causal benefit evaluation method based on the potential outcome framework and surrogate model. Figure 4 The performance of the causal benefit evaluation method based on the potential outcome framework and surrogate model is demonstrated. Figure 4 (a) The validation loss of each fold's iteration rounds shows the downward trend of the loss function of various methods during the training process of the surrogate model. Among them, the solid line is the training loss of fold 1, the long dashed line is the training loss of fold 2, the dotted line is the training loss of fold 3, and the short dashed line is the training loss of fold 4; Figure 4 (b) The curve of predicted values and true values shows the prediction results output by the surrogate model proposed in this invention. The results show that the predicted values can roughly fit the actual values.

[0078] In the modeling of the complex chloride process titanium dioxide production process, directly using traditional causal inference methods (such as the potential outcome framework) for causal benefit evaluation has high computational complexity and strong dependence on system parameters. However, the method proposed in this invention effectively reduces the computational complexity by introducing a surrogate model, making causal inference more efficient. At the same time, the method of this invention shows strong performance under the adaptation of various surrogate models (such as multi-layer perceptron, long short-term memory network, etc.). Experimental results show that the causal benefit evaluation method based on the potential outcome framework and surrogate model proposed in this invention significantly improves the accuracy of causal benefit prediction for the complex causal relationships in the chloride process titanium dioxide production process, reduces the computational cost of model training and inference, and has obvious advantages in terms of training time and result quality.

[0079] Correspondingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement a method for constructing a causal model of the chloride process titanium dioxide production process based on a neural network surrogate model as described above. As Figure 4 shown, it is a hardware structure diagram of any device with data processing capabilities where the method for constructing a causal model of the chloride process titanium dioxide production process based on a neural network surrogate model provided by an embodiment of this invention is located. In addition to Figure 5 the processors, memory, and network interfaces shown, any device with data processing capabilities where the device in the embodiment is located usually also includes other hardware according to the actual functions of the device with data processing capabilities, which will not be elaborated here.

[0080] Correspondingly, the present application further provides a computer-readable storage medium, on which computer instructions are stored. When the instructions are executed by a processor, the method for constructing a causal model of the titanium dioxide production process by chlorination method based on a neural network proxy model as described above is implemented. The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0081] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the content disclosed herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and examples are only to be considered as exemplary.

[0082] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A causal modeling optimization method for the chloride titanium dioxide production process based on a neural network agent model, characterized in that: It includes the following sub-steps: S1: Obtain the production process data of titanium dioxide by chloride process, select the fault type in the production process data of titanium dioxide by chloride process as the operating variable, select the product output variable in the production process data of titanium dioxide by chloride process as the target variable, and use the recursive feature elimination method based on cross-validation to eliminate the features of the production process data of titanium dioxide by chloride process to obtain the confounding factor features; S2: Constrain the four-layer multi-layer perceptron neural network through confidence and physical information rules to establish a proxy model; S3: The confounding factor characteristics and operation variables obtained in S1 are used as the proxy model input, the target variable is used as the proxy model output, and the proxy model is trained using the loss function constrained by the confidence and physical information rules until the loss function converges and reaches the number of iterations, and the trained proxy model is obtained, which is copied into two, namely the first proxy model and the second proxy model; S4: The confounding factor variables input into the first surrogate model and the second surrogate model are the same, the manipulated variables input into the first surrogate model and the second surrogate model are different, the output of the first surrogate model and the output of the second surrogate model are connected and output by a fully connected layer, and the fully connected layer outputs the causal benefit on the target variable caused by the change of the manipulated variable.

2. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 1 is characterized in that: In step S1, the product output variable is the particle size distribution of TiO2 particles.

3. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 2 is characterized in that: In step S1, a recursive feature elimination method based on cross-validation is used to eliminate the features of the chloride titanium dioxide production process data to obtain confounding factor features, specifically including: The genetic algorithm and greedy algorithm in the recursive feature elimination algorithm are used to obtain the importance of each feature in the chloride titanium dioxide production process data to the target variable, and the features that are irrelevant to the target variable are eliminated. Finally, the raw material delivery flow rate, titanium ore purity, coke particle size distribution, chlorine flow rate, reactor temperature, titanium tetrachloride generation rate, TiCl4 purity, hydrochloric acid recovery rate, feeder speed, raw material preheating temperature, chlorine feed temperature, coke addition amount, and reactor stirring speed variables are obtained as confounding factor features.

4. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 1 is characterized in that: In step S2, the four-layer multi-layer perceptron neural network is constrained by confidence and physical information rules to establish a proxy model, which specifically includes: The confidence is obtained using a Bayesian neural network, and the confidence size is represented by α. The physical information rule is a mathematical expression of the chloride titanium dioxide production process model. The physical information rule contains the real physical laws of the chloride titanium dioxide production process. The four-layer multi-layer perceptron neural network is constrained by the confidence and physical information rules to establish a proxy model.

5. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 4 is characterized in that: In step S3, the confounding factor characteristics and the operating variables obtained in S1 are used as the proxy model input, the target variable is used as the proxy model output, and the proxy model is trained using the loss function constrained by the confidence and physical information rules until the loss function converges and reaches the number of iterations, thereby obtaining a trained proxy model, specifically including: Use the gradient descent method to minimize the total loss function, the total loss function It is expressed as: Where y represents the actual TiO2 particle size distribution output data; Represents the output data of TiO2 particle size distribution predicted by the model; α is the confidence level; β is the weight coefficient of physical loss; is the root mean square error term, is the penalty term of physical information rule; Until the loss function converges and reaches the number of iterations, the trained proxy model is obtained.

6. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 1 is characterized in that: In step S4, the first proxy model and the second proxy model have different input operation variables, including: The input of the first proxy model is the manipulated variable in normal state, and the input of the second proxy model is the manipulated variable in fault state.

7. The causal modeling optimization method for chloride process titanium dioxide production process based on neural network agent model according to claim 1 is characterized in that: In step S4, the fully connected layer outputs the causal benefits of the target variable caused by the change of the manipulated variable, including: The difference between the outputs of the two proxy models is the causal benefit to the target variable when a failure occurs, expressed as: CE=Y intervention -Y no_intervention Among them, CE is the causal benefit; Y intervention is the output when a fault occurs; Y no_intervention This is the normal output.

8. An electronic device, comprising a memory and a processor, characterized in that: The memory is coupled to the processor; wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the causal modeling optimization method for the chloride process titanium dioxide production process based on the neural network agent model as described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the causal modeling optimization method for the chloride process titanium dioxide production process based on a neural network agent model described in any one of claims 1 to 7 is implemented.

Images