Ethylene glycol antimony response disproportionation inhibition intelligent control system

Abstract

The application discloses an ethylene glycol antimony reaction disproportionation inhibition intelligent control system and belongs to the technical field of production control, and aims to solve the problems of great fluctuation of raw materials, difficult early prevention and control of disproportionation reaction, fixed initial parameters, and unstable product quality caused by the fact that the initial parameters cannot be iteratively optimized in traditional ethylene glycol antimony preparation. According to the application, the personalized initial regulation parameter interval is generated according to the raw material component characteristics, the production process parameters are collected in real time, the process abnormalities are automatically corrected, the disproportionation reaction trend is predicted when the process is in a steady state, and the preventive inhibition is performed in advance, and the initial parameters are iteratively optimized based on the effective regulation effect. The disproportionation reaction full-process closed-loop prevention and control is realized, the quality risk is reduced from the source, and the product consistency and production stability are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of production control technology, and more specifically to an intelligent control system for inhibiting the disproportionation of antimony glycol. Background Technology

[0002] Antimony glycolate, as the most widely used antimony-based catalyst in the polyester industry, directly determines the color, transparency, and subsequent spinning performance of polyester chips through its purity and performance, making it a key raw material affecting the quality of polyester products. In the industrial preparation of antimony glycolate, the disproportionation reaction of +3 valent antimony is the core issue affecting product quality. The zero-valent antimony metal generated during the reaction leads to appearance defects such as gray haze and black spots in the product, while +5 valent antimony reduces the catalyst activity, thus affecting the reaction efficiency of subsequent polyester synthesis.

[0003] Currently, the production of antimony glycol commonly faces the problem of large batch fluctuations in raw materials. Antimony trioxide raw materials from different origins and batches exhibit significant differences in purity, impurity content, particle size distribution, and crystal structure. The moisture content and aldehyde impurity content of ethylene glycol raw materials also fluctuate. These differences in raw material characteristics directly alter the rate and extent of the disproportionation reaction, posing a significant challenge to the stable control of the production process. Traditional antimony glycol production control methods mainly rely on fixed process parameter ranges and manual periodic inspections. Adjustments can only be made passively after operating parameters exceed preset ranges or product quality issues arise, representing a reactive approach. By this time, the disproportionation reaction has already occurred, and even adjustments cannot completely eliminate the generated impurities, leading to a decrease in product qualification rate. Furthermore, manual adjustments suffer from response lag, insufficient precision, and strong subjectivity, easily triggering drastic fluctuations in the process system and further exacerbating product quality instability. Due to the lack of effective control methods that can dynamically adjust initial process parameters based on raw material characteristics and identify early signs of disproportionation reactions, manufacturers often adopt overly conservative process parameters to reduce the risk of disproportionation reactions. This leads to slower reaction rates, longer production cycles, and increased energy consumption, severely restricting the economic benefits and product competitiveness of antimony glycol production. Therefore, to overcome these limitations, this invention proposes an intelligent control system for inhibiting disproportionation in the antimony glycol reaction. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide an intelligent control system for inhibiting the disproportionation of antimony glycol, which solves the problem of unstable product quality and low production efficiency caused by the difficulty in effectively controlling the disproportionation reaction during the preparation of antimony glycol.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A smart control system for inhibiting the disproportionation of antimony glycol includes: The initialization module is used to generate an initial control parameter range based on the characteristic data of the raw material composition for the current preparation of antimony glycol and a preset mapping relationship between the raw material composition and the control parameter. The adaptive adjustment module is used to collect real-time characteristic parameters of each operation in the preparation of antimony glycol, compare them with the initial control parameter range, determine whether adaptive adjustment is triggered, and if so, identify abnormal control parameters, calculate the adjustment amount of abnormal control parameters, and generate the first control command; the real-time characteristic parameters include operation control parameters and state characterization parameters. The trend prediction module, in response to the absence of adaptive adjustment, quantifies the disproportionation response quantitative index of the current operation stage based on real-time feature parameters, and uses it to predict the disproportionation response trend within a preset time period in the future, and determines whether there is a risk of disproportionation. If so, it generates a second control instruction. The feedback self-correction module receives the first and second control instructions, performs differentiated disproportionation suppression control, calculates the control effect evaluation parameters based on the disproportionation response quantification index after the execution of the second control instruction, identifies effective control samples, self-corrects the initial control parameter range, and generates an updated control parameter range.

[0006] Specifically, the mapping relationship between raw material composition and control parameters refers to a hierarchical mapping data table that reflects the quantitative correspondence between raw material composition characteristic data and corresponding control parameter ranges, and its structure is as follows: Principal component factors were extracted from the raw material component characteristic data using principal component analysis. Each principal component factor is divided into a predetermined number of graded intervals according to its influence on the disproportionation reaction. Establish a one-to-one correspondence between all hierarchical combinations of the predetermined number of principal component factors and the control parameter ranges of each operation step.

[0007] Specifically, the steps for generating the initial control parameter range include: Obtain raw material component characteristic data and preprocess the obtained raw material component characteristic data; Extract the principal component factor vector from the preprocessed raw material component feature data; The score of each principal component factor in the principal component factor vector is compared with the corresponding principal component factor grading interval one by one to determine the grading interval to which each principal component factor belongs, and the principal component factor grading combination of the current batch of raw materials is obtained. Based on the principal component factor hierarchical combination, query the preset mapping relationship between raw material components and control parameters to obtain the initial control parameter range of each operation step that matches the current raw material components.

[0008] Specifically, the steps to determine whether adaptive adjustment is triggered include: Receive the initial control parameter range and the real-time feature parameters uploaded by each operation stage, and perform multi-dimensional data preprocessing on the real-time feature parameters; Construct a mapping table corresponding to the control parameters, the operating control parameters, and the initial control parameter range; Traverse the corresponding mapping table and compare the preprocessed multi-dimensional data operation control parameters with the upper and lower limits of the corresponding initial control parameter range one by one to identify potential abnormal operation control parameters that exceed the initial control parameter range. Continuous verification of potential abnormal operation control parameters is carried out by collecting potential abnormal operation control parameter data for a preset number of consecutive sampling periods. When the potential abnormal operation control parameter exceeds the initial control parameter range within the preset number of consecutive sampling periods, it is marked as a real process abnormal parameter. When at least one real process anomaly parameter exists, adaptive adjustment is triggered.

[0009] Specifically, the steps for generating the first control command include: Identify the abnormal control parameters corresponding to all real process abnormal parameters, and determine the adjustment order of each abnormal control parameter according to the preset control parameter priority table; According to the adjustment order, the initial control parameter range midpoint is used as the set value, and the operation control parameter value after multi-dimensional data preprocessing is used as the process variable. The basic adjustment amount is calculated for each abnormal control parameter in turn. Based on the preset maximum adjustment threshold, the calculated basic adjustment amount is divided into step-by-step adjustments to obtain a step-by-step adjustment amount sequence. Set the adjustment rate and single-step execution timeout for each adjustment step in the step-by-step adjustment sequence; Generate a first control instruction, the content of which includes an abnormal control parameter identifier, a control sequence identifier, a step-by-step control quantity sequence corresponding to each abnormal control parameter, a corresponding actuator identifier, a control rate for each control step, and a single-step execution timeout.

[0010] Specifically, the steps to determine whether there is a risk of misrepresentation include: Obtain the state characterization parameters of the current operation stage, and quantify the state characterization parameters into the disproportionation reaction quantification index of the current operation stage; the disproportionation reaction quantification index includes the current +5 valence antimony content, the current 0 valence metallic antimony content, the disproportionation reaction conversion rate, and the disproportionation reaction rate; The operation control parameter values, state characterization parameter values, and current disproportionation reaction quantitative indicators of each operation link, after multi-dimensional data preprocessing, are input into the constructed disproportionation reaction prediction model to predict the predicted values ​​of each disproportionation reaction quantitative indicator within a preset time period in the future. The predicted values ​​of the quantitative indicators of disproportionation response for each preset time period in the future are compared with the preset disproportionation risk thresholds one by one to identify abnormal quantitative indicators of disproportionation response. Based on the absolute deviation between the abnormal quantitative indicators of disproportionation response and the corresponding disproportionation risk thresholds, the disproportionation risk level is classified. When at least one abnormal disproportionation response quantitative indicator exists, a disproportionation risk is determined, and a second regulatory instruction is generated in conjunction with a pre-set graded advanced disproportionation inhibition and control strategy.

[0011] Specifically, the steps for constructing a disproportionation reaction prediction model include: A model training dataset is constructed, which includes historical batch raw material composition feature data, operation control parameter data for each operation step, state characterization parameter data, and disproportionation reaction quantitative index data; the model training dataset is preprocessed to form a standardized model training dataset. Based on the standardized model training dataset, a combination of Pearson correlation coefficient and variance filtering was used for feature selection to obtain the input feature set of the disproportionation response prediction model. Supervised learning samples for the disproportionation response prediction model are generated using a sliding window truncation method. All parameter values ​​of the input feature set at the current time are used as model input samples, and the measured values ​​of the disproportionation response quantification index corresponding to a preset time duration are used as the labels of the model input samples. The model input samples are divided into training and test sets according to a preset sample division ratio. A gradient boosting tree algorithm is used to construct a disproportionation reaction prediction model; a combination of grid search and cross-validation is used to optimize the hyperparameters of the disproportionation reaction prediction model; and a test set is used to verify the accuracy of the disproportionation reaction prediction model, thus forming the final disproportionation reaction prediction model.

[0012] Specifically, the graded advanced disproportionation suppression and regulation strategy is used to define corresponding target regulation parameters for different risk levels and combinations of abnormal disproportionation response quantitative indicator types, and to set a standardized benchmark adjustment degree for each target regulation parameter, while clarifying the adjustment direction constraints of the regulation parameters corresponding to different abnormal disproportionation response quantitative indicator types.

[0013] Specifically, the steps for generating the second control command include: Based on the quantitative index type of abnormal disproportionation response and the disproportionation risk level obtained, a pre-set graded advanced disproportionation inhibition and regulation strategy is matched. For each type of identified abnormal disproportionation response quantitative index, the absolute deviation of the index from the threshold corresponding to the current disproportionation risk level is calculated to obtain the benchmark deviation. According to the matched hierarchical advance disproportionation suppression control strategy, the initial advance adjustment amount is obtained by multiplying the baseline deviation by the baseline adjustment degree corresponding to each target control parameter, and the adjustment direction of each target control parameter is determined by combining the abnormal disproportionation response quantification index type. The absolute value of the initial advance adjustment amount is compared with the preset maximum advance adjustment amount threshold. If it exceeds the maximum advance adjustment amount threshold, the maximum advance adjustment amount threshold is used as the final advance adjustment amount. Configure appropriate advance adjustment rates and adjustment execution timeouts for the final advance adjustment amounts of each target control parameter; The system integrates the disproportionation risk level identifier, the abnormal disproportionation response quantitative indicator type identifier, the target control parameter identifier, the final advance control amount with adjustment direction, the corresponding actuator identifier, the advance control rate, and the control execution timeout time to generate a standardized second control instruction.

[0014] Specifically, the steps for calculating the evaluation parameters of the regulatory effect and identifying effective regulatory samples based on the disproportionation response quantification index after the execution of the second regulatory instruction include: After all the second control commands have been executed and the preset stabilization period has elapsed, the state characterization parameters of the reaction system are continuously collected within the preset monitoring period, and the quantitative index of the stable disproportionation reaction after the advance control is calculated. The average value of each disproportionation response quantitative indicator within the preset monitoring period is taken as the final disproportionation response quantitative indicator value after regulation, and the disproportionation risk level is classified. Based on the quantitative indicators of abnormal disproportionation response predicted before the advance regulation and the corresponding disproportionation risk level, the evaluation parameters of the regulation effect of this advance regulation are calculated; the evaluation parameters of the regulation effect include the deviation reduction rate of the quantitative indicators of disproportionation response, the risk level downgrade rate, and the stability of indicator fluctuation. Based on the evaluation parameters of the regulation effect, it is determined whether the current advance regulation is a positive and effective regulation. If so, the effective regulation sample is formed by associating the graded combination of raw material principal component factors, the initial regulation parameter range, the actual regulation parameters of the second regulation instruction, the regulation parameters of stable operation after regulation, and the evaluation parameters of the regulation effect.

[0015] Specifically, the steps for self-correcting the initial control parameter range and generating an updated control parameter range include: The archived effective control samples are classified and statistically analyzed according to the graded combination of raw material principal component factors. The number of effective control samples corresponding to each graded combination of raw material principal component factors is counted. If the preset sample accumulation threshold is reached, the self-correction process of the initial control parameter range corresponding to the graded combination of raw material principal component factors is initiated. Extract all archived effective control samples under the principal component factor classification combination of the raw material, perform statistical analysis, and calculate the statistical mean and standard deviation of each control parameter respectively; The statistical mean is used as the median of the updated initial control parameter range. The upper and lower limits of the updated initial control parameter range are respectively the statistical mean plus or minus the preset range width coefficient and the standard deviation, forming the updated control parameter range, which is then written back to the raw material composition and control parameter mapping data table.

[0016] The beneficial effects of this invention are: This application constructs a hierarchical mapping relationship based on the characteristics of raw material composition and generates personalized initial control parameter ranges. This adapts to the characteristic differences of different batches of raw materials from the source, significantly reducing the inherent risk of disproportionation reaction caused by raw material fluctuations. By performing multi-dimensional preprocessing and continuous anomaly verification on real-time operating parameters, combined with adaptive control of priority ranking and step-by-step precise adjustment, process deviations are corrected in a timely and stable manner, avoiding the aggravation of disproportionation reaction caused by abnormal parameter expansion. By quantifying the disproportionation reaction quantitative index and using a gradient boosting tree model to predict future disproportionation trends, combined with a hierarchical proactive suppression strategy to implement preventive control, the application achieves pre-emptive prevention and control of disproportionation reaction, effectively solving the problem that traditional post-event remediation cannot eliminate already generated impurities. By accumulating effective control samples and iteratively optimizing the initial control parameter range based on statistical analysis, the system continuously self-optimizes with production data, constantly improving the adaptability of initial parameters, gradually reducing the trigger frequency of preventive control, and ultimately achieving closed-loop intelligent control of disproportionation reaction throughout the entire process of antimony glycol production. This significantly improves product quality consistency and production operation stability, while reducing energy consumption and manual control costs. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an intelligent control system for inhibiting the disproportionation of antimony glycol according to the present invention; Figure 2 A flowchart for generating the initial control parameter range for this invention; Figure 3 This is a flowchart illustrating the process of determining whether adaptive adjustment is triggered according to the present invention. Figure 4 This is a flowchart for determining whether there is a risk of misinterpretation in this invention. Detailed Implementation

[0018] Please see Figure 1 This embodiment introduces an intelligent control system for inhibiting the disproportionation of antimony glycol, including an initialization module, an adaptive adjustment module, a trend prediction module, and a feedback self-correction module. The initialization module is used to generate an initial control parameter range based on the characteristic data of the raw material composition for the current preparation of antimony glycol, and based on the preset mapping relationship between the raw material composition and the control parameter, and send the initial control parameter range to the adaptive adjustment module and the trend prediction module respectively.

[0019] The raw material composition characteristic data refers to a set of parameters characterizing the purity, impurity content, and physicochemical properties of the raw materials for preparing antimony glycol, including antimony trioxide raw material composition characteristic data and ethylene glycol raw material composition characteristic data; the antimony trioxide raw material composition characteristic data refers to a set of parameters characterizing the quality and reactivity of antimony trioxide raw materials, including the purity, particle size distribution, crystal form ratio, iron impurity content, copper impurity content, and lead impurity content of antimony trioxide raw materials; the ethylene glycol raw material composition characteristic data refers to a set of parameters characterizing the purity and oxidation degree of ethylene glycol raw materials, including the purity, moisture content, glyoxal content, and total aldehyde content of ethylene glycol raw materials.

[0020] The aforementioned control parameters refer to a set of process parameters that can be actively adjusted and controlled by the actuator to directly affect the preparation process of antimony glycol and the degree of reaction disproportionation. These parameters include: raw material mixing temperature and mixing speed in the raw material pretreatment stage; reaction temperature, reaction pressure, reaction stirring speed, nitrogen flow rate, hydrogen flow rate, and inhibitor dosage in the reaction synthesis stage; crystallization temperature, crystallization time, crystallization stirring speed, and centrifugal separation speed in the crystallization separation stage; and drying temperature, drying vacuum degree, and drying time in the drying and packaging stage.

[0021] The mapping relationship between raw material components and control parameters refers to a hierarchical mapping data table that reflects the quantitative correspondence between raw material component characteristic data and corresponding control parameter ranges. Its structure is as follows: Principal component analysis (PCA) is used to extract a predetermined number of principal component factors from the raw material component characteristic data. The predetermined number of principal component factors refers to the number of core factors used to characterize the influence of raw material components on the reaction disproportionation. This number is set based on the accuracy of the balance mapping relationship and the system complexity. Too few factors will result in the loss of raw material characteristics that significantly affect the reaction disproportionation, while too many factors will lead to an exponential increase in the number of hierarchical combinations, significantly increasing data storage and computational complexity. For example, the top four principal component factors with the highest weight influencing the reaction disproportionation are extracted, in the following order: total iron and copper impurities in antimony trioxide, glyoxal content in ethylene glycol, antimony trioxide particle size distribution, and antimony trioxide crystal form ratio. Each principal component factor is divided into a predetermined number of graded intervals according to its influence on the disproportionation reaction. The predetermined number of graded intervals refers to the number of levels used to classify the influence of each principal component factor. The setting is based on balancing the control precision and the implementation difficulty. Too few grades will lead to insufficient control precision and failure to achieve accurate matching between raw material components and control parameters. Too many grades will significantly increase the workload of orthogonal experimental verification and the difficulty of maintaining the mapping table. For example, each principal component factor is divided into 3 graded intervals, corresponding to the low influence interval, medium influence interval and high influence interval, respectively. A one-to-one correspondence is established between all hierarchical combinations of the predetermined number of principal component factors and the control parameter ranges of each operation step. The complete control parameter range includes the upper and lower limits of all control parameters in each operation step, which are pre-calculated through multiple regression analysis and verified by orthogonal experiments. This is achieved by setting principal component analysis and multiple regression analysis on a predetermined number of batches of historical production data, and verifying the effectiveness of the control parameter range corresponding to each combination through orthogonal experiments. The predetermined number of batches of historical data refers to the minimum number of batches of historical production data required to construct the mapping relationship. This number is set to ensure that the mapping relationship has sufficient generalization ability and accuracy, covering the vast majority of raw material component combinations and production conditions. For example, the predetermined number of batches of historical data is no less than 1000 batches.

[0022] Please see Figure 2 Furthermore, the specific steps for generating the initial control parameter range include: Step S11: Obtain the raw material component characteristic data of the current batch and preprocess the obtained raw material component characteristic data; the preprocessing includes using the 3σ criterion to remove abnormal detection values ​​that exceed the normal fluctuation range in order to eliminate the influence of raw material detection error on subsequent calculations, and using the minimum-maximum normalization method to map all raw material component characteristic data to the [0,1] interval in order to eliminate the influence of the difference in the dimensions of different parameters on the principal component analysis results.

[0023] Step S12: Input the preprocessed raw material component feature data into a preset principal component analysis model to extract the principal component factor vector of the current batch of raw materials; the preset principal component analysis model is a dimensionality reduction statistical model used to extract principal component factors from raw material component feature data, which is pre-trained on a historical raw material component training dataset, and the specific construction process is as follows: A historical raw material composition training dataset is constructed. This dataset includes characteristic data of antimony trioxide and ethylene glycol raw materials from a predetermined number of historical data batches. Each batch of data corresponds to a complete record of the disproportionation reaction degree and product quality. The predetermined number of historical data batches refers to the minimum number of qualified production data batches required to construct the mapping relationship between raw material composition and control parameters, containing complete records of raw material composition, process operation, disproportionation reaction degree, and product quality. The basis for setting this number is to ensure that the mapping relationship has sufficient statistical representativeness and generalization ability, and can cover most raw material composition combinations, production condition fluctuations, and equipment operating status. If the number of batches is too small, the mapping relationship will not be able to fully reflect the corresponding rules of different raw material characteristics and control parameters, resulting in some raw material combinations having no matching parameters or insufficient parameter matching accuracy. If the number of batches is too large, it will significantly increase the workload of data collection, cleaning, and analysis, prolong the construction cycle of the mapping relationship, and introduce too much redundant data, affecting computational efficiency. For example, the predetermined number of historical data batches is no less than 1000 batches.

[0024] The historical raw material composition training dataset was standardized by using the Z-score standardization method to convert each raw material composition feature data into standardized data with a mean of 0 and a variance of 1, so as to eliminate the influence of differences in the scale and magnitude of different features on the analysis results. Calculate the covariance matrix of the standardized historical raw material composition data. The covariance matrix is ​​used to characterize the degree of linear correlation between different raw material composition characteristics. Its element value is the covariance of two corresponding raw material composition characteristics. Solve for all eigenvalues ​​and corresponding eigenvectors of the covariance matrix, sort all eigenvalues ​​in descending order, and assign each eigenvalue a principal component factor. The magnitude of the eigenvalue represents the proportion of the original data variance that the principal component factor can explain. The cumulative variance contribution rate of the first k principal component factors is calculated sequentially. When the cumulative variance contribution rate reaches the preset principal component variance contribution rate threshold, the preset number of principal component factors is determined to be k, and principal component extraction is stopped. The preset principal component variance contribution rate threshold is the minimum cumulative variance contribution rate threshold used to determine whether the principal component factor extraction is sufficient and can cover the core information in the original raw material component feature data that has a significant impact on reaction disproportionation. Its setting is based on balancing the information completeness of principal component factor extraction with the system's computational complexity. Setting it too low will result in an insufficient number of extracted principal component factors, losing raw material feature information that has a significant impact on reaction disproportionation, and reducing the accuracy of the mapping relationship between raw material components and control parameters. Setting it too high will result in an excessive number of extracted principal component factors, introducing a large amount of redundant information, causing the number of hierarchical combinations to grow exponentially, significantly increasing the storage capacity of the mapping data table, query complexity, and system computational burden, while also reducing the model's generalization ability. For example, when the preset principal component variance contribution rate threshold is 95%, the calculated cumulative variance contribution rate of the first 4 principal component factors is 96.2%, which meets the requirement; therefore, the preset number of principal component factors is determined to be 4.

[0025] A principal component factor loading matrix is ​​constructed, where each column corresponds to a principal component factor and each row corresponds to an original raw material component feature. The matrix elements are the loading coefficients of the corresponding original raw material component features on the corresponding principal component factors. The larger the absolute value of the loading coefficient, the greater the contribution of the original raw material component feature to the principal component factor. For example, the loading coefficients of the first principal component factor on the iron impurity content and copper impurity content in antimony trioxide are 0.89 and 0.87, respectively. Therefore, this principal component factor mainly characterizes the influence of the total iron and copper impurity content in antimony trioxide on the disproportionation reaction. The principal component analysis model was validated using a 10-fold cross-validation method to ensure that the cumulative variance contribution rate of the model on different data subsets was not lower than the preset principal component variance contribution rate threshold, thus guaranteeing the stability and generalization ability of the model.

[0026] The principal component factor vector is calculated as follows: the preprocessed raw material component feature data of the current batch is multiplied by the transpose of the principal component factor loading matrix to obtain the principal component factor vector of the current batch of raw materials. Each element in the vector corresponds to the score value of a principal component factor.

[0027] Step S13: Compare the score of each principal component factor in the principal component factor vector with the corresponding principal component factor grading interval one by one to determine the grading interval to which each principal component factor belongs, and combine them to obtain the principal component factor grading combination of the current batch of raw materials. Step S14: Based on the principal component factor classification combination, query the preset mapping relationship between raw material components and control parameters to directly obtain the initial control parameter range of each operation step that matches the current raw material components; perform boundary constraints on the initial control parameter range obtained by querying according to the process allowable range of each operation step. The process allowable range is the safe operating range of each control parameter specified by industry general standards or enterprise production specifications to ensure that the constrained range does not exceed the process allowable boundary; perform a rationality verification on the initial control parameter range of each operation step after boundary constraint. The verification criteria include that the conversion rate of the antimony glycol main reaction is not lower than the preset main reaction conversion rate threshold, the total reaction time does not exceed the preset upper limit of reaction time, and the antimony content of the product is not lower than the preset lower limit of antimony content of the product. After the verification is passed, the final initial control parameter range is output.

[0028] Specifically, the initialization module is designed to address three core issues in the preparation of antimony glycol: large fluctuations in raw material composition, ineffective integration of multi-dimensional raw material characteristics, and a lack of scientific basis for initial control parameters. In the industrial production of antimony glycol, different batches of antimony trioxide raw materials exhibit significant differences in purity, particle size distribution, crystal form ratio, and the content of heavy metal impurities such as iron, copper, and lead. Different batches of ethylene glycol raw materials also show significant fluctuations in purity, moisture content, and the content of oxidation products such as glyoxal. These differences in raw material composition directly affect the rate and extent of the disproportionation reaction of +3 antimony. Traditional preparation processes use a uniform and fixed range of initial control parameters, failing to consider the compositional differences between different batches of raw materials. This makes it impossible to achieve precise matching between raw material characteristics and process parameters, leading to premature and severe disproportionation reactions in some batches due to unreasonable initial parameter settings. This results in quality problems such as gray fog, black spots, and color differences in the product. Simultaneously, some batches of raw materials have reduced production efficiency and increased energy consumption and production costs due to overly conservative initial parameters. The initialization module achieves personalized generation of initial control parameter ranges through a systematic operation involving raw material component characteristic preprocessing, principal component factor extraction, hierarchical mapping relationship query, and parameter rationality verification. When the total iron and copper impurity content in the raw material batch is high, or the glyoxal content in ethylene glycol is high, the generated initial control parameter range will automatically lower the upper limit of the reaction temperature and increase the lower limit of the nitrogen flow rate to suppress the disproportionation reaction. When the raw material batch has high purity and low impurity content, the generated initial control parameter range will appropriately increase the upper limit of the reaction temperature and shorten the reaction time to improve production efficiency. Without the personalized initial parameter generation operation of the initialization module, the initial control parameters will be severely mismatched with the characteristics of the current batch of raw materials. Either the initial parameters are too lenient, causing the disproportionation reaction to occur in large quantities in the early stage of the reaction, and even strong suppression measures cannot salvage the product quality; or the initial parameters are too conservative, significantly prolonging the reaction time, reducing the production capacity of the unit, and increasing the control pressure and frequency of the subsequent adaptive adjustment module, affecting the stability and reliability of the entire control system.

[0029] The adaptive adjustment module is used to collect real-time characteristic parameters of each operation in the preparation of antimony glycol, compare them with the initial control parameter range, determine whether adaptive adjustment is triggered, and if so, identify abnormal control parameters, calculate the adjustment amount of abnormal control parameters, generate a first control command, and send the first control command to the feedback self-correction module.

[0030] The aforementioned operational steps refer to the production stages in the preparation of antimony glycol according to the process sequence, including the raw material pretreatment step, the reaction synthesis step, the crystallization separation step, and the drying and packaging step.

[0031] The real-time characteristic parameters refer to the set of detectable parameters passively collected by detection instruments to characterize the real-time operating status and reaction process of each operation link, including operation control parameters and status characterization parameters. The operational control parameters refer to detectable parameters that correspond one-to-one with the actively adjustable control parameters and are used to characterize the actual execution effect of each control parameter. Their values ​​directly reflect the execution result of the control command and can be changed by adjusting the operating state of the corresponding actuator. These include the actual values ​​of raw material mixing temperature and raw material mixing stirring speed in the raw material pretreatment stage; the actual values ​​of reaction temperature, reaction pressure, reaction stirring speed, nitrogen flow rate, hydrogen flow rate, and inhibitor dosage in the reaction synthesis stage; the actual values ​​of crystallization temperature, crystallization time, crystallization stirring speed, and centrifugal separation speed in the crystallization separation stage; and the actual values ​​of drying temperature, drying vacuum degree, and drying time in the drying and packaging stage. The state characterization parameters refer to detectable parameters that cannot be directly adjusted by the system through the actuator and are used to characterize the internal chemical state and reaction process of the reaction system during the preparation of antimony glycol. Their numerical changes are the result of the changes in the reaction system itself and are the core basis for judging the degree of reaction disproportionation and predicting the risk of disproportionation. They include the redox potential of the reaction system in the reaction synthesis stage, the transmittance of the reaction system at a wavelength of 420 nm, the glyoxal concentration of the reaction system, and the gas phase oxygen content in the reaction vessel.

[0032] Please see Figure 3 Furthermore, the specific steps for determining whether adaptive adjustment has been triggered include: Step S21: Receive the initial control parameter range sent by the initialization module, as well as the real-time characteristic parameters uploaded by each operation link; perform multi-dimensional data preprocessing on the received real-time characteristic parameters. First, use the 3σ criterion to remove invalid data that exceeds the instrument's range or has transmission errors. Then, use the moving average filtering method to smooth the valid data, eliminating data fluctuations caused by electromagnetic interference and random noise from the instrument in the industrial field. This solves the problems of low signal-to-noise ratio and susceptibility to false anomalies in the original acquired data.

[0033] Step S22: Construct a mapping table corresponding to the control parameters, operating control parameters and initial control parameter ranges, and uniquely bind each actively adjustable control parameter to its corresponding actual operating control parameter and initial control parameter range to clarify the system's adjustment object and judgment criteria.

[0034] Step S23: Traverse the corresponding mapping table and compare the preprocessed multi-dimensional data operation control parameters with the upper and lower limits of the corresponding initial control parameter range one by one to identify potential abnormal operation control parameters that exceed the initial control parameter range. Step S24: Continuously verify the potential abnormal operation control parameters. Collect potential abnormal operation control parameter data for a preset number of continuous sampling periods. When a potential abnormal operation control parameter exceeds the initial control parameter range within the preset number of continuous sampling periods, it is marked as a real process abnormal parameter; when it only exceeds the range momentarily, it is judged as normal process fluctuation and is not marked. The preset number of continuous sampling periods refers to the number of continuous samplings required to verify whether the abnormal operation control parameter is a continuous process abnormality. The setting is based on balancing the sensitivity and anti-interference capability of abnormality detection. Too few sampling periods will misjudge instantaneous electromagnetic interference and instrument jumps in the industrial field as real process abnormalities, leading to frequent erroneous system adjustments; too many sampling periods will prolong the abnormality detection response time, leading to an increased risk of misinterpretation. For example, the preset number of continuous sampling periods is 3 sampling periods.

[0035] Step S25: When at least one real process abnormal parameter exists, it is determined that adaptive adjustment is triggered; when all operating control parameters are within the corresponding initial control parameter range within a consecutive preset number of consecutive sampling periods, it is finally determined that adaptive adjustment is not triggered, and the steady-state signal, the smoothed values ​​of all operating control parameters, and the smoothed values ​​of all state characterization parameters are simultaneously sent to the trend prediction module to provide complete and accurate real-time status data for subsequent advanced prediction of disproportionation risk.

[0036] Furthermore, the specific steps for generating the first control command include: Step S31: Identify the abnormal control parameters corresponding to all real process abnormal parameters, and determine the adjustment order of each abnormal control parameter according to the preset control parameter priority table; the control parameter priority table is pre-defined according to the degree of influence of each parameter on the +3 antimony disproportionation reaction, wherein the reaction temperature and gas phase oxygen content in the reactor are the first priority, the reaction pressure and nitrogen flow rate are the second priority, and the remaining control parameters are the third priority, which solves the problem of chaotic adjustment order and inability to prioritize control of key risks when multiple parameters are abnormal at the same time.

[0037] Step S32: According to the adjustment sequence, with the initial control parameter range midpoint as the set value and the operating control parameter value after multi-dimensional data preprocessing as the process variable, the basic adjustment amount is calculated for each abnormal control parameter in turn through the PID control algorithm; the main loop of the PID control algorithm is used to realize the target value tracking of the control parameter, and the secondary loop is used to suppress the mechanical disturbance of the actuator itself and the interference of the external environment, which not only ensures the rapid response of the adjustment, but also improves the stability and accuracy of the adjustment.

[0038] Step S33: Based on the preset maximum adjustment threshold, the calculated basic adjustment amount is divided into steps to obtain a step-by-step adjustment amount sequence. The specific method of step-by-step adjustment is as follows: when the absolute value of the basic adjustment amount is less than or equal to the preset maximum adjustment threshold, the step-by-step adjustment amount sequence contains only one adjustment amount, the value of which is equal to the basic adjustment amount; when the absolute value of the basic adjustment amount is greater than the preset maximum adjustment threshold, the required number of adjustment steps is calculated, and the basic adjustment amount is evenly distributed to each adjustment step, so that the absolute value of each step adjustment amount does not exceed the preset maximum adjustment threshold. The preset maximum adjustment threshold is set according to the process characteristics of each control parameter to prevent excessive adjustment in a single step from impacting the production process and to avoid triggering new process fluctuations. For example, the preset maximum adjustment threshold for the reaction temperature is 2.5℃. When the calculated basic adjustment amount is 6℃, it is divided into 3 adjustment steps, with each step adjustment amount being 2℃.

[0039] Step S34: Set the adjustment rate and single-step execution timeout for each adjustment step in the stepwise adjustment sequence; the adjustment rate refers to the adjustment amplitude of the corresponding control parameter by the actuator per unit time, used to precisely control the adjustment speed of the actuator, avoiding system oscillation caused by excessively fast adjustment or increased disproportionation risk due to excessively slow adjustment; the setting is based on the different degrees of influence of each control parameter on the +3 valent antimony disproportionation reaction, using a lower adjustment rate for control parameters with long system recovery time after adjustment, and a higher adjustment rate for control parameters with fast response speed, while appropriately increasing the adjustment rate for key control parameters that directly affect the disproportionation reaction rate under the premise of ensuring system stability; for example, the adjustment rate of reaction temperature is set to 0.5℃ / min, the adjustment rate of drying vacuum degree is set to 0.005MPa / min, and the adjustment rate of nitrogen flow rate is set to 0.5m 3The stirring speed adjustment rate is set to 10 revolutions per minute. The single-step execution timeout refers to the maximum time allowed for a single-step adjustment operation to be completed, i.e., the maximum allowable time from the moment the actuator receives the single-step adjustment command until the corresponding operating control parameter reaches the single-step adjustment target value and stabilizes. This timeout is set based on the response characteristics and adjustment rate of each control parameter, ensuring that the actuator can complete the single-step adjustment within this time under normal operating conditions. If the adjustment is not completed within this time, it is considered an adjustment anomaly, requiring recalculation of the adjustment amount or troubleshooting of equipment malfunctions. For example, the single-step execution timeout for reaction temperature is set to 10 minutes, for drying vacuum degree to 15 minutes, for nitrogen flow rate to 2 minutes, and for stirring speed to 1 minute.

[0040] Step S35: Generate a first control instruction. The first control instruction includes an abnormal control parameter identifier, an adjustment sequence identifier, a sequence of step-by-step adjustment amounts corresponding to each abnormal control parameter, an identifier of the corresponding actuator, an adjustment rate for each adjustment step, and a single-step execution timeout. The adjustment sequence identifier is used to guide the feedback self-correction module to execute the step-by-step adjustment operations of each abnormal control parameter in descending order of priority, ensuring that the parameter with the greatest impact on the disproportionation reaction is corrected first.

[0041] Specifically, the adaptive adjustment module is designed to address the challenges of timely detection of real-time process parameter fluctuations in the production of antimony glycol, the potential for system oscillations due to excessive single adjustments, and the chaotic adjustment sequence when multiple parameters are simultaneously abnormal. In the continuous industrial production of antimony glycol, various uncontrollable factors, such as fluctuations in heating system power, minor leaks in the vacuum system, unstable raw material feed rates, deviations in stirring speed, and changes in ambient temperature, inevitably cause the operating parameters of each step to deviate from their initial range. Traditional production methods rely on manual timed inspections and adjustments, which suffer from inherent drawbacks such as long adjustment lag times, low adjustment accuracy, and the inability to achieve continuous monitoring. Furthermore, manual adjustments are often based on experience, easily leading to excessive single adjustments that cause system oscillations, or prioritizing non-critical parameters that amplify disproportionation risks, ultimately accelerating the disproportionation reaction of +3 antimony and resulting in substandard product quality. The adaptive adjustment module, through a systematic operation involving multi-dimensional data preprocessing, continuous anomaly verification, priority-based adjustment, and step-by-step precise control, achieves rapid detection and automated, stable correction of process parameter anomalies.

[0042] The trend prediction module, in response to the absence of adaptive adjustment, quantifies the disproportionation response quantification index of the current operation stage based on real-time feature parameters, and uses it to predict the disproportionation response trend within a preset time period in the future, and determines whether there is a disproportionation risk. If so, it generates a second control instruction and sends the second control instruction to the feedback self-correction module.

[0043] The preset duration refers to the length of time used to predict the future development trend of the disproportionation reaction. It is set based on balancing the accuracy of the prediction results and the lead time of the control operation. If the prediction duration is too short, the control operation time will be insufficient, and if the prediction duration is too long, the accuracy of the prediction results will be reduced. For example, three preset durations are set: 15 minutes, 30 minutes and 60 minutes.

[0044] Please see Figure 4 Furthermore, the specific steps to determine whether there is a risk of dissemination include: Step S41: Obtain the state representation parameters of the current operation stage.

[0045] Step S42: Quantify the state characterization parameters into disproportionation reaction quantification indicators for the current operation stage. The disproportionation reaction quantification indicators refer to a set of parameters characterizing the degree and rate of +3 valent antimony disproportionation reaction during the preparation of antimony glycol, including the current +5 valent antimony content, the current 0 valent metallic antimony content, disproportionation reaction conversion rate, and disproportionation reaction rate. The disproportionation reaction conversion rate is the ratio of the amount of +3 valent antimony that has undergone disproportionation reaction to the initial total amount of +3 valent antimony. The disproportionation reaction rate is the amount of +3 valent antimony that undergoes disproportionation reaction per unit time.

[0046] Specifically, the current +5 valence antimony content is calculated based on the redox potential of the reaction system. The calculation formula is:

[0047] in, This is a preset redox potential coefficient, with a value ranging from 0.002% / mV to 0.003% / mV. To measure the redox potential, for / The redox potential of the redox couple. The redox potential coefficient refers to the redox potential coefficient under specific temperature, ionic strength, and solvent conditions in the antimony glycol reaction system. / The offset of the redox potential of the redox couple relative to the standard potential is linearly converted into a proportionality constant of the +5 antimony mass concentration in the reaction system, which characterizes the change in +5 antimony content corresponding to a unit change in redox potential, and is determined by the standard solution gradient calibration method. The current content of 0-valent antimony is calculated based on the transmittance of the reaction system at a wavelength of 420 nm. The calculation formula is:

[0048] in, This is a preset transmittance coefficient, with a value ranging from 1.0% to 1.5%. To measure the light transmittance, The transmittance is the light transmittance at the initial moment of the reaction. The transmittance coefficient refers to the proportional constant that linearly converts the proportion of transmittance decay of the reaction system relative to the initial state into the mass concentration of 0-valent antimony in the reaction system under a specific wavelength of 420nm and a fixed optical path length. It characterizes the change in the content of 0-valent antimony corresponding to a unit decrease in transmittance and is determined by the standard suspension calibration method. Calculate the disproportionation reaction conversion rate based on the current +5 valent antimony content and the current 0 valent metallic antimony content:

[0049] in, This represents the initial total concentration of +3 valent antimony.

[0050] The disproportionation reaction rate was calculated based on the disproportionation reaction conversion rates at two consecutive sampling times. :

[0051] in, The disproportionation reaction conversion rate at the current moment. The disproportionation reaction conversion rate at the previous sampling time. This represents the time interval between two adjacent sampling times.

[0052] Step S43: Input the operation control parameter values, state characterization parameter values ​​and current disproportionation reaction quantitative indicators of each operation link after multi-dimensional data preprocessing into the constructed disproportionation reaction prediction model to predict the predicted values ​​of each disproportionation reaction quantitative indicator within a preset time period in the future.

[0053] Furthermore, the steps for constructing the disproportionation reaction prediction model include: Step S431: Construct a model training dataset, which includes historical batch raw material composition feature data, operation control parameter data for each operation step, state characterization parameter data, and disproportionation reaction quantitative index data; preprocess the model training dataset, including using the 3σ criterion to remove abnormal sampling points, and deleting invalid samples with missing data or exceeding detection limits; mapping all continuous numerical features to the [0,1] interval using minimum-maximum normalization to eliminate differences in dimensions and orders of magnitude; and performing time-series alignment at fixed time intervals to form a standardized model training dataset with a unified sampling frequency.

[0054] Step S432: Based on the standardized model training dataset, feature selection is performed using a combination of Pearson correlation coefficient and variance filtering to obtain the input feature set of the disproportionation response prediction model, balancing feature correlation and model computational efficiency. The specific operation is as follows: Calculate the Pearson correlation coefficient between each input feature parameter and each disproportionation reaction quantification index. Weakly correlated input feature parameters with absolute correlation coefficients less than a preset correlation coefficient threshold are removed, while core input feature parameters closely related to the disproportionation reaction are retained. These input feature parameters include raw material composition features, operational control features, and state characterization features. The preset correlation coefficient threshold is set based on statistical analysis of the correlation between each input feature parameter and the disproportionation reaction quantification index in historical production data. This balances feature correlation with the number of features, eliminating weakly correlated features that are almost unrelated to the disproportionation reaction and cannot provide effective information for prediction, while avoiding excessive elimination that could lead to the loss of key features. This ensures that the selected core features accurately represent the changing patterns of the disproportionation reaction, while controlling the model input dimension and reducing computational complexity. For example, the preset correlation coefficient threshold is set to 0.2.

[0055] The variance of each core input feature parameter is calculated, and redundant features with no change that have a variance less than a preset variance threshold are removed to obtain the input feature set. Redundant features with no change cannot provide effective information for model prediction and will increase the computational complexity of the model. The preset variance threshold is set based on the statistical analysis of the fluctuation range of each input feature parameter under normal production conditions, distinguishing between feature fluctuations with practical significance and valueless numerical fluctuations, removing redundant features whose values ​​remain basically stable and cannot reflect changes in the production process, while retaining effective features that can reflect raw material fluctuations and changes in operating conditions. For example, the preset variance threshold is set to 0.01.

[0056] For example, the input feature set includes: reaction temperature, gaseous oxygen content in the reactor, nitrogen flow rate, redox potential of the reaction system, transmittance at 420 nm wavelength, total content of antimony trioxide, iron, and copper impurities, glyoxal content in ethylene glycol, and current disproportionation reaction conversion rate.

[0057] Step S433: Supervised learning samples for the disproportionation response prediction model are generated using a sliding window truncation method. All parameter values ​​of the input feature set at the current time are used as model input samples, and the measured values ​​of the disproportionation response quantification index corresponding to a preset duration are used as the labels for the model input samples, thus fulfilling the core requirement of predicting future disproportionation trends from current features. The generated supervised learning samples are preprocessed to remove invalid samples with abnormal label data, mismatched input feature and label time sequences, or duplicate data, ensuring sample quality. The model input samples are divided into training and test sets according to a preset sample division ratio. The preset sample division ratio is set based on balancing the fitting effect of model training with the requirement to verify generalization ability. It ensures that the training set sample size is sufficient to support the model learning feature association patterns while also ensuring that the test set sample size can effectively verify the model's predictive performance on unknown data. For example, the preset sample division ratio is 8:2.

[0058] Step S434: Construct a data-driven disproportionation reaction prediction model using the gradient boosting tree algorithm, adapting it to the multi-parameter coupling and nonlinear fluctuation process characteristics in antimony glycol production. The mapping relationship between the input feature set and future disproportionation reaction quantification indicators is iteratively fitted. The specific operations are as follows: Set the hyperparameters of the disproportionation response prediction model. The hyperparameters include the number of decision trees, the learning rate, the maximum depth of a single tree, and the minimum number of sample leaf nodes. The setting of each hyperparameter is based on balancing the model's fitting ability and its anti-overfitting ability, and avoiding overfitting or underfitting of the model due to unreasonable hyperparameter settings. For example, the initial hyperparameters are set as follows: the number of decision trees is preset to a range of 80-150 trees, with an initial value of 120 trees; the learning rate is preset to a range of 0.05-0.1 trees, with an initial value of 0.08; the maximum depth of a single tree is preset to a range of 4-7 trees, with an initial value of 5; and the minimum number of sample leaf nodes is preset to a range of 3-5 trees, with an initial value of 4.

[0059] Mean squared error is used as the loss function for training the disproportionate response prediction model. The loss function is set based on the following criteria: it can effectively measure the deviation between the model's predicted value and the actual value of the label, adapt to the regression prediction scenario, and facilitate the model to reduce the deviation through iteration. The decision tree is trained round by round, and each newly generated decision tree fits the prediction residual of the previous model, gradually reducing the loss function value until the loss function value converges or reaches the preset number of iterations. The preset convergence threshold of the loss function value is set based on the following criteria: balancing the model training accuracy and training efficiency, and avoiding excessive iteration that leads to excessively long training time. For example, the preset convergence threshold is 0.0001.

[0060] A multi-objective regression structure design model is adopted to enable the disproportionation reaction prediction model to simultaneously output the predicted values ​​of the +5 valence antimony content, zero valence metallic antimony content, disproportionation reaction conversion rate, and disproportionation reaction rate for the future preset time, ensuring the synchronicity of the prediction results and meeting the disproportionation risk control requirements of different lead times.

[0061] Step S435: Optimize the hyperparameters of the disproportionation response prediction model using a combination of grid search and cross-validation to ensure optimal performance and stable generalization ability. The specific steps are as follows: Based on the preset range of hyperparameters, a hyperparameter combination mesh is constructed, and all hyperparameter combinations are traversed to ensure coverage of a reasonable hyperparameter range, providing comprehensive support for optimization. The cross-validation method with a preset number of folds is adopted. The preset number of folds is set based on balancing validation accuracy and validation efficiency. Too few folds will lead to unstable validation results, while too many folds will increase the validation computation. For example, the preset number of cross-validation folds is 5 folds. The training set is randomly divided into 5 parts, and 4 parts are used as the sub-training set and 1 part is used as the sub-validation set. The model training and validation are completed 5 times in a loop. With minimizing the mean absolute error under 5-fold cross-validation as the optimization objective, the optimal hyperparameter combination is selected. The mean absolute error is set based on the following criteria: it can intuitively reflect the average deviation between the model's predicted value and the measured value, is not affected by extreme values, and is suitable for the prediction scenario of the dissimilarity response quantitative index. For example, the optimized hyperparameter combination is: 130 decision trees, learning rate 0.07, maximum depth of a single tree 6, and minimum number of sample leaf nodes 4.

[0062] Step S436: Use the test set to verify the accuracy of the disproportionation reaction prediction model, forming the final disproportionation reaction prediction model, ensuring that the model's prediction accuracy meets the actual needs of industrial production of antimony glycol. The specific operation is as follows: Accuracy evaluation indicators and preset accuracy thresholds are set. The accuracy evaluation indicators are set based on the following criteria: comprehensively reflecting the accuracy and stability of the model predictions and meeting the actual needs of disproportionation reaction control. The preset accuracy thresholds are set based on the following criteria: combining the quality requirements of antimony glycol products and the accuracy requirements of disproportionation risk control, ensuring that the model prediction results can effectively support subsequent disproportionation risk judgment and control command generation. For example, the accuracy evaluation indicators and preset thresholds are: coefficient of determination R... 2 ≥Preset R 2 Threshold, preset R 2 The threshold is set to measure how well the model fits the data patterns; the closer the value is to 1, the better the fit. An example preset R0... 2Threshold = 0.95; Average relative error ≤ preset MAPE threshold. The preset MAPE threshold is set based on the following: measuring the relative deviation between the predicted value and the measured value. The smaller the value, the higher the prediction accuracy. An example preset MAPE threshold = 3%; Prediction error between +5 valent antimony content and 0 valent metallic antimony content ≤ preset content prediction threshold. The preset content prediction threshold is set based on the following: conforming to product quality control requirements and avoiding misjudgment of disparity risk due to excessive prediction error. An example preset content prediction threshold = 0.03%.

[0063] If all accuracy evaluation indicators meet the preset thresholds, the optimal hyperparameters, weight parameters, and inference structure of the disproportionation reaction prediction model will be solidified to form the final disproportionation reaction prediction model, ensuring that the model can operate stably in the industrial field. If any accuracy indicator does not meet the preset threshold, the input features will be re-selected, or the hyperparameter search range will be adjusted, and the model will be retrained and optimized until all accuracy indicators meet the standards.

[0064] Step S44: Compare the predicted values ​​of the disproportionation reaction quantitative indicators corresponding to each preset time period with the preset disproportionation risk thresholds for each disproportionation reaction quantitative indicator to identify abnormal disproportionation reaction quantitative indicators. Based on the absolute deviation between the abnormal disproportionation reaction quantitative indicators and the corresponding disproportionation risk thresholds, classify the disproportionation risk level. The disproportionation risk threshold is an abnormal judgment threshold set for each of the disproportionation reaction quantitative indicators, such as +5 valent antimony content, 0 valent metallic antimony content, disproportionation reaction conversion rate, and disproportionation reaction rate. It serves as a quantitative benchmark to distinguish between normal process fluctuations and actual disproportionation reaction anomalies. Values ​​below the disproportionation risk threshold are considered to have no disproportionation risk, while values ​​above or equal to the disproportionation risk threshold are considered to indicate the presence of disproportionation. Anomalies are identified by quantitative indicators of abnormal disproportionation reactions. The disproportionation risk threshold is set based on the following criteria: First, the critical value of the quantitative indicator for the start of irreversible decline in product quality is determined according to the national standard for antimony glycol products. Second, statistical analysis is conducted based on historical production data, and the upper limit of normal process fluctuations is determined using the 3σ principle to ensure that normal process fluctuations are not misjudged as abnormal disproportionation reactions. Third, the inflection point of the quantitative indicator change before the disproportionation reaction enters the irreversible acceleration stage is extracted as a reference benchmark, taking into account the kinetic characteristics of the antimony glycol disproportionation reaction, and a safety margin is reserved to prevent missed detection of disproportionation risks due to prediction bias. Finally, the disproportionation risk threshold is iteratively optimized through industrial field verification tests. Specifically, the steps for classifying disproportionation risk levels include: For each identified abnormal disproportionation response quantitative indicator, the absolute deviation between the predicted value of the abnormal disproportionation response quantitative indicator and the corresponding disproportionation risk threshold is calculated. The calculation formula is: Absolute deviation = |Predicted value of abnormal disproportionation response quantitative indicator - Corresponding disproportionation risk threshold|.

[0065] The absolute deviation ranges for the disproportionation risk levels of each quantitative indicator of disproportionation reaction are pre-defined. For example, when the absolute deviation is less than or equal to 0.5 times the corresponding disproportionation risk threshold, the risk level corresponding to the abnormal disproportionation reaction quantitative indicator is determined to be Level 1 risk. At this point, the disproportionation reaction is in its early nascent stage, with a slow development rate, and can be completely suppressed through small-scale preventative adjustments. When the absolute deviation is greater than 0.5 times but less than or equal to 1.0 times the corresponding disproportionation risk threshold, the risk level corresponding to the abnormal disproportionation reaction quantitative indicator is determined to be Level 2 risk. At this point, the disproportionation reaction has entered an accelerated development stage, and will rapidly deteriorate if not controlled in time, requiring moderate-scale targeted control measures. When the absolute deviation is greater than 1.0 times the corresponding disproportionation risk threshold, the risk level corresponding to the abnormal disproportionation reaction quantitative indicator is determined to be Level 3 risk. At this point, the disproportionation reaction is about to enter an irreversible and uncontrollable stage, requiring immediate high-intensity comprehensive control measures to avoid irreversible defects in product quality. The absolute deviation ranges for the disproportionation risk levels of each quantitative indicator of disproportionation reaction are set based on statistical analysis of historical abnormal disproportionation reaction samples.

[0066] If multiple abnormal disproportionation response quantitative indicators exist simultaneously, the risk level corresponding to each abnormal disproportionation response quantitative indicator is determined, and the highest level among all risk levels is selected as the overall disproportionation risk level for this case.

[0067] Step S45: When there is at least one abnormal disproportionation reaction quantitative indicator, it is determined that there is a disproportionation risk, and the second regulatory instruction is generated.

[0068] Furthermore, the specific steps for generating the second control command include: Step S51: Based on the abnormal disproportionation reaction quantitative index types identified in Step S44 and the obtained disproportionation risk levels, a preset graded advanced disproportionation suppression and control strategy is matched. This strategy is used to define corresponding target control parameters for different risk levels and abnormal disproportionation reaction quantitative index types, and to set a standardized baseline adjustment degree for each target control parameter. Simultaneously, the adjustment direction constraints for control parameters corresponding to different abnormal disproportionation reaction quantitative index types are clarified. The baseline adjustment degree is the basic adjustment range of the control parameter under the unit absolute deviation between the abnormal disproportionation reaction quantitative index and the corresponding disproportionation risk threshold. Its setting is based on the effective control patterns and process stability requirements when different disproportionation reaction quantitative indexes are abnormal in historical production. For Level 1 risk with single or multiple abnormal disproportionation reaction quantitative indexes, the matched target control parameters are reaction temperature and nitrogen flow rate. For Level 2 risk with a higher degree of index deviation, the matched target control parameters are based on the former plus reaction pressure. For example, the baseline adjustment degrees for reaction temperature, nitrogen flow rate, and reaction pressure are 0.3℃ / unit absolute deviation and 0.1m, respectively. 3 / h / unit deviation, 0.008MPa / unit deviation.

[0069] Step S52: For each type of identified abnormal disproportionation response quantitative indicator, calculate its absolute deviation from the threshold corresponding to the current disproportionation risk level. The absolute deviation is equal to the absolute value of the difference between the measured value of the abnormal disproportionation response quantitative indicator and the corresponding level threshold. If there are multiple abnormal disproportionation response quantitative indicators, calculate the deviation of each indicator separately and select the absolute deviation with the largest value as the benchmark deviation. Use this benchmark deviation as the core basis for calculating the advance adjustment amount of each target control parameter in the future, so as to ensure that the control intensity can match the most significant degree of disproportionation abnormality.

[0070] Step S53: According to the matched graded advanced disproportionation suppression control strategy, the baseline deviation is multiplied by the baseline adjustment degree corresponding to each target control parameter to obtain the initial advanced adjustment amount. Simultaneously, the adjustment direction of each target control parameter is determined by combining the abnormal disproportionation reaction quantitative index type. That is, when the +5 valence antimony and 0 valence antimony contents, disproportionation conversion rate, and rate are high, the reaction temperature is reduced, the nitrogen flow rate is increased, and the reaction pressure is slightly reduced. Then, boundary constraints are applied to the initial advanced adjustment amount. The absolute value of the initial advanced adjustment amount is compared with the preset maximum advanced adjustment amount threshold. If it exceeds the maximum advanced adjustment amount threshold, the maximum advanced adjustment amount threshold is used as the final advanced adjustment amount. The setting of this maximum advanced adjustment amount threshold is based on the fact that advanced control is a preventative intervention, and the adjustment range must be significantly smaller than adaptive control to avoid process fluctuations. For example, the maximum advanced adjustment amount thresholds for reaction temperature, nitrogen flow rate, and reaction pressure are 1℃, 0.3m, and 0.3m, respectively. 3 / h, 0.02MPa.

[0071] Step S54: Configure appropriate advance adjustment rates and adjustment execution timeouts for the final advance adjustment amounts of each target control parameter; the advance adjustment rate refers to the adjustment range of the target control parameter by the actuator per unit time, used to control the speed of preventive advance adjustment and avoid excessive adjustment causing oscillations in the process system. By following the core principle of smooth and gradual adjustment, it can be uniformly set to 40%-60% of the adaptive adjustment rate of the corresponding parameter. At the same time, the specific value is determined according to the differences in process inertia of each target control parameter: reaction temperature and reaction pressure with larger process inertia use lower adjustment rates, while nitrogen flow rate and stirring speed with smaller process inertia use moderate adjustment rates; the adjustment execution timeout... The timeout period refers to the maximum allowable time from when the actuator receives the advance adjustment command until the corresponding operating control parameter reaches the target value and stabilizes. It is used to determine whether the actuator has completed the adjustment within a reasonable time. It is determined using a standardized calculation method combining theoretical adjustment time and process response safety margin. The calculation formula is: Adjustment execution timeout = (Final advance adjustment amount of the target control parameter ÷ Corresponding advance adjustment rate) × Process response safety factor. The process response safety factor is set according to the process response characteristics of each target control parameter: for reaction temperature and reaction pressure with high process inertia, the process response safety factor can be 3.0 and 2.5 respectively; for nitrogen flow rate and stirring speed with low process inertia, the process response safety factor can be 1.5. For example, the adaptive adjustment rate of the reaction temperature is 0.5℃ / min, the corresponding advance adjustment rate is set to 0.2℃ / min, the maximum final advance adjustment amount is 1℃, the theoretical adjustment time is 5 minutes, and the adjustment execution timeout is 15 minutes; the adaptive adjustment rate of the nitrogen flow rate is 0.5m... 3 / h / minute, corresponding to an advance adjustment rate set to 0.2m. 3 / h / minute, with a maximum final lead adjustment of 0.3m. 3 / h, the theoretical adjustment time is 1.5 minutes, and the adjustment execution timeout is 3 minutes; the adaptive adjustment rate of the reaction pressure is 0.01MPa / min, the corresponding advance adjustment rate is set to 0.005MPa / min, the maximum final advance adjustment amount is 0.02MPa, the theoretical adjustment time is 4 minutes, and the adjustment execution timeout is 10 minutes.

[0072] Step S55: Integrate the disproportionation risk level identifier, abnormal disproportionation response quantitative indicator type identifier, each target control parameter identifier, final advance control amount with adjustment direction, corresponding actuator identifier, advance control rate, and control execution timeout time to generate a standardized second control instruction. This ensures that the feedback self-correction module can execute preventive advance control in an orderly manner according to the instruction, effectively blocking the development trend of disproportionation response before it occurs on a large scale.

[0073] Specifically, the trend prediction module is designed to address the problems of difficulty in identifying early signs of disproportionation reactions during the preparation of antimony glycol, the inability to intervene proactively due to reactive measures, and the resulting product quality fluctuations caused by passive adjustments. In the preparation of antimony glycol, the disproportionation reaction of +3 valent antimony is a slow process. Even when the operating parameters are still within the initial control range, state characterization parameters such as redox potential and transmittance within the reaction system begin to change, indicating potential disproportionation risks. Traditional control methods can only adjust the system after the operating parameters exceed the range or product quality problems occur, which is a reactive measure. By this time, the disproportionation reaction has already occurred, and even with adjustments, it is impossible to completely eliminate the generated 0-valent and +5-valent antimony, leading to quality defects such as haze and black spots in the product. Furthermore, the passive adjustments are often large and prone to causing system oscillations. The trend prediction module, through a systematic operation of real-time state quantification, hybrid model trend prediction, graded risk assessment, and differentiated proactive intervention, achieves early warning and proactive prevention of disproportionation reactions.

[0074] The feedback self-correction module receives the first and second control commands, executes differentiated disproportionation suppression control, calculates the control effect evaluation parameters based on the disproportionation reaction quantification index after the execution of the second control command, identifies effective control samples, self-corrects the initial control parameter range, generates an updated control parameter range, and writes the updated control parameter range back to the raw material composition and control parameter mapping data table of the initialization module, replacing the original initial control parameter range for subsequent batch production control of the same type of raw materials.

[0075] Furthermore, the specific steps for implementing differentiated disproportionation inhibition control and calculating the evaluation parameters of the control effect based on the disproportionation response quantification index after the execution of the second control instruction include: Step S61: Priority judgment is performed on the received first and second control commands. The first control command corresponds to emergency correction control when the operating control parameters have deviated from the initial control parameter range. It has a higher priority than the second control command, which is a preventive advance control. The second control command is executed only when the first control command is not received and only the second control command is triggered. According to the target control parameter identifier, adjustment direction, final advance adjustment amount, advance adjustment rate and adjustment execution timeout in the second control command, the corresponding actuator is driven to complete a smooth preventive advance adjustment operation. The changing trend of each operating control parameter is monitored in real time throughout the process to ensure that the adjustment process does not exceed the process safety operation boundary specified by industry general standards and enterprise production specifications.

[0076] Step S62: After all the second control commands have been executed and a preset stabilization period has elapsed, continuously collect the state characterization parameters of the reaction system within the preset monitoring period. After removing instantaneous fluctuation data and detection noise interference, use the same quantitative calculation method as the trend prediction module to calculate the quantitative index of stable disproportionation reaction after advance control, including +5 valence antimony content, 0 valence metallic antimony content, disproportionation reaction conversion rate and disproportionation reaction rate. Take the average value of each disproportionation reaction quantitative index within the preset monitoring period as the final value of the disproportionation reaction quantitative index after control, and classify the disproportionation risk level after control according to the same grading standard as the trend prediction module. The preset stabilization time refers to the time required for the reaction system to recover from the adjustment disturbance and reach a stable operating state after the second control command is executed. Its setting is based on the inertial characteristics of the antimony glycol preparation process, ensuring that the reaction system has sufficient time to recover from the adjustment disturbance after the actuator completes the adjustment action, and that all operating control parameters and state characterization parameters reach a stable equilibrium state. This avoids distortion in the calculation of the disproportionation reaction quantitative index after control due to system instability. Simultaneously, it avoids an excessively long stabilization time, which could affect system operating efficiency and the timeliness of subsequent correction processes. An example preset stabilization time is 10 minutes. The preset monitoring time refers to the continuous time used to collect stable state data after control and calculate the average disproportionation reaction quantitative index. Its setting is based on ensuring that the amount of data collected is sufficient to reflect the true stable operating state of the system after control, eliminating the impact of accidental fluctuations on the evaluation results. An example preset monitoring time is 5 minutes.

[0077] Step S63: Based on the predicted abnormal disproportionation reaction quantitative indicators and corresponding disproportionation risk levels before the advance control, calculate the evaluation parameters of the control effect of this advance control. These evaluation parameters are comprehensive quantitative indicators used to quantify the effectiveness of advance control in suppressing potential disproportionation risks and characterize the operational stability of the process system after control. They include the reduction rate of deviation of the disproportionation reaction quantitative indicators, the risk level downgrade rate, and the stability of indicator fluctuations. The reduction rate of deviation of the disproportionation reaction quantitative indicators refers to the percentage decrease in the deviation of the abnormal disproportionation reaction quantitative indicators relative to the corresponding risk threshold before and after control. The calculation formula is as follows: The reduction rate of deviation of the quantitative index of disproportionation response = (deviation of quantitative index of disproportionation response before regulation - deviation of quantitative index of disproportionation response after regulation) ÷ deviation of quantitative index of disproportionation response before regulation × 100%.

[0078] The risk level downgrade rate refers to the decrease in the disparity risk level after regulation relative to before regulation. It directly reflects the degree of elimination of disparity risk. The calculation formula is as follows: Risk level downgrade rate = (risk level value before regulation - risk level value after regulation) ÷ risk level value before regulation × 100%.

[0079] The stability of index fluctuation refers to the degree of numerical dispersion of the quantitative index of disproportionation reaction after regulation during the stable monitoring period, reflecting the operational stability of the process system. It is obtained by calculating the variance of the quantitative index of disproportionation reaction during the stable monitoring period.

[0080] Step S64: Based on the evaluation parameters of the control effect, determine whether the current advance control is a positive and effective control. When the deviation reduction rate of the disproportionation reaction quantitative index is ≥ the preset effective reduction rate threshold, the risk level downgrade rate is ≥ 100%, and the index fluctuation stability is ≤ the preset stability threshold, it is determined to be a positive and effective control. At this time, associate the raw material principal component factor classification combination corresponding to this production batch, the initial control parameter range, the actual control parameters of the second control instruction, the control parameters of stable operation after control, and the control effect evaluation parameters to form a standardized effective control sample and archive it. After a single positive and effective control is completed, the initial control parameter range is not immediately corrected, only the sample data is accumulated and archived. If it is determined to be an ineffective control, the sample is not archived, and the subsequent correction process is not triggered. The preset effective reduction rate threshold refers to the minimum reduction ratio of the deviation of the quantitative indicator of the disproportion response required for preemptive regulation to achieve an effective level of disproportion risk suppression. Its setting is based on balancing the actual inhibitory effect of preemptive regulation with the selection criteria for effective samples, ensuring that the selected effective regulatory samples truly exert a significant inhibitory effect on potential disproportion risk and provide a reliable basis for correcting the initial regulatory parameter range. If the effective reduction rate threshold is set too low, regulation with weak inhibitory effects will be included in the effective samples, resulting in the corrected range failing to effectively avoid disproportion risk. If the effective reduction rate threshold is set too high, the number of effective samples will be significantly reduced. Extend the interval correction period; an exemplary preset effective reduction rate threshold is 80%. The preset stability threshold refers to the maximum variance of the disproportionation reaction quantification index required for the controlled process system to reach stable operation. Its setting is based on ensuring the controlled process system operates smoothly without drastic fluctuations, avoiding new process instability caused by the control operation. If the variance exceeds this threshold, it indicates that the controlled system still has significant fluctuations; the operating parameters of such samples are not representative and cannot be used for correcting the initial control parameter interval, otherwise it will lead to a decrease in the stability of the corrected interval. An exemplary preset stability threshold is 0.0001%. 2 .

[0081] Furthermore, the specific steps for self-correcting the initial control parameter range to generate an updated control parameter range include: Step S71: Classify and statistically analyze the archived effective control samples according to the graded combination of raw material principal component factors. Count the number of effective control samples corresponding to each graded combination of raw material principal component factors. When the number of effective control samples for a certain graded combination of raw material principal component factors reaches a preset sample accumulation threshold, initiate the self-correction process of the initial control parameter range corresponding to that graded combination of raw material principal component factors. The preset sample accumulation threshold is set to ensure that the statistical sample size is sufficiently representative, can reflect the real process adaptation rules of this type of raw material, and avoids unreasonable correction results due to random deviations of a single or small number of samples. An exemplary preset sample accumulation threshold is 30 groups.

[0082] Step S72: Extract all archived effective control samples under the principal component factor classification combination of the raw material, perform statistical analysis on the stable operation control parameters in the samples that have no disproportionation risk after control, calculate the statistical mean and standard deviation of each control parameter, clarify the deviation pattern between the original initial control parameter range and the actual risk-free operation parameters, and determine the correction direction and correction magnitude of the initial control parameter range.

[0083] Step S73: Based on the statistical mean and standard deviation of each control parameter obtained from statistical analysis, the initial control parameter range corresponding to the graded combination of the main component factors of the raw material is corrected. The statistical mean is used as the median of the updated initial control parameter range, and the upper and lower limits of the updated initial control parameter range are respectively determined by adding or subtracting the preset interval width coefficient multiple of the standard deviation from the statistical mean, thus forming the updated control parameter range. The preset interval width coefficient refers to the standard deviation multiple used to determine the distance between the upper and lower limits of the updated initial control parameter range and the statistical mean. Its setting is based on balancing the coverage and safety of the interval. It should ensure that the stable operating parameters of the vast majority of effective control samples are included in the interval, while avoiding the interval being too wide, which would lead to a decrease in the disproportionation inhibition effect. An exemplary preset interval width coefficient is 1.5. The corrected and updated control parameter range is then subjected to process safety verification to ensure that it is within the process safety operating range specified by industry general standards and enterprise production specifications.

[0084] Step S74: Write the updated control parameter range after verification back to the raw material composition and control parameter mapping data table of the initialization module, and replace the original initial control parameter range corresponding to the graded combination of the main component factors of the raw material, to complete the iterative optimization of the mapping relationship; in subsequent production batches with the same graded combination of the main component factors of the raw material, the updated initial control parameter range will be directly called for initial control, reducing the probability of potential disproportionation risk from the source and significantly reducing the number of times preventive advance control is triggered.

[0085] Specifically, the feedback self-correction module is designed to address the problem of fixed initial control parameter ranges in the preparation of antimony glycol, which cannot be iteratively optimized with production data and leads to frequent triggering of advance control. In traditional production control methods, the initial control parameter range is fixed once determined and cannot be dynamically optimized based on the actual control effect in the production process. This results in some raw material batches frequently exhibiting potential disproportionation risks even when operating within the initial control parameter range, requiring repeated advance control operations, increasing the system control burden and process fluctuation risks. The feedback self-correction module only uses stable operating parameters that have been verified multiple times and can effectively suppress disproportionation risks for correcting the initial control parameter range. This avoids random deviations from single controls and achieves continuous optimization of the initial control parameter range. As the number of production batches increases, the initial control parameter range will increasingly conform to the actual process requirements of different raw material batches, gradually achieving the goal of suppressing disproportionation reactions from the source without additional advance control, significantly improving the stability of the control system and the consistency of product quality.

[0086] Working principle and its effects: This invention addresses the pain point of effectively controlling the disproportionation reaction in the preparation of antimony glycol. Through the synergistic linkage of four modules, it achieves precise control of the disproportionation reaction throughout the entire process, fundamentally improving the stability of the production process and the consistency of product quality.

[0087] Principal component analysis was used to extract the principal component factors of raw materials that significantly affect the disproportionation reaction. A hierarchical mapping relationship was constructed, and personalized initial control parameter ranges precisely matched to the characteristics of the raw materials were generated. This approach adapts to the differences between different batches of raw materials from the source, significantly reducing the inherent risk of disproportionation reaction caused by raw material fluctuations. During production, the adaptive adjustment module eliminates industrial noise interference through multi-dimensional data preprocessing and continuously verifies abnormal operating parameters to avoid misjudgments. Combined with the priority ranking of control parameters and a step-by-step precise adjustment mechanism, process deviations are corrected promptly and smoothly. This ensures timely response to abnormalities while avoiding system oscillations caused by excessive single adjustments, preventing the amplification of abnormal parameters from exacerbating the disproportionation reaction. When the process is in... In steady state, the trend prediction module quantifies the real-time state characterization parameters into disproportionation response quantitative indicators. It uses a gradient boosting tree model trained and optimized with a large amount of historical data to predict future disproportionation trends. Combined with a hierarchical advance inhibition strategy, it performs differentiated preventive regulation, intervening in advance before the disproportionation response occurs, realizing the transformation from passive remediation to proactive prevention. The feedback self-correction module quantitatively evaluates the effect of advance regulation, selects effective regulation samples and accumulates and archives them. When the sample size reaches the statistical representativeness requirement, it iteratively optimizes the initial regulation parameter range based on the statistical analysis results. This enables the system to continuously learn and optimize with production data, constantly improving the adaptability of the initial parameters and gradually reducing the trigger frequency of preventive regulation.

[0088] In summary, this invention organically integrates raw material characteristic matching, real-time adaptive adjustment, trend prediction, and feedback self-correction to form an intelligent control system with self-learning capabilities. This not only effectively solves the problems of poor adaptability to raw material fluctuations and lagging control of disproportionation reactions in traditional production, but also significantly reduces the cost and energy consumption of manual control, providing reliable technical support for the stable industrial production of antimony glycol.

[0089] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A smart control system for inhibiting the disproportionation of antimony glycol, characterized in that, include: The initialization module is used to generate an initial control parameter range based on the characteristic data of the raw material composition for the current preparation of antimony glycol and a preset mapping relationship between the raw material composition and the control parameter. The adaptive adjustment module is used to collect real-time characteristic parameters of each operation in the preparation of antimony glycol, compare them with the initial control parameter range, determine whether adaptive adjustment is triggered, and if so, identify abnormal control parameters, calculate the adjustment amount of abnormal control parameters, and generate the first control command; the real-time characteristic parameters include operation control parameters and state characterization parameters. The trend prediction module, in response to the absence of adaptive adjustment, quantifies the disproportionation response quantitative index of the current operation stage based on real-time feature parameters, and uses it to predict the disproportionation response trend within a preset time period in the future, and determines whether there is a risk of disproportionation. If so, it generates a second control instruction. The feedback self-correction module receives the first regulation instruction and the second regulation instruction, performs differentiated disproportionation suppression control, calculates regulation effect evaluation parameters based on the disproportionation response quantification index after the execution of the second regulation instruction, identifies effective regulation samples, self-corrects the initial regulation parameter range, and generates an updated regulation parameter range. The steps for determining whether there is a risk of misrepresentation include: Obtain the state characterization parameters of the current operation stage, and quantify the state characterization parameters into the disproportionation reaction quantification index of the current operation stage; the disproportionation reaction quantification index includes the current +5 valence antimony content, the current 0 valence metallic antimony content, the disproportionation reaction conversion rate, and the disproportionation reaction rate; The operation control parameter values, state characterization parameter values, and current disproportionation reaction quantitative indicators of each operation link, after multi-dimensional data preprocessing, are input into the constructed disproportionation reaction prediction model to predict the predicted values ​​of each disproportionation reaction quantitative indicator within a preset time period in the future. The predicted values ​​of the quantitative indicators of disproportionation response for each preset time period in the future are compared with the preset disproportionation risk thresholds one by one to identify abnormal quantitative indicators of disproportionation response. Based on the absolute deviation between the abnormal quantitative indicators of disproportionation response and the corresponding disproportionation risk thresholds, the disproportionation risk level is classified. When at least one abnormal disproportionation response quantitative indicator exists, a disproportionation risk is determined, and a second regulatory instruction is generated in conjunction with a pre-set graded advanced disproportionation inhibition and control strategy.

2. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 1, characterized in that, The mapping relationship between raw material components and control parameters refers to a hierarchical mapping data table that reflects the quantitative correspondence between raw material component characteristic data and corresponding control parameter ranges. Its structure is as follows: Principal component factors were extracted from the raw material component characteristic data using principal component analysis. Each principal component factor is divided into a predetermined number of graded intervals according to its influence on the disproportionation reaction. Establish a one-to-one correspondence between all hierarchical combinations of the predetermined number of principal component factors and the control parameter ranges of each operation step.

3. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 2, characterized in that, The step of generating the initial control parameter range includes: Obtain raw material component characteristic data and preprocess the obtained raw material component characteristic data; Extract the principal component factor vector from the preprocessed raw material component feature data; The score of each principal component factor in the principal component factor vector is compared with the corresponding principal component factor grading interval one by one to determine the grading interval to which each principal component factor belongs, and the principal component factor grading combination of the current batch of raw materials is obtained. Based on the principal component factor hierarchical combination, query the preset mapping relationship between raw material components and control parameters to obtain the initial control parameter range of each operation step that matches the current raw material components.

4. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 1, characterized in that, The step of determining whether adaptive adjustment is triggered includes: Receive the initial control parameter range and the real-time feature parameters uploaded by each operation stage, and perform multi-dimensional data preprocessing on the real-time feature parameters; Construct a mapping table corresponding to the control parameters, the operating control parameters, and the initial control parameter range; Traverse the corresponding mapping table and compare the preprocessed multi-dimensional data operation control parameters with the upper and lower limits of the corresponding initial control parameter range one by one to identify potential abnormal operation control parameters that exceed the initial control parameter range. Continuous verification of potential abnormal operation control parameters is carried out by collecting potential abnormal operation control parameter data for a preset number of consecutive sampling periods. When the potential abnormal operation control parameter exceeds the initial control parameter range within the preset number of consecutive sampling periods, it is marked as a real process abnormal parameter. When at least one real process anomaly parameter exists, adaptive adjustment is triggered.

5. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 4, characterized in that, The step of generating the first control command includes: Identify the abnormal control parameters corresponding to all real process abnormal parameters, and determine the adjustment order of each abnormal control parameter according to the preset control parameter priority table; According to the adjustment order, the initial control parameter range midpoint is used as the set value, and the operation control parameter value after multi-dimensional data preprocessing is used as the process variable. The basic adjustment amount is calculated for each abnormal control parameter in turn. Based on the preset maximum adjustment threshold, the calculated basic adjustment amount is divided into step-by-step adjustments to obtain a step-by-step adjustment amount sequence. Set the adjustment rate and single-step execution timeout for each adjustment step in the step-by-step adjustment sequence; Generate a first control instruction, the content of which includes an abnormal control parameter identifier, a control sequence identifier, a step-by-step control quantity sequence corresponding to each abnormal control parameter, a corresponding actuator identifier, a control rate for each control step, and a single-step execution timeout.

6. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 1, characterized in that, The steps for constructing the disproportionation reaction prediction model include: A model training dataset is constructed, which includes historical batch raw material composition feature data, operation control parameter data of each operation step, state characterization parameter data, and disproportionation reaction quantitative index data; the model training dataset is preprocessed to form a standardized model training dataset. Based on the standardized model training dataset, feature selection is performed to obtain the input feature set of the disproportionation response prediction model; Supervised learning samples for the disproportionation response prediction model are generated using a sliding window truncation method. All parameter values ​​of the input feature set at the current time are used as model input samples, and the measured values ​​of the disproportionation response quantification index corresponding to a preset time duration are used as the labels of the model input samples. The model input samples are divided into training and test sets according to a preset sample division ratio. A gradient boosting tree algorithm is used to construct a disproportionation reaction prediction model; a combination of grid search and cross-validation is used to optimize the hyperparameters of the disproportionation reaction prediction model; and a test set is used to verify the accuracy of the disproportionation reaction prediction model, thus forming the final disproportionation reaction prediction model.

7. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 1, characterized in that, The hierarchical preemptive disproportionation suppression and control strategy is used to define corresponding target control parameters for different risk levels and combinations of abnormal disproportionation response quantitative indicator types, and to set a standardized benchmark adjustment degree for each target control parameter, while clarifying the adjustment direction constraints of the control parameters corresponding to different abnormal disproportionation response quantitative indicator types.

8. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 7, characterized in that, The step of generating the second control command includes: Based on the quantitative index type of abnormal disproportionation response and the classified disproportionation risk level, a pre-set graded advanced disproportionation inhibition and regulation strategy is matched. For each type of identified abnormal disproportionation response quantitative index, the absolute deviation of the index from the threshold corresponding to the current disproportionation risk level is calculated to obtain the benchmark deviation. According to the graded advance disproportionation suppression regulation strategy, the initial advance regulation amount is obtained by multiplying the benchmark deviation by the benchmark regulation degree corresponding to each target regulation parameter, and the regulation direction of each target regulation parameter is determined by combining the abnormal disproportionation response quantification index type. If the absolute value of the initial advance adjustment exceeds the maximum advance adjustment threshold, the maximum advance adjustment threshold is used as the final advance adjustment, and a second control command is generated.

9. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 1, characterized in that, The steps of calculating the evaluation parameters of the regulatory effect and identifying effective regulatory samples based on the disproportionation response quantification index after the execution of the second regulatory instruction include: After all the second control commands have been executed and a preset stabilization period has elapsed, the state characterization parameters of the reaction system are continuously collected within the preset monitoring period, and the quantitative index of the stable disproportionation reaction after the advance control is calculated. The average value of each disproportionation response quantitative indicator within the preset monitoring period is taken as the final disproportionation response quantitative indicator value after regulation, and the disproportionation risk level is classified. Based on the quantitative indicators of abnormal disproportionation response predicted before the advance regulation and the corresponding disproportionation risk level, the evaluation parameters of the regulation effect of this advance regulation are calculated; the evaluation parameters of the regulation effect include the deviation reduction rate of the quantitative indicators of disproportionation response, the risk level downgrade rate, and the stability of indicator fluctuation. Based on the evaluation parameters of the regulation effect, it is determined whether the current advance regulation is a positive and effective regulation. If so, the effective regulation sample is formed by associating the graded combination of raw material principal component factors, the initial regulation parameter range, the actual regulation parameters of the second regulation instruction, the regulation parameters of stable operation after regulation, and the evaluation parameters of the regulation effect.

10. The intelligent control system for inhibiting the disproportionation of antimony glycol as described in claim 9, characterized in that, The step of self-correcting the initial control parameter range to generate an updated control parameter range includes: The archived effective control samples are classified and statistically analyzed according to the graded combination of raw material principal component factors. The number of effective control samples corresponding to each graded combination of raw material principal component factors is counted. If the preset sample accumulation threshold is reached, the self-correction process of the initial control parameter range corresponding to the graded combination of raw material principal component factors is initiated. Extract all archived effective control samples under the principal component factor classification combination of the raw material, perform statistical analysis, and calculate the statistical mean and standard deviation of each control parameter respectively; The statistical mean is used as the median of the updated initial control parameter range. The upper and lower limits of the updated initial control parameter range are respectively the statistical mean plus or minus the preset range width coefficient and the standard deviation, forming the updated control parameter range, which is then written back to the raw material composition and control parameter mapping data table.

Images