Phosphoric acid production whole process multivariable model prediction control method and system

Abstract

The invention relates to the technical field of phosphoric acid production control, solves the problem that in the prior art, accurate control over the whole process of phosphoric acid production cannot be achieved, and provides a multivariable model prediction control method and system for the whole process of phosphoric acid production. The method comprises the following steps: acquiring key process data of a plurality of process units in the whole phosphoric acid production process in real time; preprocessing the key process data to obtain real-time standardized data; establishing a soft measurement model according to a pre-collected historical standardized data set; inputting the real-time standardized data into the soft measurement model to obtain a soft measurement predicted value; constructing a multivariable dynamic model according to the historical standardized data set and the soft measurement predicted value; according to the multivariable dynamic model, in combination with a model prediction control algorithm, obtaining target set values of a plurality of preset operation variables; and according to the target set value, realizing control of the whole process of phosphoric acid production. According to the invention, accurate control of the whole process of phosphoric acid production can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of phosphoric acid production control, and particularly relates to a phosphoric acid production whole-process multivariable model predictive control method and system. BACKGROUND

[0002] As an important chemical raw material and fertilizer production basic raw material, phosphoric acid has a wide application in agriculture, chemical industry and food industry. With the expansion of industrialization scale and the improvement of product quality and production efficiency requirements, phosphoric acid production has developed into a complex continuous production process with high temperature and high pressure, multiple process units and multivariable coupling. The whole production process includes multiple key process links such as reaction, filtration, concentration and acid storage, and the running state and process parameters of each link have a direct impact on the final product quality and production efficiency.

[0003] In the phosphoric acid production process, due to the change of raw material properties, reaction condition fluctuation and complex coupling characteristics of multiple process units, the production process is prone to product quality deviation or energy waste. Therefore, accurate control of the whole process of phosphoric acid production can not only ensure stable product quality, but also improve resource utilization, reduce energy consumption and production cost, which is of great significance to the sustainable development of modern phosphoric acid industry.

[0004] The existing phosphoric acid production process control technology mainly relies on traditional experience rules, PID control or single variable feedback control method. In some high-end applications, soft measurement model or model predictive control method based on single process unit is also tried to be used for local optimization control. However, these methods are often aimed at single process unit or a small number of key parameters, lack of systematic modeling of dynamic coupling relationship between multiple process units in the whole process, and have limited real-time prediction and constraint optimization processing capacity for key quality parameters. Therefore, the main problems existing in the prior art include: on the one hand, the comprehensive characterization of input and output variables and dynamic coupling characteristics of multiple process units in the whole process of phosphoric acid production is lacking, so that the control strategy is difficult to fully consider the mutual influence between process units, which easily leads to insufficient control accuracy; on the other hand, the traditional control method is difficult to realize real-time prediction and rolling optimization of key quality parameters, and lacks constraint optimization means based on soft measurement prediction value, which leads to that the operation variable setting cannot meet the best control demand under complex multi-constraint conditions.

[0005] Therefore, how to realize accurate control of the whole process of phosphoric acid production through dynamic coupling modeling of each process unit in the whole process of phosphoric acid production is a technical problem to be solved. SUMMARY

[0006] Therefore, the present application provides a phosphoric acid production whole-process multivariable model predictive control method and system to solve the problem that the prior art cannot realize accurate control of the whole process of phosphoric acid production.

[0007] The technical scheme adopted by the present application is: In a first aspect, the present application provides a phosphoric acid production whole-process multivariable model predictive control method, which comprises: Real-time collection of key process data of multiple process units in the phosphoric acid production whole process, wherein the process units include reaction units, filtration units, concentration units and acid storage units, and the key process data includes temperature, pressure, flow and ingredient concentration; Preprocessing of the key process data to obtain real-time standardized data meeting modeling requirements; Establishment of a soft-sensing model for predicting key quality parameters according to a pre-collected historical standardized data set; Input of the real-time standardized data into the soft-sensing model for online prediction of the running process state of the reaction unit to obtain soft-sensing prediction values related to product quality; Construction of a multivariable dynamic model for representing dynamic coupling relationships between the process units and process constraint conditions according to the historical standardized data set and the soft-sensing prediction values; Rolling optimization calculation of the phosphoric acid production whole process according to the multivariable dynamic model in combination with a pre-set model predictive control algorithm to obtain target set values of multiple pre-set operation variables; Control of the phosphoric acid production whole process according to the target set values. Preferably, the preprocessing of the key process data to obtain standardized data meeting modeling requirements comprises: Noise filtering processing and outlier rejection of the key process data to obtain multi-source process data; Time alignment and interpolation processing of the multi-source process data according to the sampling periods of the process units to obtain time series data; Missing value compensation and consistency verification of the time series data to obtain verification data; Normalization processing of the verification data to obtain the standardized data. Preferably, the establishment of a soft-sensing model for predicting key quality parameters according to a pre-collected historical standardized data set comprises: Feature correlation analysis and principal component analysis of the historical standardized data set, and screening of feature variables related to the key quality parameters as input feature vectors according to the analysis results; Pairing of the input feature vectors with historical measurement values of corresponding key quality parameters to form a training sample set for soft-sensing modeling; Based on the training sample set, a multivariate regression method, a neural network method or a time series deep learning method is used for modeling training to obtain an initial model; The initial model is cross-validated and error-analyzed, and the model parameters are optimized and updated based on the generalization performance of the initial model to obtain the soft measurement model. Preferably, the constructing a multivariate dynamic model for characterizing the dynamic coupling relationship between the process units and the process constraint conditions according to the historical standardized data set and the soft measurement prediction value comprises: According to the historical standardized data set, the key process parameters and quality parameters of each process unit are extracted to form an initial modeling data matrix; The soft measurement prediction value is associated with the initial modeling data matrix to obtain an extended modeling data matrix, wherein the extended modeling data matrix is used to describe the input-output relationship and coupling characteristics; A preliminary dynamic relationship model characterizing the dynamic coupling relationship between the process units and the process constraint conditions is established by applying a multivariate identification technology to the extended modeling data matrix; According to the preliminary dynamic relationship model, the dynamic coupling coefficients and constraint conditions of the model are optimized and calibrated in combination with a preset fuzzy logic control rule to obtain the multivariate dynamic model. Preferably, the associating the soft measurement prediction value with the initial modeling data matrix to obtain an extended modeling data matrix comprises: According to the initial modeling data matrix, the soft measurement prediction value is matched according to the time sequence to obtain a matched soft measurement prediction value; The matched soft measurement prediction value and the corresponding key process parameters of the process unit are associated to obtain a preliminary associated soft measurement data set; According to the preliminary associated soft measurement data set, the initial modeling data matrix is column-extended, and the corresponding soft measurement prediction value column is added to the input variable and the output variable of each process unit to obtain a preliminary extended modeling data matrix; According to the preliminary extended modeling data matrix, the coupling relationship between the process unit parameters and the soft measurement prediction value is statistically analyzed, and the input-output association strength is marked and sorted to obtain the extended modeling data matrix with coupling characteristic marking. Preferably, the applying a multivariate identification technology to the extended modeling data matrix to establish a preliminary dynamic relationship model characterizing the dynamic coupling relationship between the process units and the process constraint conditions comprises: According to the extended modeling data matrix, the input variables and the output variables of each process unit are respectively feature-extracted to obtain an input feature set and an output feature set for dynamic modeling. statistical analysis is performed on potential coupling relationships between the input variables and potential coupling relationships between the output variables according to the input feature set and the output feature set, to obtain an input coupling matrix and an output coupling matrix; dynamic response characteristics between the input and the output are identified according to the input coupling matrix and the output coupling matrix, to obtain a dynamic coupling parameter set; process constraint conditions between the process units are preliminarily calibrated according to the dynamic coupling parameter set, to obtain the preliminary dynamic relationship model. Preferably, the dynamic coupling coefficients and the constraint conditions of the model are optimized and calibrated according to the multivariable dynamic relationship model in combination with preset fuzzy logic control rules, to obtain the multivariable dynamic model, which includes: preliminary dynamic coupling coefficients of each input variable and output variable are extracted according to the preliminary dynamic relationship model, to obtain an original coupling parameter set; an allowed adjustment range and an initial offset are set for each coupling coefficient according to the original coupling parameter set, to obtain a coupling coefficient optimization candidate set; a coupling sensitivity matrix is obtained by calculating the sensitivity of each coupling coefficient to the output variable according to the coupling coefficient optimization candidate set; each coupling coefficient is iteratively adjusted according to the coupling sensitivity matrix, and an optimized dynamic coupling coefficient set is obtained by minimizing the error between the predicted output and the historical or soft measurement predicted value; a preliminary constraint parameter set is obtained according to the process constraint parameters in the preliminary dynamic relationship model; an impact matrix of the constraint parameters on the process variables is obtained by performing analog simulation calculation on each process constraint parameter in the preliminary constraint parameter set according to the optimized dynamic coupling coefficient set; the constraint parameters are adjusted according to the impact matrix, so that the analog output satisfies all constraint conditions and minimizes the deviation from the soft measurement predicted value, to obtain an optimized constraint parameter set; the preliminary dynamic relationship model is updated in parameters according to the optimized dynamic coupling coefficient set and the optimized constraint parameter set, to obtain a final optimized multivariable dynamic model. Preferably, the full process of phosphoric acid production is calculated by rolling optimization according to the multivariable dynamic model in combination with a preset model predictive control algorithm, to obtain target set values of a plurality of preset operating variables, which includes: input and output parameter sets of the model are formed by extracting input variables, output variables and dynamic coupling coefficients of each process unit according to the multivariable dynamic model; According to the model input and output parameter set, a rolling prediction time window and a control time step are defined to obtain a time series planning of prediction and control; According to the rolling prediction time window and the control time step, an initial guess setting is performed on an operation variable in each time step to obtain an initial operation variable sequence; According to the initial operation variable sequence and the multivariable dynamic model, an output variable in each time step is predicted to obtain a predicted output sequence; According to the predicted output sequence, a deviation of the output variable in each time step from a target quality index is calculated to obtain a prediction error sequence; According to the prediction error sequence, an initial operation variable sequence is adjusted and optimized, and an iterative calculation is performed by using a constraint condition and a target function to obtain an optimized operation variable sequence; According to the optimized operation variable sequence, a plurality of preset operation variables in each time step in the rolling time window are reordered and calibrated to obtain the target set value. Preferably, the adjusting and optimizing of the initial operation variable sequence according to the prediction error sequence, and the iterative calculation by using the constraint condition and the target function to obtain the optimized operation variable sequence include: According to the prediction error sequence, an optimization target function of each operation variable is defined to obtain an error target mapping of the operation variable; According to the error target mapping of the operation variable, an upper and lower constraint range and an adjustable step length of each operation variable are set to obtain a constrained operation variable set; According to the constrained operation variable set, an initial iterative optimization calculation is performed to obtain an initial iterative operation variable state; According to the initial iterative operation variable state and the multivariable dynamic model, an output variable in each time step is predicted to obtain an iterative predicted output sequence; According to the iterative predicted output sequence, a deviation from a preset target quality index is calculated to obtain an iterative error sequence; According to the iterative error sequence, an iterative operation variable is adjusted, and a target function minimization and a constraint condition verification are performed to obtain an updated iterative operation variable state; When it is judged that the updated iterative operation variable state satisfies a preset iterative convergence condition, the converged operation variable sequence is obtained.

[0008] In a second aspect, the embodiments of the present application also provide a phosphoric acid production full-process multivariable model predictive control system, which comprises at least one processor, at least one memory, and computer program instructions stored in the memory, and when the computer program instructions are executed by the processor, the method of the first aspect in the above-mentioned embodiments is realized.

[0009] To sum up, the beneficial effects of the present application are as follows: The method comprises the following steps: collecting key process data of a plurality of process units in a whole phosphoric acid production process in real time, wherein the process units comprise a reaction unit, a filtration unit, a concentration unit and an acid storage unit, and the key process data comprises temperature, pressure, flow and ingredient concentration; pre-processing the key process data to obtain real-time standardized data meeting modeling requirements; establishing a soft measurement model for predicting key quality parameters according to a historical standardized data set collected in advance; inputting the real-time standardized data into the soft measurement model to perform online prediction on a running process state of the reaction unit and obtain a soft measurement prediction value related to product quality; constructing a multivariate dynamic model for characterizing dynamic coupling relationships between the process units and process constraint conditions according to the historical standardized data set and the soft measurement prediction value; performing rolling optimization calculation on the whole phosphoric acid production process according to the multivariate dynamic model and a preset model predictive control algorithm to obtain target set values of a plurality of preset operation variables; and controlling the whole phosphoric acid production process according to the target set values. The multivariate dynamic model based on the historical standardized data and the soft measurement prediction value is constructed to accurately characterize the dynamic coupling relationships between the process units and the process constraints in the whole phosphoric acid production process, thereby solving the problem that it is difficult to model the coupling between processes and constraints in the prior art. The key process data is collected in real time and pre-processed, missing data is compensated and standardized, the soft measurement model is combined to perform online prediction on the key quality parameters, the product quality state is grasped in real time, and the problem that it is difficult to predict the key quality parameters in the prior art is solved. Further, the model predictive control algorithm is used to perform rolling optimization calculation on the operation variables, and the constraints and the objective function are adjusted through iteration to realize optimal setting of the operation variables at each time step, thereby ensuring accurate control of the quality indicators in the production process, and effectively solving the technical problem that it is difficult to realize rolling optimization control under multivariate dynamic constraints in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required to be used in the embodiments of the present application will be briefly introduced as follows, and for those skilled in the art, other drawings can also be obtained without creative labor on the premise of these drawings, and these are within the protection scope of the present application.

[0011] Figure 1 The flowchart of the whole work of the phosphoric acid production whole process multivariate model predictive control method in embodiment 1 of the present application is shown in the figure. Figure 2A flowchart for constructing a multivariable dynamic model for characterizing dynamic coupling relationship and process constraint conditions between each of the process units in Example 1 of the present application; Figure 3 A flowchart for performing rolling optimization calculation on the whole phosphoric acid production process in Example 1 of the present application to obtain target set values of a plurality of preset operation variables; Figure 4 A structure diagram of a phosphoric acid production whole process multivariable model predictive control system in Example 2 of the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be noted that, in this document, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between the entities or operations. In the description of the present application, it should be understood that the orientation or positional relationship indicated by terms such as "center", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device. Without more limitations, the elements defined by the statement "include" do not exclude the presence of additional identical elements in the process, method, article or device including the elements. If there is no conflict, the embodiments of the present application and the various features in the embodiments can be combined with each other, and are all within the protection scope of the present application. EMBODIMENT

[0013] Please refer to Figure 1 Example 1 of the present application discloses a phosphoric acid production whole process multivariable model predictive control method, which comprises: Real-time acquisition of key process data of a plurality of process units in the whole phosphoric acid production process, wherein the process units include reaction units, filtration units, concentration units and acid storage units, and the key process data includes temperature, pressure, flow rate and component concentration; Specifically, real-time acquisition refers to the continuous and immediate collection of process data through sensors and automated measurement systems, ensuring that the acquired data can reflect the dynamic changes of each link in the production process. The process units include reaction units, filtration units, concentration units, and acid storage units, each of which undertakes different production functions, such as the reaction unit for chemical reaction of phosphorite and acid liquid, the filtration unit for separating solid and liquid, the concentration unit for increasing acid liquid concentration, and the acid storage unit for storing final acid liquid. Key process data includes temperature, pressure, flow rate, and component concentration, which can directly reflect production status and product quality indicators, such as abnormal changes in temperature and pressure, which may lead to decreased reaction efficiency or product deviation. The main purpose of this step is to obtain real-time information of the production process, providing a reliable data basis for subsequent modeling and prediction. Only under the premise of sufficient and accurate data, can accurate soft-sensing models and multivariate dynamic models be established, achieving optimized control of the production process. In the implementation process, temperature sensors, pressure sensors, flow meters, and online component analyzers are usually installed in each process unit, and data from each unit is transmitted to a central database through a data acquisition system. In the data acquisition process, data sampling frequency, synchronization, and anomaly detection need to be considered, such as if the reaction kettle temperature sensor fails temporarily, the anomaly should be recorded and labeled in time for subsequent processing. Through this technical feature, comprehensive monitoring of the production process can be achieved, response speed to abnormal situations can be improved, production accidents and product quality fluctuations can be reduced, and overall production stability can be improved.

[0014] The key process data is preprocessed to obtain real-time standardized data that meets the modeling requirements; Specifically, preprocessing refers to cleaning, correcting, formatting, and standardizing the collected raw data to eliminate the effects of noise, outliers, and missing values, so that the data can be directly used for modeling analysis. Standardized data refers to numerical forms with uniform dimensions and data distribution after normalization, such as converting pressure data uniformly to the interval of 0 to 1, so that different features have comparability in modeling. The purpose of this step is to improve data quality and availability, providing high-reliability input information for soft measurement models. If raw data is used directly without processing, it may lead to inaccurate model predictions or even incorrect control strategies due to noise or outliers. In the implementation process, preprocessing includes multiple sub-processes: first, noise filtering is performed on the data, such as using moving average or Kalman filter to smooth sudden data; then outliers are removed, such as instantaneous pressure fluctuations exceeding a certain threshold; then missing value compensation and time series alignment are performed to ensure synchronization of data from different sampling periods; finally, normalization or standardization is performed on each variable to make the distribution balanced and meet the modeling requirements, thereby significantly improving the stability and prediction accuracy of model training, reducing the risk of misoperation caused by data quality problems, and ensuring more reliable production control.

[0015] According to the pre-collected historical standardized data set, a soft measurement model for predicting key quality parameters is established; Specifically, the historical standardized data set refers to a representative collection of past production data after preprocessing, containing key parameters of each process unit and corresponding quality indicators. Soft measurement model is a model that maps easily measurable process parameters to difficult-to-directly-measure key quality indicators through mathematical or machine learning methods, such as using temperature, pressure, and flow to predict product concentration or purity.

[0016] The purpose of this step is to realize online prediction of key quality parameters through the model, thereby realizing indirect monitoring and early warning of product quality in actual production. Traditional methods rely on manual sampling and testing, which have lag and low efficiency, while soft measurement models can reflect the dynamic changes of product quality in real time, providing data support for subsequent control. In the implementation process, first, feature analysis is performed on historical data, such as calculating the correlation between each variable and key quality parameters, and selecting significant features as inputs; then, the input features are paired with the corresponding quality indicators to form training samples; then, multivariate regression, neural network, or time series deep learning methods are used for modeling training, and model parameters are optimized through cross-validation and error analysis; finally, a soft measurement model capable of real-time prediction of key quality indicators is obtained, thereby converting complex process data into operable quality prediction information, improving the controllability and product consistency of the production process, and effectively reducing quality fluctuations.

[0017] inputting the real-time standardized data into the soft-sensing model to perform online prediction on the running process state of the reaction unit to obtain a soft-sensing prediction value related to product quality; Specifically, online prediction refers to that the model calculates the prediction result in real time during production, rather than offline analysis. The soft-sensing prediction value refers to the value of a key quality index, such as phosphoric acid concentration or impurity content, predicted by the soft-sensing model. By using the real-time collected and standardized data, the running state of the reaction unit and other key process units is directly predicted to provide instant feedback for production control and realize dynamic monitoring and early warning of product quality. Compared with the traditional method, this online prediction can shorten the response time and improve the timeliness and accuracy of process adjustment. In the implementation process, the real-time standardized data at each collection time is input into the trained soft-sensing model as input, and the model generates the prediction value of the current key quality parameter through forward calculation or recursive prediction; meanwhile, time window data can be combined for smooth prediction to avoid the influence of instantaneous fluctuations on the result. This step can reflect the process state in real time, discover potential quality deviations in advance, and enable the operator or automatic control system to take adjustment measures in time, thereby improving production efficiency and product stability, reducing the rate of unqualified products, and reducing resource waste.

[0018] According to the historical standardized data set and the soft-sensing prediction value, a multivariate dynamic model for representing the dynamic coupling relationship between the process units and the process constraint condition is constructed; Specifically, the dynamic coupling relationship refers to the mutual influence between the process units in the production process, for example, the temperature change of the reaction unit may affect the liquid level or flow of the concentration unit. The process constraint condition refers to the operating limit that must be met in production, such as the safety range of temperature, pressure or concentration. The multivariate dynamic model is a mathematical or computational model for simultaneously describing the input-output relationship of multiple process units and the coupling characteristics changing over time. The purpose of this step is to establish a comprehensive and accurate mathematical model of the production process, combine the soft-sensing prediction value with the historical data, reflect the dynamic response of each unit, and consider the constraint condition, thereby providing a reliable basis for subsequent model predictive control. Through this modeling, the influence of each operation on the overall process can be predicted before control operation, thereby improving the controllability and safety of the production process. In the implementation process, first, the historical standardized data and the soft-sensing prediction value are time-matched to form an extended modeling data matrix; then, a multivariate identification method is applied to analyze the potential coupling relationship between the input variables and the output variables to obtain an input coupling matrix and an output coupling matrix; then, the dynamic coupling parameters between the units are extracted, and the coupling coefficients and constraint parameters are optimized in combination with the preset fuzzy logic control rules; finally, the multivariate dynamic model is established.

[0019] According to the multivariate dynamic model, a preset model predictive control algorithm is combined to perform rolling optimization calculation on the whole phosphoric acid production process to obtain target set values of a plurality of preset operation variables. Specifically, the model predictive control (MPC) algorithm is a method of predicting the behavior of a system over a certain period of time based on a mathematical model and determining the operation variables through an optimization algorithm. Rolling optimization calculation refers to the control algorithm repeatedly performing prediction and optimization in continuous time windows, so that the operation variables are continuously updated to adapt to actual production changes. The target set value refers to the optimal value of each operation variable calculated by the control system, such as pump speed, valve opening or reaction temperature setting. Through prediction and optimization, real-time adjustment of the production process is realized, so that each operation variable not only meets the dynamic coupling relationship, but also optimizes product quality and production efficiency within the constraint range. Compared with traditional fixed operation strategies, rolling optimization can cope with process fluctuations and external disturbances, improve production response speed and flexible control ability. First, the input and output variables and coupling coefficients in the multivariate dynamic model are extracted, and a rolling prediction time window and a control step are defined; then an initial guess of the operation variables is made for each time step, and the output is predicted through the multivariate dynamic model; then the deviation of the predicted output from the target quality index is calculated, and the constraint conditions and objective function are used for iterative optimization to generate an optimized operation variable sequence; finally, the operation variables at each time step in the rolling window are calibrated and sorted to obtain the final target set value, which can realize high-precision real-time control in a complex, multivariate coupled process environment, and improve production efficiency and product consistency.

[0020] According to the target set value, control of the whole phosphoric acid production process is realized.

[0021] Specifically, implementing control refers to directly applying the calculated target set value to the production equipment through the control system, such as adjusting pump speed, valve opening or heating power, to adjust the running state of each process unit and realize closed-loop control of the production process. The results of the previous model prediction and rolling optimization are converted into actual operations to realize fine and dynamic management of the production process, thereby ensuring product quality and production safety while maximizing production efficiency. Through this closed-loop control, the system can automatically respond to fluctuations and abnormalities, reducing manual intervention and improving production stability. In the implementation process, after receiving the target set value, the control system accurately adjusts each process unit through PLC, DCS or intelligent control terminal, and real-time feedback of the running state is realized through sensors, which is compared with the target set value. If there is a deviation, fine tuning is performed to form a closed-loop control. This technical feature enables the entire production process to achieve high-precision, dynamic optimization control under complex coupling conditions, thereby significantly improving product quality consistency, reducing energy consumption and reducing the rate of defective products. Preferably, the pre-processing of the key process data to obtain standardized data meeting the modeling requirements comprises: noise filtering and outlier removal on the key process data to obtain multi-source process data; Specifically, noise filtering refers to removing random fluctuations or sensor noise in measurement data through signal processing methods, such as using a low-pass filter to remove high-frequency interference; outlier removal refers to identifying and removing data points that deviate significantly from the normal range due to sensor failure or sudden disturbances. The obtained multi-source process data refers to a collection of data from different process units or different sensors, which is more reliable and consistent after processing. The purpose of this step is to ensure data quality and provide reliable input for subsequent modeling, avoiding noise and outliers causing model prediction deviation or misjudgment of process state. By removing interference information, the accuracy of soft-sensing models and multivariate dynamic models can be improved. First, apply filtering algorithms such as moving average or Kalman filtering to each process unit's sensor data to remove high-frequency noise; at the same time, set thresholds or statistical indicators to identify outliers, which are removed or filled with adjacent data. This technical feature ensures that input data is smoother and more reliable, helping to improve the stability and precision of subsequent prediction and control, thereby reducing production risk and improving product quality consistency.

[0022] time alignment and interpolation processing on the multi-source process data according to the sampling period of each process unit to obtain time series data; Specifically, the sampling period refers to the time interval of data collection for each process unit, which may differ between units, such as sampling reaction temperature every second and flow rate every minute; time alignment refers to unifying these different sampling frequencies to the same time scale; interpolation processing refers to estimating data for missing time points, such as linear interpolation or spline interpolation, to obtain continuous time series data, i.e., time series data. Integrating data from different sources and different frequencies into consistent time series provides comparability and completeness for multivariate modeling. Time alignment and interpolation ensure that input data is synchronized in the time dimension when the model performs dynamic analysis and coupling relationship identification, avoiding errors caused by time misalignment. In implementation, a unified time reference can be first determined to time-match data from each process unit, and then interpolation methods such as linear interpolation, cubic spline, or interpolation based on adjacent average values can be used to fill in missing or discontinuous data points.

[0023] missing value compensation and consistency verification on the time series data to obtain verification data; normalization processing on the verification data to obtain the standardized data.

[0024] Specifically, missing value compensation refers to filling in data missing points caused by sensor failure, communication delay, etc.; consistency check refers to detecting the consistency of data logic and value, such as whether the flow and liquid level change match; normalization processing refers to converting data of different dimensions or orders of magnitude to a unified range to meet the input requirements of the modeling algorithm, obtaining standardized data, the purpose is to ensure that the modeling data is complete, consistent and can be directly used for soft measurement model and multivariable dynamic model training and prediction, so as to improve the stability and generalization ability of the model. Data normalization can eliminate the influence of different variable dimensions, so that the optimization algorithm converges faster in rolling prediction calculation. In the implementation process, historical average value or interpolation method can be used to compensate for missing data, and then consistency check is performed through logical rules or statistical methods; finally, all feature variables are normalized to ensure that each variable is input into the model within the same scale range, so that the input data is complete and standardized, which helps to improve the prediction accuracy of soft measurement and dynamic model, and at the same time enhances the real-time performance and reliability of rolling optimization control.

[0025] Preferably, the soft measurement model for predicting the key quality parameter is established according to a pre-collected historical standardized data set, comprising: Performing feature correlation analysis and principal component analysis on the historical standardized data set, and screening out feature variables related to the key quality parameter as input feature vectors according to the analysis results; Specifically, feature correlation analysis refers to statistical linear or nonlinear relationship between each process parameter and the key quality parameter, such as calculating the correlation coefficient of temperature, pressure, flow and acid concentration to determine which parameters have greater impact on product quality; principal component analysis (PCA) is a dimension reduction method that maps original high-dimensional features to lower-dimensional space through linear combination while retaining most information. The "input feature vector" after screening refers to the set of key variables used for model training. The purpose of this step is to reduce input variable redundancy, improve modeling efficiency, and ensure that the model focuses on the most sensitive features to the quality parameter, thereby improving prediction accuracy and generalization ability. In the implementation process, the correlation coefficient of each feature and the target variable can be calculated first, and the low correlation features are removed, and then the remaining features are subjected to PCA to convert them into principal component vectors, so that the soft measurement model can be based on the most important information for modeling, reducing the calculation complexity, while reducing noise interference and improving the prediction accuracy of the model for the key quality parameter.

[0026] Pairing the input feature vectors with the corresponding historical measurement values of the key quality parameter to form a training sample set for soft measurement modeling; Specifically, the training sample set refers to an ordered data set composed of input feature vectors and corresponding output variables (key quality parameters) as the basis for model learning. For example, a temperature, pressure and flow vector at a time point and the corresponding acid concentration at that time form a training sample. The purpose is to provide sufficient and systematic learning data for the modeling algorithm, so that the model can capture the regular relationship between input variables and quality parameters. In the implementation process, the input features and corresponding quality measurement values at each collection time point can be matched in chronological order to form a matrix of feature-target pairs, and abnormal values can be removed or smoothed. This technical feature ensures that the training data is complete and the corresponding relationship is clear, thereby improving the stability of the soft measurement model training and the reliability of the prediction.

[0027] Based on the training sample set, a multivariate regression method, a neural network method or a time series deep learning method is used for modeling training to obtain an initial model. Specifically, the multivariate regression method refers to fitting the input and output relationship through linear or nonlinear regression; the neural network method refers to using artificial neural networks for nonlinear mapping; and the time series deep learning method considers the time dependence of the data, such as long short-term memory network (LSTM) or convolutional neural network (CNN-LSTM) processing sequence data. The initial model obtained by these methods can realize the prediction of the key quality parameters.

[0028] By constructing a mathematical model that can accurately predict key quality parameters, the historical features and output relationship are mapped to a function or network structure that can be used for real-time prediction. In the implementation process, an algorithm suitable for the characteristics of the data can be selected, the training sample set is input for model training, and the parameters (such as weights, biases) are iteratively optimized until the model converges on the training data, ensuring that the initial model can effectively capture the complex relationship between input features and output targets, providing a basis for subsequent model optimization and real-time prediction.

[0029] The initial model is cross-validated and error-analyzed, and the model parameters are optimized and updated based on the generalization performance of the initial model to obtain the soft measurement model.

[0030] Specifically, cross-validation refers to dividing the training samples into several subsets, using part of the data to train the model and part of the data to validate the model, to evaluate the generalization ability of the model; error analysis refers to calculating the prediction error (such as mean square error, absolute error, etc.) to judge the model performance; generalization performance refers to the prediction ability of the model on unseen data. The soft-sensing model obtained by optimization and updating can accurately predict the quality parameters in real-time data input. The purpose of this step is to improve the reliability and accuracy of the model in the actual running environment, and to prevent overfitting or large prediction bias. In the implementation process, K-fold cross-validation, grid search or Bayesian optimization can be used to adjust the model hyperparameters, and the model structure or weights can be corrected combined with the error analysis results. This technical feature ensures that the soft-sensing model can maintain high prediction accuracy when facing new real-time data, thereby providing a reliable foundation for subsequent dynamic modeling and process optimization, and improving the quality control capability of the phosphoric acid production process.

[0031] Preferably, referring to Figure 2 , the constructing a multivariate dynamic model for representing the dynamic coupling relationship between the process units and the process constraint conditions according to the historical standardized data set and the soft-sensing prediction value comprises: extracting key process parameters and quality parameters of each process unit from the historical standardized data set to form an initial modeling data matrix; Specifically, the key process parameters refer to important process variables that affect the operation of the phosphoric acid production unit, such as reaction temperature, pressure, flow rate, etc.; the quality parameters refer to indicators related to the quality of the final product, such as acid concentration, impurity content, etc.; the initial modeling data matrix is a structured data set formed by organizing the key parameters and quality parameters of multiple process units at different time points in matrix form. Systematize the dispersed historical production data, so that the model can consider the input-output relationship of each process unit, and provide a complete data foundation for subsequent dynamic modeling. In the implementation process, the historical standardized data set can be organized, and the input and output parameters of each process unit can be extracted in time sequence and arranged in matrix form, with each row corresponding to a time point and each column corresponding to a parameter or quality indicator. By ensuring the structured and analyzable data, a reliable data foundation is provided for multivariate dynamic modeling, and it is also helpful to capture the potential relationship between different units.

[0032] correlating the soft-sensing prediction value with the initial modeling data matrix to obtain an extended modeling data matrix, wherein the extended modeling data matrix is used to describe the input-output relationship and coupling characteristics; Specifically, the soft-sensing predicted value refers to a key quality parameter value predicted by the aforementioned soft-sensing model based on real-time or historical data; the extended modeling data matrix is an enhanced data set formed by adding the soft-sensing predicted value to the initial data matrix, and is used to describe the input and output and coupling characteristics of each process unit; the purpose is to combine the predicted information with the historical actual data, so that the dynamic modeling not only considers the historical observation values, but also integrates the predicted trend, thereby more accurately reflecting the coupling relationship between the units. In the implementation process, the soft-sensing predicted value at each time point can be matched with the corresponding row of the initial data matrix in time sequence, and then a new list representing the predicted value is added in the matrix to form an extended matrix. This step enables the model to utilize both historical data and predicted data, improves the description ability of the dynamic relationship, and enhances the ability to capture the process coupling relationship.

[0033] applying a multivariate identification technique to the extended modeling data matrix to establish a preliminary dynamic relationship model representing the dynamic coupling relationship between the process units and the process constraints; Specifically, the multivariate identification technique refers to establishing a mathematical relationship between input variables and output variables through statistical analysis or system identification methods, such as using recursive least squares, state space modeling or multivariate regression analysis to describe the dynamic response characteristics; the preliminary dynamic relationship model refers to a preliminary model that can represent the dynamic coupling between the process units and the constraints based on the extended data matrix. The purpose of this step is to quantify the dynamic coupling relationship between the process units and the process constraints, to provide a basic model for subsequent control strategies, and to enable the system to reflect the real process dynamics in prediction and optimization. In the implementation process, input and output characteristics can be extracted from the extended matrix, the coupling relationship between the input variables and the output variables can be identified, the system dynamic response can be fitted using a multivariate identification algorithm, and preliminary coupling coefficients and constraint parameters can be obtained, so that the model can systematically describe the interaction between the process units and the constraints, provide an accurate basis for optimization control, and improve the stability and safety of the production process.

[0034] According to the preliminary dynamic relationship model, the dynamic coupling coefficients and the constraint conditions of the model are optimized and calibrated according to the preset fuzzy logic control rules, and the multivariate dynamic model is obtained.

[0035] Specifically, the fuzzy logic control rule refers to a rule of fuzzy processing of input-output relationship based on process experience and engineering knowledge, for example, appropriately increasing the cooling water flow when the temperature is too high and the flow is low; the multivariable dynamic model is an optimized and calibrated final dynamic model which can be directly used in the control algorithm and can describe the dynamic coupling and constraint conditions of multiple process units. Through the experience rule combined with the data-driven model, the coupling coefficient and the constraint parameter are fine-tuned and optimized, so that the model is more accurate and reliable under actual working conditions. In the implementation process, the coupling parameter and the constraint coefficient of the preliminary model can be adjusted by using the fuzzy logic rule, the prediction error is minimized through iterative optimization, and the constraint condition is ensured to be met, so that the final multivariable dynamic model is obtained.

[0036] Preferably, the associating the soft-sensing prediction value with the initial modeling data matrix to obtain an extended modeling data matrix comprises: According to the initial modeling data matrix, the soft-sensing prediction value is matched in time sequence to obtain a matched soft-sensing prediction value; Specifically, the initial modeling data matrix refers to a structured data set containing historical key process parameters and quality parameters of each process unit; the soft-sensing prediction value is a key quality parameter value predicted by the foregoing soft-sensing model. The time sequence matching refers to aligning each prediction value with the historical data at the corresponding time point, so that the prediction value corresponds to the actual data in time. The purpose of this step is to ensure that the prediction data and the historical observation data correspond accurately in the time dimension, and to provide a basis for subsequent association analysis and matrix expansion. In the implementation process, the soft-sensing prediction value can be sorted according to its time stamp, and corresponded with the row of the initial modeling data matrix at the same time point one by one, to generate a time-aligned prediction value sequence, ensuring the consistency of the prediction information and the historical data in time, providing accurate input for coupling relationship analysis, and reducing errors caused by time misalignment.

[0037] The matched soft-sensing prediction value is associated with the corresponding key process parameters of the process unit to obtain a preliminary associated soft-sensing data set; Specifically, the association processing refers to establishing a corresponding relationship between the prediction value and the input and output parameters of each process unit, so as to analyze how the prediction parameter is affected by each input variable; the preliminary associated soft-sensing data set refers to a data set obtained by combining the prediction value and the corresponding process parameters at each time point. The purpose of this step is to clarify the potential relationship between the prediction value and the key parameters of each process unit, and to provide structured data support for matrix expansion and coupling analysis. In the implementation process, each matched soft-sensing prediction value can be combined with the input variables (such as temperature and flow) and the output variables (such as concentration and yield) at the corresponding time point to form a set recording the input and output and the prediction value, so that the model can quantify the association between the soft-sensing prediction value and each process parameter, and lay a foundation for subsequent multivariable analysis and optimized modeling.

[0038] According to the preliminary associated soft measurement data set, column expansion processing is performed on the initial modeling data matrix, and corresponding soft measurement prediction value columns are added to the input variables and output variables of each process unit to obtain a preliminary expanded modeling data matrix; Specifically, the column expansion processing refers to adding one or more columns after each row of the original matrix for storing soft measurement prediction values; and the preliminary expanded modeling data matrix refers to an enhanced matrix formed by combining the original historical data matrix and the soft measurement prediction values. The prediction information is embedded into the historical data matrix, so that each time point contains not only the original process parameters but also the corresponding prediction values, thereby providing rich input features for dynamic coupling modeling. In the implementation process, columns can be added to the initial modeling data matrix, and the matched soft measurement prediction values are filled into the corresponding rows in chronological order. The input and output variables of each process unit can be correspondingly increased with the addition of the prediction value columns to form an enhanced matrix. This technical feature enhances the information amount of the data matrix, so that the subsequent modeling can more comprehensively describe the process dynamics, and the prediction accuracy and the reliability of the coupling analysis are improved.

[0039] According to the preliminary expanded modeling data matrix, statistical analysis is performed on the coupling relationship between the parameters of each process unit and the soft measurement prediction values, and the input-output correlation strength is marked and arranged to obtain the expanded modeling data matrix with coupling characteristic marking.

[0040] Specifically, the coupling relationship refers to the dependence or influence degree between the input variables and the output variables or the prediction values; the correlation strength marking is represented by statistical indicators (such as correlation coefficients, regression coefficients) to indicate the strength of the relationship between the parameters; and the expanded modeling data matrix with coupling characteristic marking refers to the final data set in which the coupling relationship information between the process parameters and the prediction values is recorded in the matrix. The purpose is to quantify the mutual influence between the parameters and the prediction values, so that the model can identify important coupling paths, provide guidance for multivariate dynamic modeling, and help optimize the control strategy. In the implementation process, statistical analysis can be performed on each pair of parameters and prediction values in the expanded matrix, such as calculating the correlation coefficient or the partial correlation coefficient, and the coupling strength is stored in the matrix in the form of marking, and the matrix structure is arranged to facilitate subsequent modeling. This step makes the data matrix not only contain original and prediction data, but also reflect the coupling characteristics between the variables, thereby improving the accuracy and robustness of the multivariate dynamic model.

[0041] Preferably, the application of the multivariate identification technology to the expanded modeling data matrix to establish a preliminary dynamic relationship model representing the dynamic coupling relationship between the process units and the process constraint conditions comprises: According to the expanded modeling data matrix, feature extraction is performed on the input variables and output variables of each process unit respectively to obtain an input feature set and an output feature set for dynamic modeling; Specifically, the input variables refer to the operating parameters or environmental conditions of each process unit, such as temperature, flow rate, feed concentration, etc.; the output variables refer to the reaction results or quality indicators, such as phosphoric acid concentration, yield or impurity content; the feature extraction is to select the variables most representative or significantly influential in modeling from the original extended matrix to form a feature set. The purpose of this step is to reduce data redundancy by extracting key features while ensuring that dynamic modeling can capture the core change rules of each process unit, providing an efficient data basis for subsequent coupling analysis. In the implementation process, statistical indicators (such as variance analysis, correlation analysis) or dimensionality reduction methods (such as principal component analysis, factor analysis) can be used to select variables, and the selected input variables and output variables form the input feature set and the output feature set respectively, making the dynamic modeling more targeted, improving the calculation efficiency, and ensuring that the model can accurately reflect the process dynamic characteristics.

[0042] According to the input feature set and the output feature set, statistical analysis is performed on the potential coupling relationship between the input variables and the potential coupling relationship between the output variables, respectively, to obtain an input coupling matrix and an output coupling matrix; Specifically, the potential coupling relationship refers to the possible correlation or influence relationship between different input variables or output variables, and the coupling matrix is a structured data representing the coupling strength between variables in matrix form. The purpose of this step is to identify the interaction between process parameters and provide a coupling structure basis for dynamic modeling, so as to more accurately describe the overall dynamic behavior of the system. In the implementation process, correlation analysis, covariance calculation or partial correlation analysis methods can be used to analyze the variables in the input feature set and the output feature set, obtain their coupling coefficients, and store them in the input coupling matrix and the output coupling matrix. This technical feature helps to quantify the relationship between parameters, provides a basis for subsequent dynamic coupling parameter identification, and improves the accuracy and reliability of the model.

[0043] According to the input coupling matrix and the output coupling matrix, the dynamic response characteristics between the input and the output are identified to obtain a dynamic coupling parameter set; Specifically, the dynamic response characteristics refer to the response mode of the output variables over time in response to changes in the input variables; the dynamic coupling parameter set is a parameter set that quantifies the input-output dynamic relationship, such as transfer coefficient, time delay or gain. The purpose of this step is to establish the dynamic mathematical relationship between each process unit, so that the model can describe the mutual influence of different variables over time, thereby achieving prediction and control of the process. In the implementation process, system identification methods such as recursive least squares, state space modeling or transfer function modeling can be used to fit the output dynamic response using the data in the input coupling matrix and the output coupling matrix, and the dynamic coupling parameter set is calculated to ensure that the model not only reflects the static relationship but also captures the dynamic behavior, improving the prediction accuracy of the production process.

[0044] According to the dynamic coupling parameter set, a process constraint condition between each process unit is preliminarily calibrated to obtain the preliminary dynamic relationship model.

[0045] Specifically, the process constraint condition refers to a process operation range, a device limit or a safety constraint, such as a temperature upper limit, a flow lower limit, etc.; and the preliminary dynamic relationship model is a complete process dynamic mathematical model established in combination with the dynamic coupling parameters and the constraint condition. The purpose of this step is to incorporate the actual production limit into the model in the modeling process, so that the model can not only reflect the dynamic characteristics, but also meet the process safety and operation constraints, thereby providing a reliable model basis for subsequent optimization control. In the implementation process, the dynamic coupling parameters can be applied to the model calculation, and the constraint conditions of each process unit can be embedded into the model structure. The preliminary dynamic model is determined through parameter calibration and simulation verification, so that the prediction output meets the constraint requirements.

[0046] Preferably, the multi-variable dynamic model is obtained by optimizing and calibrating the dynamic coupling coefficients and the constraint conditions of the model according to the multi-variable dynamic relationship model in combination with preset fuzzy logic control rules. According to the preliminary dynamic relationship model, preliminary dynamic coupling coefficients of each input variable and output variable are extracted to obtain an original coupling parameter set; Specifically, the preliminary dynamic relationship model refers to the mathematical model established in the previous step and containing input variables, output variables and dynamic coupling relationships thereof; the dynamic coupling coefficient is a parameter quantifying the influence degree of the input variable on the output variable, such as a gain or a response weight; and the original coupling parameter set is a set of all coupling coefficients preliminarily extracted. The purpose of this step is to quantify the dynamic correlation between variables, to provide an initial parameter basis for subsequent optimization, and to enable the model to reflect the actual variable interaction in the process. In the implementation process, the input-output relationship matrix of the preliminary dynamic model is analyzed, the contribution or weight of each input variable to the output variable is calculated, and the original coupling parameter set is formed by summarizing the coupling coefficients. This technical feature ensures that the model has a clear variable coupling structure, provides a basis for further optimizing the coupling coefficients and improving the prediction accuracy, and can also identify key influencing factors between process units.

[0047] According to the original coupling parameter set, an allowed adjustment range and an initial offset of each coupling coefficient are set to obtain a coupling coefficient optimization candidate set; Specifically, the adjustment range refers to the upper and lower limits of each coupling coefficient that can be adjusted, the initial offset is an initial correction value set for each coupling coefficient, and the coupling coefficient optimization candidate set is a series of parameter combinations to be optimized generated according to the range and offset. The purpose of this step is to provide a search space and initial value for the optimization of the coupling coefficient, to avoid parameter out-of-bounds or falling into local optimum during the optimization process, while ensuring that the adjustment conforms to the actual feasibility of the process. In the implementation process, the upper and lower limits of each coupling coefficient can be determined according to historical data, equipment operation limits and process experience, and an initial offset is added to generate a candidate value sequence to form the coupling coefficient optimization candidate set, thereby making the subsequent optimization have constraints and safety, and improving the optimization efficiency and reliability.

[0048] According to the coupling coefficient optimization candidate set, the sensitivity of each coupling coefficient to the output variable is calculated to obtain a coupling sensitivity matrix; Specifically, the sensitivity represents the degree of response of the output variable to the change of the coupling coefficient, and the coupling sensitivity matrix is structured data in the form of a matrix that presents the sensitivity of each coupling coefficient to each output variable. The purpose of this step is to identify which coupling coefficients have a significant impact on the output, so as to focus on adjusting the key coefficients during the optimization process, thereby improving the optimization effect and efficiency. In the implementation process, the coupling sensitivity matrix can be formed by perturbing the candidate coupling coefficients slightly and simulating the output response, calculating the ratio of the output change to the coefficient change. This technical feature can quantify the influence of each coefficient, making the optimization process more targeted, while avoiding wasting of computing resources due to invalid parameter adjustment.

[0049] According to the coupling sensitivity matrix, each coupling coefficient is iteratively adjusted, and the error between the predicted output and the historical or soft measurement predicted value is minimized to obtain an optimized dynamic coupling coefficient set; Specifically, the iterative adjustment refers to continuously correcting the coupling coefficient according to the sensitivity and the error, and the minimization of the error is achieved by optimizing algorithms such as gradient descent, genetic algorithm or heuristic search to make the predicted output as close as possible to the historical observation or soft measurement predicted value. The optimized dynamic coupling coefficient set is the final coupling parameter set for the multivariate dynamic model. By fine-tuning the coupling coefficient, the prediction accuracy and stability of the model are improved, and the dynamic model can more realistically reflect the process changes. In the implementation process, the adjustment step of each coefficient in each iteration can be determined according to the coupling sensitivity matrix, and the candidate set is iteratively optimized in combination with the error minimization objective function until the final optimized parameters are obtained.

[0050] According to the process constraint parameters in the preliminary dynamic relationship model, a preliminary constraint parameter set is obtained; Specifically, the process constraint parameter refers to a physical or process parameter limited in each process unit operation and output in the phosphoric acid production process, such as an upper limit of temperature, a lower limit of pressure, a flow range or a concentration threshold, etc.; the preliminary constraint parameter set is a set of all such constraint parameters extracted from the preliminary dynamic relationship model. By explicitly defining the operation boundary of each process unit in the control process, the feasibility range is provided for subsequent optimization, and it is ensured that the model optimization does not violate the process safety or quality specification. In the implementation process, all constraint parameters are extracted and arranged into a structured set by analyzing the input-output matrix and constraint condition description in the preliminary dynamic relationship model, forming the preliminary constraint parameter set. This technical feature ensures that the model follows the actual process limit in the optimization process, improves the operability and safety of the model, and provides basic data for simulation and optimization.

[0051] According to the optimized dynamic coupling coefficient set, each process constraint parameter in the preliminary constraint parameter set is simulated and calculated to obtain an influence matrix of the constraint parameter on the process variable; Specifically, the optimized dynamic coupling coefficient set is a parameter reflecting the dynamic correlation between each input and output variable obtained by the previous step of optimization; the influence matrix of the constraint parameter on the process variable refers to the quantitative influence relationship of the change of each constraint parameter on the output of each process variable, which is represented in the form of a matrix. By quantifying the actual influence of the constraint parameter on the production process, a basis is provided for subsequent constraint optimization, so that the optimization process can take into account both safety constraints and process efficiency. In the implementation process, by applying the optimized coupling coefficient to the preliminary constraint parameter set, the constraint parameter is gradually adjusted in the simulation environment and the change of the output variable is recorded, and a matrix reflecting the influence of each constraint parameter on each process variable is generated, so that the sensitive relationship between the constraint and the variable can be intuitively presented, and the constraint parameter can be accurately adjusted, improving the optimization reliability.

[0052] According to the influence matrix, the constraint parameter is adjusted so that the simulation output satisfies all constraint conditions and minimizes the deviation from the soft measurement predicted value, to obtain an optimized constraint parameter set; Specifically, adjusting the constraint parameter refers to modifying the constraint parameter according to the influence matrix to optimize the simulation result, and minimizing the deviation refers to making the simulation output as close as possible to the soft measurement predicted value by an optimization method, and the optimized constraint parameter set is a parameter set that satisfies the process constraint and improves the accuracy of the model after adjustment. Under the premise of ensuring process safety and operation constraints, the dynamic model output is highly consistent with the soft measurement predicted value, improving the precision and controllability of the model. In the implementation process, each constraint parameter can be adjusted in combination with the influence matrix and the optimization algorithm (such as iterative search, gradient adjustment or heuristic method), and it is verified in the simulation whether the output satisfies all constraint conditions, and the deviation from the soft measurement predicted value is calculated, and the iteration is repeated until the deviation is minimized, forming the optimized constraint parameter set.

[0053] According to the optimized dynamic coupling coefficient set and the optimized constraint parameter set, parameter updating is performed on the preliminary dynamic relationship model to obtain a final optimized multivariable dynamic model.

[0054] Specifically, parameter updating refers to replacing or integrating the optimized coupling coefficients and constraint parameters into the preliminary dynamic relationship model, and the final optimized multivariable dynamic model is a complete dynamic model that can reflect the coupling relationship between variables and meet the process constraints after optimization. The purpose of this step is to form an accurate model that can be used for predictive control, so that it can effectively control multiple process units in actual production. In the implementation process, the optimized dynamic coupling coefficient set and the constraint parameter set are imported into the preliminary dynamic model to update the input-output relationship and the constraint condition, and a complete multivariable dynamic model is formed. This technical feature ensures that the model has high prediction accuracy and process constraint compliance, providing a reliable foundation for subsequent model predictive control (MPC) and achieving efficient and stable control of the entire phosphoric acid production process. Preferably, referring to Figure 3 , the rolling optimization calculation of the entire phosphoric acid production process according to the multivariable dynamic model combined with a preset model predictive control algorithm to obtain target set values of a plurality of preset operation variables includes: According to the multivariable dynamic model, input variables, output variables, and dynamic coupling coefficients of each process unit are extracted to form a model input-output parameter set; Specifically, the input variables refer to controllable process operation parameters such as reaction temperature, flow rate, and feeding rate; the output variables refer to key quality or yield indicators in the process such as phosphoric acid concentration, purity, or yield; and the dynamic coupling coefficients are a parameter set describing the time-related influence between input variables and output variables. Structuring the core information in the multivariable dynamic model forms a basic parameter set for prediction and optimization, ensuring complete data support for subsequent rolling optimization calculation. In the implementation process, by analyzing the multivariable dynamic model, the controllable input, key output, and dynamic coupling relationship between input and output of each process unit are extracted and organized into a unified model input-output parameter set.

[0055] According to the model input-output parameter set, a rolling prediction time window and a control time step are defined to obtain a time sequence plan for prediction and control; Specifically, the rolling prediction time window refers to the time range in which the output variable is continuously predicted in the future time period, the control time step refers to the interval time at which the control operation is updated each time within the time window, and the time sequence planning is the plan formed by arranging the rolling prediction and the control operation in time sequence. The purpose of this step is to provide a time structure for model predictive control, so that the control algorithm can continuously optimize the operation variable within the rolling window, realize dynamic response and real-time adjustment. In the implementation process, according to the process dynamic characteristics and control requirements, the rolling prediction window and the time step of appropriate length are set, the input and output parameters are arranged in time sequence, and the time sequence planning which can be used for MPC algorithm is formed.

[0056] According to the rolling prediction time window and the control time step, an initial guess setting is made for the operation variable in each time step, and an initial operation variable sequence is obtained. Specifically, the initial operation variable sequence refers to the initial value estimation of the operation variable in each time step within the rolling prediction time window, which is used as the starting point for optimization calculation. The purpose of this step is to provide an initial solution for subsequent prediction and optimization, avoid starting from a blank or random state, and improve the convergence speed and stability. In the implementation process, the operation variable in each time step can be initially assigned according to the historical operation data, experience setting or the optimization result of the previous rolling window, and the initial operation variable sequence is formed. This technical feature ensures that the optimization calculation has a reasonable starting point, reduces the number of calculation iterations, and takes into account the process safety and stability.

[0057] According to the initial operation variable sequence and the multivariable dynamic model, the output variable in each time step is predicted to obtain a predicted output sequence. Specifically, the predicted output sequence refers to the predicted value sequence of the output variable in each time step calculated by using the initial operation variable sequence and the multivariable dynamic model. The purpose of this step is to evaluate the influence of the initial operation variable on the output variable within the future time window, and provide a basis for subsequent optimization adjustment.

[0058] In the implementation process, the initial operation variable sequence is input into the multivariable dynamic model, and the predicted value of the output variable corresponding to each time step is calculated in sequence to form a complete predicted output sequence.

[0059] According to the predicted output sequence, the deviation of the output variable in each time step from the target quality index is calculated to obtain a predicted error sequence. Specifically, the predicted output sequence refers to the predicted values of the output variables at each time step obtained by the initial sequence of operation variables through the multivariate dynamic model; the target quality index refers to the preset key quality parameter standard in the phosphoric acid production process, such as concentration, purity, yield, etc.; and the prediction error sequence refers to the sequence of the difference between the predicted output and the target index changing over time. By quantifying the deviation of the current initial operation variable from the production quality, an accurate error reference is provided for subsequent optimization. In the implementation process, the predicted output at each time step is subtracted from the corresponding target quality index to form error data in a continuous time sequence; in actual application, the absolute error or the square error can be used to represent the error size.

[0060] According to the prediction error sequence, the initial sequence of operation variables is adjusted and optimized, and iterative calculation is performed using constraint conditions and a target function to obtain an optimized sequence of operation variables. Specifically, iterative calculation refers to adjusting the operation variables through multiple cycles to gradually approach the target quality index; the constraint conditions include process safety limits, operation ranges, and equipment capabilities, etc.; and the target function is a mathematical expression for evaluating the adjustment of the operation variables, which is usually the minimization of the sum of squares of errors. By using the prediction error information, a more reasonable sequence of operation variables is generated through a constrained optimization algorithm, so that the output variables are closer to the target quality index. In the implementation process, the initial sequence of operation variables is used as the starting point for iteration, the gradient or sensitivity is calculated according to the prediction error sequence, and the operation variables are gradually adjusted within the range of the constraint conditions, while the effect of each adjustment is evaluated by the target function; the iteration is performed until the error is minimized or the convergence condition is met.

[0061] According to the optimized sequence of operation variables, a plurality of preset operation variables at each time step within the rolling time window are reordered and calibrated to obtain the target set value.

[0062] Specifically, reordering and calibration refer to time sequence arrangement, unit correction, and setting of the final executable target value for each operation variable of the optimized sequence of operation variables; and the target set value is the operation parameter actually used in production at each time step within the rolling window. The purpose of this step is to convert the optimization result into an executable control instruction to ensure that the operation variables at each time step are correctly implemented in the production process. In the implementation process, the optimized sequence of operation variables is adjusted in time sequence to align with the rolling prediction time window, and the unit, range, upper and lower limits, etc. are calibrated according to the process requirements to generate a complete sequence of target set values. Preferably, the optimization adjustment of the initial sequence of operation variables according to the prediction error sequence includes iterative calculation using constraint conditions and a target function to obtain an optimized sequence of operation variables. According to the prediction error sequence, define an optimization objective function of each operation variable, to obtain an error target mapping of the operation variable; Specifically, the prediction error sequence represents a data sequence of the difference between the output variable of each time step and the target quality indicator changing over time; the optimization objective function is a mathematical function for quantifying the adjustment effect of each operation variable, usually in the form of sum of squared errors or weighted absolute error; and the error target mapping establishes a correlation between each operation variable and its corresponding error function. The purpose of this step is to clarify the influence of each operation variable on the production quality in the optimization process, and to provide a mathematical basis for iterative adjustment. In the implementation process, the prediction error sequence is decomposed into each operation variable, and a target function is constructed for each operation variable to map the error change range that can be produced by adjusting the operation variable; this mapping can be achieved through sensitivity analysis or partial derivative calculation. Decomposing the complex multi-variable error into an optimizable target for each operation variable makes the optimization process directional and quantifiable, thereby improving the adjustment efficiency and prediction accuracy According to the error target mapping of the operation variable, set the upper and lower constraint ranges and the adjustable step length of each operation variable, to obtain a constrained operation variable set; Specifically, the upper and lower constraint ranges refer to the safe, process or equipment limit range of the operation variable in the production process; the adjustable step length is the maximum adjustment amplitude of the operation variable in each iteration; and the constrained operation variable set contains all operation variables and their constraint information, which is used for iterative optimization. The purpose of this step is to ensure that the optimization adjustment is carried out within the safe and feasible range, to prevent the operation variable from exceeding the equipment or process allowable range, and to ensure the stability of iteration. In the implementation process, according to the process parameters and equipment capacity, the minimum value and the maximum value of each operation variable are set, and the step length of each iteration is determined; then these constraint conditions are combined to form the constrained operation variable set, which is used for subsequent iterative calculation.

[0063] According to the constrained operation variable set, perform an initialization iterative optimization calculation to obtain an initial iterative operation variable state; Specifically, the initialization iterative optimization calculation refers to performing the first round of optimization calculation based on the constrained operation variable set to generate the initial state of the operation variable that can be used for further iteration; and the initial iterative operation variable state is the preliminary value of each operation variable at the beginning of the iterative optimization. The purpose of this step is to provide a reasonable starting point for subsequent iterations, so that the optimization algorithm achieves a balance in convergence and calculation efficiency. In the implementation process, the constrained operation variable set is input into the iterative optimization algorithm, such as gradient descent, genetic algorithm or sequential quadratic programming, to calculate the initial operation variable adjustment value, and to check whether the constraint conditions are satisfied to form the initial iterative state. By providing a stable starting point for the entire optimization iteration process, the instability of the algorithm in the initial stage is reduced, and the convergence speed and optimization effect are improved.

[0064] According to the initial iteration operation variable state and the multivariable dynamic model, an iteration predicted output sequence is outputted by predicting the output variable at each time step; Specifically, the multivariable dynamic model refers to a model previously constructed to describe the input-output coupling relationship and constraint conditions among the process units; and the iteration predicted output sequence is a sequence of the output variable changing over time, which is obtained by predicting the output variable based on the initial iteration operation variable state. The purpose of this step is to evaluate the actual influence of the initial iteration operation variable on the output variable, and to provide feedback basis for the next round of iteration optimization. In the implementation process, the initial iteration operation variable state is taken as the model input, the predicted value of the output variable at each time step is calculated through the multivariable dynamic model, and a complete iteration predicted output sequence is formed; in actual application, error correction can be performed in combination with historical data or soft measurement predicted value.

[0065] According to the iteration predicted output sequence, a deviation from a preset target quality index is calculated to obtain an iteration error sequence; Specifically, the iteration predicted output sequence refers to the change sequence of the output variable at each time step, which is predicted by the multivariable dynamic model under the current iteration operation variable state; the preset target quality index is a target value of a key quality parameter specified by process design or production requirements; and the iteration error sequence represents a data sequence of the deviation between the output variable at each time step and the target quality index changing over time. The purpose of this step is to quantify the influence of the current iteration operation variable state on the production quality, and to determine which output variable has not yet reached the target, so as to provide a basis for the next round of optimization adjustment. In the implementation process, the iteration predicted output sequence is compared with the target quality index at each time step, the error value (which can be absolute error, squared error or weighted error) is calculated, the iteration error sequence is generated, and the error change trend over time can be further analyzed.

[0066] According to the iteration error sequence, the iteration operation variable is adjusted, and a target function minimization and constraint condition verification are adopted to obtain an updated iteration operation variable state; Specifically, the iteration operation variable is a control parameter of each process unit in the current iteration round; the target function minimization refers to adjusting the operation variable by a mathematical optimization algorithm to minimize the error of the output variable; the constraint condition verification refers to ensuring that the operation variable after adjustment still satisfies the process, equipment and safety constraint conditions; and the updated iteration operation variable state is a new state after adjustment, which is used for the next round of iteration or final control. The purpose of this step is to optimize and adjust the operation variable, so that the output of the production process is closer to the target quality index, while ensuring that the adjustment process is safe and feasible. In the implementation process, the iteration error sequence is taken as the optimization target, the current operation variable is taken as the optimization object, the error is minimized within the constraint condition range by iterative calculation (such as gradient descent, sequential quadratic programming or genetic algorithm), and the updated operation variable state is obtained, and the error change of each round can be recorded for convergence judgment.

[0067] When it is judged that the updated iteration operation variable state meets the preset iteration convergence condition, the converged operation variable sequence is obtained.

[0068] Specifically, the iteration convergence condition refers to that the adjustment amplitude of the operation variable or the output error in the optimization process reaches a preset threshold, indicating that further iteration cannot significantly improve the quality index; the converged operation variable sequence is the finally determined operation parameter sequence that can be used for actual production. The purpose of this step is to determine when the optimization process ends, to ensure that the operation variable reaches the best or near-optimal state, to avoid invalid iteration, and to save computing resources and time. In the implementation process, whether the change amplitude of the iteration updated operation variable, the change amplitude of the output error, or the error difference of continuous multiple iterations is lower than the preset threshold is judged, and after it is confirmed that the convergence condition is met, the current operation variable sequence is taken as the final output and is used for production control. Embodiments

[0069] In addition, in combination with Figure 1 The phosphoric acid production whole-process multivariable model predictive control method of the embodiment 1 of the application described above can be realized by a phosphoric acid production whole-process multivariable model predictive control system. Figure 4 A hardware structure schematic diagram of the phosphoric acid production whole-process multivariable model predictive control system provided by the embodiment 2 of the application is shown.

[0070] The phosphoric acid production whole-process multivariable model predictive control system can include a processor and a memory storing computer program instructions.

[0071] Specifically, the processor can include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or can be configured to implement one or more integrated circuits of the embodiment of the application.

[0072] The memory can include mass storage for data or instructions. By way of example, and not limitation, the memory can include a Hard Disk Drive (HDD), floppy disk drive, flash memory, optical disc, magnetic tape, or Universal Serial Bus (USB) drive or a combination of two or more of these. The memory can be removable or non-removable (or fixed), as appropriate. The memory can be internal or external, as appropriate. In a particular embodiment, the memory is non-volatile solid-state memory. In a particular embodiment, the memory includes read-only memory (ROM). The ROM can be mask programmed ROM, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), electrically alterable ROM (EAROM), or flash memory, or a combination of two or more of these, as appropriate.

[0073] The processor implements the phosphoric acid production whole process multivariable model predictive control method of any one of the above embodiments by reading and executing the computer program instructions stored in the memory.

[0074] In one example, the phosphoric acid production whole process multivariable model predictive control system can further include a communication interface and a bus. As shown in Figure 4 The processor, the memory, and the communication interface are connected through the bus and complete communication with each other.

[0075] The communication interface is mainly used to realize the communication between the modules, devices, units and / or equipment in the embodiments of the application.

[0076] The bus includes hardware, software or both to couple the components of the device to each other. By way of example, and not limitation, the bus can include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand (IB) interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or another suitable bus or interconnect, or a combination of two or more of these. The bus can include one or more buses, as appropriate. Although particular buses are described and shown in the embodiments of the application, the application contemplates any suitable bus or interconnect.

[0077] In summary, the embodiments of the application provide a phosphoric acid production whole process multivariable model predictive control method and system.

[0078] It is to be understood that the application is not limited to the particular configurations and processes described hereinabove and shown in the drawings. For simplicity of explanation, detailed descriptions of known methods are omitted herein. In the above-described embodiments, several specific steps are described and shown as examples. However, the method processes of the application are not limited to the specific steps described and shown, and one skilled in the art can make various changes, modifications and additions, or change the order of the steps, after understanding the spirit of the application.

[0079] The functional blocks shown in the above-described block diagrams can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, functional cards, and the like. When implemented in software, the elements of the application are program or code segments used to perform the required tasks. The program or code segments can be stored in a machine-readable medium or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. The "machine-readable medium" can include any medium capable of storing or transmitting information. Examples of the machine-readable medium include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, and the like. The code segments can be downloaded via a computer network such as the Internet, an intranet, and the like.

[0080] The user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of the relevant data need to comply with the relevant laws, regulations and standards of the relevant location, and provide corresponding operation portals for the user to choose authorization or rejection.

[0081] It is also to be understood that the exemplary embodiments mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the above steps, that is, the steps can be executed in the order mentioned in the embodiments, or in an order different from the embodiments, or several steps can be executed simultaneously.

[0082] The above merely describes specific implementation of the present application, and those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, module and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A multivariate model predictive control method for the entire phosphoric acid production process, characterized in that, The method includes: Real-time acquisition of key process data from multiple process units in the entire phosphoric acid production process, including reaction unit, filtration unit, concentration unit and acid storage unit, and key process data including temperature, pressure, flow rate and component concentration; The key process data is preprocessed to obtain real-time standardized data that meets the modeling requirements; Based on a pre-collected historical standardized dataset, a soft measurement model is built to predict key quality parameters; The real-time standardized data is input into the soft measurement model to predict the operating status of the reaction unit online, thereby obtaining soft measurement prediction values ​​related to product quality. Based on the historical standardized dataset and the soft measurement prediction values, a multivariate dynamic model is constructed to characterize the dynamic coupling relationship and process constraints between each process unit. Based on the multivariate dynamic model and combined with the preset model prediction and control algorithm, rolling optimization calculations are performed on the entire phosphoric acid production process to obtain target set values ​​for multiple preset operating variables. Based on the target set value, the entire phosphoric acid production process can be controlled.

2. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 1, characterized in that, The preprocessing of the key process data to obtain standardized data that meets the modeling requirements includes: The key process data is subjected to noise filtering and outlier removal to obtain multi-source process data; Based on the sampling period corresponding to each process unit, the multi-source process data is time-aligned and interpolated to obtain time-series data. Missing value compensation and consistency verification are performed on the time series data to obtain verification data; The verification data is normalized to obtain the standardized data.

3. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 1, characterized in that, The process of building a soft measurement model for predicting key quality parameters based on a pre-collected historical standardized dataset includes: Feature correlation analysis and principal component analysis are performed on the historical standardized dataset. Based on the analysis results, feature variables related to the key quality parameters are selected as input feature vectors. The input feature vector is paired with the historical measurement values ​​of the corresponding key quality parameters to form a training sample set for soft measurement modeling; Based on the training sample set, a modeling and training process is performed using multivariate regression, neural network, or time-series deep learning methods to obtain an initial model. Cross-validation and error analysis are performed on the initial model, and the model parameters are optimized and updated based on the generalization performance of the initial model to obtain the soft measurement model.

4. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 1, characterized in that, The step of constructing a multivariate dynamic model to characterize the dynamic coupling relationship and process constraints between each process unit based on the historical standardized dataset and the soft measurement prediction values ​​includes: Based on the historical standardized dataset, key process parameters and quality parameters of each process unit are extracted to form an initial modeling data matrix; The soft measurement prediction value is correlated with the initial modeling data matrix to obtain an extended modeling data matrix, wherein the extended modeling data matrix is ​​used to describe the input-output relationship and coupling characteristics; Multivariate identification technology is applied to the extended modeling data matrix to establish a preliminary dynamic relationship model that characterizes the dynamic coupling relationship and process constraints between each process unit; Based on the preliminary dynamic relationship model and combined with the preset fuzzy logic control rules, the dynamic coupling coefficients and constraints of the model are optimized and calibrated to obtain the multivariate dynamic model.

5. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 4, characterized in that, The step of associating the soft measurement prediction value with the initial modeling data matrix to obtain the extended modeling data matrix includes: Based on the initial modeling data matrix, the soft measurement prediction values ​​are matched according to the time series to obtain the matched soft measurement prediction values; The matched soft measurement prediction values ​​are correlated with the corresponding key process parameters of the process unit to obtain a preliminary correlated soft measurement data set; Based on the preliminarily associated soft measurement data set, the initial modeling data matrix is ​​expanded by columns, and corresponding soft measurement prediction value columns are added to the input and output variables of each process unit to obtain the preliminarily expanded modeling data matrix. Based on the preliminary extended modeling data matrix, the coupling relationship between the parameters of each process unit and the predicted values ​​of soft measurement is statistically analyzed, and the input-output correlation strength is marked and organized to obtain the extended modeling data matrix with coupling characteristic markings.

6. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 4, characterized in that, The application of multivariate identification technology to the extended modeling data matrix to establish a preliminary dynamic relationship model characterizing the dynamic coupling relationship and process constraints between various process units includes: Based on the extended modeling data matrix, feature extraction is performed on the input and output variables of each process unit to obtain the input feature set and output feature set for dynamic modeling. Based on the input feature set and the output feature set, statistical analysis is performed on the potential coupling relationships between input variables and between output variables to obtain the input coupling matrix and the output coupling matrix. Based on the input coupling matrix and the output coupling matrix, the dynamic response characteristics between the input and output are identified, and a set of dynamic coupling parameters is obtained. Based on the set of dynamic coupling parameters, the process constraints between each process unit are initially calibrated to obtain the preliminary dynamic relationship model.

7. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 4, characterized in that, The step of optimizing and calibrating the dynamic coupling coefficients and constraints of the model based on the multivariate dynamic relationship model and in conjunction with preset fuzzy logic control rules to obtain the multivariate dynamic model includes: Based on the preliminary dynamic relationship model, the preliminary dynamic coupling coefficients of each input variable and output variable are extracted to obtain the original set of coupling parameters; Based on the original set of coupling parameters, an allowable adjustment range and an initial offset are set for each coupling coefficient to obtain a candidate set for coupling coefficient optimization. Based on the optimized candidate set of coupling coefficients, the sensitivity of each coupling coefficient to the output variable is calculated to obtain the coupling sensitivity matrix; Based on the coupling sensitivity matrix, each coupling coefficient is iteratively adjusted, and by minimizing the error between the predicted output and the historical or soft measurement predicted value, an optimized set of dynamic coupling coefficients is obtained. Based on the process constraint parameters in the preliminary dynamic relationship model, obtain the preliminary constraint parameter set; Based on the optimized set of dynamic coupling coefficients, simulation calculations are performed on each process constraint parameter in the preliminary constraint parameter set to obtain the influence matrix of the constraint parameters on process variables. Based on the influence matrix, the constraint parameters are adjusted so that the simulation output satisfies all constraints and minimizes the deviation from the soft measurement prediction, thus obtaining the optimized set of constraint parameters. Based on the optimized set of dynamic coupling coefficients and the optimized set of constraint parameters, the parameters of the preliminary dynamic relationship model are updated to obtain the final optimized multivariate dynamic model.

8. The multivariate model predictive control method for the entire phosphoric acid production process according to any one of claims 1-7, characterized in that, The step of performing rolling optimization calculations on the entire phosphoric acid production process based on the multivariate dynamic model and a preset model predictive control algorithm to obtain target set values ​​for multiple preset operating variables includes: Based on the multivariate dynamic model, the input variables, output variables, and dynamic coupling coefficients of each process unit are extracted to form a set of model input and output parameters; Based on the set of input and output parameters of the model, a rolling prediction time window and a control time step are defined to obtain the time series planning of prediction and control. Based on the rolling prediction time window and control time step, the operational variables within each time step are initially guessed and set to obtain the initial operational variable sequence. Based on the initial operational variable sequence and the multivariate dynamic model, the output variable at each time step is predicted to obtain the predicted output sequence. Based on the predicted output sequence, the deviation between the output variable and the target quality index at each time step is calculated to obtain the prediction error sequence; Based on the prediction error sequence, the initial operational variable sequence is optimized and adjusted by using constraints and objective function for iterative calculation to obtain the optimized operational variable sequence. Based on the optimized sequence of operational variables, multiple preset operational variables for each time step within the rolling time window are reordered and calibrated to obtain the target set value.

9. The multivariate model predictive control method for the entire phosphoric acid production process according to claim 8, characterized in that, The optimization and adjustment of the initial operational variable sequence based on the prediction error sequence, using constraints and an objective function for iterative calculation, yields the optimized operational variable sequence, including: Based on the prediction error sequence, an optimization objective function is defined for each operational variable to obtain the error objective mapping of the operational variables; Based on the error target mapping of the operational variables, the upper and lower constraint ranges and adjustable step sizes of each operational variable are set to obtain the set of constrained operational variables; Based on the set of constraint operation variables, perform initialization iterative optimization calculations to obtain the initial iterative operation variable states; Based on the initial iterative operation variable state and the multivariate dynamic model, the output variable at each time step is predicted to obtain the iterative prediction output sequence; Based on the iterative prediction output sequence, the deviation from the preset target quality index is calculated to obtain the iterative error sequence; Based on the iterative error sequence, the iterative operation variables are adjusted, and the updated state of the iterative operation variables is obtained by minimizing the objective function and verifying the constraints. When it is determined that the updated iterative operation variable state satisfies the preset iterative convergence condition, the converged sequence of operation variables is obtained.

10. A multivariate model predictive control system for the entire phosphoric acid production process, characterized in that, include: At least one processor, at least one memory, and computer program instructions stored in the memory, which, when executed by the processor, implement the method as described in any one of claims 1-9.

Images