A method for simulating and optimizing a whole process of a wet-process phosphoric acid production process based on an agent model

Abstract

The application provides a kind of based on agent model's wet-process phosphoric acid production process whole process simulation and optimization method, the method includes: (1) using aspen plus completes the construction of process mechanism model of wet-process phosphoric acid production process flow;(2) selected process key influence variable, generates modeling sample data set by random simulation method;(3) modeling sample data set is trained using machine learning algorithm, constructs wet-process phosphoric acid production process agent model;(4) based on agent model, the integrated optimization calculation of operating parameter set of whole process of phosphoric acid production is carried out;(5) the feasibility verification of optimization scheme is carried out using the mechanism model of aspen plus of process.The method provided by the application realizes the simulation of whole process of wet-process phosphoric acid production and the fast and accurate prediction of key parameter index, and provides accurate scheme and data support for real-time optimization operation and design modification of production, saves experimental and time cost, and has strong engineering practicability.

Description

Technical Field

[0001] This invention belongs to the field of process industry production process simulation and optimization technology, and relates to a method for full-process simulation and optimization of wet-process phosphoric acid production, specifically a method for full-process simulation and optimization of wet-process phosphoric acid production based on a surrogate model. Background Technology

[0002] Wet-process phosphoric acid is the industrial foundation of the phosphate chemical industry. Its efficient and clean production is crucial for ensuring the effective utilization of phosphate rock, a vital non-renewable strategic national resource. With phosphate rock resources becoming increasingly scarce and calls for low-carbon and environmentally friendly practices growing louder, the hemihydrate-dihydrate phosphoric acid process, with its advantages of low energy consumption, high-quality phosphogypsum, and low pollution emissions, has become a direction for the industry's technological transformation and upgrading. However, this process is lengthy, with complex chemical reaction mechanisms, numerous and highly coupled operational variables, and stringent control conditions. Furthermore, the current lack of quantitative understanding and analytical methods for the overall system means that most key parameters cannot be monitored online. This results in problems such as high randomness in current production operations, low efficiency, high consumption, unstable quality, and poor overall competitiveness, which have become bottlenecks restricting the development of my country's phosphoric acid industry. To overcome these challenges, it is necessary to conduct full-process modeling and simulation of the process. The aim is to achieve energy conservation, consumption reduction, stable product quality, and improved production efficiency through accurate prediction of key parameters, optimization and adjustment of production operation parameters, and optimization and transformation of the process flow.

[0003] Aspen Plus is a large-scale general-purpose chemical process simulation platform for rigorous process mechanism calculations. Currently, some scholars both domestically and internationally have used Aspen Plus to model the mechanism of wet-process phosphoric acid production. By changing operating parameters such as sulfuric acid excess coefficient, slurry circulation ratio, temperature, and pressure, they have obtained the optimal operating range of process parameters through sensitivity analysis, thus providing guidance for the optimization design and operation control of the entire process. However, the mechanism model itself is complex and computationally time-consuming. Furthermore, Aspen Plus simulation calculations use a sequential modular method to iteratively solve each device. Directly using this mechanism model for optimization often leads to problems such as excessively long optimization iteration times and difficulty in convergence, thus failing to meet the needs of real-time production control. Therefore, it is necessary to find a proxy model with low computational complexity and high consistency with actual production to accurately and quickly predict key performance indicators such as phosphorus leaching rate and product properties in the phosphate rock acidolysis process. This model should then provide solutions and data support for the operation control and optimization design of the entire process through optimization calculations, thereby reducing enterprise production costs, improving operational management, and promoting and enhancing the automation and intelligence level of the entire phosphoric acid industry. Summary of the Invention

[0004] To address the problems of inability to monitor key parameters online, difficulties in production control, low phosphorus yield, and high costs in existing wet-process phosphoric acid production technologies, this invention aims to provide a method for simulating and optimizing the entire wet-process phosphoric acid production process based on a surrogate model. This method aims to achieve accurate and rapid prediction of key performance indicators such as phosphorus leaching rate and product properties during the acid hydrolysis process of phosphate rock, and to provide solutions and data support for real-time optimization and design modifications of the entire process, thereby achieving energy conservation, emission reduction, and improved economic benefits for enterprises.

[0005] To achieve this objective, the present invention adopts the following technical solution:

[0006] This invention provides a method for simulating and optimizing the entire process of wet-process phosphoric acid production based on a surrogate model. The method includes the following steps:

[0007] (1) The process mechanism model of the wet-process phosphoric acid production process was constructed using Aspen plus;

[0008] (2) Select key influencing variables of the process and generate a modeling sample dataset through random simulation methods;

[0009] (3) Use machine learning algorithms to train the modeling sample dataset obtained in step (2) to construct a proxy model for the wet phosphoric acid production process;

[0010] (4) Conduct integrated optimization calculations of operating parameters for the entire phosphoric acid production process based on the proxy model;

[0011] (5) The feasibility of the optimization scheme was verified by using the Aspen plus mechanism model of the process.

[0012] This invention organically combines rigorous mechanistic models, quasi-Monte Carlo simulations, machine learning modeling, and integrated optimization calculations of operating parameters. It achieves simulation of the entire wet-process phosphoric acid production process and rapid and accurate prediction of key parameters. It also provides precise solutions and data support for real-time optimization and design modifications in production, greatly saving experimental and time costs and demonstrating strong engineering practicality.

[0013] Preferably, the stochastic simulation method in step (2) includes the quasi-Monte Carlo stochastic simulation method.

[0014] As a preferred embodiment of the present invention, the construction in step (1) includes the following steps:

[0015] (1.1) Construct a kinetic model for acidolysis of phosphate rock and a kinetic model for crystallization of calcium sulfate, and fit the parameters of the kinetic model for acidolysis of phosphate rock and the kinetic model for crystallization of calcium sulfate using kinetic experimental data;

[0016] (1.2) Use Fortran to write subroutines for phosphate rock acidolysis kinetics and calcium sulfate crystallization kinetics respectively, and embed the subroutines into the phosphate rock acidolysis kinetics model and calcium sulfate crystallization kinetics model described in step (1.1) respectively to complete the simulation of phosphate rock acidolysis reaction and crystallization process in the semi-aqueous reaction section;

[0017] (1.3) Complete the construction of the process model for the production of hemihydrate-dihydrate wet phosphoric acid in the Aspen Plus platform;

[0018] (1.4) Set the global parameters of the Aspen Plus simulation environment, input the initial material data and unit operation module parameter information, and start the full-process simulation calculation;

[0019] (1.5) Collect and obtain industrial field data, including process operation data and laboratory data, verify the reliability of the whole process mechanism model and correct key parameters to obtain the corrected mechanism model.

[0020] As a preferred technical solution of the present invention, the reaction module selected for the acidolysis kinetic model of phosphate rock in step (1.1) is the RCSTR reaction module, and the reaction kinetic model is the core shrinkage model controlled by diffusion on the solid film surface.

[0021] Preferably, the reaction module selected for the calcium sulfate crystallization kinetic model in step (1.1) is the Crystallizer crystallization module, and the reaction kinetic model is a kinetic equation based on particle number balance.

[0022] As a preferred technical solution of the present invention, the process model for the production of hemihydrate-dihydrate wet phosphoric acid in step (1.3) includes a hemihydrate reaction process model, a hemihydrate filtration process model, a secondary conversion process model, and a dihydrate filtration process model.

[0023] Preferably, the hemihydrate reaction process model uses the RCSTR module to simulate the acidolysis process of phosphate rock, the Flash2 module to simulate the vacuum flash cooling and gas-liquid separation process, the Crystalizer module to simulate the crystallization process of hemihydrate calcium sulfate, and the merging and separation of the streams use the Mixer module and the Fsplit module, respectively.

[0024] Preferably, the semi-aqueous filtration process model uses the Filter module to simulate a multi-stage solid-liquid separation process, the Swash module to simulate a multi-stage washing process, the Fsplit module to simulate the separation of the filtrate and the finished phosphoric acid in the filtration tank, and the Mixer module to simulate the mixing of the liquid in the acid return tank.

[0025] The secondary conversion process model uses the RStoic module to simulate the dihydrate conversion reaction process, the Flash2 module to simulate the vacuum flash cooling and gas-liquid separation process, the Crystalize module to simulate the calcium sulfate dihydrate crystallization process, the Mixer module to simulate the mixer, and the Fsplit module to simulate the splitter.

[0026] The two-water filtration process model uses the Filter module and the Swash module to simulate the multi-stage solid-liquid separation process and the washing process, respectively, and the Fsplit module and the Mixer module to simulate the splitting and mixing of liquid in the two-water filtration tank.

[0027] As a preferred technical solution of the present invention, the environmental global parameters in step (1.4) include component selection, physical property method and flow stream type setting.

[0028] Preferably, the components selected include phosphate rock, sulfuric acid, phosphoric acid, phosphogypsum, gaseous components, and electrolyte solution.

[0029] It is worth noting that the main component of the phosphate rock described in this invention is Ca. 10 The phosphogypsum contains (PO4)6F2, CaF2, and CaCO3, with some impurities including Al2O3, Fe2O3, MgO, SiO2, and Na2O; the main components of the phosphogypsum include CaSO4, CaSO4·2H2O, and CaSO4·0.5H2O; the gaseous components include H2O, CO2, and HF; the ions in the electrolyte solution include H+. + Ca 2+ HPO4 2- PO4 3- HSO4 - SO4 2- F - CO3 2- wait.

[0030] Preferably, the property method includes a global thermodynamic property method and a filtration section property method.

[0031] Preferably, the global thermodynamic property method is the ElECNRTL method.

[0032] Preferably, the property calculation method for the filtration section is the SOLIDS property calculation method.

[0033] Preferably, the stream type includes the MIXCIPSD type.

[0034] It is worth noting that the initial material data and unit operation module parameter information settings in step (1.4) of the method described in this invention require the input of the following data:

[0035] a) Material data: total mass flow rate, temperature, pressure, composition and mass percentage of phosphate rock, and PSD particle size distribution; mass flow rate, temperature, pressure, composition and mass percentage of concentrated sulfuric acid; mass flow rate, temperature, pressure, composition and mass percentage of filtered wash water.

[0036] b) Information on all unit operation modules: volume, chemical reaction formula, reaction kinetic equation, operating pressure and temperature of the reactor module, outlet solid-liquid separation ratio, outlet pressure and temperature changes of the solid-liquid separation module, liquid-solid separation ratio and mixing efficiency of the washing module, and separation efficiency of the splitter module.

[0037] As a preferred technical solution of the present invention, the method for generating the modeling sample dataset in step (2) includes the following steps:

[0038] (2.1) Select the key influencing variables and their range of variation in the process;

[0039] (2.2) A quasi-Monte Carlo sampling method is used to generate a random sampling dataset of the input variables;

[0040] (2.3) Build the Python-Aspen Plus interface to enable Aspen simulation calculations to be started in the Python environment, as well as the input and output of simulation data;

[0041] (2.4) Start Aspen Plus to conduct random simulation experiments and generate modeling sample datasets.

[0042] As a preferred technical solution of the present invention, the key influencing variables of the process in step (2.1) include the inlet concentrated sulfuric acid mass flow rate, the circulating slurry ratio, the back acid distribution ratio, and the temperature drop of the vacuum flash cooling gas;

[0043] Preferably, the Monte Carlo sampling method in step (2.2) includes any one of the random sampling methods based on Halton sequences, Hammersley sequences, or Sobol sequences.

[0044] As a preferred technical solution of the present invention, the machine learning algorithm in step (3) includes any one of regression support vector machine (SVR), neural network (ANN) or random forest decision regression (RF) algorithm.

[0045] Preferably, the construction method in step (3) includes the following steps:

[0046] (3.1) Preprocessing of sample data; the preprocessing includes normalizing and standardizing the sample set so that the feature values ​​and output values ​​fall within the range of [-1,1]; the data is normalized using the following formula; correspondingly, the data needs to be inversely programmed after the prediction is completed;

[0047]

[0048] In the formula, x n x represents the normalized data; x represents the original data; x min Let x be the minimum value; x max Let x be the maximum value;

[0049] (3.2) Random sampling and partitioning of sample data; For the normalized data, according to the K-fold cross-validation method, (K-1) / K of the total number of samples in the dataset are randomly selected each time as the training set, and the remaining 1 / K is used as the validation set. This is repeated K times to generate K sets of training-validation dataset combinations.

[0050] (3.3) Selection of surrogate model structure, i.e., determining the list of input and output variables and the types and structures of machine learning algorithms;

[0051] (3.4) Optimal parameter settings for the proxy model;

[0052] (3.5) Fit three surrogate models, SVR, ANN, and RF, using the training sample set;

[0053] (3.6) Use test sample data to validate and evaluate the established agent model;

[0054] Preferably, the model evaluation criteria in step (3.6) include: evaluating absolute error, root mean square error, and squared correlation coefficient.

[0055] In this invention, the smaller the absolute error and root mean square error of the prediction model, the higher the R... 2 The closer the value is to 1, the higher the prediction accuracy of the model.

[0056] It is worth noting that the evaluation parameters include the absolute error (MAE), root mean square error (RMSE), and squared correlation coefficient (R²). 2 The expressions are as follows:

[0057]

[0058]

[0059]

[0060] In the formula, y iLet m be the true value of the output variable for the i-th sample in the sample set; m represents the number of original samples. These are the predicted values ​​for the corresponding samples; This is the average value of the original sample corresponding to the output variable.

[0061] As a preferred embodiment of the present invention, the optimization calculation in step (4) includes the following steps:

[0062] (4.1) Selection of optimization variables: Through sensitivity analysis, the influence of key operating parameters on the optimization objective is clarified, and they are ranked according to the degree of influence. Parameters that are easy to control on-site and have a greater degree of influence are selected as optimization variables.

[0063] (4.2) Determine the objective function and constraints, and construct the optimization model:

[0064] The objective function includes: using phosphorus yield as the optimization objective, and taking phosphoric acid quality and phosphogypsum quality as constraints; the expression for the optimization objective is as follows:

[0065]

[0066] In the formula, F pra and F phr These are the flow rates of the finished phosphoric acid product and the mass flow rates of the raw phosphate rock, respectively; C PA，pra and C PA，phr These represent the finished phosphoric acid flow rate and the P2O5 concentration in the raw phosphate rock, respectively.

[0067] The constraints include model equality constraints and restrictive inequality constraints; the model constraints refer to the surrogate model of the phosphoric acid production process; the restrictive inequality constraints include production process constraints, outlet product flow and composition constraints, plant load constraints, and boundary conditions for decision variables.

[0068] (4.3) Intelligent algorithms are used to efficiently solve the above optimization model;

[0069] The intelligent algorithm includes any one of the following: particle swarm optimization (PSO), differential algorithm (DE), or genetic algorithm (GA).

[0070] It is worth noting that the constraints described in step (4.2) of this invention specifically include:

[0071] (a) Process control constraints: Slurry component concentration constraints: Temperature constraint of slurry: T i min ≤T i ≤T i max Temperature drop constraint for vacuum flash cooling:

[0072] (b) Product Quality and Composition Constraints: Finished Product Phosphoric Acid Concentration Constraints: Component constraints for phosphogypsum products: Finished phosphoric acid flow rate constraint:

[0073] (c) Unit load constraints: Feed flow rate constraints: Outflow flow constraints:

[0074] (d) Boundary constraints of decision variables:

[0075] In the formula, F represents the mass flow rate; C represents the mass percentage of each component in the liquid phase of the reaction slurry; T represents the slurry temperature; UT represents the temperature drop of the vacuum flash cooler; X represents the optimization decision variable; subscript j = phr, sa, ra, or rs, representing the phosphate rock, sulfuric acid, and back acid entering the reaction unit, respectively; subscript k represents the slurry composition, k = PA, SA, CS, or SC, representing the phosphoric acid, sulfuric acid, and solid phase components in the slurry, respectively; subscript q represents the type of decision variable; q = phr, sa, β, or VC, representing the phosphate rock input, sulfuric acid feed rate, slurry circulation ratio, and vacuum degree of the vacuum flash cooler, respectively; subscript i represents the i-th reaction unit; subscript vc represents the vacuum flash cooler; superscripts in and out represent the inflow and outflow of the stream, respectively; superscripts min and max represent the minimum and maximum value requirements.

[0076] As a preferred technical solution of the present invention, the result standard for the feasibility verification in step (5) is:

[0077] If the simulation results do not meet the process control requirements, the constraints of the optimization model are modified, and the optimization calculation is repeated until the process control requirements are met, and the final feasible optimization solution is output.

[0078] The numerical range described in this invention includes not only the point values ​​listed above, but also any point values ​​within the numerical ranges not listed above. Due to space limitations and for the sake of brevity, this invention will not exhaustively list all the specific point values ​​included in the range.

[0079] Compared with the prior art, the present invention has the following beneficial effects:

[0080] (1) This invention applies rigorous process mechanism simulation to the study of wet-process phosphoric acid production process, making full use of the advantages of the powerful physical properties and unit model library of the existing process software Aspen Plus, as well as the convenience of user model technology to develop custom complex dynamic models, and realizes rigorous mechanism modeling of the entire process.

[0081] (2) This invention generates rich and high-quality process simulation data through the quasi-Monte Carlo stochastic simulation method, which is used to train the surrogate model. This solves the problems of high cost of acquiring industrial data, insufficient amount of effective data, and poor data quality in wet-process phosphoric acid production. It also trains a surrogate model that is highly consistent with the actual process and has low computational complexity, thus realizing the rapid and accurate simulation of the entire process.

[0082] (3) This invention organically combines the proxy model with the integrated calculation of operation parameters to obtain an optimized operation scheme. Then, it uses the Aspen Plus mechanism model simulation to verify and correct the feasibility of the scheme. This solves the problems of long time consumption and difficulty in convergence in the traditional mechanism model-based optimization process. It realizes the rapid and accurate prediction of key performance indicators such as phosphorus leaching rate and product properties in the phosphate rock acidolysis process, and provides accurate schemes and data support for real-time optimization operation and design transformation in production. Attached Figure Description

[0083] Figure 1 This is an Aspen Plus simulation diagram of the hemihydrate-dihydrate wet-process phosphoric acid production process provided in Embodiment 1 of the present invention;

[0084] Figure 2 This is a flowchart of the wet-process phosphoric acid production optimization method based on the surrogate model provided in Embodiment 1 of the present invention;

[0085] Figure 3a This is the prediction result of the proxy model provided in Embodiment 1 of the present invention;

[0086] Figure 3b It is the prediction relative error value of the surrogate model provided in Embodiment 1 of the present invention. Detailed Implementation

[0087] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the embodiments described are merely illustrative of the present invention and should not be considered as specific limitations thereof.

[0088] Example 1

[0089] This embodiment provides a method for simulating and optimizing the entire process of wet-process phosphoric acid production based on a surrogate model. The method includes the following steps:

[0090] (1) The process mechanism model of the wet-process phosphoric acid production process was constructed using Aspen plus;

[0091] (1.1) A kinetic model for acidolysis of phosphate rock and a kinetic model for crystallization of calcium sulfate were constructed, and the model parameters were fitted using kinetic experimental data. The kinetic model for acidolysis of phosphate rock was simulated using the RCSTR reaction module, and the reaction kinetic model adopted the shrinking core model controlled by diffusion on the solid film surface. The kinetic model for crystallization of calcium sulfate was simulated using the Crystallizer crystallization module, and its kinetic model adopted the kinetic equation based on particle number balance.

[0092] The kinetic model expression for the acidolysis of phosphate rock is shown in Equation (1). The effects of sulfuric acid concentration, phosphoric acid concentration, reaction temperature, reaction time, liquid-to-solid ratio, sulfuric acid excess coefficient, and particle size on phosphate leaching were comprehensively investigated through kinetic experiments. The experimental results show that under the experimental conditions, the phosphate rock leaching process is controlled by the diffusion of the surface solid film, that is, the apparent kinetic equation conforms to the shrinkage unreacted core model form of Equation (1); and the sulfuric acid excess coefficient (ξ) and liquid-to-solid ratio (L / S) have different degrees of influence on the reaction kinetics of the phosphate rock leaching process, which can be described by Equation (2).

[0093] 1-2X / 3-(1-X) 2 / 3 =K m t (1)

[0094] K m =K0ξ α [L / S] β exp(-E a / RT) (2)

[0095] In the formula, X represents the acid leaching conversion rate of phosphate rock, and K m The value represents the rate constant of the acidolysis reaction, T represents the absolute temperature, and E represents the absolute temperature. α Let E represent the activation energy, R be the universal gas constant, K0 be the pre-exponential factor, ξ be the excess sulfuric acid coefficient, L / S be the liquid-to-solid ratio, and α and β be the order of influence. This implementation case obtains E through fitting experimental data. α =27.56 kJ / mol, K0 = 0.00306, α = 1.68, β = 1.24. This apparent activation energy is greater than 20 kJ / mol, indicating that the decomposition of phosphate concentrate by sulfur-phosphorus mixed acid is controlled by solid-film surface diffusion.

[0096] The expression of the calcium sulfate crystallization kinetic model is shown in equations (3)-(4). Its crystal growth process conforms to the linear growth law that is independent of particle size. The crystallization equation can be in a semi-empirical form, the temperature can be in the Arrhenius form, and other adjustable parameters can be in the empirical power function form. The specific values ​​can be obtained by fitting multivariate parameters using crystallization kinetic experimental data.

[0097]

[0098] B = K1(C b -C * )+K2=765.95(C b -C * )-1484.2 (4)

[0099] In the formula, G represents the linear growth rate of the crystal, B represents the nucleation rate of the crystal, and C represents the linear growth rate of the crystal. b C represents the solubility of calcium sulfate. * This represents the saturation concentration of calcium sulfate in the slurry environment.

[0100] (1.2) Subroutines for phosphate rock acidolysis kinetics and calcium sulfate crystallization kinetics were written in Fortran and embedded into the RCSTR reaction module and Crystalizer crystallization module in Aspen Plus to simulate the phosphate rock acidolysis reaction and crystallization process in the hemihydrate reaction section.

[0101] (1.3) The entire wet-process phosphoric acid production model was built on the Aspen Plus platform. The simulation process is as follows: Figure 1 As shown, the process model includes four process models: half-water reaction, half-water filtration, secondary conversion, and diwater filtration.

[0102] The hemihydrate reaction process uses multiple RCSTR modules to simulate the acidolysis of phosphate rock, Flash2 modules to simulate the vacuum flash cooling and gas-liquid separation process, Crystalizer modules to simulate the crystallization process of hemihydrate calcium sulfate, and Mixer and Fsplit modules to simulate the merging and separation of streams, respectively.

[0103] The semi-aqueous filtration process uses multiple Filter modules to simulate a multi-stage solid-liquid separation process, multiple Swash modules to simulate a multi-stage washing process, an Fsplit module to simulate the separation of filtrate and finished phosphoric acid in the filtration tank, and a Mixer module to simulate the mixing of liquid in the acid return tank.

[0104] The dihydrate conversion process is simulated using the RStoic module, the Flash2 module is used to simulate the vacuum flash cooling and gas-liquid separation process, the Crystalize module is used to simulate the calcium sulfate dihydrate crystallization process, the Mixer module is used for the mixer, and the Fsplit module is used for the splitter.

[0105] The dihydrate filtration process uses multiple Filter and Swash modules to simulate the multi-stage solid-liquid separation and washing processes, and uses Fsplit and Mixer modules to simulate the liquid splitting and mixing in the dihydrate filtration tank.

[0106] (1.4) Set the global parameters of the Aspen Plus simulation environment, input the initial material data and unit operation module parameter information, and start the full-process simulation calculation;

[0107] The global parameters of the Aspen Plus simulation environment include: settings for component selection, property methods, and stream types, as detailed below:

[0108] 1) The components involved in this embodiment include phosphate rock, sulfuric acid, phosphoric acid, phosphogypsum (CaSO4, CaSO4·2H2O, CaSO4·0.5H2O), gaseous components, and electrolyte solution; the phosphate rock includes main components and some impurities, the main components including Ca 10 (PO4)6F2, CaF2, and CaCO3; the impurities include Al2O3, Fe2O3, MgO, SiO2, and Na2O; the phosphogypsum comprises any one or a combination of at least two of CaSO4, CaSO4·2H2O, or CaSO4·0.5H2O; the gaseous component comprises any one or a combination of at least two of H2O, CO2, or HF; the ions in the electrolyte solution include H+. + Ca 2+ HPO4 2- PO4 3- HSO4 - SO4 2- F - or CO3 2- Any one or at least two of them;

[0109] 2) The property methods include global thermodynamic property methods and filtration section property methods;

[0110] Specifically, the global thermodynamic property method is the ElECNRTL method; the property calculation method for the filtration section is the SOLIDS property calculation method.

[0111] 3) The stream type is selected as MIXCIPSD, which includes two sub-streams: MIXED (for processing gas and liquid phase components) and CIPSD (for processing conventional solid phase).

[0112] The initial material data and unit operation module parameter information are set as follows:

[0113] 1) The material data includes: total mass flow rate, temperature, pressure, composition and mass percentage of phosphate rock, and PSD particle size distribution; mass flow rate, temperature, pressure, composition and mass percentage of concentrated sulfuric acid; and mass flow rate, temperature, pressure, composition and mass percentage of filtered wash water.

[0114] 2) The unit operation module parameter information includes: the volume of the reactor module, chemical reaction formula, reaction kinetic equation, operating pressure and temperature, the outlet solid-liquid separation ratio, outlet pressure and temperature change of the solid-liquid separation module, the liquid-solid separation ratio and mixing efficiency of the washing module, and the separation efficiency of the splitter module.

[0115] (1.5) Collect and acquire industrial field data, including process operation data and laboratory data, verify the reliability of the whole process mechanism model and correct key parameters to obtain the corrected mechanism model;

[0116] This embodiment compares the Aspen Plus simulation results of the corrected mechanism model with actual plant data, and the results are shown in Table 1. It can be seen that the error between the simulated value and the actual value is within an interpretable and acceptable range (maximum error of 10%), indicating that the mechanism model can reflect the actual operation of the hemihydrate-dihydrate wet process phosphoric acid production process well, and can provide a reliable basis for further analysis and optimization.

[0117] Table 1

[0118]

[0119]

[0120] (2) Select key influencing variables of the process and use the quasi-Monte Carlo random simulation method to generate a modeling sample dataset;

[0121] (2.1) Select key process variables and their range of variation. The key process variables include the inlet concentrated sulfuric acid mass flow rate, the ratio of each circulating slurry, the return acid distribution ratio, and the temperature drop of each vacuum flash cooler;

[0122] (2.2) A quasi-Monte Carlo sampling method is used to generate a random sampling dataset of the input variables;

[0123] The quasi-Monte Carlo sampling method includes any one of the Halton sequence random sampling method, the Hammersley sequence random sampling method, or the Sobol sequence random sampling method;

[0124] (2.3) Build the Python-Aspen Plus interface to enable Aspen simulation calculations to be started in the Python environment, as well as the input and output of simulation data;

[0125] (2.4) Start Aspen Plus to conduct random simulation experiments and generate modeling sample datasets. In this implementation case, four input variables were selected, and the Sobol sequence sampling method with good convergence was used to generate 1500 sets of sample data;

[0126] (3) The modeling sample dataset obtained in step (2) is trained using machine learning algorithms to construct a proxy model for the wet-process phosphoric acid production process; the modeling implementation steps are as follows: Figure 2 As shown, it specifically includes the following:

[0127] (3.1) Preprocessing of sample data: In order to avoid the differences in dimensions and orders of magnitude caused by different feature data, the sample set needs to be normalized and standardized first, so that the feature value and output value fall within the range of [-1, 1]. In this embodiment, Equation (5) is used to normalize the data. Correspondingly, after the prediction is completed, the data needs to be denormalized.

[0128]

[0129] In the formula, x n x represents the normalized data; x represents the original data; x min Let x be the minimum value; x max Let x be the maximum value;

[0130] (3.2) Random sampling and partitioning of sample data: For the normalized data, according to the K-fold cross-validation method, (K-1) / K of the total number of samples are randomly selected from the dataset each time as the training set, and the remaining 1 / K is used as the validation set. This is repeated K times to generate K sets of training-validation dataset combinations. In this implementation case, 1200 sets of sample data (accounting for 80% of the total samples) are randomly selected as the training set for model building, and 300 sets (accounting for 20% of the total samples) are used as the test set for model validation.

[0131] (3.3) Selection of the proxy model structure, i.e., determining the list of input and output variables, as well as the type and structure of the machine learning algorithm. In this embodiment, key influencing factors (including the mass flow rate of imported sulfuric acid X1, the internal circulation slurry ratio X2, the temperature drop of the vacuum flash cooler X3, and the acid return ratio X4) are used as input samples, and phosphoric acid production Y is used as the input sample. pra As output samples, proxy models for the wet-process phosphoric acid production process were established using regression-based support vector machine (SVR), neural network (ANN), and random forest decision regression (RF) algorithms, respectively.

[0132] (3.4) Optimal parameter settings for the surrogate model:

[0133] In this embodiment, the basic parameters of the SVR algorithm are set as follows: the radial basis function (RBF) is selected as the kernel function, the penalty factor C is 100, and the influence factor gama is 0.01;

[0134] The basic parameters of the neural network (ANN) algorithm are set as follows: 4 nodes in the input layer, 10 nodes in the hidden layer, 1 node in the output layer, the activation function is the sigmoid function, the optimization algorithm is the Adam adaptive optimization algorithm, the maximum number of training iterations is Epoch = 500, and the target error is E0 = 0.001.

[0135] The basic parameters for the random forest decision regression are set as follows: number of decision trees n_estimators = 100, random state parameter random_state = 42;

[0136] (3.5) Fit three surrogate models, SVR, ANN, and RF, using the training sample set;

[0137] (3.6) Use the test sample data to verify and evaluate the established proxy model, select the model with the highest prediction accuracy as the final proxy model and save it; if the prediction accuracy requirement is not met, add new sample points in step three and continue to retrain the proxy model until the accuracy requirement is met.

[0138] The model evaluation metrics include evaluation absolute error (MAE), root mean square error (RMSE), and squared correlation coefficient (R²). 2 Its expression is shown in equations (6)-(8):

[0139]

[0140]

[0141]

[0142] In the formula, y i Let m be the true value of the output variable for the i-th sample in the sample set; m represents the number of original samples. These are the predicted values ​​for the corresponding samples; The average value of the original samples corresponding to the output variable; the smaller the MAE, MSE and RMSE of the prediction model, and the closer R2 is to 1, the higher the prediction accuracy of the model.

[0143] The evaluation metrics of the three proxy models established in this embodiment on the training and test sets are shown in Table 2. As can be seen from the table, the proxy model based on the RF algorithm has the smallest MSE, RMSE, and MAE on both the training and test sets, and its correlation R... 2 The closest value is 1, meaning the prediction performance is better than SVR and ANN models. Figure 3a and Figure 3bThe figures show the prediction results and relative errors of the established RF model on the training set (first 1200 sets) and the test set (last 300 sets), respectively. Analysis of the figures shows that the RF model performs well in prediction, with most prediction errors controlled within the range of [-2.5%, 2.5%], and the maximum error not exceeding 10%. This indicates that the established surrogate model has good fitting accuracy and generalization ability, and can accurately and quickly predict changes in the wet-process phosphoric acid production process.

[0144] Table 2

[0145]

[0146] (4) Conduct integrated optimization calculations of operating parameters for the entire phosphoric acid production process based on the proxy model;

[0147] (4.1) Selection of optimization variables:

[0148] Sensitivity analysis was used to clarify the degree of influence of key operating parameters on the optimization target, and these parameters were ranked according to their degree of influence. Parameters that were easy to control on-site and had a greater degree of influence were selected as optimization variables. In this embodiment, there were 14 key parameters that directly affected the yield and concentration of phosphoric acid products. Finally, 7 operating parameters with a greater degree of influence and easy to adjust on-site were selected as optimization variables. The specific parameter names and actual production fluctuation ranges are shown in Table 3.

[0149] Table 3

[0150]

[0151]

[0152] (4.2) Determine the objective function and constraints, and construct the optimization model.

[0153] The objective function includes: optimizing phosphorus yield, and considering phosphoric acid quality (P2O5 concentration and SO4 concentration). 2- The concentration of phosphorus gypsum and its quality (residual phosphorus content, moisture content) are used as constraints. The expression for the optimization objective is as follows:

[0154]

[0155] In the formula, F pra and F phr These are the flow rates of the finished phosphoric acid product and the mass flow rates of the raw phosphate rock, respectively; C PA，pra and C PA，phr These represent the finished phosphoric acid flow rate and the P2O5 concentration in the raw phosphate rock, respectively.

[0156] The constraints include model equality constraints and restrictive inequality constraints. The model equality constraints refer to the proxy model for the phosphoric acid production process. The restrictive inequality constraints include production process constraints, outlet product flow and composition constraints, plant load constraints, and boundary conditions for decision variables. Specifically, as follows:

[0157] 1) Process control index constraints

[0158] Slurry component concentration constraints:

[0159] Temperature constraint of slurry: T i min ≤T i ≤T i max (11)

[0160] Temperature drop constraint of vacuum flash cooling:

[0161] 2) Product quality and component constraints

[0162] Concentration constraints of phosphoric acid components in the finished product:

[0163] Component constraints for phosphogypsum products:

[0164] Finished phosphoric acid flow rate constraint:

[0165] 3) Load constraints of the unit:

[0166] Feed flow rate constraints:

[0167] Outflow flow constraints:

[0168] 4) Boundary constraints on decision variables:

[0169] In the formula, F represents the mass flow rate; C represents the mass percentage of each component in the liquid phase of the reaction slurry; T represents the slurry temperature; UT represents the temperature drop of the vacuum flash cooler; X represents the optimization decision variable; subscript j = phr, sa, ra, or rs, representing the phosphate rock, sulfuric acid, and back acid entering the reaction unit, respectively; subscript k represents the slurry composition, k = PA, SA, CS, or SC, representing the phosphoric acid, sulfuric acid, and solid phase components in the slurry, respectively; subscript q represents the type of decision variable; q = phr, sa, β, or VC, representing the phosphate rock input, sulfuric acid feed rate, slurry circulation ratio, and vacuum degree of the vacuum flash cooler, respectively; subscript i represents the i-th reaction unit; subscript vc represents the vacuum flash cooler; superscripts in and out represent the inflow and outflow of the stream, respectively; superscripts min and max represent the minimum and maximum value requirements.

[0170] By integrating the above optimization objectives, variables, and constraints, we can obtain the following optimization model for the entire wet-process phosphoric acid production:

[0171]

[0172]

[0173] (4.3) Intelligent algorithms are used to efficiently solve the above-mentioned optimization model. For the optimization model established above, this embodiment uses intelligent optimization algorithms to obtain the optimized operating scheme; the intelligent optimization algorithms include any one of particle swarm optimization (PSO), differential optimization (DE), or genetic optimization (GA). Furthermore, in order to effectively handle the constraints in the model, a constraint handling method based on feasibility rules is adopted;

[0174] The bond constraint conditions of the treatment method are set as follows: P2O5 concentration X in the finished phosphoric acid PA，pra ≥37%, SO4 concentration in finished phosphoric acid X SA，pra ≤5%, CaSO4 content in phosphogypsum X SC,prg ≥75%; The optimization algorithm selected was the Particle Swarm Optimization (PSO) algorithm, with the main parameters set as follows: population size 50, number of generations 100, crossover probability 0.8, and mutation uniform as the mutation function. The optimization results are shown in Table 4. The maximum phosphorus yield was 98%, the final acid concentration reached 37.5%, and the SO4 concentration was controlled at 4.31%. The corresponding optimal operating parameters x1 to x7 are shown in Table 4.

[0175] Table 4

[0176]

[0177]

[0178] (5) The feasibility of the optimization scheme was verified using the Aspen plus mechanism model of the process:

[0179] If the process control requirements are not met, the constraints of the optimization model are modified, and the optimization calculation is repeated until the process control requirements are met, and the final feasible optimization solution is output.

[0180] The method provided by this invention includes writing an external subroutine for the acidolysis reaction and crystallization kinetics of phosphate rock in Fortran and embedding it into the Aspen Plus platform, realizing a rigorous mechanistic simulation of the entire wet-process phosphoric acid production based on Aspen Plus. Based on a quasi-Monte Carlo stochastic simulation method, rich and high-quality simulation data is generated by random simulation using the Aspen Plus process mechanism model for training the surrogate model, solving the problems of high cost of industrial data acquisition, insufficient effective data volume, and poor data quality in wet-process phosphoric acid production. Furthermore, this invention uses machine learning algorithms to establish a surrogate model that highly matches actual production, and integrates the surrogate model with operational parameter optimization calculations, solving the problems of long optimization time and convergence difficulties in traditional mechanistic model-based optimization processes. This enables rapid and accurate prediction of key performance indicators such as phosphorus leaching rate and product properties in the phosphate rock acidolysis process, and provides precise solutions and data support for real-time optimization operations and design modifications in production, saving significant experimental, manpower, and time costs.

[0181] The applicant declares that the detailed structural features of the present invention are illustrated through the above embodiments, but the present invention is not limited to the above detailed structural features, that is, it does not mean that the present invention must rely on the above detailed structural features to be implemented. Those skilled in the art should understand that any improvements to the present invention, equivalent substitutions for the components selected in the present invention, additions of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

[0182] The applicant declares that the detailed process flow of this invention is illustrated by the above embodiments, but this invention is not limited to the above detailed process flow, that is, it does not mean that this invention must rely on the above detailed process flow to be implemented. Those skilled in the art should understand that any improvements to this invention, equivalent substitutions of raw materials for the product of this invention, addition of auxiliary components, and selection of specific methods, etc., all fall within the protection scope and disclosure scope of this invention.

Claims

1. A method for simulating and optimizing the entire process of wet-process phosphoric acid production based on a surrogate model, characterized in that, The method includes the following steps: (1) The process mechanism model of the wet process for producing hemihydrate-dihydrate phosphoric acid was constructed using Aspen plus; The process mechanism model of the hemihydrate-dihydrate wet process phosphoric acid production process includes a hemihydrate reaction process model, a hemihydrate filtration process model, a secondary conversion process model, and a dihydrate filtration process model. (2) Select key influencing variables of the process and generate a modeling sample dataset through random simulation methods; (3) The modeling sample dataset obtained in step (2) is trained using machine learning algorithms to construct a proxy model for the wet-process phosphoric acid production process; The machine learning algorithm includes any one of the following: regression support vector machine (SVR), neural network (ANN), or random forest decision regression (RF) algorithm; The construction method includes: (3.1) Preprocessing of sample data; the preprocessing includes normalizing and standardizing the sample set so that the feature value and output value fall between [-1, 1]; (3.2) Random sampling and partitioning of sample data; for the normalized data, according to the K-fold cross-validation method, (K-1) / K of the total number of samples in the dataset are randomly selected each time as the training set, and the remaining 1 / K is used as the validation set, and this is repeated K times to generate K sets of training-validation dataset combinations; (3.3) Selection of surrogate model structure, that is, determining the list of input and output variables and the types and structures of machine learning algorithms; (3.4) Optimal parameter setting of surrogate model; (3.5) Fitting three surrogate models, SVR, ANN, and RF, using the training sample set; (3.6) Validating and evaluating the constructed surrogate model using test sample data; (4) Conduct integrated optimization calculations of operating parameters for the entire phosphoric acid production process based on the proxy model; (5) The feasibility of the optimization scheme was verified using the Aspen plus mechanism model of the process; The criteria for the feasibility verification results are as follows: If the simulation results do not meet the process control requirements, the constraints of the optimization model are modified, and the optimization calculation is repeated until the process control requirements are met, and the final feasible optimization solution is output.

2. The method according to claim 1, characterized in that, The construction described in step (1) includes the following steps: (1.1) Construct a kinetic model for acidolysis of phosphate rock and a kinetic model for crystallization of calcium sulfate, and fit the parameters of the kinetic model for acidolysis of phosphate rock and the kinetic model for crystallization of calcium sulfate using kinetic experimental data; (1.2) Use Fortran to write subroutines for phosphate rock acidolysis kinetics and calcium sulfate crystallization kinetics, and embed the subroutines into the phosphate rock acidolysis kinetics model and calcium sulfate crystallization kinetics model described in step (1.1) respectively, to simulate the phosphate rock acidolysis reaction and crystallization process in the semi-aqueous reaction section; (1.3) Complete the construction of the process model for the production of hemihydrate-dihydrate wet phosphoric acid in the Aspen Plus platform; (1.4) Set the global parameters of the Aspen Plus simulation environment, input the initial material data and unit operation module parameter information, and start the full-process simulation calculation; (1.5) Collect and obtain industrial field data, including process operation data and laboratory data, verify the reliability of the whole process mechanism model and correct key parameters to obtain the corrected mechanism model.

3. The method according to claim 2, characterized in that, The reaction module selected for the phosphate rock acidolysis kinetic model in step (1.1) is the RCSTR reaction module, and the reaction kinetic model is the solid film surface diffusion-controlled shrinking core model. The reaction module selected for the calcium sulfate crystallization kinetic model in step (1.1) is the Crystallizer crystallization module, and the reaction kinetic model is a kinetic equation based on particle number balance.

4. The method according to claim 2, characterized in that, The process model for producing hemihydrate-dihydrate wet phosphoric acid in step (1.3) includes a hemihydrate reaction process model, a hemihydrate filtration process model, a secondary conversion process model, and a dihydrate filtration process model. The hemihydrate reaction process model uses the RCSTR module to simulate the acidolysis process of phosphate rock, the Flash2 module to simulate the vacuum flash cooling and gas-liquid separation process, and the Crystalizer module to simulate the crystallization process of hemihydrate calcium sulfate. The merging and separation of the streams are respectively handled by the Mixer module and the Fsplit module. The semi-aqueous filtration process model uses the Filter module to simulate a multi-stage solid-liquid separation process, the Swash module to simulate a multi-stage washing process, the Fsplit module to simulate the separation of filtrate and finished phosphoric acid in the filtration tank, and the Mixer module to simulate the mixing of liquid in the acid return tank. The secondary conversion process model uses the RStoic module to simulate the dihydrate conversion reaction process, the Flash2 module to simulate the vacuum flash cooling and gas-liquid separation process, the Crystalize module to simulate the dihydrate calcium sulfate crystallization process, the Mixer module to simulate the mixer, and the Fsplit module to simulate the splitter. The two-water filtration process model uses the Filter module and the Swash module to simulate the multi-stage solid-liquid separation process and the washing process, respectively, and the Fsplit module and the Mixer module to simulate the liquid splitting and mixing in the two-water filtration tank, respectively.

5. The method according to claim 2, characterized in that, The environmental global parameters mentioned in step (1.4) include component selection, property method, and stream type settings; The components selected include phosphate rock, sulfuric acid, phosphoric acid, phosphogypsum, gaseous components, and electrolyte solutions. The property methods include global thermodynamic property methods and filtration process property methods; The global thermodynamic property method is the ElECNRTL method; The physical property method for the filtration section is the SOLIDS physical property calculation method; The stream type includes the MIXCIPSD type.

6. The method according to claim 1, characterized in that, The method for generating the modeling sample dataset in step (2) includes the following steps: (2.1) Select the key influencing variables and their range of variation in the process; (2.2) A quasi-Monte Carlo sampling method is used to generate a random sampling dataset of the input variables; (2.3) Build the Python-Aspen Plus interface to enable Aspen simulation calculations to be started in the Python environment, as well as the input and output of simulation data; (2.4) Start Aspen Plus to conduct random simulation experiments and generate modeling sample datasets.

7. The method according to claim 6, characterized in that, The key influencing variables in step (2.1) include the inlet concentrated sulfuric acid mass flow rate, the circulating slurry ratio, the back acid distribution ratio, and the temperature drop of the vacuum flash cooling gas; The Monte Carlo sampling method described in step (2.2) includes random sampling methods based on Halton sequences, Hammersley sequences, or Sobol sequences.

8. The method according to claim 1, characterized in that, The model evaluation criteria described in step (3.6) include: evaluating absolute error, root mean square error, and squared correlation coefficient.

9. The method according to claim 1, characterized in that, The optimization calculation in step (4) includes the following steps: (4.1) Selection of optimization variables: The influence of key operating parameters on the optimization objective is clarified through sensitivity analysis, and the parameters are ranked according to the degree of influence. The parameters that are easy to control on site and have a greater degree of influence are selected as optimization variables. (4.2) Determine the objective function and constraints, and construct the optimization model: The objective function includes: taking phosphorus yield as the optimization objective, and using phosphoric acid quality and phosphogypsum quality as constraints. The expression for the optimization objective is as follows: In the formula, and These are the flow rates of the finished phosphoric acid product and the mass flow rates of the raw phosphate rock, respectively. and These represent the finished phosphoric acid flow rate and the P2O5 concentration in the raw phosphate rock, respectively. The constraints include model equality constraints and restrictive inequality constraints; the model constraints refer to the surrogate model of the phosphoric acid production process; the restrictive inequality constraints include production process constraints, outlet product flow and composition constraints, plant load constraints, and boundary conditions for decision variables. (4.3) Intelligent algorithms are used to efficiently solve the above optimization model; The intelligent algorithm includes any one of the following: particle swarm optimization (PSO), differential algorithm (DE), or genetic algorithm (GA).

Images