Intelligent optimization automatic control system and method for wet-process phosphoric acid fluorine absorption process

Abstract

The invention discloses an intelligent optimization automatic control system and method for a wet-process phosphoric acid fluorine absorption process. The system comprises a data acquisition module, and the data acquisition module is in communication connection with a prediction model building module through a data transmission module; the prediction model construction module is in communication connection with the control optimization module through the data transmission module, and the control optimization module is electrically connected with the control execution module; the data acquisition module acquires data to construct a model sample data set; a fluorine absorption prediction model is constructed based on the model sample data set, and the fluosilicic acid concentration and the scaling rate are predicted through the constructed fluorine absorption prediction model; based on the prediction result and the real-time data acquired by the data acquisition module, control variable optimization is carried out through a control optimization model of multi-objective optimization, and an optimal control variable value is obtained; and the control execution module receives the optimal control variable value and calls a flow control valve to perform control and adjustment, so that the wet-process phosphoric acid fluorine absorption system reaches the optimal working condition.

Description

Technical Field

[0001] The invention relates to the field of chemical process control, and in particular to an intelligent optimization automatic control system and method for a wet-process phosphoric acid fluoride absorption process. Background Art

[0002] In the production of wet phosphoric acid, the effective absorption and treatment of fluorine-containing gases is crucial. Wet phosphoric acid production usually produces a large amount of fluorine-containing gases. If these gases are not effectively absorbed and treated, they will not only pollute the environment, but also cause potential harm to equipment and operators. Therefore, how to efficiently absorb and recycle fluoride while ensuring that tail gas emissions meet emission standards is an urgent problem to be solved in the wet phosphoric acid process.

[0003] The existing fluorine absorption process usually adopts the form of a spray tower or a packed tower, which makes the fluorine-containing gas fully contact with the circulating fluorosilicic acid solution through countercurrent spraying or packing contact to achieve the purpose of absorbing fluoride. However, the existing technology has the following major problems: 1. Fluorosilicic acid concentration control is unstable: Traditional fluorine absorption systems mostly rely on the operator's experience to adjust the concentration of fluorosilicic acid by controlling the flow of inlet and circulating liquid. However, this manual adjustment method has a large hysteresis and uncertainty, resulting in large fluctuations in the concentration of fluorosilicic acid, which is difficult to maintain within the ideal range; the instability of the concentration of fluorosilicic acid directly affects the efficiency and quality of downstream fluorine chemical production. If the concentration of fluorosilicic acid is too low, the energy consumption of the subsequent concentration step needs to be increased; if the concentration is too high, it may lead to incomplete absorption of fluoride in the absorption tower, thus affecting the compliance of exhaust gas emissions.

[0004] 2. Tail gas emissions are unstable and there is a risk of exceeding the standard: During the fluorine absorption process, the concentration of fluoride in the exhaust gas must be strictly controlled to meet national emission standards. However, due to fluctuations in fluorosilicic acid concentration and absorption efficiency, the concentration of fluoride in exhaust emissions is difficult to maintain stable and often exceeds the standard; exhaust gas exceeding the standard not only brings environmental risks, but may also cause the company to face penalties from regulatory authorities, affecting the company's normal production and economic benefits. Therefore, under existing technologies, how to monitor and adjust the fluoride concentration in real time and ensure that exhaust emissions are always qualified is an important technical challenge.

[0005] 3. Large temperature fluctuations lead to serious scaling problems: During the operation of the fluorine absorption tower, temperature fluctuations will significantly affect the generation of silica gel (SiO2). When the temperature is too low, silica gel is deposited in the circulation pipes, nozzles and sealing grooves of the absorption tower to form scale. These scales will not only block the pipes and nozzles, but also reduce the gas-liquid mass transfer efficiency and reduce the absorption efficiency. In order to solve the scaling problem, the existing technology usually adopts regular chemical cleaning or mechanical cleaning methods; however, this maintenance method requires downtime, which not only increases maintenance costs, but also affects the continuity of production. At the same time, excessive cleaning will accelerate the corrosion of equipment and shorten the service life of equipment.

[0006] 4. Frequent failures and difficulty in equipment maintenance and management: The existing fluorine absorption system relies on manual operation and experience-based adjustment, which makes it difficult to monitor and predict system failures in real time. Once equipment such as pipes, nozzles and circulation pumps are blocked or fail, they often need to be shut down for maintenance, which seriously affects the continuity and economic benefits of production; traditional systems lack fault warning functions. When equipment problems occur, serious fault risks have often accumulated, resulting in extended downtime and increased maintenance costs.

[0007] In summary, the existing wet phosphoric acid fluoride absorption technology has significant deficiencies in terms of fluorosilicic acid concentration control, tail gas emission compliance, scaling problem treatment, temperature control and equipment failure prediction. These problems lead to increased production costs, low system efficiency, and increased risks to environmental protection and equipment safety during the production process. Therefore, it is necessary to design an intelligent optimization automatic control system and method for the wet phosphoric acid fluoride absorption process to solve the above problems. Summary of the invention

[0008] The technical problem to be solved by the present invention is to provide an intelligent optimization automatic control system and method for a wet-process phosphoric acid fluoride absorption process, which, under the premise of ensuring qualified tail gas emissions, increases the extraction concentration of fluorosilicic acid, slows down scaling in fluorine absorption circulation pipelines, nozzles and fluorine absorption sealing tanks, and simultaneously achieves optimized control of energy consumption, and improves the stability of the system through a fault prediction function.

[0009] In order to achieve the above technical effects, the technical solution adopted by the present invention is: An intelligent optimization automatic control system for wet-process phosphoric acid fluoride absorption process includes a data acquisition module, which is connected to a prediction model building module through a data transmission module; the prediction model building module is connected to a control optimization module through a data transmission module, and the control optimization module is electrically connected to a control execution module; The data acquisition module includes a sensor integration, which includes a high-precision density meter, an online gas analyzer, a flow meter, a temperature sensor and a pH sensor; the high-precision density meter is used to collect the real-time fluorosilicic acid concentration and scaling rate in the fluorine absorption process, the online gas analyzer is used to collect the tail gas fluoride concentration, the flow meter is used to collect the circulating liquid concentration, the temperature sensor is used to collect the temperature in the absorption tower, and the pH sensor is used to collect the pH value of the circulating liquid; the data acquisition module also includes a database interface, which is connected to the historical database of the wet-process phosphoric acid fluoride absorption central control system for collecting historical data; The prediction model building module is used to build a fluorine absorption prediction model based on random forests, providing data changes in the fluorine absorption process in the future period; The control optimization module is used to construct a control optimization model based on multi-objective optimization. The control optimization model provides optimal set values ​​for the control variables of the control execution module based on the MPC algorithm. The control execution module includes a plurality of flow control valves in the wet process phosphoric acid fluoride absorption control system, which are used to control the withdrawal flow, the chilled water flow and the circulating liquid flow; The data transmission module includes Ethernet, Bluetooth transceiver module and WiFi communication module.

[0010] Preferably, the method for intelligently optimizing the automatic control system of the wet-process phosphoric acid fluoride absorption process comprises: S1, the data acquisition module collects historical data and real-time data of the wet-process phosphoric acid fluoride absorption process to build a model sample data set; S2, the prediction model building module builds a fluorine absorption prediction model based on the model sample data set; S3, predicting the fluorosilicic acid concentration and scaling rate according to the real-time input data of the current wet phosphoric acid fluoride absorption process through the constructed fluorine absorption prediction model; S4, based on the prediction result obtained in step S3 and the real-time data collected by the data acquisition module, the control variables are optimized through the multi-objective optimization control optimization model to obtain the optimal control variable value; S5, the control execution module receives the optimal control variable value and mobilizes the flow control valve to perform control and adjustment, so that the wet phosphoric acid fluoride absorption system reaches the optimal operating condition.

[0011] Preferably, in step S1, the method for constructing the model sample data set is: S101, construct variable set x Absorption and x Scalability ;x Absorption Represents the main variables in the fluorine absorption process, including the temperature T in the absorption tower, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the fluorosilicic acid concentration C H ;x Scalabilityrepresents the key variables that affect the scaling rate, including the absorber temperature T, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the scaling rate G scale ; Each set of variables is represented as follows: x Absorption =[T,pH,v,C H ]; x Scalability =[T,pH,v,G scale ]; S102, dividing the multiple arrays in the two constructed variable sets into labeled data sets and unlabeled data sets: Labeled dataset: Extract key features from the above variables and construct a labeled dataset D labeled ={(x labeled ,y labeled )}, where x labeled A characteristic combination of temperature, pH, flow rate, fluorosilicic acid concentration and scaling rate, y labeled Indicates the actual fluorosilicic acid concentration or scaling rate; Unlabeled dataset: Unlabeled samples of fluorine absorption process collected by sensor integration x unlabeled ⊂x Process , used to generate pseudo-labeled data to improve the prediction accuracy of the model.

[0012] Preferably, in step S2, the method for constructing a fluorine absorption prediction model is: A fluorine absorption prediction model was constructed based on the semi-supervised random forest algorithm, and pseudo labels of unlabeled data were used to improve the prediction accuracy: S201, model parameter setting: Initialize the number of decision trees K to 50-100 and optimize through cross-validation to optimize the stability of prediction; Initialize the maximum tree depth to 4-7 layers to ensure generalization ability and avoid overfitting; Initialize the minimum number of split samples: set the number of samples for each leaf node to 10-20 to prevent overfitting; S202, model training and optimization: Recursive segmentation process: For each tree node, use information gain or mean square error MSE to select the optimal segmentation features and segmentation points: ; Where, R1 and R2: represent two different regions or partitions of the data set, generated in the split node of the decision tree; |R1| and |R2|: represent the number of samples in regions R1 and R2, used to normalize the error term; and : represents the average predicted value in regions R1 and R2, which is used to calculate the error; y: represents the actual label value, which is used as the true value when calculating the mean square error or information gain; Feature selection and randomness: Each layer of segmentation selects only some features including temperature, flow rate and pH value to enhance model diversity; Set tree growth stop conditions: Each tree stops growing when any of the following conditions is met: a. Reach maximum depth; b. The number of leaf node samples is less than the set threshold; c. Information gain is lower than the threshold; Pseudo-label generation and model updating; Pseudo-label generation: Use the trained random forest model to predict unlabeled data x unlabeled Pseudo labels: ; In the formula, x unlabeled : represents an unlabeled data sample, which has no real label and is processed using a pseudo-labeling method; pseudo : represents pseudo-labeled values, which are used for prediction of unlabeled data. Generate prediction labels for unlabeled data through ensemble models (such as random forests) to enhance the training effect of the model; : Represents the prediction function of the kth random forest decision tree, which is used for the unlabeled sample x unlabeled Generate forecast values; Screening effective pseudo-labels: Screening pseudo-label samples with high confidence based on the root mean square error (RMSE), updating the labeled data set, and forming a new training set D new−train ; S203, integrate the results of all decision trees to form the final prediction output: Based on the weighted average method, the output of each tree is weighted based on its prediction accuracy, and the weighted average is obtained as the model output: ; Based on the mean method, the predicted values ​​of all decision trees are averaged: ; K: represents the number of decision trees. In ensemble learning algorithms such as random forest, the prediction results of K decision trees will be integrated; : represents the final prediction output, which is usually the weighted or average result of multiple model prediction results, used to improve the stability of the overall prediction; w k : represents the weight of the kth decision tree. The weight is usually assigned according to the prediction accuracy of the decision tree to optimize the accuracy of the final prediction; : represents the prediction output of the kth decision tree, that is, the prediction value of a single tree for a sample; S204, model performance optimization: (1) Hyperparameter optimization: adjust the number of decision trees K, tree depth, and minimum number of samples through grid search or cross-validation to ensure model stability and accuracy; (2) Segmentation criterion optimization: adjust the information gain and MSE indicators to improve the accuracy of segmentation point selection, optimize the temperature, flow rate and pH value, and enhance the accuracy of capturing key variables; (3) Randomness optimization: The random subspace method RSM is introduced to select only some features in each layer to improve the diversity and generalization ability of the model; S205, based on the optimization of the deep forest regression module DFR: After generating pseudo-labels and updating the dataset, a deep forest regression model was used to make refined predictions of fluorosilicic acid concentration and scaling rate; The DFR module consists of an input layer, an intermediate layer, and an output layer, and improves the prediction effect by recursively fusing features layer by layer; (1) The input layer forest module performs sampling and sub-forest generation: new−train Perform Bootstrap and RSM sampling to generate I sub-forests; Prediction mean calculation: prediction mean for each sub-forest Weighted, select the predicted value closest to the mean to form a regression vector; (2) The middle-layer forest module processes the output of the input layer layer by layer to form a recursive regression vector. The output of each layer is used as the input of the next layer, and the prediction accuracy is enhanced recursively layer by layer. (3) The output layer forest module receives the regression vector output by the intermediate layer and makes predictions through multiple decision trees. The final result is the mean of the outputs of all sub-forests. .

[0013] In the formula, : represents the final prediction output of the deep forest regression (DFR) module. This value is obtained by averaging the prediction results of all sub-forests; I: represents the number of sub-forests, that is, the total number of sub-forests generated at the input layer of the DFR module; Represents the predicted mean of the iiith sub-forest. Each sub-forest makes independent predictions on the data and then generates a mean as the prediction result of the sub-forest.

[0014] Preferably, in step S4, the control optimization model construction method based on multi-objective optimization is: With increasing the concentration of fluorosilicic acid as the main goal, the exhaust emissions, scaling rate and energy consumption factors were comprehensively considered during the optimization process, and the objective function was constructed as follows: ; Among them, x1(k) is the concentration of fluorosilicic acid, x3(k) is the exhaust gas fluoride concentration, Gscaling(k) is the scaling rate; w1, w2, w3, w4 are weight coefficients used to adjust the importance of different objectives; by adjusting these weights, the relationship between concentration optimization, emission compliance, energy consumption control and scaling inhibition can be balanced.

[0015] Preferably, in order to ensure process safety and long-term operation of the equipment, multiple constraints are set in the control optimization model of multi-objective optimization, including: Exhaust emission constraints: ; The concentration of fluoride in tail gas shall not exceed the national standard; Scaling inhibition constraints: ; Where Gmax is the upper limit of the structural rate; monitor the flow rate, temperature and pH value of the fluorosilicic acid sealing tank, circulation pipeline and nozzle, and predict the scaling risk based on the fluorine absorption prediction model. When the scaling rate does not reach or exceed the critical value; Energy consumption control constraints: ; When the fluorosilicic acid concentration reaches the target, the energy consumption in the cooling process is minimized.

[0016] Preferably, in step S4, the method of optimizing the control variables by using the multi-objective optimization control optimization model to obtain the optimal control variable value is: S401, based on the MPC algorithm, the control variables are optimized. First, the state variables and control variables of the MPC algorithm are set: The state variables include: X1(k): Fluorosilicic acid concentration (g / cm³), measured in real time by sensor integration in the sealing tank; X2(k): temperature inside the absorption tower (°C), obtained through sensor integration; X3(k): fluoride concentration in tail gas (mg / m³), monitored by online gas analyzer; X4(k): fouling rate (g / h), generated by flow rate, temperature and fluorine absorption prediction model; Control variables: u1(k): Take-out flow rate (m³ / h), controls the discharge amount of fluorosilicic acid in the sealing groove; u2(k): Chilled water flow rate (m³ / h), used to adjust the temperature in the absorption tower to improve absorption efficiency; u3(k): circulating liquid flow rate (m³ / h), controls the gas-liquid contact time and absorption efficiency; S402, construct the state equation of the MPC algorithm: Fluorosilicic acid concentration dynamic equation: ; In the formula, Q in Indicates the liquid flow rate entering the sealing groove (m³ / h); C in Indicates the concentration of fluorosilicic acid entering the liquid; V indicates the effective volume of the sealing groove; Fluorine absorption tower temperature equation: ; In the formula, C p represents the specific heat capacity of the absorption liquid (kJ / (kg·℃)); Q heat Indicates the heat released by the gas in the absorption tower; Fluoride concentration equation of tail gas: ; In the formula, C F,in Indicates the fluoride concentration in the gas entering the absorption tower; K abs It represents the gas-liquid mass transfer coefficient, which is related to temperature and liquid flow rate factors; Scaling rate equation: ; Wherein, function f represents the effect of temperature, fluorosilicic acid concentration and pH value on the silica gel formation rate.

[0017] Preferably, based on the state equation of the constructed MPC algorithm, multi-objective optimization is performed through the objective function of the control optimization model of multi-objective optimization to maximize the fluorosilicic acid concentration, minimize the tail gas fluoride concentration and scaling rate, and reduce energy consumption. In the multi-objective function, w1 is used to increase the weight of the fluorosilicic acid concentration, w2 is used to ensure that the fluoride concentration in the tail gas meets the national emission standards, w3 is used to control the use of chilled water to reduce energy consumption, and w4 is used to limit the scaling rate and reduce the frequency of equipment maintenance.

[0018] Preferably, in step S5, mobilizing the flow control valve to perform control and regulation includes adjusting control variables, and the control variables include the extraction flow u1(k), the chilled water flow u2(k) and the circulating liquid flow u3(k).

[0019] Preferably, in step S4, the real-time data collected by the data collection module is filtered in advance by an adaptive Kalman filter to filter out noise.

[0020] The beneficial effects of the present invention are: 1. In the traditional process, the dependent variables at the front end of the wet phosphoric acid fluoride absorption system, such as the temperature of the absorption tower, the discharge amount of fluorosilicic acid in the sealing tank, and the gas-liquid contact time, all affect the final fluorosilicic acid concentration and scaling rate. However, there is a lag in this influence process. After the front-end dependent variable is adjusted, it will take a period of time for the change to be reflected in the tail-end data. Therefore, the adjustment process is blind and oscillating. The present invention predicts the fluorosilicic acid concentration and scaling rate in the future time period by constructing a fluorine absorption prediction model. After the front-end dependent variable data is adjusted, the predicted values ​​of the fluorosilicic acid concentration and scaling rate can be immediately obtained through the fluorine absorption prediction model, so that the algorithm-based multi-objective optimization can be performed according to the real-time predicted data, solving the problem of lag in wet phosphoric acid fluoride absorption data.

[0021] 2. In constructing the sample set for constructing the fluorine absorption prediction model, the present invention constructs an unlabeled data set and utilizes the unlabeled data collected by the data acquisition module, thereby solving the problem of insufficient labeled sample data in the historical database in traditional processes, thereby improving the accuracy of the constructed prediction model.

[0022] 3. The present invention constructs a multi-objective optimization model based on the MPC algorithm, cooperates with the fluorine absorption prediction model, and performs multi-objective optimization based on the measured data and the predicted data, thereby providing accurate data support for the control execution module, avoiding blind adjustment and repeated oscillation adjustment, and can accurately adjust the system to the optimal working condition through one adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is a schematic diagram of the module architecture of the present invention; Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0024] Embodiment 1: like Figure 1 As shown, the intelligent optimization automatic control system for wet phosphoric acid fluoride absorption process includes a data acquisition module, which is connected to the prediction model construction module through the data transmission module; the prediction model construction module is connected to the control optimization module through the data transmission module, and the control optimization module is electrically connected to the control execution module; The data acquisition module includes a sensor integration, which includes a high-precision density meter, an online gas analyzer, a flow meter, a temperature sensor and a pH sensor; the high-precision density meter is used to collect the real-time fluorosilicic acid concentration and scaling rate in the fluorine absorption process, the online gas analyzer is used to collect the tail gas fluoride concentration, the flow meter is used to collect the circulating liquid concentration, the temperature sensor is used to collect the temperature in the absorption tower, and the pH sensor is used to collect the pH value of the circulating liquid; the data acquisition module also includes a database interface, which is connected to the historical database of the wet-process phosphoric acid fluoride absorption central control system for collecting historical data; The prediction model building module is used to build a fluorine absorption prediction model based on random forests, providing data changes in the fluorine absorption process in the future period; The control optimization module is used to construct a control optimization model based on multi-objective optimization. The control optimization model provides optimal set values ​​for the control variables of the control execution module based on the MPC algorithm. The control execution module includes a plurality of flow control valves in the wet process phosphoric acid fluoride absorption control system, which are used to control the withdrawal flow, the chilled water flow and the circulating liquid flow; The data transmission module includes Ethernet, Bluetooth transceiver module and WiFi communication module.

[0025] like Figure 2 As shown, the method for intelligently optimizing the automatic control system of the wet phosphoric acid fluoride absorption process comprises: S1, the data acquisition module collects historical data and real-time data of the wet-process phosphoric acid fluoride absorption process to build a model sample data set; S2, the prediction model building module builds a fluorine absorption prediction model based on the model sample data set; S3, predicting the fluorosilicic acid concentration and scaling rate according to the real-time input data of the current wet phosphoric acid fluoride absorption process through the constructed fluorine absorption prediction model; S4, based on the prediction result obtained in step S3 and the real-time data collected by the data acquisition module, the control variables are optimized through the multi-objective optimization control optimization model to obtain the optimal control variable value; S5, the control execution module receives the optimal control variable value and mobilizes the flow control valve to perform control and adjustment, so that the wet phosphoric acid fluoride absorption system reaches the optimal operating condition.

[0026] Preferably, in step S1, the method for constructing the model sample data set is: S101, construct variable set x Absorption and x Scalability ;x Absorption Represents the main variables in the fluorine absorption process, including the temperature T in the absorption tower, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the fluorosilicic acid concentration C H ;xScalability represents the key variables that affect the scaling rate, including the absorber temperature T, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the scaling rate G scale ; Each set of variables is represented as follows: x Absorption =[T,pH,v,C H ]; x Scalability =[T,pH,v,G scale ]; S102, dividing the multiple arrays in the two constructed variable sets into labeled data sets and unlabeled data sets: Labeled dataset: Extract key features from the above variables and construct a labeled dataset D labeled ={(x labeled ,y labeled )}, where x labeled A characteristic combination of temperature, pH, flow rate, fluorosilicic acid concentration and scaling rate, y labeled Indicates the actual fluorosilicic acid concentration or scaling rate; Unlabeled dataset: Unlabeled samples of fluorine absorption process collected by sensor integration x unlabeled ⊂x Process , used to generate pseudo-labeled data to improve the prediction accuracy of the model.

[0027] Preferably, in step S2, the method for constructing a fluorine absorption prediction model is: A fluorine absorption prediction model was constructed based on the semi-supervised random forest algorithm, and pseudo labels of unlabeled data were used to improve the prediction accuracy: S201, model parameter setting: Initialize the number of decision trees K to 50-100 and optimize through cross-validation to optimize the stability of prediction; Initialize the maximum tree depth to 4-7 layers to ensure generalization ability and avoid overfitting; Initialize the minimum number of split samples: set the number of samples for each leaf node to 10-20 to prevent overfitting; S202, model training and optimization: Recursive segmentation process: For each tree node, use information gain or mean square error MSE to select the optimal segmentation features and segmentation points: ; Where, R1 and R2: represent two different regions or partitions of the data set, generated in the split node of the decision tree; |R1| and |R2|: represent the number of samples in regions R1 and R2, used to normalize the error term; and : represents the average predicted value in regions R1 and R2, which is used to calculate the error; y: represents the actual label value, which is used as the true value when calculating the mean square error or information gain; Feature selection and randomness: Each layer of segmentation selects only some features including temperature, flow rate and pH value to enhance model diversity; Set tree growth stop conditions: Each tree stops growing when any of the following conditions is met: a. Reach maximum depth; b. The number of leaf node samples is less than the set threshold; c. Information gain is lower than the threshold; Pseudo-label generation and model updating; Pseudo-label generation: Use the trained random forest model to predict unlabeled data x unlabeled Pseudo labels: ; In the formula, x unlabeled : represents an unlabeled data sample, which has no real label and is processed using a pseudo-labeling method; pseudo : represents pseudo-labeled values, which are used for prediction of unlabeled data. Generate prediction labels for unlabeled data through ensemble models (such as random forests) to enhance the training effect of the model; : Represents the prediction function of the kth random forest decision tree, which is used for the unlabeled sample x unlabeled Generate forecast values; Screening effective pseudo-labels: Screening pseudo-label samples with high confidence based on the root mean square error (RMSE), updating the labeled data set, and forming a new training set D new−train ; S203, integrate the results of all decision trees to form the final prediction output: Based on the weighted average method, the output of each tree is weighted based on its prediction accuracy, and the weighted average is obtained as the model output: ; K: represents the number of decision trees. In ensemble learning algorithms such as random forest, the prediction results of K decision trees will be integrated; : represents the final prediction output, which is usually the weighted or average result of multiple model prediction results, used to improve the stability of the overall prediction; w k : represents the weight of the kth decision tree. The weight is usually assigned according to the prediction accuracy of the decision tree to optimize the accuracy of the final prediction; : represents the prediction output of the kth decision tree, that is, the prediction value of a single tree for a sample; Based on the mean method, the predicted values ​​of all decision trees are averaged: ; S204, model performance optimization: (1) Hyperparameter optimization: adjust the number of decision trees K, tree depth, and minimum number of samples through grid search or cross-validation to ensure model stability and accuracy; (2) Segmentation criterion optimization: adjust the information gain and MSE indicators to improve the accuracy of segmentation point selection, optimize the temperature, flow rate and pH value, and enhance the accuracy of capturing key variables; (3) Randomness optimization: The random subspace method RSM is introduced to select only some features in each layer to improve the diversity and generalization ability of the model; S205, based on the optimization of the deep forest regression module DFR: After generating pseudo-labels and updating the dataset, a deep forest regression model was used to make refined predictions of fluorosilicic acid concentration and scaling rate; The DFR module consists of an input layer, an intermediate layer, and an output layer, and improves the prediction effect by recursively fusing features layer by layer; (1) The input layer forest module performs sampling and sub-forest generation: new−train Perform Bootstrap and RSM sampling to generate I sub-forests; Prediction mean calculation: prediction mean for each sub-forest Weighted, select the predicted value closest to the mean to form a regression vector; (2) The middle-layer forest module processes the output of the input layer layer by layer to form a recursive regression vector. The output of each layer is used as the input of the next layer, and the prediction accuracy is enhanced recursively layer by layer. (3) The output layer forest module receives the regression vector output by the intermediate layer and makes predictions through multiple decision trees. The final result is the mean of the outputs of all sub-forests. .

[0028] In the formula, : represents the final prediction output of the deep forest regression (DFR) module. This value is obtained by averaging the prediction results of all sub-forests; I: represents the number of sub-forests, that is, the total number of sub-forests generated at the input layer of the DFR module; Represents the predicted mean of the iiith sub-forest. Each sub-forest makes independent predictions on the data and then generates a mean as the prediction result of the sub-forest.

[0029] Preferably, in step S4, the control optimization model construction method based on multi-objective optimization is: With increasing the concentration of fluorosilicic acid as the main goal, the exhaust emissions, scaling rate and energy consumption factors were comprehensively considered during the optimization process, and the objective function was constructed as follows: ; Among them, x1(k) is the concentration of fluorosilicic acid, x3(k) is the exhaust gas fluoride concentration, Gscaling(k) is the scaling rate; w1, w2, w3, w4 are weight coefficients used to adjust the importance of different objectives; by adjusting these weights, the relationship between concentration optimization, emission compliance, energy consumption control and scaling inhibition can be balanced.

[0030] Preferably, in order to ensure process safety and long-term operation of the equipment, multiple constraints are set in the control optimization model of multi-objective optimization, including: Exhaust emission constraints: ; The concentration of fluoride in tail gas shall not exceed the national standard; Scaling inhibition constraints: ; Where Gmax is the upper limit of the structural rate; monitor the flow rate, temperature and pH value of the fluorosilicic acid sealing tank, circulation pipeline and nozzle, and predict the scaling risk based on the fluorine absorption prediction model. When the scaling rate does not reach or exceed the critical value; Energy consumption control constraints: ; When the fluorosilicic acid concentration reaches the target, the energy consumption in the cooling process is minimized.

[0031] Preferably, in step S4, the method of optimizing the control variables by using the multi-objective optimization control optimization model to obtain the optimal control variable value is: S401, based on the MPC algorithm, the control variables are optimized. First, the state variables and control variables of the MPC algorithm are set: The state variables include: X1(k): Fluorosilicic acid concentration (g / cm³), real-time measurement by sensor integration in the sealing groove; X2(k): temperature inside the absorption tower (°C), obtained through sensor integration; X3(k): fluoride concentration in tail gas (mg / m³), monitored by online gas analyzer; X4(k): scaling rate (g / h), generated by flow rate, temperature and fluorine absorption prediction model; Control variables: u1(k): Take-out flow rate (m³ / h), controls the discharge amount of fluorosilicic acid in the sealing groove; u2(k): Chilled water flow rate (m³ / h), used to adjust the temperature in the absorption tower to improve absorption efficiency; u3(k): circulating liquid flow rate (m³ / h), controls the gas-liquid contact time and absorption efficiency; S402, construct the state equation of the MPC algorithm: Fluorosilicic acid concentration dynamic equation: ; In the formula, Q in Indicates the liquid flow rate entering the sealing groove (m³ / h); C in Indicates the concentration of fluorosilicic acid entering the liquid; V indicates the effective volume of the sealing groove; Fluorine absorption tower temperature equation: ; In the formula, C p represents the specific heat capacity of the absorption liquid (kJ / (kg·℃)); Q heat Indicates the heat released by the gas in the absorption tower; Fluoride concentration equation of tail gas: ; In the formula, C F,in Indicates the fluoride concentration in the gas entering the absorption tower; K abs It represents the gas-liquid mass transfer coefficient, which is related to temperature and liquid flow rate factors; Scaling rate equation: ; Wherein, function f represents the effect of temperature, fluorosilicic acid concentration and pH value on the silica gel formation rate.

[0032] Preferably, based on the state equation of the constructed MPC algorithm, multi-objective optimization is performed through the objective function of the control optimization model of multi-objective optimization to maximize the fluorosilicic acid concentration, minimize the tail gas fluoride concentration and scaling rate, and reduce energy consumption. In the multi-objective function, w1 is used to increase the weight of the fluorosilicic acid concentration, w2 is used to ensure that the fluoride concentration in the tail gas meets the national emission standards, w3 is used to control the use of chilled water to reduce energy consumption, and w4 is used to limit the scaling rate and reduce the frequency of equipment maintenance.

[0033] Preferably, in step S5, mobilizing the flow control valve to perform control and regulation includes adjusting control variables, and the control variables include the extraction flow u1(k), the chilled water flow u2(k) and the circulating liquid flow u3(k).

[0034] Preferably, in step S4, the real-time data collected by the data collection module is filtered in advance by an adaptive Kalman filter to filter out noise.

[0035] Embodiment 2: Fluorosilicic acid concentration optimization: Experimental conditions: In the wet phosphoric acid production line, the intelligent optimization control system is deployed in the first fluorine absorber and the second fluorine absorber of the fluorine absorption tower to monitor the fluorosilicic acid concentration in the sealing tank in real time and automatically adjust the withdrawal flow and circulating liquid flow.

[0036] Experimental process: The system calculates the optimal removal time based on the real-time monitored fluorosilicic acid density data. When the density reaches 1.18 g / cm³ (corresponding to the target fluorosilicic acid concentration), the system automatically delays the removal time to maximize the fluorosilicic acid concentration.

[0037] During the experiment, the concentration and amount of fluorosilicic acid removed were recorded each time. Experimental data showed that under system control, the concentration of fluorosilicic acid was stable within the target range (1.17-1.18 g / cm³), which was significantly higher than the fluctuation range of traditional manual control (1.11-1.16 g / cm³).

[0038] The effect data are shown in Table 1 below:

[0039] Under system control, the concentration of fluorosilicic acid increased by an average of about 2.5%, achieving the optimization target (1%-3%).

[0040] Through automated adjustment, the system achieves precise control of the take-out time, effectively reducing operational interference and concentration fluctuations.

[0041] Embodiment three: Scaling Inhibition: Experimental conditions: In the fluorine absorption circulation pipeline and nozzle, the system integrates flow rate sensors and temperature sensors to monitor the internal temperature and liquid flow rate of the pipeline in real time. At the same time, combined with the scaling prediction model, the chilled water flow and circulating liquid flow are automatically adjusted to reduce the generation of silica gel.

[0042] Experimental process: When the system detects that the temperature is too low and may cause silica gel to form, it automatically reduces the chilled water flow rate to keep the temperature within the critical range (50-55°C). At the same time, according to the changes in flow rate and silica gel formation rate, the system intelligently adjusts the speed of the circulation pump to ensure that the liquid maintains an appropriate flow rate (1.5-2.0 m / s) in the pipeline.

[0043] In actual operation, the system predicts the generation of silica gel in the pipeline every hour. When it is found that the accumulation of silica gel reaches a critical value, the system automatically starts the cleaning program to reduce the scaling rate.

[0044] The effect data are shown in Table 2 below:

[0045] Under the regulation of the intelligent optimization control system, the silicone generation rate in the pipes and nozzles was reduced by about 35%, and the equipment cleaning cycle was extended from an average of 3 hours / time to 4.5 hours / time, greatly extending the continuous operation time of the equipment.

[0046] The system's real-time monitoring and automatic adjustment reduce the workload of operators and improve the stability of the production environment.

[0047] Embodiment 4: Energy consumption optimization: Experimental conditions: High-precision temperature sensors and flow meters are deployed in the fluorine absorption system to monitor the consumption of chilled water and circulating fluid in real time. The use of chilled water is optimized through the MPC algorithm to reduce unnecessary cooling operations.

[0048] Experimental process: When the system finds that the fluoride concentration is close to the upper limit of the national standard during exhaust gas detection, it calculates the balance point between the current fluorosilicic acid concentration and the cooling water flow rate to reduce unnecessary chilled water use.

[0049] When the system detects that the temperature has reached the optimal range for fluoride absorption efficiency (50-55°C), it automatically reduces the supply of chilled water to ensure fluoride absorption efficiency while minimizing energy consumption.

[0050] The effect data are shown in Table 3 below:

[0051] The experimental results show that after the implementation of the intelligent optimization control system, the use of chilled water was reduced by about 20% and the monthly energy consumption was saved by about 15%.

[0052] While reducing energy consumption, the system ensures that fluorosilicic acid concentration and tail gas emissions meet standards, and fluorine absorption efficiency is not negatively affected.

[0053] Embodiment five: Fault prediction and maintenance: Experimental conditions: The system was run on an actual production line for 6 months, monitoring all key parameters (including temperature, flow rate, pressure and fluosilicic acid concentration), while a fault prediction model was established using historical data.

[0054] Experimental process: The system monitors flow and temperature in real time to predict potential equipment failures. When an abnormal drop in flow rate or an abnormal rise in temperature is detected, the system calculates the risk of failure through a model and issues an early warning message.

[0055] When the system predicts that the scaling rate of a certain pipeline has reached a critical value, it will issue a cleaning and maintenance instruction in advance, and the operator can perform regular maintenance according to the system prompts.

[0056] The system successfully predicted four scaling events and two equipment failures within six months, allowing maintenance to be scheduled in advance and downtime to be avoided, saving a total of approximately 120 hours of production time.

[0057] Through intelligent prediction and maintenance optimization, the average service life of equipment has been extended by about 30%.

[0058] Fluorosilicic acid concentration: Under the automatic control of the system, the concentration increased by 1%-3%, achieving the expected target.

[0059] Scaling inhibition: effectively reduce the scaling rate, reduce the frequency of equipment cleaning, and extend the continuous operation time of the equipment.

[0060] Energy-saving effect: Under intelligent optimization control, the system energy consumption is reduced by 15%-20%, which greatly saves costs.

[0061] Fault prediction: Combined with the predictive model of machine learning, the system effectively prevents equipment failures and improves the stability and continuity of the production line.

[0062] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. Intelligent optimization automatic control system for wet phosphoric acid fluoride absorption process, characterized by: It includes a data acquisition module, which is connected to the prediction model building module through the data transmission module; the prediction model building module is connected to the control optimization module through the data transmission module, and the control optimization module is electrically connected to the control execution module; The data acquisition module includes a sensor integration, which includes a high-precision density meter, an online gas analyzer, a flow meter, a temperature sensor and a pH sensor; the high-precision density meter is used to collect the real-time fluorosilicic acid concentration and scaling rate in the fluorine absorption process, the online gas analyzer is used to collect the tail gas fluoride concentration, the flow meter is used to collect the circulating liquid concentration, the temperature sensor is used to collect the temperature in the absorption tower, and the pH sensor is used to collect the pH value of the circulating liquid; the data acquisition module also includes a database interface, which is connected to the historical database of the wet-process phosphoric acid fluoride absorption central control system for collecting historical data; The prediction model building module is used to build a fluorine absorption prediction model based on random forests, providing data changes in the fluorine absorption process in the future period; The control optimization module is used to construct a control optimization model based on multi-objective optimization. The control optimization model provides optimal set values ​​for the control variables of the control execution module based on the MPC algorithm. The control execution module includes a plurality of flow control valves in the wet process phosphoric acid fluoride absorption control system, which are used to control the withdrawal flow, the chilled water flow and the circulating liquid flow; The data transmission module includes Ethernet, Bluetooth transceiver module and WiFi communication module.

2. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 1, characterized in that: include: S1, the data acquisition module collects historical data and real-time data of the wet-process phosphoric acid fluoride absorption process to build a model sample data set; S2, the prediction model building module builds a fluorine absorption prediction model based on the model sample data set; S3, predicting the fluorosilicic acid concentration and scaling rate according to the real-time input data of the current wet phosphoric acid fluoride absorption process through the constructed fluorine absorption prediction model; S4, based on the prediction result obtained in step S3 and the real-time data collected by the data acquisition module, the control variables are optimized through the multi-objective optimization control optimization model to obtain the optimal control variable value; S5, the control execution module receives the optimal control variable value and mobilizes the flow control valve to perform control and adjustment, so that the wet phosphoric acid fluoride absorption system reaches the optimal operating condition.

3. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 2, characterized in that: In step S1, the method for constructing the model sample data set is: S101, construct variable set x Absorption and x Scalability ;x Absorption Represents the main variables in the fluorine absorption process, including the temperature T in the absorption tower, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the fluorosilicic acid concentration C H ;x Scalability represents the key variables that affect the scaling rate, including the absorber temperature T, the pH value of the circulating liquid pH, the circulating liquid flow rate v and the scaling rate G scale ; Each set of variables is represented as follows: x Absorption =[T,pH,v,C H ]; x Scalability =[T,pH,v,G scale ]; S102, dividing the multiple arrays in the two constructed variable sets into labeled data sets and unlabeled data sets: Labeled dataset: Extract key features from the above variables and construct a labeled dataset D labeled ={(x labeled ,y labeled )}, where x labeled A characteristic combination of temperature, pH, flow rate, fluorosilicic acid concentration and scaling rate, y labeled Indicates the actual fluorosilicic acid concentration or scaling rate; Unlabeled dataset: Unlabeled samples of fluorine absorption process collected by sensor integration x unlabeled ⊂x Process , used to generate pseudo-labeled data to improve the prediction accuracy of the model.

4. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 3, characterized in that: In step S2, the method for constructing a fluorine absorption prediction model is: A fluorine absorption prediction model was constructed based on the semi-supervised random forest algorithm, and pseudo labels of unlabeled data were used to improve the prediction accuracy: S201, model parameter setting: Initialize the number of decision trees K to 50-100 and optimize through cross-validation to optimize the stability of prediction; Initialize the maximum tree depth to 4-7 layers to ensure generalization ability and avoid overfitting; Initialize the minimum number of split samples: set the number of samples for each leaf node to 10-20 to prevent overfitting; S202, model training and optimization: Recursive segmentation process: For each tree node, use information gain or mean square error MSE to select the optimal segmentation features and segmentation points: ； Where, R1 and R2: represent two different regions or partitions of the data set, generated in the split node of the decision tree; |R1| and |R2|: represent the number of samples in regions R1 and R2, used to normalize the error term; and : represents the average predicted value in regions R1 and R2, which is used to calculate the error; y: represents the actual label value, which is used as the true value when calculating the mean square error or information gain; Feature selection and randomness: Each layer of segmentation selects only some features including temperature, flow rate and pH value to enhance model diversity; Set tree growth stop conditions: Each tree stops growing when any of the following conditions is met: a. Reach maximum depth; b. The number of leaf node samples is less than the set threshold; c. Information gain is lower than the threshold; Pseudo-label generation and model updating; Pseudo-label generation: Use the trained random forest model to predict unlabeled data x unlabeled Pseudo labels: ； In the formula, x unlabeled : represents an unlabeled data sample, which has no real label and is processed using a pseudo-labeling method; pseudo : represents pseudo-labeled values, which are used for prediction of unlabeled data; prediction labels are generated for unlabeled data through integrated models (such as random forests) to enhance the training effect of the model; : Represents the prediction function of the kth random forest decision tree, which is used for the unlabeled sample x unlabeled Generate forecast values; Screening effective pseudo-labels: Screening pseudo-label samples with high confidence based on the root mean square error (RMSE), updating the labeled data set, and forming a new training set D new−train ; S203, integrate the results of all decision trees to form the final prediction output: Based on the weighted average method, the output of each tree is weighted based on its prediction accuracy, and the weighted average is obtained as the model output: ； K: represents the number of decision trees. In ensemble learning algorithms such as random forest, the prediction results of K decision trees will be integrated; : represents the final prediction output, which is usually the weighted or average result of multiple model prediction results, used to improve the stability of the overall prediction; w k : represents the weight of the kth decision tree. The weight is usually assigned according to the prediction accuracy of the decision tree to optimize the accuracy of the final prediction; : represents the prediction output of the kth decision tree, that is, the prediction value of a single tree for a sample; Based on the mean method, the predicted values ​​of all decision trees are averaged: ； S204, model performance optimization: (1) Hyperparameter optimization: adjust the number of decision trees K, tree depth, and minimum number of samples through grid search or cross-validation to ensure model stability and accuracy; (2) Segmentation criterion optimization: adjust the information gain and MSE indicators to improve the accuracy of segmentation point selection, optimize the temperature, flow rate and pH value, and enhance the accuracy of capturing key variables; (3) Randomness optimization: The random subspace method RSM is introduced to select only some features in each layer to improve the diversity and generalization ability of the model; S205, based on the optimization of the deep forest regression module DFR: After generating pseudo-labels and updating the dataset, a deep forest regression model was used to make refined predictions of fluorosilicic acid concentration and scaling rate; The DFR module consists of an input layer, an intermediate layer, and an output layer, and improves the prediction effect by recursively fusing features layer by layer; (1) The input layer forest module performs sampling and sub-forest generation: new−train Perform Bootstrap and RSM sampling to generate I sub-forests; Prediction mean calculation: prediction mean for each sub-forest Weighted, select the predicted value closest to the mean to form a regression vector; (2) The middle-layer forest module processes the output of the input layer layer by layer to form a recursive regression vector. The output of each layer is used as the input of the next layer, and the prediction accuracy is enhanced recursively layer by layer. (3) The output layer forest module receives the regression vector output by the intermediate layer and makes predictions through multiple decision trees. The final result is the mean of the outputs of all sub-forests. In the formula, : represents the final prediction output of the deep forest regression (DFR) module; this value is obtained by averaging the prediction results of all sub-forests; I: represents the number of sub-forests, that is, the total number of sub-forests generated at the input layer of the DFR module; Represents the predicted mean of the iiith sub-forest; each sub-forest makes independent predictions on the data and then generates a mean as the prediction result of the sub-forest.

5. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 4, characterized in that: In step S4, the control optimization model construction method based on multi-objective optimization is: With increasing the concentration of fluorosilicic acid as the main goal, the exhaust emissions, scaling rate and energy consumption factors were comprehensively considered during the optimization process, and the objective function was constructed as follows: ； Among them, x1(k) is the concentration of fluorosilicic acid, x3(k) is the exhaust gas fluoride concentration, Gscaling(k) is the scaling rate; w1, w2, w3, w4 are weight coefficients used to adjust the importance of different objectives; by adjusting these weights, the relationship between concentration optimization, emission compliance, energy consumption control and scaling inhibition can be balanced.

6. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 5, characterized in that: To ensure process safety and long-term operation of equipment, multiple constraints are set in the control optimization model of multi-objective optimization, including: Exhaust emission constraints: ； The concentration of fluoride in tail gas shall not exceed the national standard; Scaling inhibition constraints: ； Where Gmax is the upper limit of the structural rate; monitor the flow rate, temperature and pH value of the fluorosilicic acid sealing tank, circulation pipeline and nozzle, and predict the scaling risk based on the fluorine absorption prediction model. When the scaling rate does not reach or exceed the critical value; Energy consumption control constraints: ； When the fluorosilicic acid concentration reaches the target, the energy consumption in the cooling process is minimized.

7. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 6, characterized in that: In step S4, the control variables are optimized by the multi-objective optimization control optimization model, and the method for obtaining the optimal control variable value is: S401, based on the MPC algorithm, the control variables are optimized. First, the state variables and control variables of the MPC algorithm are set: The state variables include: X1(k): Fluorosilicic acid concentration (g / cm³), real-time measurement by sensor integration in the sealing groove; X2(k): temperature inside the absorption tower (°C), obtained through sensor integration; X3(k): fluoride concentration in tail gas (mg / m³), monitored by online gas analyzer; X4(k): scaling rate (g / h), generated by flow rate, temperature and fluorine absorption prediction model; Control variables: u1(k): Take-out flow rate (m³ / h), controls the discharge amount of fluorosilicic acid in the sealing groove; u2(k): Chilled water flow rate (m³ / h), used to adjust the temperature in the absorption tower to improve absorption efficiency; u3(k): circulating liquid flow rate (m³ / h), controls the gas-liquid contact time and absorption efficiency; S402, construct the state equation of the MPC algorithm: Fluorosilicic acid concentration dynamic equation: ； In the formula, Q in Indicates the liquid flow rate entering the sealing groove (m³ / h); C in Indicates the concentration of fluorosilicic acid entering the liquid; V indicates the effective volume of the sealing groove; Fluorine absorption tower temperature equation: ； In the formula, C p represents the specific heat capacity of the absorption liquid (kJ / (kg·℃)); Q heat Indicates the heat released by the gas in the absorption tower; Fluoride concentration equation of tail gas: ； In the formula, C F,in Indicates the fluoride concentration in the gas entering the absorption tower; K abs It represents the gas-liquid mass transfer coefficient, which is related to temperature and liquid flow rate factors; Scaling rate equation: ； Wherein, function f represents the effect of temperature, fluorosilicic acid concentration and pH value on the silica gel formation rate.

8. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 7, characterized in that: Based on the state equation of the constructed MPC algorithm, multi-objective optimization is performed through the objective function of the control optimization model of multi-objective optimization to maximize the fluorosilicic acid concentration, minimize the exhaust gas fluoride concentration and scaling rate, and reduce energy consumption. In the multi-objective function, w1 is used to increase the weight of the fluorosilicic acid concentration, w2 is used to ensure that the fluoride concentration in the exhaust gas meets the national emission standards, w3 is used to control the use of chilled water to reduce energy consumption, and w4 is used to limit the scaling rate and reduce the frequency of equipment maintenance.

9. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 8, characterized in that: In step S5, the flow control valve is actuated to perform control and regulation, including adjusting control variables, wherein the control variables include the extraction flow u1(k), the chilled water flow u2(k) and the circulating liquid flow u3(k).

10. The method for intelligent optimization automatic control system of wet process phosphoric acid fluoride absorption process according to claim 9, characterized in that: In step S4, the real-time data collected by the data collection module is filtered in advance by an adaptive Kalman filter to remove noise.

Images