Phosphogypsum curing treatment method based on machine learning regulation

Abstract

The invention discloses a phosphogypsum curing treatment method based on machine learning regulation, which comprises the following steps: adding calcium oxide, an organic dispersant and water into phosphogypsum to realize phosphogypsum curing, wherein the addition amount of the calcium oxide, the addition amount of the organic dispersing agent and the addition amount of the water are determined through the following processes: collecting parameters of a phosphogypsum leaching solution sample, including soluble phosphorus concentration, soluble fluorine concentration, pH and water content; the parameters of the phosphogypsum leachate sample are input into the machine learning model of the phosphogypsum parameters and the adding parameters, the adding parameters are output, and the adding parameters comprise the calcium oxide adding amount, the organic dispersing agent adding amount and the water adding amount. Through a machine learning model, the addition amount of calcium oxide, the addition amount of an organic dispersant and the addition amount of water can be rapidly adjusted according to specific components of different raw materials, and the adaptability is higher. According to the method, the treatment effect is improved, the experiment frequency and the human input are reduced, and excessive use of the medicament is avoided through an accurate medicament formula, so that the medicament cost is saved.

Description

Technical Field

[0001] The present invention belongs to the technical field of environmental protection and resource utilization, and specifically relates to a phosphogypsum solidification treatment method, in particular a phosphogypsum solidification treatment method based on machine learning regulation. Background Art

[0002] Phosphogypsum is a byproduct of phosphoric acid production. Its main component is calcium sulfate dihydrate (CaSO4·2H2O), and it also contains a small amount of soluble phosphorus, soluble fluorine and other impurities. Since phosphoric acid production uses a large amount of phosphate ore and sulfuric acid, a large amount of phosphogypsum byproduct is produced during the reaction. According to statistics, about 4.5 tons of phosphogypsum are produced for every ton of phosphoric acid produced. At present, the total amount of phosphogypsum produced in the world each year has exceeded 300 million tons. How to handle and utilize this phosphogypsum has become a major challenge for environmental protection and resource utilization.

[0003] Improper stacking and handling of phosphogypsum can lead to a series of environmental problems. First, substances such as soluble phosphorus and soluble fluorine contained in phosphogypsum are easily dissolved, polluting the soil and water bodies and causing damage to the ecological environment. Secondly, the stacking of phosphogypsum occupies a large amount of land resources, increasing the pressure on land resources. In addition, phosphogypsum will produce dust pollution during the stacking process, affecting the health of surrounding residents. Therefore, how to effectively solidify and utilize phosphogypsum is not only related to environmental protection issues, but also to the rational use of resources.

[0004] The main purpose of phosphogypsum solidification treatment is to convert it into a stable, insoluble form to prevent the release of harmful substances. The traditional phosphogypsum solidification treatment method usually uses calcium oxide (CaO) as the main solidifying agent. However, when ordinary calcium oxide reacts with phosphogypsum, it will produce products such as monocalcium phosphate (CaHPO4) and calcium fluoride (CaF2). -7 ) and calcium fluoride (CaF2, Ksp is about 10 -11 ), these products have high solubility, which leads to the easy release of phosphorus and fluoride ions, exceeding environmental protection standards. In addition, the formation of calcium monohydrogen phosphate will cause the pH value of the curing system to fluctuate greatly. Too high or too low pH values ​​will affect the curing effect and may cause secondary pollution to the environment.

[0005] Another challenge of solidification treatment is that the composition of different batches of phosphogypsum raw materials varies greatly, resulting in the need to constantly adjust the dosage ratio of the reagent. This adjustment usually requires a large number of experiments to determine, which is time-consuming and labor-intensive, and is not conducive to large-scale industrial application. Therefore, finding an efficient, stable and adaptable solidification treatment method for different raw material compositions is the focus of current research.

[0006] In recent years, with the development of computer technology, machine learning has been widely used in the field of environmental protection. By analyzing and modeling a large amount of data, machine learning can provide accurate predictions and optimization solutions, greatly improving treatment efficiency and effectiveness. In solid waste treatment, machine learning technology has been applied to waste classification, waste generation prediction, and waste treatment process optimization.

[0007] Specifically, machine learning can analyze historical data, establish waste treatment models, and predict the best formulations for different raw material compositions. For example, regression analysis models can accurately predict the dosage ratio of agents, thereby reducing the number of experiments and costs. In addition, machine learning can also monitor the waste treatment process in real time, adjust treatment parameters according to actual conditions, and improve treatment effects. In the fields of wastewater treatment and air pollution monitoring, machine learning technology has also demonstrated its strong application potential.

[0008] The traditional method relies on a large number of experiments to determine the proportion of the reagents, which is not only time-consuming and labor-intensive, but also difficult to ensure the consistency of each treatment. At the same time, a large amount of engineering data has not been effectively statistically analyzed, which increases the workload of experimenters and project managers. Summary of the invention

[0009] In order to overcome the time-consuming and labor-intensive technical problems existing in the solidification treatment of phosphogypsum in the above-mentioned prior art, the object of the present invention is to provide a phosphogypsum solidification treatment method based on machine learning regulation.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions: A method for solidifying phosphogypsum based on machine learning regulation comprises the following steps: adding calcium oxide, an organic dispersant and water to the phosphogypsum to achieve solidification of the phosphogypsum; Among them, the amount of calcium oxide added, the amount of organic dispersant added and the amount of water added are determined by the following process: Parameters for collecting samples of phosphogypsum leachate; Input the parameters of the phosphogypsum leachate sample into the machine learning model of the parameters of the phosphogypsum and the dosing parameters, and output the dosing parameters, which include the amount of calcium oxide added, the amount of organic dispersant added, and the amount of water added; The fitting model of phosphogypsum parameters and dosing parameters is obtained through machine learning model training.

[0011] Furthermore, the fitting model of raw material parameters and dosing parameters is obtained through the following process: Parameters for collecting samples of phosphogypsum leachate; The parameters of the phosphogypsum leachate samples were processed and divided into a training set and a test set; The machine learning model is used for data training to establish a fitting model of phosphogypsum parameters and dosing parameters; the output of the fitting model is the dosing parameters.

[0012] Furthermore, the parameters of the phosphogypsum leachate samples include soluble phosphorus, soluble fluorine, pH value and water content.

[0013] Furthermore, the parameters of the phosphogypsum leachate samples are processed including data standardization and normalization.

[0014] Furthermore, the parameters of the phosphogypsum leachate samples were processed and divided into training sets and test sets using the sklearn dataset random partitioning technique.

[0015] Furthermore, a machine learning model is used for data training to establish a fitting model of phosphogypsum parameters and dosing parameters, including the following steps: The machine learning model is used for data training and prediction, and the model with the highest prediction accuracy is used to establish a fitting model for the parameters and dosage parameters of phosphogypsum.

[0016] Furthermore, the machine learning model is a linear regression model, a polynomial regression model, a support vector machine regression model, a decision tree regression model, a random forest regression model or an XGBoost regression model.

[0017] Furthermore, the parameters of the phosphogypsum leachate sample are input into a machine learning model of the phosphogypsum parameters and dosing parameters. After the dosing parameters are output, a correction step is also included.

[0018] Furthermore, among the parameters of the phosphogypsum leachate samples, the soluble phosphorus concentration is 50~1550 mg / L, the soluble fluorine concentration is 50~1000 mg / L, the pH is less than 5, and the water content is 4~28%.

[0019] Furthermore, calcium oxide, organic dispersant and water were added to the phosphogypsum, and after 3 days, the pH value was raised to 6-9 and the concentrations of soluble phosphorus and soluble fluorine were lower than 0.5 mg / L and 10 mg / L, respectively. After 6 months, the concentrations of soluble phosphorus and soluble fluorine were lower than 0.5 mg / L and 10 mg / L, respectively.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention can quickly adjust the agent formula, that is, the amount of calcium oxide added, the amount of organic dispersant added and the amount of water added according to the specific components of different batches of raw materials through a machine learning model, and has stronger adaptability. This method not only improves the treatment effect, reduces the number of experiments and manpower input, but also avoids excessive use of agents through accurate agent formulas, thereby saving agent costs. It has significant economic benefits in improving curing efficiency and effect, stabilizing and controlling pollutant indicators, adapting to different raw material components, saving costs, protecting the environment, etc., and has high application value. In the present invention, while effectively curing phosphogypsum, the calcium oxide in the agent and the soluble phosphorus and fluorine in the phosphogypsum generate fluorophosphate calcium (Ca5(PO4)3F), whose solubility product constant (Ksp) is extremely small, can effectively reduce the release concentration of phosphorus and fluoride ions, and stably control the pH value. The experimental verification results show that the formula optimized by this method can ensure that the three indicators of phosphorus, fluorine and pH are all stable within the standard range, and the prediction error is less than 0.2%, which overcomes the problems of excessive phosphorus and fluoride ion concentrations and unstable pH values ​​in traditional methods.

[0021] Furthermore, the present invention uses machine learning technology, combined with a large amount of on-site data and manual correction, to optimize the formula of phosphogypsum curing agent. This method not only improves the treatment efficiency and effect, but also ensures the accuracy and stability of the agent addition ratio, with significant economic and environmental benefits. By using machine learning technology, an accurate fitting model of phosphogypsum parameters and addition parameters is established, which can greatly reduce the number of experiments, saving time and cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The training process of building a classifier model based on MLP (multi-layer perceptron) for Example 1; wherein (a) is a training loss curve, and (b) is a verification loss curve; Figure 2 This is a real photo of the small-scale experimental verification of the phosphogypsum solidification in Example 1; Figure 3 The diagram shows the prediction process using correlation coefficient measurement; the blue curve is the predicted value of the model, and the red curve is the actual measured value; (a) is the result of using the random forest regression method, and (b) is the result of using the XGBoost regression method.

[0023] Figure 4 XRD and SEM characterization test diagrams of calcium, phosphorus and fluorine precipitation in the solidified phosphogypsum, wherein (a) is XRD, (b) is SEM after ordinary CaO solidification, and (c) is SEM after solidification by the method of the present invention. DETAILED DESCRIPTION

[0024] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] The present invention is further described in detail below with reference to the accompanying drawings.

[0027] A phosphogypsum solidification treatment method based on machine learning regulation of the present invention comprises the following steps: (1) Collect a large amount of field data from the same phosphoric acid production enterprise in the same region, extract the parameters of phosphogypsum, including soluble phosphorus, soluble fluorine, pH and moisture content, and collect the addition parameters, including the amount of calcium oxide added, the amount of organic dispersant added, and the amount of pure water added. Perform data standardization and normalization steps on the large amount of collected field data to ensure data consistency. Use the sklearn dataset random partitioning technology to divide the processed data into a training set and a test set at a ratio of 8:2.

[0028] (2) Use a variety of machine learning models (linear regression model, polynomial regression model, support vector machine regression model, decision tree regression model, random forest regression model and XGBoost regression model, etc.) to train and predict data, use the model with the highest prediction accuracy to establish a fitting model for the parameters and dosing parameters of phosphogypsum, and use the fitting model to optimize the agent formula (i.e., optimize the dosing parameters). The optimized agent formula can promote the formation of calcium fluorophosphate during the solidification process of phosphogypsum, ensuring that the solubility product constant of the product (Ksp is about 10 -60 ) is extremely small, effectively controlling the release concentration of phosphorus and fluoride ions and stabilizing the pH value.

[0029] The experiment verifies the effect of the optimized formula, ensuring that the generated products are mainly hydroxy calcium phosphate and fluoro calcium phosphate, and that the three indicators of phosphorus, fluorine and pH are all stable within the standard range.

[0030] During the experimental verification process, the prediction error of the agent dosage ratio was less than 0.2%.

[0031] (3) Manual corrections are made to a small number of data with biased predicted values ​​to optimize the drug formulation.

[0032] For the solidified phosphogypsum, calcium fluorophosphate can be observed in the precipitate through XRD and SEM characterization techniques.

[0033] Combining simulation and manual correction, the optimized formulation includes the addition of calcium oxide, organic dispersant and pure water. Specifically, on the basis of industrial calcium oxide powder (80-98% active ingredient), ester dispersants such as methyl formate and ethyl acetate are added, and the ratio of calcium oxide to dispersant is adjusted by machine learning, and water 5 to 8 times the mass of calcium oxide is added at the same time.

[0034] The pharmaceutical formulation obtained by the present invention can effectively solidify phosphogypsum, and make the calcium oxide in the pharmaceutical react with the soluble phosphorus and fluorine in the phosphogypsum to generate calcium fluorophosphate, whose Ksp is about 10 -60 .

[0035] For leachate with phosphate concentration of 150-1550 mg / L, fluoride concentration of 50-1000 mg / L and pH value lower than 5, by adding phosphogypsum solidifying agent (including calcium oxide, organic dispersant and pure water), the pH value was raised to about 8 within 3 days and the concentrations of soluble phosphorus and soluble fluorine in the treated phosphogypsum leachate were lower than 0.5 mg / L and 10 mg / L respectively, which met the national standard "Integrated Wastewater Discharge Standard" (GB 8978-1996). After 6 months of exposure to air, the concentrations of soluble phosphorus and soluble fluorine were still lower than 0.5 mg / L and 10 mg / L respectively.

[0036] The following are specific embodiments.

[0037] Example 1 2500 sets of phosphogypsum data were collected from phosphoric acid production enterprises in a certain region, covering parameters such as soluble phosphorus (30-450 mg / L), soluble fluorine (150-350 mg / L), pH value (2.5-4.5) and moisture content (4-28%). The phosphogypsum data were standardized and normalized to ensure data consistency and comparability. Forty sets of phosphogypsum data are shown in Table 1.

[0038] Table 140 Group Phosphogypsum Data

[0039] Note: In Table 1, the unit of phosphorus concentration is mg / L, the unit of fluorine concentration is mg / L, the water content is the water content in %, the modified additive 1 is CaO, and the modified additive 2 is an organic dispersant.

[0040] Using the sklearn dataset random partitioning technique, the data is divided into a training set and a test set at a ratio of 8:2.

[0041] The linear regression model, polynomial regression model, support vector machine regression model, decision tree regression model, random forest regression model and XGBoost regression model were all trained using the training set, and the best model was selected through cross-validation.

[0042] See also Figure 1 , build a model based on the neural network structure, and perform the training process to obtain the training loss and validation loss curves based on the MLP (multi-layer perceptron) classifier model. It can be seen that the training loss curve ( Figure 1 (a) in the middle has large fluctuations in the early stage, but the overall trend is that the loss gradually decreases, indicating that the model is constantly learning and optimizing. Verification loss curve ( Figure 1 (b) shows a rapid decrease in the early stage and then gradually stabilizes, indicating that the model performs relatively stably on the validation set and has good generalization ability.

[0043] The established model uses the Integrated Wastewater Discharge Standard (GB 8978-1996) as the evaluation standard, combined with the corresponding phosphogypsum raw material data and the reagent formula, and adjusts the parameters to make the predicted results meet the target expectations: the phosphorus and fluorine concentrations in the phosphogypsum leachate are controlled below 0.5 mg / L and 10 mg / L respectively, and the pH value is maintained between 6 and 9. According to the model prediction results and actual application requirements, the following reagent formula is designed: Calcium oxide: The effective ingredient is 90%, and the added amount is 2.4% of the mass of phosphogypsum; Organic dispersant: ethyl acetate, the addition amount is 0.7% of the mass of phosphogypsum; Pure water: the added amount is 15% of the mass of phosphogypsum; The above formula was applied to the phosphogypsum leachate sample for solidification treatment. The treatment results in Table 2 show that the soluble phosphorus concentration in the solidified phosphogypsum was reduced to 0.23 mg / L, the soluble fluorine concentration was reduced to 6.87 mg / L, and the pH value was stabilized at 7.8. After multiple experimental verifications, the optimized formula has excellent solidification effect, and the three indicators of phosphorus, fluorine and pH are all stable within the standard range, with a prediction error of less than 0.2%, see Table 1. At the same time, the pharmaceutical formula was compared with the phosphogypsum solidified by ordinary CaO. The morphology of soluble phosphorus and fluorine in CaO and phosphogypsum was detected by SEM and XRD characterization, and it was found that the pharmaceutical formula generated calcium fluorophosphate.

[0044] Table 2 Parameter values ​​before and after treatment

[0045] According to GB 8978-1996 Integrated Wastewater Discharge Standard and GB 18599-2020 General Industrial Solid Waste Storage and Landfill Pollution Control Standard, the standards of soluble phosphorus below 0.5 mg / L, soluble fluoride below 10 mg / L, and pH between 6 and 9 were met. After one month, the soluble phosphorus test was below the detection limit, the soluble fluoride test value was 5.38 mg / L, and the pH value was stable between 7.3 and 7.5.

[0046] See also Figure 2 It can be seen that the surface of the phosphogypsum after curing with the agent formula is more uniform and the particles are fine, indicating that the predicted agent formula has a good curing effect on the phosphogypsum. Compared with the ordinary CaO curing product, the structure is denser, which is conducive to improving the curing effect.

[0047] See also Figure 3 , it can be seen that Figure 3 (a) and (b) show the comparison curves of measured and predicted values, indicating that the model can better fit the changing trend of phosphogypsum treatment process. The correlation between the predicted and measured values ​​is high, indicating that the regression model used has a good performance in processing data.

[0048] See also Figure 4 , it can be seen that Figure 4 The XRD spectrum in (a) shows the changes in the crystal form of the product under different curing conditions, among which the formation of calcium fluorophosphate after the pharmaceutical formula is cured is obvious, indicating that the predicted pharmaceutical formula can effectively promote the formation of precipitation and stabilize the release of phosphorus and fluorine. Figure 4 The SEM images in (b) and (c) show that the predicted microstructure of the pharmaceutical formulation after curing is more regular.

[0049] Example 2 Multiple samples of phosphogypsum leachate were collected from a phosphoric acid production enterprise, totaling 2000 sets of data. Each set of data includes raw material parameters such as soluble phosphorus (20-400 mg / kg), soluble fluorine (100-7000 mg / kg), pH value (5-8) and moisture content (5-25%). The data were standardized and normalized to ensure the consistency and comparability of the data. Using the sklearn dataset random partitioning technique, the processed data was divided into a training set and a test set at a ratio of 8:2.

[0050] The linear regression model, polynomial regression model, support vector machine regression model, decision tree regression model, random forest regression model and XGBoost regression model were trained through the training set, and the best model was selected through cross-validation. According to the model prediction results, combined with manual correction, the specific ratio of calcium oxide and organic dispersant was determined. In the experiment, the following formula was selected: Calcium oxide: The active ingredient is 93%, and the added amount is 1.5% of the mass of phosphogypsum; Organic dispersant: methyl formate, added in an amount of 0.5% of the mass of phosphogypsum Pure water: The amount added is 10.8% of the mass of phosphogypsum The above formula was applied to the phosphogypsum leachate sample for solidification treatment. The results showed that the soluble phosphorus concentration in the treated phosphogypsum was reduced to 0.3 mg / L, the soluble fluorine concentration was reduced to 7.8 mg / L, and the pH value was stabilized at 7.9. The small-scale experiment verified that the solidification effect of the optimized formula was significant, and the three indicators of phosphorus, fluorine and pH all met the environmental protection standards, and the prediction error was less than 0.2%.

[0051] The method of the present invention is also applicable to different phosphogypsum leachate samples.

[0052] The present invention embodies the application potential of machine learning technology in the field of environmental protection.

[0053] The above description is only for the best embodiment of the present invention, but it should not be understood as limiting the claims. The present invention is not limited to the above embodiments, and its specific structure is allowed to be changed. However, all changes made within the protection scope of the independent claims of the present invention are within the protection scope of the present invention.

[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

Claims

1. A method for solidifying phosphogypsum based on machine learning control, characterized in that: The following steps are involved: adding calcium oxide, an organic dispersant and water to the phosphogypsum to achieve solidification of the phosphogypsum; Among them, the amount of calcium oxide added, the amount of organic dispersant added and the amount of water added are determined by the following process: Parameters for collecting samples of phosphogypsum leachate; Input the parameters of the phosphogypsum leachate sample into the machine learning model of the parameters of the phosphogypsum and the dosing parameters, and output the dosing parameters, which include the amount of calcium oxide added, the amount of organic dispersant added, and the amount of water added; The fitting model of phosphogypsum parameters and dosing parameters is obtained through machine learning model training.

2. The method for solidifying phosphogypsum based on machine learning control according to claim 1, characterized in that: The fitting model of raw material parameters and dosing parameters is obtained through the following process: Parameters for collecting samples of phosphogypsum leachate; The parameters of the phosphogypsum leachate samples were processed and divided into a training set and a test set; The machine learning model is used for data training to establish a fitting model of phosphogypsum parameters and dosing parameters; the output of the fitting model is the dosing parameters.

3. The method for solidifying phosphogypsum based on machine learning control according to claim 2, characterized in that: The parameters of phosphogypsum leachate samples include soluble phosphorus, soluble fluoride, pH value and water content.

4. The method for solidifying phosphogypsum based on machine learning control according to claim 2, characterized in that: The parameters of phosphogypsum leachate samples were processed including data standardization and normalization.

5. The method for solidifying phosphogypsum based on machine learning control according to claim 4, characterized in that: After processing the parameters of the phosphogypsum leachate samples, the sklearn dataset random partitioning technique was used to divide them into training sets and test sets.

6. The method for solidifying phosphogypsum based on machine learning control according to claim 4, characterized in that: The machine learning model is used for data training to establish a fitting model of phosphogypsum parameters and dosing parameters, including the following steps: The machine learning model is used for data training and prediction, and the model with the highest prediction accuracy is used to establish a fitting model for the parameters and dosage parameters of phosphogypsum.

7. The method for solidifying phosphogypsum based on machine learning control according to claim 4, characterized in that: The machine learning model is a linear regression model, a polynomial regression model, a support vector machine regression model, a decision tree regression model, a random forest regression model, or an XGBoost regression model.

8. The method for solidifying phosphogypsum based on machine learning control according to claim 1, characterized in that: The parameters of the phosphogypsum leachate sample are input into the machine learning model of the phosphogypsum parameters and the dosing parameters. After the dosing parameters are output, a correction step is also included.

9. The method for solidifying phosphogypsum based on machine learning control according to claim 1, characterized in that: Among the parameters of the phosphogypsum leachate samples, the soluble phosphorus concentration is 50~1550 mg / L, the soluble fluorine concentration is 50~1000 mg / L, the pH is less than 5, and the water content is 4~28%.

10. The method for solidifying phosphogypsum based on machine learning control according to claim 1, characterized in that: Calcium oxide, organic dispersant and water were added to phosphogypsum. After 3 days, the pH value was raised to 6-9 and the concentrations of soluble phosphorus and soluble fluorine were lower than 0.5 mg / L and 10 mg / L respectively. After 6 months, the concentrations of soluble phosphorus and soluble fluorine were lower than 0.5 mg / L and 10 mg / L respectively.

Images