A method for two-stage prediction of performance of a sulfur cycle coupled denitrification phosphorus removal process
By constructing a dataset and scheduling platform using a two-stage machine learning approach, the instability problem of the sulfur cycle coupled with denitrification for phosphorus removal was solved. This enabled accurate prediction of sulfate reduction and phosphorus removal rates, improving the system's operational stability and reducing costs.
Patent Information
- Application Number
- CN202310765450.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-27
- Publication Date
- 2026-01-06
- Estimated Expiration
- 2043-06-27
AI Technical Summary
Existing technologies are insufficient for the stable optimization of sulfur cycle coupled with denitrification and phosphorus removal processes. Traditional models cannot accurately predict water quality and reaction effects, leading to unstable operation and increased carbon emissions and costs.
A two-stage machine learning approach was adopted to construct datasets for sulfate reduction and phosphorus removal rates, respectively. The model was optimized using machine learning algorithms and cross-validation methods. The performance of the sulfur cycle coupled with denitrification and phosphorus removal system was predicted and optimized through a scheduling platform.
It enables precise prediction and optimized control of the sulfur cycle coupled with denitrification and phosphorus removal system, reduces carbon source demand, improves process stability and efficiency, and reduces operating costs.
Smart Images
Figure CN116861341B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wastewater treatment technology, and in particular to a method for predicting the performance of a two-stage sulfur cycle coupled with denitrification and phosphorus removal process. Background Technology
[0002] Municipal wastewater in southern my country generally has a low C / P ratio, making it difficult for biological wastewater treatment processes to meet increasingly stringent national standards for phosphate concentration. Conventional solutions to address carbon source shortages include adding external carbon sources (such as methanol) or chemical agents (such as iron and aluminum salts) to enhance phosphorus removal efficiency. However, these methods increase carbon emissions and wastewater treatment costs. Under the dual carbon goals, developing low-carbon, low-cost, highly efficient, and stable biological phosphorus removal processes is a significant challenge in the field of urban wastewater treatment. Sulfur is one of the most important elements in nature, characterized by its low price and diverse forms. Human industrial activities themselves generate large amounts of sulfur-containing wastewater and exhaust gases, which produce significant amounts of sulfide during treatment. Therefore, fully utilizing sulfide as an electron carrier to enhance biological phosphorus removal has attracted widespread attention.
[0003] The sulfur cycle coupled denitrification-phosphorus removal (DS-EBPR) process operates under alternating anaerobic / anoxic conditions, achieving simultaneous carbon, nitrogen, and phosphorus removal through the synergistic action of sulfate-reducing bacteria (SRB) and sulfur-oxidizing bacteria (SOB). It is adaptable to high-temperature and high-salinity environments and offers advantages such as saving carbon sources and aeration, and low sludge production. However, due to the unclear metabolic mechanisms, potential synergistic mechanisms, and regulatory factors driving the DS-EBPR process, the process still faces instability in actual operation. Furthermore, limited by the complex biochemical reaction processes and high-dimensional water quality, operating, sludge, and microbial parameters, the traditional activated sludge model ASM-2d cannot fully simulate and predict effluent quality and reaction effects, posing challenges to the optimization and control of the DS-EBPR process. Summary of the Invention
[0004] To address the limitations and defects of existing technologies, this invention provides a method for predicting the performance of a two-stage sulfur cycle coupled with denitrification and phosphorus removal process, comprising:
[0005] Collect process parameters of sulfur cycle coupled denitrification for phosphorus removal, and construct a dataset SR labeled with sulfate reduction amount in the anaerobic stage and a dataset P labeled with phosphorus removal rate.
[0006] Based on the dataset SR, with the anaerobic stage process parameters as input values and the sulfate reduction amount as output value, the dataset SR is divided into a training set and a test set. A machine learning algorithm is used to build a model to train the training set data. The n-fold cross-validation method and the search method are adopted. The hyperparameters are adjusted according to the root mean square error (RMSE) of the validation set to optimize the model. The model is finally evaluated through various indicators of the test set data to achieve the prediction of the sulfate reduction amount.
[0007] Based on dataset P, with the predicted value of sulfate reduction and the process parameters of the anoxic stage as input values and the phosphorus removal rate as output value, dataset P is divided into training set and test set. A machine learning algorithm is used to build a model to train the training set data. The n-fold cross-validation method and search method are adopted. The hyperparameters are adjusted according to the root mean square error (RMSE) of the validation set to optimize the model. The model is finally evaluated through various indicators of the test set data to achieve the prediction of the phosphorus removal rate.
[0008] A scheduling platform is constructed based on the trained model. The scheduling platform is used to predict the amount of sulfate reduction based on the anaerobic stage process parameters. The scheduling platform is also used to predict the phosphorus removal rate based on the predicted value of the amount of sulfate reduction and the anoxic stage process parameters.
[0009] Optionally, the process parameters include water quality parameters, operating parameters, and sludge parameters;
[0010] The water quality parameters include at least one of the following: total organic carbon concentration, acetate concentration, propionate concentration, phosphate concentration, nitrate concentration, sulfate concentration, and salinity.
[0011] The operating parameters include at least one of temperature, pH, anaerobic time, anoxic time, and working volume;
[0012] The sludge parameters include at least one of the following: mixed liquor suspended solids concentration, mixed liquor volatile suspended solids concentration, and sludge age.
[0013] Optionally, the ratio of the training set to the test set in the SR dataset is 7:3, 8:2, or 9:1.
[0014] The ratio of the training set to the test set in the dataset P is 7:3, 8:2, or 9:1.
[0015] Optionally, the machine learning model constructed according to the machine learning algorithm includes at least one of Decision Tree (DT), Random Forest (RF), Gradient Boosting Tree (GBDT), Extreme Gradient Boosting Machine (XGBoost), Lightweight Gradient Boosting Machine (LightGBM), and Categorical Feature Boosting Machine (CatBoost).
[0016] Optionally, the n-fold cross-validation method includes:
[0017] The training set is divided into n non-overlapping parts. One part is selected in turn as the validation set, and the remaining (n-1) parts are used as the training set for iterative training to obtain n process performance prediction models. The n process performance prediction models are trained and validated n times, and the root mean square error (RMSE) of the test results is finally returned.
[0018] Optionally, the value of n in the n-fold cross-validation method is in the range of 4-10.
[0019] Optionally, the search method includes at least one of the following: grid search method, random search method, Bayesian optimization algorithm, particle swarm optimization algorithm, and genetic algorithm.
[0020] Optionally, the step of optimizing the model by adopting the n-fold cross-validation method and the search method and adjusting the hyperparameters based on the root mean square error (RMSE) of the validation set includes:
[0021] Based on the search method, different combinations of hyperparameters are used to optimize the model, with the minimum mean square error (RMSE) returned by n-fold cross-validation being the optimal hyperparameter combination.
[0022] Optionally, the various metrics of the test set data include the coefficient of determination R. 2 Mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE).
[0023] Optionally, the scheduling platform includes: an anaerobic stage process parameter input module, a sulfate reduction prediction module, an anoxic stage process parameter and sulfate reduction prediction value input module, and a phosphorus removal rate prediction module.
[0024] The present invention has the following beneficial effects:
[0025] The two-stage prediction method for sulfur cycle coupled with denitrification and phosphorus removal process provided by this invention achieves prediction of the performance of the sulfur cycle coupled with denitrification and phosphorus removal system through a two-stage approach, comprehensively considering the influence of water quality parameters, operating parameters, sludge parameters, and intermediate sulfate reduction rate on phosphorus removal efficiency. Furthermore, this invention, through a two-stage machine learning method based on a tree model, can overcome the limitations of "empirical methods" for process optimization and control, achieving accurate prediction of phosphorus removal rate.
[0026] This invention provides a two-stage method for predicting and optimizing the performance of a sulfur cycle coupled with denitrification for phosphorus removal, comprising: firstly, compiling datasets for sulfate reduction and phosphorus removal rate respectively; in the first stage, using anaerobic stage process parameters as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict sulfate reduction; in the second stage, using anoxic stage process parameters and predicted sulfate reduction values as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict phosphorus removal rate; during model training, using the coefficient of determination R... 2 The model is evaluated using the root mean square error (RMSE), and the optimal model is selected based on the evaluation results. A scheduling platform is built based on the optimal model, and the performance prediction and optimized control of the sulfur cycle coupled denitrification and phosphorus removal system are realized using the scheduling platform. Attached Figure Description
[0027] Figure 1 The flowchart is for the two-stage prediction method of sulfur cycle coupled denitrification phosphorus removal process provided in Embodiment 1 of the present invention.
[0028] Figure 2 This is a schematic diagram of the test results of the CatBoost algorithm model for predicting sulfate reduction provided in Embodiment 1 of the present invention.
[0029] Figure 3 This is a schematic diagram of the test results of the GBDT algorithm prediction phosphorus removal rate model provided in Embodiment 1 of the present invention. Detailed Implementation
[0030] To enable those skilled in the art to better understand the technical solution of the present invention, the method for predicting the performance of a two-stage sulfur cycle coupled denitrification phosphorus removal process provided by the present invention will be described in detail below with reference to the accompanying drawings.
[0031] Example 1
[0032] Machine learning is a crucial method for realizing artificial intelligence applications, widely used in fields such as computer vision, natural language processing, and data mining. Due to its broad applicability and ability to describe complex and nonlinear problems without considering underlying mechanisms, machine learning has become increasingly popular. Its powerful learning and generalization capabilities offer new possibilities for the optimized control of sulfur cycle coupled denitrification-epidemic resorption (DS-EBPR). Some tree-based machine learning algorithms, such as XGBoost, are well-suited for processing tabular data and offer high interpretability. However, in DS-EBPR models, some intermediate influencing factors, such as sulfate reduction, have a significant impact on performance but are difficult to obtain directly. Therefore, this embodiment presents a two-stage DS-EBPR system performance prediction method based on machine learning.
[0033] This embodiment provides a two-stage method for predicting the performance of a sulfur cycle coupled with denitrification for phosphorus removal based on machine learning, including the following steps:
[0034] Step S1: Construct datasets; collect process parameters for sulfur cycle coupled denitrification and phosphorus removal, and construct a dataset labeled with sulfate reduction amount in the anaerobic stage, called the "SR-dataset" and a dataset labeled with phosphorus removal rate, called the "P-dataset".
[0035] Step S2, Stage 1: Based on the “SR-dataset”, with the anaerobic process parameters as input values and sulfate reduction as output values, the dataset is divided into training and testing sets. Machine learning algorithms are used to build a model training data. The n-fold cross-validation and search method are adopted. The hyperparameters are adjusted to optimize the model based on the root mean square error (RMSE) on the validation set. Finally, the model is evaluated through various indicators of the test set data to achieve the prediction of sulfate reduction.
[0036] Step S3, Stage Two: Based on the "P-dataset", with the predicted sulfate reduction rate from Stage One and the process parameters of the anoxic section as input values and the phosphorus removal rate as output value, the dataset is divided into a training set and a test set. Machine learning algorithms are used to build a model to train the data. An n-fold cross-validation and search method is adopted. The hyperparameters are adjusted to optimize the model based on the root mean square error (RMSE) on the validation set. Finally, the model is evaluated through various indicators of the test set data to achieve the prediction of phosphorus removal rate.
[0037] Step S4: Based on the optimal model, build a scheduling platform to predict the process parameters in the first stage—sulfate reduction amount and then in the second stage—phosphorus removal rate, so as to provide guidance for the optimization and control of the sulfur cycle coupled denitrification phosphorus removal system.
[0038] In this embodiment, the process parameters include water quality parameters, operating parameters, and sludge parameters. The water quality parameters include one or more of the following: influent total organic carbon concentration, influent acetate concentration, influent propionate concentration, influent phosphate concentration, influent nitrate concentration, influent sulfate concentration, and salinity. The operating parameters include one or more of the following: temperature, pH, anaerobic time, anoxic time, and operating volume. The sludge parameters include one or more of the following: sludge concentration (MLSS, MLVSS), MLVSS / MLSS, and sludge time (SRT).
[0039] In this embodiment, the ratio of the data to the training set and the test set includes, but is not limited to, 7:3, 8:2, and 9:1. The machine learning model includes one or more of the following: Decision Tree (DT), Random Forest (RF), Gradient Boosting Tree (GBDT), Extreme Gradient Boosting Machine (XGBoost), Lightweight Gradient Boosting Machine (LightGBM), and Categorical Feature Boosting Machine (CatBoost).
[0040] In this embodiment, the n-fold cross-validation method divides the training set into n non-overlapping parts, selects one part as the validation set in turn, and uses the remaining (n-1) parts as the training set for iterative training, obtaining n process performance prediction models. This can be trained and validated n times, ultimately returning the average RMSE of the test results. The value of n in the n-fold cross-validation ranges from 4 to 10. The search method includes one or more of grid search, random search, Bayesian optimization algorithm, particle swarm optimization, and genetic algorithm. Hyperparameters are adjusted based on the n-fold cross-validation and search method, and different combinations of hyperparameters are performed according to the search method. The minimum average RMSE returned by the n-fold cross-validation is taken as the optimal hyperparameter combination for subsequent test set evaluation.
[0041] In this embodiment, the various indicators of the evaluation model include the coefficient of determination R. 2 The mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are also mentioned. The coefficient of determination R... 2 The closer the value is to 1, the better the model's predictive performance. The smaller the mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE), the better the model's predictive performance.
[0042] In this embodiment, the optimal model scheduling platform includes the following modules: an anaerobic stage process parameter input module; a sulfate reduction prediction module; an anoxic stage process parameter and sulfate reduction prediction value input module; and a phosphorus removal rate prediction module.
[0043] Figure 1 This is a flowchart of the two-stage prediction method for the sulfur cycle coupled denitrification phosphorus removal process provided in Embodiment 1 of the present invention. Figure 1 As shown in the figure, this embodiment discloses a two-stage method for predicting the performance of a sulfur cycle coupled with denitrification and phosphorus removal process based on machine learning, including the following steps:
[0044] Process parameters of the DS-EBPR system were collected from the laboratory and literature. Two datasets were compiled with sulfate reduction and phosphorus removal rate as labels, which were named "SR-dataset" and "P-dataset" respectively.
[0045] Phase 1: The input parameters for the "SR-dataset" included water quality parameters (salinity, influent phosphate, influent total organic carbon, influent sulfate, influent acetate and influent propionate concentrations), operating parameters (pH, temperature, anaerobic time and working volume), and sludge parameters (MLSS, MLVSS, MLVSS / MLSS), resulting in 349 sets of valid data.
[0046] Figure 2 This is a schematic diagram illustrating the test results of the CatBoost algorithm model for predicting sulfate reduction provided in Embodiment 1 of the present invention. Figure 2 As shown, the “SR-dataset” is divided into training and testing sets according to a 9:1 ratio. The CatBoost algorithm is selected, and the hyperparameters are tuned in the hyperparameter search space in combination with nine-fold cross-validation and Bayesian optimization. The RMSE of the validation set is used as the indicator, and the hyperparameter combination corresponding to the minimum RMSE is selected for subsequent evaluation on the test set.
[0047] On the test set, the R-value between the actual and predicted values of sulfate reduction is used. 2 Using RMSE as an indicator, when R 2 The closer the value is to 1 and the smaller the RMSE, the better the model's generalization ability. The model trained using the above method is called the SR-CatBoost model.
[0048] Phase Two: The input parameters for the "P-dataset" were expanded to include anoxic time and influent nitrate concentration, resulting in a total of 498 valid data sets.
[0049] Based on the SR-CatBoost model trained in Phase 1, the data in the "P-dataset" is input to obtain the predicted value of sulfate reduction for each set of data, and this predicted value is added as the input value of the "P-dataset".
[0050] Figure 3 This is a schematic diagram illustrating the test results of the GBDT algorithm-based phosphorus removal rate prediction model provided in Embodiment 1 of the present invention. Figure 3 As shown, the "P-dataset" is divided into training and test sets according to a 9:1 ratio. The GBDT algorithm is selected, and the hyperparameters are tuned in the hyperparameter search space in combination with nine-fold cross-validation and Bayesian optimization algorithm. The RMSE of the validation set is used as the indicator, and the hyperparameter combination corresponding to the minimum RMSE is selected for subsequent evaluation on the test set.
[0051] On the test set, the R-value between the actual and predicted phosphorus removal rates is used. 2 Using RMSE as an indicator, when R 2 The closer the value is to 1 and the smaller the RMSE, the better the model's generalization ability. The model trained using the above method is called the P-GBDT model.
[0052] Based on the P-GBDT model, a scheduling platform is built, including the following modules:
[0053] Anaerobic section process parameter input module;
[0054] Sulfate reduction prediction module;
[0055] Input module for process parameters and predicted sulfate reduction in the anoxic section;
[0056] Phosphorus removal rate prediction module.
[0057] Based on the P-GBDT model, this embodiment can provide guidance for the optimization and control of sulfur cycle coupled denitrification and phosphorus removal systems, regulate environmental factors, and improve the stability of effluent quality and process operation.
[0058] This embodiment provides a two-stage method for predicting and optimizing the performance of a sulfur cycle coupled denitrification phosphorus removal process, comprising: firstly, compiling datasets for sulfate reduction and phosphorus removal rate respectively; secondly, using anaerobic stage process parameters as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict sulfate reduction; and thirdly, using anoxic stage process parameters and predicted sulfate reduction values as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict phosphorus removal rate. During model training, the model is evaluated using the coefficient of determination R² and root mean square error RMSE, and the optimal model is selected based on the evaluation results. A scheduling platform is built based on the optimal model, and the performance prediction and optimized control of the sulfur cycle coupled denitrification phosphorus removal system are achieved based on the scheduling platform.
[0059] Example 2
[0060] This embodiment discloses a two-stage method for predicting the performance of a sulfur cycle coupled with denitrification for phosphorus removal based on machine learning, including the following steps:
[0061] Process parameters of the DS-EBPR system were collected from the laboratory and literature. Two datasets were compiled with sulfate reduction and phosphorus removal rate as labels, which were named "SR-dataset" and "P-dataset" respectively.
[0062] Phase 1: The input parameters for the "SR-dataset" include water quality parameters (influent phosphate, influent total organic carbon, and influent sulfate), operating parameters (pH, temperature, anaerobic time), and sludge parameters (SRT, MLVSS / MLSS).
[0063] The “SR-dataset” was divided into training and testing sets according to an 8:2 ratio. The XGBoost algorithm was selected, and hyperparameters were tuned in the hyperparameter search space using five-fold cross-validation and grid search. The RMSE of the validation set was used as the indicator, and the hyperparameter combination corresponding to the minimum RMSE was selected for subsequent evaluation on the test set.
[0064] On the test set, the R-value between the actual and predicted values of sulfate reduction is used. 2 Using RMSE as an indicator, when R 2 The closer the value is to 1 and the smaller the RMSE, the better the model's generalization ability. The model trained using the above method is called the SR-XGBoost model.
[0065] Phase Two: The input parameters for the "P-dataset" are based on the "SR-dataset" with the addition of anoxic time and influent nitrate concentration.
[0066] Based on the SR-XGBoost model trained in Phase 1, the data in the "P-dataset" is input to obtain the predicted value of sulfate reduction for each set of data, and this predicted value is added as the input value of the "P-dataset".
[0067] The "P-dataset" is divided into training and testing sets according to an 8:2 ratio. The LightGBM algorithm is selected, and hyperparameters are tuned in the hyperparameter search space using five-fold cross-validation and random search. The RMSE of the validation set is used as the indicator, and the hyperparameter combination corresponding to the minimum RMSE is selected for subsequent evaluation on the test set.
[0068] On the test set, the R-value between the actual and predicted phosphorus removal rates is used. 2 Using RMSE as an indicator, when R 2 The closer the value is to 1 and the smaller the RMSE, the better the model's generalization ability. The model trained using the above method is called the P-LightGBM model.
[0069] Based on the P-LightGBM model, a scheduling platform was built, including the following modules:
[0070] Anaerobic section process parameter input module;
[0071] Sulfate reduction prediction module;
[0072] Input module for process parameters and predicted sulfate reduction in the anoxic section;
[0073] Phosphorus removal rate prediction module.
[0074] This embodiment provides a two-stage method for predicting and optimizing the performance of a sulfur cycle coupled with denitrification for phosphorus removal. The method includes: firstly, compiling datasets for sulfate reduction and phosphorus removal rate; secondly, using anaerobic stage process parameters as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict sulfate reduction; and thirdly, using anoxic stage process parameters and predicted sulfate reduction values as input, selecting machine learning algorithms, cross-validation methods, and search methods to predict phosphorus removal rate. During model training, the coefficient of determination R0 is used to... 2 The model is evaluated using the root mean square error (RMSE), and the optimal model is selected based on the evaluation results. A scheduling platform is built based on the optimal model, and the performance prediction and optimized control of the sulfur cycle coupled denitrification and phosphorus removal system are realized using the scheduling platform.
[0075] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.
Claims
1. A method for two-stage prediction of the performance of a sulfur cycle coupled denitrifying phosphorus removal process, characterized in that, The method comprises the following steps: Collecting process parameters of a sulfur cycle coupled denitrifying phosphorus removal process, constructing a data set SR with a label of the amount of sulfate reduction in the anaerobic stage, and a data set P with a label of the phosphorus removal rate; Based on the data set SR, the process parameters in the anaerobic stage are taken as input values, and the amount of sulfate reduction is taken as output values, the data set SR is divided into a training set and a test set, a machine learning algorithm is used to construct a model to train the training set data, an n-fold cross-validation method and a search method are adopted, the root mean square error RMSE of the validation set is used to adjust the hyperparameters to optimize the model, and finally the model is evaluated through various indicators of the test set data to realize the prediction of the amount of sulfate reduction; Based on the data set P, the predicted value of the amount of sulfate reduction and the process parameters in the anoxic stage are taken as input values, and the phosphorus removal rate is taken as output values, the data set P is divided into a training set and a test set, a machine learning algorithm is used to construct a model to train the training set data, an n-fold cross-validation method and a search method are adopted, the root mean square error RMSE of the validation set is used to adjust the hyperparameters to optimize the model, and finally the model is evaluated through various indicators of the test set data to realize the prediction of the phosphorus removal rate; A scheduling platform is constructed according to the trained model, the scheduling platform is used to predict the amount of sulfate reduction according to the process parameters in the anaerobic stage, and the scheduling platform is used to predict the phosphorus removal rate according to the predicted value of the amount of sulfate reduction and the process parameters in the anoxic stage.
2. The method for two-stage prediction of the performance of the sulfur cycle coupled denitrifying phosphorus removal process according to claim 1, characterized in that, The process parameters include water quality parameters, operation parameters and sludge parameters; The water quality parameters include at least one of the total organic carbon concentration of the influent, the acetate concentration of the influent, the propionate concentration of the influent, the phosphate concentration of the influent, the nitrate concentration of the influent, the sulfate concentration of the influent and salinity; The operation parameters include at least one of temperature, pH, anaerobic time, anoxic time and working volume; The sludge parameters include at least one of mixed liquor suspended solids concentration, mixed liquor volatile suspended solids concentration and sludge age.
3. The method for two-stage prediction of the performance of the sulfur cycle coupled denitrifying phosphorus removal process according to claim 2, characterized in that, The ratio of the training set to the test set in the division of the data set SR is 7:3 or 8:2 or 9:1; The ratio of the training set to the test set in the division of the data set P is 7:3 or 8:2 or 9:
1.
4. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 3, characterized in that, The machine learning model constructed according to the machine learning algorithm includes at least one of a decision tree DT, a random forest RF, a gradient boosting tree GBDT, an extreme gradient boosting machine XGBoost, a lightweight gradient boosting machine LightGBM and a category feature boosting machine CatBoost.
5. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 4, characterized in that, The n-fold cross-validation method comprises: Divide the training set into n non-overlapping parts, and select one part as the validation set in turn, and the remaining (n-1) parts are used as the training set for cyclic training to obtain n process performance prediction models, and the n process performance prediction models are trained and verified n times, and finally the average root mean square error RMSE of the test results is returned.
6. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 5, characterized in that, The value of n in the n-fold cross-validation method ranges from 4 to 10.
7. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 6, characterized in that, The search method includes at least one of a grid search method, a random search method, a Bayesian optimization algorithm, a particle swarm algorithm and a genetic algorithm.
8. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 7, characterized in that, The step of adjusting the hyperparameters to optimize the model according to the root mean square error (RMSE) of the verification set includes: According to the search method, the combination of different hyperparameters is carried out, and the minimum average root mean square error (RMSE) returned by the n-fold cross validation is the optimal hyperparameter combination for optimizing the model.
9. The method for two-stage prediction of performance of sulfur cycle coupled denitrifying phosphorus removal process according to claim 8, characterized in that, The various indicators of the test set data include a determination coefficient R 2 , a mean square error MSE, a root mean square error RMSE, and a mean absolute percentage error MAPE.
10. The method for two-stage prediction of the performance of the sulfur cycle coupled denitrifying phosphorus removal process according to claim 9, characterized in that, The scheduling platform comprises an anaerobic stage process parameter input module, a sulfate reduction amount prediction module, an anoxic stage process parameter and sulfate reduction amount prediction value input module, and a phosphorus removal rate prediction module.