Flocculation air flotation algae removal process optimization method based on machine learning

Through the machine learning method of the H2O AutoML platform, the automatic flocculation and algae removal process is optimized, which solves the problem of difficult to accurately control water quality changes in traditional technologies, improves treatment efficiency and water quality safety, and achieves the stability of flocculation and algae removal process Sexual and economical.

CN119025844APending Publication Date: 2024-11-26HOHAI UNIV

Patent Information

Application Number
CN202410915748.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Traditional flocculation air floatation technology is difficult to achieve precise control when treating changes in water quality, which makes it difficult to ensure treatment efficiency and water quality safety, and the instability of operating parameters and water quality fluctuations affect the stability of the treatment effect.

Method used

Using machine learning method based on the H2O AutoML platform, through automated model selection and parameter tuning, the optimization control model of flocculation air-floating algae removal process is obtained, and the flocculant dosage and air-floating conditions are dynamically adjusted to adapt to real-time water quality changes.

Benefits of technology

It improves the efficiency and water quality safety of algae water treatment, reduces the use of chemicals, reduces operating costs, enhances the stability and reliability of the system, and achieves rapid response and continuous monitoring of water quality changes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025844A_ABST
    Figure CN119025844A_ABST
Patent Text Reader

Abstract

The invention provides a machine learning-based flocculation air floatation algae removal process optimization method, which adopts an H2O automatic machine learning platform (H2O AutoML) to optimize a flocculation air floatation algae removal process, and improves algae water treatment efficiency and water quality safety. The machine learning-based flocculation air floatation algae removal process optimization method comprises the following steps: firstly, acquiring flocculation air floatation algae removal treatment historical data and dividing the data into a training set and a test set; and the H2O AutoML platform automatically executes training of various machine learning algorithms, selects an optimal model according to evaluation indexes, performs hyper-parameter optimization on the optimal model to obtain a flocculation air flotation algae removal process optimization control model, and deploys the optimization control model to a control system of a water treatment facility. And according to real-time water quality data, dynamically predicting and adjusting the adding amount of a flocculating agent, a flocculation condition and an air flotation condition. Therefore, through automatic machine learning model selection and parameter optimization, real-time changing water quality conditions are adapted, and the adding amount and operation parameters of the flocculating agent are optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of water treatment, and in particular to a method for optimizing a flocculation flotation algae removal process based on machine learning. Background Art

[0002] Flocculation flotation technology is a commonly used method for water treatment. By adding flocculants, algae and other suspended matter in the water can be aggregated into larger particles, and then these particles are separated from the water by flotation. During the flocculation flotation process, parameters such as the type of flocculant, dosage, flotation pressure and time usually need to be adjusted according to the specific water quality and treatment objectives. Traditionally, the selection of operating parameters often relies on experience or optimization through repeated trials, lacks precise control, and is difficult to adapt to changes in water quality in real time. In addition, the optimization process is cumbersome and easily affected by water quality fluctuations, making it difficult to ensure treatment efficiency and water quality safety.

[0003] Therefore, it can be seen that the traditional flocculation and flotation technology for algae removal is not only time-consuming and labor-intensive, but the flocculation and flotation process is also very sensitive to changes in water quality. Changes in the concentration and type of pollutants in the water will lead to unstable treatment effects, requiring re-optimization and adjustment of parameters, which is not only time-consuming and resource-intensive, but also prone to inconsistent treatment effects due to the subjective judgment of operators. Due to the instability of operating parameters and the influence of water quality fluctuations, the quality of the treated water fluctuates, posing challenges to the efficiency of water treatment and the safety of treated water quality. Summary of the invention

[0004] In view of the defects of the prior art, the object of the present invention is to provide a flocculation flotation algae removal process optimization method based on machine learning, and to optimize the flocculation flotation algae removal process based on H2O automatic machine learning platform (H2O AutoML) to improve the algae water treatment efficiency and water quality safety.

[0005] In the flocculation flotation algae removal process optimization method based on machine learning of the present invention, through automated machine learning model selection and parameter tuning, it is possible to adapt to real-time changing water quality conditions and scientifically optimize the flocculant dosage and operating parameters.

[0006] According to the first aspect of the present invention, a flocculation flotation algae removal process optimization method based on machine learning is proposed, comprising the following steps:

[0007] Obtain historical data on flocculation and flotation algae removal, including historical water quality data, flocculation and flotation process parameters, and algae treatment data, and obtain a training data set after preprocessing the historical data;

[0008] Divide the training data set and determine the training set and test set according to the preset ratio;

[0009] The H2O AutoML platform automatically trains multiple machine learning algorithms using the above training and test sets, and selects the best model based on pre-defined evaluation metrics;

[0010] The hyperparameters of the optimal model were optimized to obtain the optimal control model of flocculation and flotation algae removal process;

[0011] Deploy the flocculation and flotation algae removal process optimization control model into the control system of the water treatment facility, dynamically predict the flocculation and flotation operation parameters based on the real-time collected water quality data, and adjust the actual flocculant dosage, flocculation conditions and flotation conditions; and

[0012] The algae removal effect was evaluated based on the actual algae removal treatment using the actual flocculation and flotation operating parameters.

[0013] In a further embodiment, the flocculation flotation algae removal treatment historical data includes:

[0014] Historical water quality data: turbidity, algae species and quantity, pH, water temperature, conductivity;

[0015] Flocculation flotation process parameters: flotation time, coagulant dosage and stirring intensity;

[0016] Algae treatment data: effluent algae species and quantity, turbidity, and physical and chemical indicators of algae, including soluble and bound extracellular polymers and algal cell Zeta potential.

[0017] In a further embodiment, the machine learning algorithms executed on the H2O AutoML platform include GLM random forest, DRF distributed random forest, XRT extreme random forest, DeepLearning deep learning, XGboost, GBM gradient boosting machine and Stack integrated stacking model, and the optimal model is selected based on predetermined evaluation indicators.

[0018] In a further embodiment, the H2O AutoML platform automatically performs the training process configuration of multiple machine learning algorithms including:

[0019] Select the variable of the model prediction output, that is, set the response_column value;

[0020] Set the maximum running time of AutoML to 300s, that is, the max_runtime_secs parameter is set to 300s;

[0021] Adjust the maximum number of models that AutoML explores before stopping to 10, that is, the max_models parameter is set to 10;

[0022] Provide a random seed for the AutoML running process to ensure the repeatability of the experiment. The seed number is 1234, that is, the seed parameter is set to 1234;

[0023] The distribution function supported by each machine learning algorithm is selected as AUTO, that is, the distribution parameter is set to AUTO;

[0024] The number of cross-validation folds is set to 5, which helps to evaluate the stability and generalization ability of the model, that is, the nfolds parameter is set to 5;

[0025] The training stopping condition, stopping_rounds, is set to 3, which means that if the model performance does not improve within the specified number of rounds, the training process stops.

[0026] The keep_cross_validation_models parameter is set to True to keep the cross-validated model.

[0027] In a further embodiment, the step of optimizing hyperparameters of the optimal model includes:

[0028] Automatically tune model parameters using the H2O AutoML platform’s grid search or random search techniques.

[0029] In a further embodiment, the method further comprises the following steps:

[0030] The model interpretation function of the H2O AutoML platform was used to analyze the key factors and their weights that affect the algae removal efficiency.

[0031] In a further embodiment, the model interpretation methods used include residual analysis, variable importance analysis, Shapley interpretation method and partial dependence curve, which are used to interpret the prediction results of the optimal model obtained by training the H2O AutoML platform.

[0032] In a further embodiment, the method further comprises the following steps:

[0033] Water quality data, actual flocculation flotation operation parameters and algae removal effects are collected according to the preset cycle, the flocculation flotation algae removal process optimization control model is dynamically trained and updated, and the updated flocculation flotation algae removal process optimization control model is redeployed.

[0034] The flocculation flotation algae removal process optimization method based on automatic machine learning proposed in the present invention can predict indicators such as the number of algae and turbidity of the effluent according to the input water quality parameters and process parameters through a prediction model, so that the setting of flocculant dosage and flotation operating conditions is no longer blind, so as to cope with rapidly changing water quality conditions and ensure that the effluent quality continues to meet safety standards. It has good social and economic benefits and can assist water plant process managers to quickly make appropriate flocculation flotation process decisions.

[0035] The present invention introduces an automated machine learning module in the flocculation flotation process prediction, which can automatically realize model selection and hyperparameter optimization, and integrate a variety of basic machine learning algorithms, thereby improving the accuracy and generalization ability of the model; through automated model development, errors caused by human factors, such as manual data processing and parameter selection errors, are reduced, thereby improving the repeatability and reliability of the entire model development process.

[0036] Combined with the implementation of the flocculation flotation algae removal process optimization method in the above embodiment, compared with the prior art, its significant beneficial effects are:

[0037] 1. Provide automated and intelligent decision support: Use the H2O AutoML platform to automatically perform the training of multiple machine learning algorithms, and select the optimal model from a large number of possible models. Through automated decision optimization, it not only reduces the workload of professional technicians in model selection and parameter adjustment, but also improves the accuracy and efficiency of decision-making;

[0038] 2. Provide data-driven optimization process: By utilizing historical data and real-time data, this method can dynamically adjust the flocculation and flotation operating parameters, such as the dosage of flocculants, flocculation conditions and flotation conditions. The data-driven method makes the water treatment process more precise and can respond quickly to real-time changes in water quality.

[0039] 3. Improve treatment efficiency and economy: Scientifically and rationally optimized flocculation flotation operating parameters can improve algae removal effects, reduce unnecessary use of chemicals, and reduce operating costs;

[0040] 4. Enhance the stability and reliability of the system: By continuously monitoring and adjusting the operating parameters, this method can cope with the fluctuations and changes in the raw water quality and maintain the stability and reliability of the treatment effect. This is especially important in dealing with seasonal changes or changes in water quality during emergencies. The operating process parameters can be adjusted in real time according to the changes in water quality in a timely and dynamic manner to ensure the high efficiency of water treatment and water quality safety.

[0041] 5. Continuous performance monitoring and evaluation: By continuously evaluating the actual flocculation flotation operating parameters and algae removal effects, this method can monitor the performance of the entire system in real time. Continuous monitoring helps to identify problems and make necessary adjustments in a timely manner, thereby ensuring water treatment results;

[0042] 6. Easy to expand and adapt to different scenarios: Through online updating and deployment of machine learning models, the optimization method of the present invention can easily adapt to water treatment facilities of different sizes and different water quality conditions. Moreover, as more data is accumulated, the prediction ability and accuracy of the model will be further improved.

[0043] It should be understood that all combinations of the aforementioned concepts and the additional concepts described in more detail below can be considered as part of the inventive subject matter of the present disclosure as long as such concepts are not mutually inconsistent. In addition, all combinations of the claimed subject matter are considered as part of the inventive subject matter of the present disclosure.

[0044] The foregoing and other aspects, embodiments and features of the present invention can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as the features and / or beneficial effects of the exemplary embodiments, will be apparent from the following description or learned from the practice of the specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component shown in each figure may be represented by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings.

[0046] Figure 1 It is a flow chart of a flocculation flotation algae removal process optimization method based on machine learning according to an embodiment of the present invention.

[0047] Figure 2 Exploratory data analysis diagram of the raw data provided by this invention.

[0048] Figure 3 The Pearson product-moment correlation coefficient matrix diagram of the original data provided by the present invention.

[0049] Figure 4 Candidate models and their accuracy graphs during training on the H2O AutoML platform provided by the present invention.

[0050] Figure 5 This is a residual analysis diagram of the best training model provided by the present invention.

[0051] Figure 6A learning curve diagram of the integrated stacking training model provided by the present invention.

[0052] Figure 7 Variable importance graph for the best training model provided by the present invention.

[0053] Figure 8 SHAP summary diagram of the best training model provided by the present invention.

[0054] Fig. 9 Partial dependency diagram of variables for the top six models provided by the present invention. DETAILED DESCRIPTION

[0055] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0056] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed by the present invention are not limited to any implementation. In addition, some aspects disclosed by the present invention can be used alone or in any appropriate combination with other aspects disclosed by the present invention.

[0057] {Optimization method of flocculation flotation algae removal process based on machine learning}

[0058] Combination Figure 1 The flocculation flotation algae removal process optimization method based on machine learning in the illustrated embodiment includes the following steps:

[0059] Obtain historical data on flocculation and flotation algae removal, including historical water quality data, flocculation and flotation process parameters, and algae treatment data, and obtain a training data set after preprocessing the historical data;

[0060] Divide the training data set and determine the training set and test set according to the preset ratio;

[0061] The H2O AutoML platform automatically trains multiple machine learning algorithms using the above training and test sets, and selects the best model based on pre-defined evaluation metrics;

[0062] The hyperparameters of the optimal model were optimized to obtain the optimal control model of flocculation and flotation algae removal process;

[0063] Deploy the flocculation and flotation algae removal process optimization control model into the control system of the water treatment facility, dynamically predict the flocculation and flotation operation parameters based on the real-time collected water quality data, and adjust the actual flocculant dosage, flocculation conditions and flotation conditions; and

[0064] The algae removal effect was evaluated based on the actual algae removal treatment using the actual flocculation and flotation operating parameters.

[0065] In a further embodiment, the flocculation flotation algae removal treatment historical data includes:

[0066] Historical water quality data: turbidity, algae species and quantity, pH, water temperature, conductivity;

[0067] Flocculation flotation process parameters: flotation time, coagulant dosage and stirring intensity;

[0068] Algae treatment data: effluent algae species and quantity, turbidity, and physical and chemical indicators of algae, including soluble and bound extracellular polymers and algal cell Zeta potential.

[0069] The training data set is obtained through preprocessing the acquired historical data of flocculation flotation algae removal.

[0070] In an embodiment of the present invention, the aforementioned preprocessing includes: data normalization and outlier processing.

[0071] By normalizing historical data, data of different dimensions or numerical ranges in the historical data are converted to the same scale, and the contribution of features to the model is balanced, which is beneficial to the convergence performance during model training and prevents convergence problems caused by differences in feature data scales. At the same time, it improves the accuracy of the training model and makes it easier to capture subtle changes in the data, thereby improving the accuracy and robustness of the model.

[0072] As an example, the normalization method of this embodiment adopts the Min-Max Scaling or Z-score normalization method.

[0073] The aforementioned outlier processing, also known as outlier processing, refers to identifying and processing those values ​​that are significantly different from the majority of data or are obviously unreasonable, and eliminating or reducing the impact of outliers on model training through direct deletion / conditional deletion, replacement, clipping, transformation, etc. In the embodiment of the present invention, the direct deletion method is selected to propose outlier data to eliminate noise and errors in the data, improve the overall quality and signal-to-noise ratio of the data set, improve the accuracy and reliability of model training, and eliminate the model error and prediction error caused by outliers.

[0074] In an embodiment of the present invention, the aforementioned training set and test set data are divided in a ratio of 0.8:0.2.

[0075] In a further embodiment, the machine learning algorithms executed on the H2O AutoML platform include GLM random forest, DRF distributed random forest, XRT extreme random forest, DeepLearning deep learning, XGboost, GBM gradient boosting machine and Stack integrated stacking model, and the optimal model is selected based on predetermined evaluation indicators.

[0076] In an embodiment of the present invention, the segmented training set is selected as model training data, that is, specified by the training_frame parameter.

[0077] In a further embodiment, the H2O AutoML platform automatically performs the training process configuration of multiple machine learning algorithms including:

[0078] Select the variable of the model prediction output, that is, set the response_column value;

[0079] Set the maximum running time of AutoML to 300s, that is, the max_runtime_secs parameter is set to 300s;

[0080] Adjust the maximum number of models that AutoML explores before stopping to 10, that is, the max_models parameter is set to 10;

[0081] Provide a random seed for the AutoML running process to ensure the repeatability of the experiment. The seed number is 1234, that is, the seed parameter is set to 1234;

[0082] The distribution function supported by each machine learning algorithm is selected as AUTO, that is, the distribution parameter is set to AUTO;

[0083] The number of cross-validation folds is set to 5, which helps to evaluate the stability and generalization ability of the model, that is, the nfolds parameter is set to 5;

[0084] The training stopping condition, stopping_rounds, is set to 3, which means that if the model performance does not improve within the specified number of rounds, the training process stops.

[0085] The keep_cross_validation_models parameter is set to True to keep the cross-validated model.

[0086] In the embodiment of the present invention, further, during the model training process, the indicators used to evaluate the model performance are mean square error (RMSE), mean absolute error (MAE) or determination coefficient (r 2 ).

[0087] The mean square error (RMSE) measures the average of the squares of the differences between the predicted values ​​and the true values ​​in regression tasks, and is used to evaluate the accuracy of model predictions.

[0088] The mean absolute error (MAE) is the average of the absolute values ​​of the deviations of all individual observations from the true value (or arithmetic mean).

[0089] Coefficient of determination r 2 That is, the degree to which the explanatory variable (independent variable) explains the dependent variable, which is between 0 and 1. The larger the value, the better the model fit.

[0090] In the method of the present invention, mean square error, mean absolute error and coefficient of determination are selected to evaluate the performance of each model on the test set.

[0091] In a further embodiment, the step of optimizing hyperparameters of the optimal model includes:

[0092] Automatically tune model parameters using the H2O AutoML platform’s grid search or random search techniques.

[0093] In a further embodiment, the method further comprises the following steps:

[0094] The model interpretation function of the H2O AutoML platform was used to analyze the key factors and their weights that affect the algae removal efficiency.

[0095] In a further embodiment, the model interpretation methods used include residual analysis, variable importance analysis, Shapley value interpretation method and partial dependence curve, which are used to interpret the prediction results of the optimal model obtained by training the H2O AutoML platform.

[0096] Among them, residual analysis refers to the difference between the actual observed value and the model predicted value. The information provided by the residual can be used to analyze the reliability, periodicity or other interference of the data.

[0097] Variable importance analysis is a method to measure the influence of each input feature in the model on the prediction results.

[0098] The Shapley value interpretation method measures the average contribution of a single feature to the model prediction when taking into account its interactions with other features.

[0099] Partial dependence curve analysis shows how characteristic variables affect model predictions by calculating the marginal effect of one (or two) input parameters on the prediction model.

[0100] In a further embodiment, the method further comprises the following steps:

[0101] Water quality data, actual flocculation flotation operation parameters and algae removal effects are collected according to the preset cycle, the flocculation flotation algae removal process optimization control model is dynamically trained and updated, and the updated flocculation flotation algae removal process optimization control model is redeployed.

[0102] {Example 1}

[0103] In order to further illustrate the implementation of the aforementioned method of the present invention, we will combine the following examples to Figure 2-9 As shown, the specific implementation manner is further described in detail.

[0104] Step 1: Obtain 400 historical data through laboratory experimental results, including historical water quality data, flocculation flotation process parameters, and algae treatment data.

[0105] Water quality data: turbidity, algae species and abundance, pH, water temperature, conductivity, and dissolved oxygen.

[0106] Flocculation flotation process parameters: flotation time, coagulant dosage and stirring intensity.

[0107] Algae treatment data: effluent algae species and quantity, turbidity, and physical and chemical indicators of algae, including soluble and bound extracellular polymers and algal cell Zeta potential.

[0108] Step 2: Data exploration, import the seaborn package in the Python script, call the corr and pairplot functions to view the original data set distribution and the correlation between the data, such as Figure 2 and Figure 3 shown.

[0109] Specifically, download the relevant packages (pandas and matplotlib) required by seaborn through pip and configure them in the conda environment. Read CSV type data through the pandas package and build a DataFrame table data structure matrix. Use the pairplot function to perform data exploratory analysis with the parameters set to kind = "scatter" and diag_kind = "kde". Then use the corr function to calculate the Pearson correlation coefficient between all numerical columns in the DataFrame. Finally, use the show and save functions of matplotlib to display and save the image.

[0110] Step 3: Initialize the H2O AutoML platform, input the data read in step 2, split the data into training set and validation set at a ratio of 0.8:0.2, and then specify the target column for prediction (cyanobacteria removal rate) and the feature column for training (including the water quality parameters, flocculation flotation process parameters and algae treatment data mentioned in step 1).

[0111] The H2O AutoML platform automatically trains multiple models. After training, the test set is used to establish a flocculation flotation algae removal process prediction model and obtain a comparison of the performance of multiple models, such as Figure 4 shown.

[0112] Specifically, pip is used to download and automatically install the architecture required for training the model in the conda virtual environment; H2O AutoML is initialized through the init function in the python script; the upload_file function reads the data and splits the data set with the ratio of the training set to the validation set being 0.8:0.2, i.e. ratios = [0.8, 0.2]; train.columns extracts and specifies the feature columns used for training: the H2OAutoML function is used to start building the model, and finally a five-fold cross validation is used to ensure the reliability of the model.

[0113] Among them, the parameter max_runtime_secs of the H2O AutoML function is set to 300s, and the maximum running time of AutoML is 300s; max_models is set to 10, which is used to adjust the maximum number of models that AutoML can explore before stopping; seed is set to 1234, which provides a random seed for the AutoML running process to ensure the repeatability of model building; the distribution function supported by various algorithms is selected as Auto, that is, distribution is set to AUTO; the number of cross-validation folds is set to 5, that is, nfolds is set to 5; stopping_rounds is set to 3; keep_cross_validation_models is set to True, and the cross-validation model is retained.

[0114] In this embodiment, the mean square error (Formula 1), the mean absolute error (Formula 2) and the determination coefficient (Formula 3) are selected to evaluate the performance of each trained model on the test set.

[0115] The best model trained is Stack_1, with a mean square error of 0.05164, a mean absolute error of 0.05308, and a coefficient of determination of 0.972.

[0116]

[0117] Among them, SS res(Residual Sum of Squares) is the residual variation, that is, the sum of the squares of the differences between the observed and predicted values; SS tot (Total Sum of Squares) is the total variation, that is, the sum of the squares of the differences between the observations and the mean of the observations.

[0118] Step 4: After the training in step 3 is completed, the best performing model results obtained by training on the H2O AutoML platform are explained through comprehensive interpretation methods such as residual analysis, learning curves, variable importance, Shapley explanation (SHAP) summary, and partial dependence plot (PDP).

[0119] Specifically, the aforementioned interpretable method is performed directly in the Python script.

[0120] Among them, the residual analysis ( Figure 5 ) shows the residual analysis showing the residual distribution of the test dataset in the integrated stacking model. The frequency of residuals near zero is high, especially for higher algae removal rates (60%-100%), indicating that the trained integrated stacking model has obtained sufficient prediction information.

[0121] Among them, the learning curve of the integrated stacking model ( Figure 6 ) shows that in the first 20 iterations, the three curves, including the training curve, the test curve, and the cross-validation curve, dropped rapidly, indicating that the initial learning rate was fast and the model performance improved rapidly. The interval between the training curve and the cross-validation curve was small, indicating that the model did not have obvious overfitting. The performance of the model on the test curve was not much different from that on the training curve, indicating that the model has good generalization ability.

[0122] Among them, the variable importance of the optimal model ( Figure 7 ) evaluated the importance of each input variable on the algae removal rate and found that the order was flotation time > dosage > bEPS > DO > algae density > pH > turbidity > Zeta potential > dEPS. The results showed that the flotation time had the most important effect on the algae removal rate.

[0123] Among them, the SHAP summary of the optimal model ( Figure 8) to further evaluate the marginal effect of each input variable on the algae removal rate. The SHAP values ​​of 9 numerical variables, including flotation time, dosage, bEPS, DO, algae density, pH, turbidity, Zeta potential, and dEPS, were obtained in the test data set, and the SHAP values ​​varied within a certain range. The different distributions of similar variable values ​​in horizontal positions indicate that the effect of the input variable is not only determined by its variable value, but also affected by other variables. The large distribution of flotation time in the SHAP summary diagram proves its importance. At the same time, the low variable points of flotation time, dosage and bEPS are mainly distributed on the left, while the high variable points are mainly distributed on the right. These results further indicate that longer flotation time and higher dosage and bEPS are beneficial to the removal of algae.

[0124] Among them, through the partial dependency graph ( Fig. 9 ) visualizes the dependency between the most important input variables (flotation time, dosage, bEPS, and DO represented by four figures) and the algae removal rate. PDP is drawn by changing the value of the variable of interest and keeping the other variables fixed. The curves of different colors in the figure represent the results obtained by different models. For the flotation time, it can be seen that the dependency increases significantly in the first 40-60s, and then the slope gradually decreases. Based on this, the flotation time for algae removal can be significantly optimized. For the dosage, it can be seen that before 20 mg / L, the dependency increases significantly with the increase in dosage, and the subsequent dosage increases more slowly. The results show that after reaching a certain dosage, the dosage has little effect on the flocculation-flotation algae removal efficiency. For bEPS, it can be seen that the dependency increases significantly after 7.8 mg / L, indicating that high concentrations of bEPS are beneficial to the flocculation-flotation process. For DO, the dependency is strongest at low and high concentrations.

[0125] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. A person with ordinary knowledge in the technical field to which the present invention belongs may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the definition of the claims.

Claims

1. A flocculation flotation algae removal process optimization method based on machine learning, characterized in that: include: Obtain historical data on flocculation and flotation algae removal, including historical water quality data, flocculation and flotation process parameters, and algae treatment data, and obtain a training data set after preprocessing the historical data; Divide the training data set and determine the training set and test set according to the preset ratio; The H2O AutoML platform automatically trains multiple machine learning algorithms using the above training and test sets, and selects the best model based on pre-defined evaluation metrics; The hyperparameters of the optimal model were optimized to obtain the optimal control model of flocculation and flotation algae removal process; Deploy the flocculation and flotation algae removal process optimization control model into the control system of the water treatment facility, dynamically predict the flocculation and flotation operation parameters based on the real-time collected water quality data, and adjust the actual flocculant dosage, flocculation conditions and flotation conditions; and The algae removal effect was evaluated based on the actual algae removal treatment using the actual flocculation flotation operating parameters.

2. The flocculation flotation algae removal process optimization method based on machine learning according to claim 1 is characterized in that: The historical data of flocculation and flotation algae removal treatment include: Historical water quality data: turbidity, algae species and quantity, pH, water temperature, conductivity; Flocculation flotation process parameters: flotation time, coagulant dosage and stirring intensity; Algae treatment data: effluent algae species and quantity, turbidity, and physical and chemical indicators of algae, including soluble and bound extracellular polymers and algal cell Zeta potential.

3. The flocculation flotation algae removal process optimization method based on machine learning according to claim 1 is characterized in that: The machine learning algorithms executed on the H2O AutoML platform include GLM random forest, DRF distributed random forest, XRT extreme random forest, DeepLearning deep learning, XGboost, GBM gradient boosting machine and Stack integrated stacking model, and the optimal model is selected based on predetermined evaluation indicators.

4. The flocculation flotation algae removal process optimization method based on machine learning according to claim 1 is characterized in that: The H2O AutoML platform automates the training process for multiple machine learning algorithms including: Select the variable of the model prediction output, that is, set the response_column value; Set the maximum running time of AutoML to 300s, that is, the max_runtime_secs parameter is set to 300s; Adjust the maximum number of models that AutoML explores before stopping to 10, that is, the max_models parameter is set to 10; Provide a random seed for the AutoML running process to ensure the repeatability of the experiment. The seed number is 1234, that is, the seed parameter is set to 1234; The distribution function supported by each machine learning algorithm is selected as AUTO, that is, the distribution parameter is set to AUTO; The number of cross-validation folds is set to 5, which helps to evaluate the stability and generalization ability of the model, that is, the nfolds parameter is set to 5; The training stopping condition, stopping_rounds, is set to 3, which means that if the model performance does not improve within the specified number of rounds, the training process stops. The keep_cross_validation_models parameter is set to True to keep the cross-validated model.

5. The flocculation flotation algae removal process optimization method based on machine learning according to claim 1 is characterized in that: The hyperparameter optimization of the optimal model includes: Automatically tune model parameters using the H2O AutoML platform’s grid search or random search techniques.

6. The flocculation flotation algae removal process optimization method based on machine learning according to claim 1 is characterized in that: The method further comprises the following steps: The model interpretation function of the H2O AutoML platform was used to analyze the key factors and their weights that affect the algae removal efficiency.

7. The flocculation flotation algae removal process optimization method based on machine learning according to claim 6 is characterized in that: The model interpretation methods used included residual analysis, variable importance analysis, Shapley interpretation method, and partial dependence curves to interpret the prediction results of the optimal model trained by the H2O AutoML platform.

8. The method for optimizing the flocculation and flotation algae removal process based on machine learning according to any one of claims 1 to 7, characterized in that: The method further comprises the following steps: Water quality data, actual flocculation flotation operation parameters and algae removal effects are collected according to the preset cycle, the flocculation flotation algae removal process optimization control model is dynamically trained and updated, and the updated flocculation flotation algae removal process optimization control model is redeployed.

Citation Information

Patent Citations

  • Parameter setting method and device, electronic device and storage medium

    CN113822388A

  • Automatic coagulation dosing method and system based on machine vision and deep learning

    CN115925076A

  • Method and device for training soft measurement model

    CN117952228A

Cited By

  • Microalgae floating bead flotation harvesting efficiency prediction and optimization method based on machine learning

    CN119397212A

  • Prediction and optimization method of microalgae flotation recovery efficiency based on machine learning

    CN119397212B

  • Intelligent scum degassing device and process based on image online identification

    CN119841384A