Water body bacterium ultraviolet disinfection dosage prediction and optimization method driven by machine learning

Abstract

The invention discloses a machine learning-driven water body bacteria ultraviolet disinfection dosage prediction and optimization method, which comprises the following steps of: obtaining genetic information, environmental parameters and ultraviolet disinfection conditions of target bacteria, and numeralizing the genetic information into feature vectors; cleaning and coding the acquired data to form a feature matrix; based on the characteristic matrix, a machine learning algorithm is utilized to train an ultraviolet disinfection dosage prediction model, and the model takes the inactivation efficiency as a prediction target; according to bacterial genetic information, environmental parameters and ultraviolet conditions input by a user, utilizing the trained prediction model to predict the inactivation efficiency, scanning in a multi-dimensional parameter space, and determining an optimal ultraviolet disinfection condition combination meeting a target inactivation level; outputting an optimal ultraviolet disinfection condition combination; the method disclosed by the invention has cross-strain and cross-scene applicability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water treatment and environmental engineering, and particularly relates to a machine learning driven water bacterial ultraviolet disinfection dose prediction and optimization method. BACKGROUND

[0002] As an efficient and chemical residue-free water treatment technology, ultraviolet disinfection has been widely used in drinking water treatment, reclaimed water utilization and wastewater advanced treatment. Its principle is to use ultraviolet light of a specific wavelength to destroy the DNA or RNA structure of microorganisms, thereby achieving the purpose of killing pathogens. Compared with traditional chlorine disinfection, ultraviolet disinfection has the advantages of avoiding the generation of harmful disinfection by-products, rapid reaction, simple operation, etc., and therefore has become an important development direction for water plant upgrading in recent years.

[0003] However, most existing ultraviolet disinfection models rely on empirical formulas or simplified kinetic models, such as single-phase linear models, two-phase models or empirical fitting curves. Although these models can describe the disinfection dose and inactivation effect under certain conditions, they often fail to accurately reflect the shoulder effect, tailing effect and differences in genetic level of different strains. Therefore, under complex water quality conditions or for different pathogenic microorganisms, the applicability and accuracy of traditional models are obviously insufficient.

[0004] In recent years, some studies have attempted to use machine learning methods to model ultraviolet disinfection. Although these methods have improved the prediction accuracy to some extent, most existing studies only consider physical and chemical conditions (such as ultraviolet wavelength, dose, pH, etc.), ignoring the influence of microbial genetic characteristics on disinfection sensitivity. In fact, the genomic composition (such as AT content, k-mer characteristics, etc.) between different strains is closely related to their ultraviolet sensitivity, which provides a new research idea for further improving prediction ability.

[0005] Therefore, the existing technology still has the following shortcomings: (1) there is a lack of prediction methods that can consider both environmental conditions and microbial genetic characteristics; (2) water plants lack fine optimization tools based on microbial information in operation, making it difficult to achieve truly intelligent disinfection decisions. SUMMARY

[0006] The purpose of the present application is to provide a machine learning driven water bacterial ultraviolet disinfection dose prediction and optimization method, which solves the problem of relying only on empirical formulas or physical and chemical conditions in existing ultraviolet disinfection models.

[0007] Technical scheme: The machine learning driven water bacterial ultraviolet disinfection dose prediction and optimization method provided by the present application comprises the following steps:

[0008] (1) obtaining genetic information, environmental parameters and ultraviolet disinfection conditions of target bacteria, and numerically converting the genetic information into a feature vector;

[0009] (2) cleaning and encoding the obtained data to form a feature matrix;

[0010] (3) based on the feature matrix, training an ultraviolet disinfection dose prediction model using a machine learning algorithm, and the model takes inactivation efficiency as the prediction target;

[0011] (3) according to the user input of the bacterial genetic information, environmental parameters and ultraviolet conditions, using the trained prediction model to predict the inactivation efficiency, and scanning in the multi-dimensional parameter space to determine the optimal ultraviolet disinfection condition combination that meets the target inactivation level;

[0012] (4) output the optimal ultraviolet disinfection condition combination.

[0013] Further, the genetic information includes at least one of the 16S rRNA sequence of the bacteria, the genomic AT base pair content, the spore-forming ability and the gram staining attribute.

[0014] Further, the numerical conversion of the genetic information includes extracting the k-mer feature of the 16S rRNA sequence, wherein k is an integer greater than or equal to 2.

[0015] Further, the environmental parameters include at least one of pH value, temperature and initial bacterial concentration.

[0016] Further, the ultraviolet disinfection conditions include ultraviolet wavelength and ultraviolet dose range.

[0017] Further, the data preprocessing includes filling missing values, removing outliers, and one-hot encoding of categorical variables.

[0018] Further, the input features are screened using a feature selection method, and the hyperparameters of the machine learning model are optimized using a hyperparameter optimization method.

[0019] Further, the machine learning algorithm includes at least one of an ensemble learning algorithm, a decision tree algorithm, a support vector machine algorithm or a neural network algorithm.

[0020] Further, the multi-dimensional parameter space includes a three-dimensional space of ultraviolet dose, ultraviolet wavelength and pH value, and the optimal condition combination corresponding to the minimum effective dose is determined by scanning the three-dimensional space.

[0021] The machine learning driven ultraviolet disinfection dose prediction and optimization system for water body bacteria provided by the application comprises:

[0022] Data acquisition module: for acquiring genetic information of target bacteria, environmental parameters and ultraviolet disinfection conditions, and numerically converting the genetic information into a feature vector;

[0023] Data preprocessing module: for cleaning and encoding the acquired data to form a feature matrix;

[0024] Model construction module: for training an ultraviolet disinfection dose prediction model based on the feature matrix using a machine learning algorithm, with inactivation efficiency as the prediction target;

[0025] Dose prediction and optimization module: for predicting inactivation efficiency using the trained prediction model according to user input of bacterial genetic information, environmental parameters and ultraviolet conditions, and scanning in a multi-dimensional parameter space to determine the optimal ultraviolet disinfection condition combination that meets the target inactivation level;

[0026] Result output module: for outputting the optimal ultraviolet disinfection condition combination to guide the operation of the ultraviolet disinfection equipment.

[0027] Advantages: Compared with the prior art, the present application has the following significant advantages: By introducing genetic information such as 16S rRNA sequence or whole genome features, and combining with a machine learning model, the present application can more accurately predict the inactivation efficiency of different microorganisms under different ultraviolet doses, wavelengths and water quality conditions. Automated scanning and calculation in a three-dimensional parameter space (dose-wavelength-pH) can quickly search for the optimal condition combination that meets the target inactivation level (such as 4 Log removal), greatly reducing the experimental screening time. The optimal disinfection conditions obtained through model optimization can avoid excessive ultraviolet dose input, reduce energy consumption and operating costs, while ensuring stable disinfection effect. The present application is not only suitable for common water body microorganisms, but also can be used for prediction and optimization according to different microbial genetic information, with cross-species and cross-scene applicability. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flowchart of the present application. DETAILED DESCRIPTION

[0029] The technical solutions of the present application will be further described below in conjunction with the drawings.

[0030] As shown in Figure 1 , the present application provides a machine learning driven water body bacterial ultraviolet disinfection dose prediction and optimization method, comprising the following steps:

[0031] Step 1: Data acquisition and input

[0032] The system first collects and numerical processes the genetic information of target bacteria. The genetic information comes from public databases or literature reports, including but not limited to: 16S rRNA sequence (complete base sequence), genome AT base pair content (range about 40%-65%), sporulation ability (represented by binary features), and gram staining properties. For 16S rRNA sequence, the system uses k-mer feature extraction method, where 3-mer (all 64 combinations of three nucleotides) is used as the main feature unit, combined with sequence length and di-nucleotide (2-mer) frequency, to achieve comprehensive numerical encoding of genetic information. After encoding, each strain of bacteria can be converted into an 84-dimensional genetic feature vector.

[0033] At the same time, the system collects and inputs experimental conditions and physical parameters. The ultraviolet radiation dose (Dose) ranges from 0 to 1305.8 mJ / cm², and the ultraviolet wavelength (Wavelength) ranges from 214 to 310 nm; the experimental solution environmental parameters include pH 7.0-7.4 (using PBS buffer), and the temperature is standardized to 25°C. In addition, the system also records the initial bacterial concentration (log CFU / mL) to reflect the baseline level of bacteria under experimental conditions.

[0034] Through the above fusion input of genetic features, physical features and environmental conditions, the system establishes a multi-dimensional input space containing 89 feature variables, providing a complete data basis for subsequent ultraviolet disinfection dose prediction models.

[0035] Step 2: Data cleaning and preprocessing

[0036] Firstly, for missing values, some literature only reports that the experiment was carried out at "room temperature" without giving the exact value. For this, the system standardizes "room temperature" to 25°C, filling in the missing temperature data. Secondly, for abnormal values and error data, the system uses statistical-based screening method. In each wavelength range, the distribution of ultraviolet dose is analyzed by interquartile range (IQR), and extreme values outside the normal range are removed. For example, the dose values reported in some literature are significantly higher than those under similar conditions, but their inactivation effect (residual bacterial concentration) is comparable to that of low-dose experiments, such as experimental errors and are deleted. Finally, for categorical variables (such as gram properties, sporulation ability), the system uses one-hot encoding to convert them into binary numerical vectors, making it easier to input models with continuous variables.

[0037] Step 3: Prediction model construction

[0038] Based on the processed feature matrix, the system constructs multiple machine learning models for training and prediction, including CatBoost, Random Forest, Support Vector Regression (SVR), Multilayer Perceptron (MLP), and ensemble methods such as Bagging and single decision tree (Decision Tree). The prediction target of each model is the inactivation efficiency of the target microorganism under different ultraviolet irradiation conditions, and the output index is the log reduction value (Log Reduction Value, LRV).

[0039] During modeling, the system first uses the Pearson correlation coefficient and Recursive Feature Elimination (RFE) to screen the input features, eliminating features with low correlation or high redundancy to the target variable, to reduce the feature dimension and improve the model training efficiency. Then, the system uses Bayesian Optimization with Random Search to automatically optimize the key hyperparameters of each model, thereby obtaining the prediction model with the best performance under the given input features.

[0040] Model performance evaluation uses the coefficient of determination (R²) and root mean square error (RMSE) as indicators to compare the prediction results of different algorithms on the validation set and finally determine the best prediction model. In this embodiment, the best-performing model is CatBoost, and its optimized key parameters include: learning rate (learning_rate): 0.084, tree depth (depth): 8, L2 regularization coefficient (l2_leaf_reg): 2.329, random strength (random_strength): 9.332, Bagging temperature (bagging_temperature): 0.700, border count (border_count): 53, growth policy (grow_policy): Lossguide, subsample rate (subsample): 0.902.

[0041] Step 4: Dose prediction and optimization application

[0042] After determining the best prediction model, the system uses the model to predict and optimize the ultraviolet disinfection efficiency of the target microorganism.

[0043] Firstly, the user inputs the genetic characteristics of the bacteria to be predicted (16S rRNA sequence, AT content, sporulation ability, etc.), as well as experimental conditions (initial bacterial concentration, temperature) and physical parameters (ultraviolet wavelength, pre-set dose range, pH value). The system numerically processes these input characteristics and sends them to the trained CatBoost model, which outputs the inactivation efficiency index of the target microorganism under the given conditions, i.e., the Log Reduction Value (LRV).

[0044] Secondly, the system conducts systematic scanning and calculation in the three-dimensional feature space (dose-wavelength-pH). By predicting within the continuous value intervals of dose range 0-1305.8 mJ / cm², wavelength range 214-310 nm, and pH 7.0-7.4, the system can generate a complete three-dimensional dose-response surface. This surface visually demonstrates the inactivation efficiency trend with dose changes under different wavelengths and pH environments.

[0045] Based on this three-dimensional surface, the system can automatically identify the minimum effective dose that meets specific inactivation targets (such as 3-log or 4-log removal) and output the optimal wavelength-pH-dose combination scheme. In cases where multiple conditions can achieve the target inactivation effect, the system preferentially recommends the combination with the lowest energy consumption and smallest dose requirement, thereby achieving optimized control of ultraviolet disinfection.

[0046] Step 5: Output of running results and application

[0047] The optimal parameters output by the system include the optimal ultraviolet wavelength, optimal dose, and suitable pH value, and can be directly applied to water plant ultraviolet disinfection equipment to achieve automated or semi-automated operation adjustment, thereby improving the stability and economy of the disinfection process.

Claims

1. A machine learning-driven method for predicting and optimizing ultraviolet disinfection dosage for bacteria in water, characterized in that, Includes the following steps: (1) Obtain the genetic information, environmental parameters and ultraviolet disinfection conditions of the target bacteria, and convert the genetic information into a feature vector; (2) The acquired data is cleaned and encoded to form a feature matrix; (3) Based on the feature matrix, a machine learning algorithm is used to train an ultraviolet disinfection dose prediction model, with inactivation efficiency as the prediction target. (3) Based on the bacterial genetic information, environmental parameters and ultraviolet conditions input by the user, the inactivation efficiency is predicted using the trained prediction model, and the optimal combination of ultraviolet disinfection conditions that meets the target inactivation level is determined by scanning in the multidimensional parameter space. (4) Output the optimal combination of ultraviolet disinfection conditions.

2. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... Genetic information includes at least one of the following: bacterial 16S rRNA sequence, genomic AT base pair content, sporulation capacity, and Gram staining properties.

3. The machine learning-driven method for predicting and optimizing ultraviolet disinfection dosage for bacteria in water, as described in claim 2, is characterized in that... Numericalization of genetic information includes extracting k-mer features from the 16S rRNA sequence, where k is an integer greater than or equal to 2.

4. The machine learning-driven method for predicting and optimizing ultraviolet disinfection dosage for bacteria in water, as described in claim 1, is characterized in that... Environmental parameters include at least one of pH, temperature, and initial bacterial concentration.

5. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... Ultraviolet disinfection conditions include the ultraviolet wavelength and ultraviolet dose range.

6. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... Data preprocessing includes imputing missing values, removing outliers, and one-hot encoding of categorical variables.

7. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... Feature selection is used to filter input features, and hyperparameter optimization is used to tune the hyperparameters of the machine learning model.

8. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... Machine learning algorithms include at least one of ensemble learning algorithms, decision tree algorithms, support vector machine algorithms, or neural network algorithms.

9. The machine learning-driven method for predicting and optimizing ultraviolet disinfection doses for bacteria in water, as described in claim 1, is characterized in that... The multidimensional parameter space includes a three-dimensional space of ultraviolet dose, ultraviolet wavelength, and pH value. The optimal combination of conditions corresponding to the minimum effective dose is determined by scanning the three-dimensional space.

10. A machine learning-driven system for predicting and optimizing ultraviolet disinfection dosage for bacteria in water, characterized in that, include: Data acquisition module: used to acquire the genetic information, environmental parameters and ultraviolet disinfection conditions of the target bacteria, and to convert the genetic information into a feature vector; Data preprocessing module: used to clean and encode the acquired data to form a feature matrix; Model building module: used to train a UV disinfection dose prediction model based on the feature matrix and using machine learning algorithms. The model uses inactivation efficiency as the prediction target. Dosage prediction and optimization module: Based on the bacterial genetic information, environmental parameters and UV conditions input by the user, it uses a trained prediction model to predict the inactivation efficiency and scans in a multi-dimensional parameter space to determine the optimal combination of UV disinfection conditions that meet the target inactivation level. The result output module is used to output the optimal combination of ultraviolet disinfection conditions to guide the operation of ultraviolet disinfection equipment.

Images