Method for preparing acetic acid by sludge catalytic wet oxidation based on machine learning

Abstract

The present application relates to the technical field of sludge treatment, and particularly relates to a method for preparing acetic acid from sludge by catalytic wet oxidation based on machine learning. The method comprises data acquisition and preprocessing, machine learning modeling training, intelligent optimization, wet oxidation reaction, online regulation iteration; by constructing a high-precision regression model, combining SHAP interpretable analysis and GA / PSO optimization, the optimal working condition is determined; the sludge is directionally converted into acetic acid by catalytic wet oxidation, realizing high yield, high selectivity and low energy consumption. The core of the present application is to replace traditional experience debugging with data driving, adapt to the heterogeneity of sludge composition and the fluctuation of working condition, solve the problems of traditional process experience dependence, low yield, high energy consumption and poor adaptability, and is suitable for the resource utilization of municipal / industrial sludge, the acetic acid yield is increased by 15% to 25%, the energy consumption is reduced by 10% to 20%, and has significant engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of sludge treatment technology, and in particular relates to a method for preparing acetic acid by catalytic wet oxidation of sludge based on machine learning. Background Technology

[0002] Municipal and industrial sludge are produced in large quantities and have high organic matter content. Traditional landfilling and incineration can easily cause secondary pollution and resource waste. Wet oxidation (WAO) / catalytic wet oxidation (CWO) can convert proteins and polysaccharides in sludge into volatile fatty acids (VFAs) such as acetic acid, achieving both volume reduction and high-value utilization. Existing processes rely on experience-based adjustments, resulting in problems such as low acetic acid selectivity, numerous byproducts, high energy consumption, and poor adaptability to operating condition fluctuations. The heterogeneity of sludge composition and strong coupling of multiple parameters make it difficult for traditional mechanistic models to accurately predict and optimize. Existing technologies have not deeply integrated machine learning with the entire process of wet oxidation of sludge to produce acetic acid, lacking a data-driven modeling, feature importance quantification, multi-objective intelligent optimization, and online adaptive control integrated solution, making it difficult to achieve stable and efficient acid production. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing wet oxidation processes for producing acetic acid from sludge, such as low yield, poor selectivity, high energy consumption, and weak adaptability to operating conditions. This invention provides a high-precision, low-energy-consumption, and robust machine learning-driven preparation method to achieve high yield, high selectivity, and stable operation of acetic acid.

[0004] The objective of this invention is achieved through the following technical solution: This invention provides a method for preparing acetic acid by catalytic wet oxidation of sludge based on machine learning, characterized by comprising the following steps: Step 1: Collect sludge characteristic parameters, wet oxidation process parameters, and reaction product indicators to construct a dataset, and perform preprocessing and feature screening; Step 2: Build a machine learning regression model based on the preprocessed data, complete model training, validation and testing, and perform feature importance analysis using SHAP values; Step 3: Using the highest acetic acid yield or the lowest acetic acid yield and energy consumption as the optimization objective, an intelligent optimization algorithm is used to solve for the optimal combination of process parameters; Step 4: Perform catalytic wet oxidation of sludge according to the optimal process parameters. After the reaction is completed, separate and purify the product to obtain acetic acid. Step 5: Based on real-time collected operational data, iteratively update the model to achieve adaptive control of process parameters in the preparation process.

[0005] In this invention, the sludge characteristic parameters in step one include moisture content, total solids concentration, volatile solids concentration, and total chemical oxygen demand; the process parameters include reaction temperature, reaction time, oxygen partial pressure, stirring rate, pH value, and moisture content; and the reaction product indicators include acetic acid yield and the proportion of acetic acid in organic acids.

[0006] In this invention, the machine learning regression model described in step two is selected from at least one of XGBoost, CatBoost, Random Forest, Support Vector Regression (SVR), and Artificial Neural Network. The training / validation / test sets are divided in a 7:2:1 ratio, and the hyperparameters are optimized using 5-fold cross-validation. The SHAP value is combined to quantify the feature contribution, thereby making the model interpretable.

[0007] In this invention, the intelligent optimization algorithm mentioned in step three is either a genetic algorithm (GA) or a particle swarm optimization algorithm (PSO).

[0008] In this invention, the catalytic wet oxidation process conditions in step four are: temperature 240-280℃, oxygen partial pressure 1.0-1.5 MPa, pH 6.0-9.0, and reaction time 30-90 min; the catalyst is prepared in situ using sludge, and the preparation conditions are: sludge carbon catalyst is generated in situ by hydrothermal carbonization at 220-260℃, 1-2 MPa O2, and 2-4 h.

[0009] In this invention, the acetic acid prepared in step four yields ≥200mg of acetic acid per 1.0g of volatile solids, and the proportion of acetic acid in organic acids is ≥75%.

[0010] In this invention, step five involves using online sensors to acquire real-time data on sludge moisture content, total solids concentration, volatile solids concentration, total chemical oxygen demand, and operating conditions. This allows for dynamic model correction and automatic adjustment of process parameters to adapt to fluctuations in sludge composition.

[0011] In this invention, in step five, the data acquisition module collects the operating parameters of the wet oxidation reaction module in real time; the database and preprocessing module cleans the raw data and outputs a standardized dataset to the machine learning modeling and optimization module, while simultaneously storing historical data for model iteration; the machine learning modeling and optimization module trains and predicts the optimal reaction conditions based on the standardized data; the wet oxidation reaction module receives and executes control commands for data such as temperature, pressure, residence time, and oxidant addition; and the online monitoring and control module receives model optimization commands and implements linkage.

[0012] In this invention, the machine learning modeling and optimization module trains a predictive optimization model based on standardized data, outputs the optimal response conditions, sends instructions to the online monitoring and control module, and simultaneously receives online monitoring data to complete model iteration.

[0013] This invention utilizes a high-precision, low-energy-consumption, and robust machine learning-driven preparation method to achieve high yield, high selectivity, and stable operation of acetic acid. It is highly practical and possesses novelty and inventiveness. Its advantages lie in its ability to analyze the multi-parameter nonlinear correlation of sludge wet oxidation to acetic acid through machine learning modeling, thereby achieving accurate prediction of process conditions, improving acetic acid selectivity and yield, and reducing energy consumption and by-product generation. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of a machine learning-based method for preparing acetic acid by catalytic wet oxidation of sludge. Detailed Implementation

[0015] The present invention will now be described in detail with reference to specific embodiments, but these are by no means limitations on the present invention. Example 1

[0016] Using municipal sludge from urban wastewater treatment plants as raw material, an experiment was conducted to prepare acetic acid by wet oxidation of sludge according to the method of this invention. The specific implementation steps are as follows: Step 1: Data Acquisition, Preprocessing, and Feature Selection Parameter Acquisition: Sludge characteristic parameters, wet oxidation process parameters, and reaction product indicators were collected to construct the original dataset. Basic sludge characteristics: Moisture content 95%, Total Solids (TS) 45000 mg / L, Volatile Solids (VS) 26000 mg / L, Total Chemical Oxygen Demand (TCOD) 300000 mg / L; Process parameters: Reaction temperature, reaction time, oxygen partial pressure, stirring rate, pH value, moisture content; Reaction product indicators: Acetic acid yield, percentage of acetic acid in total organic acids. This embodiment collected a total of 120 sets of valid operating condition data. Data Preprocessing: Missing values ​​were imputed using the mean, outliers were removed using the 3σ principle, and Z-score normalization was performed on the original data to eliminate the influence of dimensionality and data bias. Feature Screening: Through Pearson correlation coefficient and variance analysis, feature parameters strongly correlated with acetic acid production were screened: reaction temperature, reaction time, oxygen partial pressure, stirring rate, pH value, and moisture content.

[0017] Step 2: Machine Learning Regression Model Construction, Training, and SHAP Interpretability Analysis Dataset partitioning: Strictly partitioned according to a 7:2:1 ratio, with 84 training sets, 24 validation sets, and 12 test sets. Model training and hyperparameter optimization: A CatBoost regression model was selected, with acetic acid yield as the dependent variable and 6 selection features as independent variables. A 5-fold cross-validation approach was used, minimizing the prediction error and automatically locking in the optimal hyperparameters by adjusting the learning rate (0.05~0.2), tree depth (3~7), and number of iterations (100~500). Model performance: After training, the model's performance on the test set was evaluated using the coefficient of determination R0.2 =0.95, RMSE=4.2, excellent generalization ability, can accurately predict acetic acid yield. SHAP feature importance analysis: quantifying the contribution of each parameter to acetic acid production, ranked as follows: reaction temperature (36.2%) > oxygen partial pressure (22.7%) > pH value (18.1%) > reaction time (12.3%) > stirring rate / water content (10.7%).

[0018] Step 3: Intelligent optimization algorithm for optimization With the acetic acid yield as the optimization objective, a trained CatBoost prediction model was coupled, and a particle swarm optimization (PSO) algorithm was used for global optimization. The constraints matched the process range of claim 5: temperature 240–280℃, oxygen partial pressure 1.0–1.5 MPa, pH 6.0–9.0, and reaction time 30–90 min. The optimal combination of process parameters was: reaction temperature 280℃, oxygen partial pressure 1.3 MPa, pH=8.5, reaction time 60 min, stirring rate 300 r / min, and moisture content 95%. Step 4: Separation and purification of acetic acid via catalytic wet oxidation of sludge. Municipal sludge was fed into an intermittent wet oxidation reactor, and a catalyst was prepared by in-situ hydrothermal reaction of the sludge. The preparation conditions were as follows: sludge carbon catalyst was generated in situ by hydrothermal carbonization at 220-260℃, 1-2 MPa O2, and 2-4 h. The reaction was carried out according to the optimal parameters obtained by machine learning: after feeding, the temperature of the sealed reactor was raised to 280℃; oxygen was introduced to maintain an oxygen partial pressure of 1.3 MPa, the pH of the feed solution was adjusted to 8.5, the stirring rate was 300 r / min, and the reaction was carried out at a constant temperature for 60 min. After the reaction was completed, the reactor was rapidly cooled and depressurized, and the feed solution was centrifuged, filtered through a membrane, and distilled under reduced pressure to obtain high-purity acetic acid product.

[0019] Step 5: Online Monitoring and Adaptive Model Iteration This embodiment integrates a data acquisition module, a database and preprocessing module, a machine learning modeling and optimization module, a wet oxidation reaction module, and an online monitoring and control module into a unified intelligent system. Online sensors collect real-time data on sludge moisture content, TS, VS, TCOD, operating temperature, oxygen partial pressure, and pH value. The data is transmitted back to the machine learning module in real time for incremental training and iterative updates using the CatBoost model (similar to other models such as XGBoost, random forest, support vector regression SVR, and artificial neural networks). The system dynamically and automatically corrects process parameters based on fluctuations in sludge composition, achieving adaptive and stable control throughout the entire process.

[0020] Final product specifications: Acetic acid yield: 260 mg / (g VS) (260 mg acetic acid is produced per 1.0 g volatile solids); Acetic acid content in total organic acids: 78%; High product purity and excellent resource utilization efficiency.

[0021] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for preparing acetic acid by catalytic wet oxidation of sludge based on machine learning, characterized in that, Includes the following steps: Step 1: Collect sludge characteristic parameters, wet oxidation process parameters, and reaction product indicators to construct a dataset, and perform preprocessing and feature screening; Step 2: Build a machine learning regression model based on the preprocessed data, complete model training, validation and testing, and perform feature importance analysis using SHAP values; Step 3: Using the highest acetic acid yield or the lowest acetic acid yield and energy consumption as the optimization objective, an intelligent optimization algorithm is used to solve for the optimal combination of process parameters; Step 4: Perform catalytic wet oxidation of sludge according to the optimal process parameters. After the reaction is completed, separate and purify the product to obtain acetic acid. Step 5: Based on real-time collected operational data, iteratively update the model to achieve adaptive control of process parameters in the preparation process.

2. The method according to claim 1, characterized in that, The sludge characteristic parameters mentioned in step one include moisture content, total solids concentration, volatile solids concentration, and total chemical oxygen demand; the process parameters include reaction temperature, reaction time, oxygen partial pressure, stirring rate, pH value, and moisture content; the reaction product indicators include acetic acid yield and the proportion of acetic acid in organic acids.

3. The method according to claim 1, characterized in that, The machine learning regression model described in step two is selected from at least one of XGBoost, CatBoost, Random Forest, Support Vector Regression (SVR), and Artificial Neural Network. The training / validation / test sets are divided in a 7:2:1 ratio, and the hyperparameters are optimized using 5-fold cross-validation. The SHAP value is combined to quantify the feature contribution and make the model interpretable.

4. The method according to claim 1, characterized in that, The intelligent optimization algorithm mentioned in step three is either the Genetic Algorithm (GA) or the Particle Swarm Optimization (PSO) algorithm.

5. The method according to claim 1, characterized in that, The catalytic wet oxidation process conditions in step four are: temperature 240-280℃, oxygen partial pressure 1.0-1.5 MPa, pH 6.0-9.0, and reaction time 30-90 min. The catalyst is prepared in situ using sludge, and the preparation conditions are: sludge carbon catalyst is generated in situ by hydrothermal carbonization at 220-260℃, 1-2 MPa O2, and 2-4 h.

6. The method according to claim 1, characterized in that, The acetic acid prepared in step four yields ≥200 mg of acetic acid per 1.0 g of volatile solids, and the proportion of acetic acid in the organic acid is ≥75%.

7. The method according to claim 1, characterized in that, In step five, online sensors are used to acquire real-time data on sludge moisture content, total solids concentration, volatile solids concentration, total chemical oxygen demand, and operating conditions. The model is then dynamically corrected and process parameters are automatically adjusted to adapt to fluctuations in sludge composition.

8. The method according to claim 1, characterized in that, In step five, the data acquisition module collects the operating parameters of the wet oxidation reaction module in real time. The database and preprocessing module cleans the raw data and outputs a standardized dataset to the machine learning modeling and optimization module. At the same time, historical data is stored back for model iteration. The machine learning modeling and optimization module trains and predicts the optimal reaction conditions based on the standardized data. The wet oxidation reaction module receives and executes control commands for data such as temperature, pressure, residence time, and oxidant addition. The online monitoring and control module receives model optimization instructions and implements linkage.

9. The method according to claim 8, characterized in that, The machine learning modeling and optimization module trains and predicts the optimal response conditions based on standardized data, sends instructions to the online monitoring and control module, and simultaneously receives online monitoring data to complete model iteration.

Images