Method for predicting nh3 production in organic solid waste composting process based on metabolic kinetics-gru

Abstract

The application relates to the field of environmental science and technology, and aims to solve the problem of low NH3 prediction accuracy in the existing organic solid waste composting process.The application combines a kinetic model with a recurrent neural network GRU, thereby making up for the shortcomings of the mechanism model, such as fixed change mode and inaccuracy, and solving the problem of insufficient data of the mathematical model, so that a composting process NH3 production prediction model with small error and high accuracy is obtained, and the final goal of harmless treatment, reduction, resource utilization and stabilization of organic solid waste treatment is achieved.

Description

Technical Field

[0001] This invention relates to the field of environmental science and technology, specifically to a method for predicting NH3 generation during organic solid waste composting based on metabolic kinetics-GRU. Background Technology

[0002] With the rapid development of my country's economy, the continuous improvement of people's living standards, and the accelerating urbanization process, the output of solid waste has also increased dramatically, with organic solid waste accounting for 40-60% of the total domestic waste. This type of solid waste is characterized by high water content, easy decomposition, and rich nutrients. If not treated promptly, it can easily breed pathogenic microorganisms, spread pests and diseases, and pollute soil and water sources, posing a significant threat to environmental safety and human health. Composting technology, as one of the important ways to treat and utilize organic solid waste, has advantages such as low cost, high resource utilization rate, and harmless products, and has received widespread attention from scholars at home and abroad. However, this technology still has the problem of odor emissions, such as NH3, which leads to a large loss of nitrogen, causing environmental pollution and harm to the health of operators. Therefore, there is an urgent need to develop a model that can predict the amount of NH3 generated during the composting process.

[0003] Currently, most composting models are established using kinetic methods, which describe the composting process mechanism through first-order substrate degradation kinetics and Monod-type expressions, thus constructing a kinetic model. This method can effectively establish the relationship between macroscopic operating conditions and metabolic fluxes within the system, conforming to some mechanisms of the composting process. However, due to the extreme complexity of the composting process, the numerous related factors, and the fact that the reaction mechanisms have not been fully explored, it is difficult to use fixed formulas to represent the entire process and accurately establish mechanistic models for prediction.

[0004] Research shows that neural networks are a method for simulating biological nervous systems to estimate or approximate functions. They possess self-learning capabilities and can be applied to the construction of nonlinear process models. Among them, recurrent neural networks (RNNs) are widely used due to their ability to utilize internal memory to process time-series data. GRU, as an important branch of RNNs, effectively solves the long dependency problem in RNNs and has significant advantages in language recognition, time series prediction, machine translation, and handwriting recognition. Specifically, it reduces time costs and computational load by using reset and update gates to select and update memory, effectively solving the gradient problem in long-term memory and backpropagation. However, these neural networks require a large amount of data, and the limited amount of data, numerous influencing factors, and inconsistent data quality in the composting process significantly affect simulation accuracy. Summary of the Invention

[0005] The purpose of this invention is to address the problem of low accuracy in predicting NH3 generated during organic solid waste composting in existing technologies, and to propose a method for predicting NH3 generation during organic solid waste composting based on metabolic kinetics-GRU.

[0006] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0007] A method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU includes the following steps:

[0008] Step 1: Obtain compost data;

[0009] Step 2: Based on the acquired composting data, obtain the N-cycle path in the composting process;

[0010] Step 3: Based on the N cycle pathway in the composting process, obtain the key metabolic pathways for NH3 production, and then obtain the key metabolites contained in the key metabolic pathways.

[0011] Step 4: Based on the law of conservation of mass and the cell metabolic flux balance equation, the flux balance equation when intracellular metabolism is in a quasi-steady state is obtained.

[0012] Step 5: Obtain the metabolic flux vector r corresponding to different key metabolites based on the flux balance equation when intracellular metabolism is in a quasi-steady state;

[0013] Step 6: Substitute the different key metabolites and their corresponding metabolic flux vectors r into the flux balance equation when intracellular metabolism is in a quasi-steady state to obtain the final flux balance equation.

[0014] Step 7: Based on the law of conservation of mass, establish a system to represent the concentration changes of biomass, substrate consumption, and NH3 production in relation to metabolic flux, growth rate μ, and specific substrate consumption rate r. s , specific product formation rate r p A macroscopic dynamic model of the relationship between them;

[0015] Step 8: Combine the final flux balance equation with the macroscopic dynamic model to obtain the coupled dynamic model;

[0016] Step 9: Use a coupled kinetic model to simulate the full-cycle variation of NH3 concentration and fit the data. The fitted data includes biological concentration X, substrate concentration S, temperature T, and pH value. Finally, use biological concentration X, substrate concentration S, temperature T, and pH value as inputs and the NH3 concentration corresponding to the fitted data as output to train a recurrent neural network.

[0017] Step 10: Use the trained recurrent neural network to predict NH3 generation during the organic solid waste composting process.

[0018] Furthermore, the key metabolic pathways include oxidation, intracellular synthesis, nitrification, and denitrification.

[0019] Furthermore, the key metabolite includes NH4. + NO2 - NO3 - NO, N2O, ATP, H + And NADH.

[0020] Furthermore, the cellular metabolic flux balance equation is expressed as:

[0021]

[0022] Among them, S m Let m represent the metrology matrix, d represent the metabolite, and t represent the integral.

[0023] Furthermore, the flux balance equation for intracellular metabolism in a quasi-steady state is expressed as:

[0024] S m r = 0.

[0025] Furthermore, the flux balance equation is expressed as:

[0026]

[0027] Among them, NH4 + Indicates ammonium ion; NO2 + Indicates nitro cation; NO3 - Nitrate represents nitrate; NO represents nitric oxide; N₂O represents nitrous oxide; ATP represents adenosine triphosphate; H₂O represents nitrate ions ... + represents hydrogen ions; NADH represents reduced nicotinamide adenine dinucleotide; b represents the parameter to be identified; P / O represents the phosphorus-oxygen ratio.

[0028] Furthermore, the macroscopic dynamic model is expressed as:

[0029]

[0030]

[0031]

[0032] Among them, M p The relative molecular mass of the product, specifically the relative molecular mass of NH3, is M. p =17; M s M represents the relative molecular mass of the substrate. s=74, X represents biological concentration, μ represents specific growth rate, r p The product formation rate is represented by P, the NH3 production concentration is represented by S, and the substrate concentration is represented by S.

[0033] Furthermore, the recurrent neural network includes a data processing module, a network module, and a training and verification module;

[0034] The data processing module is used for dataset preprocessing, i.e., normalization; then the dataset is split according to the time step, and the resulting data is divided into training set, validation set and test set in a 7:2:1 ratio.

[0035] The network module is used to set the structure and related hyperparameters of the recurrent neural network. Specifically, the recurrent neural network has 2 layers, 4 input layers, 16 hidden layers, 1 output layer, and the activation function is tanh with a dropout rate of 0.3.

[0036] The training and validation module is used to train and validate the model. During training, the learning rate is 0.001, and backpropagation is used for training. The parameter used for validation is the coefficient of determination R0. 2 The algorithm includes Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and a loss function, SmoothL1Loss, with Adam as the optimizer.

[0037] Furthermore, the normalization process is performed using the Sklearn StandardScaler method.

[0038] Furthermore, the normalization process is expressed as follows:

[0039]

[0040] Where x represents all sample values ​​of the dataset, x* represents the standardized values ​​of all data, u represents the mean of each data category, and σ represents the standard deviation of each data category.

[0041] The beneficial effects of this invention are:

[0042] This application combines a coupled dynamics model with a recurrent neural network (GRU), which not only overcomes the shortcomings of fixed and inaccurate change patterns in mechanistic models, but also solves the problem of insufficient data in mathematical models. This results in a more accurate and less error-prone prediction model for NH3 production during composting, ultimately achieving the goals of harmless, reduced, resource-based, and stable treatment of organic solid waste. Specifically, it includes:

[0043] ① By combining the metabolic flux balance equation with macroscopic dynamics, a coupled metabolic dynamics model was constructed, which amplified the dataset, simulated the entire composting process, and provided data support for the establishment of neural networks;

[0044] ②Based on the temporal characteristics of the recurrent neural network GRU, the prediction accuracy is effectively improved;

[0045] ③ The final model fit is R. 2 =0.73, MAPE=35.43, RMSE=1.19, meeting the correlation fit requirements;

[0046] ④ The final model can be used to predict NH3 production during composting, enabling efficient treatment and resource utilization of organic solid waste. Attached Figure Description

[0047] Figure 1 This is the overall flowchart of this application;

[0048] Figure 2 A diagram of the coupled metabolic kinetics-GRU model;

[0049] Figure 3 This is a schematic diagram of a coupled metabolic kinetic model;

[0050] Figure 4 This is a comparison chart of the actual values ​​and the predicted results. Detailed Implementation

[0051] It should be noted that, where there is no conflict, the various embodiments disclosed in this application can be combined with each other.

[0052] Specific implementation method one: Refer to Figure 1 This embodiment specifically describes the NH3 generation prediction method for organic solid waste composting based on metabolic kinetics-GRU, which includes the following steps:

[0053] Step 1: Obtain compost data;

[0054] Step 2: Based on the acquired composting data, obtain the N-cycle path in the composting process;

[0055] Step 3: Based on the N cycle pathway in the composting process, obtain the key metabolic pathways for NH3 production, and then obtain the key metabolites contained in the key metabolic pathways.

[0056] Step 4: Based on the law of conservation of mass and the cell metabolic flux balance equation, the flux balance equation when intracellular metabolism is in a quasi-steady state is obtained.

[0057] Step 5: Obtain the metabolic flux vector r corresponding to different key metabolites based on the flux balance equation when intracellular metabolism is in a quasi-steady state;

[0058] Step 6: Substitute the different key metabolites and their corresponding metabolic flux vectors r into the flux balance equation when intracellular metabolism is in a quasi-steady state to obtain the final flux balance equation.

[0059] Step 7: Based on the law of conservation of mass, establish a system to represent the concentration changes of biomass, substrate consumption, and NH3 production in relation to metabolic flux, growth rate μ, and specific substrate consumption rate r. s , specific product formation rate r p A macroscopic dynamic model of the relationship between them;

[0060] Step 8: Combine the final flux balance equation with the macroscopic dynamic model to obtain the coupled dynamic model;

[0061] Step 9: Use a coupled kinetic model to simulate the full-cycle variation of NH3 concentration and fit the data. The fitted data includes biological concentration X, substrate concentration S, temperature T, and pH value. Finally, use biological concentration X, substrate concentration S, temperature T, and pH value as inputs and the NH3 concentration corresponding to the fitted data as output to train a recurrent neural network.

[0062] Step 10: Use the trained recurrent neural network to predict NH3 generation during the organic solid waste composting process.

[0063] This application obtains basic data parameters through simulation experiments, engineering practice, and literature collection, amplifies the dataset using a coupled metabolic kinetic model, simulates the entire metabolic process, and constructs a GRU neural network model to achieve reasonable prediction of NH3 production during aerobic fermentation, which is applicable to aerobic fermentation treatment of organic solid waste.

[0064] Basic data such as time t, biomass concentration X, substrate concentration S, temperature T, pH, and NH3 production concentration P were obtained through simulation experiments, engineering practice, and literature review. Based on existing research on nitrogen metabolic networks in composting processes, the key metabolic pathways of NH3 were simplified, and a metabolic flux balance equation was established. On this basis, macroscopic kinetics were coupled to form an NH3 coupled metabolic kinetic model, which was then fitted to a full-cycle dataset of the composting process. Subsequently, based on the above work, a GRU neural network was trained and constructed using biomass concentration X, substrate concentration S, temperature T, and pH as inputs and NH3 production concentration P as the output. The coefficient of determination (R²) was used to measure the NH3 production concentration. 2 The model performance is evaluated using mean absolute percentage error (MAPE) and root mean square error (RMSE). By optimizing functions and adjusting model parameters, the error is kept within a reasonable range to ensure the model's feasibility.

[0065] Specific Implementation Method Two: This implementation method is a further explanation of Specific Implementation Method One. The difference between this implementation method and Specific Implementation Method One is that the key metabolic pathways include oxidation reactions, intracellular substance synthesis reactions, nitrification reactions, and denitrification reactions.

[0066] Specific Implementation Method 3: This implementation method is a further explanation of Specific Implementation Method 2. The difference between this implementation method and Specific Implementation Method 2 is that the key metabolite includes NH4. + NO2 - NO3 - NO, N2O, ATP, H + And NADH.

[0067] Specific Implementation Method Four: This implementation method is a further explanation of Specific Implementation Method One. The difference between this implementation method and Specific Implementation Method One is that the cell metabolic flux balance equation is expressed as follows:

[0068]

[0069] Among them, S m Let m represent the metrology matrix, d represent the metabolite, and t represent the integral.

[0070] Specific Implementation Method Five: This implementation method is a further explanation of Specific Implementation Method Four. The difference between this implementation method and Specific Implementation Method Four is that the flux balance equation when intracellular metabolism is in a quasi-steady state is expressed as follows:

[0071] S m r = 0.

[0072] Specific Implementation Method Six: This implementation method is a further explanation of Specific Implementation Method Five. The difference between this implementation method and Specific Implementation Method Five is that the flux balance equation is expressed as follows:

[0073]

[0074] Among them, NH4 + Indicates ammonium ion; NO2 + Indicates nitro cation; NO3 - Nitrate represents nitrate; NO represents nitric oxide; N₂O represents nitrous oxide; ATP represents adenosine triphosphate; H₂O represents nitrate ions ... + represents hydrogen ions; NADH represents reduced nicotinamide adenine dinucleotide; b represents the parameter to be identified; P / O represents the phosphorus-oxygen ratio.

[0075] Specific Implementation Method Seven: This implementation method is a further explanation of Specific Implementation Method Six. The difference between this implementation method and Specific Implementation Method Six is ​​that the macroscopic dynamic model is expressed as follows:

[0076]

[0077]

[0078]

[0079] Among them, M p The relative molecular mass of the product, specifically the relative molecular mass of NH3, is M. p =17; M s M represents the relative molecular mass of the substrate. s =74, X represents biological concentration, μ represents specific growth rate, r p The product formation rate is represented by P, the NH3 production concentration is represented by S, and the substrate concentration is represented by S.

[0080] Specific Implementation Method Eight: This implementation method is a further explanation of Specific Implementation Method Seven. The difference between this implementation method and Specific Implementation Method Seven is that the recurrent neural network includes a data processing module, a network module, and a training and verification module.

[0081] The data processing module is used for dataset preprocessing, i.e., normalization; then the dataset is split according to the time step, and the resulting data is divided into training set, validation set and test set in a 7:2:1 ratio.

[0082] The network module is used to set the structure and related hyperparameters of the recurrent neural network. Specifically, the recurrent neural network has 2 layers, 4 input layers, 16 hidden layers, 1 output layer, and the activation function is tanh with a dropout rate of 0.3.

[0083] The training and validation module is used to train and validate the model. During training, the learning rate is 0.001, and backpropagation is used for training. The parameter used for validation is the coefficient of determination R0. 2 The algorithm includes Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and a loss function, SmoothL1Loss, with Adam as the optimizer.

[0084] Specific Implementation Method Nine: This implementation method is a further explanation of Specific Implementation Method Eight. The difference between this implementation method and Specific Implementation Method Eight is that the normalization process is performed using the Sklearn StandardScaler method.

[0085] Specific Implementation Method Ten: This implementation method is a further explanation of Specific Implementation Method Nine. The difference between this implementation method and Specific Implementation Method Nine is that the normalization process is expressed as follows:

[0086]

[0087] Where x represents all sample values ​​of the dataset, x* represents the standardized values ​​of all data, u represents the mean of each data category, and σ represents the standard deviation of each data category.

[0088] Example:

[0089] Basic data such as time t, biological concentration X, substrate concentration S, temperature T, pH value, and NH3 production concentration P were obtained through experimental simulation, engineering practice, and literature review. The data used in this invention comes from the actual composting results of a project of our team, with a data period of 24 days.

[0090] Based on a literature review, the nitrogen cycle pathway in the composting process was clarified, and information such as key metabolic pathways and specific equations for NH3 production was summarized. These include oxidation reactions, intracellular synthesis reactions, nitrification reactions, denitrification reactions, and other reactions, totaling 14 metabolic equations and identifying 8 key metabolites, including NH4+. + NO2 - NO3 - NO, N2O, ATP, H + And NADH. At the same time, based on the law of conservation of mass, the cell metabolic flux balance equation is shown in Equation (1). When the intracellular metabolism is in a quasi-steady state, the equation can be changed to Equation (2). Therefore, by substituting the different metabolites in the simplified key pathway of NH3 metabolism and their corresponding metabolic flux vectors r into the balance equation, the flux balance equation can be obtained (see Equation (3)).

[0091]

[0092] S m r = 0 (2)

[0093] In the formula: m—metabolite;

[0094] S m —Quantitative matrix.

[0095]

[0096] Furthermore, based on the law of conservation of mass, a macroscopic kinetic model was constructed (see formulas (4)-(6)) to establish the relationship between the concentration changes of biomass, substrate consumption, and NH3 production and the metabolic flux, the specific growth rate μ, and the specific substrate consumption rate r. s , specific product formation rate r p The relationship between them.

[0097]

[0098]

[0099]

[0100] Where: M p —The relative molecular mass of the product, i.e., the relative molecular mass of NH3, M p =17;

[0101] M s —The relative molecular mass of the substrate, M s =74.

[0102] Therefore, combining the flux balance equation and the macroscopic dynamic model yields the coupled dynamic model (such as...). Figure 3 The model uses the first 3 days of data from the dataset to learn and simulate the full-cycle change of NH3 concentration over 24 days, which is then used as the input dataset for the subsequent recurrent neural network GRU.

[0103] Based on the above data, with biological concentration X, substrate concentration S, temperature T, and pH as inputs and NH3 production concentration P as output, a GRU neural network was trained and constructed. The model mainly includes a data processing module, a network module, and a training and validation module. The data processing module is mainly used for dataset preprocessing, that is, normalization using the Sklearn StandardScaler method, as shown in formula (7). Then, the dataset is divided according to the time step, and the resulting data is divided into training set, validation set, and test set according to a ratio of 7:2:1. The network module is mainly used to set the structure and related hyperparameters of the GRU network. In this invention, the GRU has 2 layers, 4 input layers, 16 hidden layers, 1 output layer, and the activation function is tanh. The dropout rate is 0.3 (to alleviate the network overfitting problem). The training and validation module is used to train and validate the model. The learning rate during training is 0.001, and backpropagation is used for training and learning. The main parameter used for validation is R. 2 The model R is trained, optimized, and validated using the Adam algorithm, with parameters including MAPE, RMSE, and SmoothL1 Loss. 2 The MAPE and RMSE values ​​were 0.73, 35.43, and 1.19, respectively, meeting the correlation fit requirements.

[0104]

[0105] In the formula: x — all sample values ​​of the dataset;

[0106] x* — The standardized value of all data;

[0107] u—mean of various data types;

[0108] σ – Standard deviation of various data types.

[0109] The innovation of this application lies in constructing a predictive model for NH3 production during organic solid waste composting by utilizing a metabolic kinetic coupled recurrent neural network (GRU). 2 Up to 0.73.

[0110] It should be noted that the specific embodiments are merely explanations and illustrations of the technical solution of the present invention and should not be used to limit the scope of protection. Any modifications made in accordance with the claims and specification of the present invention that are only partial should still fall within the protection scope of the present invention.

Claims

1. A method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU, characterized in that... Includes the following steps: Step 1: Obtain compost data; Step 2: Based on the acquired composting data, obtain the N-cycle path in the composting process; Step 3: Based on the N cycle pathway in the composting process, obtain the key metabolic pathways for NH3 production, and then obtain the key metabolites contained in the key metabolic pathways. Step 4: Based on the law of conservation of mass and the cell metabolic flux balance equation, the flux balance equation when intracellular metabolism is in a quasi-steady state is obtained. Step 5: Obtain the metabolic flux vector r corresponding to different key metabolites based on the flux balance equation when intracellular metabolism is in a quasi-steady state; Step 6: Substitute the different key metabolites and their corresponding metabolic flux vectors r into the flux balance equation when intracellular metabolism is in a quasi-steady state to obtain the final flux balance equation. Step 7: Based on the law of conservation of mass, establish a system to represent the concentration changes of biomass, substrate consumption, and NH3 production in relation to metabolic flux, growth rate μ, and specific substrate consumption rate r. s , specific product formation rate r p A macroscopic dynamic model of the relationship between them; Step 8: Combine the final flux balance equation with the macroscopic dynamic model to obtain the coupled dynamic model; Step 9: Use a coupled kinetic model to simulate the full-cycle variation of NH3 concentration and fit the data. The fitted data includes biological concentration X, substrate concentration S, temperature T, and pH value. Finally, use biological concentration X, substrate concentration S, temperature T, and pH value as inputs and the NH3 concentration corresponding to the fitted data as output to train a recurrent neural network. Step 10: Use the trained recurrent neural network to predict NH3 generation during the organic solid waste composting process.

2. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 1, characterized in that... The key metabolic pathways include oxidation, intracellular synthesis, nitrification, and denitrification.

3. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 2, characterized in that... The key metabolite includes NH4 + NO2 - NO3 - NO, N2O, ATP, H + And NADH.

4. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 1, characterized in that... The cell metabolic flux balance equation is expressed as follows: Among them, S m This represents a metrology matrix, where m represents metabolites. Represents the integral. Indicates time.

5. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU according to claim 4, characterized in that... The flux balance equation is expressed as follows: Among them, NH4 + Indicates ammonium ion; NO2 + Indicates nitro cation; NO3 - Nitrate represents nitrate; NO represents nitric oxide; N₂O represents nitrous oxide; ATP represents adenosine triphosphate; H₂O represents nitrate ions ... N₂O represents nitrous oxide; ATP represents adenosine triphosphate; H₂O represents nitrate ions; N₂O represents nitrous oxide; ATP represents + represents hydrogen ions; NADH represents reduced nicotinamide adenine dinucleotide; b represents the parameter to be identified; P / O represents the phosphorus-oxygen ratio.

6. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 5, characterized in that... The macroscopic dynamic model is expressed as follows: Among them, M p The relative molecular mass of the product, specifically the relative molecular mass of NH3, is M. p =17; M s M represents the relative molecular mass of the substrate. s =74, Indicates biological concentration. R represents the specific growth rate. p The product formation rate is represented by P, the NH3 production concentration is represented by S, and the substrate concentration is represented by S.

7. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 6, characterized in that... The recurrent neural network includes a data processing module, a network module, and a training and verification module; The data processing module is used for dataset preprocessing, i.e., normalization processing; Then the dataset is split according to the time step, and the resulting data is divided into training set, validation set and test set in a 7:2:1 ratio; The network module is used to set the structure and related hyperparameters of the recurrent neural network. Specifically, the recurrent neural network has 2 layers, 4 input layers, 16 hidden layers, 1 output layer, and the activation function is tanh with a dropout rate of 0.

3. The training and validation module is used to train and validate the model. During training, the learning rate is 0.001, and backpropagation is used for training. The parameter used for validation is the coefficient of determination R0. 2 The algorithm includes Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and a loss function, SmoothL1Loss, with Adam as the optimizer.

8. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU according to claim 7, characterized in that... The normalization process is performed using the Sklearn StandardScaler method.

9. The method for predicting NH3 generation in organic solid waste composting based on metabolic kinetics-GRU as described in claim 8, characterized in that... The normalization process is expressed as follows: Where x represents all data sample values ​​in the dataset, x* represents all standardized values ​​of the data, u represents the mean of each data category, and σ represents the standard deviation of each data category.

Images