A method for predicting direct carbon emissions of urban sewage treatment based on an adaptive fuzzy neural network
By constructing a multi-layer network structure using an adaptive fuzzy neural network, the problem of difficulty in characterizing the dynamic correlations in the wastewater treatment process is solved, enabling high-precision prediction of direct carbon emissions from urban wastewater treatment and enhancing the stability and accuracy of the model under complex operating conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING DRAINAGE GRP CO LTD
- Filing Date
- 2026-01-27
- Publication Date
- 2026-06-12
AI Technical Summary
Existing technologies struggle to accurately depict the dynamic relationship between multiple operating parameters and direct carbon emissions during wastewater treatment, and parameter fluctuations lead to prediction biases. Traditional methods lack the ability to model the dynamic correlation of carbon emissions and the suppression of interference, failing to meet the requirements for prediction accuracy and stability under complex operating conditions.
By employing an adaptive fuzzy neural network-based approach, a multi-layer network structure is constructed, and the gradient descent method is used to update the weight parameters, center parameters, and Gaussian width parameters, thereby reducing the interference of parameter fluctuations on the prediction results and achieving high-precision prediction of direct carbon emissions from urban wastewater treatment.
It effectively captures the dynamic correlation between multiple parameters and direct carbon emissions, enhances the stability and accuracy of the model in complex scenarios, and improves the accuracy and reliability of predicting direct carbon emissions from urban wastewater treatment.
Smart Images
Figure CN122201485A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of low-carbon operation and intelligent prediction technology for urban wastewater treatment, and more specifically, to a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network. Background Technology
[0002] As a core infrastructure for urban ecological environment governance, urban wastewater treatment systems are widely used in areas such as urban sewage purification, water environment quality improvement, and water resource recycling. The carbon emission level during their operation is one of the core indicators for measuring the degree of low-carbon transformation of the process. Changes in direct carbon emissions directly reflect the carbon emission intensity at the wastewater treatment site and are a key reference for formulating low-carbon control strategies and carbon reduction plans. If direct carbon emissions are not predicted in a timely and accurate manner, low-carbon operation strategies will lack specificity and may even hinder the low-carbon upgrading of the process. Therefore, predicting direct carbon emissions from urban wastewater treatment is an important means to improve the low-carbon operation level of processes and promote the low-carbon transformation of the industry.
[0003] Existing technologies for predicting direct carbon emissions from urban wastewater treatment largely rely on empirical formulas, correlation analysis of single operating parameters, or static statistical models. These methods struggle to characterize the dynamic relationships between multiple operating parameters and direct carbon emissions during wastewater treatment, and cannot effectively address the interference of parameter fluctuations on carbon emission prediction results. Under the influence of factors such as fluctuations in influent water quality, changes in process operating conditions, and variations in equipment operating status, the correlation structure between operating parameters and direct carbon emissions dynamically changes over time. The interference caused by parameter fluctuations further increases prediction bias, resulting in poor model adaptability and insufficient prediction accuracy. Traditional methods generally lack the ability to model the dynamic correlation of carbon emissions and suppress interference, making it difficult to meet the requirements for prediction accuracy and stability under complex operating conditions.
[0004] Therefore, it is necessary to develop a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network.
[0005] The information disclosed in the background section of this invention is intended only to enhance the understanding of the general background of this invention, and should not be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention
[0006] This invention proposes a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network. By establishing a fuzzy neural network prediction model with adaptive learning capabilities, extracting parameter features through a multi-layer network structure, and adaptively adjusting model parameters, the method reduces the interference of parameter fluctuations on the prediction results, thereby achieving high-precision prediction of direct carbon emissions from urban wastewater treatment.
[0007] This disclosure provides a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network, including: A direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network is constructed, including an input layer, a radial basis function layer, a normalization layer, and an output layer. The prediction model is trained using sample data, and the weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using the gradient descent method until the preset number of training rounds is reached, and the trained prediction model is output. The current wastewater treatment operation data is collected as the input variable for the trained prediction model to predict the total direct carbon emissions from urban wastewater treatment in the next moment.
[0008] Preferably, the input layer consists of multiple neurons, and the output of each neuron is:
[0009] in, x i ( t ) indicates the input layer t Time of the first i The output of each neuron z i ( t ) indicates the input layer t Time of the first i The input of each neuron.
[0010] Preferably, the radial basis function layer consists of multiple neurons, and the output of each neuron is:
[0011] in, j ( t ) is the first j The output of each radial basis function neuron c ij ( t ) and σ ij ( t ) are respectively the radial basis function layers. j In radial basis function neurons x i ( t The central parameter and Gaussian width parameter of the corresponding membership function.
[0012] Preferably, the normalization layer consists of multiple neurons, and the output of each neuron is:
[0013] in, v l( t ) is the first in the normalized layer l The output of each neuron.
[0014] Preferably, the output of the output layer is:
[0015] in, express t The total direct carbon emissions from urban wastewater treatment predicted by the time-matter model at the next time step. w l ( t ) indicates the normalized layer number. l The weight parameters between each neuron and the output layer neurons.
[0016] Preferably, the prediction model is trained using sample data, and the weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using gradient descent until a preset number of training rounds are reached. The output of the trained prediction model includes: Set the current training time to t Initialize the number of training rounds t =1, initialize the weight parameters of the prediction model; calculate t -1 moment t The predicted output of the training is used to calculate the loss function, mean absolute error, moving average error, and adaptive learning rate. The weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using gradient descent. like t Less than the preset number of training rounds, t Increase by 1, continue training, if t Once the required number of training rounds is reached, the training and updating of model parameters is terminated, and the trained prediction model is output.
[0017] Preferably, the loss function is:
[0018] in, J ( t )express t Loss of the time-matrix model express t The total direct carbon emissions from urban wastewater treatment predicted by the model at time 1 in the next time step. y ( t 1) indicates t The actual total direct carbon emissions from urban wastewater treatment at moment 1 in the next moment.
[0019] Preferably,t Time of the first t The mean absolute error of training (number of training cycles) and moving average error for:
[0020] in, express t Time of the first t -1 training moving average error, β Indicates the smoothing coefficient; The adaptive learning rate is:
[0021] in, or τ ( t )∈(0,1] means t Time of the first t Adaptive learning rate for each training session or 0 represents the base learning rate. c Indicates the error amplification factor. e It is a smoothing factor.
[0022] Preferably, the weighting parameter is:
[0023] The center parameter is:
[0024] The Gaussian width parameter is:
[0025] in, express t Time of the first t The model weight parameters for this training iteration. express t Time of the first t Loss function in the second training For model weight parameters gradient, express t Time of the first t +1 training iterations of model weight parameters express t Time of the first t The central parameters of the model after training. express t Time of the first t Loss function in the second training For model central parameters gradient, express t Time of the first t +1 training cycle of model center parameters, express t Time of the first t The Gaussian width parameter of the model trained in this iteration. express t Time of the first t Loss function in the second training Gaussian width parameter of the model gradient, express t Time of the first t Gaussian width parameters of the model trained +1 time. t The index representing the training round number.
[0026] Preferably, the wastewater treatment operation data includes the oxidation-reduction potential of the anaerobic zone, the oxidation-reduction potential of the anoxic zone, the total air flow rate, the air flow rate of the first aerobic corridor, the air flow rate of the second aerobic corridor, the ammonia nitrogen concentration, the influent biochemical oxygen demand, the influent total nitrogen, the valve opening of the air regulating valve of the first aerobic corridor, and the valve opening of the air regulating valve of the second aerobic corridor.
[0027] Its beneficial effects are as follows: 1. This invention addresses the problems in predicting direct carbon emissions from urban wastewater treatment, namely the dynamic correlation between multiple operating parameters and direct carbon emissions, the difficulty of accurately characterizing these correlations using traditional static models, and the potential for prediction bias due to parameter fluctuations. It proposes a direct carbon emission prediction method based on an adaptive fuzzy neural network. By extracting nonlinear features between parameters and carbon emissions using radial basis functions and optimizing feature distribution using a normalization layer, the method effectively captures the dynamic correlation between multiple parameters and direct carbon emissions, achieving accurate prediction of total direct carbon emissions. 2. This invention designs an adaptive learning rate mechanism, dynamically calculating the learning rate by combining the mean absolute error and the moving average error, and dynamically updating the model weight parameters, radial basis function center parameters, and Gaussian width parameters using the gradient descent method. This mechanism can adaptively reduce the interference of error fluctuations on the prediction results during parameter updates, enhance the stability of the model under complex scenarios such as influent water quality fluctuations and operating condition switching, and further improve the accuracy and reliability of predicting direct carbon emissions from urban wastewater treatment.
[0028] The method of the present invention has other features and advantages that will be apparent from or will be set forth in detail in the accompanying drawings and following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description
[0029] The above and other objects, features and advantages of the present invention will become more apparent from the more detailed description of exemplary embodiments of the invention in conjunction with the accompanying drawings, wherein the same reference numerals generally represent the same parts.
[0030] Figure 1 A flowchart illustrating the steps of a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network, according to an embodiment of the present invention, is shown.
[0031] Figure 2 A graph showing the predicted total direct carbon emissions from urban wastewater treatment according to an embodiment of the present invention is provided.
[0032] Figure 3 A graph showing the prediction error of total direct carbon emissions from urban wastewater treatment according to an embodiment of the present invention is illustrated. Detailed Implementation
[0033] Preferred embodiments of the invention will now be described in more detail. While preferred embodiments of the invention are described below, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.
[0034] Figure 1 A flowchart illustrating the steps of a method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network, according to an embodiment of the present invention, is shown.
[0035] like Figure 1 As shown, the method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network includes: Step 101: Construct a direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network, including an input layer, a radial basis function layer, a normalization layer, and an output layer; Step 102: Train the prediction model using sample data, and update the weight parameters, center parameters, and Gaussian width parameters of the prediction model using gradient descent until the preset number of training rounds is reached, and output the trained prediction model. Step 103: Collect the current sewage treatment operation data as the input variable of the trained prediction model to predict the total direct carbon emissions from urban sewage treatment in the next moment.
[0036] In one example, the input layer consists of multiple neurons, and the output of each neuron is:
[0037] in, x i ( t) indicates the input layer t Time of the first i The output of each neuron z i ( t ) indicates the input layer t Time of the first i The input of each neuron.
[0038] In one example, the radial basis function layer consists of multiple neurons, and the output of each neuron is:
[0039] in, j ( t ) is the first j The output of each radial basis function neuron c ij ( t ) and σ ij ( t ) are respectively the radial basis function layers. j In radial basis function neurons x i ( t The central parameter and Gaussian width parameter of the corresponding membership function.
[0040] In one example, the normalization layer consists of multiple neurons, and the output of each neuron is:
[0041] in, v l ( t ) is the first in the normalized layer l The output of each neuron.
[0042] In one example, the output of the output layer is:
[0043] in, express t The total direct carbon emissions from urban wastewater treatment predicted by the time-matter model at the next time step. w l ( t ) indicates the normalized layer number. l The weight parameters between each neuron and the output layer neurons.
[0044] In one example, a prediction model is trained using sample data. Gradient descent is then used to update the model's weights, center parameters, and Gaussian width until a preset number of training epochs are reached. The output of the trained prediction model includes: Set the current training time to t Initialize the number of training rounds t =1, initialize the weight parameters of the prediction model; calculate t -1 moment t The predicted output of the training is used to calculate the loss function, mean absolute error, moving average error, and adaptive learning rate. The weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using gradient descent. like t Less than the preset number of training rounds, t Increase by 1, continue training, if t Once the required number of training rounds is reached, the training and updating of model parameters is terminated, and the trained prediction model is output.
[0045] In one example, the loss function is:
[0046] in, J ( t )express t Loss of the time-matrix model express t The total direct carbon emissions from urban wastewater treatment predicted by the model at time 1 in the next time step. y ( t 1) indicates t The actual total direct carbon emissions from urban wastewater treatment at moment 1 in the next moment.
[0047] In one example t Time of the first t The mean absolute error of training (number of training cycles) and moving average error for:
[0048] in, express t Time of the first t -1 training moving average error, β Indicates the smoothing coefficient; The adaptive learning rate is:
[0049] in, or τ ( t )∈(0,1] means t Time of the first tAdaptive learning rate for each training session or 0 represents the base learning rate. c Indicates the error amplification factor. e It is a smoothing factor.
[0050] In one example, the weight parameters are:
[0051] The center parameter is:
[0052] The Gaussian width parameter is:
[0053] in, express t Time of the first t The model weight parameters for this training iteration. express t Time of the first t Loss function in the second training For model weight parameters gradient, express t Time of the first t +1 training iterations of model weight parameters express t Time of the first t The central parameters of the model after training. express t Time of the first t Loss function in the second training For model central parameters gradient, express t Time of the first t +1 training cycle of model center parameters, express t Time of the first t The Gaussian width parameter of the model trained in this iteration. express t Time of the first t Loss function in the second training Gaussian width parameter of the model gradient, express t Time of the first t Gaussian width parameters of the model trained +1 time. t The index representing the training round number.
[0054] In one example, wastewater treatment operation data includes oxidation-reduction potential in the anaerobic zone, oxidation-reduction potential in the anoxic zone, total air flow rate, air flow rate in the first aerobic corridor, air flow rate in the second aerobic corridor, ammonia nitrogen concentration, influent biochemical oxygen demand, influent total nitrogen, valve opening of the air regulating valve in the first aerobic corridor, and valve opening of the air regulating valve in the second aerobic corridor.
[0055] Specifically, wastewater treatment operation data is collected, with wastewater treatment plants as the research object. t Redox potential in the anaerobic zone at any time z 1( t Oxidation-reduction potential in the hypoxic region z 2( t Total airflow z 3( t ), air flow rate in the first corridor of the aerobic zone z 4( t ), air flow rate in the second corridor of the aerobic zone z 5( t ammonia nitrogen concentration z 6( t ), influent biochemical oxygen demand z 7( t ), total nitrogen in influent z 8( t ), Valve opening degree of the air conditioning valve in the first corridor of the aerobic zone z 9( t ), Valve opening degree of the air conditioning valve in the second corridor of the aerobic zone z 10 ( t ); Select z 1( t ), z 2( t ), z 3( t ), z 4( t ), z 5( t ), z 6( t ), z 7( t ), z 8( t ), z 9( t )and z 10 ( t As t The input variable of the urban wastewater treatment direct carbon emission prediction model based on the adaptive fuzzy neural network at each time step is the total amount of urban wastewater treatment direct carbon emissions at the next time step, which is used as the output variable of the model.
[0056] A direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network is constructed. This model consists of an input layer, a radial basis function layer, a normalization layer, and an output layer. Input layer: The input layer consists of n It consists of 10 neurons, and the output of each neuron is: (1) in, x i ( t ) indicates the input layer t Time of the first i The output of each neuron z i ( t ) indicates the input layer t Time of the first i The input of each neuron, i =1,2,…, n ; Radial basis function layer: The radial basis function layer consists of... p It consists of 16 neurons, and the output of each neuron is: (2) in, j ( t ) is the first j The output of each radial basis function neuron j =1,2,…, p , c ij ( t ) and σ ij ( t ) are respectively the radial basis function layers. j In radial basis function neurons x i ( t The corresponding membership function's center parameter and Gaussian width parameter; Normalization layer: The normalization layer consists of p It consists of 5 neurons, and the output of each neuron is: (3) in, v l ( t ) is the first in the normalized layer l The output of each neuron l =1,2,…, p ; Output layer: The output of the output layer is: (4) in, express t The total direct carbon emissions from urban wastewater treatment predicted by the time-matter model at the next time step. w l ( t ) indicates the normalized layer number. l The weight parameters between each neuron and the output layer neurons; Training a direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network: ① Define the loss function of the model as: (5) in, J ( t )express t Loss of the time-matrix model express t The total direct carbon emissions from urban wastewater treatment predicted by the model at time 1 in the next time step. y ( t 1) indicates t The actual total direct carbon emissions from urban wastewater treatment at moment 1 in the next moment; ② Set the current training time to t Initialize the number of training rounds t =1, the number of training iterations is fixed at 1000; initialize the model's weight parameters, center parameters and Gaussian width parameters. The weight parameters are randomly selected in the interval [-0.2, 0.2], the center parameters are randomly selected in the interval [-2, 2], and the Gaussian width parameters are randomly selected in the interval [0.01, 1]. ③ Definition t Time of the first t The mean absolute error of training (number of training cycles) and moving average error for: (6) (7) in, express t Time of the first t -1 training moving average error, β =0.9 indicates the smoothing coefficient; During model training, the adaptive learning rate is calculated using the following formula: (8) in, or τ (t )∈(0,1] means t Time of the first t Adaptive learning rate for each training session or 0 = 0.1 represents the base learning rate. c =0.3 indicates the error amplification factor. e =0.001 is a smoothing factor to prevent division by zero; Calculate using formulas (1)-(4) t -1 moment t The predicted output of the model after training. Calculate using formula (5) t Time of the first t The loss of the model after training. J τ ( t ), calculate using formulas (6)-(7) t Time of the first t The mean absolute error of training (number of training cycles) and moving average error The adaptive learning rate is calculated using formula (8). or τ ( t The model's weight parameters, center parameters, and Gaussian width parameters are updated using gradient descent. The calculation formula is as follows: (9) (10) (11) in, express t Time of the first t The model weight parameters for this training iteration. express t Time of the first t Loss function in the second training For model weight parameters gradient, express t Time of the first t +1 training iterations of model weight parameters express t Time of the first t The central parameters of the model after training. express t Time of the first t Loss function in the second training For model central parameters gradient, express t Time of the first t+1 training cycle of model center parameters, express t Time of the first t The Gaussian width parameter of the model trained in this iteration. express t Time of the first t Loss function in the second training Gaussian width parameter of the model gradient, express t Time of the first t Gaussian width parameters of the model trained +1 time. t Index representing the training round number, t =1,2,…,1000; ④ If the number of training rounds t <1000, t Increase by 1, proceed to step ③ and continue training; if the number of training rounds... t If the value is ≥1000, then terminate the training and update of the model parameters and output the trained prediction model.
[0057] Using a trained urban wastewater treatment direct carbon emission prediction model based on an adaptive fuzzy neural network, to... t The anaerobic zone oxidation-reduction potential, anoxic zone oxidation-reduction potential, total air flow, aerobic zone first corridor air flow, aerobic zone second corridor air flow, ammonia nitrogen concentration, influent biochemical oxygen demand, influent total nitrogen, valve opening of the aerobic zone first corridor air regulating valve, and aerobic zone second corridor air regulating valve are collected at all times as input to the model. The model is then obtained according to formulas (1)-(4). t Model output at time step Output the model As t The total direct carbon emissions from urban wastewater treatment are predicted for the next moment.
[0058] To facilitate understanding of the solutions and effects of the embodiments of the present invention, a specific application example is given below. Those skilled in the art should understand that this example is merely for the purpose of understanding the present invention, and any specific details therein are not intended to limit the present invention in any way.
[0059] example 1
[0060] The experimental data came from the actual operating wastewater treatment system of a municipal wastewater treatment plant. Operating data were collected at 15-minute sampling intervals, including oxidation-reduction potential in the anaerobic zone, oxidation-reduction potential in the anoxic zone, total air flow, air flow in the first and second aerobic corridors of the aerobic zone, ammonia nitrogen concentration, influent biochemical oxygen demand (BOD), influent total nitrogen, valve opening of the air regulating valves in the first and second aerobic corridors of the aerobic zone, etc. A total of 100 hours of data were collected, resulting in 400 sets of samples. The data were divided into training and testing sets according to time sequence, with the first 62.5 hours (250 sets) used as training samples and the last 37.5 hours (150 sets) used as testing samples.
[0061] A direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network is constructed. This model consists of an input layer, a radial basis function layer, a normalization layer, and an output layer. Input layer: The input layer consists of n It consists of 10 neurons, and the output of each neuron is: (1) in, x i ( t ) indicates the input layer t Time of the first i The output of each neuron z i ( t ) indicates the input layer t Time of the first i The input of each neuron, i =1,2,…, n ; Radial basis function layer: The radial basis function layer consists of... p It consists of 16 neurons, and the output of each neuron is: (2) in, j ( t ) is the first j The output of each radial basis function neuron j =1,2,…, p , c ij ( t ) and σ ij ( t ) are respectively the radial basis function layers. j In radial basis function neurons x i ( t The corresponding membership function's center parameter and Gaussian width parameter; Normalization layer: The normalization layer consists of p It consists of 5 neurons, and the output of each neuron is: (3) in, v l ( t ) is the first in the normalized layer l The output of each neuron l =1,2,…, p ; Output layer: The output of the output layer is: (4) in, express t The total direct carbon emissions from urban wastewater treatment predicted by the time-matter model at the next time step. w l ( t ) indicates the normalized layer number. l The weight parameters between each neuron and the output layer neurons; Training a direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network: ① Define the loss function of the model as: (5) in, J ( t )express t Loss of the time-matrix model express t The total direct carbon emissions from urban wastewater treatment predicted by the model at time 1 in the next time step. y ( t 1) indicates t The actual total direct carbon emissions from urban wastewater treatment at moment 1 in the next moment; ② Set the current training time to t Initialize the number of training rounds t =1, the number of training iterations is fixed at 1000; initialize the model's weight parameters, center parameters and Gaussian width parameters. The weight parameters are randomly selected in the interval [-0.2, 0.2], the center parameters are randomly selected in the interval [-2, 2], and the Gaussian width parameters are randomly selected in the interval [0.01, 1]. ③ Definition t Time of the first t The mean absolute error of training (number of training cycles) and moving average error for: (6) (7) in, express t Time of the first t -1 training moving average error, β =0.9 indicates the smoothing coefficient; During model training, the adaptive learning rate is calculated using the following formula: (8) in, or τ ( t )∈(0,1] means t Time of the first t Adaptive learning rate for each training session or 0 = 0.1 represents the base learning rate. c =0.3 indicates the error amplification factor. e =0.001 is a smoothing factor to prevent division by zero; Calculate using formulas (1)-(4) t -1 moment t The predicted output of the model after training. Calculate using formula (5) t Time of the first t The loss of the model after training. J τ ( t ), calculate using formulas (6)-(7) t Time of the first t The mean absolute error of training (number of training cycles) and moving average error The adaptive learning rate is calculated using formula (8). or τ ( t The model's weight parameters, center parameters, and Gaussian width parameters are updated using gradient descent. The calculation formula is as follows: (9) (10) (11) in, express t Time of the first t The model weight parameters for this training iteration. express t Time of the first t Loss function in the second training For model weight parameters gradient, expresst Time of the first t +1 training iterations of model weight parameters express t Time of the first t The central parameters of the model after training. express t Time of the first t Loss function in the second training For model central parameters gradient, express t Time of the first t +1 training cycle of model center parameters, express t Time of the first t The Gaussian width parameter of the model trained in this iteration. express t Time of the first t Loss function in the second training Gaussian width parameter of the model gradient, express t Time of the first t Gaussian width parameters of the model trained +1 time. t Index representing the training round number, t =1,2,…,1000; ④ If the number of training rounds t <1000, t Increase by 1, proceed to step ③ and continue training; if the number of training rounds... t If the value is ≥1000, then terminate the training and update of the model parameters and output the trained prediction model.
[0062] Figure 2 A graph showing the predicted total direct carbon emissions from urban wastewater treatment according to an embodiment of the present invention is provided.
[0063] Figure 3 A graph showing the prediction error of total direct carbon emissions from urban wastewater treatment according to an embodiment of the present invention is illustrated.
[0064] Using a trained urban wastewater treatment direct carbon emission prediction model based on an adaptive fuzzy neural network, to... t The anaerobic zone oxidation-reduction potential, anoxic zone oxidation-reduction potential, total air flow, aerobic zone first corridor air flow, aerobic zone second corridor air flow, ammonia nitrogen concentration, influent biochemical oxygen demand, influent total nitrogen, valve opening of the aerobic zone first corridor air regulating valve, and aerobic zone second corridor air regulating valve are collected at all times as input to the model. The model is then obtained according to formulas (1)-(4). t Model output at time step Output the model As t The predicted total direct carbon emissions from urban wastewater treatment at the next moment are as follows: Figure 2 As shown, the prediction error is as follows Figure 3 As shown.
[0065] Those skilled in the art should understand that the above description of the embodiments of the present invention is only intended to illustrate the beneficial effects of the embodiments of the present invention, and is not intended to limit the embodiments of the present invention to any of the examples given.
[0066] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.
Claims
1. A method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network, characterized in that, include: A direct carbon emission prediction model for urban wastewater treatment based on an adaptive fuzzy neural network is constructed, including an input layer, a radial basis function layer, a normalization layer, and an output layer. The prediction model is trained using sample data, and the weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using the gradient descent method until the preset number of training rounds is reached, and the trained prediction model is output. The current wastewater treatment operation data is collected as the input variable for the trained prediction model to predict the total direct carbon emissions from urban wastewater treatment in the next moment.
2. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 1, wherein, The input layer consists of multiple neurons, and the output of each neuron is: in, x i ( t ) indicates the input layer t Time of the first i The output of each neuron z i ( t ) indicates the input layer t Time of the first i The input of each neuron.
3. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 2, wherein, The radial basis function layer consists of multiple neurons, and the output of each neuron is: in, j ( t ) is the first j The output of each radial basis function neuron c ij ( t ) and σ ij ( t ) are respectively the radial basis function layers. j In radial basis function neurons x i ( t The corresponding membership function's center parameter and Gaussian width parameter.
4. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 3, wherein, The normalization layer consists of multiple neurons, and the output of each neuron is: in, v l ( t ) is the first in the normalized layer l The output of each neuron.
5. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 4, wherein, The output of the output layer is: in, express t The total direct carbon emissions from urban wastewater treatment predicted by the time-matter model at the next time step. w l ( t ) indicates the normalized layer number. l The weight parameters between each neuron and the output layer neurons.
6. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 1, wherein, The prediction model is trained using sample data, and its weight parameters, center parameters, and Gaussian width parameters are updated using gradient descent until a preset number of training epochs are reached. The output of the trained prediction model includes: Set the current training time to t Initialize the number of training rounds τ =1, initialize the weight parameters of the prediction model; calculate t -1 moment τ The predicted output of the training is used to calculate the loss function, mean absolute error, moving average error, and adaptive learning rate. The weight parameters, center parameters, and Gaussian width parameters of the prediction model are updated using the gradient descent method. like τ Less than the preset number of training rounds, τ Increase by 1, continue training, if τ Once the required number of training rounds is reached, the training and updating of model parameters is terminated, and the trained prediction model is output.
7. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 6, wherein, The loss function is: in, J ( t )express t Loss of the time-matrix model express t The total direct carbon emissions from urban wastewater treatment predicted by the model at time 1 in the next time step. y ( t 1) indicates t The actual total direct carbon emissions from urban wastewater treatment at moment 1 in the next moment.
8. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 6, wherein, t Time of the first τ The mean absolute error of training (number of training cycles) and moving average error for: in, express t Time of the first τ -1 training moving average error, β Indicates the smoothing coefficient; The adaptive learning rate is: in, η τ ( t )∈(0,1] means t Time of the first τ Adaptive learning rate for each training session η 0 represents the base learning rate. γ Indicates the error amplification factor. ε It is a smoothing factor.
9. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 6, wherein, The weight parameters are: The center parameter is: The Gaussian width parameter is: in, express t Time of the first τ The model weight parameters for this training iteration. express t Time of the first τ Loss function in the second training For model weight parameters gradient, express t Time of the first τ +1 training iterations of model weight parameters express t Time of the first τ The central parameters of the model after training. express t Time of the first τ Loss function in the second training For model central parameters gradient, express t Time of the first τ +1 training cycle of model center parameters, express t Time of the first τ The Gaussian width parameter of the model trained in this iteration. express t Time of the first τ Loss function in the second training Gaussian width parameter of the model gradient, express t Time of the first τ Gaussian width parameters of the model trained +1 time. τ The index representing the training round number.
10. The method for predicting direct carbon emissions from urban wastewater treatment based on an adaptive fuzzy neural network according to claim 1, wherein, Wastewater treatment operation data includes oxidation-reduction potential in the anaerobic zone, oxidation-reduction potential in the anoxic zone, total air flow, air flow in the first aerobic corridor, air flow in the second aerobic corridor, ammonia nitrogen concentration, influent biochemical oxygen demand, influent total nitrogen, valve opening of the air regulating valve in the first aerobic corridor, and valve opening of the air regulating valve in the second aerobic corridor.