Intelligent prediction method for total nitrogen in effluent based on event-driven fuzzy neural network

Abstract

This invention proposes an event-driven fuzzy neural network-based method for predicting total nitrogen in effluent, addressing the problems of large modeling errors and decreased prediction accuracy caused by outliers in the sampled data. The method includes: first, constructing an event-driven fuzzy neural network to predict the future dynamics of the system; second, designing an event partitioning mechanism based on the error fluctuation between the predicted and actual system output values, dividing the system state into four different events; and finally, employing an event-based adaptive parameter update strategy to effectively suppress the interference of outliers on system modeling, improve prediction accuracy, and achieve accurate prediction of the total nitrogen concentration in the effluent, thereby ensuring the efficient and reliable operation of the urban wastewater treatment process.

Description

Technical Field

[0001] This invention relates to an intelligent prediction method for total nitrogen in effluent based on an event-driven fuzzy neural network, achieving accurate prediction of total nitrogen in effluent during wastewater treatment. The effluent quality of wastewater treatment is a crucial indicator of the effectiveness of the treatment process. Predicting effluent quality improves the accuracy of achieving wastewater quality standards. This method aims to address the problem of increased modeling errors and decreased prediction accuracy caused by outliers in the sampled data. Applying the intelligent prediction model for total nitrogen in effluent based on an event-driven fuzzy neural network to wastewater quality prediction not only improves prediction accuracy and the operational efficiency of the wastewater treatment process but also effectively reduces operating costs. Background Technology

[0003] Total nitrogen (TNI) is a crucial indicator of water pollution. Excessive TNI discharge leads to eutrophication, causing a series of environmental problems such as algal blooms and water quality deterioration. Therefore, accurate and rapid detection of TNI in water is of great significance for environmental protection and remediation. Existing methods for TNI detection mainly include chemical methods, spectroscopic methods, and chromatographic methods. While these methods offer high accuracy and reliability, they also have some significant drawbacks. For example, chemical methods typically require large quantities of chemical reagents, are complex to operate, and demand highly skilled operators; spectroscopic and chromatographic methods require expensive equipment and specialized technicians, and have long detection times, making them unsuitable for real-time monitoring. In the field of wastewater treatment, artificial intelligence technologies such as neural networks are widely used in water quality monitoring and prediction due to their nonlinear mapping capabilities and strong adaptability. Therefore, researching intelligent methods for detecting TNI in effluent and achieving real-time TNI detection in effluent is of great importance.

[0004] This invention proposes an intelligent prediction method for total nitrogen in effluent based on event-driven fuzzy neural network (EDFNN). By constructing an EDFNN model, selectively selecting sampled data for model construction, and adopting different neural network parameter update strategies for data of different quality, the method achieves accurate prediction of total nitrogen concentration in effluent during wastewater treatment. Summary of the Invention

[0005] This invention proposes an intelligent prediction method for total nitrogen in effluent based on an event-driven fuzzy neural network. This method detects the quality of sampled data based on an event-driven mechanism, constructs corresponding events based on data of different quality, formulates a corresponding fuzzy neural network update strategy based on the triggered events, and obtains the predicted output of total nitrogen through the neural network. This method solves the problem of the impact of abnormal data values ​​on the accuracy of system modeling and ensures the accurate prediction of effluent water quality in urban sewage treatment processes.

[0006] The present invention adopts the following technical solution and implementation steps:

[0007] A method for intelligent prediction of total nitrogen in effluent based on event-triggered fuzzy neural networks, characterized by: determining the input and output variables of the event-triggered fuzzy neural network, establishing an event-triggered fuzzy neural network prediction model, and realizing intelligent prediction of total nitrogen in effluent based on the event-triggered fuzzy neural network; including the following steps:

[0008] (1) Determine the input and output variables of the event-triggered fuzzy neural network

[0009] Taking the wastewater treatment process as the research object, this study predicts the total nitrogen in the effluent. The output variable is the total nitrogen in the effluent, and the corresponding input variables are six variables: influent ammonia nitrogen, oxidation-reduction potential in the anaerobic tank, oxidation-reduction potential in the anoxic tank, nitrate nitrogen in the anoxic tank, suspended solids concentration in the mixed liquor of the anoxic tank, and total nitrogen concentration. The data at time k are obtained as follows: influent ammonia nitrogen h1(k), oxidation-reduction potential h2(k) in the anaerobic tank, oxidation-reduction potential h3(k) in the anoxic tank, nitrate nitrogen h4(k) in the anoxic tank, suspended solids concentration h5(k) in the mixed liquor of the anoxic tank, and total nitrogen concentration h6(k). The neural network input vector h(k) = [h1(k), h2(k), h3(k), h4(k), h5(k), h6(k)] T , T denotes the transpose of the matrix;

[0010] (2) Establish an event-triggered fuzzy neural network prediction model

[0011] The event-triggered fuzzy neural network prediction model consists of 6 input layer neurons, 6 fuzzy radial basis function layer neurons, 6 normalization layer neurons, and 1 output layer neuron. The input layer neurons convert the input data into fuzzy values ​​using fuzzy membership functions. The fuzzy radial basis function layer neurons process these fuzzy values ​​and apply fuzzy rules for inference. The normalization layer neurons normalize the fuzzy inference results. The output layer neurons defuzzify the fuzzy output into a clear prediction result. The model's output expression is as follows:

[0012]

[0013] Where ŷ(k+1) is the predicted output of total nitrogen in the effluent at time k+1, w j (k)(j=1,2,…,6) represents the weight at time k, which is randomly assigned within the range [0,1]. h i (k)(i=1,2,…,6) is the input of the prediction model at time k, c ij (k) represents the center value of the i-th neuron in the input layer corresponding to the j-th neuron in the radial basis function layer at time k, σ ij(k) represents the width value of the i-th neuron in the input layer corresponding to the j-th neuron in the radial basis function layer at time k, c ij (k) and σ ij (k) is randomly assigned a value in the range [0,1].

[0014] The parameter set for constructing the event-triggered neural network is as follows:

[0015]

[0016] in, , ,w (k)= [w1(k), w2(k), w3(k), w4(k), w5(k),w6(k)] T These represent the center value parameter set, width parameter set, and weight parameter set of the neural network, respectively.

[0017] The event-triggered parameter update process of the fuzzy neural network is as follows:

[0018] ① Calculate the error function E(k) at time k:

[0019]

[0020] Where e(k) = y(k) - ŷ(k) represents the error between the actual total nitrogen concentration and the predicted total nitrogen concentration of the system, y(k) is the true total nitrogen concentration obtained by sampling at time k, and ŷ(k) is the predicted output of the total nitrogen in the effluent at time k;

[0021] ② Calculate the error increment and change in increment at time k based on the error function E(k) at time k, as follows:

[0022]

[0023]

[0024]

[0025] Where F3(k) represents the error increment between the error function E(k) at time k and the error function E(k-3) at time k-3, G3(k) represents the increment change between the difference function F3(k) at time k and the difference function F3(k-3) at time k-3, and H3(k) represents the absolute value of the error increment at time k.

[0026] ③ Define the event state Con(k) as follows:

[0027]

[0028] In this context, state C1 indicates that the prediction error is decreasing rapidly, the data quality is excellent, and the model is quickly learning effective features; state C2 indicates that the prediction error is decreasing at a slower pace, the data quality is relatively good, the model's optimization space is narrowing, but there is still room for optimization; state C3 indicates that the prediction error is increasing, the data quality is poor, and further data feature detection mechanisms are needed to determine whether it is outlier interference or a sudden change in system operating conditions; state C4 indicates that the prediction error is very small at this point, and no neural network parameter update is required.

[0029] ④ Design a data feature detection mechanism

[0030] When C3 is triggered, a vector is constructed. Obtain the magnitude of historical neural network parameter changes:

[0031]

[0032] in (n=0, 1, …, 9) is used to quantify the magnitude of change in the total set of parameters triggered by the event at time kn:

[0033]

[0034] Among them, ||.|| F Denotes the F-norm;

[0035] Calculate the mean of the change and the standard deviation of the range of change as follows:

[0036]

[0037]

[0038] Calculate the virtual parameter update magnitude based on the current data :

[0039]

[0040] in To update the parameter set virtually, the following method is used:

[0041]

[0042] Where I represents the identity matrix;

[0043] To accurately distinguish between data feature changes caused by changes in operating conditions and data feature changes caused by outliers, a feature detection identifier is defined. as follows:

[0044]

[0045] Among them, when the conditions are met simultaneously and At that time, the feature detection identifier will be... Set to 1; otherwise, set the feature detection flag. Set it to 0 to determine that the change in prediction error was caused by a change in the data features themselves.

[0046] ⑤ Neural Network Parameter Update Strategy

[0047] The neural network parameters are updated as follows:

[0048]

[0049]

[0050]

[0051] Where U1, U2, and U3 represent different neural network parameter update strategies. When the event state triggered at time k is C1, update strategy U1 is executed, and the neural network parameters are updated according to formula (15). 2， Or when the event state triggered at time k is C3 and D1=0, execute update strategy U2 and update the neural network parameters according to formula (16); when the event state triggered at time k is C4, or when the event state triggered at time k is C3 and D1=1, execute update strategy U3 and do not update the neural network parameters.

[0052] (3) Realize intelligent prediction of total nitrogen in effluent based on event-triggered fuzzy neural network

[0053] Obtain the influent ammonia nitrogen data h1(k) at time k, the oxidation-reduction potential data h2(k) in the anaerobic tank, the oxidation-reduction potential data h3(k) in the anoxic tank, the nitrate nitrogen data h4(k) in the anoxic tank, the mixed liquor suspended solids concentration data h5(k) in the anoxic tank, and the total nitrogen concentration data h6(k). Construct the neural network input vector h(k). Input the input vector h(k) to the input layer of the event-triggered fuzzy neural network, and pass it through the fuzzy RBF layer, the normalization layer, and the output layer, as shown in the formula. As shown, the total nitrogen output ŷ(k+1) predicted by the event-triggered fuzzy neural network is obtained, in mg / L. The network parameter update strategy is as follows: -formula As shown.

[0054] The inventiveness of this invention is mainly reflected in:

[0055] (1) In view of the problem that the data quality of key process variables in the current urban sewage treatment process varies greatly and is disturbed by outliers, this invention designs an event-triggered fuzzy neural network to analyze the data quality in real time, thereby effectively reducing the adverse impact of data quality on model construction and improving the reliability of total nitrogen prediction in urban sewage treatment process.

[0056] (2) This invention addresses the problems of outliers easily interfering with modeling accuracy in urban wastewater treatment process modeling, and the inability of existing fuzzy neural networks, which use fixed parameter update modes, to adapt to dynamic changes in data quality and sudden changes in operating conditions. It designs an adaptive parameter update strategy for event-triggered fuzzy neural networks, combining a data feature detection mechanism to distinguish between sudden changes in operating conditions and data anomalies, and performs differentiated parameter updates for different event states. This enables the prediction of total nitrogen concentration in effluent under complex environments, exhibiting high modeling accuracy. Attached Figure Description

[0057] Figure 1 This is a graph showing the prediction effect of the model of this invention on the total nitrogen concentration in urban wastewater treatment processes.

[0058] Figure 2 This is a graph showing the prediction error of the total nitrogen concentration in the urban wastewater treatment process using the model of this invention. Specific implementation methods

[0059] 1. A method for intelligent prediction of total nitrogen in effluent based on event-triggered fuzzy neural networks, characterized by: determining the input and output variables of the event-triggered fuzzy neural network, establishing an event-triggered fuzzy neural network prediction model, and realizing intelligent prediction of total nitrogen in effluent based on the event-triggered fuzzy neural network; including the following steps:

[0060] (1) Determine the input and output variables of the event-triggered fuzzy neural network

[0061] Taking the wastewater treatment process as the research object, this study predicts the total nitrogen in the effluent. The output variable is the total nitrogen in the effluent, and the corresponding input variables are six variables: influent ammonia nitrogen, oxidation-reduction potential in the anaerobic tank, oxidation-reduction potential in the anoxic tank, nitrate nitrogen in the anoxic tank, suspended solids concentration in the mixed liquor of the anoxic tank, and total nitrogen concentration. The data at time k are obtained as follows: influent ammonia nitrogen h1(k), oxidation-reduction potential h2(k) in the anaerobic tank, oxidation-reduction potential h3(k) in the anoxic tank, nitrate nitrogen h4(k) in the anoxic tank, suspended solids concentration h5(k) in the mixed liquor of the anoxic tank, and total nitrogen concentration h6(k). The neural network input vector h(k) = [h1(k), h2(k), h3(k), h4(k), h5(k), h6(k)] T , T denotes the transpose of the matrix;

[0062] (2) Establish an event-triggered fuzzy neural network prediction model

[0063] The event-triggered fuzzy neural network prediction model consists of 6 input layer neurons, 6 fuzzy radial basis function layer neurons, 6 normalization layer neurons, and 1 output layer neuron. The input layer neurons convert the input data into fuzzy values ​​using fuzzy membership functions. The fuzzy radial basis function layer neurons process these fuzzy values ​​and apply fuzzy rules for inference. The normalization layer neurons normalize the fuzzy inference results. The output layer neurons defuzzify the fuzzy output into a clear prediction result. The model's output expression is as follows:

[0064]

[0065] Where ŷ(k+1) is the predicted output of total nitrogen in the effluent at time k+1, w j (k)(j=1,2,…,6) represents the weight at time k, w j (k)=[1.92, -1.77, 2.04, 1.76, -0.93, -0.68], c ij (k)= [c 1j , c 2j , c 3j , c 4j c 5j ,c 6j] = [2.21, 1.06, 1.27, -0.36, -0.85, -0.48, 1.14, -0.48, 1.00, -0.72,1.07, -0.80, 0.97, -2.79, -0.90, 1.37, -0.35, 1.10, -0.60, -2.36, -0.71,1.11, 1.04, -0.66, 0.67, 0.95, 0.47, -0.69, 1.05, 1.01, -0.55, 0.98, 0.39,0.59, 0.24, -0.84], representing the center value of the input layer corresponding to the j-th neuron of the radial basis function layer at time k, σ ij (k) =[σ 1 j ,σ 2 j ,σ 3 j , σ 4 j , σ 5 j , σ 6 j ]=[1.63, 0.93, 2.03, 1.17, 1.33, 1.00, 1.81, 1.25, 1.30, 1.07, 1.64, 1.44, 2.25, 0.77, 0.43, 0.66, 1.06, 1.24, 1.06, 0.75, 1.21, 0.82, 1.62, 1.38, 0.59, 1.02, 0.88, 0.69, 0.47, 0.35, 1.01, 0.29, 1.03, 1.69, 0.99, 1.88], representing the width value of the j-th neuron in the radial basis function layer corresponding to the i-th neuron in the input layer at time k;

[0066] The parameter set for constructing the event-triggered neural network is as follows:

[0067]

[0068] in, , ,w (k)= [w1(k), w2(k), w3(k), w4(k), w5(k),w6(k)] T These represent the center value parameter set, width parameter set, and weight parameter set of the neural network, respectively.

[0069] The event-triggered parameter update process of the fuzzy neural network is as follows:

[0070] ① Calculate the error function E(k) at time k:

[0071]

[0072] Where e(k) = y(k) - ŷ(k) represents the error between the actual total nitrogen concentration and the predicted total nitrogen concentration of the system, y(k) is the true total nitrogen concentration obtained by sampling at time k, and ŷ(k) is the predicted output of the total nitrogen in the effluent at time k;

[0073] ② Calculate the error increment and change in increment at time k based on the error function E(k) at time k, as follows:

[0074]

[0075]

[0076]

[0077] Where F3(k) represents the error increment between the error function E(k) at time k and the error function E(k-3) at time k-3, G3(k) represents the increment change between the difference function F3(k) at time k and the difference function F3(k-3) at time k-3, and H3(k) represents the absolute value of the error increment at time k.

[0078] ③ Define the event state Con(k) as follows:

[0079]

[0080] In this context, state C1 indicates that the prediction error is decreasing rapidly, the data quality is excellent, and the model is quickly learning effective features; state C2 indicates that the prediction error is decreasing at a slower pace, the data quality is relatively good, the model's optimization space is narrowing, but there is still room for optimization; state C3 indicates that the prediction error is increasing, the data quality is poor, and further data feature detection mechanisms are needed to determine whether it is outlier interference or a sudden change in system operating conditions; state C4 indicates that the prediction error is very small at this point, and no neural network parameter update is required.

[0081] ④ Design a data feature detection mechanism

[0082] When C3 is triggered, a vector is constructed. Obtain the magnitude of historical neural network parameter changes:

[0083]

[0084] in (n=0, 1, …, 9) is used to quantify the magnitude of change in the total set of parameters triggered by the event at time kn:

[0085]

[0086] Among them, ||.|| F Denotes the F-norm;

[0087] Calculate the mean of the change and the standard deviation of the range of change as follows:

[0088]

[0089]

[0090] Calculate the virtual parameter update magnitude based on the current data :

[0091]

[0092] in To update the parameter set virtually, the following method is used:

[0093]

[0094] Where I represents the identity matrix;

[0095] To accurately distinguish between data feature changes caused by changes in operating conditions and data feature changes caused by outliers, a feature detection identifier is defined. as follows:

[0096]

[0097] Among them, when the conditions are met simultaneously and At that time, the feature detection identifier will be... Set to 1; otherwise, set the feature detection flag. Set it to 0 to determine that the change in prediction error was caused by a change in the data features themselves.

[0098] ⑤ Neural Network Parameter Update Strategy

[0099] The neural network parameters are updated as follows:

[0100]

[0101]

[0102]

[0103] Where U1, U2, and U3 represent different neural network parameter update strategies. When the event state triggered at time k is C1, update strategy U1 is executed, and the neural network parameters are updated according to formula (15). 2， Or when the event state triggered at time k is C3 and D1=0, execute update strategy U2 and update the neural network parameters according to formula (16); when the event state triggered at time k is C4, or when the event state triggered at time k is C3 and D1=1, execute update strategy U3 and do not update the neural network parameters.

[0104] (3) Realize intelligent prediction of total nitrogen in effluent based on event-triggered fuzzy neural network

[0105] Obtain the influent ammonia nitrogen data h1(k) at time k, the oxidation-reduction potential data h2(k) in the anaerobic tank, the oxidation-reduction potential data h3(k) in the anoxic tank, the nitrate nitrogen data h4(k) in the anoxic tank, the mixed liquor suspended solids concentration data h5(k) in the anoxic tank, and the total nitrogen concentration data h6(k). Construct the neural network input vector h(k). Input the input vector h(k) to the input layer of the event-triggered fuzzy neural network, and pass it through the fuzzy RBF layer, the normalization layer, and the output layer, as shown in the formula. As shown, the total nitrogen output ŷ(k+1) predicted by the event-triggered fuzzy neural network is obtained, in mg / L. The network parameter update strategy is as follows: -formula As shown, Figure 1 The graph shows the predicted total nitrogen concentration of the system. X-axis: time, in days; Y-axis: total nitrogen concentration value, solid line: actual total nitrogen value; dashed line: predicted total nitrogen value, in milligrams per liter. Figure 2 The system's total nitrogen concentration error graph is shown, with the X-axis representing time in days and the Y-axis representing total nitrogen concentration in milligrams per liter. The results demonstrate the effectiveness of the method.

Claims

1. A method for intelligent prediction of total nitrogen in effluent based on event-driven fuzzy neural networks, characterized in that, Includes the following steps: (1) Determine the input and output variables of the event-triggered fuzzy neural network Taking the wastewater treatment process as the research object, this study predicts the total nitrogen in the effluent. The output variable is the total nitrogen in the effluent, and the corresponding input variables are six variables: influent ammonia nitrogen, oxidation-reduction potential in the anaerobic tank, oxidation-reduction potential in the anoxic tank, nitrate nitrogen in the anoxic tank, suspended solids concentration in the mixed liquor of the anoxic tank, and total nitrogen concentration. The data at time k are obtained as follows: influent ammonia nitrogen h1(k), oxidation-reduction potential h2(k) in the anaerobic tank, oxidation-reduction potential h3(k) in the anoxic tank, nitrate nitrogen h4(k) in the anoxic tank, suspended solids concentration h5(k) in the mixed liquor of the anoxic tank, and total nitrogen concentration h6(k). The neural network input vector h(k) = [h1(k), h2(k), h3(k), h4(k), h5(k), h6(k)] T , T denotes the transpose of the matrix; (2) Establish an event-triggered fuzzy neural network prediction model The event-triggered fuzzy neural network prediction model consists of 6 input layer neurons, 6 fuzzy radial basis function layer neurons, 6 normalization layer neurons, and 1 output layer neuron. The input layer neurons convert the input data into fuzzy values ​​using fuzzy membership functions. The fuzzy radial basis function layer neurons process these fuzzy values ​​and apply fuzzy rules for inference. The normalization layer neurons normalize the fuzzy inference results. The output layer neurons defuzzify the fuzzy output into a clear prediction result. The model's output expression is as follows: ； Where ŷ(k+1) is the predicted output of total nitrogen in the effluent at time k+1, w j (k)(j=1,2,…,6) represents the weight at time k, which is randomly assigned within the range [0,1]. h i (k)(i=1,2,…,6) is the input of the prediction model at time k, c ij (k) represents the center value of the i-th neuron in the input layer corresponding to the j-th neuron in the radial basis function layer at time k, σ ij (k) represents the width value of the i-th neuron in the input layer corresponding to the j-th neuron in the radial basis function layer at time k, c ij (k) and σ ij (k) is randomly assigned a value in the range [0,1]. The parameter set for constructing the event-triggered neural network is as follows: ； in, , ,w (k)= [w1(k), w2(k), w3(k), w4(k), w5(k), w6(k)] T These represent the center value parameter set, width parameter set, and weight parameter set of the neural network, respectively. The event-triggered parameter update process of the fuzzy neural network is as follows: ① Calculate the error function E(k) at time k: ； Where e(k) = y(k) - ŷ(k) represents the error between the actual total nitrogen concentration and the predicted total nitrogen concentration of the system, y(k) is the true total nitrogen concentration obtained by sampling at time k, and ŷ(k) is the predicted output of the total nitrogen in the effluent at time k; ② Calculate the error increment and change in increment at time k based on the error function E(k) at time k, as follows: ； ； ； Where F3(k) represents the error increment between the error function E(k) at time k and the error function E(k-3) at time k-3, G3(k) represents the increment change between the difference function F3(k) at time k and the difference function F3(k-3) at time k-3, and H3(k) represents the absolute value of the error increment at time k. ③ Define the event state Con(k) as follows: ； In this context, state C1 indicates that the prediction error is decreasing rapidly, the data quality is excellent, and the model is quickly learning effective features; state C2 indicates that the prediction error is decreasing at a slower pace, the data quality is relatively good, the model's optimization space is narrowing, but there is still room for optimization; state C3 indicates that the prediction error is increasing, the data quality is poor, and further data feature detection is needed to determine whether it is outlier interference or a sudden change in system operating conditions; state C4 indicates that the prediction error is very small at this point, and no neural network parameter update is required. ④ Design a data feature detection mechanism When C3 is triggered, a vector is constructed. Obtain the magnitude of historical neural network parameter changes: ； in (n=0, 1, …, 9) is used to quantify the magnitude of change in the total set of parameters triggered by the event at time kn: ； Among them, ||.|| F Denotes the F-norm; Calculate the mean of the change and the standard deviation of the range of change as follows: ； ； Calculate the virtual parameter update magnitude based on the current data : ； in To update the parameter set virtually, the following method is used: ； Where I represents the identity matrix; To accurately distinguish between data feature changes caused by changes in operating conditions and data feature changes caused by outliers, feature detection identifiers are defined. as follows: ； Among them, when the conditions are met simultaneously and At that time, the feature detection identifier will be... Set to 1; otherwise, set the feature detection flag. Set it to 0 to determine that the change in prediction error was caused by a change in the data features themselves. ⑤ Neural Network Parameter Update Strategy The neural network parameters are updated as follows: ； ； ； Where U1, U2, and U3 represent different neural network parameter update strategies. When the event state triggered at time k is C1, update strategy U1 is executed, and the neural network parameters are updated according to formula (15). 2， Or when the event state triggered at time k is C3 and D1=0, execute update strategy U2 and update the neural network parameters according to formula (16); when the event state triggered at time k is C4, or when the event state triggered at time k is C3 and D1=1, execute update strategy U3 and do not update the neural network parameters. (3) Realize intelligent prediction of total nitrogen in effluent based on event-triggered fuzzy neural network Obtain the influent ammonia nitrogen data h1(k) at time k, the oxidation-reduction potential data h2(k) in the anaerobic tank, the oxidation-reduction potential data h3(k) in the anoxic tank, the nitrate nitrogen data h4(k) in the anoxic tank, the mixed liquor suspended solids concentration data h5(k) in the anoxic tank, and the total nitrogen concentration data h6(k). Construct the neural network input vector h(k). Input the input vector h(k) to the input layer of the event-triggered fuzzy neural network, and pass it through the fuzzy RBF layer, the normalization layer, and the output layer, as shown in the formula. As shown, the total nitrogen output ŷ(k+1) predicted by the event-triggered fuzzy neural network is obtained, in mg / L. The network parameter update strategy is as follows: -formula As shown.

Images