An intelligent detection method for total nitrogen in effluent based on self-organizing stacked two-type fuzzy neural network

Abstract

The present application provides a kind of effluent total nitrogen intelligent detection method based on self-organizing stacked bivalent fuzzy neural network, realizes the real-time intelligent detection of effluent total nitrogen under multiple operating conditions in wastewater treatment process.For the multiple operating conditions existing in wastewater treatment process, it is impossible to express by using a single fixed mathematical model, therefore, the present application simulates the wastewater treatment process with multiple operating conditions by using stacked bivalent fuzzy neural network.The method establishes an effluent total nitrogen intelligent detection model based on stacked bivalent fuzzy neural network by analyzing wastewater treatment process data, updates the detection model parameters using urban wastewater treatment process data, adjusts the structure of detection model by evaluating the performance of bivalent fuzzy neural network, forms a hierarchical network structure, and realizes the rapid and accurate detection of effluent total nitrogen under multiple operating conditions.

Description

Technical Field

[0001] This invention, based on the operational characteristics of urban wastewater treatment processes, designs an intelligent detection model for total nitrogen in effluent using a self-organized stacked type-II fuzzy neural network. The model's parameters are updated using data from the urban wastewater treatment process. Performance evaluation of the type-II fuzzy neural network guides structural adjustments, forming a hierarchical network structure that enables intelligent detection of total nitrogen in effluent under various operating conditions in urban wastewater treatment. This intelligent detection method for total nitrogen in effluent based on a self-organized stacked type-II fuzzy neural network can achieve real-time detection of total nitrogen in effluent and belongs to the field of water treatment. Background Technology

[0002] With my country's rapid economic growth, the living standards of urban residents have greatly improved, leading to a continuous increase in domestic sewage and a growing severity of water pollution. Water pollution directly threatens public health and ecological balance, drawing widespread attention and becoming an urgent problem to solve. Therefore, sewage treatment has become a crucial measure for addressing water pollution.

[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 designs an intelligent detection method for total nitrogen in effluent based on a self-organizing stacked type-II fuzzy neural network. This method extracts feature variables, establishes an intelligent detection model for total nitrogen in effluent based on a stacked type-II fuzzy neural network, updates the detection model parameters using urban sewage treatment process data, and guides the structural adjustment of the detection model by evaluating the performance of the type-II fuzzy neural network, forming a hierarchical network structure to achieve rapid and accurate detection of total nitrogen in effluent under multiple operating conditions, meeting the actual needs of sewage treatment plants. Summary of the Invention

[0005] This invention provides an intelligent detection method for total nitrogen in effluent based on a self-organizing stacked type II fuzzy neural network. This method selects feature variables related to total nitrogen in effluent based on knowledge in the wastewater treatment field to determine the input variables of the intelligent detection model, establishes an intelligent detection model for total nitrogen in effluent based on a stacked type II fuzzy neural network, and dynamically adjusts the structure of the intelligent detection model using similarity and contribution indicators to form a hierarchical model structure, thereby achieving high-precision detection of total nitrogen in effluent under multiple operating conditions.

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

[0007] A method for intelligent detection of total nitrogen in effluent based on a self-organizing stacked type-II fuzzy neural network is characterized by selecting input variables for the intelligent detection model of total nitrogen in effluent, establishing the intelligent detection model of total nitrogen in effluent based on a stacked type-II fuzzy neural network, dynamically correcting the model parameters and structure, and detecting the total nitrogen concentration in effluent under multiple operating conditions of wastewater treatment, including the following steps:

[0008] (1) Selection of input variables for the intelligent detection model of total nitrogen in effluent

[0009] Taking the urban wastewater treatment process as the research object, chemical oxygen demand (COD), internal recirculation flow rate, oxidation-reduction potential (ORP) of the anaerobic tank, ORP of the anoxic tank, dissolved oxygen in the first aerobic tank, and dissolved oxygen in the second aerobic tank were selected as input variables for the intelligent detection model of total nitrogen in effluent; total nitrogen in effluent was used as the output variable of the intelligent detection model of total nitrogen in effluent; the input and output variables were normalized.

[0010]

[0011] Where b = 1, ..., 7; t = 1, ..., N, N is the total number of samples, U1(t) is the chemical oxygen demand at time t, in milligrams per liter, U 1,min U is the minimum chemical oxygen demand (COD) among all samples. 1,max U0 is the maximum value of chemical oxygen demand (COD) for all samples, and U2(t) is the reflux flow rate at time t, in cubic meters per hour. 2,min U is the minimum value of the internal recirculation flow rate across all samples. 2,max U is the maximum internal reflux flow rate of all samples, and U3(t) is the redox potential of the anaerobic tank at time t, in millivolts. 3,min U is the minimum redox potential of all samples in the anaerobic tank. 3,max Ut is the maximum redox potential of all samples in the anaerobic tank, and U4(t) is the redox potential of the anoxic tank at time t, in millivolts. 4,min U is the minimum redox potential of all samples in the anoxic pool. 4,max Ut is the maximum redox potential of all samples in the anoxic pool, and U5(t) is the dissolved oxygen concentration in the first aerobic pool at time t, in mg / L.5,min U is the minimum dissolved oxygen value among all samples in the first aerobic tank. 5,max Ut represents the maximum dissolved oxygen concentration of all samples in the first aerobic tank, and U6(t) represents the dissolved oxygen concentration in the second aerobic tank at time t, in mg / L. 6,min U is the minimum dissolved oxygen value among all samples from the second aerobic tank. 6,max U7(t) represents the maximum dissolved oxygen value among all samples in the second aerobic tank, and U7(t) represents the total nitrogen in the effluent at time t, in mg / L. 7,min U is the minimum total nitrogen value among all samples of effluent. 7,max The maximum value of total nitrogen in the effluent is given by all samples. u1(t) is the normalized chemical oxygen demand at time t, u2(t) is the normalized internal reflux flow rate at time t, u3(t) is the normalized oxidation-reduction potential of the anaerobic tank at time t, u4(t) is the normalized oxidation-reduction potential of the anoxic tank at time t, u5(t) is the normalized dissolved oxygen concentration of the first aerobic tank at time t, and u6(t) is the normalized dissolved oxygen concentration of the second aerobic tank at time t. The normalized total nitrogen in the effluent at time t;

[0012] (2) Establishment of intelligent detection model for total nitrogen in effluent

[0013] The intelligent detection model for total nitrogen in effluent consists of S interval type II fuzzy neural networks stacked together, where S is an integer greater than or equal to 2. The s-th interval type II fuzzy neural network includes the following five layers, s = 1, 2, ..., S.

[0014] Input layer: Consists of 6 neurons, with input variable u. s (t)=[u1(t),u2(t),…,u6(t)] T T is the value for transpose calculation;

[0015] Membership function layer: consists of 5 neurons, each neuron outputs:

[0016]

[0017] in, It is the upper bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t. It is the lower bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and n sij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, m sij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval fuzzy neural network at time t, σ sij(t) is the width of the j-th neuron of the i-th input of the s-th interval fuzzy neural network at time t, where s = 1, 2, ..., S; i = 1, ..., 6; j = 1, ..., 5;

[0018] Activation layer: Consists of 5 neurons, each neuron outputting:

[0019]

[0020] Where, φ 1sj (t) is the upper bound output of the j-th neuron in the s-th interval type-2 fuzzy neural network at time t, φ 2sj (t) is the lower bound output of the j-th neuron of the s-th interval type-2 fuzzy neural network at time t, where j = 1, ..., 5;

[0021] The consequent layer consists of 2 neurons, and its output is represented as:

[0022]

[0023] Among them, y 1s (t) is the upper bound output of the s-th interval type-2 fuzzy neural network at time t, y 2s (t) is the lower bound output of the s-th interval type-II fuzzy neural network at time t, a sij (t) is the weight coefficient of the j-th neuron at time t for the i-th input;

[0024] Sub-output layer: Consists of 1 neuron, and the output of the output layer is represented as:

[0025] y s (t)=(1-q s (t))y 1s (t)+q s (t)y 2s (t) (8)

[0026] Among them, y s (t) represents the predictor output of the s-th interval type-II fuzzy neural network at time t, and q s (t) is the scaling factor of the s-th interval type-II fuzzy neural network at time t;

[0027] The output of the intelligent detection model for total nitrogen in effluent at time t is:

[0028]

[0029] Where y(t) is the output of the intelligent detection model for total nitrogen in effluent at time t;

[0030] (3) Adjustment of parameters and structure of intelligent detection model for total nitrogen in effluent

[0031] ① Initialize the intelligent detection model for total nitrogen in effluent:

[0032] Let the current time t = 1, the current iteration number g = 1, and the upper center n of the initial time of the type II fuzzy neural network of the s-th interval of the intelligent detection model for total nitrogen in effluent. gsij (1) Randomly select values ​​in the interval [0,1], with the lower center m gsij (1) = 0.8n gsij (1), width σ gsij (1) = 1, weighting coefficient a gsij (1) Randomly select values ​​in the interval [0,1], with a proportionality coefficient q. gs (1) = 0.3, s = 1, ..., S; the initial value of S is set to 2; the maximum number of iterations is Q; Q is an integer greater than or equal to 100;

[0033] ② Update the parameters of the intelligent detection model for total nitrogen in effluent:

[0034] Input the sample at time t into the intelligent detection model for total nitrogen in the effluent, and use formulas (2)-(8) to obtain the predictor output y corresponding to the sample at time t in the g-th iteration. gs (t), and calculate the prediction error of the g-th iteration of the type II fuzzy neural network for the s-th interval at time t. s=1,…,S; The output y of the g-th iteration of the intelligent detection model for total nitrogen in water is calculated using formula (9). g (t);

[0035] The network parameter update formula for the g-th iteration of the type II fuzzy neural network in the s-th interval of the intelligent total nitrogen detection model for effluent is:

[0036]

[0037] Where, n gsij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and m is the upper center of the j-th neuron in the g-th iteration. gsij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. σ gsij (t) is the width of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gsij (t) represents the weight coefficient of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gs (t) is the scaling factor for the g-th iteration of the s-th interval type-II fuzzy neural network at time t, and n gsij (t+1) is the upper center of the j-th neuron of the i-th input in the g-th iteration of the s-th interval type-II fuzzy neural network at time t+1, m gsijc_{ij}^{sg}(t+1) is the lower center of the i-th input and j-th neuron at the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, σ gsij _{ij}^{sg}(t+1) is the width of the i-th input and j-th neuron at the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, a gsij _{ij}^{sg}(t+1) is the weight coefficient of the i-th input and j-th neuron at the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, q^{sg} gs (t+1) is the proportional coefficient at the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, ∂E(t) / ∂m_{ij}^{sg}(t) is E gs (t) versus n gsij (t), the partial derivative of ∂E(t) / ∂c_{ij}^{sg}(t) is E gs (t) versus m gsij (t), the partial derivative of ∂E(t) / ∂σ_{ij}^{sg}(t) is E gs (t) versus σ gsij (t), the partial derivative of ∂E(t) / ∂a_{ij}^{sg}(t) is E gs (t) versus a gsij (t), the partial derivative of ∂E(t) / ∂q^{sg}(t) is E gs (t) versus q gs (t), the partial derivative of; s=1,…,S;

[0038] ③ If time t<N, increase t by 1 and turn to step ②; if time t=N, turn to step ④;

[0039] ④ Input N groups of input samples into the intelligent detection model of total nitrogen in effluent sequentially, and obtain the predicted sub-output y of the g-th iteration of N groups by using formulas (2)-(8) gs (1),…,y gs (N), s=1,…,S; use formula (9) to calculate the output y corresponding to the N groups of samples in the g-th iteration of the intelligent detection model of total nitrogen in effluent g (1),…,y g (N);

[0040] The similarity C_{αβ}^{g} of the α-th interval type-2 fuzzy neural network and the β-th interval type-2 fuzzy neural network at the g-th iteration is gαβ as follows:

[0041]

[0042] where y_{α}^{g} gα (t) is the prediction output of the α-th interval type-2 fuzzy neural network at the g-th iteration at time t, y_{β}^{g} gβ (t) is the prediction output of the β-th interval type-2 fuzzy neural network at the g-th iteration at time t, α=1,…,S-1; β=i+1,…,S;

[0043] The contribution D of the type II fuzzy neural network in the s-th interval during the g-th iteration gs for:

[0044]

[0045] The output error information of the g-th iteration of the intelligent total nitrogen detection model in effluent is as follows:

[0046]

[0047] Among them, F g It is the output error of the g-th iteration of the intelligent detection model for total nitrogen in effluent;

[0048] If the number of iterations g = 1, set the variable n. (g+1)sij (1)=n gsij (t+1), m (g+1)sij (1)=m gsij (t+1), σ (g+1)sij (1)=σ gsij (t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1)=q gs (t+1), set t=1, increase g by 1, and proceed to step ②; if the number of iterations g>1, proceed to step ⑤;

[0049] ⑤ Adjust the structure of the intelligent detection model for total nitrogen in effluent:

[0050] If the β-interval type-II fuzzy neural network of the intelligent detection model for total nitrogen in effluent satisfies the following conditions:

[0051]

[0052] Delete the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent, and set the upper center n of the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent. gβij (t+1)=0, lower center m gβij (t+1)=0, width σ gβij (t+1)=0, weighting coefficient a gβij (t+1)=0, proportionality constant q gβ (t+1)=0, i=1,…,6; j=1,…,5; S decreases by 1; where, C gαβ Let D be the similarity of the α-th interval type-II fuzzy neural network and the β-th interval type-II fuzzy neural network in the g-th iteration. gα The contribution of the α-th interval type-II fuzzy neural network in the g-th iteration is given by the similarity matrix C = [C g12 Cg13 , …, C g(S-1)S , contribution degree matrix D = [D g1 , D g2 , …, D gS ;

[0053] When the intelligent effluent total nitrogen detection model meets the following conditions:

[0054] F g > F g-1 (15)

[0055] the intelligent effluent total nitrogen detection model will add an interval type-2 fuzzy neural network, and S is increased by 1; set the upper center n of the S-th interval type-2 fuzzy neural network in the intelligent effluent total nitrogen detection model gSij (t+1) is randomly valued within the interval [0,1], and the lower center m gSij (t+1) = 0.8n gSij (t+1), the width σ gSij (t+1) = 1, the weight coefficient a gSij (t+1) is randomly valued within the interval [0,1], the proportionality coefficient q gS (t+1) = 0.3, i = 1, …, 6; j = 1, …, 5;

[0056] ⑥ When the number of iterations g < Q, set the variable n (g+1)sij (1) = n gsij (t+1), m (g+1)sij (1) = m gsij (t+1), σ (g+1)sij (1) = σ gsij (t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1) = q gs (t+1), set t = 1, increase g by 1, and turn to step ②; when the number of iterations g ≥ Q, turn to step (4);

[0057] (4) Effluent total nitrogen detection based on self-organizing stacked type-2 fuzzy neural network

[0058] The system collects data online at time t during the wastewater treatment process, including: chemical oxygen demand (COD) U1(t), internal recirculation flow rate U2(t), oxidation-reduction potential (ORP) of the anaerobic tank U3(t), oxidation-reduction potential (ORP) of the anoxic tank U4(t), dissolved oxygen concentration (DOC) of the first aerobic tank U5(t), and dissolved oxygen concentration (DOC) of the second aerobic tank U6(t). These variables are then normalized to obtain the normalized COD u1(t), normalized internal recirculation flow rate u2(t), normalized ORP of the anaerobic tank u3(t), normalized ORP of the anoxic tank u4(t), normalized DOC of the first aerobic tank u5(t), and normalized DOC of the second aerobic tank u6(t) at time t. Therefore, the input variable u for the intelligent detection model of total nitrogen in the effluent at time t is obtained. s (t)=[u1(t),u2(t),…,u6(t)] T ;s=1,…,S;

[0059] The input vector is fed into the input layer of the S interval type II fuzzy neural network of the effluent total nitrogen detection model. After passing through the membership function layer, activation layer, consequent layer and sub-output layer of the effluent total nitrogen detection model, the output value y(t) of the effluent total nitrogen detection model is obtained, which is the normalized effluent total nitrogen detection value. The normalized effluent total nitrogen detection value y(t) is then denormalized to obtain the effluent total nitrogen detection value Y(t) at the current time, in milligrams per liter.

[0060] The inventiveness of this invention is mainly reflected in:

[0061] (1) In view of the cumbersome operation and long detection cycle of total nitrogen detection in wastewater treatment, this invention proposes an intelligent detection method for total nitrogen in effluent, establishes a nonlinear mathematical relationship between key variables of wastewater treatment and total nitrogen in effluent, and realizes real-time detection of total nitrogen in effluent.

[0062] (2) This invention utilizes a stacked structure to propose an intelligent detection method for total nitrogen in effluent based on stacked type II fuzzy neural networks. By allowing multiple type II fuzzy neural networks to learn data features under different working conditions, it dynamically adapts to the detection of total nitrogen in effluent under multiple working conditions of wastewater treatment.

[0063] (3) This invention proposes a self-organizing mechanism to analyze the dynamics of data during wastewater treatment, evaluate the similarity and contribution of interval type II fuzzy neural networks, guide the adjustment of the structure of the intelligent detection model for total nitrogen in effluent, and improve the detection accuracy of total nitrogen in effluent. Attached Figure Description

[0064] Figure 1This is a graph showing the change in the number of interval type II fuzzy neural networks in the intelligent detection method of this invention. The X-axis represents the number of iterations, and the Y-axis represents the number of interval type II fuzzy neural networks, in units.

[0065] Figure 2 This is a graph showing the predicted output results of the training samples of the intelligent detection method of the present invention. The solid line represents the actual value of total nitrogen in the effluent, and the dashed line represents the predicted output value of the intelligent detection model of total nitrogen in the effluent based on a self-organizing stacked type II fuzzy neural network. The X-axis represents the number of training samples, in units of individuals, and the Y-axis represents the predicted output of total nitrogen in the effluent, in milligrams per liter.

[0066] Figure 3 This is a graph showing the prediction error results of the training samples of the intelligent detection method of the present invention. The X-axis represents the number of training samples, in units of individual samples, and the Y-axis represents the prediction error of total nitrogen in the effluent, in milligrams per liter.

[0067] Figure 4 This is a graph showing the predicted output results of the test samples of the intelligent detection method of the present invention. The solid line represents the actual value of total nitrogen in the effluent, and the dashed line represents the predicted output value of the intelligent detection model of total nitrogen in the effluent based on the self-organized stacked type II fuzzy neural network. The X-axis represents the number of test samples, in units of individual samples, and the Y-axis represents the predicted output of total nitrogen in the effluent, in units of milligrams per liter.

[0068] Figure 5 This is a graph showing the prediction error results of the intelligent detection method of the present invention for test samples. The X-axis represents the number of test samples, in units of individual samples, and the Y-axis represents the prediction error of total nitrogen in the effluent, in milligrams per liter. Detailed Implementation

[0069] The experimental data came from actual data from a wastewater treatment plant in 2023. The actual test data of chemical oxygen demand, internal reflux flow rate, oxidation-reduction potential of anaerobic tank, oxidation-reduction potential of anoxic tank, dissolved oxygen of first aerobic tank, dissolved oxygen of second aerobic tank, and total nitrogen in effluent were selected as experimental sample data. After removing abnormal experimental samples, 250 sets of usable data remained. The first 180 sets of data were selected as training data, and the last 70 sets of data were selected as test data.

[0070] The present invention adopts the following technical solution and implementation steps:

[0071] A method for intelligent detection of total nitrogen in effluent based on a self-organizing stacked type-II fuzzy neural network is characterized by selecting input variables for the intelligent detection model of total nitrogen in effluent, establishing the intelligent detection model of total nitrogen in effluent based on a stacked type-II fuzzy neural network, dynamically correcting the model parameters and structure, and detecting the total nitrogen concentration in effluent under multiple operating conditions of wastewater treatment, including the following steps:

[0072] (1) Selection of input variables for the intelligent detection model of total nitrogen in effluent

[0073] Taking the urban wastewater treatment process as the research object, chemical oxygen demand (COD), internal recirculation flow rate, oxidation-reduction potential (ORP) of the anaerobic tank, ORP of the anoxic tank, dissolved oxygen in the first aerobic tank, and dissolved oxygen in the second aerobic tank were selected as input variables for the intelligent detection model of total nitrogen in effluent; total nitrogen in effluent was used as the output variable of the intelligent detection model of total nitrogen in effluent; the input and output variables were normalized.

[0074]

[0075] Where b = 1, ..., 7; t = 1, ..., 180, 180 is the total number of training samples, U1(t) is the chemical oxygen demand at time t, in mg / L, U 1,min U is the minimum chemical oxygen demand (COD) among all samples. 1,max U0 is the maximum value of chemical oxygen demand (COD) for all samples, and U2(t) is the reflux flow rate at time t, in cubic meters per hour. 2,min U is the minimum value of the internal recirculation flow rate across all samples. 2,max U is the maximum internal reflux flow rate of all samples, and U3(t) is the redox potential of the anaerobic tank at time t, in millivolts. 3,min U is the minimum redox potential of all samples in the anaerobic tank. 3,max Ut is the maximum redox potential of all samples in the anaerobic tank, and U4(t) is the redox potential of the anoxic tank at time t, in millivolts. 4,min U is the minimum redox potential of all samples in the anoxic pool. 4,max Ut is the maximum redox potential of all samples in the anoxic pool, and U5(t) is the dissolved oxygen concentration in the first aerobic pool at time t, in mg / L. 5,min U is the minimum dissolved oxygen value among all samples in the first aerobic tank. 5,max Ut represents the maximum dissolved oxygen concentration of all samples in the first aerobic tank, and U6(t) represents the dissolved oxygen concentration in the second aerobic tank at time t, in mg / L. 6,min U is the minimum dissolved oxygen value among all samples from the second aerobic tank. 6,max U7(t) represents the maximum dissolved oxygen value among all samples in the second aerobic tank, and U7(t) represents the total nitrogen in the effluent at time t, in mg / L. 7,min U is the minimum total nitrogen value among all samples of effluent. 7,max The maximum value of total nitrogen in the effluent is given by all samples. u1(t) is the normalized chemical oxygen demand at time t, u2(t) is the normalized internal reflux flow rate at time t, u3(t) is the normalized oxidation-reduction potential of the anaerobic tank at time t, u4(t) is the normalized oxidation-reduction potential of the anoxic tank at time t, u5(t) is the normalized dissolved oxygen concentration of the first aerobic tank at time t, and u6(t) is the normalized dissolved oxygen concentration of the second aerobic tank at time t. The normalized total nitrogen in the effluent at time t;

[0076] (2) Establishment of intelligent detection model for total nitrogen in effluent

[0077] The intelligent detection model for total nitrogen in effluent consists of S interval type II fuzzy neural networks stacked together, where S is an integer greater than or equal to 2. The s-th interval type II fuzzy neural network includes the following five layers, s = 1, 2, ..., S.

[0078] Input layer: Consists of 6 neurons, with input variable u. s (t)=[u1(t),u2(t),…,u6(t)] T T is the value for transpose calculation;

[0079] Membership function layer: consists of 5 neurons, each neuron outputs:

[0080]

[0081] in, It is the upper bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t. It is the lower bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and n sij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, m sij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval fuzzy neural network at time t, σ sij (t) is the width of the j-th neuron of the i-th input of the s-th interval fuzzy neural network at time t, where s = 1, 2, ..., S; i = 1, ..., 6; j = 1, ..., 5;

[0082] Activation layer: Consists of 5 neurons, each neuron outputting:

[0083]

[0084] Where, φ 1sj (t) is the upper bound output of the j-th neuron in the s-th interval type-2 fuzzy neural network at time t, φ 2sj (t) is the lower bound output of the j-th neuron of the s-th interval type-2 fuzzy neural network at time t, where j = 1, ..., 5;

[0085] The consequent layer consists of 2 neurons, and its output is represented as:

[0086]

[0087] Among them, y 1s (t) is the upper bound output of the s-th interval type-2 fuzzy neural network at time t, y2s (t) is the lower bound output of the s-th interval type-II fuzzy neural network at time t, a sij (t) is the weight coefficient of the j-th neuron at time t for the i-th input;

[0088] Sub-output layer: Consists of 1 neuron, and the output of the output layer is represented as:

[0089] y s (t)=(1-q s (t))y 1s (t)+q s (t)y 2s (t) (23)

[0090] Among them, y s (t) represents the predictor output of the s-th interval type-II fuzzy neural network at time t, and q s (t) is the scaling factor of the s-th interval type-II fuzzy neural network at time t;

[0091] The output of the intelligent detection model for total nitrogen in effluent at time t is:

[0092]

[0093] Where y(t) is the output of the intelligent detection model for total nitrogen in effluent at time t;

[0094] (3) Adjustment of parameters and structure of intelligent detection model for total nitrogen in effluent

[0095] ① Initialize the intelligent detection model for total nitrogen in effluent:

[0096] Let the current time t = 1, the current iteration number g = 1, and the upper center n of the initial time of the type II fuzzy neural network of the s-th interval of the intelligent detection model for total nitrogen in effluent. gsij (1) Randomly select values ​​in the interval [0,1], with the lower center m gsij (1) = 0.8n gsij (1), width σ gsij (1) = 1, weighting coefficient a gsij (1) Randomly select values ​​in the interval [0,1], with a proportionality coefficient q. gs (1) = 0.3, s = 1, ..., S; the initial value of S is set to 2; the maximum number of iterations is 400;

[0097] ② Update the parameters of the intelligent detection model for total nitrogen in effluent:

[0098] Input the sample at time t into the intelligent detection model for total nitrogen in the effluent, and use formulas (17)-(23) to obtain the predictor output y corresponding to the sample at time t in the g-th iteration. gs(t), and calculate the prediction error of the g-th iteration of the type II fuzzy neural network for the s-th interval at time t. s=1,…,S; The output y of the g-th iteration of the intelligent detection model for total nitrogen in water is calculated using formula (24). g (t);

[0099] The network parameter update formula for the g-th iteration of the type II fuzzy neural network in the s-th interval of the intelligent total nitrogen detection model for effluent is:

[0100]

[0101] Where, n gsij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and m is the upper center of the j-th neuron in the g-th iteration. gsij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. σ gsij (t) is the width of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gsij (t) represents the weight coefficient of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gs (t) is the scaling factor for the g-th iteration of the s-th interval type-II fuzzy neural network at time t, and n gsij (t+1) is the upper center of the j-th neuron of the i-th input in the g-th iteration of the s-th interval type-II fuzzy neural network at time t+1, m gsij (t+1) is the lower center of the j-th neuron of the i-th input in the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, σ gsij (t+1) is the width of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t+1, and a gsij (t+1) is the weight coefficient of the j-th neuron at the i-th input of the s-th interval type-II fuzzy neural network at time t+1, and q gs (t+1) is the scaling factor for the g-th iteration of the s-th interval fuzzy neural network at time t+1. It is E gs (t) for n gsij The partial derivative of (t), It is E gs (t) for m gsij The partial derivative of (t), It is E gs (t) for σ gsij The partial derivative of (t), It is E gs (t) for a gsijThe partial derivative of (t), It is E gs (t) for q gs The partial derivatives of (t); s = 1, ..., S;

[0102] ③ If time t < 180, increment t by 1 and proceed to step ②; if time t = 180, proceed to step ④.

[0103] ④ Input the 180 sets of input samples into the intelligent detection model for total nitrogen in effluent in sequence, and use formulas (17)-(23) to obtain the predicted sub-output y of the g-th iteration of the 180 sets. gs (1),…,y gs (180), s=1,…,S; The output y corresponding to the 180 samples in the g-th iteration of the intelligent detection model for total nitrogen in water is calculated using formula (24). g (1),…,y g (180);

[0104] The similarity C between the α-th interval type-II fuzzy neural network and the β-th interval type-II fuzzy neural network at the g-th iteration. gαβ for:

[0105]

[0106] Among them, y gα (t) represents the prediction output of the α-th interval type-II fuzzy neural network at time t, in the g-th iteration. gβ (t) represents the predicted output of the g-th iteration of the β-th interval type-II fuzzy neural network at time t, where α = 1, ..., S-1; β = i+1, ..., S;

[0107] The contribution D of the type II fuzzy neural network in the s-th interval during the g-th iteration gs for:

[0108]

[0109] The output error information of the g-th iteration of the intelligent total nitrogen detection model in effluent is as follows:

[0110]

[0111] Among them, F g This is the output error information of the g-th iteration of the total nitrogen detection model in effluent;

[0112] If the number of iterations g = 1, set the variable n. (g+1)sij (1)=n gsij (t+1), m (g+1)sij (1)=m gsij (t+1), σ (g+1)sij (1)=σ gsij(t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1)=q gs (t+1), set t=1, increase g by 1, and proceed to step ②; if the number of iterations g>1, proceed to step ⑤;

[0113] ⑤ Adjust the structure of the intelligent detection model for total nitrogen in effluent:

[0114] If the β-interval type-II fuzzy neural network of the intelligent detection model for total nitrogen in effluent satisfies the following conditions:

[0115]

[0116] Delete the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent, and set the upper center n of the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent. gβij (t+1)=0, lower center m gβij (t+1)=0, width σ gβij (t+1)=0, weighting coefficient a gβij (t+1)=0, proportionality constant q gβ (t+1)=0, i=1,…,6; j=1,…,5; S decreases by 1; where, C gαβ Let D be the similarity of the α-th interval type-II fuzzy neural network and the β-th interval type-II fuzzy neural network in the g-th iteration. gα The contribution of the α-th interval type-II fuzzy neural network in the g-th iteration is given by the similarity matrix C = [C g12 C g13 ,…,C g(S-1)S ], Contribution matrix D = [D g1 D g2 ,…,D gS ];

[0117] If the intelligent detection model for total nitrogen in effluent meets the following conditions:

[0118] F g >F g-1 (30)

[0119] The intelligent detection model for total nitrogen in effluent will add an interval type II fuzzy neural network, and S will increase by 1; the center n of the S-th interval type II fuzzy neural network in the intelligent detection model for total nitrogen in effluent will be set. gSij (t+1) takes a random value in the interval [0,1], with the lower center m. gSij (t+1)=0.8n gSij (t+1), width σ gSij (t+1)=1, weighting coefficient a gSij(t+1) takes a random value in the interval [0,1], and the proportionality coefficient q gS (t+1)=0.3, i=1,…,6; j=1,…,5;

[0120] ⑥ If the number of iterations g < 400, set the variable n (g+1)sij (1)=n gsij (t+1), m (g+1)sij (1)=m gsij (t+1), σ (g+1)sij (1)=σ gsij (t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1)=q gs (t+1), set t=1, increase g by 1, and proceed to step ②; if the number of iterations g≥400, proceed to step (4);

[0121] (4) Total nitrogen detection in effluent based on self-organizing stacked type-II fuzzy neural network

[0122] The system collects data online at time t during the wastewater treatment process, including: chemical oxygen demand (COD) U1(t), internal recirculation flow rate U2(t), oxidation-reduction potential (ORP) of the anaerobic tank U3(t), oxidation-reduction potential (ORP) of the anoxic tank U4(t), dissolved oxygen concentration (DOC) of the first aerobic tank U5(t), and dissolved oxygen concentration (DOC) of the second aerobic tank U6(t). These variables are then normalized to obtain the normalized COD u1(t), normalized internal recirculation flow rate u2(t), normalized ORP of the anaerobic tank u3(t), normalized ORP of the anoxic tank u4(t), normalized DOC of the first aerobic tank u5(t), and normalized DOC of the second aerobic tank u6(t) at time t. Therefore, the input variable u for the intelligent detection model of total nitrogen in the effluent at time t is obtained. s (t)=[u1(t),u2(t),…,u6(t)] T ;s=1,…,S;

[0123] The input vector is fed into the input layer of the S interval type II fuzzy neural network of the effluent total nitrogen detection model. After passing through the membership function layer, activation layer, consequent layer and sub-output layer of the effluent total nitrogen detection model, the output value y(t) of the effluent total nitrogen detection model is obtained, which is the normalized effluent total nitrogen detection value. The normalized effluent total nitrogen detection value y(t) is then denormalized to obtain the effluent total nitrogen detection value Y(t) at the current time, in milligrams per liter.

[0124] The variation in the number of interval type II fuzzy neural networks in intelligent detection methods is as follows: Figure 1As shown, the X-axis represents the number of iterations, and the Y-axis represents the number of interval type II fuzzy neural networks. The predicted output of the training samples of the intelligent detection method is shown below. Figure 2 As shown, the X-axis represents the number of training samples (in units), and the Y-axis represents the predicted output of total nitrogen in the effluent (in milligrams per liter). The solid line represents the actual value of total nitrogen in the effluent, and the dashed line represents the predicted output value of the intelligent detection model based on a self-organizing stacked type II fuzzy neural network. The prediction error results of the training samples of the intelligent detection method are shown below. Figure 3 X-axis: Number of training samples (in units); Y-axis: Prediction error of total nitrogen in effluent (in milligrams per liter); The prediction output of the intelligent detection method for test samples is as follows: Figure 4 As shown, the X-axis represents the number of test samples (in units), and the Y-axis represents the predicted output of total nitrogen in the effluent (in milligrams per liter). The solid line represents the actual value of total nitrogen in the effluent, and the dashed line represents the predicted output value of the intelligent detection model based on a self-organizing stacked type II fuzzy neural network. The prediction error results of the intelligent detection method for the test samples are shown below. Figure 5 X-axis: Number of test samples, in units; Y-axis: Prediction error of total nitrogen in effluent, in milligrams per liter.

Claims

1. A method for intelligent detection of total nitrogen in effluent based on a self-organizing stacked type-II fuzzy neural network, characterized in that, Comprising the following steps: (1) Selection of input variables for intelligent effluent total nitrogen detection model Taking the municipal sewage treatment process as the research object, chemical oxygen demand, internal reflux flow, anaerobic tank oxidation-reduction potential, anoxic tank oxidation-reduction potential, dissolved oxygen in the first aerobic tank and dissolved oxygen in the second aerobic tank are selected as input variables of the intelligent effluent total nitrogen detection model; effluent total nitrogen is taken as the output variable of the intelligent effluent total nitrogen detection model; normalization processing is performed on the input variables and the output variable: Where b = 1, ..., 7; t = 1, ..., N, N is the total number of samples, U1(t) is the chemical oxygen demand at time t, in milligrams per liter, U 1,min U is the minimum chemical oxygen demand (COD) of all samples. 1,max U0 is the maximum value of chemical oxygen demand (COD) for all samples, and U2(t) is the reflux flow rate at time t, in cubic meters per hour. 2,min U is the minimum value of the internal recirculation flow rate across all samples. 2,max U is the maximum internal reflux flow rate of all samples, and U3(t) is the redox potential of the anaerobic tank at time t, in millivolts. 3,min U is the minimum redox potential of all samples in the anaerobic tank. 3,max Ut is the maximum redox potential of all samples in the anaerobic tank, and U4(t) is the redox potential of the anoxic tank at time t, in millivolts. 4,min U is the minimum redox potential of all samples in the anoxic pool. 4,max Ut is the maximum redox potential of all samples in the anoxic pool, and U5(t) is the dissolved oxygen concentration in the first aerobic pool at time t, in mg / L. 5,min U is the minimum dissolved oxygen value among all samples in the first aerobic tank. 5,max Ut represents the maximum dissolved oxygen concentration of all samples in the first aerobic tank, and U6(t) represents the dissolved oxygen concentration in the second aerobic tank at time t, in mg / L. 6,min U is the minimum dissolved oxygen value among all samples from the second aerobic tank. 6,max U7(t) represents the maximum dissolved oxygen value among all samples in the second aerobic tank, and U7(t) represents the total nitrogen in the effluent at time t, in mg / L. 7,min U is the minimum total nitrogen value among all samples of effluent. 7,max The maximum value of total nitrogen in the effluent is given by all samples. u1(t) is the normalized chemical oxygen demand at time t, u2(t) is the normalized internal reflux flow rate at time t, u3(t) is the normalized oxidation-reduction potential of the anaerobic tank at time t, u4(t) is the normalized oxidation-reduction potential of the anoxic tank at time t, u5(t) is the normalized dissolved oxygen concentration of the first aerobic tank at time t, and u6(t) is the normalized dissolved oxygen concentration of the second aerobic tank at time t. The normalized total nitrogen in the effluent at time t; (2) Establishment of intelligent effluent total nitrogen detection model The intelligent effluent total nitrogen detection model is formed by stacking S interval type-2 fuzzy neural networks, wherein S is an integer greater than or equal to 2, the s-th interval type-2 fuzzy neural network comprises the following five layers, and s=1,2,…,S; Input layer: Consists of 6 neurons, with input variable u. s (t)=[u1(t),u2(t),…,u6(t)] T T is the value for transpose calculation; Membership function layer: consisting of 5 neurons, and the output of each neuron is: in, It is the upper bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t. It is the lower bound output of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and n sij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, m sij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval fuzzy neural network at time t, σ sij (t) is the width of the j-th neuron of the i-th input of the s-th interval fuzzy neural network at time t, where s = 1, 2, ..., S; i = 1, ..., 6; j = 1, ..., 5; Activation layer: consisting of 5 neurons, and the output of each neuron is: Where, φ 1sj (t) is the upper bound output of the j-th neuron in the s-th interval type-2 fuzzy neural network at time t, φ 2sj (t) is the lower bound output of the j-th neuron of the s-th interval type-2 fuzzy neural network at time t, where j = 1, ..., 5; Consequent layer: consisting of 2 neurons, and the output of the consequent layer is expressed as: Among them, y 1s (t) is the upper bound output of the s-th interval type-2 fuzzy neural network at time t, y 2s (t) is the lower bound output of the s-th interval type-II fuzzy neural network at time t, a sij (t) is the weight coefficient of the j-th neuron at time t for the i-th input; Sub-output layer: consisting of 1 neuron, and the output of the output layer is expressed as: y s (t)=(1-q s (t))y 1s (t)+q s (t)y 2s (t) (8) Among them, y s (t) represents the predictor output of the s-th interval type-2 fuzzy neural network at time t, q s (t) is the scaling factor of the s-th interval type-II fuzzy neural network at time t; The output of the intelligent effluent total nitrogen detection model at time t is: wherein y(t) is the output of the intelligent effluent total nitrogen detection model at time t; (3) Parameter and structure adjustment of intelligent effluent total nitrogen detection model ① Initialize the intelligent effluent total nitrogen detection model: Let the current time t = 1, the current iteration number g = 1, and the upper center n of the initial time of the type II fuzzy neural network of the s-th interval of the intelligent detection model for total nitrogen in effluent. gsij (1) Randomly select values ​​in the interval [0,1], with the lower center m gsij (1) = 0.8n gsij (1), width σ gsij (1) = 1, weighting coefficient a gsij (1) Randomly select values ​​in the interval [0,1], with a proportionality coefficient q. gs (1) = 0.3, s = 1, ..., S; the initial value of S is set to 2; the maximum number of iterations is Q; Q is an integer greater than or equal to 100; ② Update parameters of the intelligent effluent total nitrogen detection model: Input the sample at time t into the intelligent detection model for total nitrogen in the effluent, and use formulas (2)-(8) to obtain the predictor output y corresponding to the sample at time t in the g-th iteration. gs (t), and calculate the prediction error of the g-th iteration of the type II fuzzy neural network for the s-th interval at time t. The output y of the g-th iteration of the intelligent total nitrogen detection model in water is calculated using formula (9). g (t); The network parameter update formula for the g-th iteration in the s-th interval type-2 fuzzy neural network of the intelligent effluent total nitrogen detection model is: Where, n gsij (t) is the upper center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, and m is the upper center of the j-th neuron in the g-th iteration. gsij (t) is the lower center of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. σ gsij (t) is the width of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gsij (t) represents the weight coefficient of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t, in the g-th iteration. gs (t) is the scaling factor for the g-th iteration of the s-th interval type-II fuzzy neural network at time t, and n gsij (t+1) is the upper center of the j-th neuron of the i-th input in the g-th iteration of the s-th interval type-II fuzzy neural network at time t+1, m gsij (t+1) is the lower center of the j-th neuron of the i-th input in the g-th iteration of the s-th interval type-2 fuzzy neural network at time t+1, σ gsij (t+1) is the width of the j-th neuron at the i-th input of the s-th interval type-2 fuzzy neural network at time t+1, and a gsij (t+1) is the weight coefficient of the j-th neuron at the i-th input of the s-th interval type-II fuzzy neural network at time t+1, and q gs (t+1) is the scaling factor for the g-th iteration of the s-th interval fuzzy neural network at time t+1. It is E gs (t) for n gsij The partial derivative of (t), It is E gs (t) for m gsij The partial derivative of (t), It is E gs (t) for σ gsij The partial derivative of (t), It is E gs (t) for a gsij The partial derivative of (t), It is E gs (t) for q gs The partial derivatives of (t); s = 1, ..., S; ③ If time t<N, increase t by 1 and go to step ②; if time t=N, go to step ④; ④ Input the N sets of input samples into the intelligent detection model for total nitrogen in effluent in sequence, and use formulas (2)-(8) to obtain the predicted sub-output y of the g-th iteration of the N sets. gs (1),…,y gs (N), s=1,…,S; The output y corresponding to the N groups of samples in the g-th iteration of the intelligent detection model for total nitrogen in water is calculated using formula (9). g (1),…,y g (N); The similarity C between the α-th interval type-II fuzzy neural network and the β-th interval type-II fuzzy neural network at the g-th iteration. gαβ for: Among them, y gα (t) represents the prediction output of the α-th interval type-II fuzzy neural network at time t, in the g-th iteration. gβ (t) represents the predicted output of the g-th iteration of the β-th interval type-II fuzzy neural network at time t, where α = 1, ..., S-1; β = i+1, ..., S; The contribution D of the type II fuzzy neural network in the s-th interval during the g-th iteration gs for: The output error information of the intelligent effluent total nitrogen detection model at the g-th iteration is: Among them, F g It is the output error of the g-th iteration of the intelligent detection model for total nitrogen in effluent; If the number of iterations g = 1, set the variable n. (g+1)sij (1)=n gsij (t+1), m (g+1)sij (1)=m gsij (t+1), σ (g+1)sij (1)=σ gsij (t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1)=q gs (t+1), set t=1, increase g by 1, and proceed to step ②; if the number of iterations g>1, proceed to step ⑤; ⑤ Adjust the structure of the intelligent effluent total nitrogen detection model: When the β-th interval type-2 fuzzy neural network of the intelligent effluent total nitrogen detection model satisfies the following conditions: Delete the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent, and set the upper center n of the β-th interval type II fuzzy neural network of the intelligent total nitrogen detection model in effluent. gβij (t+1)=0, lower center m gβij (t+1)=0, width σ gβij (t+1)=0, weighting coefficient a gβij (t+1)=0, proportionality constant q gβ (t+1)=0, i=1,…,6; j=1,…,5; S decreases by 1; where, C gαβ Let D be the similarity of the α-th interval type-II fuzzy neural network and the β-th interval type-II fuzzy neural network in the g-th iteration. gα The contribution of the α-th interval type-II fuzzy neural network in the g-th iteration is given by the similarity matrix C = [C g12 C g13 ,…,C g(S-1)S ], Contribution matrix D = [D g1 D g2 ,…,D gS ]; When the intelligent effluent total nitrogen detection model satisfies the following conditions: F g >F g-1 (15) The intelligent detection model for total nitrogen in effluent will add an interval type II fuzzy neural network, and S will increase by 1; the center n of the S-th interval type II fuzzy neural network in the intelligent detection model for total nitrogen in effluent will be set. gSij (t+1) takes a random value in the interval [0,1], with the lower center m. gSij (t+1)=0.8n gSij (t+1), width σ gSij (t+1)=1, weighting coefficient a gSij (t+1) takes a random value in the interval [0,1], and the proportionality coefficient q gS (t+1)=0.3, i=1,…,6; j=1,…,5; ⑥ If the number of iterations g is less than Q, set the variable n (g+1)sij (1) = n gsij (t+1), m (g+1)sij (1) = m gsij (t+1), σ (g+1)sij (1) = σ gsij (t+1), a (g+1)sij (1) = a gsij (t+1), q (g+1)s (1) = q gs (t+1), set t = 1, increase g by 1, and go to step ②; if the number of iterations g ≥ Q, go to step (4); (4) Effluent total nitrogen detection based on self-organizing stacked type-2 fuzzy neural network The system collects data online at time t during the wastewater treatment process, including: chemical oxygen demand (COD) U1(t), internal recirculation flow rate U2(t), oxidation-reduction potential (ORP) of the anaerobic tank U3(t), oxidation-reduction potential (ORP) of the anoxic tank U4(t), dissolved oxygen concentration (DOC) of the first aerobic tank U5(t), and dissolved oxygen concentration (DOC) of the second aerobic tank U6(t). These variables are then normalized to obtain the normalized COD u1(t), normalized internal recirculation flow rate u2(t), normalized ORP of the anaerobic tank u3(t), normalized ORP of the anoxic tank u4(t), normalized DOC of the first aerobic tank u5(t), and normalized DOC of the second aerobic tank u6(t) at time t. Therefore, the input variable u for the intelligent detection model of total nitrogen in the effluent at time t is obtained. s (t)=[u1(t),u2(t),…,u6(t)] T ;s=1,…,S; Inputting an input vector into the input layers of S interval type-2 fuzzy neural networks of the effluent total nitrogen detection model, and after transmission through the membership function layer, activation layer, consequent layer and sub-output layer of the effluent total nitrogen detection model, obtaining the output value y(t) of the effluent total nitrogen detection model, which is a normalized effluent total nitrogen detection value; performing inverse normalization processing on the normalized effluent total nitrogen detection value y(t) to obtain the detection value Y(t) of effluent total nitrogen at the current time, with the unit of milligram per liter.

Images