A total nitrogen removal amount detection method based on a multi-attribute type-2 fuzzy neural network

Abstract

The present invention proposes a method for detecting the total nitrogen removal amount based on a multi-attribute type-2 fuzzy neural network to achieve online intelligent detection of the total nitrogen removal amount in the sewage treatment process. Due to changes in influent flow rate, influent composition, weather, and operating processes, there are various operating conditions in the sewage treatment process. However, aiming at the sewage treatment process with multiple operating conditions that is difficult to comprehensively model, the present invention designs an intelligent detection method for the total nitrogen removal amount based on a multi-attribute type-2 fuzzy neural network. By establishing an interval type-2 fuzzy neural network total nitrogen removal amount detection model, using multi-attribute fuzzy rules to express the correlation between sewage treatment process variables and the total nitrogen removal amount, and correcting the parameters of the total nitrogen removal amount detection model based on the gradient descent algorithm, accurate detection of the total nitrogen removal amount under multiple conditions is achieved.

Description

Technical Field

[0001] The present invention designs a total nitrogen removal amount detection method based on a multi-attribute type-2 fuzzy neural network, establishes a type-2 fuzzy neural network total nitrogen removal amount detection model, corrects the parameters of the total nitrogen removal amount detection model, and realizes the accurate detection of the total nitrogen removal amount. The intelligent detection of the total nitrogen removal amount is an effective way to monitor the sewage treatment process and is also a basic link for realizing denitrification control. It belongs to both the water treatment field and the control field. Background Art

[0002] The domestic sewage of humans and the wastewater discharged from industrial production are the main sources of water pollution. The wastewater contains various pollutants, such as total phosphorus, total nitrogen, biochemical oxygen demand, and chemical oxygen demand. The increase in the total nitrogen content in water causes large-scale reproduction of microorganisms and algae in the water, directly resulting in a decrease in the oxygen content in the water, leading to the death of fish and other organisms in the water due to a large amount of oxygen deficiency, and affecting the ecological balance. Therefore, the total nitrogen content is an important indicator for measuring the eutrophication and pollution degree of water bodies, and is also an important indicator for evaluating whether the effluent water quality meets the standards. Denitrification treatment is the main means to ensure the water quality meets the standards. Therefore, the total nitrogen removal amount is used to evaluate the nitrogen removal ability of the sewage treatment process and is an important indicator for measuring the quality of urban sewage treatment. At present, according to the calculation principle of the total nitrogen removal amount, the detection of the total nitrogen removal amount requires obtaining the water quality flow rate, the influent total nitrogen content, and the effluent total nitrogen content. Among them, the detection of the total nitrogen content mainly uses the spectrophotometric method, which has the characteristics of high detection accuracy and high detection environment requirements. The spectrophotometric method still has great defects in terms of detection time and is difficult to meet the on-line real-time detection requirements of total nitrogen in sewage, resulting in the inability to quickly evaluate the total nitrogen removal amount in the sewage treatment process. Therefore, it is of great significance to establish a fast and accurate intelligent detection method to model the sewage treatment process and real-time evaluate the total nitrogen removal amount in the sewage treatment process to improve the detection level of the sewage treatment process.

[0003] The present invention designs a total nitrogen removal amount detection method based on a multi-attribute type-2 fuzzy neural network. This method uses an interval type-2 fuzzy neural network as a carrier to establish a detection model, and realizes the accurate detection of the total nitrogen removal amount by correcting the parameters of the total nitrogen removal amount detection model. Summary of the Invention

[0004] The present invention obtains a total nitrogen removal amount detection method based on a multi-attribute type-2 fuzzy neural network. This method is a type-2 fuzzy neural network total nitrogen removal amount detection model, corrects the parameters of the total nitrogen removal amount detection model, ensures the accuracy of extracting the operation characteristics of the sewage treatment process, and improves the detection accuracy of the total nitrogen removal amount.

[0005] The present invention adopts the following technical solutions and implementation steps:

[0006] A total nitrogen removal amount detection method based on a multi-attribute type-2 fuzzy neural network, characterized in that a type-2 fuzzy neural network total nitrogen removal amount detection model is established, the parameters of the total nitrogen removal amount detection model are corrected, and accurate detection of the total nitrogen removal amount is realized, including the following steps:

[0007] (1) Selection of input variables for the total nitrogen removal amount detection model

[0008] Taking the urban sewage treatment process as the research object, select the pH value, anaerobic pond oxidation-reduction potential, dissolved oxygen in the first aerobic pond, dissolved oxygen in the second aerobic pond, and internal reflux flow rate as the input variables of the total nitrogen removal amount detection model; normalize the data of each variable:

[0009]

[0010] Among them, V i,min is the minimum value of the i-th variable, V i,max is the maximum value of the i-th variable, i = 1, 2,..., 6; V1(t) is the pH value at time t, v1(t) is the normalized pH value at time t, V2(t) is the anaerobic pond oxidation-reduction potential at time t, in millivolts, v2(t) is the normalized anaerobic pond oxidation-reduction potential at time t, V3(t) is the dissolved oxygen concentration in the first aerobic pond at time t, in milligrams per liter, v3(t) is the normalized dissolved oxygen in the first aerobic pond at time t, V4(t) is the dissolved oxygen concentration in the second aerobic pond at time t, in milligrams per liter, v4(t) is the normalized dissolved oxygen in the second aerobic pond at time t, V5(t) is the internal reflux flow rate at time t, in cubic meters per hour, v5(t) is the normalized internal reflux flow rate at time t, V6(t) is the total nitrogen removal amount at time t, in kilograms per hour, y0(t) = v6(t) is the normalized total nitrogen removal amount at time t, t = 1, 2,..., D, D is the total number of training samples;

[0011] (2) Establishment of the total nitrogen removal amount detection model

[0012] The type-2 fuzzy neural network total nitrogen removal amount detection model consists of an input layer, three hidden layers, and an output layer. The hidden layer is composed of a membership function layer, an activation layer, and a consequent layer. The specific establishment is as follows:

[0013] The input layer of the detection model consists of 5 neurons, and the input is [v1(t), …, v5(t)] T , T is the transpose;

[0014] The membership function layer of the detection model consists of 8 neurons. The upper bound output and the lower bound output of the m-th input variable of the n-th neuron in this layer are calculated as follows:

[0015]

[0016]

[0017] Among them, is the lower center of the m-th input of the n-th neuron at time t, is the upper center of the m-th input of the n-th neuron at time t, and σ mn (t) is the width of the m-th input of the n-th neuron at time t, and β mn (t) is the average center of the m-th input of the n-th neuron at time t.

[0018] The activation layer of the detection model consists of 8 neurons, and the output of this layer is:

[0019]

[0020]

[0021] Among them, is the upper bound output of the n-th neuron at time t, is the lower bound output of the n-th neuron at time t;

[0022] The consequent layer of the detection model consists of 2 neurons, and the upper bound output y1(t) and the lower bound output y2(t) of this layer are calculated as follows:

[0023]

[0024]

[0025] Among them, a mn (t) is the weight coefficient of the m-th input of the n-th neuron at time t;

[0026] The detection value of the total nitrogen removal amount is the calculation result of the output layer, and the formula is as follows:

[0027] y(t) = q(t)y2(t) + (1 - q(t))y1(t) (8)

[0028] Among them, y(t) is the detection value of the total nitrogen removal amount, and q(t) is the proportionality coefficient of the detection model at time t;

[0029] (3) Calibration of the parameters of the total nitrogen removal amount detection model

[0030] ① Initialize the parameters of the total nitrogen removal amount detection model: The initial upper center is randomly taken in [0, 1], the initial lower center The initial width σ mn (1) = 1, and the initial weight coefficient a mn(1) = 0.5, the initial proportionality coefficient q(1) = 0.3; set the current time t = 1;

[0031] ② The calculation formula for the continuous sensitivity of neurons is as follows:

[0032]

[0033] Among them, H n (t) is the continuous sensitivity of the nth neuron at time t; when t ≤ 20, the identifier L = t; when t > 20, the identifier L = 20; the continuous sensitivity matrix of neurons at time t, H(t) = [H1(t), …, H8(t)], H1(t) is the continuous sensitivity of the 1st neuron at time t, and H8(t) is the continuous sensitivity of the 8th neuron at time t;

[0034] Sort the 8 continuous sensitivities in H(t) at time t from largest to smallest to obtain the permutation order number h n (t) of the nth neuron at time t; if h n (t) ≤ 4, then the nth neuron is a plastic neuron; if h n (t) > 4, then the nth neuron is a stable neuron, n = 1, 2, …, 8;

[0035] ③ The calculation formula for the error information O1(t) of the total nitrogen removal amount detection model is:

[0036]

[0037] The calculation formula for the parameters corresponding to plastic neurons is:

[0038]

[0039] Among them, Ω1(t) is the parameter vector of plastic neurons at time t, and Ω1(t + 1) is the parameter vector corresponding to plastic neurons at time t + 1, is the partial derivative of the parameter vector corresponding to plastic neurons at time t;

[0040] ④ The calculation formula for the parameter offset O2(t) of the total nitrogen removal amount detection model is:

[0041]

[0042] Among them, ||||2 represents the two - norm, β(t) = [β 11 (t),..., β 58 (t)], β(t - 1) = [β 11 (t - 1),..., β 58 (t - 1)], a(t) = [a 11 (t),..., a58 (t)], a(t - 1) = [a 11 (t - 1),..., a 58 (t - 1)]; The parameter calculation formula for stable neurons is as follows:

[0043]

[0044] Among them, max() represents taking the maximum value, min() represents taking the minimum value, Ψ1(t + 1) is the upper center vector of the stable neuron at time t + 1, Ψ1(t) is the upper center vector of the stable neuron at time t, is the partial derivative of the upper center of the stable neuron at time t, Ψ2(t + 1) is the lower center vector of the stable neuron at time t + 1, Ψ2(t) is the lower center vector of the stable neuron at time t, is the partial derivative of the lower center of the stable neuron at time t, Ψ3(t + 1) is the set vector of the width and weight coefficient of the stable neuron at time t + 1, Ψ3(t) is the set of the width and weight coefficient of the stable neuron at time t, is the partial derivative of the set vector of the width and weight coefficient of the stable neuron at time t;

[0045] ⑤ If the time t < D, t is incremented by 1, and it turns to step ②; if the time t ≥ D, the loop ends; thus, the calibration of the parameters of the total nitrogen removal amount detection model is completed;

[0046] (4) Intelligent detection of total nitrogen removal amount

[0047] Online collect the pH value V1(t), anaerobic oxidation-reduction potential V2(t), dissolved oxygen concentration V3(t) in the first aerobic tank, dissolved oxygen concentration V4(t) in the second aerobic tank, and internal reflux flow rate V5(t) during the sewage treatment process at the current time t; Normalize each data to [0, 1], and respectively obtain the normalized pH value v1(t), the normalized anaerobic oxidation-reduction potential v2(t), the normalized dissolved oxygen v3(t) in the first aerobic tank, the normalized dissolved oxygen v4(t) in the second aerobic tank, and the normalized internal reflux flow rate v5(t); Therefore, the input of the total nitrogen removal amount detection model is v(t) = [v1(t), v2(t), …, v5(t)];

[0048] Input the input vector v(t) into the input layer of the total nitrogen removal amount detection model, and after passing through the membership function layer, activation layer, and consequent layer of the total nitrogen removal amount detection model, obtain the output value y(t) of the total nitrogen removal amount detection model, which is the normalized total nitrogen removal amount detection value; Perform anti-normalization processing on the normalized total nitrogen removal amount detection value y(t) to obtain the detection value Y(t) of the total nitrogen removal amount at the current time, with the unit of kg / h.

[0049] The creativity of the present invention is mainly reflected in:

[0050] (1) Aiming at the problem that the total nitrogen removal amount is difficult to obtain in real time, the present invention proposes a detection method for the total nitrogen removal amount based on a type-2 fuzzy neural network to realize real-time detection of the total nitrogen removal amount;

[0051] (2) Aiming at the complex operation characteristics of the sewage treatment process, the present invention proposes a multi-attribute type-2 fuzzy neural network detection model, which uses plastic neurons and stable neurons to comprehensively express the operation information of the sewage treatment process and improve the detection accuracy of the model. Description of the Drawings

[0052] Figure 1 is the error information change diagram of the total nitrogen removal amount detection method of the multi-attribute type-2 fuzzy neural network of the present invention;

[0053] Figure 2 is the parameter offset change diagram of the total nitrogen removal amount detection method of the multi-attribute type-2 fuzzy neural network of the present invention;

[0054] Figure 3 is the total nitrogen removal amount prediction result diagram of the total nitrogen removal amount detection method of the multi-attribute type-2 fuzzy neural network of the present invention, where the solid line is the actual output value of the total nitrogen removal amount, and the dotted line is the predicted output value of the total nitrogen removal amount detection model;

[0055] Figure 4 is the total nitrogen removal amount prediction error diagram of the total nitrogen removal amount detection method of the multi-attribute type-2 fuzzy neural network of the present invention; Detailed Embodiments

[0056] The experimental data comes from the actual data of a sewage treatment plant; the influent flow rate, pH value, chemical oxygen demand, redox potential of the anaerobic tank, redox potential of the anoxic tank, dissolved oxygen of the first aerobic tank, dissolved oxygen of the second aerobic tank, temperature, internal reflux flow rate and total nitrogen removal amount are respectively taken as experimental samples, and 600 groups of available data are left after removing abnormal samples.

[0057] The present invention adopts the following technical solutions and implementation steps:

[0058] A detection method for the total nitrogen removal amount based on a multi-attribute type-2 fuzzy neural network, characterized in that a type-2 fuzzy neural network total nitrogen removal amount detection model is established, the parameters of the total nitrogen removal amount detection model are corrected, and accurate detection of the total nitrogen removal amount is realized, including the following steps:

[0059] (1) Selection of input variables of the total nitrogen removal amount detection model

[0060] Taking the urban sewage treatment process as the research object, the pH value, the oxidation-reduction potential of the anaerobic tank, the dissolved oxygen in the first aerobic tank, the dissolved oxygen in the second aerobic tank, and the internal reflux flow rate are selected as the input variables of the total nitrogen removal detection model; the data of each variable is normalized as follows:

[0061]

[0062] Among them, V i,min is the minimum value of the i-th variable, and V i,max is the maximum value of the i-th variable, where i = 1, 2, …, 6; V1(t) is the pH value at time t, v1(t) is the normalized pH value at time t, V2(t) is the oxidation-reduction potential of the anaerobic tank at time t, in millivolts, v2(t) is the normalized oxidation-reduction potential of the anaerobic tank at time t, V3(t) is the dissolved oxygen concentration in the first aerobic tank at time t, in milligrams per liter, v3(t) is the normalized dissolved oxygen in the first aerobic tank at time t, V4(t) is the dissolved oxygen concentration in the second aerobic tank at time t, in milligrams per liter, v4(t) is the normalized dissolved oxygen in the second aerobic tank at time t, V5(t) is the internal reflux flow rate at time t, in cubic meters per hour, v5(t) is the normalized internal reflux flow rate at time t, V6(t) is the total nitrogen removal at time t, in kilograms per hour, and y0(t) = v6(t) is the normalized total nitrogen removal at time t, where t = 1, 2, …, 600;

[0063] (2) Establishment of the total nitrogen removal detection model

[0064] The type-2 fuzzy neural network total nitrogen removal detection model consists of an input layer, three hidden layers, and an output layer. The hidden layer is composed of a membership function layer, an activation layer, and a consequent layer. The specific establishment is as follows:

[0065] The input layer of the detection model consists of 5 neurons, and the input is [v1(t), …, v5(t)] T , where T is the transpose;

[0066] The membership function layer of the detection model consists of 8 neurons. The upper bound output and the lower bound output of the m-th input variable of the n-th neuron in this layer are calculated as follows:

[0067]

[0068]

[0069] Among them, is the lower center of the m-th input of the n-th neuron at time t, is the upper center of the m-th input of the n-th neuron at time t, and σ mn(t) is the width of the m-th input of the n-th neuron at time t, β mn (t) is the average center of the m-th input of the n-th neuron at time t,

[0070] The activation layer of the detection model consists of 8 neurons, and the output of this layer is:

[0071]

[0072]

[0073] Among them, is the upper bound output of the n-th neuron at time t, is the lower bound output of the n-th neuron at time t;

[0074] The consequent layer of the detection model consists of 2 neurons. The upper bound output y1(t) and the lower bound output y2(t) of this layer are calculated as follows:

[0075]

[0076]

[0077] Among them, a mn (t) is the weight coefficient of the m-th input of the n-th neuron at time t;

[0078] The measured value of the total nitrogen removal amount is the calculation result of the output layer, and the formula is as follows:

[0079] y(t) = q(t)y2(t) + (1 - q(t))y1(t) (21)

[0080] Among them, y(t) is the measured value of the total nitrogen removal amount, and q(t) is the proportionality coefficient of the detection model at time t;

[0081] (3) Calibration of the parameters of the total nitrogen removal amount detection model

[0082] ① Initialize the parameters of the total nitrogen removal amount detection model: The initial upper center is randomly taken in [0, 1], the initial lower center The initial width σ mn (1) = 1, the initial weight coefficient a mn (1) = 0.5, the initial proportionality coefficient q(1) = 0.3; Set the current time t = 1;

[0083] ② The calculation formula for the continuous sensitivity of neurons is as follows:

[0084]

[0085] Among them, H n(t) is the continuous sensitivity of the nth neuron at time t; when t ≤ 20, the identifier L = t; when t > 20, the identifier L = 20; the continuous sensitivity matrix H(t) of neurons at time t = [H1(t), …, H8(t)], where H1(t) is the continuous sensitivity of the 1st neuron at time t and H8(t) is the continuous sensitivity of the 8th neuron at time t;

[0086] Sort the continuous sensitivities of the 8 neurons in H(t) at time t from largest to smallest to obtain the permutation order number h of the nth neuron at time t n (t); if h n (t) ≤ 4, then the nth neuron is a plastic neuron; if h n (t) > 4, then the nth neuron is a stable neuron, where n = 1, 2, …, 8;

[0087] ③ The calculation formula for the error information O1(t) of the total nitrogen removal amount detection model is:

[0088]

[0089] The calculation formula for the parameters corresponding to plastic neurons is:

[0090]

[0091] where Ω1(t) is the parameter vector of plastic neurons at time t, and Ω1(t + 1) is the parameter vector corresponding to plastic neurons at time t + 1, is the partial derivative of the parameter vector corresponding to plastic neurons at time t;

[0092] ④ The calculation formula for the parameter offset O2(t) of the total nitrogen removal amount detection model is:

[0093]

[0094] where ||||2 represents the two-norm, β(t) = [β 11 (t),..., β 58 (t)], β(t - 1) = [β 11 (t - 1),..., β 58 (t - 1)], a(t) = [a 11 (t),..., a 58 (t)], a(t - 1) = [a 11 (t - 1),..., a 58 (t - 1)]; the calculation formula for the parameters corresponding to stable neurons is:

[0095]

[0096] Among them, max() represents taking the maximum value, min() represents taking the minimum value, Ψ1(t + 1) is the upper center vector of the stable neurons at time t + 1, and Ψ1(t) is the upper center vector of the stable neurons at time t. is the partial derivative of the upper center of the stable neurons at time t, Ψ2(t + 1) is the lower center vector of the stable neurons at time t + 1, and Ψ2(t) is the lower center vector of the stable neurons at time t. is the partial derivative of the lower center of the stable neurons at time t, Ψ3(t + 1) is the set vector of the width and weight coefficients of the stable neurons at time t + 1, and Ψ3(t) is the set of the width and weight coefficients of the stable neurons at time t. is the partial derivative of the set vector of the width and weight coefficients of the stable neurons at time t;

[0097] ⑤ If the time t < D, increase t by 1 and turn to step ②; if the time t ≥ D, end the loop; thus, the calibration of the parameters of the total nitrogen removal amount detection model is completed.

[0098] (4) Intelligent detection of total nitrogen removal amount

[0099] Online collect the pH value V1(t), anaerobic pond oxidation-reduction potential V2(t), dissolved oxygen concentration V3(t) in the first aerobic pond, dissolved oxygen concentration V4(t) in the second aerobic pond, and internal reflux flow rate V5(t) during the sewage treatment process at the current time t; normalize each data to [0, 1], and respectively obtain the normalized pH value v1(t), the normalized anaerobic pond oxidation-reduction potential v2(t), the normalized dissolved oxygen v3(t) in the first aerobic pond, the normalized dissolved oxygen v4(t) in the second aerobic pond, and the normalized internal reflux flow rate v5(t); therefore, the input of the total nitrogen removal amount detection model is v(t) = [v1(t), v2(t), …, v5(t)].

[0100] Input the input vector v(t) into the input layer of the total nitrogen removal amount detection model, and after passing through the membership function layer, activation layer, and consequent layer of the total nitrogen removal amount detection model, obtain the output value y(t) of the total nitrogen removal amount detection model, which is the normalized total nitrogen removal amount detection value; perform anti-normalization processing on the normalized total nitrogen removal amount detection value y(t) to obtain the detection value Y(t) of the total nitrogen removal amount at the current time, with the unit of kg / h.

[0101] Using the total nitrogen removal amount detection method based on the multi-attribute type-2 fuzzy neural network, the error information results are as Figure 1 , X-axis: training samples, unit is piece, Y-axis: error information; the parameter offset results are as Figure 2 , X-axis: training samples, unit is piece, Y-axis: parameter offset; the total nitrogen removal amount prediction results are as Figure 3As shown, the X-axis represents the test samples in units of pieces, and the Y-axis represents the predicted output in units of kg / h. The solid line is the actual value of the total nitrogen removal amount, and the dashed line is the predicted output value of the total nitrogen removal amount; the error between the actual output and the predicted output of the total nitrogen removal amount is as Figure 4 , where the X-axis represents the test samples in units of pieces and the Y-axis represents the prediction error in units of kg / h.

Claims

1. A detection method for total nitrogen removal amount based on a multi-attribute type-2 fuzzy neural network, characterized in that It includes the following steps: (1) Selection of input variables for the total nitrogen removal detection model Taking the urban sewage treatment process as the research object, select pH, redox potential of the anaerobic tank, dissolved oxygen in the first aerobic tank, dissolved oxygen in the second aerobic tank, and internal reflux flow rate as the input variables of the total nitrogen removal detection model; normalize the data of each variable: Among them, V i,min is the minimum value of the i-th variable, and V i,max is the maximum value of the i-th variable, where i = 1, 2, …, 6; V1(t) is the pH value at time t, v1(t) is the normalized pH value at time t, V2(t) is the oxidation-reduction potential of the anaerobic tank at time t in millivolts, v2(t) is the normalized oxidation-reduction potential of the anaerobic tank at time t, V3(t) is the dissolved oxygen concentration of the first aerobic tank at time t in mg / L, v3(t) is the normalized dissolved oxygen of the first aerobic tank at time t, V4(t) is the dissolved oxygen concentration of the second aerobic tank at time t in mg / L, v4(t) is the normalized dissolved oxygen of the second aerobic tank at time t, V5(t) is the internal reflux flow rate at time t in m³ / h, v5(t) is the normalized internal reflux flow rate at time t, V6(t) is the total nitrogen removal amount at time t in kg / h, and y0(t) = v6(t) is the normalized total nitrogen removal amount at time t, where t = 1, 2, …, D and D is the total number of training samples; (2) Establishment of the total nitrogen removal detection model The type-2 fuzzy neural network total nitrogen removal detection model consists of an input layer, three hidden layers, and an output layer. The hidden layer is composed of a membership function layer, an activation layer, and a consequent layer. The specific establishment is as follows: The input layer of the detection model consists of 5 neurons, and the input is [v1(t),…,v5(t)] T , where T represents the transpose; The membership function layer of the detection model consists of 8 neurons. The upper bound output of the m-th input variable of the n-th neuron in this layer and the lower bound output are calculated as follows: Among them, is the lower center of the m-th input of the n-th neuron at time t, is the upper center of the m-th input of the n-th neuron at time t, σ mn (t) is the width of the m-th input of the n-th neuron at time t, β mn (t) is the average center of the m-th input of the n-th neuron at time t, The activation layer of the detection model consists of 8 neurons, and the output of this layer is: where, is the upper bound output of the nth neuron at time t, is the lower bound output of the nth neuron at time t; The consequent layer of the detection model consists of 2 neurons. The upper bound output y1(t) and the lower bound output y2(t) of this layer are calculated as follows: where a mn (t) is the weight coefficient of the m-th input of the n-th neuron at time t; The detected value of the total nitrogen removal amount is the calculation result of the output layer, and the formula is as follows: y(t) = q(t)y2(t) + (1 - q(t))y1(t) (8) Where, y(t) is the detected value of the total nitrogen removal amount, and q(t) is the proportionality coefficient of the detection model at time t; (3) Parameter correction of the total nitrogen removal detection model ① Initialize the parameters of the total nitrogen removal detection model: the initial upper center Randomly take values in [0,1], the initial lower center The initial width σ mn (1) = 1, the initial weight coefficient a mn (1) = 0.5, the initial proportionality coefficient q(1) = 0.3; Set the current time t = 1; ② The calculation formula for the continuous sensitivity of neurons is as follows: Among them, H n (t) is the continuous sensitivity of the nth neuron at time t; when t ≤ 20, the identifier L = t; when t > 20, the identifier L = 20; the continuous sensitivity matrix of neurons at time t, H(t) = [H1(t), …, H8(t)], where H1(t) is the continuous sensitivity of the 1st neuron at time t and H8(t) is the continuous sensitivity of the 8th neuron at time t; Sort the continuous sensitivities of the 8 neurons in H(t) at time t from largest to smallest, and obtain the permutation order number h of the nth neuron at time t n (t); if h n (t) ≤ 4, then the nth neuron is a plastic neuron; if h n (t) > 4, then the nth neuron is a stable neuron, where n = 1, 2, …, 8; ③ The calculation formula for the error information O1(t) of the total nitrogen removal detection model is: The calculation formula for the parameters corresponding to the plastic neurons is: Among them, Ω1(t) is the parameter vector of the plastic neuron at time t, and Ω1(t + 1) is the parameter vector corresponding to the plastic neuron at time t + 1. is the partial derivative of the parameter vector corresponding to the plastic neuron at time t. ④ The calculation formula for the parameter offset O2(t) of the total nitrogen removal detection model is: Among them, ||||2 represents the second norm, β(t) = [β 11 (t),..., β 58 (t)], β(t - 1) = [β 11 (t - 1),..., β 58 (t - 1)], a(t) = [a 11 (t),..., a 58 (t)], a(t - 1) = [a 11 (t - 1),..., a 58 (t - 1)]; The calculation formula for the parameters corresponding to the stable neurons is: Among them, max() represents taking the maximum value, min() represents taking the minimum value, Ψ1(t + 1) is the upper center vector of the stable neuron at time t + 1, and Ψ1(t) is the upper center vector of the stable neuron at time t. is the partial derivative of the upper center of the stable neuron at time t, Ψ2(t + 1) is the lower center vector of the stable neuron at time t + 1, and Ψ2(t) is the lower center vector of the stable neuron at time t. is the partial derivative of the lower center of the stable neuron at time t, Ψ3(t + 1) is the set vector of the width and weight coefficient of the stable neuron at time t + 1, and Ψ3(t) is the set of the width and weight coefficient of the stable neuron at time t. is the partial derivative of the set vector of the width and weight coefficient of the stable neuron at time t; ⑤ If the time t < D, t is incremented by 1, and go to step ②; if the time t ≥ D, the loop ends; thus, the parameter correction of the total nitrogen removal detection model is completed; (4) Intelligent detection of the total nitrogen removal amount Online collect the pH V1(t), redox potential V2(t) of the anaerobic tank, dissolved oxygen concentration V3(t) in the first aerobic tank, dissolved oxygen concentration V4(t) in the second aerobic tank, and internal reflux flow rate V5(t) during the sewage treatment process at the current time t; normalize each data to [0, 1], and obtain the normalized pH v1(t), the normalized redox potential v2(t) of the anaerobic tank, the normalized dissolved oxygen v3(t) in the first aerobic tank, the normalized dissolved oxygen v4(t) in the second aerobic tank, and the normalized internal reflux flow rate v5(t) respectively; therefore, the input of the total nitrogen removal detection model is v(t) = [v1(t), v2(t),..., v5(t)]; Input the input vector v(t) into the input layer of the total nitrogen removal detection model, and after passing through the membership function layer, activation layer, and consequent layer of the total nitrogen removal detection model, obtain the output value y(t) of the total nitrogen removal detection model, which is the normalized detected value of the total nitrogen removal amount; perform anti-normalization processing on the normalized detected value y(t) of the total nitrogen removal amount to obtain the detected value Y(t) of the total nitrogen removal amount at the current time, with the unit kg / hour.

Images