A Multi-Operating Condition Double-Layer Optimal Control Method for Sewage Treatment Process Based on Task Clustering

Abstract

A multi-condition double-layer optimization control method for sewage treatment process based on task clustering belongs to the field of sewage treatment. In order to achieve multi-condition double-layer optimization control in the sewage treatment process, the present invention establishes a data-based multi-condition double-layer optimization model for the sewage treatment process, which respectively describes the relationship between the optimized set value and the double-layer optimization objectives of each condition, including the greenhouse gas emission model of the leadership optimization objective and the operating energy consumption model of the follower layer optimization objective in each condition. It studies an optimization setting method based on task clustering, solves the optimized set values of dissolved oxygen and nitrate nitrogen in the sewage treatment process, and designs a fuzzy neural network controller to complete the tracking control of the optimized set value, thereby promoting the multi-condition double-layer operation optimization control of the sewage treatment process.

Description

Technical Field

[0001] The present invention designs a multi-condition double-layer optimization control method for a sewage treatment process based on task clustering based on the analysis of the operation characteristics of the sewage treatment process, wherein a multi-condition double-layer optimization model for a sewage treatment process based on data is established, and the relationship between the optimization setting value and the double-layer optimization target of each condition is described respectively, including the greenhouse gas emission model of the leading layer optimization target and the operating energy consumption model of the following layer optimization target in each condition, and the optimization setting method based on task clustering is studied, and the optimization setting values of dissolved oxygen and nitrate nitrogen in the sewage treatment process are solved, and a fuzzy neural network controller is designed to complete the tracking control of the optimization setting value, so as to promote the multi-condition double-layer operation optimization control of the sewage treatment process. This multi-condition double-layer optimization control method for a sewage treatment process based on task clustering belongs to the field of water treatment. Background Art

[0002] The sewage treatment process is actually a carbon emission process. A large amount of greenhouse gases such as carbon dioxide, methane and nitrous oxide will be emitted during the sewage treatment process. Carbon emissions from the sewage treatment industry account for about 1% of the total emissions of the whole society, which accounts for the largest proportion in the environmental protection industry. Therefore, how to reduce greenhouse gas emissions in the sewage treatment process while ensuring that the effluent quality meets the standards and reduce the operating costs of the sewage treatment plant is a challenging problem. However, the sewage treatment process has a variety of working conditions with different characteristics. Establishing a two-level optimization model for the sewage treatment process for different working conditions is of great significance for accurately describing the relationship between key variables and the two-level optimization objectives of different working conditions; at the same time, how to design an optimization control strategy to achieve multi-condition two-level optimization control of the sewage treatment process is the key to achieving energy conservation and emission reduction optimization operation of the sewage treatment process.

[0003] The present invention designs a multi-condition two-layer optimization control method for a sewage treatment process based on task clustering. It mainly establishes a multi-condition two-layer optimization model for a sewage treatment process based on data, studies the multi-condition optimization setting method for a sewage treatment process based on task clustering, and designs a fuzzy neural network controller to complete the optimization setting value tracking control, thereby realizing the multi-condition two-layer operation optimization control of the sewage treatment process. Summary of the invention

[0004] The present invention obtains a multi-condition two-layer optimization control method for a sewage treatment process based on task clustering. The method establishes a multi-condition two-layer optimization model for a sewage treatment process based on data, obtains the relationship between the optimization setting value and the two-layer optimization target of each condition, studies the optimization setting method based on task clustering to solve the optimization setting values of dissolved oxygen and nitrate nitrogen in the sewage treatment process, and designs a fuzzy neural network controller to complete the optimization setting value tracking control, thereby realizing multi-condition two-layer operation optimization control of the sewage treatment process.

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

[0006] 1. A multi-condition double-layer optimization control method for sewage treatment process based on task clustering, characterized in that a data-based multi-condition double-layer optimization model for sewage treatment process is established, an optimization setting method based on task clustering is studied, and a fuzzy neural network controller is designed to complete the tracking control of the optimization setting value, specifically including the following steps:

[0007] (1) Design of a data-based multi-condition double-layer optimization model for sewage treatment process

[0008] Using the operation data to deeply analyze the operation trend of sewage treatment, and obtaining various operation conditions with different influent water volume characteristics according to different time periods; specifically divided into 4 conditions: Condition 1 is the water reduction condition from 3 am to 7 am, Condition 2 is the small water volume condition from 7 am to 10 am, Condition 3 is the water increase condition from 10 am to 12 noon, and Condition 4 is the large water volume condition from 12 noon to 3 am;

[0009] The multi-condition double-layer optimization model for sewage treatment process uses a data-driven method to respectively describe the relationship between the optimization setting value and the double-layer optimization objectives of each condition, including the greenhouse gas emission model of the leadership optimization objective and the operation energy consumption model of the follower layer optimization objective in each condition;

[0010] Establish a data-based greenhouse gas emission model for the leadership optimization objective of the sewage treatment process:

[0011]

[0012] Among them, f u,n (t) is the leadership greenhouse gas emission model of the nth condition of the sewage treatment process at time t, n = 1, 2, 3, 4; A 1,n (t) is the output offset of the greenhouse gas emission model f u,n (t) of the nth condition of the sewage treatment process at time t and its value range is [-2, 2], P is the number of kernel functions of the double-layer optimization model of the nth condition of the sewage treatment process and its value range is [2, 20], W 1,p,n (t) is the weight of the pth kernel function of the greenhouse gas emission model of the nth condition of the sewage treatment process at time t and its value range is [-3, 3], L 1,p,n (t) is the kernel function related to the greenhouse gas emission model of the nth condition of the sewage treatment process:

[0013]

[0014] Among them, x(t) = [x u (t), x l(t) is the input variable at time t, x u (t) is the decision variable of the optimization objective of the leadership level at time t, the dissolved oxygen concentration S at the aerobic end O (t), and its value range is [0, 3], unit mg / L; x l (t) is the decision variable of the optimization objective of the following layer at time t, the nitrate nitrogen concentration S at the anaerobic end NO (t), and its value range is [0, 2], unit mg / L; c 1,p,n (t) = [c 1,p,n,1 (t), c 1,p,n,2 (t)] is the center of the p-th kernel function of the greenhouse gas emission model of the n-th working condition in the sewage treatment process at time t, and c 1,p,n,1 (t) and c 1,p,n,2 (t) both have a value range of [-1, 1], σ 1,p,n (t) is the width of the p-th kernel function of the greenhouse gas emission model of the n-th working condition in the sewage treatment process at time t, and its value range is [0, 2];

[0015] Establish a data-based operating energy consumption model for the optimization objective of the following layer in the sewage treatment process:

[0016]

[0017] Among them, f l,n (t) is the following layer operating energy consumption model of the n-th working condition in the sewage treatment process at time t, A 2,n (t) is the output offset of the operating energy consumption model f l,n (t) of the n-th working condition in the sewage treatment process at time t, and its value range is [-2, 2], W 2,p,n (t) is the weight of the p-th kernel function of the operating energy consumption model of the n-th working condition in the sewage treatment process at time t, and its value range is [-3, 3], L 2,p,n (t) is the kernel function related to the energy consumption model of the n-th working condition in the sewage treatment process operation:

[0018]

[0019] Among them, c 2,p,n (t) = [c 2,p,n,1 (t), c 2,p,n,2 (t)] is the center of the p-th kernel function of the energy consumption model of the n-th working condition in the sewage treatment process at time t, and c 2,p,n,1 (t) and c 2,p,n,2 (t) both have a value range of [-1, 1]; σ 2,p,n (t) is the width of the p-th kernel function of the energy consumption model of the n-th working condition in the sewage treatment process at time t, and its value range is [0, 2];

[0020] The multi - condition double - layer optimization objective model for the sewage treatment process based on data is as follows:

[0021] minimize f u,n (x u (t),x l (t))

[0022] The constraint is to minimize f l,n (x u (t),x l (t))(5) Constraint where f u,n (x u (t),x l (t)) is the optimization objective model of the leadership level for the n - th condition of the sewage treatment process at time t, f l,n (x u (t),x l (t)) is the optimization objective model of the follower level for the n - th condition of the sewage treatment process at time t, BOD5(t) is the concentration of biochemical oxygen demand in the effluent at time t and its value range is [0, 50], unit mg / L, COD(t) is the concentration of chemical oxygen demand in the effluent at time t and its value range is [0, 250], unit mg / L, MLSS(t) is the concentration of mixed suspended solids in the effluent at time t and its value range is [40, 120], unit mg / L, S NH (t) is the concentration of ammonia nitrogen in the effluent at time t and its value range is [0, 2.5], unit mg / L, S TN (t) is the concentration of total nitrogen in the effluent at time t and its value range is [0, 15], unit mg / L;

[0023] Parameter update of the double - layer optimization objective model for the n - th condition of the sewage treatment process:

[0024]

[0025] where A n (t)=[A 1,n (t),A 2,n (t),W 1,p,n (t),W 2,p,n (t),c 1,p,n (t),c 2,p,n (t),σ 1,p,n (t),σ 2,p,n (t)] are the parameters of the double - layer optimization objective model for the n - th condition at time t; α n is the learning rate of the model parameters for the n - th condition and its value range is [0, 1]; e u,n (t)=y n (t)-y u(t) is the prediction error of the double - layer optimization objective model for the n - th working condition of the sewage treatment process at time t, y n (t)=[f u,n (t), f l,n (t)] is the output of the double - layer optimization objective model for the n - th working condition of the sewage treatment process at time t, y u (t)=[GHG(t), EC(t)] is the actual output value of the sewage treatment process at time t, GHG(t) is the actual greenhouse gas emission of the sewage treatment process at time t, and EC(t) is the actual energy consumption value of the sewage treatment process at time t;

[0026] (2) Solving the optimal set - point based on the task clustering algorithm

[0027] Use the task clustering algorithm to solve the double - layer optimization problem. According to the similarity of the candidate solutions of the leadership layer, simplify all follower - layer optimizations into parallel optimizations of core tasks to achieve the accelerated convergence of the double - layer optimization problem and obtain the optimal set - point;

[0028] ① Set the total number of iterations of the multi - task particle swarm optimization process to γ max =500, the particle swarm size to N = 100, the number of tasks (i.e., the number of clusters) to K, and initialize the external archive U(0) as an empty set;

[0029] ② Take the double - layer optimization objective model of the sewage treatment process based on data as the optimization objective. The evolution starts from the first generation, and use the position information of the particles x t (γ)=[S Ot (γ), S NOt (γ)] as the input;

[0030] ③ Calculate the fitness and skill factor of the particles, divide the particles into different groups according to the skill factor, and sort the particles according to their fitness;

[0031] ④ Calculate the similarity of the candidate solution particles of the leadership layer:

[0032]

[0033] where D j,h (γ) is the similarity between the j - th and h - th candidate solutions of the leadership layer in the γ - th iteration; f l,j (γ) is the follower - layer fitness function corresponding to the j - th particle of the leadership layer in the γ - th iteration, f l,h (γ) is the follower - layer fitness function corresponding to the h - th particle of the leadership layer in the γ - th iteration, σ(·) represents the standard deviation of the fitness sequence, M is the number of fitness values of the follower - layer function and its value range is [10, 50], f m l,j(γ) is the m-th fitness value of the follower layer function corresponding to the j-th particle in the γ-th iteration of the leadership layer, is the average fitness of the follower layer function corresponding to the j-th particle in the γ-th iteration of the leadership layer, d m (γ) is the m-th f l,j (γ)'s fitness to f l,h (γ)'s fitness of the closest distance, is d m (γ)'s mean value;

[0034] ⑤ Design similarity clustering strategy:

[0035]

[0036] Among them, K(γ) is the number of clusters in the γ-th iteration, that is, a multi-task optimization problem containing K optimization tasks is established, and round(·) represents the rounding function; the similarity information of all candidate solutions is sorted in ascending order, and K particles are evenly selected as the initial cluster centers;

[0037] Calculate the similarity between each particle and each cluster center, and classify the particles,

[0038] k = argmin{D i,k (γ)}, k = 1, 2,..., K (9)

[0039] Among them, D i,k (γ) is the similarity between the i-th particle in the leadership layer and the k-th cluster center in the γ-th iteration. The k value corresponding to the minimum similarity between the i-th particle and the cluster center is the category to which the particle belongs, i = 1, 2,..., 100;

[0040] Particle velocity update formula:

[0041]

[0042] Among them, v ti (γ + 1) is the velocity of the i-th particle at time t in the (γ + 1)-th iteration, v ti (γ) is the velocity of the i-th particle at time t in the γ-th iteration; x ti (γ + 1) is the position of the i-th particle at time t in the (γ + 1)-th iteration, x ti (γ) is the position of the i-th particle at time t in the γ-th iteration; is the individual optimal position of the i-th particle at time t in the γ-th iteration, is the global optimal position at time t in the γ-th iteration, is the knowledge transfer term at the γ-th iteration at time t; ω is the inertia weight with a value of 0.8; c1 is the individual experience acceleration constant with a value of 0.25, c2 is the social experience acceleration constant with a value of 0.25, and c3 is the knowledge transfer term acceleration constant with a value of 0.25; r1 is the individual experience random vector with a value range of [0, 1], r2 is the social experience random vector with a value range of [0, 1], and r3 is the knowledge transfer term random vector with a value range of [0, 1];

[0043] ⑥ Compare the individual optimal position at the γ-th iteration at time t with the archive U t in the (γ - 1)-th iteration, U t (γ - 1) = [U 1,t (γ - 1), U 2,t (γ - 1), …, U δ,t (γ - 1), …, U 100,t (γ - 1)], U δ,t (γ - 1) is the δ-th optimal solution in the archive at the (γ - 1)-th iteration at time t, δ = 1, 2, …, 100; F(U δ,t (γ - 1)) is the fitness vector of U δ,t (γ - 1), is the fitness vector of, if then save into the archive to generate the new γ-th generation archive U t (γ); if then the γ-th generation archive U t (γ) is the same as the (γ - 1)-th generation archive U t (γ - 1);

[0044] ⑦ Judge whether to stop the iteration: If the current iteration number γ ≥ γ max , then terminate the iteration process and go to step ⑧, otherwise, increase the iteration number γ by 1 and return to step ③;

[0045] ⑧ Randomly select a solution from the archive U t (γ max ) as the optimized set value u * n (t) = [S * O (t), S * NO (t)], where S * O (t) is the dissolved oxygen optimized set value at the n-th working condition at time t, S * NO(t) is the optimized set value of nitrate nitrogen under the nth operating condition at time t;

[0046] (3) Optimization set value tracking control based on a fuzzy neural network controller

[0047] Design a fuzzy neural network controller to track the optimized set values S * O (t) and S * NO (t) of the dissolved oxygen and nitrate nitrogen concentrations under the nth operating condition, using the real-time control errors of the dissolved oxygen concentration and nitrate nitrogen concentration and the changes in the control errors as inputs, and the oxygen transfer coefficient K L a(t) of the fifth partition and the change in the internal circulation flow rate Q a (t) as outputs;

[0048] The fuzzy neural network controller performs tracking control on the optimized set value S * O (t) and S * NO (t):

[0049]

[0050] where Δu(t) = [ΔQ a (t), ΔK L a(t)] T is the manipulated variable matrix, T is the transpose of a vector or matrix, ΔQ a (t) is the change in the internal circulation flow rate of the sewage treatment, and ΔK L a(t) is the change in the oxygen transfer coefficient of the fifth partition; Z is the number of neurons in the fuzzy neural network and its value range is [2, 20], and θ q (t) = [θ1 q (t), θ2 q (t), …, θ Z q (t)] represents the weights between the output layer and the normalization layer and their value ranges are all [-2, 2], q = 1, 2;

[0051] ψ z (t) represents the normalized output of the zth neuron, z = 1, 2, …, 10,

[0052]

[0053] where ε(t) = [e SO (t), Δe SO (t), e SNO (t), Δe SNO (t)] is the input variable, and e SO\((t)=S * O (t)-S O (t) is the control error between the set value and the actual value of the dissolved oxygen concentration, e SNO (t)=S * NO (t)-S NO (t) is the control error between the set value and the actual value of the nitrate nitrogen concentration, Δe SO (t)=e SO (t + 1)-e SO (t) is the change in the control error of the dissolved oxygen concentration, Δe SNO (t)=e SNO (t + 1)-e SNO (t) is the change in the control error of the nitrate nitrogen concentration, μ z (t)=[μ 1,z (t), μ 2,z (t), μ 3,z (t), μ 4,z (t)] is the center vector of the z-th neuron, and μ 1,z (t), μ 2,z (t), μ 3,z (t) and μ 4,z (t) both have a value range of [-1, 1], b z (t) is the width of the z-th neuron and has a value range of [0, 2];

[0054] Parameter update of the fuzzy neural network controller:

[0055]

[0056] Among them, B(t)=[μ z (t), b z (t), θ 1 (t), θ 2 (t)] are the controller parameters at time t; β is the learning rate of the model parameters and has a value range of [0, 1]; e(t)=y * (t)-y(t) is the control error at time t, y * (t)=[S * O (t), S * NO (t)] is the optimized set value at time t, y(t)=[S O (t), S NO (t)] is the actual output value at time t;

[0057] Adjust the dissolved oxygen transfer coefficient and the internal reflux flow:

[0058] KL a(t + 1)=K L a(t)+△K L a(t) (14)

[0059] Q a (t + 1)=Q a (t)+△Q a (t) (15)

[0060] where K L a(t) is the dissolved oxygen transfer coefficient at time t, and Q a (t) is the internal reflux flow rate at time t; by using a frequency converter to adjust the frequencies of the oxygen supply pump and the reflux pump, the nitrate nitrogen concentration will be adjusted to S * NO (t), and the dissolved oxygen concentration will be adjusted to S * O (t); thus, the multi - condition double - layer optimal control of the sewage treatment process is realized.

[0061] The creativity of the present invention is mainly reflected in:

[0062] Design a multi - condition double - layer optimal control method for the sewage treatment process based on task clustering. Among them, establish a multi - condition double - layer optimal model for the sewage treatment process based on data, respectively describe the relationship between the optimal set value and the double - layer optimal objectives of each condition, including the greenhouse gas emission model of the leadership - level optimal objective and the operating energy consumption model of the follower - level optimal objective in each condition, study the optimal set method based on task clustering, solve the optimal set values of dissolved oxygen and nitrate nitrogen in the sewage treatment process, and design a fuzzy neural network controller to complete the tracking control of the optimal set values, so as to promote the multi - condition double - layer operation optimal control of the sewage treatment process.

[0063] (1) Aiming at the double - layer optimal control problem of multiple conditions in the sewage treatment process, the present invention constructs a multi - condition double - layer optimal model for the sewage treatment process, respectively describes the relationship between the optimal set value and the double - layer optimal objectives of each condition, including the greenhouse gas emission model of the leadership - level optimal objective and the operating energy consumption model of the follower - level optimal objective in each condition, designs an optimal set method for the sewage treatment process based on task clustering, obtains the optimal set values of dissolved oxygen and nitrate nitrogen and conducts tracking control. This method completes the multi - condition double - layer optimal control of the sewage treatment process and realizes the energy conservation and emission reduction of the sewage treatment process.

[0064] (2) The multi - condition two - layer optimization control method for sewage treatment process based on task clustering designed by the present invention optimizes the dissolved oxygen and nitrate nitrogen in the sewage treatment process. This method divides the working conditions of the sewage treatment process according to the influent water volume, takes minimizing greenhouse gas emissions as the optimization problem of the leadership layer, and minimizing the operating energy consumption of the sewage treatment process as the problem of the following layer for multi - condition two - layer optimization. It promotes the accelerated convergence of the two - layer optimization problem through the task clustering strategy, and then obtains the optimized set values of the sewage treatment process. A fuzzy neural network controller is designed to track and control the optimized set values to obtain better control effects. Description of the Drawings

[0065] Figure 1 It is the dissolved oxygen tracking control result diagram of the optimization control method of the present invention

[0066] Figure 2 It is the nitrate nitrogen tracking control result diagram of the optimization control method of the present invention

[0067] Figure 3 It is the greenhouse gas emissions result diagram of the optimization control method of the present invention

[0068] Figure 4 It is the operating energy consumption result diagram of the optimization control method of the present invention Detailed Implementation Modes

[0069] The present invention adopts the following technical solutions and implementation steps:

[0070] A multi - condition two - layer optimization control method for sewage treatment process based on task clustering, characterized in that a data - based multi - condition two - layer optimization model for sewage treatment process is established, an optimization setting method based on task clustering is studied, and a fuzzy neural network controller is designed to complete the tracking control of the optimized set values, specifically including the following steps:

[0071] (1) Design of the data - based multi - condition two - layer optimization model for sewage treatment process

[0072] The operation data is used to deeply analyze the operation trend of sewage treatment. According to different time periods, multiple operation conditions with different influent water volume characteristics are obtained; specifically divided into 4 conditions: Condition 1 is the water - reduction condition from 3:00 am to 7:00 am, Condition 2 is the small - water - volume condition from 7:00 am to 10:00 am, Condition 3 is the water - increase condition from 10:00 am to 12:00 noon, and Condition 4 is the large - water - volume condition from 12:00 noon to 3:00 am;

[0073] The multi - condition two - layer optimization model for sewage treatment process uses the data - driven method to respectively describe the relationship between the optimized set values and the two - layer optimization objectives of each condition, including the greenhouse gas emissions model of the leadership - layer optimization objective and the operation energy consumption model of the following - layer optimization objective in each condition;

[0074] Build a data-based optimization target greenhouse gas emissions model for the leadership level of the sewage treatment process:

[0075]

[0076] where f u,n (t) is the greenhouse gas emissions model of the nth working condition leadership level of the sewage treatment process at time t, n = 1, 2, 3, 4; A 1,n (t) is the output offset of the greenhouse gas emissions model f u,n (t) of the nth working condition of the sewage treatment process at time t and its value range is [-2, 2], A 1,n (0) = -1.21, P is the number of kernel functions of the double-layer optimization model of the nth working condition of the sewage treatment process and its value is 10, W 1,p,n (t) is the weight of the pth kernel function of the greenhouse gas emissions model of the nth working condition of the sewage treatment process at time t and its value range is [-3, 3], W 1,p,n (0) = 1.8, L 1,p,n (t) is the kernel function related to the greenhouse gas emissions model of the nth working condition of the sewage treatment process:

[0077]

[0078] where x(t) = [x u (t), x l (t)] is the input variable at time t, x u (t) is the decision variable of the leadership optimization target, the dissolved oxygen concentration S O (t) at the aerobic end at time t and its value range is [0, 3], unit mg / L, S O (0) = 1.5 mg / L; x l (t) is the decision variable of the follower layer optimization target, the nitrate nitrogen concentration S NO (t) at the anaerobic end at time t and its value range is [0, 2], unit mg / L, S NO (0) = 1 mg / L; c 1,p,n (t) = [c 1,p,n,1 (t), c 1,p,n,2 (t)] is the center of the pth kernel function of the greenhouse gas emissions model of the nth working condition of the sewage treatment process at time t, and c 1,p,n,1 (t) and c 1,p,n,2 (t) both have a value range of [-1, 1], c 1,p,n (0) = [0.51, 0.34] T , σ 1,p,n (t) is the width of the pth kernel function of the greenhouse gas emissions model of the nth working condition of the sewage treatment process at time t and its value range is [0, 2], σ1,p,n f(0) = 1.69;

[0079] Build an operation energy consumption model for optimizing the follow-up layer of the sewage treatment process based on data:

[0080]

[0081] where f l,n f_n(t) is the operation energy consumption model of the nth working condition follow-up layer of the sewage treatment process at time t, and A 2,n A_n(t) is the output offset of the operation energy consumption model f l,n f_n(t) at time t, and its value range is [-2, 2], and A 2,n A_n(0) = -0.97, and W 2,p,n W_{n,p}(t) is the weight of the pth kernel function of the operation energy consumption model of the nth working condition of the sewage treatment process at time t, and its value range is [-3, 3], and W 2,p,n W_{n,p}(0) = 1.48, and L 2,p,n L_n(t) is the kernel function related to the operation energy consumption model of the nth working condition of the sewage treatment process:

[0082]

[0083] where c 2,p,n c_n(t)=[c_{n,p1}(t), c_{n,p2}(t)] is the center of the pth kernel function of the operation energy consumption model of the nth working condition of the sewage treatment process at time t, and c 2,p,n,1 c_{n,p1}(t) and c 2,p,n,2 c_{n,p2}(t) both have a value range of [-1, 1], and c 2,p,n,1 c_{n,p1}(0)=[-0.27, 0.75] 2,p,n,2 ; and σ 2,p,n σ_{n,p}(t) is the width of the pth kernel function of the operation energy consumption model of the nth working condition of the sewage treatment process at time t, and its value range is [0, 2], and σ T ; σ_{n,p}(0)=0.84; 2,p,n 2,p,n

[0084] Construct a multi-working-condition double-layer optimization objective model for the sewage treatment process based on data as:

[0085] minimize f u,n f(x u (t), x l (t))

[0086] Subject to minimize f l,n f(x u (t), x l (t))(20) Subject to where f u,n f(x​​u (t), x l (t)) is the nth working condition leadership optimization objective model of the sewage treatment process at time t, f l,n (x u (t), x l (t)) is the nth working condition follower layer optimization objective model of the sewage treatment process at time t. BOD5(t) is the effluent biochemical oxygen demand concentration at time t and its value range is [0, 50], unit mg / L, BOD5(0) = 25 mg / L. COD(t) is the effluent chemical oxygen demand concentration at time t and its value range is [0, 250], unit mg / L, COD(0) = 125 mg / L. MLSS(t) is the effluent mixed suspended solids concentration at time t and its value range is [40, 120], unit mg / L, MLSS(0) = 75 mg / L, S NH (t) is the effluent ammonia nitrogen concentration at time t and its value range is [0, 2.5], unit mg / L, S NH (0) = 1 mg / L, S TN (t) is the effluent total nitrogen concentration at time t and its value range is [0, 15], unit mg / L, S TN (0) = 12 mg / L;

[0087] Parameter update of the double-layer optimization objective model for the nth working condition of the sewage treatment process:

[0088]

[0089] Among them, A n (t) = [A 1,n (t), A 2,n (t), W 1,p,n (t), W 2,p,n (t), c 1,p,n (t), c 2,p,n (t), σ 1,p,n (t), σ 2,p,n (t)] is the parameter of the double-layer optimization objective model for the nth working condition at time t; α n is the learning rate of the model parameters for the nth working condition and its value range is [0, 1]; e u,n (t) = y n (t) - y u (t) is the prediction error of the double-layer optimization objective model for the nth working condition of the sewage treatment process at time t, y n (t) = [f u,n (t), f l,n (t)] is the output of the double-layer optimization objective model for the nth working condition of the sewage treatment process at time t, y u(t) = [GHG(t), EC(t)] is the actual output value of the sewage treatment process at time t, where GHG(t) is the actual greenhouse gas emissions of the sewage treatment process at time t, and EC(t) is the actual energy consumption value of the sewage treatment process at time t;

[0090] (2) Solving the optimized setpoint based on the task clustering algorithm

[0091] Use the task clustering algorithm to solve the two-layer optimization problem. According to the similarity of the candidate solutions of the leadership layer, simplify all the follower layer optimizations into parallel optimizations of the core tasks to achieve the accelerated convergence of the two-layer optimization problem and obtain the optimized setpoint;

[0092] ① Set the total number of iterations of the multi-task particle swarm optimization process to γ max = 500, the particle swarm size is N = 100, the number of tasks (i.e., the number of clusters) is K, and initialize the external archive U(0) as an empty set;

[0093] ② Use the two-layer optimization objective model of the sewage treatment process based on data as the optimization objective. The evolution will start from the first generation, and use the position information x of the particle t (γ) = [S Ot (γ), S NOt (γ)] as the input;

[0094] ③ Calculate the fitness and skill factor of the particle, divide the particles into different groups according to the skill factor, and sort the particles according to their fitness;

[0095] ④ Calculate the similarity of the candidate solution particles of the leadership layer:

[0096]

[0097] Among them, D j,h (γ) is the similarity between the j-th and h-th candidate solutions of the leadership layer in the γ-th iteration; f l,j (γ) is the follower layer fitness function corresponding to the j-th particle of the leadership layer in the γ-th iteration, f l,h (γ) is the follower layer fitness function corresponding to the h-th particle of the leadership layer in the γ-th iteration, σ(·) represents the standard deviation of the fitness sequence, M is the number of fitness values of the follower layer function and takes the value of 20, f m l,j (γ) is the m-th fitness value of the follower layer function corresponding to the j-th particle of the leadership layer in the γ-th iteration, is the average fitness of the follower layer function corresponding to the j-th particle of the leadership layer in the γ-th iteration, d m (γ) is the closest distance from the fitness of the m-th f l,j (γ) to the fitness of f l,h (γ), is dm The mean value of (γ);

[0098] ⑤ Design a similarity clustering strategy:

[0099]

[0100] Among them, K(γ) is the number of clusters in the γ-th iteration, that is, a multi-task optimization problem including K optimization tasks is established, and round(·) represents the rounding function; the similarity information of all candidate solutions is sorted in ascending order, and K particles are evenly selected as the initial cluster centers;

[0101] Calculate the similarity between each particle and each cluster center, and classify the particles.

[0102] k = argmin{D i,k (γ)}, k = 1, 2,..., K (24)

[0103] Among them, D i,k (γ) is the similarity between the i-th particle and the k-th cluster center of the leadership layer in the γ-th iteration. The k value corresponding to the minimum similarity between the i-th particle and the cluster center is the category to which the particle belongs, i = 1, 2,..., 100;

[0104] Particle velocity update formula:

[0105]

[0106] Among them, v ti (γ + 1) is the velocity of the i-th particle at time t in the (γ + 1)-th iteration, and v ti (γ) is the velocity of the i-th particle at time t in the γ-th iteration; x ti (γ + 1) is the position of the i-th particle at time t in the (γ + 1)-th iteration, and x ti (γ) is the position of the i-th particle at time t in the γ-th iteration; is the individual optimal position of the i-th particle at time t in the γ-th iteration, is the global optimal position at time t in the γ-th iteration, is the knowledge transfer term at time t in the γ-th iteration; ω is the inertia weight and its value is 0.8; c1 is the individual experience acceleration constant and its value is 0.25, c2 is the social experience acceleration constant and its value is 0.25, c3 is the knowledge transfer term acceleration constant and its value is 0.25; r1 is the individual experience random vector and its value range is [0, 1], r2 is the social experience random vector and its value range is [0, 1], r3 is the knowledge transfer term random vector and its value range is [0, 1];

[0107] ⑥ The individual optimal position at time t in the γ-th iteration compared with the archive U at the (γ - 1)-th iteration t with the solutions in U(γ - 1), U(γ - 1) = [U(γ - 1), U(γ - 1), …, U(γ - 1), …, U(γ - 1), …, U(γ - 1)] t (γ - 1) = [U 1,t (γ - 1), U 2,t (γ - 1), …, U δ,t (γ - 1), …, U 100,t (γ - 1)] and U(γ - 1) is the δ-th optimal solution in the archive at the (γ - 1)-th iteration at time t, δ = 1, 2, …, 100; F(U(γ - 1)) is the fitness vector of U(γ - 1), δ,t (γ - 1) is the δ-th optimal solution in the archive at the (γ - 1)-th iteration at time t, δ = 1, 2, …, 100; F(U δ,t (γ - 1)) is the fitness vector of U δ,t (γ - 1), is the fitness vector of, if then save to the archive to generate the new γ-th generation archive U(γ); if t then the γ-th generation archive U (γ) is the same as the (γ - 1)-th generation archive U t (γ - 1); t (γ - 1) same;

[0108] ⑦ Judge whether to stop iteration: If the current iteration number γ ≥ γ max , then terminate the iteration process and go to step ⑧, otherwise, increase the iteration number γ by 1 and return to step ③;

[0109] ⑧ Randomly select a solution from the archive U(γ t ) as the optimal setting value u(t) of the n-th working condition at time t, u(t) = [S max (t), S * n (t)] where S * O (t) is the optimal setting value of dissolved oxygen of the n-th working condition at time t, and S * NO (t) is the optimal setting value of nitrate nitrogen of the n-th working condition at time t; * O (t) is the optimal setting value of dissolved oxygen of the n-th working condition at time t, and S * NO (t) is the optimal setting value of nitrate nitrogen of the n-th working condition at time t;

[0110] (3) Optimal setting value tracking control based on fuzzy neural network controller

[0111] Design a fuzzy neural network controller for the optimal setting values S * O (t) and S * NO(t) Perform tracking control, taking the real-time control errors of dissolved oxygen concentration and nitrate nitrogen concentration and the change of the control error as inputs, and the oxygen transfer coefficient K of the fifth zone L a(t) and the change of the internal circulation flow rate Q a (t) as the outputs;

[0112] The fuzzy neural network controller performs tracking control on the optimized set value S * O (t) and S * NO (t):

[0113]

[0114] Among them, Δu(t) = [ΔQ a (t), ΔK L a(t)] T is the manipulated variable matrix, T is the transpose of a vector or matrix, ΔQ a (t) is the change in the internal circulation flow rate of sewage treatment, ΔK L a(t) is the change in the oxygen transfer coefficient of the fifth zone, Z is the number of neurons of the fuzzy neural network and takes the value of 10; θ q (t) = [θ1 q (t), θ2 q (t), …, θ Z q (t)] represents the weights between the output layer and the normalization layer and their value ranges are all [-2, 2], q = 1, 2, θ q (0) = [0.58, 0.96, 0.25, 0.74, -0.63, 1.38, 1.08, -0.43, -1.07, 1.64];

[0115] ψ z (t) represents the normalized output of the z-th neuron, z = 1, 2, …, 10,

[0116]

[0117] Among them, ε(t) = [e SO (t), Δe SO (t), e SNO (t), Δe SNO (t)] are the input variables, e SO (t) = S * O (t) - S O (t) is the control error between the set value and the actual value of the dissolved oxygen concentration, e SO (0) = 0 mg / L, e SNO (t) = S* NO (t)-S NO (t) is the control error between the set value and the actual value of the nitrate nitrogen concentration, e SNO (0) = 0 mg / L, Δe SO (t) = e SO (t + 1)-e SO (t) is the change in the control error of the dissolved oxygen concentration, Δe SNO (t) = e SNO (t + 1)-e SNO (t) is the change in the control error of the nitrate nitrogen concentration, μ z (t) = [μ 1,z (t), μ 2,z (t), μ 3,z (t), μ 4,z (t)] is the center vector of the z-th neuron, and μ 1,z (t), μ 2,z (t), μ 3,z (t) and μ 4,z (t) both have a value range of [-1, 1], μ z (0) = [0.59, -0.76, 0.51, 0.49], b z (t) is the width of the z-th neuron and has a value range of [0, 2], b z (0) = 0.97;

[0118] Parameter update of the fuzzy neural network controller:

[0119]

[0120] Among them, B(t) = [μ z (t), b z (t), θ 1 (t), θ 2 (t)] are the controller parameters at time t; β is the learning rate of the model parameters and has a value range of [0, 1]; e(t) = y * (t)-y(t) is the control error at time t, y * (t) = [S * O (t), S * NO (t)] is the optimized set value at time t, y(t) = [S O (t), S NO (t)] is the actual output value at time t;

[0121] Adjust the dissolved oxygen transfer coefficient and the internal reflux flow rate:

[0122] KL a(t + 1)=K L a(t)+△K L a(t) (29)

[0123] Q a (t + 1)=Q a (t)+△Q a (t) (30)

[0124] Wherein, K L a(t) is the dissolved oxygen transfer coefficient at time t, Q a (t) is the internal reflux flow rate at time t; By using a frequency converter to adjust the frequencies of the oxygen supply pump and the reflux pump, the nitrate nitrogen concentration will be adjusted to S * NO (t), and the dissolved oxygen concentration will be adjusted to S * O (t). Thus, the multi-condition double-layer optimal control of the sewage treatment process is realized.

[0125] The output results of a multi-condition double-layer optimal control system for sewage treatment process based on task clustering are the dissolved oxygen concentration and the nitrate nitrogen concentration. Figure 1 It is the dissolved oxygen tracking control result graph, where the solid line is the optimized set value and the dashed line is the actual output value. Horizontal axis: time, unit: day; Vertical axis: dissolved oxygen concentration, unit: mg / L. Figure 2 The nitrate nitrogen tracking control result graph, where the solid line is the optimized set value and the dashed line is the actual output value. Horizontal axis: time, unit: day; Vertical axis: nitrate nitrogen concentration, unit: mg / L. Figure 3 It is the greenhouse gas emission result graph. Horizontal axis: time, unit: day; Vertical axis: greenhouse gas emission, unit: kg CO₂ equivalent / m³. Figure 3 It is the operating energy consumption result graph. Horizontal axis: time, unit: day; Vertical axis: operating energy consumption, unit: kWh / day. The experimental results show the effectiveness of the multi-condition double-layer optimal control method for sewage treatment process based on task clustering.

Claims

1. A multi-condition double-layer optimal control method for sewage treatment process based on task clustering, characterized in that, Build a multi-condition double-layer optimization model for the sewage treatment process based on data, study the optimization setting method based on task clustering, and design a fuzzy neural network controller to complete the tracking control of the optimized setting value. The specific steps are as follows: (1) Design of the multi-condition double-layer optimization model for the sewage treatment process based on data Use the operation data to deeply analyze the operation situation of sewage treatment, and obtain various operation conditions with different influent water volume characteristics according to different time periods; The multi-condition double-layer optimization model for the sewage treatment process uses a data-driven method to describe the relationship between the optimized setting value and the double-layer optimization objectives of each condition, including the greenhouse gas emission model of the leadership optimization objective and the operation energy consumption model of the follower layer optimization objective in each condition; Build a greenhouse gas emission model for the leadership optimization objective of the sewage treatment process based on data: Among them, f u,n (t) is the greenhouse gas emission model of the nth working condition leadership level in the sewage treatment process at time t, where n = 1, 2, 3, 4; A 1,n (t) is the output offset of the greenhouse gas emission model f u,n (t) at the nth working condition in the sewage treatment process at time t, and its value range is [-2, 2]. P is the number of kernel functions of the double-layer optimization model in the nth working condition of the sewage treatment process, and its value range is [2, 20]. W 1,p,n (t) is the weight of the pth kernel function of the greenhouse gas emission model at the nth working condition in the sewage treatment process at time t, and its value range is [-3, 3]. L 1,p,n (t) is the kernel function related to the greenhouse gas emission model in the nth working condition of the sewage treatment process: Build an operation energy consumption model for the follower layer optimization objective of the sewage treatment process based on data: Among them, f l,n (t) is the operating energy consumption model of the nth operating condition following layer in the sewage treatment process at time t, A 2,n (t) is the operating energy consumption model of the nth operating condition in the sewage treatment process at time t, f l,n (t) is the output offset of (t) and its value range is [-2, 2], W 2,p,n (t) is the weight of the pth kernel function of the operating energy consumption model of the nth operating condition in the sewage treatment process at time t and its value range is [-3, 3], L 2,p,n (t) is the kernel function related to the energy consumption model of the nth operating condition in the sewage treatment process: Construct a multi-condition double-layer optimization objective model for the sewage treatment process based on data as: minimize f u，n (x u (t), x l (t)) minimize f l,n (x u (t), x l (t)) where f u,n (x u (t), x l (t)) is the optimization objective model of the leadership level for the nth working condition of the sewage treatment process at time t, and f l,n (x u (t), x l (t)) is the optimization objective model of the follower level for the nth working condition of the sewage treatment process at time t; (2) Solve the optimized setting value based on the task clustering algorithm Use the task clustering algorithm to solve the double-layer optimization problem. According to the similarity of the leadership candidate solutions, simplify all follower layer optimizations into parallel optimizations of core tasks to achieve the accelerated convergence of the double-layer optimization problem and obtain the optimized setting value; ① Set the total number of iterations of the multi-task particle swarm optimization process to γ max = 500, the particle swarm size is N = 100, the number of tasks (i.e., the number of clusters) is K, and initialize the external archive U(0) as an empty set; ②Taking the double-layer optimization objective model of the sewage treatment process based on data as the optimization objective, the evolution will start from the first generation, and the position information x of the particles will be used as the input. t (γ) = [S Ot (γ), S NOt (γ)] as the input; ③ Calculate the fitness and skill factor of the particles, divide the particles into different groups according to the skill factor, and sort the particles according to their fitness; ④ Calculate the similarity of the leadership candidate solution particles: Among them, D j,h (γ) is the similarity between the j-th and h-th candidate solutions of the leadership layer in the γ-th iteration; f l,j (γ) is the fitness function of the follower layer corresponding to the j-th particle in the leadership layer in the γ-th iteration, f l,h (γ) is the fitness function of the follower layer corresponding to the h-th particle in the leadership layer in the γ-th iteration, σ(·) represents the standard deviation of the fitness sequence, M is the number of fitness values of the follower layer function and its value range is [10, 50], f m l,j (γ) is the m-th fitness value of the follower layer function corresponding to the j-th particle in the leadership layer in the γ-th iteration, is the average fitness of the follower layer function corresponding to the j-th particle in the leadership layer in the γ-th iteration, d m (γ) is the m-th f l,j (γ)'s fitness to f l,h (γ)'s fitness of the nearest distance, is the mean of d m (γ); ⑤ Design a similarity clustering strategy: Among them, K(γ) is the number of clusters in the γ-th iteration, that is, a multi-task optimization problem containing K optimization tasks is established, and round(·) represents the rounding function; sort the similarity information of all candidate solutions in ascending order, and evenly select K particles as the initial clustering centers; Calculate the similarity between each particle and each clustering center, and classify the particles, k = argmin{D i,k (γ)}, k = 1, 2, ..., K (5) Among them, D i,k (γ) is the similarity between the i-th particle of the leadership layer in the γ-th iteration and the k-th cluster center. The k value corresponding to the minimum similarity between the i-th particle and the cluster center is the category to which the particle belongs, where i = 1, 2, …, 100; (3) Tracking control of the optimized setting value based on the fuzzy neural network controller Design a fuzzy neural network controller to optimize the set values S * O (t) and S * NO (t) for tracking control. Use the real-time control errors of the dissolved oxygen concentration and the nitrate nitrogen concentration and the changes in the control errors as inputs, and the oxygen transfer coefficient K L a(t) and the internal circulation flow rate Q a (t) for changes as outputs; use a frequency converter to adjust the frequencies of the oxygen supply pump and the reflux pump, then the nitrate nitrogen concentration will be adjusted to S * NO (t), and the dissolved oxygen concentration will be adjusted to S * O (t); thus, multi-condition double-layer optimal control of the sewage treatment process is achieved.

2. The method according to claim 1, wherein: Specifically, it is divided into 4 working conditions: Working condition 1 is the water reduction working condition from 3 am to 7 am, Working condition 2 is the small water volume working condition from 7 am to 10 am, Working condition 3 is the water increase working condition from 10 am to 12 noon, and Working condition 4 is the large water volume working condition from 12 noon to 3 am.

3. The method according to claim 1, wherein: where x(t) = [x u (t), x l (t)] is the input variable at time t, x u (t) is the decision variable of the optimization goal of the leadership level at time t, the dissolved oxygen concentration S O (t) at the aerobic end, and its value range is [0, 3], unit mg / L; x l (t) is the decision variable of the optimization goal of the follower level at time t, the nitrate nitrogen concentration S NO (t) at the anaerobic end, and its value range is [0, 2], unit mg / L; c 1,p,n (t) = [c 1,p,n,1 (t), c 1,p,n,2 (t)] is the center of the p-th kernel function of the greenhouse gas emission model in the n-th working condition of the sewage treatment process at time t, and the value ranges of c 1,p,n,1 (t) and c 1,p,n,2 (t) are both [-1, 1], and σ 1,p,n (t) is the width of the p-th kernel function of the greenhouse gas emission model in the n-th working condition of the sewage treatment process at time t, and its value range is [0, 2].

4. The method according to claim 1, wherein: where c 2,p,n (t) = [c 2,p,n,1 (t), c 2,p,n,2 (t)] is the center of the p-th kernel function of the energy consumption model under the n-th working condition of the sewage treatment process at time t, and the value ranges of both c 2,p,n,1 (t) and c 2,p,n,2 (t) are [-1, 1]; σ 2,p,n (t) is the width of the p-th kernel function of the energy consumption model under the n-th working condition of the sewage treatment process at time t and its value range is [0, 2].

5. The method according to claim 1, wherein The constraints of the multi-condition double-layer optimization objective model for the sewage treatment process are as follows: Constraints The constraints of the multi-condition double-layer optimization objective model for the sewage treatment process are as follows: BOD5(t) is the effluent biochemical oxygen demand concentration at time t, and its value range is [0, 50], with the unit of milligram per liter; COD(t) is the effluent chemical oxygen demand concentration at time t, and its value range is [0, 250], with the unit of milligram per liter; MLSS(t) is the effluent mixed suspended solid concentration at time t, and its value range is [40, 120], with the unit of milligram per liter; S NH (t) is the effluent ammonia nitrogen concentration at time t, and its value range is [0, 2.5], with the unit of milligram per liter; S TN (t) is the effluent total nitrogen concentration at time t, and its value range is [0, 15], with the unit of milligram per liter.

6. The method according to claim 1, wherein Parameter update of the double-layer optimization objective model of the n-th working condition of the sewage treatment process: Among them, A n (t) = [A 1,n (t), A 2,n (t), W 1,p,n (t), W 2,p,n (t), c 1,p,n (t), c 2,p,n (t), σ 1,p,n (t), σ 2,p,n (t)] are the parameters of the double - layer optimization objective model for the nth working condition at time t; α n is the learning rate of the model parameters for the nth working condition and its value range is [0, 1]; e u,n (t) = y n (t) - y u (t) is the prediction error of the double - layer optimization objective model for the nth working condition in the sewage treatment process at time t, y n (t) = [f u,n (t), f l,n (t)] is the output of the double - layer optimization objective model for the nth working condition in the sewage treatment process at time t, y u (t) = [GHG(t), EC(t)] is the actual output value in the sewage treatment process at time t, GHG(t) is the actual greenhouse gas emission in the sewage treatment process at time t, and EC(t) is the actual energy consumption value in the sewage treatment process at time t.

7. The method according to claim 1, wherein Particle velocity update formula: where v ti (γ + 1) is the velocity of the i-th particle at the (γ + 1)-th iteration at time t, and v ti (γ) is the velocity of the i-th particle at the γ-th iteration at time t; x ti (γ + 1) is the position of the i-th particle at the (γ + 1)-th iteration at time t, and x ti (γ) is the position of the i-th particle at the γ-th iteration at time t; is the personal best position of the i-th particle at the γ-th iteration at time t, is the global best position at the γ-th iteration at time t, is the knowledge transfer term at the γ-th iteration at time t; ω is the inertia weight with a value of 0.8; c1 is the personal experience acceleration constant with a value of 0.25, c2 is the social experience acceleration constant with a value of 0.25, and c3 is the knowledge transfer term acceleration constant with a value of 0.25; r1 is the personal experience random vector with a value range of [0, 1], r2 is the social experience random vector with a value range of [0, 1], and r3 is the knowledge transfer term random vector with a value range of [0, 1]; The individual optimal position at the γ-th iteration at time t is compared with the archive U at the (γ - 1)-th iteration t in (γ - 1), where U t (γ - 1) = [U 1,t (γ - 1), U 2,t (γ - 1), …, U δ,t (γ - 1), …, U 100,t (γ - 1)], and U δ,t (γ - 1) is the δ-th optimal solution in the archive at the (γ - 1)-th iteration at time t, where δ = 1, 2, …, 100; F(U δ,t (γ - 1)) is the fitness vector of U δ,t (γ - 1), is the fitness vector of, and if then is saved to the archive to generate the new γ-th generation archive U t (γ); if then the γ-th generation archive U t (γ) is the same as the (γ - 1)-th generation archive U t (γ - 1). Determine whether to stop iteration: If the current iteration count γ ≥ γ max , then terminate the iteration process and go to step ⑧; otherwise, increment the iteration count γ by 1 and return to step ③; In the archive U t (γ max ) randomly selects a solution as the optimal setpoint u of the nth operating condition at time t * n (t) = [S * O (t), S * NO (t)], where S * O (t) is the optimal dissolved oxygen setpoint of the nth operating condition at time t, and S * NO (t) is the optimal nitrate nitrogen setpoint of the nth operating condition at time t.

8. The method according to claim 1, wherein The tracking control of the optimized setting value based on the fuzzy neural network controller is specifically: The fuzzy neural network controller performs tracking control on the optimized set value S * O (t) and S * NO (t): where, Δu(t) = [ΔQ a (t), ΔK L a(t)] T is the manipulated variable matrix, T is the transpose of a vector or matrix, ΔQ a (t) is the change in the internal circulation flow rate of sewage treatment, ΔK L a(t) is the change in the oxygen transfer coefficient of the fifth zone; Z is the number of neurons of the fuzzy neural network and its value range is [2, 20]; θ q (t) = [θ1 q (t), θ2 q (t), …, θ Z q (t)] represents the weights between the output layer and the normalization layer and their value ranges are all [-2, 2], q = 1, 2; ψ z (t) represents the normalized output of the z-th neuron, where z = 1, 2, …, 10, where ε(t) = [e SO (t), Δe SO (t), e SNO (t), Δe SNO (t)] is the input variable, e SO (t) = S * O (t) - S O (t) is the control error between the set value and the actual value of the dissolved oxygen concentration, e SNO (t) = S * NO (t) - S NO (t) is the control error between the set value and the actual value of the nitrate nitrogen concentration, Δe SO (t) = e SO (t + 1) - e SO (t) is the change in the control error of the dissolved oxygen concentration, Δe SNO (t) = e SNO (t + 1) - e SNO (t) is the change in the control error of the nitrate nitrogen concentration, μ z (t) = [μ 1,z (t), μ 2,z (t), μ 3,z (t), μ 4,z (t)] is the center vector of the z-th neuron, and μ 1,z (t), μ 2,z (t), μ 3,z (t) and μ 4,z (t) all have a value range of [-1, 1], b z (t) is the width of the z-th neuron and has a value range of [0, 2]; Parameter update of the fuzzy neural network controller: Among them, B(t) = [μ z (t), b z (t), θ 1 (t), θ 2 (t)] are the controller parameters at time t; β is the learning rate of the model parameters and its value range is [0, 1]; e(t) = y * (t) - y(t) is the control error at time t, y * (t) = [S * O (t), S * NO (t)] is the optimized set value at time t, y(t) = [S O (t), S NO (t)] is the actual output value at time t; Adjust the dissolved oxygen transfer coefficient and the internal reflux ratio: K L a(t + 1) = K L a(t) + ΔK L a(t) (13) Q a (t + 1)=Q a (t)+ΔQ a (t)(14) Among them, K L a(t) is the dissolved oxygen transfer coefficient at time t, and Q a (t) is the internal reflux flow rate at time t.

Images