Knowledge and data driven multi-temporal-spatial-scale optimization control method for denitrification process of sewage treatment

Abstract

According to the knowledge and data driven multi-temporal-spatial-scale optimization control method for the sewage treatment denitrification process, optimization control over the dissolved oxygen concentration and the nitrate nitrogen concentration of different spatial galleries through multiple temporal scales is achieved, and the purposes of improving the effluent quality and reducing aeration energy consumption and pumping energy consumption are achieved. The method comprises the following steps: establishing a knowledge and data-driven sewage treatment denitrification process multi-temporal-spatial scale evaluation model, and obtaining a relationship among dissolved oxygen concentration, nitrate nitrogen concentration, effluent quality, aeration energy consumption and pumping energy consumption of multi-temporal scale different spatial galleries; designing a multi-source knowledge collaborative multi-task optimization algorithm to obtain set values of dissolved oxygen concentration and nitrate nitrogen concentration capable of effectively improving the effluent quality and reducing aeration energy consumption and pumping energy consumption; and designing a multivariable proportional-integral-differential controller to realize optimal set value tracking control.

Description

Technical Field

[0001] This invention addresses the optimization of nitrogen removal processes in wastewater treatment by designing a knowledge- and data-driven multi-temporal-scale optimization control method. The method is characterized by: establishing a knowledge- and data-driven multi-temporal-scale evaluation model for the nitrogen removal process; designing a multi-source knowledge-coordinated, multi-objective, multi-task optimization algorithm; and designing a multivariable proportional-integral-derivative controller to achieve optimal setpoint tracking control. This method reduces operational energy consumption while ensuring effluent quality, which is significant for the efficient and stable operation of wastewater treatment processes. This invention belongs to both the field of water research and the field of intelligent optimization control. Background Technology

[0002] With the continuous increase in urban population, the scale of sewage treatment is growing, and the problem of domestic sewage treatment is becoming increasingly prominent. Sewage treatment can purify water quality, improve the ecological environment, and promote water resource recycling, which is of great significance to urban development. As a typical high-energy-consuming industry, sewage treatment plants are consuming increasingly more energy to ensure effluent quality. The denitrification process in sewage treatment oxidizes ammonia nitrogen to nitrate nitrogen, and then reduces it back to nitrogen gas and releases it into the atmosphere. This is an indispensable part of the sewage treatment process. To improve the operational effectiveness and efficiency of sewage treatment, optimized control strategies have been widely applied in the sewage treatment process.

[0003] The goal of optimal control in wastewater treatment denitrification processes is to ensure effluent quality meets standards while reducing operational energy consumption. However, the operational mechanism of wastewater treatment processes is complex, and there is a strong conflict between effluent quality and operational energy consumption. Therefore, balancing the relationship between effluent quality and operational energy consumption to ensure effluent quality meets standards while reducing operational energy consumption is an important research topic. In establishing wastewater quality and energy consumption models, the complex nonlinear dynamics of wastewater treatment processes make it difficult to accurately express the optimization objective model using mechanistic models. Therefore, employing data-driven modeling methods to accurately describe the optimization objectives of wastewater treatment processes is of great significance. Furthermore, multiple objectives conflict with each other in wastewater treatment processes, affecting the quality of the optimal setpoint solution. Simultaneously, the objectives in wastewater treatment processes are dynamically changing, and the speed of solving the optimal setpoint significantly impacts the performance of optimal control. Therefore, designing knowledge- and data-driven multi-temporal-scale optimal control methods for wastewater treatment denitrification processes can not only improve the quality and speed of solving the optimal setpoint, reducing energy consumption while ensuring effluent quality, but is also crucial for the stable and efficient operation of wastewater treatment processes.

[0004] This invention analyzes the characteristics of the nitrogen removal process in wastewater treatment, establishes operational performance indicators and constraints including effluent quality, aeration energy consumption, and pumping energy consumption, and designs a multi-source knowledge collaborative multi-objective multi-task optimization algorithm to obtain effective reductions in aeration and pumping energy consumption and increases in the dissolved oxygen concentration setpoint of the effluent, thus achieving efficient and stable operation of the wastewater treatment aeration process. Summary of the Invention

[0005] This invention proposes a knowledge- and data-driven multi-temporal-scale optimization control method for wastewater treatment denitrification processes. The method is characterized by: establishing a knowledge- and data-driven multi-temporal-scale evaluation model for wastewater treatment denitrification processes; designing a multi-source knowledge collaborative multi-objective multi-task optimization algorithm; designing a multi-variable proportional-integral-derivative controller; and achieving optimized setpoint tracking control.

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

[0007] (1) Establish a knowledge- and data-driven multi-temporal scale evaluation model for wastewater treatment denitrification processes.

[0008] Due to the physical properties of the sensor and the reaction characteristics of the denitrification process in wastewater treatment, the mixed suspended solids concentration (MLSS) is collected at a slow time scale t1 with a sampling period of 2 hours. Other relevant variables of the optimization control model are collected at a fast time scale t2 with a sampling period of 0.5 hours. MLSS is a key variable in the pumping energy consumption optimization control model. Therefore, pumping energy consumption, aeration energy consumption, and effluent water quality are optimized at time t1, and aeration energy consumption and effluent water quality are optimized at time t2.

[0009] At time t1, the optimization control tasks for the wastewater treatment denitrification process include denitrification and nitrification. The optimization objectives for the denitrification task are pumping energy consumption and effluent quality.

[0010]

[0011] Among them, J PE (z PE (t1)) is the pumping energy consumption evaluation model at time t1, z PE (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],J 1 EQ (z 1 EQ (t1)) is the effluent water quality evaluation model for the denitrification task at time t1, z 1 EQ (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],SNO (t1) represents the decision variable for the denitrification nitrogen removal task at time t1, S NO (t1) represents the nitrate nitrogen concentration at time t1, Q in (t1) represents the inflow rate at time t1, W PE,h (t1) represents the connection weight of the h-th kernel function of the pumping energy consumption at time t1, W 1 EQ,h (t1) represents the link weight of the h-th kernel function for the effluent quality of the denitrification process at time t1. Let h be the center value of the kernel function for the pumping energy consumption and effluent quality of the denitrification task at time t1. Let d be the center value of the h-th kernel function of the effluent quality of the denitrification process at time t1. PE,h (t1) represents the width of the h-th kernel function of the pumping energy consumption at time t1, d 1 EQ,h (t1) represents the width of the h-th kernel function of the effluent quality of the denitrification task at time t1;

[0012] The optimization objectives for nitrification denitrification are aeration energy consumption and effluent water quality:

[0013]

[0014] Among them, J AE (z AE (t1)) is the aeration energy consumption evaluation model at time t1, z AE (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],J 2 EQ (z 2 EQ (t1)) is the effluent water quality evaluation model for the nitrification and denitrification task at time t1, z 2 EQ (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],S O,3 (t1),S O,4 (t1),S O,5 (t1) represents the decision variable for the nitrification denitrification task at time t1, S O,3 (t1) represents the dissolved oxygen concentration in the third corridor at time t1, S O,4(t1) represents the dissolved oxygen concentration in the fourth corridor at time t1, S O,5 (t1) represents the dissolved oxygen concentration in the fifth corridor at time t1, S NH (t1) represents the ammonia nitrogen concentration at time t1, and SS(t1) represents the suspended solids concentration at time t1.

[0015] The constraints on the nitrification denitrification task at time t1 in the wastewater treatment denitrification process are:

[0016]

[0017] Where a4, b4, c4 are related to S O,4 (t1) are the least squares regression coefficients related to S, where a5, b5, and c5 are the coefficients related to S. O,5 The least squares regression coefficients related to (t1), g1(t1) and g2(t1) are constraints at time t1 established based on mechanistic knowledge. The material balance equations for dissolved oxygen in multiple spatial corridors during the nitrification reaction in the wastewater treatment denitrification process are as follows:

[0018]

[0019] Where, r j Let Q be the reaction rate of the j-th aerobic corridor. j Let K be the flow velocity of the j-th aerobic corridor. L a) j Let S be the oxygen transfer coefficient of the j-th aerobic corridor, where j = 4, 5. O,j-1 and S O,j The nonlinear relationship is as follows:

[0020] S O,j (t)=a j S O,j-1 (t)+b j S O,j (t-1)+c j (7)

[0021] Among them, a j ,b j ,c j The least squares regression coefficients related to the dissolved oxygen concentration in the j-th corridor.

[0022]

[0023] Where τ = 1, 2, ..., R, and R is the total number of sample data;

[0024] At time t2, the optimization objectives for the wastewater treatment denitrification process are aeration energy consumption and effluent quality:

[0025]

[0026] Among them, J EQ (z EQ (t2)) is the effluent water quality evaluation model at time t2, z EQ (t2)=[S O,3 (t2),S O,4 (t2),S O,5 (t2),S NO (t2),S NH (t2),Q in [(t2),SS(t2)],S O,3 (t2),S O,4 (t2),S O,5 (t2) and S NO (t2) is the decision variable at time t2, and the constraint is:

[0027]

[0028] in, The predicted pumping energy consumption at time t2

[0029]

[0030] in, This represents the actual pumping value for t2-1;

[0031] (2) Design a multi-objective, multi-task optimization algorithm for multi-source knowledge collaboration.

[0032] The total number of iterations for solving the optimization setpoints is set to 500, the particle swarm size is 50, and x... 1,i (n) represents the position vector of particle i in the denitrification task, evolving to the nth generation, x 2,i (n)=[x 2,i 1 (n),x 2,i 2 (n),x 2,i 3 [n] represents the position vector of particle i in the nitrification task at generation n, and the population information entropy measures the evolutionary state.

[0033]

[0034] Among them, H k (n) represents the population information entropy of the nth generation of task k, q(x) k,i (n))) is x k,i The probability of (n) occurring is used as the adaptive threshold α for classifying the evolutionary stage of the population. k (n) and β k (n) is

[0035]

[0036] Where, ΔH k (n)=H k (n)-H k (n-1) is the difference between the information entropy of the nth generation and the information entropy of the (n-1)th generation of the population for task k;

[0037] If H k (n)≥α k (n)H k,max H k,max The initial population's information entropy is [value], the population is in the first evolutionary stage, and the transferred knowledge term is [value].

[0038]

[0039] Among them, K k,i (n) represents the transferred knowledge item of particle i in the nth generation of task k, S * k (n) represents the decision knowledge of task k in the nth generation.

[0040]

[0041] Where, x k l (n) represents the decision variables of task k in the l-th dimension of the decision space. For the probabilistic model of task k in the l-th dimension decision space

[0042]

[0043] Wherein, P1(x l k (n) represents the probabilistic model of the nitration task in the l-th dimension of the nth generation decision space, P2(x l k (n)) represents the probabilistic model of the denitrification task in the l-th dimension of the nth generation decision space, and * represents the multiplication operation of the probabilistic model;

[0044] If β k (n)H k,max ≤H k (n)<α k (n)H k,max The population is in the second stage of evolution, and the transferred knowledge items are:

[0045]

[0046] Among them, O * k,i (n) represents the decision knowledge of task k particle i in the nth generation.

[0047]

[0048] Among them, pBest k,i (n-γ) represents the optimal position of task particle i in the n-γ generation, η k,i (n) represents the contribution of particle i to task k.

[0049]

[0050] Where m1 = {PE, EQ}, m2 = {AE, EQ};

[0051] If H k (n)<β k (n)H k,max The population is in the third stage of evolution, and the transferred knowledge items are:

[0052]

[0053] Where λ is the similarity between the nitrification and denitrification tasks calculated by the Spearman correlation coefficient;

[0054] When rand ≤ 0.3, and rand is a random number in the range [0,1], the particle velocity update equation is:

[0055] v k,i (n+1)=0.7v k,i (n)+0.25τ1(pBest k,i (n)-x k,i (n))+0.25τ2(gBest k (n)-x k,i (n))(23)

[0056] Among them, v k,i (n+1) is the velocity vector of particle i in the nth generation of task k, τ1 is the individual evolution coefficient vector with a value range of [0,1], τ2 is the population evolution coefficient vector with a value range of [0,1], gBest k (n) represents the global optimal position of task k in the nth generation;

[0057] When rand > 0.3, the particle velocity update equation is:

[0058]

[0059] Wherein, τ3 is the population evolution coefficient vector and its value range is [0,1];

[0060] The particle's position is updated as follows:

[0061] x k,i (n+1)=x k,i (n)+vk,i (n+1) (25)

[0062] Where, x k,i (n+1) is the position vector of particle i in the nth generation of task k;

[0063] The evolutionary process is checked to see if it meets the stopping condition: if the evolutionary generation n is less than 500, the evolutionary generation n is increased by 1 and the particle's velocity and position are updated again; if the evolutionary generation n equals 500, the evolutionary process is terminated, and a solution from the 500th generation is selected as the optimal setpoint y for the process variable. * (t)=[S * O,3 (t),S * O,4 (t),S * O,5 (t),S * NO [(t)],S * O,3 (t) represents the optimal dissolved oxygen setpoint for the third corridor at time t, S * O,4 (t) represents the optimal dissolved oxygen setting value for the fourth corridor at time t, S * O,5 (t) represents the optimal dissolved oxygen setting for the fifth corridor at time t, S * NO (t) represents the optimal setpoint for nitrate nitrogen at time t;

[0064] (3) Design a multivariable proportional-integral-derivative controller

[0065] Design a multivariable proportional-integral-derivative controller to optimize the setpoint y. * (t) Perform tracking control:

[0066]

[0067] Where C = [200, 200, 200, 100] is the proportionality coefficient, L = [15, 15, 15, 9] is the integral time constant, F = [2, 2, 2, 1] is the differential time constant, and e(t) = y(t) - y * y(t) is the control error vector at time t, and y(t) = [S O,3 (t),S O,4 (t),S O,5 (t),S NO [(t)] represents the actual output at time t, Δu(t) = [ΔK] L a3(t),ΔK L a4(t),ΔK L a5(t),ΔQ a(t)] T Let ΔK be the input to the wastewater treatment aeration process optimization control system at time t. L a3(t) represents the change in the dissolved oxygen transfer coefficient of the third corridor at time t, ΔK L a4(t) represents the change in the dissolved oxygen transfer coefficient of the fourth corridor at time t, ΔK L a5(t) represents the change in the dissolved oxygen transfer coefficient of the fifth corridor at time t, ΔQ a (t) represents the change in return flow rate at time t, using ΔK L a3(t) will S O,3 (t) Adjusted to S * O,3 (t), using ΔK L a4(t) will S O,4 (t) Adjusted to S * O,4 (t), using ΔK L a5(t) will S O,5 (t) Adjusted to S * O,5 (t), using Q a (t) will S NO (t) Adjusted to S * NO (t). Attached Figure Description

[0068] Figure 1 The nitrate nitrogen concentration S of this invention NO Optimize the control result graph and error graph.

[0069] Figure 2 The dissolved oxygen concentration S in the third corridor of this invention O,3 Optimize the control result graph and error graph.

[0070] Figure 3 The fourth corridor dissolved oxygen concentration S of this invention O,4 Optimize the control result graph and error graph.

[0071] Figure 4 The fifth corridor dissolved oxygen concentration S of this invention O,5 Optimize the control result graph and error graph. Detailed Implementation

[0072] 1. A knowledge- and data-driven multi-temporal-scale optimization control method for wastewater treatment denitrification processes, specifically including the following steps:

[0073] (1) Establish a knowledge- and data-driven multi-temporal scale evaluation model for wastewater treatment denitrification processes.

[0074] Due to the physical properties of the sensor and the reaction characteristics of the denitrification process in wastewater treatment, the mixed suspended solids concentration (MLSS) is collected at a slow time scale t1 with a sampling period of 2 hours. Other relevant variables of the optimization control model are collected at a fast time scale t2 with a sampling period of 0.5 hours. MLSS is a key variable in the pumping energy consumption optimization control model. Therefore, pumping energy consumption, aeration energy consumption, and effluent water quality are optimized at time t1, and aeration energy consumption and effluent water quality are optimized at time t2.

[0075] At time t1, the optimization control tasks for the wastewater treatment denitrification process include denitrification and nitrification. The optimization objectives for the denitrification task are pumping energy consumption and effluent quality.

[0076]

[0077] Among them, J PE (z PE (t1)) is the pumping energy consumption evaluation model at time t1, z PE (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],J 1 EQ (z 1 EQ (t1)) is the effluent water quality evaluation model for the denitrification task at time t1, z 1 EQ (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],S NO (t1) represents the decision variable for the denitrification nitrogen removal task at time t1, S NO (t1) represents the nitrate nitrogen concentration at time t1, Q in (t1) represents the inflow rate at time t1, W PE,h (t1) represents the connection weight of the h-th kernel function of the pumping energy consumption at time t1, W 1 EQ,h (t1) represents the link weight of the h-th kernel function for the effluent quality of the denitrification process at time t1. Let h be the center value of the kernel function for the pumping energy consumption and effluent quality of the denitrification task at time t1. Let d be the center value of the h-th kernel function of the effluent quality of the denitrification process at time t1. PE,h (t1) represents the width of the h-th kernel function of the pumping energy consumption at time t1, d 1 EQ,h (t1) represents the width of the h-th kernel function of the effluent quality of the denitrification task at time t1;

[0078] The optimization objectives for nitrification denitrification are aeration energy consumption and effluent water quality:

[0079]

[0080] Among them, J AE (z AE (t1)) is the aeration energy consumption evaluation model at time t1, z AE (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],J 2 EQ (z 2 EQ (t1)) is the effluent water quality evaluation model for the nitrification and denitrification task at time t1, z 2 EQ (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],S O,3 (t1),S O,4 (t1),S O,5 (t1) represents the decision variable for the nitrification denitrification task at time t1, S O,3 (t1) represents the dissolved oxygen concentration in the third corridor at time t1, S O,4 (t1) represents the dissolved oxygen concentration in the fourth corridor at time t1, S O,5 (t1) represents the dissolved oxygen concentration in the fifth corridor at time t1, S NH (t1) represents the ammonia nitrogen concentration at time t1, and SS(t1) represents the suspended solids concentration at time t1.

[0081] The constraints on the nitrification denitrification task at time t1 in the wastewater treatment denitrification process are:

[0082]

[0083] Where a4, b4, c4 are related to S O,4 (t1) are the least squares regression coefficients related to S, where a5, b5, and c5 are the coefficients related to S. O,5 The least squares regression coefficients related to (t1), g1(t1) and g2(t1) are constraints at time t1 established based on mechanistic knowledge. The material balance equations for dissolved oxygen in multiple spatial corridors during the nitrification reaction in the wastewater treatment denitrification process are as follows:

[0084]

[0085] Where, r j Let Q be the reaction rate of the j-th aerobic corridor. j Let K be the flow velocity of the j-th aerobic corridor. L a) j Let S be the oxygen transfer coefficient of the j-th aerobic corridor, where j = 4, 5. O,j-1 and S O,j The nonlinear relationship is as follows:

[0086] S O,j (t)=a j S O,j-1 (t)+b j S O,j (t-1)+c j (7)

[0087] Among them, a j ,b j ,c j The least squares regression coefficients related to the dissolved oxygen concentration in the j-th corridor.

[0088]

[0089] Where τ = 1, 2, ..., R, and R is the total number of sample data;

[0090] At time t2, the optimization objectives for the wastewater treatment denitrification process are aeration energy consumption and effluent quality:

[0091]

[0092] Among them, J EQ (z EQ (t2)) is the effluent water quality evaluation model at time t2, z EQ (t2)=[S O,3 (t2),S O,4 (t2),S O,5 (t2),S NO (t2),S NH (t2),Q in [(t2),SS(t2)],S O,3 (t2),S O,4 (t2),S O,5 (t2) and S NO (t2) is the decision variable at time t2, and the constraint is:

[0093]

[0094] in, The predicted pumping energy consumption at time t2

[0095]

[0096] in, This represents the actual pumping value for t2-1;

[0097] (2) Design a multi-objective, multi-task optimization algorithm for multi-source knowledge collaboration.

[0098] The total number of iterations for solving the optimization setpoints is set to 500, the particle swarm size is 50, and x... 1,i (n) represents the position vector of particle i in the denitrification task, evolving to the nth generation, x 2,i (n)=[x 2,i 1 (n),x 2,i 2 (n),x 2,i 3 [n] represents the position vector of particle i in the nitrification task at generation n, and the population information entropy measures the evolutionary state.

[0099]

[0100] Among them, H k (n) represents the population information entropy of the nth generation of task k, q(x) k,i (n))) is x k,i The probability of (n) occurring is used as the adaptive threshold α for classifying the evolutionary stage of the population. k (n) and β k (n) is

[0101]

[0102] Where, ΔH k (n)=H k (n)-H k (n-1) is the difference between the information entropy of the nth generation and the information entropy of the (n-1)th generation of the population for task k;

[0103] If H k (n)≥α k (n)H k,max H k,max The initial population's information entropy is [value], the population is in the first evolutionary stage, and the transferred knowledge term is [value].

[0104]

[0105] Among them, K k,i (n) represents the transferred knowledge item of particle i in the nth generation of task k, S * k (n) represents the decision knowledge of task k in the nth generation.

[0106]

[0107] Where, x k l (n) represents the decision variables of task k in the l-th dimension of the decision space. For the probabilistic model of task k in the l-th dimension decision space

[0108]

[0109] Wherein, P1(x l k (n) represents the probabilistic model of the nitration task in the l-th dimension of the nth generation decision space, P2(x l k (n)) represents the probabilistic model of the denitrification task in the l-th dimension of the nth generation decision space, and * represents the multiplication operation of the probabilistic model;

[0110] If β k (n)H k,max ≤H k (n)<α k (n)H k,max The population is in the second stage of evolution, and the transferred knowledge items are:

[0111]

[0112] Among them, O * k,i (n) represents the decision knowledge of task k particle i in the nth generation.

[0113]

[0114] Among them, pBest k,i (n-γ) represents the optimal position of task particle i in the n-γ generation, η k,i (n) represents the contribution of particle i to task k.

[0115]

[0116] Where m1 = {PE, EQ}, m2 = {AE, EQ};

[0117] If H k (n)<β k (n)H k,max The population is in the third stage of evolution, and the transferred knowledge items are:

[0118]

[0119] Where λ is the similarity between the nitrification and denitrification tasks calculated by the Spearman correlation coefficient;

[0120] When rand ≤ 0.3, and rand is a random number in the range [0,1], the particle velocity update equation is:

[0121] v k,i (n+1)=0.7v k,i (n)+0.25τ1(pBest k,i (n)-x k,i (n))+0.25τ2(gBest k (n)-x k,i (n))(23)

[0122] Among them, v k,i (n+1) is the velocity vector of particle i in the nth generation of task k, τ1 is the individual evolution coefficient vector with a value range of [0,1], τ2 is the population evolution coefficient vector with a value range of [0,1], gBest k (n) represents the global optimal position of task k in the nth generation;

[0123] When rand > 0.3, the particle velocity update equation is:

[0124]

[0125] Wherein, τ3 is the population evolution coefficient vector and its value range is [0,1];

[0126] The particle's position is updated as follows:

[0127] x k,i (n+1)=x k,i (n)+v k,i (n+1) (25)

[0128] Where, x k,i (n+1) is the position vector of particle i in the nth generation of task k;

[0129] The evolutionary process is checked to see if it meets the stopping condition: if the evolutionary generation n is less than 500, the evolutionary generation n is increased by 1 and the particle's velocity and position are updated again; if the evolutionary generation n equals 500, the evolutionary process is terminated, and a solution from the 500th generation is selected as the optimal setpoint y for the process variable. * (t)=[S * O,3 (t),S * O,4 (t),S * O,5 (t),S * NO [(t)],S * O,3 (t) represents the optimal dissolved oxygen setpoint for the third corridor at time t, S* O,4 (t) represents the optimal dissolved oxygen setting value for the fourth corridor at time t, S * O,5 (t) represents the optimal dissolved oxygen setting for the fifth corridor at time t, S * NO (t) represents the optimal setpoint for nitrate nitrogen at time t;

[0130] (3) Design a multivariable proportional-integral-derivative controller

[0131] Design a multivariable proportional-integral-derivative controller to optimize the setpoint y. * (t) Perform tracking control:

[0132]

[0133] Where C = [200, 200, 200, 100] is the proportionality coefficient, L = [15, 15, 15, 9] is the integral time constant, F = [2, 2, 2, 1] is the differential time constant, and e(t) = y(t) - y * y(t) is the control error vector at time t, and y(t) = [S O,3 (t),S O,4 (t),S O,5 (t),S NO [(t)] represents the actual output at time t, Δu(t) = [ΔK] L a3(t),ΔK L a4(t),ΔK L a5(t),ΔQ a (t)] T Let ΔK be the input to the wastewater treatment aeration process optimization control system at time t. L a3(t) represents the change in the dissolved oxygen transfer coefficient of the third corridor at time t, ΔK L a4(t) represents the change in the dissolved oxygen transfer coefficient of the fourth corridor at time t, ΔK L a5(t) represents the change in the dissolved oxygen transfer coefficient of the fifth corridor at time t, ΔQ a (t) represents the change in return flow rate at time t, using ΔK L a3(t) will S O,3 (t) Adjusted to S * O,3 (t), using ΔK L a4(t) will S O,4 (t) Adjusted to S * O,4 (t), using ΔK L a5(t) will S O,5 (t) Adjusted to S *O,5 (t), using Q a (t) will S NO (t) Adjusted to S * NO (t).

Claims

1. A knowledge- and data-driven multi-temporal-scale optimization control method for wastewater treatment denitrification processes, characterized in that: A knowledge- and data-driven evaluation model for the nitrogen removal process in wastewater treatment across multiple time and space scales is established. A multi-source knowledge-based, multi-objective, multi-task optimization algorithm is designed, and a multivariable proportional-integral-derivative controller is designed to achieve optimal setpoint tracking control. The specific steps include: (1) Establish a knowledge- and data-driven multi-temporal scale evaluation model for wastewater treatment denitrification processes. Due to the physical properties of the sensor and the reaction characteristics of the denitrification process in wastewater treatment, the mixed suspended solids concentration (MLSS) is collected at a slow time scale t1 with a sampling period of 2 hours. Other relevant variables of the optimization control model are collected at a fast time scale t2 with a sampling period of 0.5 hours. MLSS is a key variable in the pumping energy consumption optimization control model. Therefore, pumping energy consumption, aeration energy consumption, and effluent water quality are optimized at time t1, and aeration energy consumption and effluent water quality are optimized at time t2. At time t1, the optimization control tasks for the wastewater treatment denitrification process include denitrification and nitrification. The optimization objectives for the denitrification task are pumping energy consumption and effluent quality. Among them, J PE (z PE (t1)) is the pumping energy consumption evaluation model at time t1, z PE (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],J 1 EQ (z 1 EQ (t1)) is the effluent water quality evaluation model for the denitrification task at time t1, z 1 EQ (t1)=[S NO (t1),Q in [(t1),MLSS(t1)],S NO (t1) represents the decision variable for the denitrification nitrogen removal task at time t1, S NO (t1) represents the nitrate nitrogen concentration at time t1, Q in (t1) represents the inflow rate at time t1, W PE,h (t1) represents the connection weight of the h-th kernel function of the pumping energy consumption at time t1, W 1 EQ,h (t1) represents the link weight of the h-th kernel function for the effluent quality of the denitrification process at time t1. Let h be the center value of the kernel function for the pumping energy consumption and effluent quality of the denitrification task at time t1. Let d be the center value of the h-th kernel function of the effluent quality of the denitrification process at time t1. PE,h (t1) represents the width of the h-th kernel function of the pumping energy consumption at time t1, d 1 EQ,h (t1) represents the width of the h-th kernel function of the effluent quality of the denitrification task at time t1; The optimization objectives for nitrification denitrification are aeration energy consumption and effluent water quality: Among them, J AE (z AE (t1)) is the aeration energy consumption evaluation model at time t1, z AE (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],J 2 EQ (z 2 EQ (t1)) is the effluent water quality evaluation model for the nitrification and denitrification task at time t1, z 2 EQ (t1)=[S O,3 (t1),S O,4 (t1),S O,5 (t1),S NH (t1),Q in [(t1),SS(t1)],S O,3 (t1),S O,4 (t1),S O,5 (t1) represents the decision variable for the nitrification denitrification task at time t1, S O,3 (t1) represents the dissolved oxygen concentration in the third corridor at time t1, S O,4 (t1) represents the dissolved oxygen concentration in the fourth corridor at time t1, S O,5 (t1) represents the dissolved oxygen concentration in the fifth corridor at time t1, S NH (t1) represents the ammonia nitrogen concentration at time t1, and SS(t1) represents the suspended solids concentration at time t1. The constraints on the nitrification denitrification task at time t1 in the wastewater treatment denitrification process are: Where a4, b4, c4 are related to S O,4 (t1) are the least squares regression coefficients related to S, where a5, b5, and c5 are the coefficients related to S. O,5 The least squares regression coefficients related to (t1), g1(t1) and g2(t1) are constraints at time t1 established based on mechanistic knowledge. The material balance equations for dissolved oxygen in multiple spatial corridors during the nitrification reaction in the wastewater treatment denitrification process are as follows: Where, r j Let Q be the reaction rate of the j-th aerobic corridor. j Let K be the flow velocity of the j-th aerobic corridor. L a) j Let S be the oxygen transfer coefficient of the j-th aerobic corridor, where j = 4, 5. O,j-1 and S O,j The nonlinear relationship is as follows: S O,j (t)=a j S O,j-1 (t)+b j S O,j (t-1)+c j (7) Among them, a j ,b j ,c j The least squares regression coefficients related to the dissolved oxygen concentration in the j-th corridor. Where τ = 1, 2, ..., R, and R is the total number of sample data; At time t2, the optimization objectives for the wastewater treatment denitrification process are aeration energy consumption and effluent quality: Among them, J EQ (z EQ (t2)) is the effluent water quality evaluation model at time t2, z EQ (t2)=[S O,3 (t2),S O,4 (t2),S O,5 (t2),S NO (t2), S NH (t2),Q in [(t2),SS(t2)],S O,3 (t2),S O,4 (t2),S O,5 (t2) and S NO (t2) is the decision variable at time t2, and the constraint is: in, The predicted pumping energy consumption at time t2 in, This represents the actual pumping value for t2-1; (2) Design a multi-objective, multi-task optimization algorithm for multi-source knowledge collaboration. The total number of iterations for solving the optimization setpoints is set to 500, the particle swarm size is 50, and x... 1,i (n) represents the position vector of particle i in the denitrification task, evolving to the nth generation, x 2,i (n)=[x 2,i 1 (n),x 2,i 2 (n),x 2,i 3 [n] represents the position vector of particle i in the nitrification task at generation n, and the population information entropy measures the evolutionary state. Among them, H k (n) represents the population information entropy of the nth generation of task k, q(x) k,i (n))) is x k,i The probability of (n) occurring is used as the adaptive threshold α for classifying the evolutionary stage of the population. k (n) and β k (n) is Where, ΔH k (n)=H k (n)-H k (n-1) is the difference between the information entropy of the nth generation and the information entropy of the (n-1)th generation of the population for task k; If H k (n)≥α k (n)H k,max H k,max The initial population's information entropy is [value], the population is in the first evolutionary stage, and the transferred knowledge term is [value]. Among them, K k,i (n) represents the transferred knowledge item of particle i in the nth generation of task k, S * k (n) represents the decision knowledge of task k in the nth generation. Where, x k l (n) represents the decision variables of task k in the l-th dimension of the decision space. For the probabilistic model of task k in the l-th dimension decision space Wherein, P1(x l k (n) represents the probabilistic model of the nitration task in the l-th dimension of the nth generation decision space, P2(x l k (n)) represents the probabilistic model of the denitrification task in the l-th dimension of the nth generation decision space, and * represents the multiplication operation of the probabilistic model; If β k (n)H k,max ≤H k (n)<α k (n)H k,max The population is in the second stage of evolution, and the transferred knowledge items are: Among them, O * k,i (n) represents the decision knowledge of task k particle i in the nth generation. Among them, pBest k,i (n-γ) represents the optimal position of task particle i in the n-γ generation, η k,i (n) represents the contribution of particle i to task k. Where m1 = {PE, EQ}, m2 = {AE, EQ}; If H k (n)<β k (n)H k,max The population is in the third stage of evolution, and the transferred knowledge items are: Where λ is the similarity between the nitrification and denitrification tasks calculated by the Spearman correlation coefficient; When rand ≤ 0.3, and rand is a random number in the range [0,1], the particle velocity update equation is: v k,i (n+1)=0.7v k,i (n)+0.25τ1(pBest k,i (n)-x k,i (n))+0.25τ2(gBest k (n)-x k,i (n))(23) where, v k,i (n+1) is the velocity vector of particle i in the nth generation of task k, τ1 is the individual evolution coefficient vector with a value range of [0,1], τ2 is the population evolution coefficient vector with a value range of [0,1], gBest k (n) represents the global optimal position of task k in the nth generation; When rand > 0.3, the particle velocity update equation is: Wherein, τ3 is the population evolution coefficient vector and its value range is [0,1]; The particle's position is updated as follows: x k,i (n+1)=x k,i (n)+v k,i (n+1) (25) Where, x k,i (n+1) is the position vector of particle i in the nth generation of task k; The evolutionary process is checked to see if it meets the stopping condition: if the evolutionary generation n is less than 500, the evolutionary generation n is increased by 1 and the particle's velocity and position are updated again; if the evolutionary generation n equals 500, the evolutionary process is terminated, and a solution from the 500th generation is selected as the optimal setpoint y for the process variable. * (t)=[S * O,3 (t),S * O,4 (t),S * O,5 (t),S * NO [(t)],S * O,3 (t) represents the optimal dissolved oxygen setpoint for the third corridor at time t, S * O,4 (t) represents the optimal dissolved oxygen setting value for the fourth corridor at time t, S * O,5 (t) represents the optimal dissolved oxygen setting for the fifth corridor at time t, S * NO (t) represents the optimal setpoint for nitrate nitrogen at time t; (3) Design a multivariable proportional-integral-derivative controller Design a multivariable proportional-integral-derivative controller to optimize the setpoint y. * (t) Perform tracking control: Where C = [200, 200, 200, 100] is the proportionality coefficient, L = [15, 15, 15, 9] is the integral time constant, F = [2, 2, 2, 1] is the differential time constant, and e(t) = y(t) - y * y(t) is the control error vector at time t, and y(t) = [S O,3 (t),S O,4 (t),S O,5 (t),S NO [(t)] represents the actual output at time t, Δu(t) = [ΔK] L a3(t),ΔK L a4(t),ΔK L a5(t),ΔQ a (t)] T Let ΔK be the input to the wastewater treatment aeration process optimization control system at time t. L a3(t) represents the change in the dissolved oxygen transfer coefficient of the third corridor at time t, ΔK L a4(t) represents the change in the dissolved oxygen transfer coefficient of the fourth corridor at time t, ΔK L a5(t) represents the change in the dissolved oxygen transfer coefficient of the fifth corridor at time t, ΔQ a (t) represents the change in return flow rate at time t, using ΔK L a3(t) will S O,3 (t) Adjusted to S * O,3 (t), using ΔK L a4(t) will S O,4 (t) Adjusted to S * O,4 (t), using ΔK L a5(t) will S O,5 (t) Adjusted to S * O,5 (t), using Q a (t) will S NO (t) Adjusted to S * NO (t).

Images