A sewage treatment process adaptive constraint multi-objective operation optimization control method

Abstract

The application provides a sewage treatment process adaptive constraint multi-objective operation optimization control method, and relates to the technical field of sewage treatment. The method first establishes energy consumption, effluent quality and effluent index models through a recursive bilinear subspace identification modeling method; then optimizes the energy consumption and effluent quality and processes effluent index constraints through an adaptive constraint multi-objective evolution method to obtain optimized setting values of nitrate nitrogen S NO2 and dissolved oxygen D O5 concentrations; finally, model-free adaptive predictive control is adopted to realize real-time control of S NO2 and dissolved oxygen D O5 concentrations. The method completes high-performance and accurate control of the optimized setting values of S NO2 and D O5 concentrations, can effectively improve the operation efficiency and stability of the sewage treatment process, and realizes energy saving and consumption reduction of the sewage treatment process on the basis that the effluent indexes meet the discharge standards.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment technology, and in particular to an adaptive constrained multi-objective operation optimization control method for wastewater treatment processes. Background Technology

[0002] Wastewater treatment involves a series of physicochemical reactions and biochemical oxidation reactions by microorganisms to filter, precipitate, decompose, and absorb pollutants in wastewater, thereby treating and purifying the wastewater to ensure that the final effluent meets national discharge standards. The efficient and stable operation of wastewater treatment plants is crucial for improving effluent quality, reducing energy consumption, and achieving water resource recycling. This effectively alleviates current societal pressure on water resources and promotes the protection of the ecological water environment.

[0003] Wastewater treatment is a complex, nonlinear, dynamic process involving numerous physicochemical reactions and microbial biochemical oxidation reactions. It is highly susceptible to environmental and external uncertainties, exhibiting strong nonlinearity, time-varying characteristics, and uncertainty. Therefore, it is difficult to establish accurate mathematical models, making it challenging for conventional control methods to achieve satisfactory results in the control and optimization of wastewater treatment processes. Optimization control of wastewater treatment processes is a typical constrained multi-objective optimization problem. Minimizing energy consumption and optimizing effluent quality are two conflicting performance indicators. Effluent indicators have strict emission constraints and vary with the complex changes in the biochemical reaction process. Furthermore, the mechanistic relationships between energy consumption, effluent quality, effluent indicators, and optimization variables are unclear, making it difficult to establish clear mechanistic optimization models and effluent indicator constraint models. Nitrate nitrogen (S) in wastewater treatment... NO2 and dissolved oxygen D O5 Nitrate concentration is a key parameter affecting the final effluent quality and energy consumption. It is crucial to rationally adjust nitrate nitrogen (S) based on real-time changes in the wastewater treatment operation status. NO2 and dissolved oxygen D O5 Nitrate concentration plays a crucial role in achieving energy conservation and emission reduction in wastewater treatment while ensuring that effluent indicators meet discharge standards. Therefore, it is necessary to seek a new adaptive constrained multi-objective optimization control method. This method first establishes a model relating wastewater treatment energy consumption, effluent quality, effluent indicator constraints, and optimization variables, deriving the mathematical expression of the constrained multi-objective optimization problem for wastewater treatment. Then, through the constrained multi-objective optimization method, it optimizes energy consumption and effluent quality while handling complex effluent indicator constraints, obtaining the desired nitrate nitrogen concentration. NO2 and dissolved oxygen D O5 The optimized setpoint for concentration is then precisely controlled through a high-precision process control method. This is key to improving effluent quality, ensuring that effluent indicators meet standards, and simultaneously increasing the operational efficiency of the wastewater treatment process, thereby achieving energy conservation and consumption reduction. Summary of the Invention

[0004] The technical problem this invention aims to solve is to address the shortcomings of the existing technology by providing an adaptive constrained multi-objective operation optimization control method for wastewater treatment processes. This method optimizes energy consumption and effluent quality and handles complex effluent index constraints through an adaptive constrained multi-objective evolutionary approach, thereby obtaining nitrate nitrogen S. NO2 and dissolved oxygen D O5 Optimized setpoint for concentration; model-free adaptive predictive control is used to achieve control of S. NO2 and dissolved oxygen D O5 Real-time control of concentration.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: an adaptive constrained multi-objective operation optimization control method for wastewater treatment processes, which establishes energy consumption, effluent quality, and effluent index models through a recursive bilinear subspace identification and modeling method; optimizes energy consumption and effluent quality and handles effluent index constraints through an adaptive constrained multi-objective evolutionary method to obtain nitrate nitrogen S. NO2 and dissolved oxygen D O5 Optimized setpoint for concentration; model-free adaptive predictive control is used to achieve control of S. NO2 and dissolved oxygen D O5 Real-time concentration control includes the following steps:

[0006] Step 1: Select nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is the optimization variable, with energy consumption (EC) and effluent water quality (EQ) as optimization objectives, and effluent ammonia nitrogen (S) as the optimization target. NHe and total nitrogen (N) in effluent tote To establish a constrained multi-objective optimization problem for the wastewater treatment process, with effluent constraints as the main metric;

[0007] Step 1.1: Collect optimization variable S using a sliding window data acquisition method. NO2 and D O5 Optimize target EC and EQ and effluent constraint S NHe and N tote Real-time data is used to construct the database required for building the optimization target model and the effluent constraint model;

[0008] The sliding window data acquisition method is as follows: the original database used for modeling is constructed from the sewage treatment operation data within a fixed period. As the sewage treatment plant operates, every optimization cycle, a set of old data is deleted from the database and a set of the same number of real-time new data is added. The total amount of data in the database remains unchanged, thereby realizing data acquisition and database updates and improving the real-time performance of modeling data.

[0009] Step 1.2: Establish models for energy consumption, effluent quality, and effluent constraint indicators using the recursive bilinear subspace identification online modeling method, as shown in the following formula:

[0010]

[0011]

[0012]

[0013]

[0014] in, and These represent the energy consumption and effluent water quality model outputs, respectively. and The following represent the effluent constraint index, ammonia nitrogen (S). NHe and total nitrogen N tote Model output, u f Indicates the input optimization variable S NO2 and D O5 , These are energy consumption, effluent water quality, and S. NHe N tote Model parameters with respect to the input data These are energy consumption, effluent water quality, and S. NHe N tote Model parameters w based on historical data pm These are energy consumption, effluent water quality, and S. NHe N tote The model's historical input and output data, These represent energy consumption, effluent water quality, and S, respectively. NHe N tote Model parameters with respect to the bilinear subspace matrix of the model input and output data. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The model's input and output data are bilinear subspace matrices. These are energy consumption, effluent water quality, and S. NHe N tote The model parameter matrix of the model. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The real-time data vectors used in model building, and the model parameters Capable of using real-time data vectors Online learning and adaptive updating are performed using the following recursive least squares method with a forgetting factor:

[0015]

[0016] in, This represents the model parameter matrix at time k, representing the established model for energy consumption, effluent quality, and effluent constraint indicators. k+1 This represents the model output at time k+1, corresponding to the established model for energy consumption, effluent quality, and effluent constraint indicators. This represents the real-time data used at time k+1 by the established model of energy consumption, effluent quality, and effluent constraint indicators. k+1 P represents the gain vector. k+1 Let λ be the covariance matrix, λ be the forgetting factor, and I be the identity matrix.

[0017] Step 1.3: Using equations (1) and (2) as the optimized performance index model, and equations (3) and (4) as the effluent index constraint model, with S... NO2 and D O5 To optimize the variables, a constrained multi-objective optimization problem for wastewater treatment is obtained;

[0018] Step 2: Design an optimization method based on an adaptive constrained multi-objective evolutionary algorithm, and adaptively select the optimization variable S. NO2 and D O5 Optimized settings;

[0019] Step 2.1: In the crossover evolution of individuals in a multi-objective evolutionary algorithm population, the neighborhood of an individual is defined using a comprehensive distance approach;

[0020] The specific formula for calculating the overall distance is as follows:

[0021]

[0022] Where sd represents vector x s and x j The combined distance between them, x s =[S NO2 D O5 ] T and x v =[S NO2 D O5 ] T Let s and v represent the s-th individual and v-th individual in the population, respectively, where s, v = 1, ..., N, N represents the population size, and 0 ≤ α ≤ 1 is a weighting factor used to adjust the distance similarity. And angular similarity cosθ(x) s ,x v The proportion of d(x) in the overall distance, e is the natural index, and d(x) is the natural index. s ,x v ) represents x s and x j The distance between them, θ(x) k ,x v ) represents x s and xv The angle between them, d(x) s ,x v ) and θ(x s ,x v The calculation formulas for ) are as follows:

[0023] d(x s ,x v )=||x s -x v ||2 (7)

[0024]

[0025] Where ||·||2 represents the vector 2 norm, Cos represents the dot product of vectors. -1 (·) represents the inverse cosine operation;

[0026] Step 2.2: The i-th individual x in the g-th generation of the population iteration ig When performing crossover operations, the formula for calculating the comprehensive distance is used to calculate x. ig The T individuals with the smallest combined distances to each individual in the current generation population are selected to form individual x. ig In the current generation of comprehensive distance neighborhood, x is calculated simultaneously. ig The T individuals with the smallest combined distances to every individual in the previous generation population are selected to form individual x. ig In the previous generation's comprehensive distance neighborhood, individual x ig The indices of the combined distance neighborhood members in the current generation and the previous generation are denoted as {S}. i,g,1 ,S i,g,2 ,…,S i,g,T} and {S i,g-1,1 ,S i,g-1,2 ,…,S i,g-1,T}, where T is the size of the neighborhood;

[0027] Step 2.3: Perform cross-generational information crossover / mutation based on the comprehensive distance neighborhood with a 50% probability;

[0028] Set x ig Let x be the i-th parent individual in the current generation population. ij,g Individual x ig The j-th element, y ig Indicates based on parent individual x ig The newly generated mutant individuals undergo cross-generational information cross-operation based on the comprehensive distance neighborhood, as shown in the following formula:

[0029] y ij,g =0.5×[(1-γ j )xr1n,j,g +(1+γ j )x r2n,j,g-1 (9)

[0030]

[0031] Among them, y ij,g Represents the mutant individual y ig The j-th element, where the index g represents the current generation population, g-1 represents the previous generation population, x r1n,g From parent individual x ig A neighboring individual randomly selected from the current generation's neighborhood, x r1n,j,g For x r1n,g The j-th element, r1n∈{S i,g,1 ,S i,g,2 ,…,S i,g,T}, and x r2n,g-1 From parent individual x ig A neighboring individual randomly selected from the previous generation's neighborhood, x r2n,j,g-1 It is its j-th element, r2n∈{S i,g-1,1 ,S i,g-1,2 ,…,S i,g-1,T}, μ j For a random number that follows a uniform distribution, η c >0 represents the cross-distribution factor, γ j It is determined by the cross-distribution factor η c Dynamically generated random numbers;

[0032] Step 2.4: Perform cross-generational information crossover / mutation based on the population with a 50% probability;

[0033] The cross-generational information crossover operation based on population is shown in the following formula:

[0034] y ij,g =0.5×[(1-γ j )x r1p,j,g +(1+γ j )x r2p,j,g-1 (11)

[0035] Where, x r1p,g It is an individual randomly selected from all individuals in the current generation population, x. r1p,j,g For x r1p,g The j-th element, r1p∈{1,2,…,N}, x r2p,g-1 It is an individual randomly selected from all individuals in the previous generation population, x r2p,j,g-1 Let r2p represent its j-th element, where r2p∈{1,2,…,N};

[0036] Step 2.5: Perform polynomial mutation on the mutant individuals;

[0037] The specific formula for polynomial mutation is:

[0038]

[0039] Among them, off ij The off-number of off-numbers generated by the final generation of crossover mutation i The j-th element, p m δ represents the probability of mutation. j δ is the polynomial variation increment. j The specific calculation formula is as follows:

[0040]

[0041] Where, η m Indicates the distribution factor of variation;

[0042] Step 2.6: Offset the offspring individuals after crossover mutation i Together with all other individuals in the population, the fitness function value of each individual is calculated using an adaptive penalty method, and a penalty coefficient is adaptively applied to individuals that do not meet the constraints.

[0043] Individual off i The formula for calculating the fitness function value is:

[0044]

[0045]

[0046] Among them, f j,fitness (off i ) indicates individual off i The fitness function value of the j-th objective function, f j (off i ) indicates individual off i The value of the j-th objective function, It is a value determined by the mean of the j-th objective function in the population. <f j (off i The value of ) represents the mean of the j-th objective function for all individuals in the current generation of the population. Indicates the kth f One constraint It is the kth f The penalty coefficient for each constraint, It can be calculated using the following formula:

[0047]

[0048] in, <vl (off i )> is the mean of the number of times the l-th constraint is violated in the population, and K is the total number of constraints;

[0049] Step 2.7: Calculate individual off using the method of equation (14). i The fitness function values ​​of all other individuals in the population are used, and these fitness function values ​​are used to replace the individual's off-value. i The objective function values ​​corresponding to all other individuals in the population are used to evaluate the off-off values ​​of off-off values ​​for off-off ... i Screening was performed to complete population P. g =[x1,…,x N Update of ];

[0050] Step 2.8: If the current iteration step g is less than the maximum iteration step G, go to step 2.2; if the current iteration step g is greater than or equal to the maximum iteration step G, go to step 2.9.

[0051] Step 2.9: Based on the effluent water quality threshold EQT principle, if the individual x with the lowest energy consumption in the population after the iteration is... EC,min The corresponding effluent water quality target value is less than or equal to the EQT value, and the effluent index S NHe N tote If the emission standards are met, then select individual x. EC,min For S NO2 and D O5 The optimal setting value is determined; otherwise, the individual x with the lowest effluent quality in the population after the iteration ends is selected. EQ,min For S NO2 and D O5 Optimize the settings to achieve S NO2 and D O5 Optimize adaptive selection of setpoints;

[0052] Step 3: Design a control method for the process control loop based on model-free adaptive predictive control;

[0053] Step 3.1: Using the dissolved oxygen conversion coefficient K La5 Internal return flow Q a As the input to the process control loop, nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is used as the output of the process control loop, in the form of nitrate nitrogen S. NO2 and dissolved oxygen D O5 The optimal setpoint for concentration is used as the desired output of the process control loop;

[0054] Step 3.2: Collect dissolved oxygen conversion coefficient K La5 Internal return flow Q a Nitrate nitrogen SNO2 Dissolved oxygen D O5 Based on real-time data, estimate the pseudo-Jacobi parameter matrix Φ of the improved model-free adaptive predictive control at the current moment. c (k);

[0055] Pseudo-Jacobi matrix Φ c The estimation formula for (k) is:

[0056]

[0057] in, Φ c The estimate of (k), Φ c (k-1) denotes the pseudo-Jacobi matrix of the previous time step. Φ c The estimate of (k-1), where Δy(k) represents the output S at time k. NO2 and D O5 The increment, Δu(k-1), represents the input K at time k-1. La5 and Q a The increment, μ represents the weighting factor, μ>0, and η represents the step size factor, 0<η≤2;

[0058] Step 3.3: Estimate the Jacobian matrix at the current time. To reset, use the formula shown below:

[0059]

[0060] in, Representative matrix The elements, i′=1,2, j′=1,2, yes The initial values ​​are: |(·)| is the absolute value, sign(·) is the sign function, α, b1 and b2 are constants greater than 0, and satisfy α≥1, b2>b1(2α+1)(m′-1), m′=2 represents the number of input variables;

[0061] Step 3.4: Utilize the pseudo-Jacobi parameter matrix Φ at the known time. c (1),…Φ c (k), the pseudo-Jacobi parameter matrix Φ for predicting future times based on an autoregressive model. c (k+j0),j0=1,…N u -1;

[0062] The formula for the autoregressive model is:

[0063]

[0064] Where, N uTo control the time domain, n represents the coefficients of the autoregressive model. p The order of the model is given by... The coefficient vector matrix formed Calculated using the following formula:

[0065]

[0066] in, δ is a weight parameter greater than 0, δ∈(0,1];

[0067] Step 3.5: Obtain the pseudo-Jacobi parameter matrix for the predicted future times. To reset, use the formula shown below:

[0068]

[0069] in, Representative matrix elements, yes The initial values ​​are i = 1, 2, j = 1, 2;

[0070] Step 3.6: Based on the pseudo-Jacobi parameter matrix Calculate the control increment in the control time domain based on the input and output data of the process control loop. The control input u(k) at time k;

[0071] The formulas for calculating u(k) are as follows:

[0072]

[0073]

[0074] in, The vector form representing the input increment in the control time domain, where ρ represents the step size factor. It is 2N u A 2-row block matrix, where A1(k) represents the block coefficient matrix, is defined as follows:

[0075]

[0076] Among them, 0 2×2 It is a 2x2 matrix of zeros;

[0077] Step 3.7: Input the obtained control into the dissolved oxygen conversion coefficient K La5 Internal return flow Q a The value applied to the wastewater treatment system S NO2 and dissolved oxygen D O5Optimize the control of setpoints;

[0078] Step 4: Obtain the S-value of the wastewater treatment process operation optimization control method. NO2 and dissolved oxygen D O5 The actual concentration value was calculated, along with energy consumption, effluent water quality, and effluent ammonia nitrogen (S). NHe and total nitrogen (N) in effluent tote index.

[0079] The beneficial effects of adopting the above technical solution are as follows: The adaptive constrained multi-objective operation optimization control method for sewage treatment process provided by the present invention (1) addresses the problem that sewage treatment processes with strong nonlinear time-varying characteristics, uncertainties, and non-stationary operating conditions are difficult to maintain efficient and stable operation, and the actual sewage treatment plant needs to process sewage S NO2 and dissolved oxygen D O5 The concentration is controlled within a suitable range, so that S NHe and N tote To ensure that effluent indicators are within specified emission standards and to improve the operational efficiency of the wastewater treatment process, an adaptive constrained multi-objective operation optimization control method for the wastewater treatment process was designed. This method can effectively improve the operational efficiency and stability of the wastewater treatment process and achieve energy conservation and consumption reduction while meeting the emission standards.

[0080] (2) To address the problem that the energy consumption, effluent quality, effluent index constraints and optimization variable mechanism relationships are unclear and the model is difficult to establish in wastewater treatment processes with non-stationary operating conditions, a sliding window-based recursive bilinear subspace identification modeling method is designed to establish an online model of energy consumption, effluent quality, and effluent index constraints. This effectively improves the accuracy of the optimization performance index and effluent constraint model, and can obtain a more accurate constrained multi-objective optimization problem for the wastewater treatment process.

[0081] (3) An adaptive constrained multi-objective optimization method based on comprehensive distance neighborhood, cross-generational information crossover, and adaptive constraint penalty was designed. This method can effectively improve the optimization performance of multi-objective optimization problems and increase the processing efficiency of complex effluent constraints. It achieves optimization of energy consumption and effluent quality on the basis of satisfying the effluent indicators. Furthermore, a method based on effluent quality thresholds was adopted to obtain a suitable S from the multi-objective optimization results. NO2 and D O5 Optimized concentration settings;

[0082] (4) A model-free adaptive predictive control method for wastewater treatment was designed, and the control of S was completed. NO2 and D O5 High-performance, precise control of concentration optimization settings. Attached Figure Description

[0083] Figure 1 A flowchart of an adaptive constrained multi-objective operation optimization control method for a wastewater treatment process provided in an embodiment of the present invention;

[0084] Figure 2 The following are the model and modeling error probability density function PDF distribution curves of energy consumption EC and effluent water quality EQ provided in the embodiments of the present invention, wherein (a) is the recursive bilinear subspace identification model of energy consumption EC, (b) is the probability density function PDF distribution shape of the energy consumption EC model, (c) is the recursive bilinear subspace identification model of effluent water quality EQ, and (d) is the probability density function PDF distribution shape of the effluent water quality EQ model.

[0085] Figure 3 The effluent index constraint S provided for the embodiments of the present invention NHe and N tote The model and the probability density function PDF distribution curve of the modeling error are shown, where (a) is the effluent index ammonia nitrogen S. NHe The recursive bilinear subspace identification model, (b) is the effluent index S NHe The probability density function PDF distribution shape of the model, (c) represents the effluent index N. tote The recursive bilinear subspace identification model, (d) represents the effluent index N. tote The shape of the probability density function (PDF) distribution of the model;

[0086] Figure 4 S provided for embodiments of the present invention NO2 Real-time control curves;

[0087] Figure 5 D provided for the embodiments of the present invention O5 The real-time control curve. Detailed Implementation

[0088] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0089] In this embodiment, an adaptive constrained multi-objective operation optimization control method for a wastewater treatment process is proposed. This method establishes energy consumption, effluent quality, and effluent index models using a recursive bilinear subspace identification and modeling method. It then optimizes energy consumption and effluent quality while handling complex effluent index constraints using an adaptive constrained multi-objective evolutionary method to obtain nitrate nitrogen S. NO2 and dissolved oxygen D O5 Optimized setpoint for concentration; model-free adaptive predictive control is used to achieve control of S. NO2 and dissolved oxygen D O5 Real-time control of concentration, such as Figure 1As shown, it includes the following steps:

[0090] Step 1: Select nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is the optimization variable, with energy consumption (EC) and effluent water quality (EQ) as optimization objectives, and effluent ammonia nitrogen (S) as the optimization target. NHe and total nitrogen (N) in effluent tote To establish a constrained multi-objective optimization problem for the wastewater treatment process, with effluent constraints as the main metric;

[0091] Step 1.1: Collect optimization variable S using a sliding window data acquisition method. NO2 and D O5 Optimize target EC and EQ and effluent constraint S NHe and N tote Real-time data is used to construct the database required for building the optimization target model and the effluent constraint model;

[0092] The sampling period is 15 minutes. The data acquisition method of the sliding window is as follows: the original database used for modeling is constructed with sewage treatment operation data within 7 days. As the sewage treatment plant operates, every 2 hours of optimization cycle, a set of old data is deleted from the database and a set of new real-time data of the same amount is added. The total amount of data in the database remains unchanged. This is to achieve data acquisition and database update, and improve the real-time performance of modeling data.

[0093] Step 1.2: Establish models for energy consumption, effluent quality, and effluent constraint indicators using the Recursive Bilinear Subspace Identification (RBLSI) online modeling method, as shown in the following formula:

[0094]

[0095]

[0096]

[0097]

[0098] in, and These represent the energy consumption and effluent water quality model outputs, respectively. and The following represent the effluent constraint index, ammonia nitrogen (S). NHe and total nitrogen N tote Model output, u f Indicates the input optimization variable S NO2 and D O5 , These are energy consumption, effluent water quality, and S. NHe N toteModel parameters with respect to the input data These are energy consumption, effluent water quality, and S. NHe N tote Model parameters w based on historical data pm These are energy consumption, effluent water quality, and S. NHe N tote The model's historical input and output data, These represent energy consumption, effluent water quality, and S, respectively. NHe N tote Model parameters with respect to the bilinear subspace matrix of the model input and output data. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The model's input and output data are bilinear subspace matrices. These are energy consumption, effluent water quality, and S. NHe N tote The model parameter matrix of the model. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The real-time data vectors used in model building, and the model parameters Capable of using real-time data vectors Online learning and adaptive updating are performed using the following recursive least squares method with a forgetting factor:

[0099]

[0100] in, This represents the model parameter matrix at time k, representing the established model for energy consumption, effluent quality, and effluent constraint indicators. k+1 This represents the model output at time k+1, corresponding to the established model for energy consumption, effluent quality, and effluent constraint indicators. This represents the real-time data used at time k+1 by the established model of energy consumption, effluent quality, and effluent constraint indicators. k+1 P represents the gain vector. k+1 Let λ be the covariance matrix, λ be the forgetting factor, and I be the identity matrix.

[0101] Step 1.3: Using equations (1) and (2) as the optimized performance index model, and equations (3) and (4) as the effluent index constraint model, with S... NO2 and D O5 To optimize the variables, a constrained multi-objective optimization problem for wastewater treatment is obtained;

[0102] Step 2: Design an optimization method based on an adaptive constrained multi-objective evolutionary algorithm, and adaptively select the optimization variable S. NO2 and D O5 Optimized settings;

[0103] Step 2.1: In the crossover evolution of individuals in a multi-objective evolutionary algorithm population, the neighborhood of an individual is defined using a comprehensive distance method. The specific formula for calculating the comprehensive distance is as follows:

[0104]

[0105] Where sd represents vector x s and x j The combined distance between them, x s =[S NO2 D O5 ] T and x v =[S NO2 D O5 ] T Let s and v represent the s-th individual and v-th individual in the population, respectively, where s, v = 1, ..., N, N = 350 represents the population size, and 0 ≤ α ≤ 1 is the weighting factor used to adjust the distance similarity. And angular similarity cosθ(x) s ,x v In this embodiment, α = 0.89, e is the natural exponent, and d(x) represents the proportion of the overall distance. s ,x v ) represents x s and x j The distance between them, θ(x) k ,x v ) represents x s and x v The angle between them, d(x) s ,x v ) and θ(x s ,x v The calculation formulas for ) are as follows:

[0106] d(x s ,x v )=||x s -x v ||2 (7)

[0107]

[0108] Where ||·||2 represents the vector 2 norm, Cos represents the dot product of vectors. -1 (·) represents the inverse cosine operation;

[0109] Step 2.2: The i-th individual x in the g-th generation of the population iteration ig When performing cross-operation, the formulas for calculating the comprehensive distance (6) to (8) are used to calculate x. igThe individuals x are selected from the T individuals with the smallest combined distances to each individual in the current generation (gth generation). ig In the current generation of comprehensive distance neighborhood, x is calculated simultaneously. ig The individuals x are selected from the T individuals with the smallest combined distances to each individual in the previous generation (g-1 generation). ig In the previous generation's comprehensive distance neighborhood, individual x ig The indices of the combined distance neighborhood members in the current generation and the previous generation are denoted as {S}. i,g,1 ,S i,g,2 ,…,S i,g,T} and {S i,g-1,1 ,S i,g-1,2 ,…,S i,g-1,T}, where T = 20 is the size of the neighborhood;

[0110] Step 2.3: Perform cross-generational information crossover / mutation based on the comprehensive distance neighborhood with a 50% probability;

[0111] Set x ig Let x be the i-th parent individual in the current generation population. ij,g Individual x ig The j-th element, y ig Indicates based on parent individual x ig The newly generated mutant individuals undergo cross-generational information cross-operation based on the comprehensive distance neighborhood, as shown in the following formula:

[0112] y ij,g =0.5×[(1-γ j )x r1n,j,g +(1+γ j )x r2n,j,g-1 (9)

[0113]

[0114] Among them, y ij,g Represents the mutant individual y ig The j-th element, where the index g represents the current generation population, g-1 represents the previous generation population, x r1n,g From parent individual x ig A neighboring individual randomly selected from the current generation's neighborhood, x r1n,j,g For x r1n,g The j-th element, r1n∈{S i,g,1 ,S i,g,2 ,…,S i,g,T}, and x r2n,g-1 From parent individual x ig A neighboring individual randomly selected from the previous generation's neighborhood, x r2n,j,g-1It is its j-th element, r2n∈{S i,g-1,1 ,S i,g-1,2 ,…,S i,g-1,T}, μ j For a random number that follows a uniform distribution, η c >0 represents the cross-distribution factor, and in this embodiment, η c =2,γ j It is determined by the cross-distribution factor η c Dynamically generated random numbers;

[0115] Step 2.4: Perform cross-generational information crossover / mutation based on the population with a 50% probability;

[0116] The cross-generational information crossover operation based on population is shown in the following formula:

[0117] y ij,g =0.5×[(1-γ j )x r1p,j,g +(1+γ j )x r2p,j,g-1 (11)

[0118] Where, x r1p,g It is an individual randomly selected from all individuals in the current generation population, x. r1p,j,g For x r1p,g The j-th element, r1p∈{1,2,…,N}, x r2p,g-1 It is an individual randomly selected from all individuals in the previous generation population, x r2p,j,g-1 Let r2p represent its j-th element, where r2p∈{1,2,…,N};

[0119] Step 2.5: Perform polynomial mutation on the mutant individuals. The specific polynomial mutation formula is as follows:

[0120]

[0121] Among them, off ij The off-number of off-numbers generated by the final generation of crossover mutation i The j-th element, p m δ represents the probability of mutation. j δ is the polynomial variation increment. j The specific calculation formula is as follows:

[0122]

[0123] Where, η m =5 indicates the variation distribution factor;

[0124] Step 2.6: Offset the offspring individuals after crossover mutation iTogether with all other individuals in the population, the fitness function value of each individual is calculated using an adaptive penalty method, and a penalty coefficient is adaptively applied to individuals that do not meet the constraints.

[0125] This embodiment uses individual off i For example, the formula for calculating the fitness function value is:

[0126]

[0127]

[0128] Among them, f j,fitness (off i ) indicates individual off i The fitness function value of the j-th objective function, f j (off i ) indicates individual off i The value of the j-th objective function, It is a value determined by the mean of the j-th objective function in the population. <f j (off i The value of ) represents the mean of the j-th objective function for all individuals in the current generation of the population. Indicates the kth f One constraint It is the kth f The penalty coefficient for each constraint, It can be calculated using the following formula:

[0129]

[0130] in, <v l (off i )> is the mean of the number of times the l-th constraint is violated in the population, and K=2 is the total number of constraints;

[0131] Step 2.7: Calculate individual off using the method of equation (14). i The fitness function values ​​of all other individuals in the population are used, and these fitness function values ​​are used to replace the individual's off-value. i The objective function values ​​corresponding to all other individuals in the population are used to evaluate the off-off values ​​of off-off values ​​for off-off ... i Screening was performed to complete population P. g =[x1,…,x N Update;

[0132] Step 2.8: If the current iteration step g is less than the maximum iteration step G = 300, go to step 2.2; if the current iteration step g is greater than or equal to the maximum iteration step G, go to step 2.9.

[0133] Step 2.9: Based on the effluent water quality threshold EQT principle, EQT = 6200, unit is kg poll·d⁻¹, if the individual x with the lowest energy consumption in the population after the iteration is... EC,min The corresponding effluent water quality target value is less than or equal to the EQT value, and the effluent index S NHe N tote If the emission standards are met, then select individual x. EC,min For S NO2 and D O5 The optimal setting value is determined; otherwise, the individual x with the lowest effluent quality in the population after the iteration ends is selected. EQ,min For S NO2 and D O5 Optimize the settings to achieve S NO2 and D O5 Optimize adaptive selection of setpoints;

[0134] Step 3: Design a control method for the process control loop based on model-free adaptive predictive control;

[0135] Step 3.1: Using the dissolved oxygen conversion coefficient K La5 Internal return flow Q a As the input to the process control loop, nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is used as the output of the process control loop, in the form of nitrate nitrogen S. NO2 and dissolved oxygen D O5 The optimal setpoint for concentration is used as the desired output of the process control loop;

[0136] Step 3.2: Collect dissolved oxygen conversion coefficient K La5 Internal return flow Q a Nitrate nitrogen S NO2 Dissolved oxygen D O5 Based on real-time data, estimate the pseudo-Jacobi parameter matrix Φ of the improved model-free adaptive predictive control at the current moment. c (k), pseudo-Jacobi matrix Φ c The estimation formula for (k) is:

[0137]

[0138] in, Φ c The estimate of (k), Φ c (k-1) denotes the pseudo-Jacobi matrix of the previous time step. Φ c The estimate of (k-1), where Δy(k) represents the output S at time k. NO2 and D O5 The increment, Δu(k-1), represents the input K at time k-1. La5and Q a The increment is μ, which represents the weighting factor (μ>0) and η, which represents the step size factor (0<η≤2). In this embodiment, μ=0.6 and η=0.5.

[0139] Step 3.3: Estimate the Jacobian matrix at the current time. To reset, use the formula shown below:

[0140]

[0141] in, Representative matrix The elements, i′=1,2, j′=1,2, yes The initial values ​​are: |(·)| is the absolute value, sign(·) is the sign function, α, b1 and b2 are constants greater than 0 and satisfy α≥1, b2>b1(2α+1)(m′-1), m′=2 represents the number of input variables; in this embodiment, a=1.2, b1=0.52, b2=0.8.

[0142] Step 3.4: Utilize the pseudo-Jacobi parameter matrix Φ at the known time. c (1),…Φ c (k), the pseudo-Jacobi parameter matrix Φ for predicting future times based on an autoregressive model. c (k+j0),j0=1,…N u -1;

[0143] The formula for the autoregressive model is:

[0144]

[0145] Where, N u =3 represents the control time domain. n represents the coefficients of the autoregressive model. p =3 represents the order of the model, from The coefficient vector matrix formed Calculated using the following formula:

[0146]

[0147] in, δ is a weight parameter greater than 0, δ∈(0,1]; in this embodiment, δ=0.8.

[0148] Step 3.5: Obtain the pseudo-Jacobi parameter matrix for the predicted future times. To reset, use the formula shown below:

[0149]

[0150] in, Representative matrix elements, yes The initial value;

[0151] Step 3.6: Based on the pseudo-Jacobi parameter matrix Calculate the control increment in the control time domain based on the input and output data of the process control loop. The control input u(k) at time k;

[0152] The formulas for calculating u(k) are as follows:

[0153]

[0154]

[0155] in, This represents the vector form of the input increment in the control time domain, where ρ = 0.95 represents the step size factor. It is 2N u A 2-row block matrix, where A1(k) represents the block coefficient matrix, is defined as follows:

[0156]

[0157] Among them, 0 2×2 It is a 2x2 matrix of zeros;

[0158] Step 3.7: Input the obtained control into the dissolved oxygen conversion coefficient K La5 Internal return flow Q a The value applied to the wastewater treatment system S NO2 and dissolved oxygen D O5 Optimize the control of setpoints;

[0159] Step 4: Obtain the S-value of the wastewater treatment process operation optimization control method. NO2 and dissolved oxygen D O5 The actual concentration value was calculated, along with energy consumption, effluent water quality, and effluent ammonia nitrogen (S). NHe and total nitrogen (N) in effluent tote index.

[0160] In this embodiment, after 7 days of operation, the S of the wastewater treatment process using the method of the present invention is obtained. NO2 and dissolved oxygen D O5 The real-time control value of the concentration is obtained, and simultaneously, the energy consumption, effluent water quality, and effluent ammonia nitrogen S are obtained for 7 days of operation of the wastewater treatment process under the method of this invention. NHe and total nitrogen (N) in effluent tote Indicator value.

[0161] To verify the effectiveness of the method of the present invention, a simulation experiment for optimized control was designed and implemented. Using the method of the present invention, energy consumption, effluent water quality models, and effluent constraint index S were established. NHe and N tote Data-driven model; obtaining S NO2 and dissolved oxygen D O5 Optimize the setpoints and track and control them; calculate the energy consumption, effluent quality, and effluent ammonia nitrogen (S) of the wastewater treatment process operating for 7 days under the method of this invention. NHe and total nitrogen (N) in effluent tote The specific indicator values ​​are as follows:

[0162] ① Set up S in the experiment NO2 and dissolved oxygen D O5 The concentration optimization period was 2 hours (h), the sampling period of the process control loop was set to 15 minutes, i.e., 0.25 hours (h), and the experimental running time was 7 days (d), i.e., 168 hours (h). Using the method of this invention, energy consumption, effluent water quality models, and effluent constraint index S were established. NHe and N tote Data-driven model; obtaining S NO2 and dissolved oxygen D O5 Optimize the setpoints and track and control them;

[0163] In this embodiment, under the adaptive constrained multi-objective operation optimization control method, the recursive bilinear subspace identification model of wastewater treatment process energy consumption (EC), effluent water quality (EQ), and the probability density function (PDF) distribution shape of modeling error are as follows: Figure 2 As shown, Figure 2 (a) Represents the recursive bilinear subspace identification model of energy consumption EC, with the X-axis representing time in days (d) and the Y-axis representing energy consumption EC values ​​in kilowatt-hours per day (kW·h·d). -1 (The solid line represents the actual EC value, and the dashed line represents the EC model output value.) Figure 2 (b) Represents the shape of the probability density function PDF distribution of the energy consumption EC model. The X-axis represents the modeling error of energy consumption EC, expressed in kilowatt-hours per day (kW·h·d). -1 Y-axis: PDF of energy consumption EC modeling error, unitless; Figure 2 (c) Represents the recursive bilinear subspace identification model for effluent water quality EQ, with the X-axis representing time in days (d) and the Y-axis representing effluent water quality EQ values ​​in kilograms per millisecond per day (kg poll·d). -1 (The solid line represents the actual EQ value, and the dashed line represents the EQ model output value.) Figure 2 (d) represents the shape of the probability density function PDF distribution of the effluent water quality EQ model. The X-axis represents the modeling error of energy consumption effluent water quality EQ, expressed in kilograms per millisecond per day (kg poll·d). -1Y-axis: PDF of effluent water quality EQ modeling error, unitless; Figure 3 (a) indicates the effluent index ammonia nitrogen (S) NHe The recursive bilinear subspace identification model, X-axis: time, in days (d), Y-axis: effluent index ammonia nitrogen S NHe Values, in milligrams per liter (mg / L), solid lines represent S. NHe Actual value, dashed line represents S NHe Model output values; Figure 3 (b) indicates the effluent index S NHe The probability density function (PDF) distribution shape of the model, X-axis: effluent index S NHe Modeling error, in milligrams per liter (mg / L), Y-axis: effluent index S NHe PDF of modeling error, unitless; Figure 3 (c) indicates the effluent index N tote The recursive bilinear subspace identification model, X-axis: time, unit is days (d), Y-axis: water output index N tote Values, in milligrams per liter (mg / L), solid line represents N. tote Actual value, dashed line represents N tote Model output values; Figure 3 (d) indicates the effluent index N tote The probability density function (PDF) distribution shape of the model, X-axis: effluent index N tote Modeling error, in milligrams per liter (mg / L), Y-axis: effluent index N tote PDF of modeling error, unitless; Figure 4 The display shows that under the adaptive constrained multi-objective operation optimization control method, the wastewater treatment process S NO2 Real-time control results, such as Figure 4 As shown, the X-axis represents time, in days (d), and the Y-axis represents space (S). NO2 The concentration value is expressed in milligrams per liter (mg / L), with the solid line representing S. NO2 The control result value, the dashed line is S NO2 The optimized setpoint; under the adaptive constrained multi-objective operation optimization control method, the wastewater treatment process D O5 Real-time control results, such as Figure 5 As shown, the X-axis represents time, in days (d), and the Y-axis represents D. O5 The concentration value is expressed in milligrams per liter (mg / L), with the solid line representing D. O5 The control result value, the dashed line is D O5 Optimized settings.

[0164] ② Obtain the energy consumption, effluent quality, and effluent ammonia nitrogen (S) of the wastewater treatment process operating for 7 days using the method of this invention. NHe and total nitrogen (N) in effluenttote Indicator value;

[0165] The energy consumption for wastewater treatment using this invention for 7 days was 3631.37 kWh / day. -1 Compared to the energy consumption of conventional PID control methods (3917.73 kW·h·d), -1 The concentration of pollutants decreased significantly; after 7 days of operation using this invention, the effluent quality of the treated wastewater was 5988.86 kg / milliseconds per day (kg poll·d). -1 Compared to the effluent quality of conventional PID control methods (6071.87 kg poll·d), -1 There has been a significant improvement; the effluent index S NHe It is 2.44 mg / L, which meets the S requirement. NHe The discharge standard is 4 mg / L, and the effluent index is N. tote The concentration is 15.94 mg / L, which meets the N requirement. tote The emission standard is 18 mg / L. The energy consumption, effluent quality, and effluent index results of this invention show that it effectively improves the operating efficiency of sewage treatment and achieves energy conservation and consumption reduction while ensuring that the effluent index meets the national emission standards.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. An adaptive constrained multi-objective operation optimization control method for a wastewater treatment process, characterized in that: Models for energy consumption, effluent quality, and effluent indicators are established using a recursive bilinear subspace identification and modeling method. An adaptive constrained multi-objective evolutionary method was used to optimize energy consumption and effluent quality, and to address effluent index constraints, thereby obtaining nitrate nitrogen S. NO2 and dissolved oxygen D O5 Optimized concentration settings; Model-free adaptive predictive control is used to achieve control of S NO2 and dissolved oxygen D O5 Real-time concentration control; Includes the following steps: Step 1: Select nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is the optimization variable, with energy consumption (EC) and effluent water quality (EQ) as optimization objectives, and effluent ammonia nitrogen (S) as the optimization target. NHe and total nitrogen (N) in effluent tote To establish a constrained multi-objective optimization problem for the wastewater treatment process, with effluent constraints as the main metric; Step 2: Design an optimization method based on an adaptive constrained multi-objective evolutionary algorithm, and adaptively select the optimization variable S. NO2 and D O5 Optimized settings; Step 2.1: In the crossover evolution of individuals in a multi-objective evolutionary algorithm population, the neighborhood of an individual is defined using a comprehensive distance approach; Step 2.2: The i-th individual in the g-th generation of the population iteration When performing crossover operations, the formula for calculating the comprehensive distance is used. The T individuals with the smallest combined distance from each individual in the current generation population are selected to form an individual population. In the current generation of comprehensive distance neighborhood, simultaneously calculate The T individuals with the smallest combined distances to every individual in the previous generation population are selected to form an individual population. In the previous generation's integrated distance neighborhood, individuals The indices of the combined distance neighborhood members in the current generation and the previous generation are respectively denoted as... and T is the size of the neighborhood; Step 2.3: With a 50% probability, perform cross-generational information crossover / mutation based on the comprehensive distance neighborhood; Step 2.4: Perform cross-generational information crossover / mutation based on the population with a 50% probability; Step 2.5: Perform polynomial mutation on the mutant individuals; Step 2.6: Offset the offspring individuals after crossover mutation i Together with all other individuals in the population, the fitness function value of each individual is calculated using an adaptive penalty method, and a penalty coefficient is adaptively applied to individuals that do not meet the constraints. Step 2.7: Calculate individual off i The fitness function values ​​of all other individuals in the population are used, and these fitness function values ​​are used to replace the individual's off-value. i The objective function values ​​corresponding to all other individuals in the population are used to evaluate the off-off values ​​of off-off values ​​for off-off ... i Screening was performed to complete population P. g =[x 1g ,…, x Ng Update of ]; Step 2.8: If the current iteration step g is less than the maximum iteration step G, go to step 2.2; if the current iteration step g is greater than or equal to the maximum iteration step G, go to step 2.

9. Step 2.9: Based on the effluent water quality threshold EQT principle, if the individual x with the lowest energy consumption in the population after the iteration is... EC,min The corresponding effluent water quality target value is less than or equal to the EQT value, and the effluent index S NHe N tote If the emission standards are met, then select individual x. EC,min For S NO2 and D O5 The optimal setting value is determined; otherwise, the individual x with the lowest effluent quality in the population after the iteration ends is selected. EQ,min For S NO2 and D O5 Optimize the settings to achieve S NO2 and D O5 Optimize adaptive selection of setpoints; Step 3: Design a control method for the process control loop based on model-free adaptive predictive control; Step 4: Obtain the S-value of the wastewater treatment process operation optimization control method. NO2 and dissolved oxygen D O5 The actual concentration value was calculated, along with energy consumption, effluent water quality, and effluent ammonia nitrogen (S). NHe and total nitrogen (N) in effluent tote index.

2. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 1, characterized in that: The specific method for step 1 is as follows: Step 1.1: Collect optimization variable S using a sliding window data acquisition method. NO2 and D O5 Optimize target EC and EQ and effluent constraint S NHe and N tote Real-time data is used to construct the database required for building the optimization target model and the effluent constraint model; Step 1.2: Establish models for energy consumption, effluent quality, and effluent constraint indicators using the recursive bilinear subspace identification online modeling method, as shown in the following formula: (1) (2) (3) (4) in, and These represent the energy consumption and effluent water quality model outputs, respectively. and The following represent the effluent constraint index, ammonia nitrogen (S). NHe and total nitrogen N tote Model output, Indicates the input optimization variable S NO2 and D O5 , These are energy consumption, effluent quality, and S. NHe N tote Model parameters with respect to the input data These are energy consumption, effluent quality, and S. NHe N tote Model parameters based on historical data, These are energy consumption, effluent water quality, and S. NHe N tote The model's historical input and output data, These represent energy consumption, effluent water quality, and S, respectively. NHe N tote Model parameters with respect to the bilinear subspace matrix of the model input and output data. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The model's input and output data are bilinear subspace matrices. These are energy consumption, effluent water quality, and S. NHe N tote The model parameter matrix of the model. These represent energy consumption, effluent water quality, and S, respectively. NHe N tote The real-time data vectors used in model building, and the model parameters Capable of using real-time data vectors Online learning and adaptive updating are performed using the following recursive least squares method with a forgetting factor: (5) in, This represents the model parameter matrix at time k corresponding to the established model of energy consumption, effluent quality, and effluent constraint indicators. This represents the model output at time k+1, corresponding to the established model for energy consumption, effluent quality, and effluent constraint indicators. This represents the real-time data used at time k+1 by the established model of energy consumption, effluent quality, and effluent constraint indicators. k+1 P represents the gain vector. k+1 Let λ be the covariance matrix, λ be the forgetting factor, and I be the identity matrix. Step 1.3: Using equations (1) and (2) as the optimized performance index model, and equations (3) and (4) as the effluent index constraint model, with S... NO2 and D O5 To optimize the variables, we obtain a constrained multi-objective optimization problem for wastewater treatment.

3. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 2, characterized in that: The data acquisition method of the sliding window is as follows: the original database used for modeling is constructed from the sewage treatment operation data within a fixed period. As the sewage treatment plant operates, every optimization cycle, a set of old data is deleted from the database and a set of the same number of real-time new data is added. The total amount of data in the database remains unchanged, thereby realizing data acquisition and database updates and improving the real-time performance of the modeling data.

4. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 3, characterized in that: The comprehensive distance calculation formula mentioned in step 2.1 is as follows: (6) in, Represents vector x s and The combined distance between them and Let s and v represent the s-th individual and v-th individual in the population, respectively. N represents the population size. This is a weighting factor used to adjust distance similarity. and angle similarity The weighting in the overall distance, where e is the natural index. x represents s and The distance between them x represents s and x v The angle between them and The calculation formulas are as follows: (7) (8) in, Describing the L2 norm of a vector, Represents the dot product of vectors. This represents the inverse cosine operation.

5. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 4, characterized in that: The specific method for step 2.3 is as follows: Set x ig Let x be the i-th parent individual in the current generation population. ij,g Representative individual x ig The j-th element, y ig Indicates based on parent individual x ig The newly generated mutant individuals undergo cross-generational information cross-operation based on the comprehensive distance neighborhood, as shown in the following formula: (9) (10) Among them, y ij,g Represents the mutant individual y ig The j-th element, where the index g represents the current generation population and g-1 represents the previous generation population. From parent individual x ig A neighboring individual randomly selected from the current generation's neighborhood. for The j-th element, ,and From parent individual x ig A neighboring individual randomly selected from the previous generation's neighborhood. It is its j-th element. μ j For a random number that follows a uniform distribution, η c >0 represents the cross-distribution factor. It is determined by the cross-distribution factor η c Dynamically generated random numbers.

6. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 5, characterized in that: The cross-generational information crossover operation based on the population described in step 2.4 is shown in the following formula: (11) in, It is an individual randomly selected from all individuals in the current generation population. for The j-th element, , It is an individual randomly selected from all individuals in the previous generation population. Represents its j-th element, ; The polynomial mutation formula mentioned in step 2.5 is as follows: (12) Among them, off ij The off-number of off-numbers generated by the final generation of crossover mutation i The j-th element, p m Indicates the probability of mutation. For polynomial variation increments, The specific calculation formula is as follows: (13) Where, η m This represents the variation distribution factor.

7. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 6, characterized in that: Step 2.6 describes the individual off i The formula for calculating the fitness function value is: (14) (15) in, Indicates individual off i The fitness function value of the j-th objective function. Indicates individual off i The value of the j-th objective function, The value is determined by the mean of the j-th objective function in the population. Let $\mathbf{j}$ represent the mean of the objective function for all individuals in the current generation of the population. Indicates the first One constraint It is the first The penalty coefficient for each constraint, It can be calculated using the following formula: (16) in, is the mean of the number of times the l-th constraint is violated in the population, and K is the total number of constraints.

8. The adaptive constrained multi-objective operation optimization control method for a wastewater treatment process according to claim 7, characterized in that: The specific method for step 3 is as follows: Step 3.1: Using the dissolved oxygen conversion coefficient K La5 Internal return flow Q a As the input to the process control loop, nitrate nitrogen S NO2 and dissolved oxygen D O5 Concentration is used as the output of the process control loop, in the form of nitrate nitrogen S. NO2 and dissolved oxygen D O5 The optimal setpoint for concentration is used as the desired output of the process control loop; Step 3.2: Collect dissolved oxygen conversion coefficient K La5 Internal return flow Q a Nitrate nitrogen S NO2 Dissolved oxygen D O5 Based on real-time data, estimate the pseudo-Jacobi parameter matrix of the improved model-free adaptive predictive control at the current moment. ; pseudo-Jacobi parameter matrix The estimation formula is: (17) in, Represents the pseudo-Jacobi parameter matrix The estimate, This represents the pseudo-Jacobi parameter matrix of the previous time step. express The estimate, The output S at time k NO2 and D O5 The increment, The control input K at time k-1 is represented by La5 and Q a The increment, Indicates the weighting factor. , Indicates the step size factor. ; Step 3.3: Calculate the pseudo-Jacobi parameter matrix at the current time. Estimate To reset, use the formula shown below: (18) in, represent elements, , yes initial value, It is to find the absolute value. It is a sign function, where α, b1, and b2 are constants greater than 0, and satisfy α ≥ 1. , =2 indicates the number of input variables; Step 3.4: Utilize the pseudo-Jacobi parameter matrix at known times Predicting the pseudo-Jacobi parameter matrix for future times based on an autoregressive model. ; The formula for the autoregressive model is: (19) in, To control the time domain, n represents the coefficients of the autoregressive model. p The order of the model is given by... The coefficient vector matrix formed Calculated using the following formula: (20) in, , The weight parameter is greater than 0. ; Step 3.5: Estimation of the pseudo-Jacobi parameter matrix for the predicted future times To reset, use the formula shown below: (21) in, represent elements, yes Initial value; Step 3.6: Estimation based on the pseudo-Jacobi parameter matrix , Calculate the control increment in the control time domain based on the input and output data of the process control loop. The control input u(k) at time k; The formulas for calculating u(k) are as follows: (22) (23) in, The vector form representing the input increment in the control time domain. Indicates the step size factor. yes A 2-row block matrix, where A1(k) represents the block coefficient matrix, is defined as follows: (24) in, It is a 2x2 matrix of zeros; Step 3.7: Input the obtained control into the dissolved oxygen conversion coefficient K La5 Internal return flow Q a The value applied to the wastewater treatment system S NO2 and dissolved oxygen D O5 Optimize the control of set values.

Images