Dynamic multi-objective optimization control method for sewage treatment process based on probability and dimension joint prediction

Abstract

The invention provides a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction, and relates to the technical field of sewage treatment. Comprising the following steps: acquiring process data of a sewage treatment process, constructing an optimization objective function according to the predictor based on support vector regression and the process data; solving the objective function by using a dynamic multi-objective optimization algorithm and a fast elite multi-objective genetic algorithm based on probability and dimension joint prediction to obtain a group of Pareto optimal solutions; according to a group of Pareto optimal solutions, determining optimal set values of the dissolved oxygen concentration of the fifth zone and the nitrate nitrogen concentration of the second zone; and carrying out tracking control on the actual dissolved oxygen concentration of the fifth zone and the nitrate nitrogen concentration value of the second zone by utilizing a PID (Proportion Integration Differentiation) controller and the optimized set values of the dissolved oxygen concentration of the fifth zone and the nitrate nitrogen concentration of the second zone. The problem of low control accuracy of corresponding sewage treatment in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment, and in particular to a dynamic multi-objective optimization control method for a sewage treatment process based on joint prediction of probability and dimension. Background Art

[0002] The sewage treatment process has complex biological, physical and chemical characteristics, and it is difficult to achieve effective control due to the strong coupling relationship between dependent variables and highly nonlinear characteristics. At present, my country generally adopts the activated sludge method to remove pollutants in water. This biochemical reaction process uses pumping for sludge return and aeration, which consumes a lot of electricity and causes huge operating costs. With the increase in sewage discharge and the increasingly stringent emission standards, the cost of sewage treatment continues to rise. Therefore, how to reduce energy consumption while ensuring that the effluent quality meets the standards has become a key issue that needs to be urgently solved in sewage treatment.

[0003] In order to improve the operation performance of urban sewage treatment and achieve a balance between effluent water quality and operating energy consumption, optimization control strategies have emerged. In order to verify the effectiveness of different optimization control methods in the sewage treatment process, the International Water Quality Association and the European Union Scientific and Technical Cooperation Organization jointly developed the Benchmark Simulation Model No. 1 (BSM1) for sewage treatment. BSM1 consists of a biochemical reaction tank and a secondary sedimentation tank. The biochemical reaction tank is divided into five units. The first two units are anoxic zones, which mainly carry out denitrification reactions, and the last three units are aerobic zones, which mainly carry out nitrification reactions. After the sewage is clarified in the secondary sedimentation tank, the upper clear liquid flows out, while part of the bottom sludge enters the anoxic zone through external reflux, and the other part is discharged as residual sludge. The basic idea of ​​optimization control of the urban sewage treatment process is to optimize the operating indicators such as effluent water quality and operating energy consumption by designing appropriate optimization strategies and control strategies. In order to accurately describe the dynamic relationship between effluent water quality, operating energy consumption and their key variables, it is necessary to construct an optimization model for the operation indicators of the urban sewage treatment process. In addition, in the actual sewage treatment process, the changes in the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition will directly affect the efficiency of the nitrification and denitrification processes. Therefore, there are two control loops, the first loop adjusts the dissolved oxygen concentration through the oxygen transfer coefficient of the fifth partition, and the second loop adjusts the nitrate nitrogen concentration through the internal recirculation flow of the second partition. Therefore, it is necessary to design a reasonable dynamic optimization control method for the sewage treatment process, which can not only calculate the optimal set values ​​of the dissolved oxygen concentration and the nitrate nitrogen concentration in real time, but also realize the tracking control of the optimal set values ​​of the dissolved oxygen concentration and the nitrate nitrogen concentration. Summary of the invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a dynamic multi-objective optimization control method for a sewage treatment process based on joint prediction of probability and dimension.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction includes:

[0007] Obtaining process data of the sewage treatment process, wherein the process data includes: dissolved oxygen concentration, nitrate nitrogen concentration, suspended solids concentration, ammonia nitrogen concentration and inflow flow;

[0008] An optimization objective function is constructed according to a predictor based on support vector regression and the process data, wherein the optimization objective function includes: an effluent water quality index model and an energy consumption performance index model, and the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition are decision variables of the optimization objective function;

[0009] A set of Pareto optimal solutions are obtained by solving the objective function using a dynamic multi-objective optimization algorithm based on probability and dimensionality joint prediction and a fast elite multi-objective genetic algorithm;

[0010] Determine the optimal set values ​​of dissolved oxygen concentration in the fifth partition and nitrate nitrogen concentration in the second partition according to the set of Pareto optimal solutions;

[0011] The actual dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition are tracked and controlled by using a PID controller and the optimized set values ​​of the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition.

[0012] Preferably, the expression of the effluent water quality index model is:

[0013]

[0014] Among them, f EC (x(t)) is the effluent water quality model at time t, y(i) is the target variable, and the input variable is x(t) = [S O,sat (t),S NO,e (t),MLSS(t),S NH (t),Q in (t)],S O , sat(t) is the dissolved oxygen concentration at time t, S NO,e (t) is the nitrate nitrogen concentration at time t, MLSS(t) is the suspended solids concentration at time t, S NH (t) is the ammonia nitrogen concentration at time t, Q in (t) is the water inflow rate at time t, <·,·> is expressed as a dot product, φ(x) is a nonlinear mapping, α i1 , are the first Lagrange multiplier and the second Lagrange multiplier respectively, ε1 is the first loss deviation, K1(x(i), x(c)) is the first Gaussian RBF kernel function, w1 is the first weight vector, b1 is the first threshold, α c1 , They are the fifth Lagrange multiplier and the sixth Lagrange multiplier respectively.

[0015] 3. According to the method of dynamic multi-objective optimization control of sewage treatment process based on probability and dimension joint prediction according to claim 2, it is characterized in that the expression of energy consumption performance index model is:

[0016]

[0017] Among them, K2(x(i),x(c)) is the second Gaussian RBF kernel function, α i2 , are the third and fourth Lagrange multipliers respectively, w2 is the second weight vector, b2 is the second threshold, α c2 , They are the seventh and eighth Lagrange multipliers respectively.

[0018] Preferably, the dynamic multi-objective optimization algorithm based on probability and dimensionality joint prediction and the fast elite multi-objective genetic algorithm are used to solve the objective function to obtain a set of Pareto optimal solutions, including:

[0019] Set the population size and maximum number of iterations;

[0020] According to the population size, a preset number of populations are selected and the fitness values ​​of all individuals in the currently selected population, i.e., the objective function values ​​of the individuals, are calculated;

[0021] Obtaining an environmental change result according to the adaptation value and the environmental change detection formula;

[0022] It is judged whether the environmental change result is greater than the environmental change threshold to obtain a judgment result. If the judgment result is yes, a population under a new environment is generated and a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions. If the judgment result is no, a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions.

[0023] Preferably, the expression of the environmental change detection formula is:

[0024]

[0025] Where m0 is the number of samples drawn to detect environmental changes, N is the population size, k is a positive integer between 1 and t-1, and ||·|| is the Euclidean distance. i ,t) and F(X i ,k) are the objective function values ​​of the individuals at time t and time k respectively.

[0026] Preferably, the method of generating a population under a new environment and optimizing the objective function using a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions includes:

[0027] A probability prediction model is established based on the Gamma distribution to generate the first initial sub-population;

[0028] The second initial sub-population is generated by establishing a denoising autoencoder for dimensionality prediction;

[0029] Merging the first initial subpopulation and the second initial subpopulation to obtain a population under a new environment;

[0030] A corresponding set of Pareto optimal solutions is obtained according to the population under the new environment.

[0031] Preferably, a probability prediction model is established based on the Gamma distribution to generate a first initial subpopulation, including:

[0032] Construct a Gamma distribution based on the population in the previous environment;

[0033] Calculate the mean and covariance of the population in the previous environment according to the Gamma distribution;

[0034] Obtaining shape parameters and scale parameters of the Gamma distribution according to the mean and covariance;

[0035] Adjusting the mean value according to the disturbance factor to obtain an adjusted mean value;

[0036] The shape parameter and scale parameter of the Gamma distribution are updated using the adjusted mean to obtain a first initial subpopulation in a new environment.

[0037] Preferably, the step of generating a second initial subpopulation by establishing a denoising autoencoder for dimensionality prediction includes:

[0038] Obtain the populations corresponding to adjacent moments in the previous environment, and obtain the population at the first moment and the population at the second moment;

[0039] Using fast non-dominated sorting and crowding distance to screen the solutions of the first moment population and the second moment population to obtain a first screened solution and a second screened solution;

[0040] Using the second screened solution as the input of the first layer of the denoising autoencoder, and using the first screened solution as the output of the first layer of the denoising autoencoder to obtain a first dimension change rule;

[0041] Set a nonlinear function, obtain the input of the second layer of the denoising autoencoder according to the first dimension change law and the second screening solution, and use the first screening solution as the output of the second layer of the denoising autoencoder to obtain the second dimension change law;

[0042] According to the first dimension change law, the second dimension change law and the first screening solution, individuals of each dimension of the corresponding population in the new environment are obtained;

[0043] The individuals in each dimension are dimensionally spliced ​​to obtain a second initial sub-population in a new environment.

[0044] Preferably, the expression of the PID is:

[0045]

[0046] Where u(t)=[K L a5(t),Q a (t)] T , K L a5(t) is the oxygen transfer coefficient of the fifth partition at time t, Q a (t) is the internal recirculation flow of the second partition at time t, K p , K i , K d They are proportional coefficient matrix, integral coefficient matrix, and differential coefficient matrix respectively. e(t) is the control error between the actual output of dissolved oxygen or nitrate nitrogen concentration at time t and the optimized set value.

[0047] The present invention discloses the following technical effects:

[0048] The present invention provides a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction, comprising: obtaining process data of a sewage treatment process, wherein the process data comprises: dissolved oxygen concentration, nitrate nitrogen concentration, suspended solids concentration, ammonia nitrogen concentration and inlet flow rate; constructing an optimization objective function according to a predictor based on support vector regression and the process data, wherein the optimization objective function comprises: an effluent water quality index model and an energy consumption performance index model, wherein the dissolved oxygen concentration of the fifth partition and the nitrate nitrogen concentration of the second partition are decision variables of the optimization objective function; solving the objective function by using a dynamic multi-objective optimization algorithm based on probability and dimension joint prediction and a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions; determining the optimal set values ​​of the dissolved oxygen concentration of the fifth partition and the nitrate nitrogen concentration of the second partition according to the set of Pareto optimal solutions; and tracking and controlling the actual dissolved oxygen concentration of the fifth partition and the nitrate nitrogen concentration of the second partition by using a PID controller and the optimal set values ​​of the dissolved oxygen concentration of the fifth partition and the nitrate nitrogen concentration of the second partition. The present invention establishes an effluent water quality and energy consumption model based on a support vector regression algorithm and uses it as an objective function. The dynamic multi-objective optimization algorithm based on probability and dimension joint prediction is used to obtain the optimal set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration. The selection and tracking control of dissolved oxygen and nitrate nitrogen concentrations are carried out through the intelligent decision-making algorithm, thus solving the problem of difficult effluent water quality to meet the standard and high operating energy consumption; the optimization control method combining the dynamic multi-objective optimization algorithm based on probability and dimension joint prediction and the intelligent decision-making algorithm is used to achieve the optimal set value of dissolved oxygen concentration and nitrate nitrogen concentration and tracking control, which reduces the operating cost while ensuring that the effluent water quality meets the standard. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0050] Figure 1 A flow chart of a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimensionality joint prediction provided by an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of dynamic optimization control measurement provided by an embodiment of the present invention;

[0052] Figure 3 A diagram showing the dissolved oxygen concentration setting value and tracking result provided by an embodiment of the present invention;

[0053] Figure 4A dissolved oxygen concentration tracking control error diagram provided by an embodiment of the present invention;

[0054] Figure 5 A graph showing the nitrate nitrogen concentration setting value and tracking result provided by an embodiment of the present invention;

[0055] Figure 6 This is a nitrate nitrogen concentration tracking control error diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] like Figure 1-2 As shown, the present invention provides a dynamic multi-objective optimization control method for a sewage treatment process based on joint prediction of probability and dimension, comprising:

[0059] Step 100: Acquire process data of the sewage treatment process, wherein the process data includes: dissolved oxygen concentration, nitrate nitrogen concentration, suspended solid concentration, ammonia nitrogen concentration and influent flow;

[0060] Step 200: constructing an optimization objective function according to a predictor based on support vector regression and the process data, wherein the optimization objective function includes: an effluent water quality index model and an energy consumption performance index model, and the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition are decision variables of the optimization objective function;

[0061] Step 300: solving the objective function using a dynamic multi-objective optimization algorithm based on probability and dimensionality joint prediction and a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions;

[0062] Step 400: Determine the optimal set values ​​of the dissolved oxygen concentration in the fifth subarea and the nitrate nitrogen concentration in the second subarea according to the set of Pareto optimal solutions;

[0063] Step 500: Using a PID controller and the optimized set values ​​of the dissolved oxygen concentration in the fifth subarea and the nitrate nitrogen concentration in the second subarea to track and control the actual dissolved oxygen concentration in the fifth subarea and the nitrate nitrogen concentration in the second subarea.

[0064] Furthermore, it is clear that the main goal of sewage treatment is to reduce energy consumption while ensuring that water quality meets standards. Therefore, the conflicting effluent quality EQ and energy consumption EC are selected as optimization targets. The dynamic behavior characteristics and operation data of the sewage treatment process are analyzed to determine the main parameters affecting effluent quality and energy consumption: including dissolved oxygen concentration S O,sat , nitrate nitrogen concentration S NO,e , suspended solids concentration MLSS, ammonia nitrogen concentration S NH And the water flow Q in .

[0065] A predictor based on support vector regression (SVR) was used to approximate EC and EQ, and the effluent water quality and energy consumption performance index models were established.

[0066] Furthermore, the expression of the effluent water quality index model is:

[0067]

[0068] Among them, f EC (x(t)) is the effluent water quality model at time t, y(i) is the target variable, and the input variable is x(t) = [S O,sat (t),S NO,e (t),MLSS(t),S NH (t),Q in (t)],S O , sat(t) is the dissolved oxygen concentration at time t, S NO,e (t) is the nitrate nitrogen concentration at time t, MLSS(t) is the suspended solids concentration at time t, S NH (t) is the ammonia nitrogen concentration at time t, Q in (t) is the water inflow rate at time t, <·,·> is expressed as a dot product, φ(x) is a nonlinear mapping, α i1 , are the first Lagrange multiplier and the second Lagrange multiplier respectively, ε1 is the first loss deviation, K1(x(i), x(c)) is the first Gaussian RBF kernel function, w1 is the first weight vector, b1 is the first threshold, α c1 , They are the fifth and sixth Lagrange multipliers respectively.

[0069] Furthermore, the expression of the energy consumption performance index model is:

[0070]

[0071] Among them, K2(x(i),x(c)) is the second Gaussian RBF kernel function, α i2 , are the third and fourth Lagrange multipliers respectively, w2 is the second weight vector, b2 is the second threshold, α c2 , They are the seventh and eighth Lagrange multipliers respectively.

[0072] Furthermore, the objective function is solved by using a dynamic multi-objective optimization algorithm based on joint prediction of probability and dimension and a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions, including:

[0073] Set the population size and maximum number of iterations;

[0074] According to the population size, a preset number of populations are selected and the fitness values ​​of all individuals in the currently selected population, ie, the objective function value, are calculated.

[0075] Obtaining an environmental change result according to the adaptation value and the environmental change detection formula;

[0076] It is judged whether the environmental change result is greater than the environmental change threshold to obtain a judgment result. If the judgment result is yes, a population under a new environment is generated and a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions. If the judgment result is no, a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions.

[0077] Specifically, set the population size to N∈[200,250] and the maximum number of iterations to G max ∈[40,50];

[0078] Set the effluent water quality model f EC (x(t)) and energy consumption model f EQ (x(t)) is the optimization target, and the decision variable is x(t) = [x1(t), x2(t)], where x1(t) and x2(t) are the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition, respectively. The population is initialized and the objective function value is calculated;

[0079] The optimization period is set to T, that is, every T hours, 20% of the individuals in the population are randomly selected for re-evaluation to detect changes in the environment. At time t, m0 individuals are randomly selected from the current population as the candidate solution set S = {X i,1≤i≤m0}, calculate the fitness values ​​of all individuals in the candidate solution set. Substitute the fitness value into the environmental change detection formula. If s(t,k)≥0.005, it is considered that the environment has changed. The dynamic multi-objective optimization algorithm based on probability and dimension joint prediction is used to obtain the optimal setting values ​​of dissolved oxygen and nitrate nitrogen concentrations; when the environment has not changed, jump to the optimization step, that is, directly use the fast elite multi-objective genetic algorithm NSGA-II to optimize the initial population.

[0080] Furthermore, the expression of the environmental change detection formula is:

[0081]

[0082] Where m0 is the number of samples drawn to detect environmental changes, N is the population size, k is a positive integer between 1 and t-1, and ||·|| is the Euclidean distance. i ,t) and F(X i ,k) are the objective function values ​​of the individuals at time t and time k respectively.

[0083] Furthermore, the method of generating a population under a new environment and optimizing the objective function using a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions includes:

[0084] A probability prediction model is established based on the Gamma distribution to generate the first initial sub-population;

[0085] The second initial sub-population is generated by establishing a denoising autoencoder for dimensionality prediction;

[0086] Merging the first initial subpopulation and the second initial subpopulation to obtain a population under a new environment;

[0087] A corresponding set of Pareto optimal solutions is obtained according to the population under the new environment.

[0088] Specifically, the initial population of the new environment consists of two parts. First, a probability prediction model is established based on the Gamma distribution to generate 50% of the initial population. Second, a denoising autoencoder is established to perform dimensionality prediction to generate the remaining 50% of the initial population.

[0089] Furthermore, a probability prediction model is established based on the Gamma distribution to generate a first initial sub-population, including:

[0090] Construct a Gamma distribution based on the population in the previous environment;

[0091] Calculate the mean and covariance of the population in the previous environment according to the Gamma distribution;

[0092] Obtaining shape parameters and scale parameters of the Gamma distribution according to the mean and covariance;

[0093] Adjusting the mean value according to the disturbance factor to obtain an adjusted mean value;

[0094] The shape parameter and scale parameter of the Gamma distribution are updated using the adjusted mean to obtain a first initial subpopulation in a new environment.

[0095] Specifically, a probability prediction model is established based on the Gamma distribution: when the environment changes, the population and Pareto optimal solution POP at time t-1 are obtained t-1 and POS t-1 , get the Pareto optimal solution POS at time t-2 of the environment t-2 , used to predict the initial population of the environment at time t. First, based on the population POP of the previous environment t-1 To construct a Gamma distribution; secondly, in order to track the dynamic trend of the population, a disturbance factor λ based on historical information is introduced to control the disturbance level so as to obtain the initial population in the new environment. t-1 ={x i :1≤i≤N}, where N represents the number of solutions, x i represents a solution with d-dimensional decision variables. Assuming the population follows a Gamma distribution, calculate POP t-1 The mean μ and covariance σ of 2 ;

[0096]

[0097] According to the mean μ and covariance σ 2 , we can get the shape parameter α and scale parameter β of the Gamma distribution of the population in the environment at time t-1:

[0098]

[0099] The perturbation factor λ is introduced to adjust the mean μ to help track the new optimization direction. The adjusted mean is:

[0100] μ * =μ+λ=μ+(C t-1 -C t-2 )

[0101] Among them C t-1 and C t-2 Respectively represented as POS t-1 and POS t-2 The center of mass.

[0102] Use the adjusted mean μ* , update the shape parameter α and scale parameter β of the Gamma distribution, and obtain the probability density function corresponding to the population at time t in the new environment as follows:

[0103]

[0104] Where Γ(α) is the Gamma function, and the new solution sampled from the Gamma distribution is regarded as the predicted solution of the environment at time t, denoted as POP gd .

[0105] Furthermore, the step of generating a second initial subpopulation by establishing a denoising autoencoder for dimensionality prediction includes:

[0106] Obtain the populations corresponding to adjacent moments in the previous environment, and obtain the population at the first moment and the population at the second moment;

[0107] Using fast non-dominated sorting and crowding distance to screen the solutions of the first moment population and the second moment population to obtain a first screened solution and a second screened solution;

[0108] Using the second screened solution as the input of the first layer of the denoising autoencoder, and using the first screened solution as the output of the first layer of the denoising autoencoder to obtain a first dimension change rule;

[0109] Set a nonlinear function, obtain the input of the second layer of the denoising autoencoder according to the first dimension change law and the second screening solution, and use the first screening solution as the output of the second layer of the denoising autoencoder to obtain the second dimension change law;

[0110] According to the first dimension change law, the second dimension change law and the first screening solution, individuals of each dimension of the corresponding population in the new environment are obtained;

[0111] The individuals in each dimension are dimensionally spliced ​​to obtain a second initial sub-population in a new environment.

[0112] Specifically, a denoising autoencoder is established for dimensional prediction: Obtain the population POP at time t-1 and t-2 of the environment t-1 and POP t-2 , used to predict the initial population of the environment at time t. Using fast non-dominated sorting (FNS) and crowding distance (CD), POP t-1 and POP t-2 Half of the solutions are regarded as high-quality solutions, which are expressed as and Where N represents the population size. Use denoising autoencoder to extract NDS t-1 and NDS t-2 The dynamic change rules of each dimension. As the input of the first layer of the denoising autoencoder, As the first layer output, the extracted arrive The changing law of the jth dimension as follows:

[0113]

[0114] Where T is the matrix transposition operation, j = 1, ..., d, d is the dimension of the decision variable. In order to improve the model's ability to capture complex linear relationships, the nonlinear function tanh() is introduced. Set As the input of the second layer of the denoising autoencoder, As the second layer output, the extracted arrive The changing law of the jth dimension as follows:

[0115]

[0116] Based on the change patterns in each dimension learned by the denoising autoencoder, high-quality single-dimensional individuals in the new environment are generated. At time t, the j-th dimension individual of the predicted population is:

[0117]

[0118] Among them, Inipop j and It is a matrix with N / 2 rows and 1 column. By splicing these single-dimensional individuals in different dimensions, the predicted population Inipop at time t in the environment is generated, denoted as POP dl .

[0119] POP dl =Inipop=[Inipop 1 Inipop 2 … Inipop d ];

[0120] The POP obtained based on the probability prediction model gd And the solution POP based on the dimensional prediction model dl Merge to obtain the initial population of the new environment at time t.

[0121] The initial population is optimized using the fast elite multi-objective genetic algorithm NSGA-II to obtain a set of Pareto optimal solutions. Determine whether the algorithm has reached the set maximum number of evolutionary iterations G max , if reached, the iterative evolution process is terminated.

[0122] Furthermore, the expression of the PID is:

[0123]

[0124] Where u(t)=[K L a5(t),Q a (t)] T , K L a5(t) is the oxygen transfer coefficient of the fifth partition at time t, Q a (t) is the internal recirculation flow of the second partition at time t, K p , K i , K d They are proportional coefficient matrix, integral coefficient matrix, and differential coefficient matrix respectively. e(t) is the control error between the actual output of dissolved oxygen or nitrate nitrogen concentration at time t and the optimized set value.

[0125] In order to determine S O ,sat(t) and S NO,e (t) is the optimal set point, and a non-dominated solution needs to be selected from the obtained Pareto optimal set. However, the optimal solutions on the Pareto boundary do not dominate each other, and the specific selection of the optimal solution directly affects the final operating performance of the process. Therefore, it is necessary to make appropriate decisions on these optimal solutions to select a set of solutions that best meet the current operating conditions and specific process requirements as the final set values ​​of the proportional integral differential (PID) controller. The primary goal of a wastewater treatment plant is to ensure that water quality parameters meet discharge standards. However, ammonia nitrogen S NH (t) and total nitrogen N tot (t) is likely to exceed the corresponding emission standards and should be given special attention. Therefore, this paper designs an intelligent decision-making algorithm that takes into account specific technical requirements. First, based on support vector regression SVR, an S NH (t) and N tot (t) is the model output, S O,sat (t) and S NO,e (t) is the prediction model input by the model. For the obtained Pareto optimal solution, the S NH (t) and N tot (t) Whether the predicted value exceeds the emission standard. If the predicted value exceeds the standard, the solution with the smallest EQ is selected from the Pareto optimal set as the final set point of the controller. Otherwise, the inflection point in the Pareto optimal solution is selected as the final set point of the controller.

[0126] After determining the optimal set values ​​of key control variables dissolved oxygen and nitrate nitrogen, the PID controller is used for tracking control. The output of the dynamic multi-objective optimization control system for sewage treatment process based on probability and dimension joint prediction is the actual dissolved oxygen S O,sat (t) and nitrate nitrogen S NO,e (t) concentration value, Figure 3 This is a graph of dissolved oxygen concentration setting value and tracking results. The solid line is the optimized setting value of dissolved oxygen concentration, and the dotted line is the actual dissolved oxygen output concentration value. The horizontal axis is time, unit: day, and the vertical axis is dissolved oxygen concentration, unit: mg / L. Figure 4 This is the dissolved oxygen concentration tracking control error diagram. The horizontal axis is time, unit: day, and the vertical axis is the dissolved oxygen concentration error value, unit: mg / L. Figure 5 This is a graph of nitrate nitrogen concentration setting values ​​and tracking results. The solid line is the optimized setting value of nitrate nitrogen concentration, and the dotted line is the actual nitrate nitrogen output concentration value. The horizontal axis is time, unit: day, and the vertical axis is nitrate nitrogen concentration, unit: mg / L. Figure 6 It is the nitrate nitrogen concentration tracking control error diagram, horizontal axis: time, unit: day, vertical axis: nitrate nitrogen concentration error value, unit: mg / L.

[0127] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0128] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A dynamic multi-objective optimization control method for sewage treatment process based on probability and dimension joint prediction, characterized in that: include: Obtaining process data of the sewage treatment process, wherein the process data includes: dissolved oxygen concentration, nitrate nitrogen concentration, suspended solids concentration, ammonia nitrogen concentration and inflow flow; An optimization objective function is constructed according to a predictor based on support vector regression and the process data, wherein the optimization objective function includes: an effluent water quality index model and an energy consumption performance index model, and the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition are decision variables of the optimization objective function; A set of Pareto optimal solutions are obtained by solving the objective function using a dynamic multi-objective optimization algorithm based on probability and dimensionality joint prediction and a fast elite multi-objective genetic algorithm; Determine the optimal set values ​​of dissolved oxygen concentration in the fifth partition and nitrate nitrogen concentration in the second partition according to the set of Pareto optimal solutions; The actual dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition are tracked and controlled by using a PID controller and the optimized set values ​​of the dissolved oxygen concentration in the fifth partition and the nitrate nitrogen concentration in the second partition.

2. According to claim 1, a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction is characterized in that: The expression of the effluent water quality index model is: Among them, f EC (x(t)) is the effluent water quality model at time t, y(i) is the target variable, and the input variable is x(t) = [S O,sat (t),S NO,e (t),MLSS(t),S NH (t),Q in (t)],S O , sat(t) is the dissolved oxygen concentration at time t, S NO,e (t) is the nitrate nitrogen concentration at time t, MLSS(t) is the suspended solids concentration at time t, S NH (t) is the ammonia nitrogen concentration at time t, Q in (t) is the water inflow rate at time t, <·,·> is expressed as a dot product, φ(x) is a nonlinear mapping, α i1 , are the first Lagrange multiplier and the second Lagrange multiplier respectively, ε1 is the first loss deviation, K1(x(i), x(c)) is the first Gaussian RBF kernel function, w1 is the first weight vector, b1 is the first threshold, α c1 , They are the fifth Lagrange multiplier and the sixth Lagrange multiplier respectively.

3. According to claim 2, a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction is characterized in that: The expression of the energy consumption performance index model is: Among them, K2(x(i),x(c)) is the second Gaussian RBF kernel function, α i2 , are the third and fourth Lagrange multipliers respectively, w2 is the second weight vector, b2 is the second threshold, α c2 , They are the seventh and eighth Lagrange multipliers respectively.

4. According to claim 3, a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction is characterized in that: The objective function is solved by using a dynamic multi-objective optimization algorithm based on joint prediction of probability and dimension and a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions, including: Set the population size and maximum number of iterations; According to the population size, a preset number of populations are selected and the fitness values ​​of all individuals in the currently selected population, i.e., the objective function values ​​of the individuals, are calculated; Obtaining an environmental change result according to the adaptation value and the environmental change detection formula; It is judged whether the environmental change result is greater than the environmental change threshold to obtain a judgment result. If the judgment result is yes, a population under a new environment is generated and a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions. If the judgment result is no, a fast elite multi-objective genetic algorithm is used to optimize the objective function to obtain a set of Pareto optimal solutions.

5. According to claim 4, a dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction is characterized in that: The expression of the environmental change detection formula is: Where m0 is the number of samples drawn to detect environmental changes, N is the population size, k is a positive integer between 1 and t-1, and ||·|| is the Euclidean distance. i ,t) and F(X i ,k) are the objective function values ​​of the individuals at time t and time k respectively.

6. A dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction according to claim 4, characterized in that: The method of generating a population under a new environment and optimizing the objective function using a fast elite multi-objective genetic algorithm to obtain a set of Pareto optimal solutions includes: A probability prediction model is established based on the Gamma distribution to generate the first initial sub-population; The second initial sub-population is generated by establishing a denoising autoencoder for dimensionality prediction; Merging the first initial subpopulation and the second initial subpopulation to obtain a population under a new environment; A corresponding set of Pareto optimal solutions is obtained according to the population under the new environment.

7. A dynamic multi-objective optimization control method for a sewage treatment process based on probability and dimension joint prediction according to claim 6, characterized in that: A probability prediction model is established based on the Gamma distribution to generate the first initial sub-population, including: Construct a Gamma distribution based on the population in the previous environment; Calculate the mean and covariance of the population in the previous environment according to the Gamma distribution; Obtaining shape parameters and scale parameters of the Gamma distribution according to the mean and covariance; Adjusting the mean value according to the disturbance factor to obtain an adjusted mean value; The shape parameter and scale parameter of the Gamma distribution are updated using the adjusted mean to obtain a first initial subpopulation in a new environment.

8. The method for dynamic multi-objective optimization control of a sewage treatment process based on probability and dimension joint prediction according to claim 6 is characterized in that: The method of generating a second initial subpopulation by establishing a denoising autoencoder for dimensionality prediction includes: Obtain the populations corresponding to adjacent moments in the previous environment, and obtain the population at the first moment and the population at the second moment; Using fast non-dominated sorting and crowding distance to screen the solutions of the first moment population and the second moment population to obtain a first screened solution and a second screened solution; Using the second screened solution as the input of the first layer of the denoising autoencoder, and using the first screened solution as the output of the first layer of the denoising autoencoder to obtain a first dimension change rule; Set a nonlinear function, obtain the input of the second layer of the denoising autoencoder according to the first dimension change law and the second screening solution, and use the first screening solution as the output of the second layer of the denoising autoencoder to obtain the second dimension change law; According to the first dimension change law, the second dimension change law and the first screening solution, individuals of each dimension of the corresponding population in the new environment are obtained; The individuals in each dimension are dimensionally spliced ​​to obtain a second initial sub-population in a new environment.

9. The method for dynamic multi-objective optimization control of a sewage treatment process based on probability and dimension joint prediction according to claim 1 is characterized in that: The expression of the PID is: Where u(t)=[K L a5(t),Q a (t)] T , K L a5(t) is the oxygen transfer coefficient of the fifth partition at time t, Q a (t) is the internal recirculation flow of the second partition at time t, K p , K i , K d They are proportional coefficient matrix, integral coefficient matrix, and differential coefficient matrix respectively. e(t) is the control error between the actual output of dissolved oxygen or nitrate nitrogen concentration at time t and the optimized set value.

Images