Collaborative optimization control method for sewage treatment process based on knowledge transfer constrained MOEA / D

Abstract

The present invention discloses a collaborative optimization control method for a sewage treatment process based on knowledge transfer constrained MOEA / D. The method first uses a T-S fuzzy neural network to establish a soft measurement model for effluent indicators, and takes the balance of effluent energy consumption and water quality as the optimization objective function. Then, the collaborative optimization idea is used to adjust the control cycle through the optimized COD effluent indicator. Next, the dynamic neighborhood multi-objective optimization algorithm is used to optimize energy consumption and effluent water quality and to optimize the set value. In order to improve the optimization efficiency, the Pareto optimal solution is used to construct knowledge information, and the knowledge transfer strategy is used to generate the initial population. Finally, the optimized solution is used as the set value of the fuzzy logic controller to achieve precise control of the sewage treatment process, so as to improve the operating efficiency and stability, save energy and reduce consumption, and ensure that the effluent indicators meet the emission standards. This method improves the fixed-cycle optimization control strategy and the traditional MOEA / D algorithm, and effectively improves the operating efficiency and stability of the sewage treatment process.

Description

Technical Field

[0001] The present invention relates to the technical field of sewage treatment optimization control, and in particular to a sewage treatment process collaborative optimization control method based on knowledge transfer type constraint MOEA / D. Background Art

[0002] In recent years, my country's social economy has developed rapidly, but the water pollution problem that has come with it needs to be solved urgently. The total amount of sewage discharged in my country remains at a high level, and both surface water and groundwater have been polluted to varying degrees, and the pollution control situation is severe. Therefore, protecting water resources, controlling water pollution, and improving the water environment are important measures in the process of achieving sustainable development. In addition, with the development of society and the growth of population, the amount of sewage treatment has also doubled. The water quality standards for sewage treatment are becoming increasingly stringent, and the cost of sewage treatment plants is also increasing. Sewage treatment plants are used to collect and treat sewage, and treat sewage so that organic pollutants, nitrogen, phosphorus, etc. in it meet the discharge standards. Sewage treatment can reduce pollution to the water environment; at the same time, the recycled water that meets the standards is used for irrigation and planting, urban cleaning, and can promote the recycling of water resources. Therefore, building urban sewage treatment plants and improving sewage treatment capacity have become an inevitable choice to improve the urban ecological environment and alleviate water shortages.

[0003] The ultimate goal of sewage treatment is to ensure that various effluent water quality indicators such as suspended solids concentration, ammonia nitrogen concentration, chemical oxygen demand, biochemical oxygen demand, etc. can meet the standards. The huge energy consumed in controlling dissolved oxygen and nitrate nitrogen concentrations during sewage treatment has led to an increase in operating costs for sewage treatment plants while they are pursuing energy reduction to ensure water quality. Therefore, on the premise of ensuring that various effluent parameters meet the standards, how to effectively reduce the cost consumption in the sewage treatment process and achieve optimal control of the sewage treatment process has become an urgent problem to be solved. Summary of the invention

[0004] Purpose of the invention: In view of the problems pointed out in the background technology, the present invention proposes a wastewater treatment process collaborative optimization control method based on knowledge transfer constraint MOEA / D, and applies it to the wastewater treatment process to optimize the dissolved oxygen concentration and nitrate nitrogen concentration, so as to improve the effluent water quality and reduce energy consumption. O,5 、The nitrate nitrogen concentration in the second zone S NO,2 The optimal value is tracked and controlled in real time.

[0005] Technical solution: The present invention provides a collaborative optimization control method for a sewage treatment process based on knowledge transfer constraint MOEA / D, comprising the following steps:

[0006] Step 1: Collect real-time data during the sewage treatment process to ensure data accuracy and timeliness, which will help monitor system operation status, analyze performance and optimize it. Relevant data include nitrate nitrogen concentration (S NO,2 ), dissolved oxygen concentration (S O,5 ), dissolved oxygen conversion coefficient (K La, ), internal reflux (Q a ), operating energy consumption (EC, including pumping energy consumption and aeration energy consumption) and effluent water quality (EQ) and other data;

[0007] Step 2: Use TS fuzzy neural network to establish soft measurement model of total energy consumption and effluent water quality for real-time collected sewage treatment parameters, and use it as optimization objective function to predict total energy consumption and effluent water quality of the system; in the control cycle after the initial modeling, a model will be established based on the existing characteristic data and monitoring indicators to describe the new treatment process. The transfer learning method can be used to transfer the knowledge of the existing model to the new model, thereby accelerating the establishment and training process of the new model;

[0008] Step 3: Processing effluent quality constraints through adaptive constrained multi-objective evolutionary method;

[0009] Step 4: Use the knowledge transfer strategy to autonomously generate the initial population for this optimization process. A randomly initialized population is used in the first optimization. In the next round of optimization, the useful knowledge information of the previous optimization cycle is used to accelerate the new round of optimization process by calculating the similarity index between Pareto solutions and selecting a part of Pareto solutions with high relevance to the current problem as the initial population. Solutions with higher similarity usually have better heuristic information and can help accelerate the optimization process. This method can reduce the time and resources required for randomly initializing the population and improve optimization efficiency.

[0010] Step 5: Dynamic neighborhood MOEA / D algorithm based on density estimation. This algorithm improves the original MOEA / D by utilizing the angle evolution state judgment mechanism, distance density estimation strategy, dynamic neighborhood strategy and penalty factor dynamic adjustment strategy. The improved multi-objective algorithm is used to optimize the energy consumption and effluent water quality objective functions, obtain the Pareto solution set, and optimize the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration.

[0011] Step 6: Design and decide the optimal solution for each group of solutions by using the fuzzy membership function method, and obtain the nitrate nitrogen S NO,2 and dissolved oxygen S O,5 The optimal setting value of concentration achieves the best balance between the total energy consumption EC of effluent and the effluent water quality EQ;

[0012] Step 7: The optimized solutions of nitrate nitrogen concentration and dissolved oxygen concentration are used as tracking set values ​​of the fuzzy logic controller, and the fuzzy logic controller is used to output the real-time control values ​​of the dissolved oxygen conversion coefficient and the internal recirculation flow rate to control the nitrate nitrogen concentration and dissolved oxygen concentration in the sewage treatment process;

[0013] Step 8: Use the existing characteristic data and monitoring indicators, combined with the characteristics of the new treatment process, to adjust the treatment control cycle. By monitoring the COD value of the effluent, the stability problems of the treatment system can be discovered in time, and the treatment control cycle can be adjusted according to the changes in the COD value to ensure the stability of the effluent water quality. According to the monitoring indicator data of the existing treatment process, set the monitoring indicator threshold and detect the COD concentration of the water quality indicator. When the COD value of the monitoring indicator data of the new treatment process continues to rise or exceeds the threshold, the treatment control cycle needs to be shortened; if the COD value continues to decline or stabilizes at a low level, you can consider extending the treatment control cycle.

[0014] Beneficial effects: This scheme improves the fixed-cycle optimization control strategy and the traditional MOEA / D algorithm, proposes a collaborative optimization control method, a dynamic neighborhood adjustment MOEA / D, and a knowledge migration strategy, and applies them to the sewage treatment process to optimize the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration, completing the S NO,2 and S O,5 High-performance and precise control of the concentration optimization set value. This method can effectively improve the operating efficiency and stability of the sewage treatment process, and achieve energy saving and consumption reduction in the sewage treatment process on the basis that the effluent indicators meet the emission standards. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a diagram of the optimized control architecture of the sewage treatment process of the present invention;

[0016] Figure 2 A flowchart of the collaborative optimization control method for a sewage treatment process based on knowledge transfer constraint MOEA / D provided by the present invention;

[0017] Figure 3 For the present invention O Tracking control effects;

[0018] Figure 4 For the present invention NO Tracking control effects;

[0019] Figure 5 This is the effluent water quality tracking status of the present invention. DETAILED DESCRIPTION

[0020] The following further describes the implementation of the present invention in detail with reference to the accompanying drawings and specific examples.

[0021] The present invention provides a sewage treatment process collaborative optimization control method based on knowledge transfer constraint MOEA / D, comprising the following steps:

[0022] Step 1: Collect real-time data during the sewage treatment process to ensure data accuracy and timeliness, which will help monitor system operation status, analyze performance and optimize it. Relevant data include nitrate nitrogen concentration (S NO,2 ), dissolved oxygen concentration (S O,5 ), dissolved oxygen conversion coefficient (K La, ), internal reflux (Q a ), operating energy consumption (EC, including pumping energy consumption and aeration energy consumption) and effluent water quality (EQ) and other data;

[0023] Step 2: Use TS fuzzy neural network to establish soft measurement model of total energy consumption and effluent water quality for real-time collected sewage treatment parameters, and use it as optimization objective function to predict total energy consumption and effluent water quality of the system; in the control cycle after the initial modeling, a model will be established based on the existing characteristic data and monitoring indicators to describe the new treatment process. The transfer learning method can be used to transfer the knowledge of the existing model to the new model, thereby accelerating the establishment and training process of the new model;

[0024] The specific algorithm process is as follows:

[0025] Step 2.1: normalize the acquired modeling data to obtain normalized data;

[0026] Step 2.2: Divide the normalized data into training set and test set;

[0027] Step 2.3: Construct a TS fuzzy neural network (FNN) model, use the training set to train the random weight neural network, obtain the trained random weight neural network model as the model of energy consumption and water quality, and use the test set to test the model of energy consumption and water quality. Use the TS model to construct a 5-layer FNN consisting of an input layer, a membership function layer, a rule layer, a normalization layer, and an output layer;

[0028] The fuzzy rules of the TS FNN fuzzy neural network model are described as follows:

[0029] R j :If x1 is A 1j ,...,x n is A nj , then y is f(x1,…,x n )

[0030] Where j = 1, 2, ..., r, r represents the number of fuzzy rules, f(x1, ..., xn ) is the input variable x i A linear function of

[0031] Input layer: Each node has an input x=[x1,x2,…,x n ] T are connected, n is the number of input variables, that is, the number of nodes;

[0032] Membership function layer: Each node represents a linguistic variable value, which is used to calculate the membership of each input component to the fuzzy set of each linguistic variable value, using the Gaussian membership function:

[0033]

[0034] Among them, μ ij is the membership function of the i-th input variable belonging to the j-th fuzzy set, c ij With σ ij are the center and width of the corresponding Gaussian membership function, respectively;

[0035] Rule layer: Each node represents a fuzzy rule, which is used to match the premise of the fuzzy rule and calculate the applicability of each rule in the form of a product:

[0036] α j (x i )=μ 1j (x1)μ 2j (x2)...μ nj (x n )

[0037] Among them, j = 1, 2,…, r, r is the number of fuzzy rules.

[0038] Normalization layer: The number of nodes is the same as the third layer, and the output of the third layer is normalized:

[0039]

[0040] Output layer: also called defuzzification layer:

[0041]

[0042] y j (x i )=p 0j +p 1j x1+...+p nj x n ,j=1,2,...,r

[0043] Among them, y(x i ) is the network output, y j (xi ) is the consequent output of the jth rule, p ij (i=0,1,…,n,j=1,2,…,r) are coefficients, n is the number of input variables, and r is the number of fuzzy rules.

[0044] Step 3: Process the effluent quality constraints through the adaptive constrained multi-objective evolutionary method. Determine the optimization objectives and constraints of sewage treatment. The optimization objectives of sewage treatment are effluent quality and sewage treatment energy consumption.

[0045] Since a large amount of electricity is needed to maintain the aeration operation of the blower in the aerobic tank and the pumping operation of the sludge during the sewage treatment process, aeration energy consumption (AE) and pumping energy consumption (PE) are the main performance indicators, accounting for about 60% to 70% of the total energy consumption. The calculation formula is as follows:

[0046] EC=PE+AE

[0047]

[0048]

[0049] From the working process of the blower and pump, we can know that the factors affecting AE and PE are K La And the circulation flow rate Q in the system a (t), external circulation reflux Q r (t) and sludge discharge Q w (t).

[0050] The effluent quality (EQ) performance index helps to evaluate the effectiveness of sewage purification and refers to the fee to be paid for discharging pollutants. The EQ calculation formula is as follows:

[0051]

[0052] The relevant variables are described as follows:

[0053] SS e --Effluent suspended solids concentration; COD e --Effluent chemical oxygen demand concentration; BOD e --Biochemical oxygen demand concentration of effluent; S NKj，e --Kjeldahl nitrogen concentration in effluent; S NO，e --Effluent nitrate nitrogen concentration.

[0054] Due to the influence of BSM1 effluent quality limit, the establishment of a multi-objective model of sewage treatment process constraints includes:

[0055] min F(x)=(f EQ (x),f EC (x) T

[0056] Its inequality constraints are:

[0057]

[0058] The constructed sewage treatment optimization model is a multi-objective optimization model with effluent water quality constraints. Since the sewage treatment optimization model with constraints is not conducive to the algorithm's optimization when used as the optimization objective function, in order to solve the constrained optimization problem, the penalty function method is used to transform the constrained optimization model into an unconstrained optimization model.

[0059] The constraint penalty term is defined as:

[0060] f penalty (x)=max{g1(x)-4,0}+max{g2(x)-18,0}

[0061] Among them, g1(x) is the calculated value of ammonia nitrogen in the effluent water quality, and g2(x) is the calculated value of total nitrogen in the effluent water quality.

[0062] The penalty mechanism is used to limit the part of the effluent mass concentration that exceeds the limit, so that each average exceeding the limit part is multiplied by a larger penalty factor. Therefore, if a part exceeds the limit more seriously, the corresponding optimization objective function value will be larger, and it will be eliminated in the optimization process. Therefore, the multi-objective model of the sewage treatment process after sorting is introduced as follows:

[0063] min F(x)=(f EQ (x)+e·f penalty (x),f EC (x)+e·f penalty (x) T

[0064] Among them, ε is the penalty factor, f EQ (x) is the energy consumption soft sensor optimization model, f EC (x) is the water quality soft sensor optimization model.

[0065] Step 4: Use the knowledge transfer strategy to autonomously generate the initial population for this optimization process. In the first optimization, a random initialization population is used. In the next round of optimization, the useful knowledge information of the previous optimization cycle is used to accelerate the new round of optimization process by calculating the similarity index between Pareto solutions and selecting a part of Pareto solutions with high relevance to the current problem as the initial population.

[0066] The specific algorithm process is as follows:

[0067] Step 4.1: Obtain the corresponding environmental monitoring quantity under the actual environment and detect the actual water quality index S O,5 and SNO,2 ; Collect the Pareto solutions with the best performance in the multi-objective optimization problem obtained in the previous optimization process.

[0068] Step 4.2: Calculate the similarity s between the water quality index under actual environment and the Pareto solution of the optimal performance solution set i :

[0069]

[0070] In the formula, x k is the corresponding environmental monitoring quantity in the actual environment, x i is the solution to the ith subproblem of the Pareto solution in the multi-objective optimization problem, similarity(x i ,x k ) is the similarity distance calculation between the Pareto solution and the surrounding solution set.

[0071] Step 4.3: Based on the similarity distribution of water quality indicators and Pareto solutions in the actual environment, perform knowledge transfer based on the similarity index between each pair of solutions in the similarity matrix. Set the similarity threshold to s = 0.5. If s is satisfied i ≥s, the i-th Pareto solution is considered similar to the working condition of the actual working condition environment. Otherwise, the i-th Pareto solution is not considered as a similar working condition of the actual working condition environment and does not have the same knowledge information. Only when the similarity between two solutions is greater than this threshold, they are considered similar.

[0072] Step 4.4: Select similar solutions, traverse the Pareto solution set, and select solutions with similarity higher than the threshold as part of the initial population based on the calculated similarity. When more than half of the solutions in the Pareto solution set have similarity higher than the threshold, select the top 50 solutions with higher similarity based on the ranking of similarity. Based on the similarity matrix and the set similarity threshold, select the Pareto solutions with higher relevance to the current problem. Only solutions with similarity higher than the set threshold will be selected.

[0073] Step 4.5: According to the determined knowledge transfer strategy, use the Pareto solution to generate the initial population of this optimization process. The Pareto solution can be used as the seed solution, and then these seed solutions are selected according to the knowledge transfer strategy to generate new individuals as the initial population.

[0074] Step 5: Dynamic neighborhood MOEA / D algorithm based on density estimation. This algorithm improves the original MOEA / D by utilizing the angle evolution state judgment mechanism, distance density estimation strategy, dynamic neighborhood strategy and penalty factor dynamic adjustment strategy. The improved multi-objective algorithm is used to optimize the energy consumption and effluent water quality objective functions, obtain the Pareto solution set, and optimize the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration.

[0075] The specific algorithm process is as follows:

[0076] Step 5.1: The angle-based evolutionary state judgment mechanism uses the vector between the new solution and the old solution of the j-th problem and the weight vector λ j The angle between the vector X-X' and the weight vector λ is used to determine the evolutionary state of the population iteration. j The angle δ j , the weight vector is used as the guiding direction of population evolution. If the vector X-X' between the new solution and the old solution of a problem in the population is equal to the weight vector λ of the problem j The angle θ between i When it is an obtuse angle, it means that the new solution has good diversity and convergence in the iteration process. Otherwise, the diversity and convergence are relatively poor. Angle δ j The calculation formula is as follows:

[0077]

[0078] Among them, λ j is the weight vector of the j-th subproblem, and X-X' is the vector from the old solution X to the new solution X'.

[0079] Based on the evolutionary state of the angle, consider the following 3 cases:

[0080] (d) If δ i When the range is (90°, 130°), the convergence of the new solution is better than that of the old solution, but the diversity of the new solution is better than the convergence. In this case, the population has good diversity, and the population evolution tends to be diverse during the evolution process. Although the diversity develops well during evolution, the solution set is far away from the PF, and the convergence cannot meet the requirements. Appropriately reducing the neighborhood size is conducive to improving the convergence speed and reducing the computational cost while ensuring diversity.

[0081] (e) If δ i When the range is (130°, 140°), the convergence and diversity of the new solution are better than the old solution, and the diversity and convergence show the same performance trend. In this case, in order to maintain the current evolutionary state of the algorithm, the neighborhood size should remain unchanged. The population obtains good diversity and convergence, and the solutions obtained are evenly distributed near the actual PF, indicating that diversity and convergence show the same performance trend.

[0082] (f) If δ i When the range is (140°, 180°), the new solution has a certain diversity and convergence compared with the old solution, but the convergence is better than the diversity. In the process of evolution, the population evolution tends to converge. Although the solution set approaches the PF convergence and develops well in the evolution, the population diversity is affected. The development of diversity does not meet the expectations and cannot meet the requirements of evolution. Therefore, it is necessary to increase the neighborhood size appropriately. Increasing the neighborhood size is conducive to optimizing the unknown area in the evolution process, increasing the probability of generating new individuals, and improving the diversity of the population.

[0083] Step 5.2: Calculate the population density based on the distance-based density estimation scheme, which is an important basis for adjusting the neighborhood size. Calculate the vertical distance between the solution and the two adjacent weight vectors. Then determine the distribution range of the solution by comparing the two vertical distances. Solution x i With adjacent weight vector λ j The vertical distances are p ij 、p i,j+1 , where λ j is the weight vector of the jth subproblem. Solution x i With the weight vector λ j The vertical distance is less than the solution x i With adjacent weight vector λ j+1 The vertical distance of i is considered as the aggregation weight vector λ j The distribution solution of .

[0084] The local density of each subproblem is determined by the number of dominant and non-dominant solutions near each subproblem. The local solution density of the j-th subproblem is defined as:

[0085] In the density estimation strategy designed in this paper, the subproblem population density estimation will determine the population density of each subproblem based on the principle of MOEA / D according to the number of dominated solutions and non-dominated solutions near each decomposed subproblem. The solution density formula of each subproblem corresponding to the subproblem weight vector is:

[0086]

[0087] Among them, ld j is the local solution density of the jth subproblem, that is, it falls on the weight vector λ j The number of solutions within the range. min andld max They represent the minimum and maximum number of solutions respectively. ε is 10 -6 This is to prevent the denominator from being zero and to prevent the population density from being invalid.

[0088] ld jThe calculation formula is as follows:

[0089]

[0090] Among them, ξ ij is a logical variable used to represent the solution x i Whether to cluster on the weight vector λ j around.

[0091] If the solution is x i Aggregated on the weight vector λ j Around, then ij =1, otherwise ξ ij =0.ξ ij The formula is as follows:

[0092]

[0093] Among them, p ij is the solution vector x i To the adjacent weight vector λ j The vertical distance between the vertical distance p ij The calculation formula is:

[0094]

[0095] For simplicity, p th Set to x i The average of the vertical distances to two adjacent weight vectors. The specific definition is as follows:

[0096]

[0097] Among them, p ij is the solution vector x i To the adjacent weight vector λ j Vertical distance, p i,j+1 is the solution vector x i To the adjacent weight vector λ j+1 Vertical distance, p th is the solution vector x j To the adjacent weight vector λ i and λ i+1 The mean vertical distance.

[0098]

[0099] From this calculation, we can conclude that if the solution vector x in the population i To the adjacent weight vector λ j is less than the vertical distance to another adjacent weight vector, then x i Aggregated on the weight vector λ j around.

[0100] Local Density Measurement Model,Due to the characteristic of MOEA / D in selecting parents in the neighborhood during population evolution,,subregions are the evolutionary regions that will divide each subproblem according to the neighborhood size.,First, we propose a local density measurement model to estimate the local density of each subregion.,Decompose the target space into K subregions S1, S2,,...,S N . S j (l=1,...,N) is defined as:

[0101]

[0102] Among them, λ l is the unit direction vector, <u,λ l > is the weight vector λ corresponding to u and the lth subproblem l The angle between them, that is, the sub-area angle, T l is the size of the neighborhood corresponding to the lth subproblem, β th Set to be the angle between any two adjacent weight vectors.

[0103] β th The formula is as follows:

[0104]

[0105] Since each solution has a corresponding subregion, the local density of the subregion can be calculated by calculating the solutions belonging to the subregion. The solution density of the jth subregion can be expressed as:

[0106]

[0107] After obtaining the sub-region population density, in order to illustrate the difference in solutions corresponding to adjacent target vectors, after calculating the solution density of the sub-region, an indicator is defined to quantify the difference between any two solutions in the sub-region, which is defined as follows:

[0108]

[0109] Among them, h m is the minimum average fitness difference, S j is the jth sub-region, x i and x j are two solution vectors.

[0110] Finally, sub-region S j The total fitness difference is calculated as follows:

[0111]

[0112] Step 5.3: Dynamic Neighborhood Strategy

[0113] After determining the upward or downward trend of the neighborhood size according to the angle-based evolutionary state judgment mechanism in step 5.1, the neighborhood size is further updated according to the calculated neighborhood solution density. According to the three situations of the angle evolution state, the strategy of dynamic neighborhood adjustment is as follows:

[0114] (a) When δ j When the angle range is (90°, 130°), population diversity is better than convergence.

[0115] In this case, to increase the convergence speed, reduce the neighborhood size as follows:

[0116]

[0117] Among them, . is the upper limit function, △T j t is the change in neighborhood adjustment for the j-th subproblem at the n-th iteration, defined as follows:

[0118]

[0119] in, is the population density in the neighborhood of the j-th subproblem at the t-th iteration, is the median population density obtained by sorting the population density in the neighborhood of all subproblems at the tth iteration, The maximum population density is obtained after sorting the population density in the neighborhood of all subproblems at the tth iteration.

[0120] (b) When δ j When the angle range is (130°, 140°), diversity and convergence show the same evolutionary trend. In this case, in order to maintain the optimal balance between convergence and diversity, the neighborhood size Remain unchanged.

[0121] (c) When δ j When the angle range is (140°, 180°), population convergence is better than diversity. In this case, in order to enhance diversity, increase the neighborhood size as follows:

[0122]

[0123] Step 5.4: Dynamic adjustment strategy of penalty factor. In response to the changes in the population evolution state, a dynamic adjustment scheme of the penalty factor closely related to the population evolution state is proposed. This scheme can adjust the penalty factor θ in time at each iteration to ensure a dynamic balance between diversity and convergence.

[0124] The penalty factor dynamic adjustment strategy is as follows:

[0125] (d) When δ j When the angle range is (90°, 130°), population diversity is better than convergence. In this case, in order to improve the convergence speed, the penalty factor at the tth iteration is Reduce as follows:

[0126]

[0127] in, are the penalty factors for the j-th subproblem at the t-th and t-1-th iterations respectively. The population fitness difference at the t-1th iteration is calculated using Eq.

[0128] (e) When δ j When the angle range is (130°, 140°), diversity and convergence show the same evolutionary trend. In this case, in order to maintain the optimal balance between convergence and diversity, the penalty factor at the tth iteration is Remain unchanged.

[0129] (f) When δ j When the angle range is (140°, 180°), the population convergence is better than the diversity. In this case, in order to enhance the diversity, the penalty factor at the tth iteration is Add as follows:

[0130]

[0131] The above adjustments can well implement the dynamic adjustment strategy of the penalty factor. Timely adjustment of the penalty factor according to the evolutionary state is conducive to maintaining the dynamic balance of the convergence and diversity of the population, and is conducive to improving the overall performance of the MOEA / D algorithm. By improving the original MOEA / D, the improved dynamic neighborhood MOEA / D algorithm based on density estimation is used to optimize the sewage multi-objective problem, achieving the purpose of the best balance between the total energy consumption EC of effluent and the effluent water quality EQ.

[0132] Step 6: Design and decide the optimal solution for each group of solutions by using the fuzzy membership function method, and obtain the nitrate nitrogen S NO,2 and dissolved oxygen S O,5 The optimal setting value of concentration achieves the best balance between the total energy consumption EC of effluent and the effluent water quality EQ;

[0133] By adopting the fuzzy membership function method, a solution that can determine the optimal solution is designed, as follows:

[0134] The jth objective function f j The Pareto solution x s The satisfaction degree is represented by the membership function, which is defined as follows:

[0135]

[0136] x s The normalized membership function τ s Calculated by the following formula:

[0137]

[0138] Among them, |Num| is the number of POS solutions obtained by the algorithm. In this study, we tend to choose the solution with the largest τ * The Pareto optimal solution is the solution.

[0139] Step 7: The optimized solutions of nitrate nitrogen concentration and dissolved oxygen concentration are used as tracking set values ​​of the fuzzy logic controller, and the fuzzy logic controller is used to output the real-time control values ​​of the dissolved oxygen conversion coefficient and the internal recirculation flow rate to control the nitrate nitrogen concentration and dissolved oxygen concentration in the sewage treatment process;

[0140] A fuzzy logic controller is used to control S O,5 and S NO,2 Best optimization settings:

[0141] △u(t)=-K △u Fuzzy(K e e,K △e △e)

[0142] Among them, K e , K Δe and K Δu are the quantization factors of error, error change and Δu respectively. In addition, the input and output are defined as 7 fuzzy subsets, which are defined as follows:

[0143]

[0144] This control uses Gaussian membership function to realize the transformation from clear and precise value to fuzzy language, and assumes that the values ​​of the domain E, ΔE and ΔU are [-6,6].

[0145] Next, it is necessary to design rules to further process these memberships. According to the experience of sewage treatment control experts, the fuzzy rules are established based on the principle of large deviations accelerating the approach to the set value and small deviations ensuring control accuracy.

[0146] Table 1 Fuzzy rules

[0147]

[0148] According to the fuzzy rule operation and fuzzy reasoning, ΔU is obtained. Then, the control quantity is defuzzified, and ΔU*K Δu Make the value of Δu precise.

[0149] Compared with PID controller, the strategy combining optimization algorithm with fuzzy control can make the water quality concentration of the discharged water decrease overall, especially S NH 、N tot Table 2 and Table 3 show the control effect changes of various parameters under PID control and the improved MOEA / D algorithm in a sunny environment. O,5 and S NO,2 The concentrations were set at 2 mg / L and 1 mg / L respectively.

[0150] Table 2 Comparison of two control effects on the target under sunny conditions

[0151]

[0152] Table 3 Comparison of different control effects on average effluent water quality under sunny conditions

[0153]

[0154] From Table 2 and Table 3, it can be seen that the improved MOEA / D algorithm combined with the fuzzy control strategy reduces the treatment energy consumption, and the effluent water quality meets the discharge standard. In addition, compared with the effect without control, the effluent S NH 、N tot The removal rates of BOD5, COD and TSS were 87.98%, 66.05%, 94.71%, 68.86% and 93.13% respectively. Figure 3 and Figure 4 It can be seen that S O,5 and S NO,2 The set value can be adjusted according to the actual situation, indicating that the improved algorithm has the advantage of real-time optimization. Therefore, the present invention is an efficient and practical intelligent optimization control method for sewage treatment process, and the results prove the effectiveness of the method.

[0155] Step 8: Use the existing characteristic data and monitoring indicators, combined with the characteristics of the new treatment process, to adjust the treatment control cycle. By monitoring the COD value of the effluent, the stability problem of the treatment system can be discovered in time, and the treatment control cycle can be adjusted according to the change of the COD value to ensure the stability of the effluent water quality.

[0156] The specific algorithm process is as follows:

[0157] Step 8.1: Take 2 days as a unit, collect COD value data for each optimization cycle in 2 days, establish the trend of COD value changes, and establish the trend of COD value changes over time based on the monitored COD value data. The initial optimization cycle time is set to optimize once every 2 hours, and the moving average window size is set to 2 hours. Use the exponentially weighted moving average method to reveal the trend characteristics of the time series and calculate the trend of COD value changes within 2 days. The calculation method of the exponentially weighted moving average method is as follows:

[0158]

[0159]

[0160] Where t is the number of moving average window sizes within a unit, 0<α<1 is the smoothing factor, and (1-α) i The weight increases exponentially, and the closer the period is to the prediction time, the greater the weight.

[0161] By comparing the exponentially weighted moving average with the COD value, we can observe the trend of COD value changes. By observing the changes in the moving average, we can also more clearly understand the overall trend of COD value changes. COD Specific mathematical description of the COD value trend:

[0162] V COD =x i -EMA

[0163] where x i It is the COD value after the last optimization in the test unit.

[0164] When V COD When <-5, it means that the recent value is lower than the average level in the past period of time. At this time, the COD value shows a downward trend and the COD value may continue to decline.

[0165] When -5 <V COD When <5, it means that the recent numerical changes are relatively stable and fluctuate within the average level. At this time, the COD value will continue to fluctuate within a stable range.

[0166] When V COD When >5, it means that the recent value is higher than the average level in the past period of time. At this time, the COD value shows an upward trend and the COD value may continue to rise.

[0167] Step 8.2: Set a reasonable COD value threshold according to the actual situation and emission standards. Generally speaking, different treatment processes and effluent water quality standards will put forward different requirements for the stability threshold range of COD. For domestic sewage treatment, according to the "Pollutant Discharge Standard for Urban Sewage Treatment Plants" (GB 18918-2002), the COD emission concentration limit of the first-level A standard is 50mg / L, and the COD emission concentration limit of the first-level B standard is 100mg / L. Therefore, for domestic sewage treatment systems, the stability threshold range of COD may be set between 40mg / L and 60mg / L. If the COD value exceeds the set threshold, it is necessary to consider adjusting the treatment control cycle.

[0168] Step 8.3: According to the trend of COD value changes and the situation of exceeding the threshold, formulate the corresponding cycle adjustment strategy. Set the upper and lower limits of the cycle adjustment time to 0.5 hours and 4 hours. When the upper limit is exceeded, the cycle optimization time is automatically adjusted to the upper limit. When it is lower than the lower limit, the cycle optimization time is automatically adjusted to the lower limit.

[0169] If the COD value continues to rise or exceeds the set threshold, this may indicate that the processing capacity of the treatment system has reached its limit or the treatment system has encountered an abnormal situation. In this case, you can consider shortening the treatment control cycle, and the optimization cycle in the next 2 days will be shortened by half an hour based on the existing optimization time t.

[0170] If the COD value is stable or within the threshold range, this may indicate that the treatment capacity of the treatment system matches the sewage quality. In this case, the existing optimization cycle duration is maintained unchanged.

[0171] If the COD value continues to decrease or stabilizes at a low level, this may indicate that the treatment system has a certain amount of excess processing capacity or that the treatment system is operating stably. In this case, you can consider extending the treatment control cycle, and then increase the optimization cycle within 2 days by half an hour based on the existing optimization time t.

[0172] Step 8.4: According to the established adjustment strategy, adjust the treatment control cycle accordingly. When the treatment control cycle is shortened due to abnormal conditions, the aeration time, aeration intensity, and sludge return ratio can be increased to increase the treatment capacity of the treatment system and ensure that the COD value can be stabilized within a reasonable range. When the system runs stably and the treatment control cycle is extended, the aeration time, aeration intensity, and sludge return ratio can be reduced to save energy and chemical use, and ensure that the treatment system can still maintain a good treatment effect.

[0173] After adjusting the treatment control cycle, it is necessary to continuously monitor the changes in COD values ​​to ensure that the treatment system can respond to changes in COD values ​​in a timely manner. And continuously optimize and adjust the strategy according to actual conditions to ensure that the effluent water quality is stable within a reasonable range.

[0174] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A collaborative optimization control method for sewage treatment process based on knowledge transfer constrained MOEA / D, characterized in that: The steps include: Step 1: Collect real-time data during sewage treatment; The real-time data includes nitrate nitrogen concentration S NO,2 , dissolved oxygen concentration S O,5 , dissolved oxygen conversion coefficient K La 、Internal reflux flow Q a , total effluent energy consumption EC and effluent water quality EQ, wherein the total effluent energy consumption EC includes pumping energy consumption and aeration energy consumption; Step 2: Use TS fuzzy neural network to establish the effluent energy consumption and effluent water quality soft measurement model for the real-time collected sewage treatment parameters, and use it as the optimization objective function to predict the system energy consumption and effluent water quality; Step 3: Processing effluent quality constraints through adaptive constrained multi-objective evolutionary method; Step 4: Use the knowledge transfer strategy to autonomously generate the initial population for this optimization process; Step 5: The dynamic neighborhood MOEA / D algorithm based on density estimation is used to optimize the energy consumption and effluent water quality objective functions, obtain the Pareto solution set, and optimize the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration; The dynamic neighborhood MOEA / D algorithm for density estimation improves the original MOEA / D by utilizing the angle evolution state judgment mechanism, distance density estimation strategy, dynamic neighborhood strategy and penalty factor dynamic adjustment strategy; Step 6: Use the fuzzy membership function method to design an intelligent decision-making scheme, use the intelligent decision-making scheme to determine the optimal solution for each group of solutions, and obtain the optimal solution for nitrate nitrogen and dissolved oxygen concentrations to achieve the best balance between total effluent energy consumption and effluent water quality; Step 7: Using the optimized solutions of nitrate nitrogen and dissolved oxygen concentrations as tracking set values ​​of the fuzzy logic controller, and using the fuzzy logic controller to output the real-time control values ​​of the dissolved oxygen conversion coefficient and the internal recirculation flow rate to control the nitrate nitrogen concentration and dissolved oxygen concentration in the sewage treatment process; Step 8: According to the monitoring index data of the existing treatment process, set the monitoring index threshold and detect the COD concentration of the water quality index. When the COD value of the monitoring index data of the new treatment process continues to rise or exceeds the threshold, the treatment control cycle needs to be shortened; if the COD value continues to decline or stabilizes at a low level, consider extending the treatment control cycle; In step 5, the dynamic neighborhood MOEA / D algorithm based on density estimation optimizes the energy consumption and effluent water quality objective functions, obtains the Pareto solution set, and optimizes the set values ​​of dissolved oxygen concentration and nitrate nitrogen concentration. The specific steps of algorithm improvement are as follows: Step 5.1: The angle-based evolutionary state judgment mechanism uses the vector between the new solution and the old solution of the j-th problem and the weight vector λ j The angle δ between i To judge the evolutionary state of population iteration and use it as the basis for subsequent adjustments; Step 5.2: Calculate population density using a distance-based density estimation scheme and adjust neighborhood size; Step 5.3: Dynamic neighborhood strategy, after determining the upward or downward trend of the neighborhood size according to the angle-based evolutionary state judgment mechanism in step 5.1, the neighborhood size is further updated according to the calculated neighborhood solution density; Step 5.4: The penalty factor dynamic adjustment strategy is to propose a dynamic adjustment scheme of the penalty factor related to the population evolution state in response to the change of the population evolution state. This scheme adjusts the penalty factor θ at each iteration to ensure a dynamic balance between diversity and convergence; The penalty factor dynamic adjustment strategy is as follows: (a) When δ j When the angle range is (90°, 130°), the penalty factor at the tth iteration is Reduce as follows: in, are the penalty factors of the j-th subproblem at the t-th and t-1-th iterations, respectively. is the population fitness difference at the t-1th iteration; is the population density in the neighborhood of the j-th subproblem at the t-1th iteration; (b) When δ j When the angle range is (130°, 140°), the penalty factor at the tth iteration remain unchanged; (c) When δ j When the angle range is (140°, 180°), the penalty factor at the tth iteration is Add as follows:

2. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: The specific steps in step 2 are as follows: Step 2.1: normalize the acquired modeling data to obtain normalized data; Step 2.2: Divide the normalized data into training set and test set; Step 2.3: Construct a TS fuzzy neural network model, use the training set to train the random weight neural network, obtain the trained random weight neural network model as the water discharge index soft measurement model, and use the test set to test the water discharge index soft measurement model; The TS fuzzy neural network model is used to construct a five-layer FNN including input layer, membership function layer, rule layer, normalization layer and output layer.

3. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: In step 3, the effluent water quality constraint is processed by the adaptive constrained multi-objective evolutionary method, specifically: determining the optimization target and constraint conditions of sewage treatment, the optimization target of sewage treatment is effluent water quality and sewage treatment energy consumption, and performing constraint processing of effluent water quality; The penalty function method is used to transform the constrained optimization model into an unconstrained optimization model; The constraint penalty term is defined as: f penalty (x)=max{g1(x)-4,0}+max{g2(x)-18,0} Among them, g1(x) is the calculated value of ammonia nitrogen in the effluent water quality, and g2(x) is the calculated value of total nitrogen in the effluent water quality; The penalty mechanism is used to limit the part of the effluent mass concentration that exceeds the limit, so that each average exceeding the limit part is multiplied by the penalty factor. Therefore, if a part exceeds the limit more seriously, the corresponding optimization objective function value will be larger, and it will be eliminated in the optimization process; the multi-objective model of the sewage treatment process after sorting is as follows: min F(x)=(f EQ (x)+ε·f penalty (x),f EC (x)+e·f penalty (x)) T Among them, ε is the penalty factor, f EQ (x) is the energy consumption soft sensor optimization model, f EC (x) is the water quality soft sensor optimization model.

4. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: The specific steps in step 4 are as follows: Step 4.1: Obtain the corresponding environmental monitoring quantity under the actual environment and detect the actual water quality index S O,5 and S NO,2 ; Collect the Pareto solutions with the best performance in the multi-objective optimization problem obtained in the previous optimization process; Use random initialization population in the first optimization; Step 4.2: Calculate the similarity s between the water quality index under actual environment and the Pareto solution of the optimal performance solution set i : In the formula, x k is the corresponding environmental monitoring quantity in the actual environment, x i is the solution to the ith subproblem of the Pareto solution in the multi-objective optimization problem, similarity(x i ,x k ) is the similarity distance calculation between the Pareto solution and the surrounding solution set; Step 4.3: Based on the similarity distribution of water quality indicators and Pareto solutions in the actual environment, knowledge transfer is performed according to the similarity indicators between each pair of solutions in the similarity matrix; Set the similarity threshold to s = 0.

5. If s is satisfied i ≥s, then the i-th Pareto solution is similar to the working condition of the actual working environment; Otherwise, the i-th Pareto solution is not a similar condition to the actual working condition environment and does not have the same knowledge information; only when the similarity between two solutions is greater than the threshold, they are considered similar; Step 4.4: Select similar solutions, traverse the Pareto solution set, and select solutions with similarities higher than the threshold as part of the initial population based on the calculated similarity; when more than half of the solutions in the Pareto solution set have similarities higher than the threshold, select the top 50 solutions with higher similarity based on the ranking of similarity; based on the similarity matrix and the set similarity threshold, screen out Pareto solutions with higher relevance to the current problem; only solutions with similarities higher than the set threshold will be selected; Step 4.5: Use the Pareto solution to generate the initial population for this optimization process; by taking the Pareto solution as the seed solution and then selecting these seed solutions according to the knowledge transfer strategy, new individuals are generated as the initial population.

5. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: In step 6, the fuzzy membership function method is used to design an intelligent decision-making solution as follows: The jth objective function f j The Pareto solution x s The satisfaction degree is represented by the membership function, which is defined as follows: x s The normalized membership function τ s Calculated by the following formula: Among them, |Num| is the number of POS solutions obtained by the algorithm.

6. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: The step 7 comprises: A fuzzy logic controller is used to control S O,5 and S NO,2 Best optimization settings: △u(t)=-K △u Fuzzy(K e e,K △e △e) Among them, K e , K Δe and K Δu are the quantization factors of error, error change and Δu respectively; in addition, the input and output are defined as 7 fuzzy subsets, which are defined as follows: The control uses Gaussian membership function to achieve the transformation from clear and precise value to fuzzy language.

7. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: The specific steps in step 8 are as follows: Step 8.1: According to the monitored COD value data, establish the change trend of COD value over time, use the exponential weighted moving average method to reveal the trend characteristics of the time series, and calculate the change trend of COD value. The specific calculation method is as follows: Where t is the number of moving average window sizes within a unit, 0<α<1 is the smoothing factor, and (1-α) i The weight increases exponentially, and the closer the period is to the prediction time, the greater the weight; Comparing the exponentially weighted moving average with the COD value, the COD value trend can be observed. COD Specific mathematical description of the COD value trend: V COD =x t -EMA where x t COD value after the last optimization in the test unit; Step 8.2: Set a reasonable COD value threshold according to the actual situation and emission standards; Step 8.3: Develop corresponding cycle adjustment strategies based on the changing trend of COD values ​​and situations where the threshold is exceeded; Step 8.4: According to the formulated adjustment strategy, make corresponding adjustments to the treatment control cycle. After adjusting the treatment control cycle, continuously monitor the changes in COD values, and continuously optimize the adjustment strategy according to actual conditions to ensure that the effluent water quality is stable within a reasonable range.

8. The collaborative optimization control method for sewage treatment process based on knowledge transfer constraint MOEA / D according to claim 1 is characterized in that: The angle-based evolution status judgment mechanism considers the following three situations: (a) If δ i When the range is (90°, 130°), reduce the neighborhood size; (b) If δ i When the range is (130°, 140°), the neighborhood size remains unchanged; (c) If δ i When the range is (140°, 180°), increase the neighborhood size.

Images