Soft measurement and multi-objective optimization control method, device and medium for sewage treatment process
By using regularized online sequential limit learning machine and volume Kalman filtering algorithm for soft measurement in sewage treatment, combined with multi-objective adaptive guidance differential evolution and fractional controller, the costly and single-objective optimization of traditional sewage treatment methods is solved. The difficult-to-balance problem has achieved high-precision sewage treatment and energy consumption optimization.
Patent Information
- Application Number
- CN202410727975.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-06
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2044-06-06
AI Technical Summary
Traditional sewage treatment methods are expensive and difficult to maintain, and single-target optimization is difficult to find the best balance between water quality and energy consumption, resulting in unsatisfactory sewage treatment results.
Soft measurement is performed based on regularized online sequential extreme learning machine (ReOSELM) and volumetric Kalman filtering (CKF) algorithm, and multi-objective optimization control is performed by combining multi-objective adaptive boot differential evolution (MOAGDE) algorithm and fractional order controller (FOPID) , realize high-precision prediction and optimization of effluent water quality and energy consumption.
The prediction accuracy and control response speed of the sewage treatment system are improved, the optimal balance between water quality and energy consumption is achieved, and the energy consumption and pollution of the system are reduced.
Smart Images
Figure CN118625667B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the research field of sewage treatment process optimization control, and specifically relates to a sewage treatment process soft measurement and multi-objective optimization control method, device and medium. Background Art
[0002] Wastewater treatment is an important project to protect the environment and human health. Its purpose is to remove or transform harmful substances in sewage through physical, chemical and biological methods to meet discharge standards while reducing pollution to the natural environment. In the sewage treatment process, monitoring and controlling key parameters are the key to ensuring treatment results.
[0003] However, traditional measurement methods usually require a large number of sensors and instruments, which are costly and difficult to maintain, and are prone to failure during long-term operation. In contrast, soft measurement technology uses mathematical models and calculation methods to estimate and predict parameters that cannot be directly measured using a small amount of sensor data, thereby achieving real-time monitoring of the sewage treatment process. Considering that the sewage treatment system is a typical nonlinear system with characteristics such as large lag and strong coupling, it is difficult for traditional soft measurement models to accurately describe its complex dynamic behavior; the soft measurement model based on neural network technology has a strong nonlinear modeling capability, which can better capture the nonlinear characteristics and time-varying nature of the sewage treatment system, and can learn and extract hidden information in the data, significantly improving the accuracy and stability of the soft model.
[0004] The sewage treatment process often encounters problems such as excessive energy consumption and substandard effluent quality. Traditional control methods usually only consider a single optimization target, such as water quality or energy consumption, while ignoring the mutual influence and trade-offs between different targets. However, in the actual sewage treatment process, indicators such as water quality and energy consumption are often conflicting, and single-target optimization often fails to achieve the desired effect. Therefore, a multi-objective optimization control scheme is crucial to improving the overall performance of the sewage treatment system. It can comprehensively consider the relationship between multiple targets and find the optimal balance between them to achieve efficient operation of the sewage treatment system. Summary of the invention
[0005] Purpose of the invention: In view of the problems pointed out in the background technology, the present invention discloses a soft measurement and multi-objective optimization control method, device and medium for sewage treatment process, which can realize online, high-precision soft measurement and multi-objective optimization control of key parameters of sewage treatment process.
[0006] Technical solution: The soft measurement and multi-objective optimization control method for a sewage treatment process described in the present invention comprises the following steps:
[0007] (1) Collect data on dissolved oxygen, nitrate nitrogen, effluent quality and energy consumption during sewage treatment;
[0008] (2) A sewage soft-sensing model based on the regularized online sequential extreme learning machine (ReOSELM) was established, and the cubature Kalman filter (CKF) algorithm was used to update the hidden layer output weights of ReOSELM to achieve high-precision prediction of effluent water quality and energy consumption.
[0009] (3) Establish a multi-objective adaptive guided differential evolution (MOAGDE) algorithm, use piecewise linear chaotic mapping (PWLCM) initialization and multivariate learning strategy to improve the MOAGDE algorithm, and obtain the IMOAGDE algorithm, which improves the overall performance of the algorithm and the diversity of the obtained Pareto solutions;
[0010] (4) Taking water quality and energy consumption as optimization targets, the optimal set values of dissolved oxygen and nitrate nitrogen are obtained through the iteration of the IMOAGDE algorithm;
[0011] (5) The integral time absolute error, adjustment time and overshoot are used as optimization objectives, and the IMOAGDE algorithm is used to tune the parameters of the fractional-order controller FOPID to find the optimal parameter combination, and the dissolved oxygen and nitrate nitrogen are tracked and controlled through FOPID.
[0012] Furthermore, the implementation process of step (2) is as follows:
[0013] (21) Define the data sample as Q(x,e), the number of hidden layer nodes N and the activation function f(x), and randomly generate the input weights And bias υ, calculate the initial hidden layer output weight φ0 and hidden layer output matrix A0; the specific process is as follows:
[0014]
[0015] Among them, E0 is the target matrix; U0 is the set matrix;
[0016] (22) A regularization parameter is introduced between the output weight norm optimization and the training error to prevent the model from overfitting; the formula is as follows:
[0017]
[0018] Where ρ represents the regularization parameter; I is the identity matrix;
[0019] (23) Use the CKF algorithm to update the hidden layer output weights of ReOSELM;
[0020] (24) Through the online learning capability of the ReOSELM model, online real-time prediction of effluent quality and energy consumption in the sewage treatment process can be achieved.
[0021] Furthermore, the implementation process of step (23) is as follows:
[0022] (231) Define the hidden layer output weight as the state variable to be estimated, and the output of the ReOSELM model as the observation;
[0023] (232) Initialize the initial state estimate and error covariance matrix of the CKF algorithm; when (k+1) samples arrive, calculate the hidden layer output weight φ k+1 And the output layer node outputs Y k+1 :
[0024]
[0025] Among them, a k and b k are the process noise covariance matrix and the observation noise covariance matrix respectively;
[0026] (233) According to the actual observation data, the CKF algorithm is used to update the hidden layer output weights of the ReOSELM model; in each sample training, the volumetric Kalman gain is calculated and used to update the state estimation.
[0027] Furthermore, the implementation process of step (233) is as follows:
[0028] Calculate the volumetric Kalman gain V k+1 :
[0029]
[0030] in, is the cross-covariance, representing the correlation between the state estimation error and the observation error; is the innovation covariance, which is used to evaluate the consistency between the observation and the state estimate;
[0031] Update the hidden layer output weights:
[0032] φ k+1 =φ k +V k+1 ·z' k+1 (6)
[0033] Among them, z' k+1 is the difference between the observed value and its estimated value; k+1 Represents the estimated value of the hidden layer output weight corresponding to the k+1th sample;
[0034] Update the error covariance matrix estimate P k+1 :
[0035]
[0036] Furthermore, the implementation process of step (3) is as follows:
[0037] (31) Improvements were made to the population initialization part of the MOAGDE algorithm, using piecewise linear chaotic mapping (PWLCM) to initialize the population of the algorithm; by generating a random number sequence with good uniform distribution characteristics, it helps to improve the uniformity of population distribution and the global search ability of the algorithm; the specific improvements are as follows:
[0038]
[0039] Among them, the segment control factor η∈(0,0.5); X i+1 ∈(0,1),X i is the random iteration value generated;
[0040] (32) A multivariate learning strategy is used to replace the mutation operation of the algorithm itself, and the search is guided by the best individual in the population to improve the diversity of the population. The population is randomly divided into two parts, one part of the individuals learns from the current population, and the other individuals learn from the best individual in the current population, as shown in the following formula:
[0041]
[0042] Where h is a random individual in the population and h≠i; X h,j represents the value of the jth variable of individual h in the population; X best,j represents the value of the jth variable of the best individual in the population; F is the mutation rate; e1 and e2 are random numbers from 0 to 1; M i,j Represents the value of an individual obtained through a multivariate learning strategy.
[0043] Furthermore, the implementation process of step (4) is as follows:
[0044] (41) The soft sensing model of effluent water quality and energy consumption is used as the optimization objective function, and the specific expression is as follows:
[0045] min R OCI (τ),R EQI (τ) (10)
[0046] Among them, R OCI (τ) and R EQI (τ) are the output values of the soft sensor model for energy consumption and effluent water quality, respectively; τ is the vector composed of dissolved oxygen and nitrate nitrogen;
[0047] (42) Set the population size NP, mutation rate F, and maximum number of iterations G of the IMOAGDE algorithm max and the upper and lower limits of the search space; initialize the algorithm's population and execute the multivariate learning strategy, perform the crossover operation, cross-recombine the generated individuals and the target individuals, and generate the test individuals; the process is as follows:
[0048]
[0049] in, represents the experimental individual; G is the current iteration number; Represents the target individual; is the crossover rate, which is adaptively adjusted by formula (11); g1 and g2 are the selection probabilities, g1+g2=1; j rand is the random dimension of the variable;
[0050] (43) Calculate the fitness value of the individual and compare the quality of the target individual with the test individual. The better individual is retained as the target individual of the next generation:
[0051]
[0052] Among them, R(·) represents the fitness value; Represents the value of the target individual of the next generation;
[0053] (44) If the current number of iterations G is less than the maximum number of iterations G max , then let G = G + 1, update the Pareto solution set, and return to step (42) to continue iteration, otherwise end the iteration and output the Pareto solution set;
[0054] (45) A set of better solutions are selected from the Pareto solution set as the set values of dissolved oxygen and nitrate nitrogen in the FOPID controller.
[0055] Furthermore, the implementation process of step (5) is as follows:
[0056] (51) Initialize the parameters of the IMOAGDE algorithm, determine the parameters of FOPID: proportional gain, integral gain, differential gain, integral order and differential order, and set the upper and lower limits of the parameters;
[0057] (52) The three optimization objectives of integral time absolute error ITAE, adjustment time T and overshoot σ are used as the fitness function of the IMOAGDE algorithm;
[0058] (53) The individual optimization and iteration method of the IMOAGDE algorithm is used to update the population and evaluate the fitness function to obtain the optimal parameters of the current FOPID;
[0059] (54) Repeat step (53) to gradually adjust the controller parameters until the algorithm reaches the maximum number of iterations, output the parameter combination that optimizes the system performance, and achieve parameter tuning of FOPID;
[0060] (55) The dissolved oxygen conversion coefficient and the internal recirculation rate are adjusted by the FOPID controller to achieve tracking control of the dissolved oxygen concentration and the nitrate nitrogen concentration.
[0061] A device according to the present invention includes a memory and a processor, wherein:
[0062] A memory for storing computer programs that can be run on the processor;
[0063] The processor is used to execute the steps of the soft measurement and multi-objective optimization control method for the sewage treatment process as described above when running the computer program.
[0064] A storage medium described in the present invention stores a computer program, and when the computer program is executed by at least one processor, the steps of the soft measurement and multi-objective optimization control method for the sewage treatment process as described above are implemented.
[0065] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:
[0066] 1. The present invention adopts the ReOSELM model to predict the effluent quality and energy consumption of the sewage treatment process. As an improved model of the online sequential extreme learning machine (OSELM) model, it not only has the online learning ability to process a large amount of data in real time, but also the added regularization term further improves the generalization ability of the model; at the same time, the CKF algorithm is used to update the hidden layer output weights of ReOSELM, which can effectively track the dynamic changes of the sewage treatment system and improve the prediction accuracy of the soft sensor model;
[0067] 2. The present invention proposes a multi-objective adaptive guided differential evolution (IMOAGDE) algorithm integrating multiple improvement strategies. The algorithm's population is initialized by piecewise linear chaotic mapping (PWLCM), so that it can effectively cover the search space, help the algorithm find a better solution faster, and accelerate the convergence speed; a multivariate learning strategy is used to replace the mutation operation of the algorithm itself, and the search is guided by the optimal individual in the population to improve the comprehensive performance of the algorithm and the diversity of the obtained Pareto solution, so as to more effectively find the optimal setting values of effluent water quality and energy consumption;
[0068] 3. The present invention uses the IMOAGDE algorithm for FOPID parameter setting. By adjusting the parameters of the FOPID controller, the optimal parameter combination is found, thereby improving the response speed, stability and robustness of the system; the FOPID controller is used to accurately track and control key water quality parameters in the sewage treatment process such as dissolved oxygen and nitrate nitrogen. Compared with traditional PID control, FOPID can better adapt to the nonlinear, time-varying and complex characteristics in sewage treatment, improve the dynamic response and anti-interference ability of the control system, and achieve more effective sewage treatment optimization control. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] Figure 1 Soft measurement and multi-objective optimization control flow chart for sewage treatment process;
[0070] Figure 2 It is a schematic diagram of the ReOSELM model structure;
[0071] Figure 3 Flowchart of the IMOAGDE algorithm. DETAILED DESCRIPTION
[0072] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0073] The present invention proposes a soft measurement and multi-objective optimization control method for sewage treatment process. Figure 1 As shown, the specific steps include:
[0074] Step 1: Collect data on dissolved oxygen, nitrate nitrogen, effluent quality and energy consumption during the sewage treatment process.
[0075] The sewage treatment process consists of three parts: an anaerobic tank, an aerobic tank and a secondary sedimentation tank; the anaerobic tank consists of unit 1 and unit 2; the aerobic tank consists of unit 3, unit 4 and unit 5; by placing corresponding sensors and instruments, real-time data such as dissolved oxygen, nitrate nitrogen, effluent water quality and energy consumption in the sewage treatment process are collected; and pre-processing operations such as data cleaning, interpolation and removal of outliers are performed on the data to improve data quality.
[0076] Step 2: Establish a sewage soft sensor model based on Regularized Online Sequential Extreme Learning Machine (ReOSELM), such as Figure 2 As shown in the figure, the CKF (Cubic Kalman Filter) algorithm is used to update the hidden layer output weights of ReOSELM to achieve high-precision prediction of effluent water quality and energy consumption.
[0077] Step 2.1: Define the data sample collected in step 1 as Q(x,e), the number of hidden layer nodes N and the activation function f(x), and randomly generate input weights And bias υ, calculate the initial hidden layer output weight φ0 and hidden layer output matrix A0; the specific process is as follows:
[0078]
[0079] Among them, E0 is the target matrix; U0 is the set matrix.
[0080] Step 2.2: A regularization parameter is introduced between the output weight norm optimization and the training error to prevent the model from overfitting; the formula is as follows:
[0081]
[0082] Where ρ represents the regularization parameter and I is the identity matrix.
[0083] Step 2.3: Use the CKF algorithm to update the hidden layer output weights of ReOSELM;
[0084] The hidden layer output weights are defined as the state variables to be estimated, and the output of the ReOSELM model is taken as the observed value.
[0085] Initialize the initial state estimate and error covariance matrix of the CKF algorithm; when k+1 samples arrive, calculate the hidden layer output weight φ k+1 And the output layer node outputs Y k+1 ;
[0086]
[0087] Among them, a k and b k are the process noise covariance matrix and the observation noise covariance matrix, respectively.
[0088] According to the actual observation data, the CKF algorithm is used to update the hidden layer output weights of the ReOSELM model. In each sample training, the volumetric Kalman gain is calculated and used to update the state estimation.
[0089] 1) Calculate the volumetric Kalman gain V k+1 :
[0090]
[0091] in, is the cross-covariance, representing the correlation between the state estimation error and the observation error; is the innovation covariance, which is used to evaluate the consistency between the observation and the state estimate;
[0092] 2) Update the hidden layer output weights:
[0093] φ k+1 =φ k +V k+1 ·z' k+1 (6)
[0094] Among them, z' k+1 is the difference between the observed value and its estimated value; k+1 Represents the estimated value of the hidden layer output weight corresponding to the k+1th sample;
[0095] 3) Update the error covariance matrix estimate P k+1 :
[0096]
[0097] Step 2.4: Through the online learning capability of the ReOSELM model, online real-time prediction of effluent quality and energy consumption of the sewage treatment process is achieved.
[0098] Step 3: Establish a multi-objective adaptive guided differential evolution (MOAGDE) algorithm, use piecewise linear chaotic mapping (PWLCM) initialization and multi-learning strategy to improve the MOAGDE algorithm, and obtain the IMOAGDE algorithm to improve the comprehensive performance of the algorithm and the diversity of the obtained Pareto solutions.
[0099] Step 3.1: Improve the population initialization part of the MOAGDE algorithm by using piecewise linear chaotic mapping (PWLCM) to initialize the algorithm's population; by generating a random number sequence with good uniform distribution characteristics, it helps to improve the uniformity of population distribution and the global search ability of the algorithm; the specific improvements are as follows:
[0100]
[0101] Among them, the segment control factor η∈(0,0.5); X i+1 ∈(0,1),X i is the random iteration value generated.
[0102] Step 3.2: Use a multivariate learning strategy to replace the mutation operation of the algorithm itself, guide the search through the best individual in the population, and improve the diversity of the population; randomly divide the population into two parts, one part of the individuals learn from the current population, and the rest of the individuals learn from the best individual in the current population, as shown in the following formula:
[0103]
[0104] Where h is a random individual in the population and h≠i; X h,j represents the value of the jth variable of individual h in the population; X best,j represents the value of the jth variable of the best individual in the population; F is the mutation rate; e1 and e2 are random numbers from 0 to 1; M i,j Represents the value of an individual obtained through a multivariate learning strategy.
[0105] Step 4: Taking water quality and energy consumption as optimization targets, the optimal set values of dissolved oxygen and nitrate nitrogen are obtained through the iteration of the IMOAGDE algorithm, such as Figure 3 As shown, the implementation process is as follows:
[0106] Step 4.1: Use the soft sensor model of effluent water quality and energy consumption in step 2 as the optimization objective function. The specific expression is as follows:
[0107] min R OCI (τ),R EQI (τ) (10)
[0108] Among them, R OCI (τ) and R EQI (τ) are the output values of the soft sensor model for energy consumption and effluent water quality respectively; τ is the vector composed of dissolved oxygen and nitrate nitrogen.
[0109] Step 4.2: Set the population size NP, mutation rate F, and maximum number of iterations G of the IMOAGDE algorithm max and the upper and lower limits of the search space; use step 3.1 to initialize the algorithm's population; use step 3.2 to execute the multi-learning strategy.
[0110] Step 4.3: Perform a crossover operation to cross-recombine the individuals generated in step 4.2 and the target individuals to generate test individuals. The process is as follows:
[0111]
[0112] in, represents the experimental individual; G is the current iteration number; Represents the target individual; is the crossover rate, which is adaptively adjusted by formula (11); g1 and g2 are the selection probabilities, g1+g2=1; j rand is the random dimension of the variable.
[0113] Step 4.4: Perform the selection operation; calculate the fitness value of the individual, and compare the quality of the target individual and the test individual. The better individual is retained as the target individual of the next generation;
[0114]
[0115] Among them, R(·) represents the fitness value; Represents the value of the target individual of the next generation.
[0116] Step 4.5: If the current number of iterations G is less than the maximum number of iterations G max , then let G=G+1, update the Pareto solution set, and return to step 4.2 to continue the iteration, otherwise end the iteration and output the Pareto solution set.
[0117] Step 4.6: Select a set of better solutions from the Pareto solution set as the set points for dissolved oxygen and nitrate nitrogen in the FOPID controller.
[0118] Step 5: Using the integral time absolute error, adjustment time and overshoot as optimization objectives, the IMOAGDE algorithm is used for parameter tuning of the fractional order controller (FOPID) to find the optimal parameter combination, and the dissolved oxygen and nitrate nitrogen are tracked and controlled through FOPID.
[0119] Step 5.1: Initialize the parameters of the IMOAGDE algorithm, determine the parameters of FOPID: proportional gain, integral gain, differential gain, integral order and differential order, and set the upper and lower limits of the parameters;
[0120] Step 5.2: Use the three optimization objectives of integral time absolute error ITAE, adjustment time T and overshoot σ as the fitness function of the IMOAGDE algorithm; the specific expression is as follows:
[0121] min ITAE,T,σ(14)
[0122] Step 5.3: Use the individual optimization and iteration method of the IMOAGDE algorithm to update the population and evaluate the fitness function to obtain the optimal parameters of the current FOPID.
[0123] Step 5.4: Repeat step 5.3, gradually adjusting the controller parameters until the algorithm reaches the maximum number of iterations, outputting the parameter combination that optimizes the system performance, and achieving parameter tuning of FOPID.
[0124] Step 5.5: The dissolved oxygen conversion coefficient and the internal recirculation volume are adjusted by the FOPID controller to achieve tracking control of the dissolved oxygen concentration in the 5th unit and the nitrate nitrogen concentration in the 2nd unit.
[0125] The present invention also provides an apparatus including a memory and a processor, wherein the memory is used to store a computer program that can be run on the processor; the processor is used to execute the steps of the soft measurement and multi-objective optimization control method for the sewage treatment process as described above when running the computer program.
[0126] The present invention also provides a storage medium having a computer program stored thereon, and when the computer program is executed by at least one processor, the steps of the soft measurement and multi-objective optimization control method for the sewage treatment process as described above are implemented.
[0127] 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 soft measurement and multi-objective optimization control method for sewage treatment process, It is characterized in that Here are the steps: (1) Collect data on dissolved oxygen, nitrate nitrogen, effluent quality and energy consumption during sewage treatment; (2) A sewage soft-sensing model based on the regularized online sequential extreme learning machine (ReOSELM) was established, and the cubature Kalman filter (CKF) algorithm was used to update the hidden layer output weights of ReOSELM to achieve high-precision prediction of effluent water quality and energy consumption. (3) Establish a multi-objective adaptive guided differential evolution (MOAGDE) algorithm, use piecewise linear chaotic mapping (PWLCM) initialization and multivariate learning strategy to improve the MOAGDE algorithm, and obtain the IMOAGDE algorithm, which improves the overall performance of the algorithm and the diversity of the obtained Pareto solutions; (4) Taking water quality and energy consumption as optimization targets, the optimal set values of dissolved oxygen and nitrate nitrogen are obtained through the iteration of the IMOAGDE algorithm; (5) Taking the integral time absolute error, adjustment time and overshoot as optimization objectives, the IMOAGDE algorithm is used for parameter tuning of the fractional-order controller FOPID to find the optimal parameter combination, and the dissolved oxygen and nitrate nitrogen are tracked and controlled by FOPID; The implementation process of step (2) is as follows: (21) Define the data sample as Q(x,e), the number of hidden layer nodes N and the activation function f(x), and randomly generate the input weights And bias υ, calculate the initial hidden layer output weight φ 0 And the hidden layer output matrix A 0 ; The specific process is as follows: Among them, E 0 is the target matrix; U 0 is the set matrix; (22) A regularization parameter is introduced between the output weight norm optimization and the training error to prevent the model from overfitting; the formula is as follows: Where ρ represents the regularization parameter; I is the identity matrix; (23) Use the CKF algorithm to update the hidden layer output weights of ReOSELM; (24) Through the online learning capability of the ReOSELM model, online real-time prediction of effluent quality and energy consumption in the sewage treatment process is achieved; The implementation process of step (23) is as follows: (231) Define the hidden layer output weight as the state variable to be estimated, and the output of the ReOSELM model as the observation; (232) Initialize the initial state estimate and error covariance matrix of the CKF algorithm; when (k+1) samples arrive, calculate the hidden layer output weight φ k+1 And the output layer node outputs Y k+1 : Among them, a k and b k are the process noise covariance matrix and the observation noise covariance matrix respectively; (233) Based on the actual observation data, the CKF algorithm is used to update the hidden layer output weights of the ReOSELM model; in each sample training, the volumetric Kalman gain is calculated and the gain is used to update the state estimation; The implementation process of step (233) is as follows: Calculate the volumetric Kalman gain V k+1 : in, is the cross-covariance, representing the correlation between the state estimation error and the observation error; is the innovation covariance, which is used to evaluate the consistency between the observation and the state estimate; Update the hidden layer output weights: φ k+1 =φ k +V k+1 ·z′ k+1 (6) Among them, z′ k+1 is the difference between the observed value and its estimated value; φ k+1 Represents the estimated value of the hidden layer output weight corresponding to the k+1th sample; Update the error covariance matrix estimate P k+1 : The implementation process of step (3) is as follows: (31) Improvements were made to the population initialization part of the MOAGDE algorithm, using piecewise linear chaotic mapping (PWLCM) to initialize the population of the algorithm; by generating a random number sequence with good uniform distribution characteristics, it helps to improve the uniformity of population distribution and the global search ability of the algorithm; the specific improvements are as follows: Among them, the segment control factor η∈(0,0.5); X i+1 ∈(0,1),X i is the random iteration value generated; (32) A multivariate learning strategy is used to replace the mutation operation of the algorithm itself, and the search is guided by the best individual in the population to improve the diversity of the population. The population is randomly divided into two parts, one part of the individuals learns from the current population, and the other individuals learn from the best individual in the current population, as shown in the following formula: Where h is a random individual in the population and h≠i; X h,j represents the value of the jth variable of individual h in the population; X best,j represents the value of the jth variable of the best individual in the population; F is the mutation rate; e 1 and e 2 is a random number from 0 to 1; M i,j Represents the individual value obtained through the multi-learning strategy; The implementation process of step (4) is as follows: (41) The soft sensing model of effluent water quality and energy consumption is used as the optimization objective function, and the specific expression is as follows: min R OCI (t),R EQI (t)(10) Among them, R OCI (τ) and R EQI (τ) are the output values of the soft sensor model for energy consumption and effluent water quality, respectively; τ is the vector composed of dissolved oxygen and nitrate nitrogen; (42) Set the population size NP, mutation rate F, and maximum number of iterations G of the IMOAGDE algorithm max and the upper and lower limits of the search space; initialize the algorithm's population and execute the multivariate learning strategy, perform the crossover operation, cross-recombine the generated individuals and the target individuals, and generate the test individuals; the process is as follows: in, represents the experimental individual; G is the current iteration number; Represents the target individual; is the crossover rate, which is adaptively adjusted through formula (11); g 1 and g 2 is the selection probability, g 1 +g 2 =1;j rand is the random dimension of the variable; (43) Calculate the fitness value of the individual and compare the quality of the target individual with the test individual. The better individual is retained as the target individual of the next generation: Among them, R(·) represents the fitness value; Represents the value of the target individual of the next generation; (44) If the current number of iterations G is less than the maximum number of iterations G max , then let G = G + 1, update the Pareto solution set, and return to step (42) to continue iteration, otherwise end the iteration and output the Pareto solution set; (45) A set of better solutions are selected from the Pareto solution set as the set values of dissolved oxygen and nitrate nitrogen in the FOPID controller.
2. According to claim 1, a soft measurement and multi-objective optimization control method for a sewage treatment process, It is characterized in that The implementation process of step (5) is as follows: (51) Initialize the parameters of the IMOAGDE algorithm, determine the parameters of FOPID: proportional gain, integral gain, differential gain, integral order and differential order, and set the upper and lower limits of the parameters; (52) The three optimization objectives of integral time absolute error ITAE, adjustment time T and overshoot σ are used as the fitness function of the IMOAGDE algorithm; (53) The individual optimization and iteration method of the IMOAGDE algorithm is used to update the population and evaluate the fitness function to obtain the optimal parameters of the current FOPID; (54) Repeat step (53) to gradually adjust the controller parameters until the algorithm reaches the maximum number of iterations, output the parameter combination that optimizes the system performance, and achieve parameter tuning of FOPID; (55) The dissolved oxygen conversion coefficient and the internal recirculation rate are adjusted by the FOPID controller to achieve tracking control of the dissolved oxygen concentration and the nitrate nitrogen concentration.
3. A device, It is characterized in that comprising a memory and a processor, wherein: A memory for storing computer programs that can be run on the processor; A processor is used to execute the steps of the soft measurement and multi-objective optimization control method for a sewage treatment process as described in any one of claims 1 to 2 when running the computer program.
4. A storage medium, It is characterized in that The storage medium stores a computer program, which, when executed by at least one processor, implements the steps of the soft measurement and multi-objective optimization control method for a sewage treatment process as described in any one of claims 1 to 2.