Flue gas desulfurization system prediction control method and system based on self-organizing fuzzy neural network

Abstract

The invention discloses a flue gas desulfurization system prediction control method and system based on a self-organizing fuzzy neural network, and relates to the field of artificial intelligence driven complex dynamic system model representation and state control, and the method comprises the steps: proposing a control method framework based on the dynamic characteristics of a wet flue gas desulfurization system variable evolution process; constructing a prediction model based on the control method framework; and completing predictive control of the flue gas desulfurization system by using the predictive model. The invention provides a prediction model and a control framework based on a self-organizing fuzzy neural network aiming at the problems of strong nonlinearity of SO2 and pH states, non-stability in operation, difficulty in control and the like in a wet flue gas desulfurization system. The structure of the network can be dynamically adjusted, the network adapts to operation changes, and accurate modeling and prediction of SO2 and pH are achieved. And on the basis of prediction information, an optimal control law is dynamically solved in a rolling optimization mode, real-time tracking and accurate control of SO2 and pH states are achieved, and the control effect of the flue gas desulfurization process is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of complex dynamic system model characterization and state control driven by artificial intelligence, and in particular to a flue gas desulfurization system predictive control method and system based on a self-organizing fuzzy neural network. Background Art

[0002] Sulfur dioxide (SO2) emissions are the main cause of air pollution and respiratory diseases, and its concentration variation characteristics are key indicators of the gas quality of the chemical tail gas desulfurization process. Therefore, in order to improve the intelligent operation level of the chemical tail gas desulfurization process and ensure that the gas quality meets the standards, it is particularly important to predict the trend of SO2 concentration changes and control its behavior during the desulfurization process.

[0003] At the same time, the country's high standards and strict requirements for SO2 emissions from chemical processes have brought new challenges to existing complex chemical processes: (1) Due to the multivariable coupling, strong nonlinearity and complex dynamic characteristics of the wet flue gas desulfurization system, the existing prediction model cannot change its structure in real time to adapt to the dynamic characteristics, making the model predictive control scheme ineffective in the modeling and control of the wet flue gas desulfurization system. (2) In the traditional model predictive control scheme, only the discrete sampling points of the prediction model output sequence may be passed to the loss function. During the rolling optimization, especially in the process of minimizing the loss function, the information cannot be derived, making it difficult to achieve convergence.

[0004] Aiming at the problem that the key parameters of the desulfurization process of the wet flue gas desulfurization system are difficult to track in real time and accurately control, a flue gas desulfurization system predictive control method based on a self-organizing fuzzy neural network is proposed. First, the original data is obtained from the flue gas desulfurization plant to obtain the task sample set to be processed, and the auxiliary variables with a strong correlation with the SO2 concentration and pH value in the desulfurization process are found. Secondly, the structure and parameters of the fuzzy neural network are learned according to the task sample set by self-organizing learning to achieve accurate prediction of the key parameters of the desulfurization process. Finally, a loss function with the one-step output of the prediction model and the control law as independent variables is designed, and the optimal control law sequence is obtained by minimizing the loss function through rolling optimization and gradient descent method. The present invention uses self-organizing learning to learn the structure and parameters according to the task sample set, takes the fuzzy neural network as a supervised learning and prediction model, and uses the loss function containing the one-step output of the prediction model and the control law as independent variables to obtain the optimal control law sequence, thereby realizing model predictive control and improving the real-time tracking capability and predictive control accuracy of the key parameters in the desulfurization process. Summary of the invention

[0005] Aiming at the problem that the key parameters of the desulfurization process of the wet flue gas desulfurization system are difficult to track and control in real time, a flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network is proposed. The self-organizing fuzzy neural network is used as a prediction model to predict SO 2 The dynamic characteristics of the flue gas desulfurization system are used to construct the one-step output of the prediction model and the loss function of the control law as independent variables. The optimal control law sequence is obtained in the form of rolling solution through the gradient descent algorithm, thereby realizing accurate tracking and predictive control of the key parameters of the flue gas desulfurization system.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] The flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network is characterized in that the steps include:

[0008] Based on the dynamic characteristics of the evolution process of the wet flue gas desulfurization system variables, a control method framework is proposed; the steps include: analyzing the dynamic characteristics of the evolution process of the key variables of the wet flue gas desulfurization system, converting SO 2 Describe the concentration and pH control process and objectives as a closed-loop control problem, and design the control method and framework;

[0009] Based on the control method framework, a prediction model is constructed; the steps include: firstly establishing a fuzzy neural network model, including an input layer, a membership function layer, a rule layer and an output layer; based on the fuzzy neural network model, establishing a self-organizing fuzzy neural network; setting a stop criterion for adding and deleting structures for the self-organizing fuzzy neural network;

[0010] The prediction model is used to complete the prediction control of the flue gas desulfurization system; the steps include: using the one-step output of the prediction model and the current control law change as independent variables to construct a loss function, combining rolling optimization and gradient descent algorithms, and solving the optimal control law sequence by minimizing the loss function.

[0011] Preferably, the dynamic characteristics include:

[0012] ,

[0013] ,

[0014] ,

[0015] Based on the dynamic characteristics, the control method framework is designed, which includes: continuously detecting the SO at the top of the absorption zone 2 The closed-loop feedback information is obtained by controlling the concentration and pH of the slurry in the oxidation zone. in and f re To implement SO 2 Maximize absorption.

[0016] Preferably, the fuzzy neural network model includes:

[0017] ,

[0018] ,

[0019] ,

[0020] ,

[0021] In the formula, i, r, j represent natural numbers; and represents the center and width vector of the jth neuron, and , where R represents the number of neurons in the membership function layer and the rule layer; m represents the dimension of the input data; represents the output vector of the normalization layer; c ir represents the central value of the rth neuron in the regular layer under the action of the i-th input data; x i represents the i-th component in the input vector; , and They represent the center and width of the neurons in the membership function layer and the output matrix respectively; η represents the learning rate.

[0022] Preferably, the self-organizing fuzzy neural network comprises:

[0023] ,

[0024] ,

[0025] ,

[0026] In the formula, AS represents the activation sensitivity of neurons in the regular layer, and is the input and output of the i-th neuron in the regular layer; , and Respectively represent the center, width and output weight of the newly added neurons; ; Indicates the expected output value; Represents the output error; In continuous The reduction in step As a stopping criterion for structural growth.

[0027] Preferably, the loss function includes:

[0028] ,

[0029] In the formula, represents the control law; Indicates the change of control law; represents the reference input; P and Q represent constant matrices; γ represents the learning rate of the solution process; Represents the predicted output.

[0030] The present invention also provides a flue gas desulfurization system prediction control system based on a self-organizing fuzzy neural network, the system is used to implement the above method, including: a design module, a construction module and a prediction module;

[0031] The design module is used to propose a control method framework based on the dynamic characteristics of the evolution process of the wet flue gas desulfurization system variables; the steps include: analyzing the dynamic characteristics of the evolution process of the key variables of the wet flue gas desulfurization system, 2 Describe the concentration and pH control process and objectives as a closed-loop control problem, and design the control method and framework;

[0032] The construction module is used to construct a prediction model based on the control method framework; the steps include: firstly establishing a fuzzy neural network model, including an input layer, a membership function layer, a rule layer and an output layer; based on the fuzzy neural network model, establishing a self-organizing fuzzy neural network; setting a stop criterion for adding and deleting structures for the self-organizing fuzzy neural network;

[0033] The prediction module is used to complete the predictive control of the flue gas desulfurization system using the prediction model; the steps include: constructing a loss function using the one-step output of the prediction model and the current control law change as independent variables, combining rolling optimization and gradient descent algorithms, and solving the optimal control law sequence by minimizing the loss function.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] (1) The present invention is aimed at wet flue gas desulfurization system SO 2 In order to solve the problems of strong nonlinearity and non-stationary operation in the evolution of pH state, a SO 2 and pH state prediction model. The self-organizing fuzzy neural network can adapt to the non-stationary operation by dynamically adjusting the structure, thereby achieving SO 2 Accurate modeling and prediction of pH states.

[0036] (2) The present invention is aimed at wet flue gas desulfurization system SO 2 The problem of pH state behavior control being difficult to track and accurately control in real time is solved by constructing a predictive control framework based on self-organizing fuzzy neural network (such asFigure 1 Compared with the existing flue gas desulfurization process control method, the predictive control method based on the self-organizing model neural network can not only accurately predict SO 2 The dynamic evolution characteristics of the pH value can also be used to dynamically solve the optimal control law in the form of rolling optimization based on the prediction information, thereby achieving SO 2 and tracking control of pH state behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1 It is a schematic diagram of the operation principle of a wet flue gas desulfurization system of the control method framework of an embodiment of the present invention;

[0039] Figure 2 A predictive control framework for a flue gas desulfurization system based on a self-organizing fuzzy neural network according to an embodiment of the present invention;

[0040] Figure 3 The control method framework of the embodiment of the present invention is SO 2 Concentration predictions and errors;

[0041] Figure 4 The prediction result and error of pH value of the control method framework of the embodiment of the present invention;

[0042] Figure 5 The control method framework of the embodiment of the present invention is SO 2 Concentration tracking results and errors;

[0043] Figure 6 The tracking result and error of pH value of the control method framework of the embodiment of the present invention;

[0044] Figure 7 The control method framework of the embodiment of the present invention is SO 2 Control variable change curve of concentration;

[0045] Figure 8 This is a control variable variation curve of pH value in the control method framework of an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Embodiment 1

[0049] This embodiment provides a flue gas desulfurization system predictive control method based on a self-organizing fuzzy neural network, the steps comprising:

[0050] S1. Based on the dynamic characteristics of the variable evolution process of the wet flue gas desulfurization system, a control method framework is proposed.

[0051] The dynamic characteristics of the evolution process of key variables in the wet flue gas desulfurization system are analyzed, the control process and target of SO2 concentration and pH value are described as a closed-loop control problem, and the proposed control method and framework are given. Specifically, the wet flue gas desulfurization process is a complex dynamic reaction system, such as Figure 2 As the flue gas to be treated enters the absorption zone of the spray tower, the dissolved and evenly stirred limestone slurry flows at a flow rate of f in From the stirring tank into the spray tower, SO 2 After being absorbed and oxidized, it turns into calcium sulfate dihydrate (gypsum). In order to make the absorption and oxidation process more thorough, the intermediate product calcium sulfite slurry in the reaction process will also flow at a flow rate of f re The circulation sprays from the top of the absorption zone downwards.

[0052] , (1)

[0053] , (2)

[0054] , (3)

[0055] From formulas (1)-(3), it can be seen that in order to completely absorb SO in flue gas, the amount of oxygen required for the complete absorption of SO in flue gas is 2 , it is only necessary to reasonably adjust and control the flow rate of the limestone slurry f in and the flow rate of calcium sulfite slurry f re , that is, the control variable is f in and f reTherefore, in the process of wet flue gas desulfurization, by continuously detecting the SO 2 The closed-loop feedback information is obtained by controlling the concentration and pH of the slurry in the oxidation zone. in and f re To implement SO 2 Maximize absorption.

[0056] S2. Build a prediction model based on the control method framework.

[0057] The original data is obtained from the actual flue gas desulfurization process and preprocessed to obtain sample data sets (training set and test set), and the SO 2 Then, based on the above control method framework, a prediction model based on self-organizing fuzzy neural network is established, and the auxiliary variables and SO are strongly correlated based on the training set. 2 The nonlinear mapping relationship between SO 2 The specific process includes:

[0058] S201. First, establish a fuzzy neural network model (FNN), which includes input layer, membership function layer, rule layer and output layer. 2 The auxiliary variables for and pH are: HSO 3 - , HSO 4 - , H + OH - , HCO 3 - 、SO 4 2- 、SO 3 2- , Ca + The first input layer is used to receive auxiliary variables. and ; Assume that the second membership function layer contains R neurons, each neuron is regarded as a radial basis function, and its state output φ j for:

[0059] , (4)

[0060] The third regular layer contains R neurons, and the output state of each neuron can be expressed as:

[0061] , (5)

[0062] Assume that the fourth output layer has P neurons and the output is ;in, , is the output matrix, and . represents the output of the pth neuron, which can be expressed as:

[0063] · , (6)

[0064] The training process of FNN is the process of continuously updating and optimizing parameters. The parameter updating formula is:

[0065] , (7)

[0066] In formulas (4) to (7), i, r, and j represent natural numbers; and represents the center and width vector of the jth neuron, and , where R represents the number of neurons in the membership function layer and the rule layer; m represents the dimension of the input data; represents the output vector of the normalization layer; c ir represents the central value of the rth neuron in the regular layer under the action of the i-th input data; x i represents the i-th component in the input vector; , and Respectively represent the center and width of the neurons in the membership function layer and the output matrix; Represents the learning rate.

[0067] S202. Establish a self-organizing fuzzy neural network (SOFNN).

[0068] (1) FNN structure growth stage.

[0069] Define variables that can reflect the real-time performance of the trained FNN model as shown in the following formula:

[0070] , (8)

[0071] Where AS represents the activation sensitivity of neurons in the regular layer, and It is the first The input and output of a neuron; when AS exceeds a certain value and the training error decreases rapidly, this neuron will split into two new neurons, and the relevant parameters of the new neurons are updated as follows:

[0072] , (9)

[0073] in , and They represent the center, width and output weight of the newly added neurons respectively.

[0074] (2) FNN structure deletion stage.

[0075] When AS does not exceed a certain value and the training error gradually decreases, this neuron needs to be deleted, and the parameters of the remaining neurons are updated according to formula (7).

[0076] S203. Set the stopping criteria for structure addition and deletion.

[0077] Calculate output error using validation data , and In continuous The reduction in step As the stopping criterion of structural growth, it is expressed as:

[0078] , (10)

[0079] in ,when When it reaches 0.01 for the first time, the process of adding and removing structures stops; Indicates the expected output value; Represents the predicted output.

[0080] S3. Use the prediction model to complete the predictive control of the flue gas desulfurization system.

[0081] The one-step output of the prediction model and the change of the current control law are used as independent variables to construct the loss function. Combining the rolling optimization and gradient descent algorithms, the optimal control law sequence is solved by minimizing the loss function, that is, a predictive control framework for the flue gas desulfurization system based on self-organizing fuzzy neural network is constructed.

[0082] In the wet flue gas desulfurization system, the SO 2 The closed-loop feedback information is obtained by controlling the concentration and pH of the slurry in the oxidation zone. in and f re To implement SO 2 Assume that , and Respectively represent SO 2 Expected values ​​of concentration and pH, and SOFNN model for SO 2 Predicted values ​​of concentration and pH, and the control variable f in and f re The loss function is defined as follows:

[0083] , (11)

[0084] , , (12)

[0085] in, and represents a constant matrix, Indicates the maximum value of the control law increase; represents the control law; Indicates the change of control law; and Represent the prediction time domain and control time domain respectively; T represents the transposed matrix. The optimal control variable is solved by the gradient descent algorithm in the process of minimizing the loss function. The formula is as follows:

[0086] , (13)

[0087] in, Represents the learning rate of the solution process. Combining formula (11) and formula (13), the optimal control variable can be expressed as:

[0088] , (14)

[0089] In the formula, Indicates the reference input.

[0090] It can be seen from formula (14) that the optimal control variable sequence can be solved in the rolling optimization and iteration process, thereby forming the SO 2 Closed-loop control framework for concentration and pH, such as Figure 2 shown.

[0091] Embodiment 2

[0092] In this embodiment, the proposed model predictive control method is tested and verified on a test set, and its modeling and predictive control performance is evaluated through different evaluation indicators.

[0093] (1) To evaluate the prediction performance of the SOFNN model, the following evaluation indicators are defined:

[0094] , (15)

[0095] in, Indicates the number of training samples; and Represent the actual output and the predicted output of SOFNN respectively; MAE represents the mean absolute error.

[0096] (2) In order to evaluate the performance of the self-organizing fuzzy neural network-based predictive control method (SOFNN-MPC), the following evaluation indicators are defined:

[0097] , (16)

[0098] , (17)

[0099] Among them, IAE and ISE represent the absolute error integral index and the square error integral index respectively.

[0100] First, 14 days of SO was obtained from an actual desulfurization plant 2 The designed SOFNN model was trained with two sets of data: concentration and its auxiliary variables, pH and its auxiliary variables. The data of the first 7 days were used as training samples, and the data of the last 7 days were used as test samples. The optimal structure scales for the model to achieve convergence were 10-10-10-1 and 10-8-8-1, respectively, and the parameters were set to , . Figure 3 and Figure 4 SO 2 Table 1 shows the prediction results and errors of the SOFNN model and other models in 20 independent repeated experiments. Figure 3 , Figure 4 As can be seen from Table 1, SOFNN is superior to other prediction models in terms of prediction accuracy, stability and computational complexity.

[0101] Table 1

[0102] .

[0103] In the predictive control process, the time domain of prediction and control is: , . Will SO 2 The desired values ​​of concentration and pH were set to 0.5 mg / L and 6.8. Figure 5 and Figure 6 SO 2 Concentration and pH values ​​track their expected values ​​and their errors, Figure 7 and Figure 8 The control variables (lime slurry input flow rate and circulating slurry flow rate) curves for achieving the above tracking results are given respectively. Table 2 shows the comparison of the results of SOFNN-MPC and other control methods in 20 independent repeated experiments. Figure 5 , Figure 6 , Figure 7 , Figure 8As can be seen from Table 2, SOFNN-MPC achieves SO 2 It can maximize the absorption and achieve relatively fast and stable control, providing theoretical and technical support for operators of flue gas desulfurization process to take reasonable control measures.

[0104] Table 2

[0105] .

[0106] Embodiment 3

[0107] This embodiment also provides a flue gas desulfurization system prediction control system based on a self-organizing fuzzy neural network, including: a design module, a construction module and a prediction module; the design module is used to propose a control method framework based on the dynamic characteristics of the evolution process of the wet flue gas desulfurization system variables; the steps include: analyzing the dynamic characteristics of the evolution process of the key variables of the wet flue gas desulfurization system, and converting SO 2 The control process and target description of concentration and pH value is described as a closed-loop control problem, and the control method and framework are designed; the construction module is used to construct a prediction model based on the control method framework; the steps include: firstly, a fuzzy neural network model is established, including an input layer, a membership function layer, a rule layer and an output layer; based on the fuzzy neural network model, a self-organizing fuzzy neural network is established; the stopping criteria for structure addition and deletion are set for the self-organizing fuzzy neural network; the prediction module is used to complete the predictive control of the flue gas desulfurization system using the prediction model; the steps include: constructing a loss function with the one-step output of the prediction model and the current control law change as independent variables, combining the rolling optimization and gradient descent algorithms, and solving the optimal control law sequence by minimizing the loss function.

[0108] The following will describe in detail how the present invention solves the technical problems in practical work in conjunction with this embodiment.

[0109] Firstly, the design module proposes a control method framework based on the dynamic characteristics of the variable evolution process of the wet flue gas desulfurization system.

[0110] The dynamic characteristics of the evolution process of key variables in the wet flue gas desulfurization system are analyzed, the control process and target of SO2 concentration and pH value are described as a closed-loop control problem, and the proposed control method and framework are given. Specifically, the wet flue gas desulfurization process is a complex dynamic reaction system, such as Figure 2 As the flue gas to be treated enters the absorption zone of the spray tower, the dissolved and evenly stirred limestone slurry flows at a flow rate of f in From the stirring tank into the spray tower, SO 2After being absorbed and oxidized, it turns into calcium sulfate dihydrate (gypsum). In order to make the absorption and oxidation process more thorough, the intermediate product calcium sulfite slurry in the reaction process will also flow at a flow rate of f re The circulation sprays from the top of the absorption zone downwards.

[0111] , (18)

[0112] , (19)

[0113] , (20)

[0114] From formulas (18)-(20), it can be seen that in order to completely absorb SO in flue gas, the 2 , it is only necessary to reasonably adjust and control the flow rate of the limestone slurry f in and the flow rate of calcium sulfite slurry f re , that is, the control variable is f in and f re Therefore, in the process of wet flue gas desulfurization, by continuously detecting the SO 2 The closed-loop feedback information is obtained by controlling the concentration and pH of the slurry in the oxidation zone. in and f re To implement SO 2 Maximize absorption.

[0115] The building module builds a prediction model based on the control method framework.

[0116] The original data is obtained from the actual flue gas desulfurization process and preprocessed to obtain sample data sets (training set and test set), and the SO 2 Then, based on the above control method framework, a prediction model based on self-organizing fuzzy neural network is established, and the auxiliary variables and SO are strongly correlated based on the training set. 2 The nonlinear mapping relationship between SO 2 The specific process includes:

[0117] S201. First, establish a fuzzy neural network model (FNN), which includes input layer, membership function layer, rule layer and output layer. 2 The auxiliary variables for and pH are: HSO 3 - , HSO 4 - , H + OH -, HCO 3 - 、SO 4 2- 、SO 3 2- , Ca + The first input layer is used to receive auxiliary variables. and ; Assume that the second membership function layer contains R neurons, each neuron is regarded as a radial basis function, and its state output φ j for:

[0118] , (twenty one)

[0119] The third regular layer contains R neurons, and the output state of each neuron can be expressed as:

[0120] , (twenty two)

[0121] Assume that the fourth output layer has P neurons and the output is ;in, , is the output matrix, and . represents the output of the pth neuron, which can be expressed as:

[0122] , (twenty three)

[0123] The training process of FNN is the process of continuously updating and optimizing parameters. The parameter updating formula is:

[0124] , (twenty four)

[0125] In formulas (21) to (24), i, r, and j represent natural numbers; and represents the center and width vector of the jth neuron, and , where R represents the number of neurons in the membership function layer and the rule layer; m represents the dimension of the input data; represents the output vector of the normalization layer; c ir represents the central value of the rth neuron in the regular layer under the action of the i-th input data; x i represents the i-th component in the input vector; , and They represent the center and width of the neurons in the membership function layer and the output matrix respectively; η represents the learning rate.

[0126] S202. Establish a self-organizing fuzzy neural network (SOFNN).

[0127] (1) FNN structure growth stage.

[0128] Define variables that can reflect the real-time performance of the trained FNN model as shown in the following formula:

[0129] , (25)

[0130] Where AS represents the activation sensitivity of neurons in the regular layer, and It is the first The input and output of a neuron; when AS exceeds a certain value and the training error decreases rapidly, this neuron will split into two new neurons, and the relevant parameters of the new neurons are updated as follows:

[0131] , (26)

[0132] in , and They represent the center, width and output weight of the newly added neurons respectively.

[0133] (2) FNN structure deletion stage.

[0134] When AS does not exceed a certain value and the training error gradually decreases, this neuron needs to be deleted, and the parameters related to the remaining neurons are updated according to formula (24).

[0135] S203. Set the stopping criteria for structure addition and deletion.

[0136] Calculate output error using validation data , and In continuous The reduction in step As the stopping criterion of structural growth, it is expressed as:

[0137] , (27)

[0138] in ,when When it reaches 0.01 for the first time, the process of adding and removing structures stops; Indicates the expected output value; Represents the predicted output.

[0139] The prediction module uses the prediction model to complete the predictive control of the flue gas desulfurization system.

[0140] The one-step output of the prediction model and the change of the current control law are used as independent variables to construct the loss function. Combining the rolling optimization and gradient descent algorithms, the optimal control law sequence is solved by minimizing the loss function, that is, a predictive control framework for the flue gas desulfurization system based on self-organizing fuzzy neural network is constructed.

[0141] In the wet flue gas desulfurization system, the SO 2 The closed-loop feedback information is obtained by controlling the concentration and pH of the slurry in the oxidation zone. in and f re To implement SO 2 Assume that , and Respectively represent SO 2 Expected values ​​of concentration and pH, and SOFNN model for SO 2 Predicted values ​​of concentration and pH, and the control variable f in and f re The loss function is defined as follows:

[0142] , (28)

[0143] , , (29)

[0144] in, and represents a constant matrix, Indicates the maximum value of the control law increase; Indicates the change of control law; represents the control law; and Represent the prediction time domain and control time domain respectively; T represents the transposed matrix. The optimal control variable is solved by the gradient descent algorithm in the process of minimizing the loss function. The formula is as follows:

[0145] , (30)

[0146] in, Represents the learning rate of the solution process. Combining formula (11) and formula (13), the optimal control variable can be expressed as:

[0147] , (31)

[0148] In the formula, Indicates the reference input.

[0149] It can be seen from formula (31) that the optimal control variable sequence can be solved in the rolling optimization and iteration process, thereby forming the SO 2 Closed-loop control framework for concentration and pH, such as Figure 3 shown.

[0150] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A flue gas desulfurization system predictive control method based on a self-organizing fuzzy neural network, characterized in that the steps include: Based on the dynamic characteristics of the evolution process of the wet flue gas desulfurization system variables, a control method framework is proposed; the steps include: analyzing the dynamic characteristics of the evolution process of the key variables of the wet flue gas desulfurization system, describing the control process and target of SO2 concentration and pH value as a closed-loop control problem, and designing the control method and framework; Based on the control method framework, a prediction model is constructed; the steps include: firstly establishing a fuzzy neural network model, including an input layer, a membership function layer, a rule layer and an output layer; based on the fuzzy neural network model, establishing a self-organizing fuzzy neural network; setting a stop criterion for adding and deleting structures for the self-organizing fuzzy neural network; The prediction model is used to complete the prediction control of the flue gas desulfurization system; the steps include: using the one-step output of the prediction model and the current control law change as independent variables to construct a loss function, combining rolling optimization and gradient descent algorithms, and solving the optimal control law sequence by minimizing the loss function.

2. The flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network according to claim 1 is characterized in that: The kinetic characteristics include: ， ， ， Based on the dynamic characteristics, the control method framework is designed, which includes: obtaining closed-loop feedback information by continuously detecting the SO2 concentration at the top of the absorption zone and the pH of the slurry in the oxidation zone, and then controlling f in and f re To achieve maximum absorption of SO2.

3. The flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network according to claim 1 is characterized in that: The fuzzy neural network model includes: ， ， ， ， In the formula, i, r, j represent natural numbers; and represents the center and width vector of the jth neuron, and , where R represents the number of neurons in the membership function layer and the rule layer; m represents the dimension of the input data; represents the output vector of the normalization layer; c ir represents the central value of the rth neuron in the regular layer under the action of the i-th input data; x i represents the i-th component in the input vector; , and They represent the center and width of the neurons in the membership function layer and the output matrix respectively; η represents the learning rate.

4. The flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network according to claim 3 is characterized in that: The self-organizing fuzzy neural network comprises: ， ， ， In the formula, AS represents the activation sensitivity of neurons in the regular layer, and It is the first The input and output of each neuron; , and Respectively represent the center, width and output weight of the newly added neurons; ; Indicates the expected output value; Represents the output error; In continuous The reduction in step As a stopping criterion for structural growth.

5. The flue gas desulfurization system predictive control method based on self-organizing fuzzy neural network according to claim 4 is characterized in that: The loss function includes: ， In the formula, represents the control law; Indicates the change of control law; represents the reference input; P and Q represent constant matrices; γ represents the learning rate of the solution process; Represents the predicted output.

6. A flue gas desulfurization system predictive control system based on a self-organizing fuzzy neural network, the system being used to implement the method according to any one of claims 1 to 5, characterized in that: include: Design module, construction module and prediction module; The design module is used to propose a control method framework based on the dynamic characteristics of the evolution process of the wet flue gas desulfurization system variables; the steps include: analyzing the dynamic characteristics of the evolution process of the key variables of the wet flue gas desulfurization system, describing the control process and target of SO2 concentration and pH value as a closed-loop control problem, and designing a control method and framework; The construction module is used to construct a prediction model based on the control method framework; the steps include: firstly establishing a fuzzy neural network model, including an input layer, a membership function layer, a rule layer and an output layer; based on the fuzzy neural network model, establishing a self-organizing fuzzy neural network; setting a stop criterion for adding and deleting structures for the self-organizing fuzzy neural network; The prediction module is used to complete the predictive control of the flue gas desulfurization system using the prediction model; the steps include: constructing a loss function using the one-step output of the prediction model and the current control law change as independent variables, combining rolling optimization and gradient descent algorithms, and solving the optimal control law sequence by minimizing the loss function.

Images