Intelligent flue gas oxygen content control method based on interval type 3 fuzzy width learning

Abstract

The invention belongs to the technical field of urban solid waste incineration, and provides an intelligent flue gas oxygen content control method based on interval type 3 fuzzy width learning, which comprises the steps of IT3FBLS controller module construction, PID controller module construction, controller output module construction, NSWI process regulation and control and hyper-parameter updating. According to the invention, the IT3FBLS controller module and the PID controller module are integrated, so that the response speed is high, overshoot is avoided, the control performance is good, and the deviation between the flue gas oxygen content stabilized in an interference environment and a set value is small; according to the method, hyper-parameter updating is carried out through the BO algorithm, when the set value of the flue gas oxygen content changes, rapid and stable tracking can be carried out, the tracking error is small, and efficient, stable and accurate control can be achieved under the complex working condition.

Description

Technical Field

[0001] The present invention relates to the technical field of municipal solid waste incineration, and in particular to an intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning. Background Art

[0002] MSW is the abbreviation for municipal solid waste worldwide; MSW incineration is abbreviated as MSWI; dioxins are abbreviated as DXN; and sulfur dioxide is abbreviated as SO2. The production of MSW has increased significantly with the acceleration of urbanization. Currently, the main methods for disposing MSW are landfill and incineration. However, landfill disposal has become less common due to its large land occupation and environmental pollution. MSWI, with its advantages of harmlessness, resource utilization, and volume reduction, has become the primary method for MSW treatment. In the MSWI process, flue gas oxygen content is a key indicator of the excess air coefficient and is subject to strict control. Low flue gas oxygen content leads to increased incomplete heat loss and the production of toxic and hazardous gases such as DXN and SO2; high flue gas oxygen content reduces incineration efficiency. Research has shown that only when the oxygen content of the flue gas at the waste heat boiler outlet is maintained within an appropriate range can MSW and combustible flue gas be fully burned. Therefore, to maintain efficient and stable operation of the MSWI process, research on precise control of flue gas oxygen content is essential.

[0003] During the MSWI process, traditional control methods rely heavily on expert experience to adjust the controlled variables. The inherent subjectivity and arbitrariness of manual control methods lead to unstable on-site control of flue gas oxygen content. Commonly used neural network or PID control methods often perform poorly in highly nonlinear and uncertain environments, suffering from issues such as poor robustness, poor real-time performance, high computational complexity, and difficulty adjusting parameters. Summary of the Invention

[0004] In order to overcome the shortcomings of the existing technology, the purpose of the present invention is to provide an intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning to solve the problems of unstable control effect of manual mode and poor performance of neural network or PID control mode in dealing with highly nonlinear and uncertain environments.

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

[0006] An intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning, comprising:

[0007] Build the IT3FBLS controller module;

[0008] Build the PID controller module;

[0009] Associating the IT3FBLS controller module with the PID controller module to obtain a controller output module;

[0010] The controller output module is used to control the NSWI process, and the sensor is used to monitor the NSWI process to obtain an actual output value of the flue gas oxygen content;

[0011] The hyperparameters of the controller output module are updated using the BO algorithm according to the actual output value of the flue gas oxygen content, and the process returns to the step of "using the controller output module to regulate the NSWI process and using a sensor to monitor the NSWI process to obtain the actual output value of the flue gas oxygen content".

[0012] Preferably, an electronic device comprises: at least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions that can be executed by the processor, and the instructions are executed by the processor so that the processor can execute the aforementioned intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning.

[0013] Preferably, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the aforementioned intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning.

[0014] The present invention discloses the following technical effects:

[0015] The present invention provides an intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning. By integrating the IT3FBLS controller module and the PID controller module, the problem of unstable control effect in manual mode is solved, and control with fast response speed and no overshoot is achieved. By using the BO algorithm to update hyperparameters, the problem of poor performance of neural network or PID control mode in dealing with highly nonlinear and uncertain environments is solved, and efficient, stable and precise control is achieved under complex working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 any creative work.

[0017] Figure 1 A schematic diagram of the intelligent control process of flue gas oxygen content based on interval type 3 fuzzy width learning provided by an embodiment of the present invention;

[0018] Figure 2 The MSWI process flow chart provided by the embodiment of the present invention;

[0019] Figure 3 A schematic diagram of the Pearson correlation coefficient provided by an embodiment of the present invention;

[0020] Figure 4 Schematic diagram of the flue gas oxygen content control strategy based on IT3FBLS-PID provided in an embodiment of the present invention;

[0021] Figure 5 A structural diagram of the IT3FBLS provided in an embodiment of the present invention;

[0022] Figure 6 A schematic diagram of a BO process objective function change curve provided by an embodiment of the present invention;

[0023] Figure 7 Schematic diagram of constant set value tracking experimental results under different methods provided by the embodiment of the present invention, Figure 7 (a) is a schematic diagram of the flue gas oxygen content tracking curve in the constant set value tracking experiment under different methods. Figure 7 (b) Schematic diagram of the tracking error curve of flue gas oxygen content in the constant set value tracking experiment under different methods;

[0024] Figure 8 A schematic diagram of the constant set value tracking experiment results provided by an embodiment of the present invention, Figure 8 (a) is a schematic diagram of the primary air volume increment curve in the constant set value tracking experiment. Figure 8 (b) is a schematic diagram of the primary air volume correction curve for the constant set value tracking experiment;

[0025] Figure 9 Schematic diagram of experimental results of setting value tracking under different methods provided by the embodiment of the present invention, Figure 9 (a) is a schematic diagram of the flue gas oxygen content tracking curve in the set value tracking experiment under different methods. Figure 9 (b) Schematic diagram of the tracking error curve of flue gas oxygen content in the set value tracking experiment with different methods;

[0026] Figure 10 Schematic diagram of the experimental results of the variable set value tracking provided by the embodiment of the present invention, Figure 10 (a) is a schematic diagram of the primary air volume increment curve in the variable set value tracking experiment. Figure 10 (b) Schematic diagram of the primary air volume correction curve in the variable set value tracking experiment. DETAILED DESCRIPTION

[0027] Figure 1 The flow chart of intelligent control of flue gas oxygen content based on interval type 3 fuzzy width learning provided by the embodiment of the present invention is as follows: Figure 1As shown, the present invention provides an intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning, comprising:

[0028] Step 100: Build the IT3FBLS controller module;

[0029] Step 200: Construct a PID controller module;

[0030] Step 300: Associating the IT3FBLS controller module with the PID controller module to obtain a controller output module;

[0031] Step 400: Regulating the NSWI process using the controller output module and monitoring the NSWI process using a sensor to obtain an actual output value of flue gas oxygen content;

[0032] Step 500: Update the hyperparameters of the controller output module using the BO algorithm according to the actual output value of the flue gas oxygen content, and return to the step of "using the controller output module to control the NSWI process and using the sensor to monitor the NSWI process to obtain the actual output value of the flue gas oxygen content."

[0033] refer to Figure 2 , MSWI process description for flue gas oxygen content control, MSWI process description. The process flow of a typical grate furnace MSWI mainly includes six process stages: storage and fermentation, solid waste combustion, waste heat exchange, steam power generation, flue gas treatment, and flue gas emission. The functions of each stage are described as follows:

[0034] 1) Storage and fermentation stage: The raw MSW is fermented in the solid waste storage tank for 3 to 7 days to increase the calorific value, and will be put into the hopper by the operator through the grab bucket.

[0035] 2) Solid waste combustion stage: The fermented MSW is converted into high-temperature flue gas and solid residue under the coupling of multi-phases such as solid-gas-liquid and multi-fields such as heat flow. It can be divided into three sub-stages: drying, combustion and burnout.

[0036] ① Drying sub-stage: The surface moisture gradually evaporates as the temperature rises, and is completely evaporated when the temperature rises to 100°C; at the same time, the internal moisture gradually precipitates and absorbs a large amount of heat energy.

[0037] ② Combustion sub-stage: From the initiation of MSW combustion to high-intensity luminous heating, and finally ending with an oxidation reaction. From the perspective of chemical reaction mechanisms, this sub-stage involves strong oxidation, pyrolysis, and atomic radical collision reactions. Among them, strong oxidation reaction refers to the complete combustion reaction of combustible components with oxygen; pyrolysis reaction refers to the destruction of chemical bonds between elements of carbon-containing polymer compounds or their reorganization by thermal radiation energy under oxygen-free or near-oxygen-free conditions, and then oxidation reaction after the precipitation of volatiles; atomic radical collision reaction refers to the transition of atomic radical electron energy, molecular rotation and vibration, etc., which generate infrared thermal radiation, visible light, and ultraviolet light, and then form flame morphology. Therefore, this sub-stage is closely related to oxygen content.

[0038] ③ Burnout sub-stage: The residual combustible components after combustion are primarily composed of coke. Under high temperature conditions, the coke undergoes oxidation reactions with O₂ in the primary air and gasifies with CO₂, water vapor, and other substances. Furthermore, inert materials gradually accumulate until all MSW is converted to ash, and combustion subsequently weakens and ceases completely. Prolonging the burnout sub-stage generally effectively increases the thermal reduction rate of MSW and improves the overall reduction level.

[0039] 3) Waste Heat Exchange Stage: First, the high-temperature flue gas passes through the water-cooled wall for initial cooling. Next, with the help of related equipment, the heat energy enters the boiler through radiation and convection. Then, within the boiler, the water is converted into high-pressure superheated steam, which then enters the steam-based power generation stage. Finally, the flue gas temperature at the boiler outlet drops rapidly to 200°C.

[0040] 4) Steam power generation stage: The high-temperature steam generated by the waste heat boiler is used to drive the steam turbine generator to realize the conversion of mechanical energy into electrical energy.

[0041] 5) Flue gas treatment stage: First, the selective non-catalytic reduction (SNCR) system removes NO at a temperature range of 850°C to 1100°C x Next, a semi-dry deacidification process uses lime and water to neutralize the acidic gases. Activated carbon is then injected to adsorb DXN and heavy metals in the flue gas. Finally, a bag filter removes particulate matter from the flue gas, neutralizes the reactants, and adsorbs the activated carbon, completing the purification process.

[0042] 6) Flue gas emission stage: After treatment, the flue gas that meets the national emission standard GB18485-2014 is pulled by the induced draft fan and discharged into the atmosphere through the chimney.

[0043] Further analysis of factors influencing flue gas oxygen content reveals that, according to the above description of the MSWI process, in the combustion sub-stage, combustible components undergo a vigorous oxidation reaction with oxygen; in the burnout sub-stage, under the influence of high temperature and primary air, coke undergoes an oxidation reaction with oxygen. Therefore, the flue gas oxygen content at the waste heat boiler outlet is used as a key controlled variable in the MSWI process. Controlling the flue gas oxygen content within the process settings ensures: sufficient combustion of solid waste on the grate; sufficient oxidation of combustible gases within the furnace; and compliance with flue gas emission standards. Therefore, precise control of flue gas oxygen content is crucial for the stable operation of the combustion process.

[0044] Furthermore, the five key variables related to flue gas oxygen content are primary air volume, secondary air volume, feeder average speed, drying grate average speed and ammonia injection rate. The Pearson correlation coefficient (PCC) is used to evaluate the correlation between key variables and CV, such as Figure 3 As shown in the figure, with the exception of ammonia injection, the PCC values ​​of secondary and primary air volumes relative to flue gas oxygen content are relatively large. Furthermore, considering the actual MSWI process, the primary air volume is selected as the MV, and the remaining key variables are considered as interference variables.

[0045] Specifically, the control strategy is based on the overall control strategy of BO-IT3FBLS-PID. Figure 4 As shown. Where t represents the tth moment; y r (t) is the set value of flue gas oxygen content at time t; y(t) represents the actual output value of flue gas oxygen content at time t; e(t) = [e1(t), e2(t), e3(t)] = [e o (t),e p (t),e d (t)] represents the input characteristics of the controller, e o (t) represents the system error at time t, e p (t) represents the incremental error of the proportional term at time t, e d (t) represents the incremental error of the differential term at time t; e1(t), e2(t), and e3(t) represent the first error, second error, and third error at time t, respectively; [e1(t), e2(t), e3(t)] = [e o (t),e p (t),e d(t)] indicates that the first error, second error, and third error at time t are assigned to the system error, proportional incremental error, and differential incremental error, respectively; Δu1(t) indicates the output of the IT3FBLS controller at time t; Δu2(t) indicates the output of the PID controller at time t; u(t) = [u(t), d(t)] = [u(t), d1(t), …, d4(t)] indicates the value of the key variable at time t, u(t) indicates the value of the MV at time t, and d(t) indicates the value of the interference variable at time t; u(t-1) indicates the value of the MV at time t-1; Γ indicates the value of the BO process objective function; X indicates the hyperparameter to be optimized.

[0046] Furthermore, the functional description of each module in the control strategy is as follows:

[0047] 1) IT3FBLS controller module: This module is implemented based on prior knowledge. The prior knowledge is used as a training set to train IT3FBLS. The trained IT3FBLS is used as a controller to track the oxygen content of the flue gas.

[0048] 2) PID controller module: Tracks the oxygen content of flue gas based on an incremental PID controller;

[0049] 3) Controller output module: The IT3FBLS controller is connected in parallel with the PID controller to obtain the manipulated variable increment and further obtain the manipulated variable;

[0050] 3) Hyperparameter optimization module: This module updates the hyperparameters of the IT3FBLS-PID controller based on the BO algorithm to obtain a controller with better performance.

[0051] Specifically, the structure of the IT3FBLS controller module is as follows: Figure 5 As shown. Among them, W fe =[ω1,ω2,…,ω j ,…ω J ] represents the connection weight from the output of the IT3FNN subsystem node to the enhancement layer; ω j represents the connection weight between the IT3FNN layer and the jth enhancement node; j = 1, 2, ..., J represents the index of the enhancement node, and J represents the number of enhancement nodes; ω f is the connection weight between the IT3FNN layer output and the output layer; ω e is the connection weight between the enhancement layer output and the output layer; K (k = 1, 2, ..., K) represents the number of IT3FNN subsystems; f represents the output vector of the IT3NN layer; θ is the enhancement layer output vector.

[0052] Furthermore, this module is implemented based on prior knowledge, and the prior knowledge is used as a training set to train IT3FBLS, and the trained IT3FBLS is used as a controller. Assume that during the training of IT3FBLS, the training samples are Among them, l = 1, 2, ..., L represents the index of the number of samples in the prior knowledge, L represents the total number of samples contained in the prior knowledge, m = 1, 2, ..., M represents the index of the number of sample features in the prior knowledge, M = 3 represents the total number of sample features in the prior knowledge, Indicates that Assign values ​​to The IT3FBLS controller mainly includes three processes: forward calculation, parameter learning, and controller construction, which are described in detail below. The forward calculation and parameter learning are based on the first sample as an example.

[0053] Furthermore, the forward calculation is performed. IT3FBLS mainly includes the data input layer, IT3FNN layer, enhancement layer and output layer. The specific description of each layer is as follows:

[0054] 1) Data input layer: receiving input

[0055] 2) IT3FNN layer: This layer contains K (k = 1, 2, ..., K) IT3FNN subsystems. The subsystems mainly include the input layer, membership function layer, fuzzy calculation layer, reduction layer and output layer.

[0056] Input layer: used for information transmission, weight is 1, receiving input

[0057] Membership function layer: Assumption The value of Consider the fuzzy set for each input is the s(1,2,…,S)th fuzzy set (or fuzzy rule) of the lth sample m input, and calculates its slice μ=α at different levels h ,h=1,2,…H membership, and its membership in α h The horizontal slice at is a type 2 fuzzy set between The calculation is as follows:

[0058]

[0059] in, and Respectively In horizontal slices and α h The upper and lower bounds of the membership degree, and αh for:

[0060]

[0061] Among them, λ≥1 is a constant, indicating that The uncertainty of the standard partition of the Gaussian vertical slice at , the standard partition of the slice is is the variance associated with this vertical slice, indicating that the input value is The distribution uncertainty of the membership function is calculated as follows:

[0062]

[0063] in, is a constant used to adjust the uncertainty measure, The calculation is as follows:

[0064]

[0065] in, and The calculation is as follows:

[0066]

[0067] in, and Represent fuzzy sets Center, upper width and lower width at the transverse horizontal tangent μ=α0.

[0068] Fuzzy computation layer: In this layer, rules are defined to represent the local linear relationship between input variables and outputs. Define S (s = 1, 2, ..., S) fuzzy rules. The sth rule is as follows:

[0069]

[0070] in, υ s and It represents the weight of the upper and lower bound results corresponding to the sth rule. The activation strength of the rule is calculated as follows:

[0071]

[0072]

[0073] Among them, T n represents the T-norm.

[0074] Reduction layer: By performing the reduction (TR) operation, the lower and upper bounds of the subsystem's output value are obtained as follows:

[0075]

[0076] Output layer: Calculate the output of the kth subsystem as follows:

[0077]

[0078] The output of the IT3NN layer composed of K subsystems is:

[0079] f=[f1,L,f k ,L,f K ]

[0080] 3) Enhancement layer: Perform nonlinear transformation on the output of the IT3FNN layer. Taking the j-th enhancement node as an example, its output is calculated as follows:

[0081] θ j =tanh(fω j +b j )

[0082] Among them, ω j =[ω 1j ,…,ω Kj ] T represents the connection weight between the IT3FNN layer and the j-th enhancement node; b j Represents the bias coefficient between the IT3FNN layer and the j-th enhancement node.

[0083] 4) Output layer: The final output is obtained by linearly combining the output of the IT3FNN layer and the enhancement layer, as follows:

[0084]

[0085] Among them, ω f (t) = [ω f1 ,…,ω fK ] T is the connection weight between the output of IT3FNN layer and the output layer, θ=[θ1,θ2,…,θ J ] T is the enhancement layer output vector, ω e (t) = [ω e1 ,…,ω eJ ] T is the connection weight between the enhancement layer output and the output layer.

[0086] Preferably, the ridge regression approximation algorithm is used to calculate the parameter ω f and ω e The initial value of is solved as follows:

[0087]

[0088] in, I is the identity matrix, β is the regularization coefficient, A matrix consisting of the actual output values ​​of all samples.

[0089] Furthermore, parameter learning is performed. In order to reduce the performance function and improve network performance, the parameters in IT3FBLS are updated. j and b j The update is based on gradient descent, ω f and ω e The update is based on ridge regression. The error function of the IT3FBLS model is as follows:

[0090]

[0091] The parameter ω is calculated based on the gradient descent method. j and b j Make the update as follows:

[0092] First, define the performance function:

[0093]

[0094] Then, the gradient of the performance function with respect to the parameters is calculated as follows:

[0095]

[0096]

[0097] Finally, gradient descent is used to adjust the parameter ω j and b j Update as follows:

[0098]

[0099] Where η represents the parameter ω j and b j The learning rate when performing updates;

[0100] According to the previous formula, the parameter ω based on ridge regression is realized f and ω e Updates.

[0101] Specifically, the construction of IT3FBLS controller. The trained IT3FBLS is used as the controller, e=[e1,e2,e3]=[e o ,e p ,e d] as its input, and the output of the controller is Δu1(t), where e1, e2, and e3 are the first error, the second error, and the third error respectively; [e1, e2, e3] = [e o ,e p ,e d ] means assigning e1, e2, and e3 to e respectively o , e p and e d ;e o , e p and e d The calculations are as follows:

[0102] e o (t) = y r (t)-y(t)

[0103] e p (t) = e o (t)-e o (t-1)

[0104] e d (t) = e o (t)-2*e o (t-1)+e o (t-2)

[0105] Preferably, the PID controller module. This embodiment adopts an incremental PID controller, whose input is e=[e1,e2,e3]=[e o ,e p ,e d ], e o , e p and e d The calculation formula is shown below. The output value of the PID controller Δu2(t) is as follows:

[0106] Δu2(t)=K p (t)*e p (t)+K i (t)*e o (t)+K d (t)*e d (t)

[0107] Among them, K p , K i and K d They represent the proportional gain, integral time constant, and differential time constant respectively.

[0108] Specifically, the controller output module. The parallel controller MV increment (manipulated variable increment) is composed of the output of the IT3FBLS controller and the output of the PID controller, as follows:

[0109] Δu(t)=Δu1(t)+Δu2(t)

[0110] The manipulated variables are further obtained as follows:

[0111] u(t)=u(t-1)+Δu1(t)+Δu2(t)

[0112] Preferably, a hyperparameter optimization module based on the BO algorithm. BO, as one of the optimization methods based on machine learning, achieves optimization by minimizing the objective function. This embodiment adopts the BO algorithm, selects 9 hyperparameters as targets to optimize the controller, and uses the integral of squared error (ISE) as the objective function. The process of the BO algorithm is as follows:

[0113] 1) Set the boundary space

[0114] Set the value range of the BO input sample X = {x1; L; x9}, where L represents the number of hyperparameters to be optimized, including the upper bound (UB) and lower bound (LB) of the sample space, as follows:

[0115]

[0116] Among them, {x1; L; x9} are respectively represented as follows:

[0117]

[0118] 2) Construct the initial dataset

[0119] Randomly initialize P init input samples And observe the objective function value corresponding to the sample Construct the initial data set.

[0120] 3) Build / train the GPR model

[0121] The Gaussian process regression (GPR) model is constructed using the dataset. This model assumes that the known data follows a multivariate Gaussian distribution as the prior probability distribution, as follows:

[0122] Γ~N(ζ0,Σ0)

[0123] Among them, ζ0=[ζ0(x1),…,ζ0(x p )] is the mean function, p is the number of samples in the current dataset, and Σ0 is the covariance matrix, which is expressed as follows:

[0124]

[0125] Among them, σ 2 is the variance of the measurement noise, Σ0(x a ,xb ) is the kernel function between the a-th and b-th sample points, calculated as follows:

[0126]

[0127] Among them, x ao represents the o-th dimension feature of the a-th sample, σ Γ is the kernel amplitude of the kernel function, η o Indicates the length ratio of the o-th dimension of the sample.

[0128] The posterior probability distribution of the new sample point in the sample space can be derived from the prior probability distribution. According to the definition of GP, the joint Gaussian distribution Γ of the known data and the predicted value It also obeys the Gaussian distribution and can be expressed as:

[0129]

[0130] in, is the covariance of the training and testing sets, is the test set covariance.

[0131] Since the training set is known, the given Γ can be calculated based on (38) The conditional distribution of is as follows:

[0132]

[0133] Among them, p and as follows:

[0134]

[0135] Among them, p is the posterior mean, is the posterior variance.

[0136] 4) Estimate the parameters of the GPR model;

[0137] Maximum Likelihood Estimation (MLE) is used to provide key support for hyperparameter optimization in GP, ​​thereby improving model accuracy and ultimately achieving more efficient optimization of the objective function.

[0138] 5) Sampling;

[0139] This paper uses the expected improvement (EI) sampling function to sample the objective function to find the optimal point x op .

[0140] 6) Update the dataset;

[0141] Calculate the new sample point x * The corresponding objective function value Γ(x* ), and update the data set X={x1;…;x p ;x *} and Γ=[Γ(x1),…,Γ(x p ),Γ(x * )] T .

[0142] Repeat steps 3) to 6) until the preset maximum number of iterations is reached, and find the optimal sample point x that meets the requirements from the data set. op .

[0143] Specifically, the symbols and their meanings used in this embodiment are: e: input characteristics of the controller; X: data set of the BO algorithm; Γ: objective function value; u: input of the controlled object; W fe : The connection weight from the output of the IT3FNN subsystem node to the enhancement layer; E IT3 : training sample of IT3FBLS; m: index of sample feature number in prior knowledge; M: total number of sample features in prior knowledge; The constant value of λ: The uncertainty of the standard partition of the Gaussian vertical slice at W f,e (t) is is f(t)|θ(t); I: identity matrix; β: regularization coefficient; The matrix of the actual output values ​​of all samples; η: parameter ω j and b j Learning rate; UB: upper bound of sample space; LB: lower bound of sample space; P init : Randomly initialize the initial sample number; ζ0: mean function; p: number of samples in the current dataset; Σ0: covariance matrix; σ 2 : variance of measurement noise; Σ0(x a ,x b ): kernel function between the a-th and b-th sample points; x ao : the o-th dimension feature of the a-th sample; σ Γ : kernel amplitude of the kernel function; η o : The length ratio of the oth dimension of the sample; The predicted value of the objective function; Training and testing set covariance; Test set covariance; ζ p : posterior mean; Posterior variance; x op :Make the objective function value Γ(x op ) minimizes the optimal point x op ;x *: new sample point; Q: number of optimization iterations; Ω: EI sampling function; ratio: exploration ratio; T: total number of controller iterations; t0: initial time of controller iteration.

[0144] Furthermore, the experimental verification was conducted. The control performance evaluation indexes were ISE, integral of absolute error (IAE) and maximum deviation of set value (Devmax) to analyze the control performance. The calculations were as follows:

[0145]

[0146] Dev max =max{|e o (t)|}

[0147] Where T represents the total number of controller iterations, and t0 represents the initial time of controller iteration.

[0148] This example uses operational data from an actual MSWI power plant from 8:00 AM to 12:00 PM on a single day. The preprocessed dataset consists of 857 samples, including four disturbance variables (secondary air volume, average feeder speed, average drying grate speed, and ammonia injection rate), one MV (primary air volume), and one CV (flue gas oxygen content). The flue gas oxygen content model is constructed using a BO-based primary-compensation integrated strategy.

[0149] The experimental results show that the objective function change curve in the BO process is as follows: Figure 6 As shown in Figure 2, in the BO process, the value range of the objective function ISE is [1.7723E-03, 2.2274E-4].

[0150] Furthermore, the experiment employed a control method based on the BO-IT3FBLS-PID to control the flue gas oxygen content of an actual MSWI power plant. To evaluate the control performance of the proposed strategy, two simulation experiments were designed: one with a constant flue gas oxygen content (6.8) and one with a variable setpoint (6.4 to 7.05). During the experiments, 60dBW Gaussian white noise was applied as interference. The hyperparameters of the IT3FBLS-PID controller obtained based on the BO algorithm are shown in Table 1:

[0151] Table 1

[0152]

[0153]

[0154] In order to further illustrate the performance of the controller proposed in this embodiment, PID control, TSFNN-PID control, IT3FBLS-PID control and BO-IT3FBLS control are set as comparative experiments. The hyperparameters of the comparative experiments are set as follows: PID control: K p =0.6, K i =0.2, K d =0.02; TSFNN-PID control: K p =0.7, K i =0.5, K d =0.001,η c =0.01,η b =0.01,η a =0.2, where η c , η b and η a Represent the center, width and learning rate of the subsequent coefficient of TSFNN respectively; IT3FBLS-PID control: β=5e-06,S=5,K=5,H=1,J=500,η=0.3,K p =0.5, K i =0.3, K d =0.2; BO-IT3FBLS control: β=1.12e-06, S=16, K=8, H=5, J=307, η=0.68.

[0155] Specifically, the constant value tracking experiment results. In the constant value tracking experiment, the set value of the flue gas oxygen content is set to 6.8, and the number of iterations is set to 1000 times. Figure 7 (a) Figure 7 (b) Figure 8 (a) Figure 8 (b).

[0156] In the constant setpoint tracking experiment, the comparison of the performance indicators of different controllers is shown in Table 2:

[0157] Table 2

[0158]

[0159] refer to Figure 7 Compared with the PID controller, TSFNN-PID controller, IT3FBLS-PID controller and BO-IT3FBLS controller, the BO-IT3FBLS-PID controller proposed in this embodiment has a fast response speed, no overshoot, and good control performance. In an interference environment, the deviation between the stabilized flue gas oxygen content and the set value is small, maintained at ±0.2214, indicating that the controller has strong anti-interference ability.

[0160] As shown in Table 2, the BO-IT3FBLS-PID proposed in this example has small ISE and IAE values, demonstrating that the proposed control method strikes a good balance between response speed and steady-state performance, resulting in superior performance. Furthermore, it exhibits advantages over other controllers in handling uncertainty. Comparisons with a PID controller, an IT3FBLS-PID controller, and a BO-IT3FBLS controller demonstrate the effectiveness of the IT3FBLS controller module, the BO module, and the PID controller module, respectively. This demonstrates the controller's strong stability and robustness.

[0161] refer to Figure 8 ,In the tracking process, the IT3FBLS controller and the PID controller ,play a comparable role, reflecting the effectiveness and ,rationality of the parallel controller design, which can achieve efficient, ,stable and precise control under complex working conditions.

[0162] Furthermore, the experimental results of the variable set value are given. In order to verify the performance of the proposed controller, the flue gas oxygen content is set to change gradually in the range of 6.4 to 7.05 to verify the adaptability of the controller when the set value changes in the MSWI process control. Figure 9 (a) Figure 9 (b) Figure 10 (a) Figure 10 (b) and Table 3.

[0163] Table 3

[0164]

[0165] refer to Figure 9 In the experiment of changing the set value, the BO-IT3FBLS controller is unable to track the controlled variable when the set value of the flue gas oxygen content changes. Compared with the PID controller, the TSFNN-PID controller and the BO-IT3FBLS controller, the BO-IT3FBLS-PID controller proposed in this embodiment can quickly and stably track the flue gas oxygen content when the set value changes, and the tracking error is small, within the range of ±0.6535.

[0166] Referring to Table 3, compared with other methods, the control method proposed in this embodiment has smaller ISE and IAE, indicating that the proposed control method shows a better balance between the controller in the transition process and the steady-state process, which can not only respond to changes quickly but also avoid the accumulation of errors in long-term operation; at the same time, it is superior to other controllers in dealing with uncertainty problems; by comparing with the PID controller, IT3FBLS-PID controller and BO-IT3FBLS controller, the effectiveness of the IT3FBLS controller module, BO module and PID module are respectively explained. The above shows that the controller proposed in this embodiment has strong stability, robustness and adaptability. In addition, according to Figure 10 It can be seen that in the tracking process, the IT3FBLS controller and the PID controller play a comparable role, reflecting the effectiveness and rationality of the parallel controller design, which can achieve efficient, stable and precise control under complex working conditions.

[0167] The beneficial effects of the present invention are as follows:

[0168] The present invention integrates the IT3FBLS controller module and the PID controller module, has a fast response speed, no overshoot, and good control performance. The deviation between the flue gas oxygen content after stabilization and the set value in an interference environment is small. By using the BO algorithm to update the hyperparameters, when the set value of the flue gas oxygen content changes, it can track quickly and stably with a small tracking error, and can achieve efficient, stable and precise control under complex working conditions.

[0169] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0170] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning, characterized in that: include: Build the IT3FBLS controller module; Build the PID controller module; Associating the IT3FBLS controller module with the PID controller module to obtain a controller output module; The controller output module is used to control the NSWI process, and the sensor is used to monitor the NSWI process to obtain an actual output value of the flue gas oxygen content; The hyperparameters of the controller output module are updated using the BO algorithm according to the actual output value of the flue gas oxygen content, and the process returns to the step of "controlling the NSWI process using the controller output module and monitoring the NSWI process using a sensor to obtain the actual output value of the flue gas oxygen content." 2. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 1 is characterized in that: Build the IT3FBLS controller module, including: Build the data input layer; Construct K IT3FNN subsystems to obtain the IT3FNN layer; Construct an enhancement layer; the expression of the enhancement layer is: θ j =tanh(fω j +b j ); where θ is the output vector of the enhancement layer; f is the output vector of the IT3FNN layer; ω j is the connection weight between the IT3FNN layer and the j-th enhancement node; b j is the bias term coefficient between the IT3FNN layer and the j-th enhancement node; Construct a data output layer; the expression of the data output layer is: Where t represents the time; is the predicted output corresponding to the lth training sample in the training process; ω f is the connection weight between the IT3FNN layer and the data output layer; ω e is the connection weight between the enhancement layer and the data output layer; k is the index of the IT3FNN subsystem; f k (t) is the output of the kth IT3FNN subsystem at time t; ω fk (t) is the connection weight from the output of the IT3FNN subsystem node to the enhancement layer at time t; J is the number of enhancement nodes; j is the index of the enhancement node; ω ej (t) is the connection weight from the enhancement layer to the output layer of IT3FBLS at time t; The data input layer, the IT3FNN layer, the enhancement layer, and the data output layer are sequentially connected to obtain the IT3FBLS control model.

3. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 2 is characterized in that: Construct K IT3FNN subsystems and obtain the IT3FNN layer, including: Construct information input layer; Construct a membership function layer; the expression of the membership function layer includes: in, is the sth fuzzy set of the lth sample m input; α h is the membership degree of the h-th horizontal slice; They are In horizontal slices The upper and lower bounds of the membership degree; They are In horizontal slices α h The upper and lower bounds of the membership degree; for Membership degree at horizontal slice α0; The input value is The distribution uncertainty of the membership function; is a constant used to adjust the measure of uncertainty; Fuzzy set At the center of the horizontal tangent μ = α0; μ is the membership of the horizontal slice; Construct a fuzzy calculation layer; the expression of the fuzzy calculation layer includes: Where z is the activation strength of the rule; T n is the T-norm; s is the weight of the upper bound result corresponding to the sth rule; is the weight of the lower bound result corresponding to the sth rule; S is the number of fuzzy rules; s is the index of the fuzzy rule; f is the output of the IT3FNN subsystem; Construct a degradation layer; the expression of the degradation layer includes: in, f h 、 are the output lower bound and output upper bound of the IT3FNN subsystem respectively; Construct an information output layer; the expression of the information output layer is: Where H is the number of horizontal slices; h is the index of the horizontal slice.

4. The intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 3 is characterized in that: Build the IT3FBLS controller module, including: Construct an IT3FBLS error function; the expression of the IT3FBLS error function is: Wherein, ε is the output of the IT3FBLS error function; is the true value of the lth training sample in the training process; is the predicted output corresponding to the lth training sample in the training process; A performance function is constructed based on the IT3FBLS error function; the expression of the performance function is: Among them, E u (t) is the output of the performance function; The parameters of the enhancement layer are updated using a gradient descent strategy according to the performance function, and the parameters of the data output layer are updated using a ridge regression approximation algorithm to obtain the trained IT3FBLS control model.

5. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 4 is characterized in that: Building the IT3FBLS controller module also includes: A controller is constructed based on the output of the trained IT3FBLS control model to obtain the IT3FBLS controller module; the expression of the IT3FBLS controller module is: e=[e o ,e p ,e d ]; where e o (t) = y r (t)-y(t);e p (t) = e o (t)-e o (t-1); e d (t) = e o (t)-2*e o (t-1)+e o (t-2); e is the input feature of the controller; e o 、e p 、e d They are system error, proportional incremental error, and differential incremental error respectively; y r (t) and y(t) are the set value and actual output value of the flue gas oxygen content at time t respectively.

6. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 5 is characterized in that: Build the PID controller module, including: A controller is constructed according to the output of the trained IT3FBLS control model to obtain the PID controller module; the expression of the PID controller module is: Δu2(t)=K p (t)*e p (t)+K i (t)*e o (t)+K d (t)*e d (t); Wherein, Δu2(t) is the output of the PID controller module at time t; K p , K i , K d They are proportional gain, integral time constant, and differential time constant respectively.

7. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 6 is characterized in that: The IT3FBLS controller module and the PID controller module are associated to obtain a controller output module, including: The IT3FBLS controller module and the PID controller module are summed to obtain a manipulated variable increment; the manipulated variable increment is expressed as: Δu(t)=Δu1(t)+Δu2(t); wherein Δu(t) is the manipulated variable increment at time t; Δu1(t) is the output of the IT3FBLS controller module at time t; The manipulated variable is constructed according to the manipulated variable increment to obtain the controller output module; the expression of the controller output module is: u(t)=u(t-1)+Δu1(t)+Δu2(t); wherein u(t) is the manipulated variable value at time t.

8. The intelligent control method of flue gas oxygen content based on interval type 3 fuzzy width learning according to claim 7 is characterized in that: The hyperparameters of the controller output module are updated using the BO algorithm according to the actual output value of the flue gas oxygen content, and the process returns to the step of "controlling the NSWI process using the controller output module and monitoring the NSWI process using a sensor to obtain the actual output value of the flue gas oxygen content", including: Select the hyperparameters to be optimized; Setting the upper and lower bounds of each hyperparameter to be optimized to obtain a value range; Initializing input samples according to the value range and the hyperparameter to be optimized, and observing the objective function value corresponding to the input sample to obtain an initial data set; Constructing a Gaussian process regression model and training the Gaussian process regression model using the initial data set; According to the trained Gaussian process regression model, the objective function is sampled using maximum likelihood estimation and expected lifting sampling function to obtain a temporary optimal point; The initial data set is updated according to the temporary optimal point, and the process returns to the step of "constructing a Gaussian process regression model and training the Gaussian process regression model using the initial data set" until a preset maximum number of iterations is reached to obtain the optimal sample point; The optimal sample points are used to update the hyperparameters of the controller output module.

9. An electronic device, characterized in that: include: At least one processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning according to any one of claims 1 to 8.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the intelligent control method for flue gas oxygen content based on interval type 3 fuzzy width learning according to any one of claims 1 to 8.

Images