A method for predicting oxygen content in flue gas during MSWI processes

Abstract

This invention provides a method for predicting the oxygen content of flue gas in MSWI processes, relating to the field of urban solid waste incineration technology. The method includes: constructing a final flue gas oxygen content prediction model; inputting manipulated variables into the final flue gas oxygen content prediction model to obtain the flue gas oxygen content; the method for constructing the final flue gas oxygen content prediction model is as follows: constructing a master model of flue gas oxygen content based on EnTSFRT; constructing a compensation model based on LRDT and the error values ​​of the master model; constructing an initial flue gas oxygen content prediction model based on the compensation model and the master model; and optimizing the hyperparameters of the initial model using the BO algorithm to obtain the final flue gas oxygen content prediction model. This invention solves the problems of poor interpretability, low model accuracy, and complex manual parameter tuning in existing MSWI process flue gas oxygen content prediction technologies.

Description

Technical Field

[0001] This invention relates to the field of urban solid waste incineration technology, and in particular to a method for predicting the oxygen content of flue gas in a MSWI process. Background Technology

[0002] The oxygen content in flue gas is one of the important parameters in the municipal solid waste incineration (MSWI) process. Its stable control is closely related to the incineration efficiency and pollutant emissions of the MSWI process. However, existing methods have problems such as poor interpretability, low model accuracy and complex manual parameter tuning. Summary of the Invention

[0003] To overcome the shortcomings of the prior art, the purpose of this invention is to provide a method for predicting the oxygen content of flue gas in MSWI processes. This invention solves the problems of poor interpretability, low model accuracy, and complex manual parameter tuning in the prior art.

[0004] To achieve the above objectives, the present invention provides the following solution:

[0005] A method for predicting oxygen content in flue gas during MSWI processes includes:

[0006] Construct a prediction model for the final oxygen content of flue gas;

[0007] Input manipulated variables into the final flue gas oxygen content prediction model to obtain the flue gas oxygen content;

[0008] The method for constructing the final flue gas oxygen content prediction model is as follows:

[0009] Construct a master model for flue gas oxygen content based on EnTSFRT;

[0010] Based on LRDT, a compensation model is constructed according to the error value of the master model of flue gas oxygen content;

[0011] An initial flue gas oxygen content prediction model is constructed based on the aforementioned compensation model and the main model of flue gas oxygen content.

[0012] The parameters of the initial flue gas oxygen content prediction model were optimized using the BO algorithm to obtain the final flue gas oxygen content prediction model.

[0013] Preferably, the step of constructing the master model for flue gas oxygen content based on EnTSFRT includes:

[0014] Construct multiple TSFRT sub-models;

[0015] EnTSFRT is obtained by parallel integration of the multiple TSFRT sub-models;

[0016] A master model for flue gas oxygen content was constructed based on EnTSFRT.

[0017] Preferably, the expression for the master model of flue gas oxygen content is:

[0018]

[0019] in, The output value of the main model for flue gas oxygen content is given by F, where F represents the EnTSFRT model. Let U, K, and θ be the weight matrix of TSFRT in the EnTSFRT model. leaf Let be the minimum number of samples, l be the width matrix of the membership function, δ be the center matrix of the membership function, and ω be the consequent weight parameter matrix.

[0020] Preferably, optimizing the parameters of the initial flue gas oxygen content prediction model using the BO algorithm includes:

[0021] Construct the final optimized model;

[0022] Input the hyperparameter set into the final optimized model to obtain the optimized hyperparameter set;

[0023] The method for constructing the optimization model is as follows:

[0024] Set the boundary space and construct the initial dataset within the set space;

[0025] Construct an initial optimization model based on the initial dataset;

[0026] The parameters of the initial optimization model are adjusted using maximum likelihood estimation to obtain the final optimization model.

[0027] Preferably, the expression for the final flue gas oxygen content prediction model is as follows:

[0028]

[0029] in, is the output value of the final flue gas oxygen content prediction model, and is the error output value of the compensation model.

[0030] Preferably, the hyperparameters include:

[0031] The output parameters, minimum number of samples, and number of decision trees of the main model for flue gas oxygen content, and the minimum number of samples and regularization coefficients of the compensation model.

[0032] The present invention discloses the following technical effects:

[0033] This invention provides a method for predicting the oxygen content of flue gas in a MSWI process, comprising: constructing a final flue gas oxygen content prediction model; inputting manipulated variables into the final flue gas oxygen content prediction model to obtain the flue gas oxygen content; the method for constructing the final flue gas oxygen content prediction model is as follows: constructing a master model of flue gas oxygen content based on EnTSFRT; constructing a compensation model based on LRDT and the error value of the master model of flue gas oxygen content; constructing an initial flue gas oxygen content prediction model based on the compensation model and the master model of flue gas oxygen content; and optimizing the parameters of the initial flue gas oxygen content prediction model using the BO algorithm to obtain the final flue gas oxygen content prediction model. This invention improves the interpretability of the model by using a master-complement ensemble model based on a tree algorithm; it proposes a hyperparameter selection method for the master-complement tree ensemble model based on the BO algorithm, reducing the complexity and time consumption of manual parameter tuning. This enhances model performance; and it is the first to construct a BO-optimized master-complement tree ensemble model for flue gas oxygen content, providing support for related intelligent control research. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 A flowchart of a method for predicting the oxygen content in flue gas during an MSWI process, provided in an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of the overall modeling strategy provided in an embodiment of the present invention;

[0037] Figure 3 A schematic diagram of the master modeling strategy based on EnTSFRT provided in this embodiment of the invention;

[0038] Figure 4 A schematic diagram of the modeling strategy for the compensation model based on LRDT provided in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown, this invention provides a method for predicting the oxygen content in flue gas during an MSWI process, comprising:

[0042] Step 100: Construct the final flue gas oxygen content prediction model;

[0043] Step 200: Input manipulated variables into the final flue gas oxygen content prediction model to obtain the flue gas oxygen content;

[0044] The method for constructing the final flue gas oxygen content prediction model is as follows:

[0045] Step 101: Construct the master model of flue gas oxygen content based on EnTSFRT;

[0046] Step 102: Based on LRDT, construct a compensation model according to the error value of the master model of flue gas oxygen content;

[0047] Step 103: Construct an initial flue gas oxygen content prediction model based on the compensation model and the main model of flue gas oxygen content;

[0048] Step 104: Optimize the parameters of the initial flue gas oxygen content prediction model using the BO algorithm to obtain the final flue gas oxygen content prediction model.

[0049] Furthermore, such as Figure 2 As shown, the expression for the master model of flue gas oxygen content is:

[0050]

[0051] in, The output value of the main model for flue gas oxygen content is given by F, where F represents the EnTSFRT model. Let U, K, and θ be the weight matrix of TSFRT in the EnTSFRT model. leaf Let l be the minimum number of samples, l be the width matrix of the membership function, δ be the center matrix of the membership function, ω be the consequent weight parameter matrix, U be the input variable matrix, and K be the total number of fuzzy rules.

[0052] Specifically, Figure 2 In this context, u represents MV, the manipulated variable, and y represents the actual oxygen content in the flue gas. This represents the output value of the flue gas oxygen content from the EnTSFRT master model. This represents the error value of the EnTSFRT master model. This represents the error output value of the LRDT compensation model. This represents the error value of the LRDT compensation model. The output value of flue gas oxygen content for BO-EnTSFRT-LRDT is represented by the following variables:

[0053]

[0054]

[0055] Furthermore, the construction of the master model for flue gas oxygen content based on EnTSFRT includes:

[0056] Construct multiple TSFRT sub-models;

[0057] EnTSFRT is obtained by parallel integration of the multiple TSFRT sub-models;

[0058] A master model for flue gas oxygen content was constructed based on EnTSFRT.

[0059] The EnTSFRT model requires parallel integration of multiple TSFRT sub-models. Each TSFRT sub-model primarily consists of a filtering layer and a fuzzy inference layer. The filtering layer is used for feature selection, while the fuzzy inference layer utilizes TS fuzzy inference. For example... Figure 3 As shown, Figure 3 middle As input to EnTSFRT, u n Let MV (manipulated variable) represent the input sample, n represent the nth input sample, y represent the true value of oxygen content in flue gas, N represent the total number of samples in the dataset, M represent the dimension of the input feature, and U represent the value of the input feature. j Let D represent the input variable matrix of the j-th subset sample. j This represents the j-th training subset obtained by randomly sampling features from D. and f represents the membership degree on the (t+1)th and (t+2)th non-leaf nodes. TSFRT Represents the TSFRT model. This indicates the output value of EnTSFRT.

[0060] Furthermore, features are randomly sampled from D to obtain a training subset D of M features. j It can be represented as:

[0061]

[0062] In the formula, D j This represents the j-th training subset. For the input variable matrix, This represents the true value vector of oxygen content in flue gas.

[0063] Accordingly, from the perspective of input features, u containing M input featuresj It can be represented as:

[0064]

[0065] The subset {D} can be obtained 1 ,…,D j ,…D Tn}

[0066] Furthermore, the construction of the j-th TSFRT sub-model will be described as an example.

[0067] Dataset D j This serves as input to the filtering layer and determines the optimal segmentation position.

[0068] Obtain D j For each recursive coordinate point (j,m), calculate its mean square error (MSE) value, and determine the segmentation node by minimizing the MSE. The estimation process of the loss function value is as follows:

[0069]

[0070] In the formula, Φ represents the dataset D. j The loss value, Represents the left subset (or right subset) The MSE value of ) Represents the left subset The true vector, Represents the right subset The true vector, Indicates input u j The clear membership function at the t-th non-leaf node Represents the left subset The average of the target values, Represents the right subset The average of the target values. Here, will... and Simultaneously expressed as The calculation is as follows:

[0071]

[0072] In the formula, express (or The number of ).

[0073] In formula (5) The calculation is as follows:

[0074]

[0075] In the formula, μ cs (u j ) indicates the input u j Clear membership function, membership function Indicates sample u j The splitting operation on the t-th non-leaf node, φ t The segmentation threshold can be expressed as φ t =u j,m Assuming the TSFRT model of this invention consists of T nodes, the number of non-leaf nodes is... The sum of membership degree sets is

[0076] D j The data is divided into left and right subsets as follows:

[0077]

[0078] In the formula, This indicates that the clear set is obtained using the minimum MSE. The first membership degree is defined by equation (7) for the first element of the clarity set, and the clarity set is described as follows:

[0079]

[0080] Repeat the above steps until the minimum number of samples is reached at the leaf node. When the set threshold is reached, the first A clear set can be represented as:

[0081]

[0082] This can be further expressed as:

[0083]

[0084] according to Determine the training data for the nodes And construct TS fuzzy inference, It is expressed as follows:

[0085]

[0086] In the formula, This is the training dataset for fuzzy reasoning. Indicates the first leaf nodes, Indicates the first The number of samples in each leaf node.

[0087] In the fuzzy inference layer, K rules are defined to represent the local linear relationship between the input variables and the target. The k-th rule is defined as follows:

[0088]

[0089] In the formula, y k Let (13) represent the output of the consequent of the fuzzy rule, i.e., the weight of the k-th fuzzy rule. Further simplification of (13) is as follows:

[0090]

[0091] in, express Variables in express The membership function describes right The degree of membership.

[0092] This invention uses a Gaussian function as the membership function, as follows:

[0093]

[0094] In the formula, δ m,k Indicates the center of the membership function. This indicates the width of the membership function.

[0095] The activation strength of the k-th fuzzy rule among m input variables is calculated as follows:

[0096]

[0097] In the formula, α k This represents the product output of the k-th fuzzy rule. This represents the membership degree of the k-th fuzzy rule.

[0098] Further normalization operations are performed:

[0099]

[0100] Based on (13) and (17), y k and Perform a linear combination and then sum the results to obtain the predicted output value. as follows:

[0101]

[0102] Where g(·) represents the consequent output function, and ω represents the consequent weight parameter.

[0103] Repeat the above steps until you obtain... Each leaf node is obtained by... A fuzzy inference model is constructed using a multi-input single-output model, resulting in the following simplified TSFRT model:

[0104]

[0105] In the formula, Let U represent the j-th TSFRT model, where U = {U} 1 ,…,U j ,…,U Tn}, θ leaf Let ω represent the minimum number of samples, where ω = {ω1, ω2, ..., ω} m} represents the consequent weight parameter matrix, δ represents the center matrix of the membership function, l represents the width matrix of the membership function, and j represents the j-th TSFRT model.

[0106] Furthermore, this invention employs the least squares method to update the weights of successors.

[0107] Expand and merge (18) as follows:

[0108]

[0109] In the formula,

[0110] Given the input and output matrices, the weights of the TS consequents can be estimated as follows:

[0111] ω=((U * ) T U * ) -1 (U * ) T y (21)

[0112] In the formula, U * by u * constitute.

[0113] The parameters are learned using the least squares method, and the recursive calculation is as follows:

[0114]

[0115]

[0116] In the formula, ω0 is the weight of the consequent at the initial time step, is a random vector, i represents the i-th recursive process, and H i As an intermediate variable in the recursive process, H0 is defined as H0 = γI, where λ represents a large positive value and I represents the identity matrix.

[0117] Furthermore, firstly, based on the TSFRT modeling process, Tn TSFRT models are constructed. Defined as

[0118] Then, in order to estimate the weights with the minimum training error, this invention employs the following optimal problem to compute the pseudo-inverse:

[0119]

[0120] Where ν represents the regularization coefficient.

[0121] Using ridge regression theory, specifically the Moore-Penrose inverse matrix, to calculate the weight matrix, the results are as follows:

[0122]

[0123] The final output of EnTSFRT is as follows:

[0124]

[0125] Furthermore, the compensation model based on LRDT (the compensation model modeling module based on LRDT):

[0126] This module consists of two parts: feature selection and linear regression, with the following structure: Figure 4 As shown:

[0127] Figure 4 The feature selection module performs feature filtering to determine the input features of the leaf nodes, where the input feature of the s-th leaf node is... The linear regression module uses linear regression to calculate the predicted output of the leaf nodes, as shown in equation (30), where samples represents the number of node samples. Indicates an indicator function, This represents the node identifier for the first non-leaf node. This represents the input feature of the s-th leaf node. express The number of samples, express Dimensions Let J(·) represent the weight vector, and J(·) represent the cost function.

[0128] Furthermore, firstly, given the dataset This represents the true error value of the EnTSFRT master model.

[0129] D E Divide into left and right subsets, as follows:

[0130]

[0131] Among them, the difference from (8) is that in equation (27) Indicator functions are represented as follows:

[0132]

[0133] Based on (5), the node arbiter can be determined, and then the coordinates of the optimal splitting position can be determined.

[0134] There are T paths from the first non-leaf node to the leaf node. The input features of the leaf node are determined based on the nodes on the paths, as follows:

[0135]

[0136] Furthermore, the linear regression module: This module uses the linear regression method to calculate the predicted output of the leaf nodes, as follows:

[0137]

[0138] To ensure that the weight vector does not diverge during the modeling of flue gas oxygen content, this invention employs a Tikhonov regularized least squares cost function. The weight vector is solved using the following method:

[0139]

[0140] Compared to The gradient can be expressed as follows:

[0141]

[0142] make It can be obtained as follows:

[0143]

[0144] Finally, the output value of the leaf node can be calculated using (30).

[0145] Furthermore, the optimization of the parameters of the initial flue gas oxygen content prediction model using the BO algorithm includes:

[0146] Construct the final optimized model;

[0147] Input the hyperparameter set into the final optimized model to obtain the optimized hyperparameter set;

[0148] The method for constructing the optimization model is as follows:

[0149] Set the boundary space and construct the initial dataset within the set space;

[0150] Construct an initial optimization model based on the initial dataset;

[0151] The parameters of the initial optimization model are adjusted using maximum likelihood estimation to obtain the final optimization model.

[0152] Specifically, BO is a type of machine learning-based optimization method that aims to optimize the objective function, i.e., minimize the objective function. This invention uses the BO algorithm to optimize the RMSE values ​​of five hyperparameters in the master-complement tree ensemble model, where LSM... α MinSamples1 and Tn are the hyperparameters of the principal model, while MinSamples2 and β are the hyperparameters of the compensation model.

[0153] The optimization process of the BO algorithm is as follows:

[0154] Step 1: Set the boundary space;

[0155] Set the input samples for BO The range of values ​​for includes the upper boundary (UB) and lower boundary (LB) of the sample space, where D represents the number of sample features, set as follows:

[0156]

[0157] Where {x1; L; x5} represent the following respectively:

[0158]

[0159] Step 2: Construct the initial dataset;

[0160] Randomly initialize P init Input Samples And observe the objective function value corresponding to the sample. This constitutes the initial dataset.

[0161] Step 3: Build / train the GPR model;

[0162] A Gaussian process regression (GPR) model is constructed using the dataset. The GPR model treats the known data as a prior probability distribution that satisfies a multivariate normal distribution, as described below:

[0163] Γ~N(ζ0,Σ0)(36)

[0164] Among them, ζ0=[ζ0(x1),L,ζ0(x p ] is the mean function, p is the number of samples in the dataset at the current time, and Σ0 is the covariance matrix, as shown below:

[0165]

[0166] Where, σ 2 To measure the variance of noise, Σ0(x a ,x b Let be the kernel function between the a-th and b-th sample points, calculated as follows:

[0167]

[0168] Where, x ao Let σ represent the o-th dimension feature of the a-th sample. Γ Let η be the kernel amplitude of the kernel function. o This represents the length proportion of the o-th dimension of the sample.

[0169] For a new sample point in the sample space, the posterior probability distribution of that sample point can be calculated from the prior probability distribution. According to the definition of Gaussian distribution (GP), the joint Gaussian distribution Γ of the known data and the predicted value... Also following a Gaussian distribution, it can be represented as:

[0170]

[0171] in, To train the test set covariance, To test the set covariance.

[0172] Since the training set is known, the given Γ can be calculated based on (39). The conditional distribution is shown below:

[0173]

[0174] Where, ζ p and as follows:

[0175]

[0176] Where, ζ p The posterior mean is... This represents the posterior variance.

[0177] Step 4: Estimate the parameters of the GPR model (optimization model);

[0178] Estimate the hyperparameters in the mean function and covariance matrix. Maximum likelihood estimation (MLE) is then used to estimate these parameters, thereby improving the model's fitting and predictive performance.

[0179] Step 5: Sampling;

[0180] To find the optimal point x op Make the objective function value Γ(x) opTo minimize the objective function, the expected improvement (EI) sampling function is used to sample the objective function.

[0181] Step 6: Update the dataset;

[0182] Calculate the new sample point x * The corresponding objective function value Γ(x) * ), and update the dataset X = {x1; L; x p ;x *} and Γ=[Γ(x1),L,Γ(x p ),Γ(x * )] T .

[0183] Repeat steps 3-6 until the preset maximum number of iterations is reached, and find the optimal sample point x that meets the requirements from the dataset. op .

[0184] Furthermore, Loop represents the number of optimization iterations, Ω represents the EI sampling function, ratio represents the exploration ratio, and q represents the q-th optimization iteration. This indicates that when the input to BO is x p The output value corresponding to the time-mixed model.

[0185] Furthermore, the pseudocode of the algorithm used in this invention is shown in Table 1, which is as follows:

[0186] Table 1. Pseudocode of the BO-EnTSFRT-LRDT algorithm

[0187]

[0188]

[0189]

[0190] Furthermore, this experiment uses the actual operating data of a MSWI plant in Beijing on a certain day. The dataset contains a total of 857 samples. After empirical processing, it includes 4 MV (primary air volume, secondary air volume, feeder average speed, drying grate average speed) and one controlled variable (flue gas oxygen content).

[0191] The evaluation metrics used in this invention are RMSE and Mean Absolute Error (MAE), as follows:

[0192]

[0193] The parameters of BO are set as Loop = 20, Ω represents the sampling function, and this invention uses the EI function with a ratio of 0.5, resulting in the optimal hyperparameter LSM. α =1.7504, MinSamples1=42, Tn=200, MinSamples2=30 and β=0.30434.

[0194] To further illustrate the effectiveness of the proposed method, a comparative experiment was conducted with Random Forest (RF), LRDT, EnTSFRT-LRDT, and BO-EnTSFRT. The parameters for each model were set as follows: For Random Forest, the minimum number of samples was set to 50, and the number of features was 4; for LRDT, the minimum number of samples was set to 9, and the regularization coefficient was 0.4; for EnTSFRT-LRDT, LSM was set... α =2.4, MinSamples1=2, Tn=132, MinSamples2=9, β=0.5; During the BO-EnTSFRT modeling process, LSM is set. α =1.2266, MinSamples1=66, Tn=137.

[0195] Based on the above model, tests were conducted on the test set, and the test was repeated 20 times. The performance evaluation results of different methods are shown in Table 2.

[0196] Table 2 Statistical Results of Performance Evaluation Indicators for Different Methods

[0197]

[0198] Table 2 shows that, compared with the BO-EnTSFRT method, the proposed method has lower RMSE and MAE values ​​on the training, validation, and test sets, demonstrating the effectiveness of the LRDT-based error modeling module. Compared with the EnTSFRT-LRDT method, the proposed method also has lower RMSE and MAE values ​​on the validation and test sets, indicating the feasibility of using the BO algorithm to select optimal parameters. Furthermore, comparisons with LRDT and RF show that the proposed method has lower RMSE and MAE values ​​on both the validation and test sets than the LRDT method, and lower RMSE and MAE values ​​on all three sets than the RF method.

[0199] The above comparative experiments demonstrate the effectiveness and feasibility of the LRDT module and BO method of this invention, and also prove the accuracy and effectiveness of the method proposed in this invention.

[0200] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for predicting oxygen content in flue gas during MSWI processes, characterized in that, include: Construct a prediction model for the final oxygen content of flue gas; Input manipulated variables into the final flue gas oxygen content prediction model to obtain the flue gas oxygen content; The method for constructing the final flue gas oxygen content prediction model is as follows: Construct a master model for flue gas oxygen content based on EnTSFRT; Based on LRDT, a compensation model is constructed according to the error value of the master model of flue gas oxygen content; An initial flue gas oxygen content prediction model is constructed based on the aforementioned compensation model and the main model of flue gas oxygen content. The hyperparameters of the initial flue gas oxygen content prediction model were optimized using the BO algorithm to obtain the final flue gas oxygen content prediction model. Manipulated variables include: primary air volume, secondary air volume, feeder average velocity, and drying grate average velocity; The hyperparameters include: The output parameters, minimum number of samples, and number of decision trees of the main model for flue gas oxygen content, and the minimum number of samples and regularization coefficients of the compensation model.

2. The method for predicting oxygen content in MSWI process flue gas according to claim 1, characterized in that, The construction of the master model for flue gas oxygen content based on EnTSFRT includes: Construct multiple TSFRT sub-models; EnTSFRT is obtained by parallel integration of the multiple TSFRT sub-models; A master model for flue gas oxygen content was constructed based on EnTSFRT.

3. The method for predicting the oxygen content of flue gas in an MSWI process according to claim 1, characterized in that, The expression for the master model of flue gas oxygen content is: ； in, The output value of the main model for flue gas oxygen content is given by F, where F represents the EnTSFRT model. Let U, K, and K be the weight matrix of TSFRT in the EnTSFRT model. Minimum sample size The width matrix of the membership functions. The central matrix of the membership functions, This is the consequent weight parameter matrix. For the input variable matrix, This represents the total number of fuzzy rules.

4. The method for predicting the oxygen content in flue gas during an MSWI process according to claim 1, characterized in that, The optimization of the parameters of the initial flue gas oxygen content prediction model using the BO algorithm includes: Construct the final optimized model; Input the hyperparameter set into the final optimized model to obtain the optimized hyperparameter set; The method for constructing the optimization model is as follows: Set the boundary space and construct the initial dataset within the set space; Construct an initial optimization model based on the initial dataset; The parameters of the initial optimization model are adjusted using maximum likelihood estimation to obtain the final optimization model.

5. The method for predicting the oxygen content of flue gas in an MSWI process according to claim 3, characterized in that, The final expression for the flue gas oxygen content prediction model is: ； in, This is the output value of the final flue gas oxygen content prediction model. This is the output value to compensate for the model's error.

Images