Multi-objective integrated stochastic configuration network furnace temperature prediction method for municipal solid waste incineration process

Abstract

The application provides a kind of urban solid waste incineration process multi-objective integrated random configuration network furnace temperature prediction method, relating to urban solid waste incineration process furnace temperature intelligent prediction technical field, the method is to the urban solid waste incineration process data is pretreated, by self-service sampling strategy, obtain S training data set;Using random configuration network algorithm, base model construction is carried out, and the base model corresponding to each training data set is obtained;Using global negative correlation integration strategy based, S base model integrated model is trained in parallel, and the learning ability of integrated model to multi-region furnace temperature is realized by introducing group sparse regularization technology, and the trained integrated model is obtained;Using the trained integrated model, the urban solid waste incineration process data is analyzed, and the furnace temperature prediction result is obtained, to complete the prediction of furnace temperature.The application solves the problem of low accuracy and stability of the existing furnace temperature prediction method.

Description

Technical Field

[0001] This specification relates to the field of intelligent furnace temperature prediction technology for urban solid waste incineration processes, and in particular to a multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration processes. Background Technology

[0002] With urbanization and improved living standards, the volume of urban solid waste collected is increasing year by year. Incineration, which can reduce, render harmless, and recycle urban solid waste, has become a major means of solid waste disposal. Among the key parameters for controlling the incineration process, furnace temperature needs to be maintained above 850℃ to ensure the complete decomposition of harmful pollutants. However, the complex operating mechanism of the incineration process makes it difficult to fully describe using precise mathematical models. Furthermore, while on-site temperature sensors can effectively monitor furnace temperature changes in real time, the inherent lag during incineration can lead to untimely temperature sensing. Therefore, researching data-driven methods for predicting furnace temperature in urban solid waste incineration is of significant practical importance for precise control of furnace temperature and ensuring the safe and stable operation of the incineration process.

[0003] Currently, research on data-driven modeling methods is booming. Compared to traditional data analysis methods such as multivariate statistical analysis and partial least squares, artificial neural networks, with their powerful nonlinear mapping capabilities, self-learning, and adaptive characteristics, have become an important research direction in the field of industrial data analysis. However, parameter learning methods based on error backpropagation generally suffer from slow convergence speed and susceptibility to local optima, making it difficult for such methods to meet the timeliness and computational burden requirements of industrial processes. In recent years, neural network training methods based on stochastic learning have been widely applied in the field of industrial process data analysis due to their advantages such as low computational cost and fast learning speed. Among them, stochastic configuration networks (SCNs) can adaptively determine the input weights and biases of the model through a supervision mechanism oriented towards training data, thereby ensuring the model's universal approximation property of nonlinear mapping. Although its variants have achieved varying degrees of improvement in robustness, sparsity, and adaptability, single SCN models still face challenges in generalization ability and stability in the modeling of furnace temperature prediction in urban solid waste incineration processes.

[0004] Neural network ensembles, by constructing multiple base learner models and effectively fusing them, can comprehensively utilize the advantages of different base models, significantly improving the stability and generalization ability of the overall model. Common methods include: employing a bootstrap ensemble strategy, using an improved SCN as the base model, and constructing an ensemble SCN model through weighted averaging, which has achieved good application results in the field of industrial parameter range prediction; an adaptive SCN ensemble model has been proposed, which improves the accuracy and reliability of the model in structural reliability analysis by using an adaptive sampling strategy. In addition, the improved SCN ensemble method has also been applied to time series prediction, fault diagnosis, and other fields. To address the problem that the homogenization of base models weakens the ensemble effect, neural network ensemble modeling methods based on the negative correlation learning (NCL) strategy have been studied. These methods minimize the correlation of prediction errors between base models by learning the complementary features of the base models. However, the learning objective of the above-mentioned NCL-based ensemble SCN algorithms is to minimize the error of the base models, rather than directly optimizing the global error of the ensemble model, which limits the improvement of model accuracy.

[0005] Furthermore, traditional furnace temperature prediction modeling methods typically focus only on the average flue gas temperature of one combustion chamber, neglecting the coupling relationships between furnace temperatures in multiple regions. It is worth noting that leveraging information shared by relevant objective parameters can improve parameter estimation and enhance the model's generalization performance. A multi-objective SCN modeling method has been proposed, utilizing the structure matrix... L 2,1 Norms reveal the correlations among multiple objectives, and an alternating optimization algorithm is used to solve for the model's output weights. In addition, a novel multi-objective SCN model addresses this from the perspective of network structure design and sparse constraints on parameter sets, achieving collaborative constraints on output weights by constructing a generalized sparse matrix elastic network. While the above methods can achieve rapid learning of multi-objective parameters, further exploration is needed for multi-objective modeling strategies in neural network ensemble models. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration processes, which solves the problems of low accuracy and stability in existing furnace temperature prediction methods.

[0007] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration processes, comprising:

[0008] S1: Preprocess the data of urban solid waste incineration process and obtain S training datasets through a self-sampling strategy;

[0009] S2: Use the randomized network algorithm to construct the base model and obtain the base model corresponding to each training dataset;

[0010] S3: Using a global negative correlation-based ensemble strategy, the ensemble model of S base models is trained in parallel. Group sparse regularization technology is introduced to enable the ensemble model to learn the furnace temperature of multiple regions, and a well-trained ensemble model is obtained.

[0011] S4: Analyze the data of the urban solid waste incineration process using the trained ensemble model to obtain the furnace temperature prediction results and complete the prediction of furnace temperature.

[0012] The beneficial effects of this invention are as follows: This invention provides a multi-objective ensemble stochastic configuration network (SCN) method for predicting furnace temperature in urban solid waste incineration processes. By initializing the base model and performing multi-objective ensemble processing, an ensemble model with a stochastic configuration network is obtained. A multi-objective ensemble algorithm based on a global network control language is used to train the ensemble model, resulting in a trained ensemble model. This improves the accuracy and stability of multi-region furnace temperature prediction. Specifically, a self-adoption strategy is used to sample incineration process operation data to construct a diverse dataset for neural network ensemble, and the base model is initialized using SCN. A global negative correlation learning strategy is used to synchronously train the base model, ensuring its accuracy and stability. Furthermore, multi-objective sparse regularization technology is used to optimize the synchronous training algorithm of the ensemble model, and an alternating optimization algorithm is used to solve the non-smoothness problem in solving the output weights of the ensemble model, improving the model's ability to learn the correlation of multi-objective parameters in industrial processes. Experiments using real operational data from urban solid waste incineration processes demonstrate that the proposed method has good accuracy and stability.

[0013] Furthermore, the expression for the base model ensemble model is:

[0014] ;

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] in, The output of the base model ensemble model is represented. Indicates the number of base models. This represents the output of the s-th base model. This represents the hidden layer output matrix. This represents the hidden layer output weight matrix of the s-th base model. Represent the identity. This represents the output weight matrix of the first hidden layer of the s-th base model. This represents the output weight matrix of the Lth hidden layer of the s-th base model. Let s represent the s-th training dataset. This represents the input sample of the s-th training dataset. This represents the output sample of the s-th training dataset. This represents the nth input sample in the s-th training dataset. This represents the nth output sample from the s-th training dataset, where N represents the number of samples. This represents the preprocessed training dataset. Indicates the input sample. This indicates the output sample.

[0020] Furthermore, the expression for the output weights of the ensemble model is:

[0021] ;

[0022] in, This represents the output weight matrix after the first base model is updated. This represents the updated output weight matrix of the S-th base model. This indicates the minimum value of the loss function. , The loss function of a multi-objective ensemble model. Represents the base model ensemble model. This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of base models, where s represents the base model index and S represents the number of base models.

[0023] Furthermore, the expression for the loss function of the ensemble model is:

[0024] ;

[0025] in, The loss function of a multi-objective ensemble model. Represents the base model ensemble model. C This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of the base model. Indicates the number of base models. This represents the regularization parameter used to control the effect of the regularization term. This represents the regularization parameter used for sparsity in group sparsity regularization. Indicates the first k The output weight matrix of each base model express L 2,1 Norm calculation This indicates the calculation of the F-norm. s Indicates the base model number. k The base model requires, Indicates the first k The output of each base model Indicates except the first k The difference between the output of the integrated model other than the base model and the output sample. Indicates except the first k The output of the integration of all models except the base model. Indicates the output of the s-th base model and The difference, Represents all independent loss functions The calculation item. Attached Figure Description

[0026] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein:

[0027] Figure 1 This is an exemplary flowchart illustrating a multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration processes, according to some embodiments of this specification. Detailed Implementation

[0028] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0029] Example

[0030] Figure 1 This is an exemplary flowchart illustrating a multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration processes, according to some embodiments of this specification. Figure 1 As shown, the process includes the following steps. In some embodiments, the process may be executed by a processor.

[0031] S1: Preprocess the data of urban solid waste incineration process and obtain S training datasets through a self-sampling strategy.

[0032] Data on the urban solid waste incineration process are monitoring data of core control parameters during the urban solid waste incineration process.

[0033] The training dataset is the dataset used to build the base model.

[0034] S2: Use the randomized network algorithm to construct the base model and obtain the base model corresponding to each training dataset.

[0035] The base model is the self-healing network architecture (SCN) model.

[0036] In some embodiments, the processor can utilize the training dataset, employ a bootstrapping technique to sample the training set data, and rapidly construct a base model using the SC-III algorithm.

[0037] S3: Using a global negative correlation-based ensemble strategy, the ensemble model of S base models is trained in parallel. Group sparse regularization technology is introduced to enable the ensemble model to learn the furnace temperature of multiple regions, and a well-trained ensemble model is obtained.

[0038] The base model ensemble model is an ensemble model with a randomly configured network whose number of hidden layer neurons in the base model is determined after initialization.

[0039] In some embodiments, the processor can use a validation set and a trial-and-error method to determine the number of hidden layer neurons in the base model to ensure the generalization ability of the base model.

[0040] The output of the ensemble stochastic model network is calculated using a weighted approach, primarily for the sake of simplicity in subsequent derivations. Further optimization of the weights of each base model, such as evolutionary algorithms, can positively impact the accuracy of the algorithm.

[0041] In some embodiments, the expression for the base model ensemble model is:

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] in, The output of the base model ensemble model is represented. Indicates the number of base models. This represents the output of the s-th base model. This represents the hidden layer output matrix. This represents the hidden layer output weight matrix of the s-th base model. Represent the identity. This represents the output weight matrix of the first hidden layer of the s-th base model. This represents the output weight matrix of the Lth hidden layer of the s-th base model. Let s represent the s-th training dataset. This represents the input sample of the s-th training dataset. This represents the output sample of the s-th training dataset. This represents the nth input sample in the s-th training dataset. This represents the nth output sample from the s-th training dataset, where N represents the number of samples. This represents the preprocessed training dataset. Indicates the input sample. This indicates the output sample.

[0048] In some embodiments, based on the bias-variance-covariance decomposition theorem, a global NCL architecture is used to optimize the output weights of each SCN base model, while incorporating group sparse regularization to enhance the ensemble model's ability to learn multi-objective parameters of industrial processes. This can further improve the generalization ability and stability of the ensemble SCN.

[0049] In some embodiments, the expression for the output weights of the trained ensemble model is:

[0050] ;

[0051] in, This represents the output weight matrix after the first base model is updated. This represents the updated output weight matrix of the S-th base model. This indicates the minimum value of the loss function. , The loss function of a multi-objective ensemble model. Represents the base model ensemble model. This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of base models, where s represents the base model index and S represents the number of base models.

[0052] In some embodiments, the expression for the loss function of the ensemble model is:

[0053] ;

[0054] in, The loss function of a multi-objective ensemble model. Represents the base model ensemble model. C This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of the base model. Indicates the number of base models. This represents the regularization parameter used to control the effect of the regularization term. This represents the regularization parameter used for sparsity in group sparsity regularization. Indicates the first k The output weight matrix of each base model express L 2,1 Norm calculation This indicates the calculation of the F-norm. s Indicates the base model number. k The base model requires, Indicates the first k The output of each base model Indicates except the first k The difference between the output of the integrated model other than the base model and the output sample. Indicates except the first k The output of the integration of all models except the base model. Indicates the output of the s-th base model and The difference, Represents all independent loss functions The calculation item.

[0055] In some embodiments, the derivation process of the loss function of the ensemble model is as follows:

[0056] For the initialized SCN base model, with fixed input weights and biases in the hidden layers, the output weights of the ensemble model can be calculated using the above formula based on the global NCL architecture:

[0057] ;

[0058] Among them, the loss function of the integrated SCN model It can be represented as:

[0059] ;

[0060] Compared to the traditional NCL algorithm, the global NCL ensemble architecture focuses on optimizing the global ensemble rather than the individual fitness of the base models, thus reducing the fitting error term. Represented as:

[0061] ;

[0062] In addition, diversity constraints It can be represented as:

[0063] ;

[0064] Regularization term Using group sparse regularization

[24] , it can be expressed as:

[0065]

[0066] Here, α>0 is a balance parameter used to adjust the contribution of the two regularization methods.

[0067] To facilitate subsequent solutions, the above equation needs to be decomposed into two parts to isolate the first part. k Parameters of an SCN machine model k =1,2, …, S. Specifically, define... This indicates that the ensemble model is missing the first... k The output for a base model is shown below:

[0068] ;

[0069] The fitting error term can be decomposed into:

[0070] ;

[0071] The diversity constraint term is decomposed into:

[0072] ;

[0073] The first term of equation (19) can be rewritten as:

[0074] ;

[0075] The second term can be represented as:

[0076] ;

[0077] Therefore, the diversity constraint term can be expressed as:

[0078] ;

[0079] in, Independent of .

[0080] The regularization term can be decomposed into:

[0081] ;

[0082] Therefore, the global loss function of the multi-objective ensemble SCN model can be expressed as:

[0083] ;

[0084] in and All independent of Furthermore, due to The formula can be further expressed as:

[0085] ;

[0086] Therefore, when the parameters α, C, and λ satisfy the following constraints, the first k The analytical solution for the output weights of each SCN basis model can be obtained by minimizing the formula.

[0087] ;

[0088] The minimum value can be obtained by letting get:

[0089] ;

[0090] in,

[0091] ;

[0092] in, Indicates the first k The Lth row of output weights from the SCN basis model. Since the above equation has no analytical solution, an alternating optimization method is used to solve it here. For simplicity, the following two parameters are defined:

[0093] ;

[0094] Therefore, combining the above equations, we get:

[0095] ;

[0096] ;

[0097] For s = 1,2,…, S, the above equation can be expressed in linear matrix form as follows:

[0098] ;

[0099] in,

[0100] ;

[0101] ;

[0102] Where t represents the number of iterations, and I represents the identity matrix. express, express, express, Represents the regularization term, Represents the s-th base model. Represents the fitted term, Representing diversity terms, This represents the output samples in the training set. Indicates the first k The output weights corresponding to the first hidden layer neuron in the base model's input weight matrix. Indicates the first k The output weights corresponding to the Lth hidden layer neuron in the base model's input weight matrix. This indicates a custom parameter. This indicates a custom parameter. This indicates a custom parameter. Indicates the first k Hidden layer outputs of each base model This indicates a custom parameter. This represents the hidden layer output matrix of the first base model. Let represent the hidden layer output matrix of the s-th base model. This represents the hidden layer output matrix of the S-th base model. This represents the output weight of the S-th base model after the (t+1)-th update.

[0103] S5: Analyze the data of urban solid waste incineration process using the trained ensemble model to obtain furnace temperature prediction results and complete the prediction of furnace temperature.

[0104] The furnace temperature prediction results are the prediction results of the furnace temperature during the incineration of urban solid waste.

[0105] A self-adoption strategy is employed to sample incineration process data, constructing a diverse dataset for neural network ensembles. The base model is initialized using SCN (Self-Directed Learning Network). A global negative correlation learning strategy is used for synchronous training of the base model, ensuring its accuracy and stability. Furthermore, multi-objective sparse regularization is used to optimize the synchronous training algorithm of the ensemble model, and an alternating optimization algorithm is employed to address the non-smoothness problem in solving the output weights of the ensemble model, enhancing the model's ability to learn the correlations of multi-objective parameters in industrial processes. Experiments using real-world operational data from urban solid waste incineration processes demonstrate that the proposed method exhibits good accuracy and stability.

[0106] In some embodiments, the experimental data that the processor can use originates from a solid waste incineration power plant and includes two datasets, DB1 and DB2, described in Table 1. The datasets contain 42 inputs and 3 furnace temperature outputs. The test RMSEs on DB1 and DB2 are 0.0818±0.0015 and 0.1716±0.0029, respectively; the test MAEs are 0.0640±0.0012 and 0.1339±0.0023, respectively; and the R² values ​​are 0.9900±0.0004 and 0.9652±0.0012, respectively.

[0107] Table 1 Dataset Description Table

[0108]

[0109] In some embodiments of this specification, a multi-objective ensemble stochastic configuration network (SCN) method for predicting furnace temperature in urban solid waste incineration processes is provided. This method involves initializing a base model and performing multi-objective ensemble processing to obtain an ensemble model with a stochastic configuration network. A multi-objective ensemble algorithm based on a global network control language is then used to train the ensemble model, resulting in a trained ensemble model. This improves the accuracy and stability of furnace temperature prediction in multiple regions. Specifically, a self-adoption strategy is used to sample incineration process operation data to construct a diverse dataset for neural network ensemble, and the base model is initialized using a SCN. A global negative correlation learning strategy is used to synchronously train the base model, ensuring its accuracy and stability. Furthermore, multi-objective sparse regularization is employed to optimize the synchronous training algorithm of the ensemble model, and an alternating optimization algorithm is used to address the non-smoothness problem in solving the output weights of the ensemble model, improving the model's ability to learn the correlation of multi-objective parameters in the industrial process. Experiments using real operational data from urban solid waste incineration processes demonstrate that the proposed method exhibits good accuracy and stability.

Claims

1. A multi-objective integrated stochastic configuration network method for predicting furnace temperature in urban solid waste incineration processes, characterized in that, include: S1: Preprocess the data of urban solid waste incineration process and obtain S training datasets through a self-sampling strategy; S2: Use the randomized network algorithm to construct the base model and obtain the base model corresponding to each training dataset; S3: Using a global negative correlation-based ensemble strategy, the ensemble model of S base models is trained in parallel. Group sparse regularization technology is introduced to enable the ensemble model to learn the furnace temperature of multiple regions, and a well-trained ensemble model is obtained. The expression for the loss function of the ensemble model is: ； in, The loss function of a multi-objective ensemble model. Represents the base model ensemble model. C This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of the base model. Indicates the number of base models. This represents the parameter used to control the effect of the regularization term. This represents the parameter used for sparsity in group sparsity regularization. Indicates the first k The output weight matrix of each base model express L 2,1 Norm calculation This indicates the calculation of the F-norm. s Indicates the base model number. k Indicates the base model number. Indicates the first k The output of each base model Indicates except the first k The difference between the output of the integrated model other than the base model and the output sample. Indicates except the first k The output of the integration of all models except the base model. Indicates the output of the s-th base model and The difference, Represents all independent loss functions Calculation items; S4: Analyze the data of the urban solid waste incineration process using the trained ensemble model to obtain the furnace temperature prediction results and complete the prediction of furnace temperature.

2. The multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration process according to claim 1, characterized in that, The expression for the base model ensemble model is: ； ； ； ； ； in, The output of the base model ensemble model is represented. Indicates the number of base models. This represents the output of the s-th base model. This represents the hidden layer output matrix. This represents the hidden layer output weight matrix of the s-th base model. Represent the identity. This represents the output weight matrix of the first hidden layer of the s-th base model. This represents the output weight matrix of the Lth hidden layer of the s-th base model. Let s represent the s-th training dataset. This represents the input sample of the s-th training dataset. This represents the output sample of the s-th training dataset. This represents the nth input sample in the s-th training dataset. This represents the nth output sample from the s-th training dataset, where N represents the number of samples. This represents the preprocessed training dataset. Indicates the input sample. This indicates the output sample.

3. The multi-objective integrated stochastic configuration network furnace temperature prediction method for urban solid waste incineration process according to claim 1, characterized in that, The expression for the output weights of the ensemble model is: ； in, This represents the output weight matrix after the first base model is updated. This represents the updated output weight matrix of the S-th base model. This indicates the minimum value of the loss function. , The loss function of a multi-objective ensemble model. Represents the base model ensemble model. This represents the regularization parameter used to control the effect of the fitted term. This represents the regularization parameter used to control the diversity of base models, where s represents the base model index and S represents the number of base models.

Images