Hearth temperature robust forecasting method in urban solid waste incineration process
By using the skewed t distribution to simulate the prior distribution of outliers during urban solid waste incineration, and iteratively optimize the output weight and hyperparameters of the model, the problem of inaccurate furnace temperature forecast is solved, and the asymmetric outliers are achieved Robust processing and real-time accurate forecasting.
Patent Information
- Application Number
- CN202411065150.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-05
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-05
AI Technical Summary
The furnace temperature forecast during the incineration of urban solid waste is inaccurate, and there are asymmetric outliers, which are difficult to effectively deal with in the prior art.
The skewed t distribution is used to simulate the prior distribution of outliers, and iteratively optimize the output weights and hyperparameters of the model through maximum likelihood estimation and the expected condition maximization algorithm to establish a robust furnace temperature forecast model.
The robustness of the furnace temperature forecast model for abnormal data is improved, and real-time and accurate forecast of the furnace temperature during urban solid waste incineration is achieved.
Smart Images

Figure CN119026466A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of furnace temperature prediction, and in particular relates to a method for robustly predicting furnace temperature in a municipal solid waste incineration process. Background Art
[0002] Urban solid waste refers to solid waste generated in daily life or in activities that provide services for daily life. However, due to the characteristics of strong coupling, large lag, and large inertia in the incineration process, it is difficult to accurately predict the changes in furnace temperature by relying solely on on-site temperature thermocouples. Therefore, it is necessary to build a data-driven furnace temperature prediction model to achieve accurate prediction and forecasting of furnace temperature changes. In addition, in terms of data collection, data recording, and data storage, the historical operating data of the incineration process may contain outliers with unknown distributions, and the distribution of such outliers is usually asymmetric, so the furnace temperature prediction method is required to be robust to such outliers. Summary of the invention
[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for robustly predicting furnace temperature in a municipal solid waste incineration process, which solves the problem of inaccurate prediction of furnace temperature in a municipal solid waste incineration process in the prior art.
[0004] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for robustly predicting furnace temperature in a municipal solid waste incineration process, comprising the following steps:
[0005] S1. Establishing an initial furnace temperature model according to a preset training set;
[0006] S2. Assuming that the prior distribution of noise in the training data follows a skewed t distribution, the output weights and hyperparameters of the model are iteratively optimized by the maximum likelihood estimation and expectation conditional maximization algorithm to establish a furnace temperature prediction model;
[0007] S3. Predict the furnace temperature during the incineration process through the furnace temperature prediction model.
[0008] Further: In S1, the training set X and y are training samples;
[0009] The expression of the initial furnace temperature model is as follows:
[0010]
[0011] In the formula, is the output of the noise included in the initial furnace temperature model, x n is the input of the initial furnace temperature model, x n ∈X,y n is the output of the initial furnace temperature model, y n ∈y,h(xn ) is the model hidden layer output matrix, β is the output weight matrix, ε n is the noise contained in the nth sample, and N is the number of training samples.
[0012] Further: S2 comprises the following sub-steps:
[0013] S21, assuming that the prior distribution of noise in the training data follows a skewed t distribution, introducing latent variables, and using a hierarchical method to represent the output probability density function of the initial furnace temperature model;
[0014] S22. Based on the output probability density function of the initial furnace temperature model and according to Bayesian theorem, a posterior distribution of the output weight of the initial furnace temperature model under the condition of a given complete data set is established;
[0015] S23, performing a logarithmic operation on the posterior distribution of the output weight of the furnace temperature model, and establishing a general calculation formula for the output weight of the model and a set of hyperparameters according to the maximum a posteriori estimation algorithm;
[0016] S24. The calculation formula of the output weights and hyperparameters of the model is solved by the ECM algorithm (Expectation Conditional Maximization), and the output weights and hyperparameters of the model are iteratively updated. When the expected rate of change meets the termination condition, a furnace temperature prediction model is established according to the output weights and hyperparameters of the final model.
[0017] Further: In S21, assuming that the prior distribution of noise in the training data follows a skewed t distribution, the expression of the probability density function p(ε) of the noise is specifically as follows:
[0018] p(ε)=ST(0,σ 2 ,λ,v)
[0019] In the formula, σ 2 is the scale parameter, σ 2 ∈(0,+∞), λ is the skew parameter, λ∈R, R is a real number, v is the degree of freedom parameter, v∈(0,+∞).
[0020] Further: In S21, the probability density function p(y n ;x n ,β,σ 2 ,λ,v) is as follows:
[0021]
[0022] In the formula, γ n and τ nis a latent variable, v∈(0,+∞), δ λ is the first intermediate parameter, The mean is h(x n )β+δ λ γ n The standard deviation is The normal distribution of is a truncated normal distribution with a truncation interval of (0,∞), and Γ(·,·) is a gamma distribution.
[0023] Further: In S22, the posterior distribution p(β; T, σ 2 ,λ,v) is expressed as follows:
[0024]
[0025] In the formula, ST(h(x n )β,σ 2 ,λ,v) is the position parameter h(x n )β, the scale parameter is σ 2 , a skewed t distribution with skew parameter λ and v degrees of freedom, is the prior distribution of the output weight, and its specific expression is as follows:
[0026]
[0027] Where L is the number of neurons in the hidden layer of the furnace temperature model, is the variance of the prior distribution of the output weights.
[0028] Further: In S23, the set of output weights and hyperparameters of the model The calculation formula is as follows:
[0029]
[0030] In the formula, lnp(β;T,σ 2 ,λ,v) is the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model, To find the parameters β,σ corresponding to the maximum value of the function 2 ,λ,v,
[0031] Further: In S24, the set of output weights and hyperparameters of the model at the kth iteration The specific method for iterative updating is:
[0032] SA1. Calculate the expected function of the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model;
[0033] SA2. Based on the given output weights and hyperparameters of the model at the k-th iteration, update the output weights and hyperparameters of the model at the k-th iteration according to the expected function.
[0034] Further: In SA1, the expected function The specific expression is as follows:
[0035]
[0036] In the formula, c3 is a constant, is the second intermediate parameter, and its specific expression is:
[0037]
[0038]
[0039] in,
[0040] Where DG(·) represents the digamm function;
[0041] In SA2, update and The specific expression is as follows:
[0042]
[0043] In the formula, h T (x n ) is the transpose of the model hidden layer output matrix, is the output weight matrix at the k+1th iteration, is the variance of the prior distribution at the k+1th iteration;
[0044] renew The specific expression is as follows:
[0045]
[0046] In the formula, is the scale parameter at the k+1th iteration;
[0047] renew The specific method is: and Calculated by the following formula The value of is used as the skew parameter for the k+1th iteration
[0048]
[0049] In the formula, is the scale parameter at the k+1th iteration;
[0050] renew The specific method is: and Calculated by the following formula The value of is used as the degree of freedom parameter for the k+1th iteration
[0051] Further: In S24, the expression of the expected change rate satisfying the termination condition is specifically:
[0052]
[0053] Here, κ is a positive number approaching 0.
[0054] The beneficial effects of the present invention are as follows: the present invention provides a method for robust prediction of furnace temperature in a municipal solid waste incineration process, by acquiring process data of the municipal solid waste incineration process in real time, establishing a training set for the collected process data, and establishing a furnace temperature prediction model based on a robust random configuration network, simulating the prior distribution of asymmetric outliers in the incineration process operation data through a skewed t distribution with a heavy-tailed characteristic, and solving the output weights of the furnace temperature prediction model with the help of the maximum likelihood estimation method, thereby improving the robustness of the furnace temperature prediction model to abnormal data, and realizing real-time and accurate prediction of the furnace temperature in the municipal solid waste incineration process. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 The present invention is a flow chart of a method for robustly predicting furnace temperature in a municipal solid waste incineration process. DETAILED DESCRIPTION
[0056] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0057] like Figure 1 As shown, in one embodiment of the present invention, a method for robustly predicting furnace temperature in a municipal solid waste incineration process comprises the following steps:
[0058] S1. Establishing an initial furnace temperature model according to a preset training set;
[0059] S2. Assuming that the prior distribution of noise in the training data follows a skewed t distribution, the output weights and hyperparameters of the model are iteratively optimized by the maximum likelihood estimation and expectation conditional maximization algorithm to establish a furnace temperature prediction model;
[0060] S3. Predict the furnace temperature during the incineration process through the furnace temperature prediction model.
[0061] In order to comprehensively improve the robustness of the furnace temperature prediction model in the incineration process to asymmetric outliers and the adaptability to fluctuations in solid waste components, a robust SCN modeling method is proposed to establish the furnace temperature prediction model. The SC-III algorithm is used to initialize the model. It is assumed that the prior distribution of the abnormal data in the training data follows the skewed t distribution hypothesis, and the maximum likelihood estimation and expectation conditional maximization algorithm are used to iteratively optimize the hyperparameters of the model.
[0062] In S1, for a given training set X and y are training samples;
[0063] The expression of the initial furnace temperature model is as follows:
[0064]
[0065] In the formula, is the output of the noise included in the initial furnace temperature model, x n is the input of the initial furnace temperature model, x n ∈X,y n is the output of the initial furnace temperature model, y n ∈y,h(x n ) is the model hidden layer output matrix, β is the output weight matrix, ε n is the noise contained in the nth sample, and N is the number of training samples.
[0066] The S2 comprises the following sub-steps:
[0067] S21, assuming that the prior distribution of noise in the training data follows a skewed t distribution, introducing latent variables, and using a hierarchical method to represent the output probability density function of the initial furnace temperature model;
[0068] S22. Based on the output probability density function of the initial furnace temperature model and according to Bayesian theorem, a posterior distribution of the output weight of the initial furnace temperature model under the condition of a given complete data set is established;
[0069] S23, performing a logarithmic operation on the posterior distribution of the output weight of the furnace temperature model, and establishing a general calculation formula for the output weight of the model and a set of hyperparameters according to the maximum a posteriori estimation algorithm;
[0070] S24. Solve the calculation formula of the output weights and hyperparameters of the model through the ECM algorithm, and iteratively update the output weights and hyperparameters of the model. When the expected rate of change meets the termination condition, establish a furnace temperature prediction model based on the final output weights and hyperparameters of the model.
[0071] In this embodiment, since the distribution of outliers or noise in the operation data of complex industrial processes is usually unbalanced, the present invention adopts a skewed t distribution with a heavy-tail characteristic as the prior distribution of noise. In S21, it is assumed that the prior distribution of noise in the training data follows the skewed t distribution, and the expression of the probability density function p(ε) of the noise is specifically as follows:
[0072] p(ε)=ST(0,σ 2 ,λ,v)
[0073] In the formula, σ 2 is the scale parameter, σ 2 ∈(0,+∞), λ is the skew parameter, λ∈R, R is a real number, v is the degree of freedom parameter, v∈(0,+∞).
[0074] The output probability density function of the original initial furnace temperature model is expressed as:
[0075] p(y n ;x n ,β,σ 2 ,λ,v)=ST(h(x n )β,σ 2 ,λ,v)
[0076] In order to facilitate subsequent calculations, the present invention introduces latent variables to obtain a complete data set T = {X, y, γ, τ} and establish the probability density function of the output of the initial furnace temperature model. In S21, the probability density function p(y n ;x n ,β,σ 2 ,λ,v) is as follows:
[0077]
[0078] In the formula, γ n and τ n is a latent variable, v∈(0,+∞), δ λ is the first intermediate parameter, The mean is h(x n )β+δ λ γ n The standard deviation is The normal distribution of is a truncated normal distribution with a truncation interval of (0,∞), and Γ(·,·) is a gamma distribution.
[0079] In S22, the posterior distribution p(β; T,σ 2 ,λ,v) is expressed as follows:
[0080]
[0081] In the formula, ST(h(x n )β,σ 2 ,λ,v) is the position parameter h(x n )β, the scale parameter is σ 2 , a skewed t distribution with skew parameter λ and v degrees of freedom, is the prior distribution of the output weight, and its specific expression is as follows:
[0082]
[0083] Where L is the number of neurons in the hidden layer of the furnace temperature model, is the variance of the prior distribution of the output weights.
[0084] In S23, the output weights and hyperparameters of the model are The calculation formula is as follows:
[0085]
[0086] In the formula, lnp(β;T,σ 2 ,λ,v) is the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model, To find the parameters β,σ corresponding to the maximum value of the function 2 ,λ,v,
[0087] Since the output weights and the set of hyperparameters of the model are unknown, the above formula has no analytical solution. Therefore, the present invention adopts the ECM algorithm to solve the output weights and the set of hyperparameters of the model.
[0088] In this embodiment, the expected function is obtained by performing a logarithmic operation on the posterior distribution of the output weights of the initial furnace temperature model. The specific operation is as follows:
[0089]
[0090] In the formula, c1 and c2 are constants, η n =(y n -h(x n )β) / σ.
[0091] In S24, the set of output weights and hyperparameters of the model at the kth iteration is The specific method for iterative updating is:
[0092] SA1. Calculate the expected function of the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model;
[0093] SA2. Based on the given output weights and hyperparameters of the model at the k-th iteration, update the output weights and hyperparameters of the model at the k-th iteration according to the expected function.
[0094] In SA1, the expected function The specific expression is as follows:
[0095]
[0096] In the formula, c3 is a constant, is the second intermediate parameter, and its specific expression is:
[0097]
[0098]
[0099] in,
[0100] Where DG(·) represents the digamm function;
[0101] In SA2, update and The specific expression is as follows:
[0102]
[0103]
[0104] In the formula, h T (x n ) is the transpose of the model hidden layer output matrix, is the output weight matrix at the k+1th iteration, is the variance of the prior distribution at the k+1th iteration;
[0105] In this embodiment, the above formula is obtained by maximizing the expected function with respect to β and Establish.
[0106] renew The specific expression is as follows:
[0107]
[0108] In the formula, is the scale parameter at the k+1th iteration;
[0109] In this embodiment, the above formula is obtained by maximizing the expected function with respect to σ 2 Establish.
[0110] renew The specific method is: Calculated by the following formula The value of is used as the skew parameter for the k+1th iteration
[0111]
[0112] In the formula, is the scale parameter at the k+1th iteration, and the above formula is established by maximizing the expectation function with respect to λ.
[0113] renew The specific method is: and Calculated by the following formula The value of is used as the degree of freedom parameter for the k+1th iteration
[0114] In this embodiment, the above formula is established with respect to v by maximizing the expectation function.
[0115] In S24, the expression for the expected change rate to satisfy the termination condition is specifically:
[0116]
[0117] Wherein, κ is a positive number close to 0, and in this embodiment, the value is 10 -4 .
[0118] In this embodiment, the maximum number of hidden layer neurons of the furnace temperature prediction model is 50, the maximum number of model configurations is 200, the skew parameter of the skew t distribution is 0.1, the scale parameter is 0.1, the degree of freedom is 5, and the prior variance of the output weight is 0.15.
[0119] The furnace temperature prediction model has 600 training data and 200 test data. The number of input feature variables is 49 and the output variable is 1. After 50 experiments, the experimental results are given in the form of mean ± standard deviation, as follows:
[0120] The test RMSE is 4.91±0.19℃
[0121] The test MAE is 3.91±0.15℃;
[0122] The test R2 coefficient was 92.64±0.56.
[0123] It can be seen from this that the furnace temperature prediction model designed in the present invention can accurately calculate the furnace temperature.
[0124] The beneficial effects of the present invention are as follows: the present invention provides a method for robust prediction of furnace temperature in a municipal solid waste incineration process, by acquiring process data of the municipal solid waste incineration process in real time, establishing a training set for the collected process data, and establishing a furnace temperature prediction model based on a robust random configuration network, simulating the prior distribution of asymmetric outliers in the incineration process operation data through a skewed t distribution with a heavy-tailed characteristic, and solving the output weights of the furnace temperature prediction model with the help of the maximum likelihood estimation method, thereby improving the robustness of the furnace temperature prediction model to abnormal data, and realizing real-time and accurate prediction of the furnace temperature in the municipal solid waste incineration process.
[0125] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.
Claims
1. A robust prediction method for furnace temperature in municipal solid waste incineration process, characterized in that: The following steps are involved: S1. Establishing an initial furnace temperature model according to a preset training set; S2. Assuming that the prior distribution of noise in the training data follows a skewed t distribution, the output weights and hyperparameters of the model are iteratively optimized by the maximum likelihood estimation and expectation conditional maximization algorithm to establish a furnace temperature prediction model; S3. Predict the furnace temperature during the incineration process through the furnace temperature prediction model.
2. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 1 is characterized in that: In S1, the training set X and y are training samples; The expression of the initial furnace temperature model is as follows: In the formula, is the output of the noise included in the initial furnace temperature model, x n is the input of the initial furnace temperature model, x n ∈X,y n is the output of the initial furnace temperature model, y n ∈y,h(x n ) is the model hidden layer output matrix, β is the output weight matrix, ε n is the noise contained in the nth sample, and N is the number of training samples.
3. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 2 is characterized in that: The S2 comprises the following sub-steps: S21, assuming that the prior distribution of noise in the training data follows a skewed t distribution, introducing latent variables, and using a hierarchical method to represent the output probability density function of the initial furnace temperature model; S22. Based on the output probability density function of the initial furnace temperature model and according to Bayesian theorem, a posterior distribution of the output weight of the initial furnace temperature model under the condition of a given complete data set is established; S23, performing a logarithmic operation on the posterior distribution of the output weight of the furnace temperature model, and establishing a general calculation formula for the output weight of the model and a set of hyperparameters according to the maximum a posteriori estimation algorithm; S24. Solve the calculation formula of the output weights and hyperparameters of the model through the ECM algorithm, and iteratively update the output weights and hyperparameters of the model. When the expected rate of change meets the termination condition, establish a furnace temperature prediction model based on the final output weights and hyperparameters of the model.
4. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 3 is characterized in that: In S21, it is assumed that the prior distribution of the noise in the training data follows a skewed t distribution, and the expression of the probability density function p(ε) of the noise is specifically as follows: p(ε)=ST(0,σ 2 ,λ,v) In the formula, σ 2 is the scale parameter, σ 2 ∈(0,+∞), λ is the skew parameter, λ∈R, R is a real number, v is the degree of freedom parameter, v∈(0,+∞).
5. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 4 is characterized in that: In S21, the probability density function p(y n ;x n ,β,σ 2 ,λ,v) is as follows: In the formula, γ n and τ n is a latent variable, v∈(0,+∞), δ λ is the first intermediate parameter, The mean is h(x n )β+δ λ γ n The standard deviation is The normal distribution of is a truncated normal distribution with a truncation interval of (0,∞), and Γ(·,·) is a gamma distribution.
6. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 5 is characterized in that: In S22, the posterior distribution p(β; T,σ 2 ,λ,v) is expressed as follows: In the formula, ST(h(x n )β,σ 2 ,λ,v) is the position parameter h(x n )β, the scale parameter is σ 2 , a skewed t distribution with skew parameter λ and v degrees of freedom, is the prior distribution of the output weight, and its specific expression is as follows: Where L is the number of neurons in the hidden layer of the furnace temperature model, is the variance of the prior distribution of the output weights.
7. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 6 is characterized in that: In S23, the output weights and hyperparameters of the model are The calculation formula is as follows: In the formula, lnp(β;T,σ 2 ,λ,v) is the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model, To find the parameters β,σ corresponding to the maximum value of the function 2 ,λ,v, 8. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 7 is characterized in that:In S24, the set of output weights and hyperparameters of the model at the kth iteration is The specific method for iterative updating is: SA1. Calculate the expected function of the logarithmic operation result of the posterior distribution of the output weight of the furnace temperature model; SA2. Based on the given output weights and hyperparameters of the model at the k-th iteration, update the output weights and hyperparameters of the model at the k-th iteration according to the expected function.
9. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 8 is characterized in that: In SA1, the expected function The specific expression is as follows: In the formula, c3 is a constant, is the second intermediate parameter, and its specific expression is: in, Where DG(·) represents the digamm function; In SA2, update and The specific expression is as follows: In the formula, h T (x n ) is the transpose of the model hidden layer output matrix, is the output weight matrix at the k+1th iteration, is the variance of the prior distribution at the k+1th iteration; renew The specific expression is as follows: In the formula, is the scale parameter at the k+1th iteration; renew The specific method is: and Calculated by the following formula The value of is used as the skew parameter for the k+1th iteration In the formula, is the scale parameter at the k+1th iteration; renew The specific method is: and Calculated by the following formula The value of is used as the degree of freedom parameter for the k+1th iteration 10. The method for robust prediction of furnace temperature in the municipal solid waste incineration process according to claim 9, characterized in that: In S24, the expression for the expected change rate to satisfy the termination condition is specifically: Here, κ is a positive number approaching 0.
Citation Information
Patent Citations
Method for forecasting flue gas temperature of primary combustion chamber in municipal solid waste incineration process
CN113191078A
Coking diagnosis method based on Bayesian t distribution mixed regression model
CN117216723A
Variable grid sparse Bayesian direction-of-arrival estimation method based on noise integration
CN118036759A
Furnace temperature interval prediction method and system in solid waste incineration process
CN118171238A
Refuse burning state detecting method using fire grate temperature, and refuse incineration control method and fire grate temperature control method using it in stoker type refuse incinerator
JP2007010258A
Cited By
Industrial hearth temperature prediction system based on reasoning human learning optimization algorithm
CN119416896A
Industrial furnace temperature prediction system based on inference human learning optimization algorithm
CN119416896B
