Coal-fired unit complex working condition superheater wall temperature prediction method based on multi-model fusion

Through the integrated wall temperature prediction method of multi-model fusion and attention mechanism, the problem of poor generalization ability of the existing model under complex operating conditions is solved, and the accurate prediction and risk control of the wall temperature of coal-fired units is achieved, and the operation is improved. stability and safety.

CN119026459APending Publication Date: 2024-11-26LIYANG RES INST OF SOUTHEAST UNIV +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411036276.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The existing wall temperature prediction model is difficult to reflect dynamic characteristics under complex operating conditions of coal-fired units, and has poor generalization capabilities, so it is unable to effectively deal with the risk of ultra-temperature pipe bursting during peak regulating and frequency regulation.

Method used

Using multi-model fusion method, the data set is divided according to the working conditions through Gaussian mixed model, a single prediction model is established, and the model output is empowered by the multi-head additive attention mechanism, and an integrated wall temperature prediction model is constructed to improve the adaptability of the model. and precision.

Benefits of technology

It realizes accurate prediction of wall temperature under complex working conditions, improves the generalization ability and prediction accuracy of the model, provides strategic guidance for operators, reduces the risk of pipe burst caused by overtemperature, and improves the stability of coal-fired units. and security.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026459A_ABST
    Figure CN119026459A_ABST
Patent Text Reader

Abstract

The invention discloses a coal-fired unit complex working condition superheater wall temperature prediction method based on multi-model fusion, and the method comprises the steps: collecting the related parameters of the wall temperature of a rear screen superheater of a coal-fired unit, a maximum information coefficient (MIC) algorithm and a correlation feature selection (CFS) algorithm are utilized to screen feature variables of related parameters, and a Gaussian mixture model is utilized to divide a data set according to working conditions to form subsets; establishing an independent single prediction model for the subset through a hybrid network; an output result of the single prediction model is weighted by using a multi-head additive attention mechanism, different working conditions are matched to establish an integrated rear platen superheater wall temperature prediction model, and the model is trained; and verifying the performance of the model by using field acquired data. According to the method, an integrated wall temperature prediction model adapting to dynamic characteristics of different working conditions can be constructed, wall temperature prediction under complex working conditions can be realized, reference guidance is provided for operators, and accidents such as pipe explosion caused by overtemperature of a rear screen superheater pipeline are prevented.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of wall temperature prediction of a superheater of a rear panel of a coal-fired unit, and in particular relates to a method for predicting the wall temperature of a superheater of a coal-fired unit under complex operating conditions based on multi-model fusion. Background Art

[0002] The problem of pipe burst caused by long-term over-temperature operation of the heating surface in the boiler of coal-fired units has always been one of the pain points threatening its safe and stable operation. Especially in the context of large-scale access to renewable energy, the frequent peak and frequency regulation of thermal power units has increased the risk of over-temperature and pipe burst of the heating surface. Traditional control methods are difficult to meet real-time needs, and it is urgent to build an efficient and accurate wall temperature prediction model.

[0003] Existing wall temperature prediction models include mechanism modeling and data-driven modeling. Mechanism modeling usually uses numerical simulation methods for prediction, but its calculation is complex, depends on parameter selection and initial condition setting, has low robustness, and is difficult to implement in engineering applications. Data-driven modeling is more efficient than numerical simulation and can predict online. Existing shallow neural networks and deep neural networks such as LSTM and TCN have certain accuracy in prediction, but the establishment of a single model is often difficult to reflect the dynamic characteristics of different operating conditions. The generalization ability of the model is poor and it cannot cope with the complex operating conditions of coal-fired units during peak and frequency regulation. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for predicting the superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion. The data set is accurately divided according to the operating conditions through a Gaussian mixture model and a corresponding single prediction model is established. Finally, the attention mechanism is used to focus on the operating conditions of the data, and the single model is weighted to match different operating conditions, so that the integrated prediction model can adapt to the dynamic characteristics of different operating conditions, improve the generalization ability, and also improve the prediction accuracy, providing guidance and reference for operators to take strategic adjustments to deal with overtemperature.

[0005] Technical solution: A method for predicting the superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion of the present invention comprises the following steps:

[0006] Step 1: Collect relevant parameters of the wall temperature of the rear screen superheater of the coal-fired unit, and use the maximum information coefficient algorithm MIC and the correlation feature selection algorithm CFS to screen the characteristic variables of the relevant parameters;

[0007] Step 2: Use the Gaussian mixture model to divide the data set according to the working conditions to form subsets;

[0008] Step 3: Establish an independent single prediction model for the subset through the PA-TCN-LSTM network;

[0009] Step 4: Use the multi-head additive attention mechanism to weight the output results of the single prediction model, match different working conditions to establish an integrated rear-panel superheater wall temperature prediction model, and train the model;

[0010] Step 5: Based on the trained integrated rear-screen superheater wall temperature prediction model, the model performance is verified using field collected data to predict the superheater wall temperature of the coal-fired unit under complex operating conditions.

[0011] Furthermore, in step 1, the relevant parameters of the wall temperature of the rear screen superheater of the coal-fired unit include unit load, fuel calorific value, primary air volume, temperature and pressure, secondary air volume, temperature and pressure, cooling water flow temperature and pressure at each level, total air volume, feed water flow temperature and pressure, flue gas damper opening, outlet flue gas oxygen content, main steam flow temperature and pressure.

[0012] Furthermore, step 1 is specifically as follows:

[0013] The maximum information coefficient MIC is used to analyze the correlation between the selected parameters and the target variable, and highly correlated variables are preliminarily screened out according to the correlation. Then, the correlation between the screened input variables with high correlation with the target variable is analyzed, and the variables x and y are set. The MIC formula is:

[0014]

[0015]

[0016] Where m and n are the interval sizes of the grid G, p(x,y) is the joint probability density function, p(x), p(y) are the marginal probability density functions, and I m×n (x, y) is the maximum information number under the grid G, Mic is the normalized maximum mutual information number, where m×n is the constraint condition for the total number of grid divisions, and B is usually taken as 0.6 power of the total number of data;

[0017] Considering the correlation between input variables and the correlation between input and output variables, the correlation feature selection algorithm CFS is used to screen out the optimal feature combination, so that the correlation between input variables is minimized and the correlation between input variables and output variables is maximized. The expression of the CFS algorithm is:

[0018]

[0019] Where w is the number of features in the feature combination, is the average MIC value between the input features and the output features in Q, is the average MIC value among the input features in Q, P Q is the evaluation score of the feature combination Q.

[0020] Furthermore, step 2 is specifically as follows:

[0021] Initialize the Gaussian mixture model. The data sets under different working conditions are non-Gaussian. The Gaussian mixture model is a new distribution obtained by weighted averaging of multiple Gaussian distributions, which can characterize non-Gaussian data. The specific formula is:

[0022]

[0023]

[0024]

[0025] Among them, p(x) is the probability density function of the Gaussian mixture model, N is the number of Gaussian components, N(xμ k ,∑ k ) is the probability density function of the kth Gaussian component, α k is the weight coefficient of the kth Gaussian component, μ k is the mean of the kth Gaussian component, ∑ k is the covariance of the kth Gaussian component;

[0026] The unknown parameter α in the Gaussian mixture model is estimated iteratively by using the EM maximum expectation method k , μ k ,∑ k , clustering is completed, and the formula of EM maximum expectation method is as follows:

[0027]

[0028]

[0029]

[0030]

[0031] where γ t (z j ) is the posterior probability density function.

[0032] Furthermore, step 3 is specifically as follows:

[0033] Using PA-TCN-LSTM to establish an independent single prediction model for the subset includes: using load characteristics as clustering criteria to divide the subset through a Gaussian mixture model, and using MIC-CFS to obtain input feature variables of the single model, and training the PA-TCN-LSTM single prediction model;

[0034] The specific calculation formula of LSTM is as follows:

[0035] f t=σ(W f [h t-1 ,x t ]+b f )

[0036] i t =σ(W i [h t-1 ,x t ]+b i )

[0037] o t =σ(W o [h t-1 ,x t ]+b o )

[0038] c t =i t tanh(W c [h t-1 ,x t ]+b c )+f t c t-1

[0039] h t =o t tanh(c t )

[0040]

[0041]

[0042] Among them, x t is the input at the current moment, c t-1 , c t The memory cell state at the previous moment and the current moment, h t-1 ,h t is the output of the LSTM layer at the previous moment and the current moment; σ is the sigmoid function, tanh represents the tangent hyperbolic function, and f t ,i t , o t Corresponding to the input gate, forget gate and output gate; W f , W i , W o The weights of the sigmoid function in the corresponding input gate, forget gate and output gate; b f , b i , b o are the corresponding biases respectively; W c , b c are the weights and biases in the tanh activation function;

[0043] The calculation formula of TCN network is as follows:

[0044]

[0045] Among them, f l (i) is the convolution kernel weight of the lth layer, r l is the expansion rate, y l (t) is the output of the last layer at time t, y l -1 is the output of the previous layer, and the input of the 0th layer is the input feature x;

[0046] The calculation formula for the parallel structure is as follows:

[0047] Y(t)=[y l (t)h t ]

[0048] Among them, y l (t) is the output of the TCN network at time t, h t is the output of the LSTM layer at time t.

[0049] Furthermore, step 4 is specifically as follows:

[0050] The output features of different trained single models are extracted, and the correlation between them and load features in different dimensions is analyzed through the multi-head additive attention mechanism. The integrated prediction model is trained by weighting the output features of different models. The calculation formula of the additive attention mechanism is:

[0051] α i =softmax(V T tanh(W f f i +W L L))

[0052]

[0053] where f i is the characteristic variable, L is the load characteristic, W f , W L , V T is the linear mapping weight matrix, tanh, softmax nonlinear activation function, α i is the output weight of different feature variables, c is the attention output result;

[0054] The calculation formula of the multi-head attention mechanism is:

[0055] multi=concat(c1,c2,…,c n )W O

[0056] Among them, c n Make the attention output results of different heads, concat refers to feature fusion, W O is the output weight matrix.

[0057] Furthermore, step 5 is specifically as follows:

[0058] The root mean square error RMSE, mean absolute error MAE and determination coefficient R are used. 2 The prediction effect of the integrated model is evaluated, and the calculation formula of the evaluation index is:

[0059]

[0060]

[0061]

[0062] Among them, y i is the actual value; is the predicted value; is the average of the actual values; n is the number of samples.

[0063] The present invention also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0064] The present invention also discloses a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method of the present invention are implemented.

[0065] The present invention also discloses a computer program product, comprising a computer program / instruction, which implements the steps of the method of the present invention when executed by a processor.

[0066] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0067] The present invention establishes an integrated coal-fired unit rear screen superheater wall temperature prediction model based on Gaussian mixture model and attention mechanism, collects relevant parameters of the coal-fired unit rear screen superheater wall temperature, and uses MIC-CFS algorithm to screen the model input feature variables; uses Gaussian mixture model to divide the data set according to the working condition; establishes an independent single prediction model for the subset through PA-TCN-LSTM; uses multi-head additive attention mechanism to weight the output result of the single model to establish an integrated model; based on the trained integrated rear screen superheater wall temperature prediction model, the model performance is verified using field collected data. The present invention can construct an integrated wall temperature prediction model with dynamic characteristics that adapt to different working conditions, can realize wall temperature prediction under complex working conditions, provide reference guidance for operating personnel, and prevent the occurrence of accidents such as pipe burst caused by overheating of the rear screen superheater pipeline, which is of great significance to the improvement of the stability and safety of coal-fired unit operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 An architecture diagram of an integrated coal-fired unit rear panel superheater wall temperature prediction model based on a Gaussian mixture model and an attention mechanism provided by the present invention;

[0069] Figure 2 A single model structure diagram for predicting the wall temperature of the rear panel superheater of a coal-fired unit based on a Gaussian mixture model and an attention mechanism provided by the present invention;

[0070] Figure 3 A Gaussian mixture model clustering diagram of an integrated coal-fired unit rear panel superheater wall temperature prediction model based on a Gaussian mixture model and an attention mechanism provided by the present invention;

[0071] Figure 4 A comparison diagram of the overall prediction effect of a single model of an integrated coal-fired unit rear panel superheater wall temperature prediction model based on a Gaussian mixture model and an attention mechanism provided by the present invention;

[0072] Figure 5 A comparison chart of the overall prediction effects of an integrated model and a single model of an integrated coal-fired unit rear panel superheater wall temperature prediction model based on a Gaussian mixture model and an attention mechanism provided by the present invention;

[0073] Figure 6 A comparison chart of the prediction effects of an integrated model and a single model under different operating conditions of an integrated coal-fired unit rear panel superheater wall temperature prediction model based on a Gaussian mixture model and an attention mechanism provided by the present invention. DETAILED DESCRIPTION

[0074] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0075] like Figures 1-2As shown, the integrated coal-fired unit rear panel superheater wall temperature prediction model based on Gaussian mixture model and attention mechanism in the embodiment of the present invention includes the following steps:

[0076] S1: Collect relevant parameters of the wall temperature of the rear screen superheater of the coal-fired unit, and use the MIC-CFS algorithm to screen the model input characteristic variables.

[0077] According to one embodiment of the present invention, the parameters related to the wall temperature of the rear screen superheater include: unit load, fuel calorific value, primary air volume, temperature and pressure, secondary air volume, temperature and pressure, temperature and pressure of cooling water flow at each level, total air volume, feed water flow temperature and pressure, flue gas damper opening, outlet flue gas oxygen content, main steam flow temperature and pressure.

[0078] According to one embodiment of the present invention, the use of the MIC-CFS algorithm to screen the model input feature variables includes: using the maximum information coefficient (MIC) to analyze the correlation between the selected parameters and the target variable, preliminarily screening out highly correlated variables based on the correlation, and then analyzing the correlation between the screened input variables that are highly correlated with the target variable, setting variables x, y, and the MIC formula is:

[0079]

[0080]

[0081] Where m and n are the interval sizes of the grid G, p(x,y) is the joint probability density function, p(x), p(y) are the marginal probability density functions, and I m×n (x,y) is the maximum information number under the grid G, Mic is the normalized maximum mutual information number, where m×n is the constraint condition for the total number of grid divisions, and B is usually taken as 0.6 of the total number of data.

[0082] Considering the correlation between input variables and the correlation between input and output variables, the correlation feature selection algorithm (CFS) is used to screen out the optimal feature combination so that the correlation between input variables is minimized and the correlation between input variables and output variables is maximized. The expression of the CFS algorithm is:

[0083]

[0084] Where w is the number of features in the feature combination, is the average MIC value between the input features and the output features in Q, is the average MIC value among the input features in Q, P Q is the evaluation score of the feature combination Q.

[0085] S2: Use Gaussian mixture model to divide the data set according to working conditions.

[0086] According to an embodiment of the present invention, the use of a Gaussian mixture model to divide the data set according to the working condition includes: initializing the Gaussian mixture model, the data sets of different working conditions are non-Gaussian, and the Gaussian mixture model is a new distribution obtained by weighted averaging of multiple Gaussian distributions, which can characterize non-Gaussian data. The specific formula is:

[0087]

[0088]

[0089]

[0090] Among them, p(x) is the probability density function of the Gaussian mixture model, N is the number of Gaussian components, N(xμ k ,∑ k ) is the probability density function of the kth Gaussian component, α k is the weight coefficient of the kth Gaussian component, μ k is the mean of the kth Gaussian component, ∑ k is the covariance of the kth Gaussian component.

[0091] The unknown parameter α in the Gaussian mixture model is estimated iteratively by using the EM maximum expectation method k , μ k ,∑ k , clustering is completed, and the formula of EM maximum expectation method is as follows:

[0092]

[0093]

[0094]

[0095]

[0096] where γ t (z j ) is the posterior probability density function.

[0097] S3: According to one embodiment of the present invention, the use of PA-TCN-LSTM to establish an independent single prediction model for a subset includes: dividing the subset through a Gaussian mixture model based on load characteristics as the clustering standard, and using MIC-CFS to obtain the input feature variables of the single model, and training the PA-TCN-LSTM single prediction model.

[0098] It should be noted that the single model combines the advantages of the LSTM neural network and the TCN neural network, and integrates them through a parallel structure, which can improve the temporal feature extraction ability of the single model. The specific calculation formula of LSTM is as follows:

[0099] f t =σ(W f [h t-1 ,x t ]+b f )

[0100] i t =σ(W i [h t-1 ,x t ]+b i )

[0101] o t =σ(W o [h t-1 ,x t ]+b o )

[0102] c t =i t tanh(W c [h t-1 ,x t ]+b c )+f t c t-1

[0103] h t =o t tanh(c t )

[0104]

[0105]

[0106] Among them, x t is the input at the current moment, c t-1 , c t The memory cell state at the previous moment and the current moment, h t-1 ,h t is the output of the LSTM layer at the previous moment and the current moment; σ is the sigmoid function, tanh represents the tangent hyperbolic function, and f t ,i t , o t Corresponding to the input gate, forget gate and output gate; W f , W i , W o The weights of the sigmoid function in the corresponding input gate, forget gate and output gate; bf , b i , b o are the corresponding biases respectively; W c , b c are the weights and biases in the tanh activation function.

[0107] The calculation formula of TCN network is as follows:

[0108]

[0109] Among them, f l (i) is the convolution kernel weight of the lth layer, r l is the expansion rate, y l (t) is the output of the last layer at time t, y l -1 is the output of the previous layer, and the input of the 0th layer is the input feature x.

[0110] The calculation formula for the parallel structure is as follows:

[0111] Y(t)=[y l (t)h t ]

[0112] Among them, y l (t) is the output of the TCN network at time t, h t is the output of the LSTM layer at time t

[0113] S4: Use the multi-head additive attention mechanism to weight the output of a single model and build an integrated model.

[0114] According to one embodiment of the present invention, the use of PA-TCN-LSTM to establish an independent single prediction model for a subset includes: dividing the subset through a Gaussian mixture model based on load characteristics as the clustering standard, and using MIC-CFS to obtain the input feature variables of the single model, and training the PA-TCN-LSTM single prediction model.

[0115] According to an embodiment of the present invention, the method of using a multi-head additive attention mechanism to weight the output results of a single model and establish an integrated model includes: extracting the output features of different trained single models, analyzing the correlation between them and load features in different dimensions through a multi-head additive attention mechanism, and training an integrated prediction model by weighting the output features of different models, wherein the calculation formula of the additive attention mechanism is:

[0116] α i =softmax(V T tanh(W f f i +W L L))

[0117]

[0118] where f i is the characteristic variable, L is the load characteristic, W f , W L , V T is the linear mapping weight matrix, tanh, softmax nonlinear activation function, α i is the output weight of different feature variables, and c is the attention output result;.

[0119] The calculation formula of the multi-head attention mechanism is:

[0120] multi=concat(c1,c2,…,c n )W O

[0121] Among them, c n Make the attention output results of different heads, concat refers to feature fusion, W O is the output weight matrix.

[0122] S5: Based on the trained integrated rear screen superheater wall temperature prediction model, the model performance is verified using field collected data. It should be noted that:

[0123] The root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R 2 ) is used to evaluate the prediction effect of the integrated model. The calculation formula of the evaluation index is:

[0124]

[0125]

[0126]

[0127] Among them, y i is the actual value; is the predicted value; is the average of the actual values; n is the number of samples.

[0128] It should be noted that the integration model based on the attention mechanism is different from the general integration learning that adopts average weighting and fixed weights. The integration model can update the output weights online according to the input data.

[0129] In conjunction with a specific embodiment of the present invention, the beneficial effects of the integrated coal-fired unit rear panel superheater wall temperature prediction model based on the Gaussian mixture model and the attention mechanism of the present invention are verified below.

[0130] The actual collected data of the wall temperature of the rear panel superheater of the coal-fired unit is used for the example test. In this embodiment, the relevant parameters of the wall temperature of the rear panel superheater of the coal-fired unit are collected, and the MIC-CFS algorithm is used to screen the model input characteristic variables. First, the input characteristic variables are preliminarily screened by the MIC method with a threshold of 0.5, as shown in Table 1.

[0131] Table 1: MIC value analysis table

[0132]

[0133]

[0134] According to Table 1 and the CFS algorithm, the feature combination with the smallest correlation between input variables and the largest correlation between input and output variables is selected: II-level superheater A desuperheating water flow, II-level superheater B desuperheating water flow, II-level superheater A desuperheater inlet steam temperature 3, II-level superheater A desuperheater inlet steam temperature 1, fuel mean, total air volume, feed water temperature, furnace outlet oxygen content as the input parameters of the single prediction model

[0135] The Gaussian mixture model is used to divide the data set according to the working conditions, such as Figure 3 As shown in the figure, the data before clustering presents an obvious non-Gaussian type with multiple peaks. When the number of clusters is set to 2, the probability density distribution curve of its subset shows a smooth and single Gaussian distribution shape, indicating that the two clusters have good homogeneity and separability, and are accurately divided according to working conditions.

[0136] Then, based on the selected input feature variables and the divided subsets, a single prediction model of the PA-TCN-LSTM structure is trained, such as Figure 4 As shown in the figure, the prediction effects of the single prediction model of PA-TCN-LSTM and the single models of LSTM, TCN and TCN-LSTM with a tandem structure are compared.

[0137] The multi-head additive attention mechanism is used to weight the output results of a single model and establish an integrated model to match different working conditions;

[0138] Finally, based on the trained integrated rear screen superheater wall temperature prediction model, the model performance was verified using field collected data, such as Figure 5-6 As shown in Table 2, the overall prediction effect of the integrated model is compared with that of the single model. At the same time, the prediction effects of the integrated model and the single model under different working conditions are also compared. The comparison results are shown in Table 2.

[0139] Table 2: Comparison of prediction model effects

[0140] Prediction Model RMSE MAE <![CDATA[R 2 ]]> TCN 2.5700 2.0495 0.9139 LSTM 2.7522 2.1330 0.9012 TCN-LSTM 2.6355 1.9989 0.9094 PA-TCN-LSTM 2.3077 1.7508 0.9305 Ensemble Model 1.8469 1.4471 0.9555

[0141] According to the integrated coal-fired unit rear screen superheater wall temperature prediction model of the Gaussian mixture model and attention mechanism established by the present invention, the relevant parameters of the coal-fired unit rear screen superheater wall temperature are collected, and the model input feature variables are screened using the MIC-CFS algorithm; the data set is divided according to the working conditions using the Gaussian mixture model; an independent single prediction model is established for the subset through PA-TCN-LSTM; the output results of the single model are weighted using the multi-head additive attention mechanism to establish an integrated model; based on the trained integrated rear screen superheater wall temperature prediction model, the model performance is verified using field collected data. The present invention can construct an integrated wall temperature prediction model with dynamic characteristics that adapt to different working conditions, can realize wall temperature prediction under complex working conditions, provide reference guidance for operating personnel, and prevent the occurrence of accidents such as pipe burst caused by overheating of the rear screen superheater pipeline, which is of great significance to the improvement of the stability and safety of coal-fired unit operation.

[0142] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0143] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0144] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0145] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0146] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0147] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0148] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0149] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0150] In addition, each functional unit in each embodiment of the present invention may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0151] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion, characterized in that: The steps include: Step 1: Collect relevant parameters of the wall temperature of the rear screen superheater of the coal-fired unit, and use the maximum information coefficient algorithm MIC and the correlation feature selection algorithm CFS to screen the characteristic variables of the relevant parameters; Step 2: Use the Gaussian mixture model to divide the data set according to the working conditions to form subsets; Step 3: Establish an independent single prediction model for the subset through the PA-TCN-LSTM network; Step 4: Use the multi-head additive attention mechanism to weight the output results of the single prediction model, match different working conditions to establish an integrated rear-panel superheater wall temperature prediction model, and train the model; Step 5: Based on the trained integrated rear-screen superheater wall temperature prediction model, the model performance is verified using field collected data to predict the superheater wall temperature of the coal-fired unit under complex operating conditions.

2. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: In step 1, the relevant parameters of the wall temperature of the rear screen superheater of the coal-fired unit include unit load, fuel calorific value, primary air volume, temperature and pressure, secondary air volume, temperature and pressure, cooling water flow temperature and pressure at each level, total air volume, feed water flow temperature and pressure, flue gas damper opening, outlet flue gas oxygen content, main steam flow temperature and pressure.

3. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: Step 1 is as follows: The maximum information coefficient MIC is used to analyze the correlation between the selected parameters and the target variable, and highly correlated variables are preliminarily screened out according to the correlation. Then, the correlation between the screened input variables with high correlation with the target variable is analyzed, and the variables x and y are set. The MIC formula is: Where m and n are the interval sizes of the grid G, p(x,y) is the joint probability density function, p(x), p(y) are the marginal probability density functions, and I m×n (x, y) is the maximum information number under the grid G, Mic is the normalized maximum mutual information number, where m×n is the constraint condition for the total number of grid divisions, and B is usually taken as 0.6 power of the total number of data; Considering the correlation between input variables and the correlation between input and output variables, the correlation feature selection algorithm CFS is used to screen out the optimal feature combination, so that the correlation between input variables is minimized and the correlation between input variables and output variables is maximized. The expression of the CFS algorithm is: Where w is the number of features in the feature combination, is the average MIC value between the input features and the output features in Q, is the average MIC value among the input features in Q, P Q is the evaluation score of the feature combination Q.

4. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: Step 2 is as follows: Initialize the Gaussian mixture model. The data sets under different working conditions are non-Gaussian. The Gaussian mixture model is a new distribution obtained by weighted averaging of multiple Gaussian distributions, which can characterize non-Gaussian data. The specific formula is: Among them, p(x) is the probability density function of the Gaussian mixture model, N is the number of Gaussian components, N(xμ k ,∑ k ) is the probability density function of the kth Gaussian component, α k is the weight coefficient of the kth Gaussian component, μ k is the mean of the kth Gaussian component, ∑ k is the covariance of the kth Gaussian component; The unknown parameter α in the Gaussian mixture model is estimated iteratively by using the EM maximum expectation method k , μ k ,∑ k , clustering is completed, and the formula of EM maximum expectation method is as follows: where γ t (z j ) is the posterior probability density function.

5. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: Step 3 is as follows: Using PA-TCN-LSTM to establish an independent single prediction model for the subset includes: using load characteristics as clustering criteria to divide the subset through a Gaussian mixture model, and using MIC-CFS to obtain input feature variables of the single model, and training the PA-TCN-LSTM single prediction model; The specific calculation formula of LSTM is as follows: f t =σ(W f [h t-1 ,x t ]+b f ) i t =σ(W i [h t-1 ,x t ]+b i ) the t =σ(W o [h t-1 ,x t ]+b o ) c t =i t tanh(W c [h t-1 ,x t ]+b c )+f t c t-1 h t =o t fishy t ) Among them, x t is the input at the current moment, c t-1 , c t The memory cell state at the previous moment and the current moment, h t-1 ,h t is the output of the LSTM layer at the previous moment and the current moment; σ is the sigmoid function, tanh represents the tangent hyperbolic function, and f t ,i t , o t Corresponding to the input gate, forget gate and output gate; W f , W i , W o The weights of the sigmoid function in the corresponding input gate, forget gate and output gate; b f , b i , b o are the corresponding biases respectively; W c , b c are the weights and biases in the tanh activation function; The calculation formula of TCN network is as follows: Among them, f l (i) is the convolution kernel weight of the lth layer, r l is the expansion rate, y l (t) is the output of the last layer at time t, y l-1 is the output of the previous layer, and the input of the 0th layer is the input feature x; The calculation formula for the parallel structure is as follows: Y(t)=[y l (t)h t ] Among them, y l (t) is the output of the TCN network at time t, h t is the output of the LSTM layer at time t.

6. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: Step 4 is as follows: The output features of different trained single models are extracted, and the correlation between them and load features in different dimensions is analyzed through the multi-head additive attention mechanism. The integrated prediction model is trained by weighting the output features of different models. The calculation formula of the additive attention mechanism is: α i =softmax(V T tanh(W f f i +W L L)) where f i is the characteristic variable, L is the load characteristic, W f , W L , V T is the linear mapping weight matrix, tanh and softmax are nonlinear activation functions, α i is the output weight of different feature variables, c is the attention output result; The calculation formula of the multi-head attention mechanism is: multi=concat(c1,c2,...,c n )W O Among them, c n Make the attention output results of different heads, concat refers to feature fusion, W O is the output weight matrix.

7. The method for predicting superheater wall temperature of a coal-fired unit under complex operating conditions based on multi-model fusion according to claim 1 is characterized in that: Step 5 is as follows: The root mean square error RMSE, mean absolute error MAE and determination coefficient R are used. 2 The prediction effect of the integrated model is evaluated, and the calculation formula of the evaluation index is: Among them, y i is the actual value; is the predicted value; is the average of the actual values; n is the number of samples.

8. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Citation Information

Cited By

  • Data anomaly detection method based on multi-scale feature extraction

    CN120316629A

  • Data anomaly detection method based on multi-scale feature extraction

    CN120316629B

  • Variable-working-condition coal-fired unit high-temperature superheater outlet steam temperature prediction method and device

    CN120493774A