A prediction method for high-temperature corrosion of boiler water walls based on data fusion and driving

Abstract

The present invention relates to the technical field of boiler water-cooled wall high-temperature corrosion prediction, and specifically discloses a boiler water-cooled wall high-temperature corrosion prediction method based on data fusion and drive, comprising the following steps: S1: collecting target boiler design and operation parameters; S2: constructing an organic sulfur evolution model, a pyrite evolution model and a tube wall temperature coupled heat transfer model; S3: building a boiler water-cooled wall high-temperature corrosion model in combination with a corrosion rate correlation formula; S4: calculating high-temperature corrosion conditions under several different parameters to obtain a typical condition simulation database; S5: constructing a transposed convolutional neural network prediction model based on the database and taking various boiler parameters as input; S6: estimating the degree of thinning of the water-cooled wall surface; the present invention develops a complete water-cooled wall high-temperature corrosion prediction CFD model, which for the first time considers all key parameters affecting the corrosion rate, such as H2S concentration in the water-cooled wall area, FeS deposit distribution on the wall surface and tube wall temperature.

Description

Technical Field

[0001] The present invention belongs to the technical field of boiler water wall high temperature corrosion prediction, and in particular relates to a boiler water wall high temperature corrosion prediction method based on data fusion and drive. Background Art

[0002] In recent years, coal-fired boilers have generally adopted air staged combustion to reduce NO X Emissions, but also significantly aggravated the problem of high-temperature corrosion of the water-cooled wall. High-temperature corrosion is strongly affected by three key factors: H2S concentration in the water-cooled wall area, FeS deposits on the wall, and tube wall temperature.

[0003] However, to date, most studies on high-temperature corrosion have only focused on the reducing atmosphere and H2S concentration distribution in the furnace, while ignoring two important factors: FeS deposits and tube wall temperature. In addition, due to the lack of corrosion measurement methods and the time-consuming numerical simulation methods, there is still a lack of a technical means to accurately reflect the high-temperature corrosion rate distribution of the water-cooled wall during boiler operation in real time. Therefore, a prediction system for boiler water-cooled wall high-temperature corrosion based on data fusion and data-driven is designed. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for predicting high-temperature corrosion of boiler water-cooled walls based on data fusion and driving, so as to solve the problems raised in the above background technology.

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

[0006] A method for predicting high-temperature corrosion of boiler water-cooled walls based on data fusion and driving includes the following steps:

[0007] S1: Collect target boiler design and operating parameters, including combustion system design, heat exchange system design and online monitoring data;

[0008] S2: Construct organic sulfur evolution model, pyrite evolution model and pipe wall temperature coupled heat transfer model to calculate the detailed distribution of H2S concentration, FeS deposition and wall temperature respectively;

[0009] S3: Build a boiler water wall high-temperature corrosion model based on the corrosion rate correlation formula;

[0010] S4: Calculate high-temperature corrosion conditions under several different parameters, obtain a typical condition simulation database, and verify it using field measured data;

[0011] S5: Based on the database and taking various boiler parameters as input, a transposed convolutional neural network prediction model is constructed;

[0012] S6: Estimate the thinning degree of the water-cooled wall surface, and realize the real-time prediction of the high-temperature corrosion rate of the boiler water-cooled wall and the real-time prediction of the cumulative thinning of the wall thickness.

[0013] Preferably, the organic sulfur evolution model includes the thermal desorption of organic sulfur from pulverized coal and the gas phase reaction process of sulfur-containing components in the flue gas, which is used to predict the concentration distribution of H2S gas in the furnace;

[0014] The pyrite evolution model includes the decomposition and oxidation of pyrite particles during combustion and the deposition of its incomplete oxidation product, FeS, and is used to predict the deposition distribution of FeS.

[0015] The coupled heat transfer model includes a three-dimensional CFD model that simulates the combustion and heat transfer process in the furnace and a one-dimensional hydrodynamic model that simulates the steam flow and heat transfer process in the tube, which is used to predict the tube wall temperature distribution.

[0016] Preferably, in said S4, the operating parameters are boiler load, excess air coefficient, burnout air rate, air volume of each burner and coal volume.

[0017] Preferably, in said S3, the water-cooled wall high-temperature corrosion distribution is composed of H2S corrosion rate distribution, theoretical FeS deposit corrosion rate distribution and FeS deposition rate distribution.

[0018] Preferably, in S5, the model adopts 6-fold cross validation, and a total of 6 rounds of cross validation training are performed.

[0019] Preferably, the training process of the neural network specifically includes:

[0020] S51: The training data is randomly divided into 6 groups. In each round of training, one group is selected as the unfamiliar test set, and the remaining five groups are used as training sets.

[0021] S52: Input the training set data of this round into the neural network with initialized parameters and output the result;

[0022] S53: The error between the output and the true value is evaluated by the loss function, and the optimizer updates the neural network parameters according to the error;

[0023] S54: Repeat the training process of S52-S53 until the accuracy of this round reaches the optimal value;

[0024] S55: Repeat the training process of S52-S54, and select the model whose loss function is close to the average loss function as the final model.

[0025] Preferably, in S5, the structure of the transposed convolutional neural network is:

[0026] 200 neurons are used to re-encode the input vector. A fully connected layer is used to connect the input vector and the encoded vector. The encoded vector is reshaped into a 10×20 pixel feature map.

[0027] Use 64 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 1 on the feature map, and output a 64-channel feature map of 10×20 pixels;

[0028] Use 32 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 2 on the new feature map, and output a 32-channel feature map of 20×40 pixels;

[0029] Repeat the above operation using 16 convolution kernels of size 3×3 to output a 16-channel feature map of 40×80 pixels;

[0030] A convolution kernel of size 3×3 is used to perform a transposed convolution operation with a stride of 2, and a single-channel feature map of 80×160 pixels is output to obtain the target image.

[0031] Preferably, in S6, the process of estimating the degree of water-cooled wall corrosion includes:

[0032] S61: The H2S corrosion rate distribution, theoretical FeS deposit corrosion rate distribution, and FeS deposition rate distribution were input into the neural network prediction model to obtain three sets of corresponding network parameters. All output data were converted into a single-channel feature map of size 80×160 pixels.

[0033] S62: Integrate the FeS deposition rate over time to obtain the historical FeS deposition distribution. Combined with the theoretical FeS deposition corrosion rate distribution under the current working conditions, the actual FeS deposition corrosion rate distribution is obtained. Combined with the H2S corrosion rate distribution, the real-time total high-temperature corrosion rate distribution is obtained.

[0034] S63: Integrate the real-time high-temperature corrosion rate over time to obtain the current cumulative thinning distribution of the water-cooled wall thickness.

[0035] Preferably, in said S51, normalization processing is performed before the data is input into the artificial neural network.

[0036] Preferably, in S53, the loss function is named LMSE, specifically:

[0037]

[0038] Where M is the number of samples, y i is the model prediction value, is the true value, y max The maximum working range.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) This paper develops a complete high-temperature corrosion prediction model for water-cooled walls. For the first time, it considers all key parameters affecting the corrosion rate, such as H2S concentration in the water-cooled wall area, FeS deposit distribution on the wall surface, and tube wall temperature. Combined with the corrosion rate correlation formula, it realizes the accurate prediction of the detailed distribution of high-temperature corrosion rate.

[0041] (2) The present invention realizes the real-time prediction of the three-dimensional distribution of high-temperature corrosion of the water-cooled wall of a coal-fired boiler by combining numerical simulation results with artificial neural networks. The model prediction results are in good agreement with the field data, with an error LMSE of only 0.32 and a response time of less than 0.01s. The real-time prediction of the detailed distribution of the high-temperature corrosion rate of the water-cooled wall and the real-time prediction of the cumulative thinning distribution of the water-cooled wall tube during boiler operation is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of the method for predicting high-temperature corrosion of boiler water walls of the present invention;

[0043] Figure 2 This is a convolutional neural network workflow diagram of the present invention;

[0044] Figure 3 The distribution diagram of H2S corrosion, theoretical FeS corrosion and FeS deposition rate of the water-cooled wall of the present invention;

[0045] Figure 4 The training set error and the validation set error of the present invention are plotted against the number of training rounds;

[0046] Figure 5 Schematic diagram of the transposed convolutional neural network structure of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] Example:

[0049] See also Figure 1-Figure 5 As shown in FIG, a method for predicting high-temperature corrosion of boiler water-cooled walls based on data fusion and driving includes:

[0050] S1: Collect target boiler design and operating parameters, including combustion system design, heat exchange system design and online monitoring data;

[0051] S2: Construct an organic sulfur evolution model, a pyrite evolution model, and a tube wall temperature coupled heat transfer model to calculate the detailed distribution of H2S concentration, FeS deposition, and wall temperature, respectively.

[0052] S3: Build a boiler water wall high-temperature corrosion model based on the corrosion rate correlation formula;

[0053] Among them, the corrosion rate correlation formula is formula (1) and formula (2):

[0054] Carbon steel and low alloy steel corrosion rates:

[0055]

[0056] Where, CR H2S is the H2S corrosion rate, mm / year; T w is the tube wall temperature, K; [H2S] is the volume fraction of H2S in the flue gas, ppm; Cr% is the chromium content of the tube wall metal;

[0057] FeS deposit corrosion rate:

[0058]

[0059] Where: CR dep is the corrosion rate of FeS deposits, mm / year; [O2] and [CO] are the volume fractions of O2 and CO in the flue gas, respectively, %;

[0060] S4: Calculate high-temperature corrosion conditions under several different parameters, obtain a typical condition simulation database, and verify it using field measured data;

[0061] S5: Based on the database and taking various boiler parameters as input, a transposed convolutional neural network prediction model is constructed;

[0062] S6: Estimate the thinning degree of the water-cooled wall surface, and realize the real-time prediction of the high-temperature corrosion rate of the boiler water-cooled wall and the real-time prediction of the cumulative thinning of the wall thickness.

[0063] In one embodiment of the present invention, the organic sulfur evolution model includes the thermal desorption of organic sulfur from pulverized coal and the gas phase reaction process of sulfur-containing components in flue gas, and is used to predict the concentration distribution of H2S gas in the furnace;

[0064] The pyrite evolution model includes the decomposition and oxidation of pyrite particles during combustion and the deposition of its incomplete oxidation product, FeS, and is used to predict the deposition distribution of FeS.

[0065] The coupled heat transfer model includes a three-dimensional CFD model to simulate the combustion and heat transfer process in the furnace and a one-dimensional hydrodynamic model to simulate the steam flow and heat transfer process in the tube, which is used to predict the temperature distribution on the tube wall.

[0066] The 3D CFD model uses ANSYS Fluent as the computing platform, and includes a complete description of the processes including gas-solid two-phase flow, radiation heat transfer, gas-phase combustion, pulverized coal pyrolysis, and carbon particle combustion in the furnace.

[0067] The gas flow is described by the Favre-averaged conservation equations of mass, momentum, energy, and species mass fractions;

[0068] The Realizable k-ε model is used for turbulence closure;

[0069] The radiative heat transfer was simulated using the discrete ordinate method (DO), with angular discretization of two partitions and four pixels in the polar and azimuthal directions, respectively. The weighted sum grey gas model (WSGGM) was used to calculate the smoke radiation absorption coefficient.

[0070] The motion of pulverized coal particles is tracked in the Lagrangian framework using the random trajectory method. In the process of tracking the motion trajectory of coal particles, the mass, momentum and energy exchange between the gas flow field and the coal particles are calculated using the particle source (PSIC) method.

[0071] The pulverized coal particle size is assumed to follow the Rosin-Rammler distribution, with an average diameter of 62 μm and a uniformity index of 1.1;

[0072] The pyrolysis process of coal is described by a first-order single reaction rate model, and the volatile matter released by coal pyrolysis is assumed to be a single virtual component C a H b O c S d N e , the volatile components (a, b, c, d, e values) can be calculated by the conservation of mass of each element;

[0073] The surface combustion reaction rate of carbon particles formed by pyrolysis of pulverized coal and O2 is calculated using a kinetic / diffusion controlled model.

[0074] The metal tube wall temperature is determined by the heating effect of the flue gas side and the cooling effect of the steam side in the furnace, which can be expressed as:

[0075]

[0076] In formula (3): q is the wall heat flux, W / m 2 ;T w 、T s are the metal tube wall temperature, K; T s is the steam temperature in the tube, K; H s is the convection heat transfer coefficient of steam flow in the tube, W / (m 2 ·K); Δ mis the thickness of the metal pipe wall, m; k m is the thermal conductivity of the metal pipe wall, W / (m·K);

[0077] Wall temperature T w The steam temperature in the tube is T s , wall heat flux q and convection heat transfer coefficient H s The distribution of T s 、H S , q and T w They are solved together as key parameters of the high temperature corrosion model;

[0078] A one-dimensional hydrodynamic model is used to simulate the flow and heat transfer process of steam in the water-cooled wall tube to obtain the steam temperature T s and the convective heat transfer coefficient H s , and further combined with the distribution of water-cooled wall heat flux q obtained by the three-dimensional CFD numerical model of the boiler flue gas side, the tube wall temperature T w Detailed distribution of

[0079] One-dimensional hydrodynamic model for simulating steam flow and heat transfer in tubes:

[0080] One-dimensional hydrodynamic model of steam flow in water-cooled wall tubes SE software is a computing platform that discretizes the pipe wall into several control volumes along the spiral direction and solves the conservation equations of mass, momentum, and energy in each control volume:

[0081]

[0082]

[0083]

[0084] In the above formula: x is the pipe length, m; V is the steam flow rate, m / s; p is the steam density, kg / m3; p is the steam pressure, Pa; g is the gravitational acceleration constant, m / s 2 ; z is the elevation of the control body, m; f is the friction coefficient; D is the pipe diameter, m; h is the steam specific enthalpy, J / kg; Q H is the heat absorbed by steam per unit volume, W / m 3 , depends on the local heat flux q at the corresponding wall position of each control volume;

[0085] Q of each control volume H The value is given by the solution of the three-dimensional CFD model of the boiler flue gas side according to its position on the water-cooled wall. HAfter the value of is determined, the one-dimensional hydrodynamic model solves the conservation equations starting from the upstream control body of the water-cooled wall steam flow, and transfers the calculated parameters at the outlet of each control body to the downstream control body as its inlet boundary condition, sequentially obtaining the steam temperature and other key parameters of all control bodies of the entire spiral water-cooled wall;

[0086] Iterative calculation process of the coupled heat transfer model:

[0087] First, assume that the initial steam temperature is T s,0 For average distribution, then the flue gas side 3D CFD model can be based on T s,0 Calculate the water wall heat flux distribution q j According to q j The heat absorption Q of each control body in formula (6) can be determined by the position of each control body. H ,Then, the one-dimensional flow model calculates the steam temperature of each control volume by solving the ,control equations (4)-(6) of steam flow;

[0088] New steam temperature distribution T s,j+1 It is passed back to the 3D CFD model to update the wall thermal boundary conditions for the next iterative calculation;

[0089] The calculation process is repeated until the interval between two iterations is T s,j The difference is less than 1K;

[0090] The coupled heat transfer model also gives the steam temperature T in the tube. s and the convergent solution of the wall heat flux q. On this basis, the wall temperature T can be further calculated according to formula (3): w In formula (3), the convection heat transfer coefficient of steam in the tube is H s Calculated by Flownex software based on steam state parameters, and Δ m and k m Determined by the material used for the water-cooled wall.

[0091] In one embodiment of the present invention, in S4, the operating parameters are 41 operating parameters including boiler load, excess air coefficient, burnout air rate, air volume of each burner, and coal volume. The operating parameters are shown in the following table:

[0092]

[0093] In one embodiment of the present invention, in S3, the water wall high temperature corrosion distribution is composed of the H2S corrosion rate distribution, the theoretical FeS deposit corrosion rate distribution, and the FeS deposition rate distribution.

[0094] In one embodiment of the present invention, in S5, the model adopts 6-fold cross validation, and a total of 6 rounds of cross validation training are performed.

[0095] In one embodiment of the present invention, the training process of the neural network specifically includes:

[0096] S51: The training data is randomly divided into 6 groups. In each round of training, one group is selected as the unfamiliar test set, and the remaining five groups are used as training sets.

[0097] S52: Input the training set data of this round into the neural network with initialized parameters and output the result;

[0098] S53: The error between the output and the true value is evaluated by the loss function, and the optimizer updates the neural network parameters according to the error;

[0099] S54: Repeat the training process of S52-S53 until the accuracy of this round reaches the optimal value;

[0100] S55: Repeat the training process of S52-S54, and select the model whose loss function is close to the average loss function as the final model.

[0101] In one embodiment of the present invention, in S53, the loss function is named LMSE, specifically:

[0102]

[0103] In formula (7), M is the number of samples, y i is the model prediction value, is the true value, y max The maximum range for each work;

[0104] Reference Figure 4 As shown in (a), (b), and (c), they are the training error change curves of the H2S corrosion rate distribution, the theoretical FeS deposit corrosion rate distribution, and the FeS deposition rate distribution, respectively. After 100 steps of training, the errors of the three variables are all less than 0.01. The error of the FeS deposition rate distribution model validation set is close to that of the training set, and the LMSE errors of each item are all less than 0.32.

[0105] Reference Figure 5 As shown in S5, the structure of the transposed convolutional neural network is:

[0106] 200 neurons are used to re-encode the input vector. A fully connected layer is used to connect the input vector and the encoded vector. The encoded vector is reshaped into a 10×20 pixel feature map.

[0107] Use 64 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 1 on the feature map, and output a 64-channel feature map of 10×20 pixels;

[0108] Use 32 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 2 on the new feature map, and output a 32-channel feature map of 20×40 pixels;

[0109] Repeat the above operation using 16 convolution kernels of size 3×3 to output a 16-channel feature map of 40×80 pixels;

[0110] A convolution kernel of size 3×3 is used to perform a transposed convolution operation with a stride of 2, and a single-channel feature map of 80×160 pixels is output to obtain the target image.

[0111] In one embodiment of the present invention, in S6, the process of estimating the degree of water wall corrosion includes:

[0112] S61: The H2S corrosion rate distribution, theoretical FeS deposit corrosion rate distribution, and FeS deposition rate distribution were input into the neural network prediction model to obtain three sets of corresponding network parameters. All output data were converted into a single-channel feature map of size 80×160 pixels.

[0113] S62: Integrate the FeS deposition rate over time to obtain the historical FeS deposition distribution. Combined with the theoretical FeS deposition corrosion rate distribution under the current working conditions, the actual FeS deposition corrosion rate distribution is obtained. Combined with the H2S corrosion rate distribution, the real-time total high-temperature corrosion rate distribution is obtained.

[0114] S63: Integrate the real-time high-temperature corrosion rate over time to obtain the current cumulative thinning distribution of the water-cooled wall thickness.

[0115] In one embodiment of the present invention, in S51, the data is normalized before being input into the artificial neural network;

[0116] All input and output parameters are linearly normalized, that is, the original data is linearly transformed and the data values ​​are mapped to the range [0,1], which can be expressed as follows:

[0117]

[0118] In formula (8), x is the original data and x' is the normalized data.

[0119] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting high-temperature corrosion of boiler water-cooled walls based on data fusion and driving, characterized in that: The following steps are involved: S1: Collect target boiler design and operating parameters, including combustion system design, heat exchange system design and online monitoring data; S2: Construct organic sulfur evolution model, pyrite evolution model and pipe wall temperature coupled heat transfer model to calculate the detailed distribution of H2S concentration, FeS deposition and wall temperature respectively; S3: Build a boiler water wall high-temperature corrosion model based on the corrosion rate correlation formula; S4: Calculate high-temperature corrosion conditions under several different parameters, obtain a typical condition simulation database, and verify it using field measured data; S5: Based on the database and taking various boiler parameters as input, a transposed convolutional neural network prediction model is constructed; S6: Estimate the degree of water wall thinning, and realize the real-time prediction of the high-temperature corrosion rate of the boiler water wall and the real-time prediction of the cumulative thinning of the wall thickness; The organic sulfur evolution model includes the thermal desorption of organic sulfur from pulverized coal and the gas-phase reaction of sulfur-containing components in flue gas, and is used to predict the concentration distribution of H2S gas in the furnace; The pyrite evolution model includes the decomposition and oxidation of pyrite particles during combustion and the deposition of its incomplete oxidation product, FeS, and is used to predict the deposition distribution of FeS. The coupled heat transfer model includes a three-dimensional CFD model to simulate the combustion and heat transfer process in the furnace and a one-dimensional hydrodynamic model to simulate the steam flow and heat transfer process in the tube, which is used to predict the temperature distribution on the tube wall. In S5, the structure of the transposed convolutional neural network is: 200 neurons are used to re-encode the input vector. A fully connected layer is used to connect the input vector and the encoded vector. The encoded vector is reshaped into a 10×20 pixel feature map. Use 64 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 1 on the feature map, and output a 64-channel feature map of 10×20 pixels; Use 32 convolution kernels of size 3×3 to perform a transposed convolution operation with a stride of 2 on the new feature map, and output a 32-channel feature map of 20×40 pixels; Repeat the above operation using 16 convolution kernels of size 3×3 to output a 16-channel feature map of 40×80 pixels; A convolution kernel of size 3×3 is used to perform a transposed convolution operation with a stride of 2, and a single-channel feature map of 80×160 pixels is output to obtain the target image.

2. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 1 is characterized by: In said S4, the operating parameters are boiler load, excess air coefficient, burnout air rate, air volume of each burner and coal volume.

3. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 1 is characterized by: In the S3, the water-cooled wall high-temperature corrosion distribution is composed of the H2S corrosion rate distribution, the theoretical FeS deposit corrosion rate distribution and the FeS deposition rate distribution.

4. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 1 is characterized by: In S5, the model adopts 6-fold cross validation, and a total of 6 rounds of cross validation training are performed.

5. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 1 is characterized by: The training process of the neural network specifically includes: S51: The training data is randomly divided into 6 groups. In each round of training, one group is selected as the unfamiliar test set, and the remaining five groups are used as training sets. S52: Input the training set data of this round into the neural network with initialized parameters and output the result; S53: The error between the output and the true value is evaluated by the loss function, and the optimizer updates the neural network parameters according to the error; S54: Repeat the training process of S52-S53 until the accuracy of this round reaches the optimal value; S55: Repeat the training process of S52-S54, and select the model whose loss function is close to the average loss function as the final model.

6. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 1 is characterized by: In S6, the process of estimating the degree of water wall corrosion includes: S61: The H2S corrosion rate distribution, theoretical FeS deposit corrosion rate distribution, and FeS deposition rate distribution were input into the neural network prediction model to obtain three sets of corresponding network parameters. All output data were converted into a single-channel feature map of size 80×160 pixels. S62: Integrate the FeS deposition rate over time to obtain the historical FeS deposition distribution. Combined with the theoretical FeS deposition corrosion rate distribution under the current working conditions, the actual FeS deposition corrosion rate distribution is obtained. Combined with the H2S corrosion rate distribution, the real-time total high-temperature corrosion rate distribution is obtained. S63: Integrate the real-time high-temperature corrosion rate over time to obtain the current cumulative thinning distribution of the water-cooled wall thickness.

7. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 5 is characterized by: In S51 , the data is normalized before being input into the artificial neural network.

8. The method for predicting high-temperature corrosion of boiler water walls based on data fusion and driving according to claim 5 is characterized by: In the S53, the loss function is named LMSE, specifically: Where M is the number of samples, y i is the model prediction value, is the true value, y max The maximum working range.

Images