Boiler water cooling wall temperature prediction method, over-temperature early warning method, system and medium

Abstract

The invention discloses a boiler water-cooled wall temperature prediction method, an overtemperature early warning method, a system and a medium, and belongs to the technical field of boiler safe operation, and the boiler water-cooled wall temperature prediction method comprises the following steps: 1, evaluating the characteristic importance of an auxiliary variable on the water-cooled wall temperature through a Gain mode of a Light GBM algorithm, and screening out the first D key variables; 2, optimizing delay time of each auxiliary variable and a target variable by adopting an AMI (Average Maximum Information) and DE (Differential Evolution) algorithm, and reconstructing input variable data; step 3, dynamically optimizing hidden layer node parameters of the random configuration network SCN based on a hiking algorithm HOA, constructing an HOA-SCN model, and introducing L2 regularization to calculate an output weight; 4, the trained HOA-SCN model is used for predicting the wall temperature of the water cooling wall, the prediction precision and robustness are remarkably improved through the prediction method, and effective support is provided for safe operation of a boiler.

Description

Technical Field

[0001] The present invention belongs to the technical field of boiler safe operation, and particularly relates to a method for predicting the wall temperature of a boiler water wall, an over-temperature early warning method, a system and a medium. Background Technique

[0002] Ultra (ultra)-critical units have become the main units in the power grid because of their high efficiency. However, there are serious tube burst problems in the "four tubes" of the boiler, which greatly affect the safe operation of the units. Thermal power units need to frequently participate in deep peak shaving, which increases the combustion instability of the boiler, seriously affects the heat transfer characteristics in the furnace, endangers the safety of the water wall, and even causes heat transfer surface leakage accidents and even explosions, thus limiting the unit's ability to participate in peak shaving. Accurately predicting the inner wall temperature of the furnace is the key to solving the over-temperature problem.

[0003] In recent years, using data-driven methods to establish a boiler combustion model and predict the wall temperature has become a research hotspot in this field.

[0004] As is well known, traditional neural networks lack interpretability, and the process of iteratively searching for a set of appropriate weights and biases is time-consuming and computationally intensive, which is difficult for efficiently processing large-scale industrial data, especially power production process data.

[0005] Stochastic Configuration Networks (SCN) is an incremental learning method with universal approximation properties. It uses a supervised mechanism to randomly configure the parameters of hidden layer nodes under a set of inequality constraints to adaptively select their value ranges. This method has the significant advantages of artificial intervention in the network structure and adaptive selection of the number of hidden layer nodes. After determining the hidden layer parameters through the stochastic configuration algorithm, the output weights are calculated based on the pseudo-inverse theory. This mechanism makes the training speed of SCN faster than traditional gradient-based iterative algorithms. Currently, stochastic configuration networks have been applied to establish models for various industrial scenarios, and this invention is the first application of stochastic configuration networks to model the metal wall temperature of boiler heating surfaces. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to implement the application of the stochastic configuration network SCN to model the wall temperature of the boiler water wall. And to improve the efficiency of the traditional stochastic configuration network SCN in searching for weights and biases and improve the generalization and prediction performance of the model, the present invention introduces the Hiking Optimization Algorithm (HOA) into the stochastic configuration network, and establishes a prediction model for the wall temperature of the boiler water wall based on HOA-SCN, which can enable boiler operators to predict the metal temperature of the heating surface in advance and provide guidance for adjusting the over-temperature of the pipe wall.

[0007] The present invention proposes a method for predicting the wall temperature of a boiler water wall, including the following steps:

[0008] Step 1: Select D auxiliary variables that have an important impact on the water wall temperature.

[0009] Step 2: Construct a water wall temperature prediction model as follows: Step 2.1: Given the objective function , assume that L - 1 neural nodes in the hidden layer of the randomly configured network (SCN) have been constructed; where represents the input dimensional space, represents the output dimensional space, D is the dimension of the input samples, i.e., the number of auxiliary variables, and m is the dimension of the output samples; Step 2.2: Calculate the input weight of the j-th node in the hidden layer and the bias of the j-th node in the hidden layer, and then obtain the optimal output weights of all nodes in the SCN ; where 1 ≤ j ≤ L - 1; Step 2.3: The predicted value of the water wall temperature prediction model is: ; where represents the weight matrix between the input layer and the hidden layer in the SCN, represents the weight matrix of the input layer in the SCN, represents the weight matrix of the (L - 1)-th hidden layer in the SCN;

[0010] Step 3: Train the water wall temperature prediction model to obtain a trained water wall temperature prediction model for predicting the water wall temperature.

[0011] Further, the specific steps for selecting the auxiliary variables that have an important impact on the water wall temperature in Step 1 are as follows:

[0012] Step 1.1: Collect historical data of the auxiliary variables;

[0013] Step 1.2: Perform data preprocessing on the collected historical data of the auxiliary variables. The preprocessing includes abnormal data processing and normalization processing;

[0014] Step 1.3: Use the Gain mode in the LightGBM algorithm to evaluate the feature importance of the auxiliary variables on the wall temperature, and select the top D auxiliary variables with the highest feature importance as the auxiliary variables that have an important impact on the water wall temperature.

[0015] Further, the evaluation of the feature importance of the auxiliary variables on the wall temperature is as shown in the following formula:

[0016] (1)

[0017] (2)

[0018] Among them, is the reduction of the loss caused by the metal temperature of the water wall tube wall during node splitting, measured by the mean square error; is the mean square error of the parent node, and are the mean square errors of the left child node and the right child node.

[0019] Furthermore, the auxiliary variables selected in step 1 include the main steam pressure, fuel quantity, static vane and moving vane positions of the induced draft fan, load, superheat degree, secondary air of the F baffle of the B3 burner, secondary air of the F baffle of the B1 burner, secondary air of the F baffle of the A3 burner, secondary air of the F baffle of the F1 burner, and the flue gas temperature at the outlet of the air preheater; there are a total of D = 10 auxiliary variables.

[0020] Furthermore, the auxiliary variables in step 1 are the reconstructed auxiliary variables after considering the delay time between the auxiliary variables and the water wall wall temperature;

[0021] Define the th variable input to the water wall wall temperature prediction model as , is the number of samples; x i ( t ) = [ x i , 1 ( t ), x i , 2 ( t ),..., x i , d ( t ),..., x i , D ( t )] , the target variable is expressed as ;

[0022] Since there is a delay between the target variable and the input variable, use to represent the dth input variable of the input and the target variable The delay time between;

[0023] The input variable is reconstructed as shown in the following formula:

[0024] X i ( t ) = [ x ^ i , 1 ( t ), x ^ i , 2 ( t ),..., x ^ i , d ( t ),..., x ^ i , D ( t )] (3)

[0025] (4)

[0026] Among them, represents the value of the dth input variable after reconstruction.

[0027] Delay time The calculation method of is as follows:

[0028] The average mutual information AMI method combined with the differential evolution DE algorithm is used to calculate the delay time between each auxiliary variable and the water wall wall temperature.

[0029] Furthermore, the steps for constructing the water wall wall temperature prediction model described in Step 2 are as follows:

[0030] Step 2.1, Given the objective function , assuming that L - 1 neural nodes in the hidden layer of the SCN have been constructed, then the network output expression of the current hidden layer is:

[0031] (5)

[0032] where β j = [ β j , 1 ,..., β j , m ] T represents the output weight of the j - th node in the hidden layer; represents the activation function; and are the input weight and bias of the j - th node in the hidden layer respectively, represents the input of the current hidden layer;

[0033] Step 2.2, Calculate the input weight of the j - th node in the hidden layer and the bias of the j - th node in the hidden layer through the Hiking Optimization Algorithm (HOA), and then obtain the optimal output weight , the specific steps are as follows:

[0034] Step 2.2.1, Initialize the position of the traveler;

[0035] Initialize the position of traveler z , as shown in the following equation:

[0036] (6)

[0037] where is a uniformly distributed number in the range [0, 1]; represents the current position of traveler z at time t; and represent the upper and lower bounds of the j - th dimension of the decision variable of the optimization problem;

[0038] Step 2.2.2, Compare q with , q represents the q - th traveler, , is the population size, that is, the number of travelers;

[0039] If q ≤ , calculate the fitness function , and enter Step 2.2.3;

[0040] If q > , determine the best position of the traveler, that is, the minimum fitness function , the last element in the minimum fitness function is the bias , and the transpose of the other elements are the input weights , and when It > Maxlt, It is the current iteration number, Maxlt is the maximum iteration number, so as to determine the optimal output weights of all nodes through formula (11) , enter step 2.2.6;

[0041] If q > , and when It ≤ Maxlt, enter step 2.2.3;

[0042] (1) If q ≤ , then calculate the fitness function as:

[0043] (7)

[0044] (8)

[0045] h L ( X ) = [ g L ( ω L T x 1 + b L ), g L ( ω L T x 2 + b L ),..., g L ( ω L T x N + b L )] T (9)

[0046] In the formula is the network residual of the L-th hidden layer node under the n-th sample, is a function with the regularization parameter real value range being positive; is the output of the L-th hidden layer node, is the activation function of the L-th hidden layer node, , respectively represent the input weight and bias of the L-th hidden layer node; is the introduced value, which is a measurement standard, and its formula is as follows:

[0047] (10)

[0048] where is the regularization parameter;

[0049] (2) If q > , determine the best position of the traveler, that is, the minimum fitness function ;

[0050] Judge It and Maxlt. If It > Maxlt, store the best position at each iteration, obtain the input weights and biases and determine the optimal output weights of all nodes :

[0051] (11)

[0052] Among them, denote the Moore-Penrose generalized inverse matrix of, and H L = [ h 1 , h 2 ,..., h L ] denote the weight matrices of the input layer and the hidden layer, denote the optimal output weights of all nodes, = [ β 1 * , β 2 * ,..., β L * ] T , denote the output weights of all nodes, is the Frobenius norm, O represents the target value of each input sample, K represents the identity matrix, denote the regularization strength.

[0053] Step 2.2.3, compare the number z of travelers with the maximum number of travelers ;

[0054] If z > , then assign the minimum fitness function, i.e., the best position of the traveler, to the (z + 1)-th traveler, and return to Step 2.2.2 to judge It again;

[0055] If z ≤ , obtain the best position of traveler z at the It-th iteration number , and enter Step 2.2.4;

[0056] Step 2.2.4, calculate , is the velocity of traveler z at time t, as shown in the following formula:

[0057] (12)

[0058] is the slope of the path or terrain; and is represented by formula (13):

[0059] (13)

[0060] where and respectively represent the elevation difference and the distance traveled by the hiker; is the inclination angle of the path or terrain; and θ z , t ∈ [ 0 , 50 ] ; and the velocity of traveler z at time t + 1 is:

[0061] (14)

[0062] where is a uniformly distributed number within the range [0, 1]; and represent the speeds of the hiker at times t + 1 and t respectively; is the position of the leading hiker, is the scanning factor of hiker z at time t; then, update the position of the hiker and enter step 2.2.5, as shown in the following formula:

[0063] (15)

[0064] Step 2.2.5, let be the new fitness function, is the current fitness function, if ≤ , then store the best position , and return to step 2.2.3; if > , then directly return to step 2.2.3;

[0065] Step 2.2.6, add new nodes until or the node , where is the allowable error; is the maximum number of hidden neurons;

[0066] Step 2.3, finally obtain the predicted value y of the water wall temperature prediction model as:

[0067] (16)

[0068] The present invention also proposes a method for warning of over-temperature of the water wall temperature of a boiler. Based on the water wall temperature prediction model constructed by the present invention, the water wall temperature is predicted to obtain the water wall temperature; the water wall temperature is compared with the maximum allowable temperature of the water wall material. If the water wall temperature is higher than the maximum allowable temperature of the water wall material, an over-temperature warning signal is issued.

[0069] The present invention also proposes a boiler water wall temperature prediction system, including:

[0070] A data acquisition module for obtaining historical data of auxiliary variables from a distributed control system DCS.

[0071] A preprocessing module for performing abnormal data smoothing replacement and normalization processing on the historical data of the auxiliary variables to obtain preprocessed historical data of the auxiliary variables.

[0072] A delay time optimization module for performing delay processing on the preprocessed historical data of the auxiliary variables to obtain delayed historical data of the auxiliary variables.

[0073] A prediction module uses a water wall temperature prediction model to predict the water wall temperature based on the historical data of the delayed auxiliary variables.

[0074] The present invention also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, the method for predicting the water wall temperature of a boiler according to the present invention is implemented.

[0075] Advantageous effects: In the prior art, feature selection mainly relies on linear analysis methods, ignoring other feature variables with high non-linearity, which limits the expression ability of the prediction model for the predicted temperature. The present invention uses the LightGBM dimensionality reduction method to screen out feature variables with high correlation.

[0076] There is a time delay in the coal grinding system, and the control variables are not synchronized with the wall temperature changes. The present invention uses a combination of the differential evolution DE algorithm and the average mutual information method AIM to handle the time delay problem and ensure high-quality feature input.

[0077] HOA-SCN can select the optimal hyperparameters for each new node in the hidden layer. At the same time, L2 regularization is introduced in the output layer, which not only improves the prediction performance of the model but also enhances the robustness of the model. Description of the Drawings

[0078] Figure 1 is the data processing flow chart of the HOA-SCN model of the present invention;

[0079] Figure 2 is the historical data graph collected;

[0080] Figure 3 is the trend graph of the historical data after abnormal data processing and normalization;

[0081] Figure 4 is the importance graph of the top ten input variables obtained using LightGBM;

[0082] Figure 5 is the delay time between the input variable and the wall temperature estimation graph;

[0083] Figure 6 is the network structure diagram of the stochastic configuration network SCN;

[0084] Figure 7 is the comparison graph of the iterative loss values of HOA-SCN under different activation functions.

[0085] Figure 8 is the input variable graph extracted from the auxiliary variables using MIC;

[0086] Figure 9It is a graph of the evaluation results of the first 10 input variables using RF;

[0087] Figure 10 It is a graph of the prediction performance of HOA-SCN of the present invention under four variable processing methods;

[0088] Figure 11 It is a comparison graph of the prediction results of HOA-SCN with and without delay time;

[0089] Figure 12 It is a comparison graph of the prediction results of HOA-SCN, LSTM, and BP on the test set;

[0090] Figure 13 It is a comparison graph of the prediction errors of HOA-SCN, LSTM, and BP;

[0091] Figure 14 It is a scatter comparison graph of the prediction results of HOA-SCN, GSCN, and SCN on the test set;

[0092] Figure 15 It is a comparison graph of the prediction errors of HOA-SCN, SCN, and GSCN. Detailed implementation manners

[0093] Traditional neural networks lack interpretability. The process of iteratively searching for a set of appropriate weights and biases is time-consuming and computationally intensive, which is particularly difficult for processing large-scale industrial data. The method of the present invention for constructing a neural network structure based on the random configuration network strategy is used to establish a prediction model for the water wall wall temperature, where the water wall wall temperature refers to the metal wall temperature of the boiler heating surface. The random configuration network SCN is an incremental learning method with universal approximation properties. It uses a supervised mechanism to randomly configure the parameters of the hidden layer nodes under a set of inequality constraints to adaptively select their value ranges. This method has the significant advantages of artificial intervention in the network structure and adaptive selection of the number of hidden layer nodes. After determining the hidden layer parameters through the random configuration algorithm, the output weights are calculated based on the pseudoinverse theory. This mechanism makes the training speed of SCN faster than traditional gradient-based iterative algorithms. Currently, the random configuration network SCN has been applied to establish models for various industrial scenarios, and the present invention is the first application of the random configuration network SCN to the modeling of the metal wall temperature of the boiler heating surface.

[0094] To improve the efficiency of the traditional random configuration network SCN in searching for weights and biases and improve the generalization and prediction performance of the model, the present invention introduces the hiking HOA algorithm into the random configuration network to establish a prediction model for the water wall wall temperature of the boiler based on HOA-SCN. HOA-SCN can select the optimal hyperparameters for each newly added node in the hidden layer. At the same time, L2 regularization is introduced in the output layer, which not only improves the prediction performance of the water wall wall temperature prediction model of the boiler but also enhances the robustness of the water wall wall temperature prediction model of the boiler.

[0095] In this specific embodiment, a W-flame boiler with a rated load of 660 MW is selected as the target boiler. The boiler is equipped with 6 double-inlet and double-outlet coal mills, and each coal mill is equipped with 4 double-cyclone pulverized coal burners. 24 pulverized coal burners are arranged in a row on the front and rear wall arches of the lower furnace; 26 over-fire air (OFA) nozzles are also arranged on the upper parts of the front and rear wall water-cooled walls, and the air volume is controlled by 4 air regulators. The secondary air of each burner is individually controlled, and the air distribution device consists of upper and lower air boxes. The secondary air dampers of each burner are adjusted separately, and the specific functions are shown in Table 1.

[0096] Table 1 Functions and operating procedures of secondary air dampers

[0097]

[0098] The wall temperature of the boiler water-cooled wall is also affected by the coal quality. Since the target boiler is not equipped with a corresponding on-line coal quality detection system, the improved coal quality data of the fuel part cannot be synchronized with the sampling period of the auxiliary variables. Therefore, the auxiliary variables do not consider the coal characteristics and it is assumed that the coal quality is stable during the simulation.

[0099] A method for predicting the wall temperature of the water-cooled wall of a pulverized coal boiler provided by the present invention specifically includes the following steps:

[0100] Step 1: Select auxiliary variables that can affect the wall temperature of the water-cooled wall, preprocess the auxiliary variables, evaluate the feature importance of the auxiliary variables to the wall temperature, and select important auxiliary variables as the input features of the boiler water-cooled wall temperature prediction model;

[0101] Step 1.1: Select auxiliary variables that can affect the wall temperature of the water-cooled wall;

[0102] Selecting various medium parameters such as wind, smoke, water, and steam as auxiliary variables for predicting the wall temperature of the water-cooled wall can comprehensively reflect the complex heat exchange process inside the boiler and the influence of various factors on the wall temperature. These parameters provide information from different angles, which helps to capture the dynamic response of the boiler thermal system, thereby improving the accuracy of the water-cooled wall temperature prediction, optimizing the operation control of the boiler, and improving the energy utilization efficiency and the safety of the boiler.

[0103] The adjustment of induced draft and forced draft affects the temperature distribution in the furnace, especially during the combustion process. When the air volume is too small or the air flow distribution is uneven, the heat distribution inside the boiler is uneven, and local overheating or too low wall temperature will occur. Therefore, the changes in air distribution parameters such as air volume and air pressure have an important impact on the water wall wall temperature. The flue gas temperature and flow rate directly affect the heat distribution, and thus affect the temperature of the water wall. The secondary air and the volumetric air of the coal mill are very important control parameters in the boiler, which have a direct impact on the boiler combustion, heat distribution, furnace temperature, and the stability and uniformity of the water wall wall temperature. Taking the secondary air and volumetric air as auxiliary variables of the water wall wall temperature can better reflect the change of the wall temperature and provide valuable information for the operation optimization of the boiler. Based on the above impacts of the boiler furnace on the water wall wall temperature, combustion control laws, and the suggestions of operation engineers, a list of auxiliary variables for dynamic prediction of the wall temperature based on the HOA-SCN model is established, as shown in Table 2, with a total of 60 auxiliary variables. The selection of auxiliary variables is to capture the basic characteristics of the combustion process and consider the impact of key operation factors on the wall temperature.

[0104] Table 2 List of Auxiliary Variables

[0105]

[0106] Step 1.2, collect historical data of auxiliary variables;

[0107] In this specific embodiment, historical data for 7 days was collected from the Distributed Control System (DCS) database of the target boiler at a time granularity of 1 minute. And the data collected in these 7 days covers most situations, such as load increase intervals, load decrease intervals, variable load intervals, and steady state intervals, to comprehensively reflect the change of the water wall wall temperature, and the shutdown interval should be excluded. The original historical data is as Figure 2 shown, with approximately 10,080 sampling points. The first 80% is used as the training set for the water wall wall temperature prediction model of the boiler, and the last 20% is used as the test set.

[0108] Step 1.3, perform data preprocessing on the collected historical data of auxiliary variables;

[0109] Step 1.3.1, abnormal data processing;

[0110] Due to the complex on-site operation environment of the boiler, the signals are interfered by noise, resulting in abnormal values in the original data stored in the DCS. In this specific embodiment, the moving smoothing replacement method is adopted, as shown in Equation (1):

[0111] (1)

[0112] where N is the window size, which is set to 10 in this specific embodiment. is the th data point of the original historical data. is the smoothed data value, representing the smoothed value at time t; replace the original historical data at time t with the smoothed value at time t .

[0113] Step 1.3.2, normalization processing;

[0114] To improve the training efficiency of wall temperature dynamic prediction and avoid error fluctuations in the output layer, the min-max scaling method is used to normalize the auxiliary variables after abnormal data processing:

[0115] (2)

[0116] where , represent the normalized value and the th data point after abnormal data processing respectively, while , represent the maximum and minimum values of the data after abnormal data processing respectively. Figure 3 It shows a partial trend graph of the original data after normalization.

[0117] Step 1.4, dimensionality reduction of the auxiliary variables;

[0118] Light Gradient Boosting Machine (LightGBM) is a gradient-boosted decision tree (GBDT) based algorithm proposed and released by Microsoft in 2017, which is widely used in classification and regression tasks and also widely used in data mining. In this specific implementation manner, the feature importance of the auxiliary variables is evaluated based on LightGBM.

[0119] LightGBM has made several major optimizations on the traditional GDNT algorithm, including the Histogram-based decision tree algorithm, Gradient-based One-Side Sampling (GOSS), Exclusive Feature Bundling (EFB), and the Leaf-wise leaf growth strategy with depth limit. LightGBM has the ability to perform parallel training and can quickly process large amounts of industrial data, so it is suitable for processing industrial objects such as boilers that are non-linear and have large amounts of data. In addition, when calculating feature importance, the LightGBM algorithm has two models, one is Gain and the other is Split. Among them, Split measures the number of times each feature is used in all trees and focuses on the frequency of feature appearance. While Gain measures the contribution of each feature to the model performance, usually obtained by calculating the information gain at the split points of the feature in the tree model. In LightGBM, the greater the gain, the greater the contribution of the feature to the model. This specific implementation mode selects to use the Gain mode to evaluate the feature importance of the auxiliary variables for the wall temperature, as shown in the following formula:

[0120] (3)

[0121] (4)

[0122] Among them, is the reduction in loss (error) brought by feature f (i.e., the metal temperature of the water wall tube wall) during node splitting, measured by the mean squared error (MSE). is the mean squared error of the parent node, and are the mean squared errors of the left child node and the right child node respectively.

[0123] The LightGBM algorithm is used to evaluate the feature importance of the 60 auxiliary variables in Table 2. At the same time, in order to find the optimal hyperparameters, grid search is used for the LightGBM algorithm. It is found from the running results that when the number of decision trees is 500, the tree iteration depth is 7, and the learning rate is 0.05, the effect is the best. Figure 4 Shows the importance of the top 10 feature variables in the Gain mode. The total weight ratio of the importance of the top 10 feature variables (auxiliary variables) reaches 0.9696.

[0124] Analyze the necessity of variables in the modeling of water - cooled wall temperature according to the evaluation results of importance. The main steam pressure is directly related to steam flow, heat exchange efficiency, unit load, etc., and its importance reaches 0.6190. Too high or too low main steam pressure will lead to a decrease in thermal efficiency or boiler damage. Therefore, its impact on the water - cooled wall temperature is very significant. In addition, the fuel quantity directly determines the combustion intensity of the boiler, affecting the heat output and flue gas temperature. When the fuel quantity increases, the heat generated by combustion increases, which will increase the furnace temperature and then affect the wall temperature of the water - cooled wall. For the load, an increase in load usually means more heat output is required, which will increase the furnace temperature and the wall temperature of the water - cooled wall. The positions of the static and moving blades of the induced draft fan control the air flow and air distribution, which have a great impact on the combustion process. Excessive or insufficient air supply will affect the combustion efficiency and the temperature distribution of the boiler. Good air flow ensures complete combustion and avoids local overheating or uneven cooling. The superheat degree affects the steam temperature control of the boiler and then affects the heat conduction characteristics of the water - cooled wall. The secondary air volume of the burner directly affects the combustion efficiency and temperature distribution. Insufficient secondary air volume may lead to incomplete combustion, thus affecting the wall temperature.

[0125] From the analysis of these characteristic variables, the main steam pressure, total fuel quantity, and load are the most direct factors affecting the water - cooled wall temperature of the boiler because they determine the heat output and heat load of the boiler. The secondary air of the burner (such as the secondary air of the static and moving blades of the induced draft fan and the F baffle of burner) optimizes the combustion process by adjusting the air flow, and then controls the local temperature distribution of the boiler. The superheat degree, flue gas temperature at the outlet of the air preheater, etc. reflect the heat exchange efficiency of the boiler, which will also indirectly affect the temperature of the water - cooled wall. Through the above analysis, the importance of the variables calculated by LightGBM follows the variation law of the water - cooled wall temperature of the boiler. Compromising between the model complexity and universality, in this specific implementation manner, the top 10 auxiliary variables in importance are selected as the input variables of the water - cooled wall temperature prediction model, specifically including: main steam pressure, fuel quantity, positions of the static and moving blades of the induced draft fan, load, superheat degree, secondary air of the F baffle of burner B3, secondary air of the F baffle of burner B1, secondary air of the F baffle of burner A3, secondary air of the F baffle of burner F1, flue gas temperature at the outlet of the air preheater.

[0126] Step 2, estimate the variable delay time;

[0127] In the process of coal combustion in the furnace, chemical energy is converted into the thermal energy of water, and there are different degrees of delay times between various auxiliary variables and the wall temperature. Therefore, calculating the delay time between the input variables and the target variable helps to establish a more accurate dynamic prediction of the water-cooled wall temperature. However, estimating the time delay solely from mechanism analysis has uncertainties. Therefore, the present invention uses the average mutual information (AMI) method to calculate the delay time between the input variables and the wall temperature. Due to the interaction and coupling relationship between each input variable, calculating the MI value alone has a large error. Therefore, all input variables need to be considered as a whole.

[0128] Ten input variables are obtained through dimensionality reduction by LightGBM, and the variables input into the water-cooled wall temperature prediction model are defined as , representing the d-th input variable of the i-th input sample, i = is the number of samples. Among them x i , d ( t ) = [ x i , 1 ( t ), x i , 2 ( t ),..., x i , d ( t ),..., x i , 10 ( t )] , and the target variable is represented as . For this specific embodiment, the target variable refers to the water-cooled wall temperature.

[0129] Due to the delay between the target variable and the input variables, the value of the input variable before time is related to , that is and . Among them, is the input variable and the target variable The delay time between them.

[0130] The input variable matrix is reconstructed as shown in the following formula:

[0131] X i ( t ) = [ x ^ i , 1 ( t ), x ^ i , 2 ( t ),..., x ^ i , d ( t ),..., x ^ i , 10 ( t )] (5)

[0132] (6)

[0133] Among them, represents the value of the d-th input variable after reconstruction.

[0134] In order to estimate the optimal delay time, this specific embodiment uses the differential evolution (DE) algorithm to optimize the delay time so that it satisfies the maximum condition of the average mutual information (AMI). This method can perform global search under constraints to find the optimal delay time for each auxiliary variable. The formula for the average mutual information AMI is:

[0135] (7)

[0136] Among them is the mutual information MI at and The value between, and N represents the number of samples.

[0137] Figure 5 is the delay time between the 10 input variables and the target variable obtained. The change of the main steam pressure usually affects the heat transfer efficiency of the boiler relatively quickly. When the main steam pressure increases, the heat output of the boiler increases, which in turn increases the temperature of the water wall. However, due to the physical delay of steam flow and heat transfer, the temperature of the water wall usually lags slightly behind the change of the main steam pressure. After a 2-minute delay, the fluctuation of the main steam pressure will be transmitted to the boiler wall and affect the temperature of the water wall for a period of time; the fuel quantity and load directly determine the combustion intensity and heat output of the boiler. Increasing the fuel quantity or load will immediately increase the heat in the furnace chamber, thereby increasing the temperature of the water wall. However, there is an obvious time delay between the coal feeding process of the coal mill and the heat supply, and it takes a certain amount of time for the combustion heat to conduct from the furnace to the water wall. There is a large lag in the propagation of temperature changes between the furnace chamber of the boiler and various components of the system. The 9-minute delay duration is related to factors such as coal transportation, combustion reaction time, boiler heat capacity, and steam flow regulation. The change of the secondary air will affect the rate and efficiency of the combustion reaction, and it takes some time to be fully reflected in the heat load of the boiler and the temperature of the water wall; the heat release and temperature distribution during the combustion process gradually stabilize and it takes several minutes to be transmitted to the water wall. Coupled with the fact that the temperature change in the boiler usually has a certain inertia, especially in large boilers where heat transfer from the furnace to the water wall and other components takes some time, the influence of secondary air regulation on the temperature of the water wall will not be immediately apparent. Therefore, it is completely reasonable that the delay time of the secondary air is 1 to 4 minutes.

[0138] Step 3, construct a HOA-SCN water wall temperature prediction model for temperature prediction, and input historical data into the water wall temperature prediction model to train the model;

[0139] The Hiking Optimization Algorithm is inspired by hiking, identifying the similarity between the search landscape of the optimization problem and the mountainous terrain traversed by hikers. In the Hiking Optimization Algorithm (HOA), by simulating the characteristics involved in a hiking adventure, the ultimate goal is to reach the peak.

[0140] Step 3.1, construct auxiliary variables and input them into the water wall temperature model of the present invention, as Figure 1 shown, including the following steps:

[0141] Step 3.1.1, given a set of training data , where is the input sample of HOA-SCN, where represents the i-th sample input into the SCN, , in this specific embodiment, D = 10, representing ten auxiliary variables; the output sample is , . Where N is the number of samples, D is the dimension of the input sample, and m is the dimension of the output sample. For this specific embodiment, the output sample of this application is the predicted water wall temperature, which is one-dimensional data, so m = 1.

[0142] Step 3.1.2, given the objective function , assuming that L - 1 neural nodes in the hidden layer of the SCN have been constructed, then the network output expression of the current hidden layer is:

[0143] (8)

[0144] Where β j = [ β j , 1 ,..., β j , m ] T represents the output weight of the j - th node in the hidden layer; represents the activation function; and are the input weight and bias of the j - th node in the hidden layer respectively.

[0145] Step 3.1.3, as Figure 6 shown, calculate the input weight of the j - th node in the hidden layer and the bias of the j - th node in the hidden layer by the Hiking Optimization Algorithm (HOA), and then determine the optimal output weight , the specific steps are as follows:

[0146] Step 3.1.3.1, initialize the position of the traveler z, as shown in the following equation:

[0147] (9)

[0148] Where is a uniformly distributed number in the range [0, 1]; represents the current position of the traveler z at time t; and represent the upper and lower bounds of the j - th dimension of the decision variable of the optimization problem.

[0149] Step 3.1.3.2, if q ≤ , calculate the fitness function , and enter Step 3.1.3.3;

[0150] q represents the q - th traveler, , is the population size, that is, the number of travelers, which is set to 60 in this paper;

[0151] If q > , determine the best position of the traveler, i.e., the minimum fitness function , the last element in the minimum fitness function is the bias , and the transpose of the other elements is the input weight , and when It > Maxlt, It is the current iteration number, Maxlt is the maximum iteration number, and then the optimal output weights of all nodes are determined through formula (14) , enter step 3.1.3.6;

[0152] If q > , and when It ≤ Maxlt, It is the current iteration number, Maxlt is the maximum iteration number, enter step 3.1.3.3;

[0153] (1) If q ≤ , then calculate the fitness function as:

[0154] (10)

[0155] (11)

[0156] h L ( X ) = [ g L ( ω L T x 1 + b L ), g L ( ω L T x 2 + b L ),..., g L ( ω L T x N + b L )] T (12)

[0157] In the formula is the network residual of the L-th hidden layer node under the n-th sample, is a function with the regularization parameter real value range being positive. is the output of the L-th hidden layer node, is the activation function of the L-th hidden layer node, , respectively represent the input weight and bias of the L-th hidden layer node. is a specially introduced value, a measurement standard for judging whether each candidate node meets specific conditions. Its formula is as follows:

[0158] (13)

[0159] Where is the regularization parameter.

[0160] (2) If q > , determine the best position of the traveler, i.e., the minimum fitness function .

[0161] Judge It and Maxlt. If It > Maxlt, where It is the current iteration number and Maxlt is the maximum iteration number, then store the best position at each iteration and determine the output weights of all nodes. :

[0162] (14)

[0163] Where, denotes the Moore-Penrose generalized inverse matrix of, and H L = [ h 1 , h 2 ,..., h L ] denotes the weight matrix of the input layer and the hidden layer, denotes the optimal output weights of all nodes, = [ β 1 * , β 2 * ,..., β L * ] T , denotes the output weights of all nodes, is the Frobenius norm, O represents the target value of each input sample, K represents the identity matrix, denotes the regularization strength.

[0164] Step 3.1.3.3, judge z and . z is the number of travelers, the maximum number of travelers.

[0165] If z > , then assign the minimum fitness function, i.e., the best position of the traveler, to the (z + 1)-th traveler, and return to Step 3.1.3.2 to judge It again.

[0166] If z ≤ , obtain the best position of traveler z at the It-th iteration , and enter Step 3.1.3.4

[0167] Step 3.1.3.4, calculate , where is the velocity of traveler z at time t, as shown in the following formula:

[0168] (15)

[0169] is the slope of the path or terrain. And is represented by formula (16):

[0170] (16)

[0171] Where, and respectively represent the elevation difference and the distance traveled by the hiker; is the inclination angle of the path or terrain. And θ z , t ∈ [ 0 , 50 ] . The speed of traveler z at time t + 1 is:

[0172] (17)

[0173] where is a uniformly distributed number within the range [0, 1]; and respectively represent the speeds of the hiker at times t + 1 and t; is the position of the leading hiker, is the scanning factor (SF) of hiker z at time t. Then, update the position of the traveler and enter step 3.1.3.5, as shown in the following formula:

[0174] (18)

[0175] Step 3.1.3.5, let be the new fitness function, is the current fitness function. If ≤ , then store the best position , and return to step 3.1.3.3. If > , then directly return to step 3.1.3.3.

[0176] Step 3.1.3.6, add new nodes until is satisfied or the number of nodes , where is the allowable error; is the maximum number of hidden neurons.

[0177] Finally, the predicted value y of the water wall temperature prediction model is:

[0178] (19)

[0179] Step 3.2, regarding the ill - posed problem of the stochastic configuration network SCN

[0180] During the construction of the stochastic configuration network SCN, the input weights and biases of the neural network are randomly assigned under the inequality supervision mechanism. The output weights of the stochastic configuration network SCN are calculated using the least - squares method. The output weights of the stochastic configuration network SCN are:

[0181] (20)

[0182] There is a defect in the randomly configured network SCN. If the dimension of the input data is high and the amount of data is large, it will cause an increase in the number of hidden layer nodes in the randomly configured network SCN model, resulting in a very easy occurrence of collinearity among the output matrix variables of the hidden layer nodes, leading to an ill-conditioned problem. Furthermore, the generalization ability of the randomly configured network SCN model will drop sharply, resulting in overfitting. Therefore, in this specific implementation, the L2 regularization method is used to improve the output weights of the randomly configured network SCN to alleviate the ill-conditioned problem of the randomly configured network SCN. The optimal output weights of all nodes in the randomly configured network SCN Changed to:

[0183] (21)

[0184] Step 3.3 Optimization of the parameters of the water wall temperature prediction model

[0185] After being verified by multiple simulation experiments, in the HOA-SCN, the maximum number of iterations for hiking HOA is set to 20 times, and the population size is set to 60 times, Set to 0.9999999, and the regularization coefficient is set to 0.05. The activation function determines the activation state of the neuron and allows the model to map complex non-linear relationships. In this specific implementation, the three activation functions Tanh, Sigmoid, and ReLU are used to compare and verify the model loss function values on the validation set and the training set. As Figure 7 shown, Figure 7 In (a) is the curve graph of the model loss function values on the validation set and the training set obtained when using the Sigmoid activation function, Figure 7 In (b) is the curve graph of the model loss function values on the validation set and the training set obtained when using the ReLU activation function, Figure 7 In (c) is the curve graph of the model loss function values on the validation set and the training set obtained when using the Tanh activation function. The loss curve shows that the Sigmoid activation function is more suitable for HOA-SCN.

[0186] Step 4, input the auxiliary parameters into the HOA-SCN water wall temperature prediction model for temperature prediction.

[0187] Considering that the impact of adjusting the combustion conditions in the furnace on the wall temperature has a certain time delay, the present invention calculates the temperature rise δt in the next 2 minutes according to the average rate (°C / min) of the temperature rise of the water-cooled wall metal wall temperature measurement point in the most recent 1 minute, and then adds δt to the water-cooled wall metal wall temperature obtained from the prediction model to obtain the overtemperature warning temperature; the obtained overtemperature warning temperature is compared with the maximum allowable temperature of the water-cooled wall material for judgment. When the value of the overtemperature warning temperature is greater than or equal to the maximum allowable temperature threshold of the material, an overtemperature warning signal is sent to remind the boiler operator to take measures to prevent overtemperature as early as possible (such as reducing the fuel input and increasing the oxygen setting value), that is, changing the air-coal ratio from the combustion side, thereby adjusting the combustion conditions and reducing the number of times of overtemperature of the water-cooled wall metal wall temperature of the boiler. For the secondary air damper, adjust the opening of the F-layer secondary air damper above the furnace, and at the same time dynamically increase the bias of the feed water flow rate setting value (and restore it after a short period of time), so as to affect the fuel-water ratio and further reduce the water-cooled wall temperature. The present invention is beneficial to promoting the safe operation of power plant boilers and extending the service life of equipment.

[0188] Correspondingly, the present invention also provides a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for predicting the water-cooled wall temperature of the boiler according to the present invention are realized.

[0189] To evaluate the performance of the HOA-SCN water-cooled wall temperature prediction model, different modeling methods were compared, and the average value and standard deviation of the model prediction error were obtained from 20 independent simulation experiments.

[0190] (1) Comparison of dimensionality reduction methods

[0191] Based on HOA-SCN, the influence of whether there is LightGBM feature dimensionality reduction and other dimensionality reduction methods on the prediction performance was compared. When not dimensionally reduced, there are 59 input variables, and the other dimensionality reduction methods use the mainstream MIC and RF dimensionality reduction methods. For the convenience of comparison, 10 features are also selected as the input variables. Figure 8 and Figure 9 are the first ten variables under MIC and RF feature dimensionality reduction respectively.

[0192] The mean values of the prediction performance of HOA-SCN under 4 variable processing methods are as Figure 10 shown. Table 3 shows the mean values and standard deviations of the performance of 20 tests. From the results, the LightGBM dimensionality reduction method is better than MIC and RF, which indicates that the LightGBM dimensionality reduction method is better than the other two in dealing with non-linear relationships and feature interactions. And the training time under dimensionality reduction is significantly faster than that without dimensionality reduction.

[0193] Table 3 Evaluation indexes of HOA-SCN under four variable processing methods

[0194]

[0195] (2) Delay time estimation

[0196] To verify the influence of the delay time of input variables on the model prediction performance, a comparative experiment with and without delay time was carried out based on HOA-SCN. Figure 11 is the prediction curve for variables with and without delay time, and Table 4 shows the corresponding evaluation indexes. Figure 11 In (a) is the comparison curve between the predicted value and the true value of the variable with delay time, Figure 11 In (c) is the comparison curve between the predicted value and the true value of the variable without delay time. The prediction effect considering time delay is significantly better than that without considering time delay. Figure 11 In (b) is the absolute error between the predicted value and the true value when the variable has delay time, Figure 11 In (d) is the absolute error between the predicted value and the true value when the variable has no delay time. When considering time delay, the absolute error between the predicted value and the true value of half of the sampling points is lower than 1.10439 °C, while for the case without considering time delay, this data is 1.43922 °C. Therefore, considering the delay time of input variables can effectively improve the prediction performance of the model.

[0197] Table 4 Prediction performance of HOA-SCN with and without delay time of input variables

[0198]

[0199] (3) Different modeling methods

[0200] To evaluate the generalization performance of HOA-SCN, the traditional model BP neural network and the long short-term memory neural network LSTM were compared with HOA-SCN. The hyperparameters of the BP network and the long short-term memory neural network LSTM are shown in Table 5.

[0201] Table 5 Hyperparameter settings of BP and LSTM neural networks

[0202]

[0203] Figure 12 In (a) is the prediction curve of the water wall wall temperature of the HOA-SCN, LSTM, and BP models, Figure 12 In (b) is the corresponding unit load curve, Figure 12 In (c) is the prediction curve of the water wall wall temperature in the load reduction interval, Figure 12 In (d) is the prediction curve of the water wall wall temperature in the load increase interval, Figure 12 In (e) is the prediction curve of the water wall wall temperature under the stable load state. From Figure 12It can be found from (c), (d) and (e) that when the unit load decreases, the wall temperature also decreases accordingly. When the load increases, the wall temperature of the boiler water wall also increases accordingly. When the load is stable, the wall temperature tends to fluctuate smoothly. From the operation mechanism of the boiler, a decrease in the unit load means a weakening of the combustion intensity in the furnace and a reduction in the heat generated by combustion. Moreover, a decrease in load is usually accompanied by the adjustment of the induced draft fan and forced draft fan, and the air flow and flue gas flow of the boiler also change. These factors cause the wall temperature to decrease. The reason when the load increases is similar to when it decreases. When the load is stable, the internal heat load and combustion process of the boiler tend to be stable, the fluctuation of the wall temperature will decrease, and it will tend to a stable state within a period of time. The thermal inertia and adjustment mechanism of the boiler system keep the temperature of the water wall within a relatively stable range, but there may still be slight fluctuations. From Figure 12 It can be clearly seen from the error statistics of the three models in (f), (g) and (h) that the error values of HOA-SCN are the most and densest at the sample points near 0, followed by LSTM, and finally BP. The traditional model BP has a simple structure and limited ability to extract variable features, resulting in poor prediction performance. Due to its own memory module, LSTM will cause the prediction error to gradually accumulate, thereby affecting the prediction performance.

[0204] Figure 13 For the prediction errors of the three models, as Figure 13 shown in (b), the RMSE of HOA-SCN is 2.06043±0.02241℃, the MAE is 1.62609±0.03779℃, the MAPE is 0.42372±0.01128%, and R 2 is 0.97553±0.00053. As Figure 13 shown in (a), the RMSE of BP is 4.87107±1.26054℃, the MAE is 3.90565±1.07193℃, the MAPE is 0.97775±0.35495%, and R2 is 0.85407±0.07467. As Figure 13 shown in (c), the RMSE of LSTM is 2.68964±0.47341℃, the MAE is 2.18015±0.46092℃, the MAPE is 0.56336±0.11967%, and R 2 is 0.95701±0.0159. Compared with BP, the various prediction indicators of HOA-SCN have increased by 57.70%, 58.36%, 56.67%, and 12.45% respectively. Compared with LSTM, the various prediction indicators of HOA-SCN have increased by 23.39%, 25.41%, 24.79%, and 1.90% respectively.

[0205] (4) Comparison under the SCN architecture

[0206] To verify the impact of the hiking HOA algorithm and L2 regularization on the model prediction performance, this specific implementation conducted a comparative experiment on the baseline model SCN, the optimized model GSCN, and HOA-SCN. To ensure the accuracy of the comparison results, the hyperparameters of the three are kept consistent. Among them, the shrinkage factor in HOA-SCN and GSCN is 0.999999 for both, and the in GSCN is set to 10 -6 (Note that here GSCN-II is used for GSCN instead of GSCN-I). For SCN, it is selected from the set .

[0207] Figure 14 Figure (c) in Figure 14 shows that the experimental data is closely close to both sides of the Perfect line, and the 95% confidence band interval is very narrow, indicating that the predicted values of HOA-SCN are closer to the true values. From Figure 14 Figures (a) and (b) in it, it can be clearly found that the confidence bands of SCN and GSCN are relatively wide, and the predicted points of SCN deviate far from the true values, indicating that the prediction effect of SCN is not good. From the slopes of the fitting curves, it can also be seen that the slopes of HOA-SCN, GSCN, and SCN are 1.03, 0.93, and 0.80 respectively, indicating that the generalization ability of HOA-SCN is the strongest and that of SCN is the weakest.

[0208] Figure 15 For the evaluation indicators of the three SCN models, the prediction effect of HOA-SCN is better than that of SCN and GSCN. Compared with SCN, the various prediction indicators (RMSE, MAE, MAPE, R 2 ) of HOA-SCN are improved by 54.05%, 49.52%, 56.66%, and 10.23% respectively. It can be seen that the hiking HOA algorithm effectively improves the prediction performance of the SCN model.

[0209] To accurately predict the wall temperature of the boiler water wall under variable working conditions, this specific implementation proposes a new method based on the LightGBM dimensionality reduction method, delay time estimation, and HOA-SCN model. This method has a simple and effective network structure and can significantly improve the prediction performance. To evaluate the generalization ability of the model, a simulation experiment was conducted on an in-service supercritical 660MW tangentially fired boiler. The conclusions are as follows:

[0210] (1) Through the comparative experiment with and without dimensionality reduction by LightGBM, it is found that the LightGBM dimensionality reduction method can not only improve the training speed but also enhance the model accuracy. The comparative experiment with different dimensionality reduction methods such as MIC and RF shows that the LightGBM dimensionality reduction method is more suitable for the data of non-linear and high-dimensional industrial objects like boilers.

[0211] (2) The comparative experiment with and without delay time shows that estimating the delay time of input variables can improve the model prediction accuracy.

[0212] (3) Although the traditional BP neural network modeling method has high performance in dealing with steady-state load, when the load changes, the wall temperature changes with the load, and the traditional modeling method has limited ability to capture variable characteristics, resulting in low prediction accuracy under transient load. The memory module of LSTM will lead to the accumulation of prediction errors, thus reducing the prediction performance. HOA-SCN shows good prediction performance both in steady state and transient state.

[0213] (4) Introducing the HOA optimization algorithm into SCN significantly improves the prediction accuracy without significantly increasing the training complexity. Compared with SCN and GSCN, the RMSE of HOA-SCN is improved by 54.05% and 43.76% respectively. 2 They are improved by 10.23% and 5.39% respectively.

Claims

1. A method for predicting boiler water wall temperature, characterized in that: The steps include: Step 1, selecting auxiliary variables that have an important influence on the water-cooled wall temperature, wherein the auxiliary variables are D in total; Step 2: Build a water-cooled wall temperature prediction model. The steps are as follows: Step 2.1, given the objective function , assuming that the hidden layer of the random configuration network SCN has been constructed with L-1 neural nodes; among them, represents the dimensional space of the input, represents the dimensional space of the output, D is the dimension of the input sample, that is, the number of auxiliary variables, and m is the dimension of the output sample; Step 2.2, calculate the input weight of the jth node in the hidden layer and the bias of the jth node in the hidden layer , and then get the optimal output weights of all nodes in SCN ; Where 1≤j≤L-1; Step 2.3, water wall temperature prediction model prediction value for: ； in, represents the weight matrix of the input layer and hidden layer in SCN, represents the weight matrix of the input layer in SCN, represents the weight matrix of the L-1th hidden layer in SCN; Step 3: training the water-cooled wall temperature prediction model to obtain a trained water-cooled wall temperature prediction model for predicting the water-cooled wall temperature.

2. The method for predicting boiler water wall temperature according to claim 1, characterized in that: The auxiliary variables that have an important influence on the water-cooled wall temperature are selected in step 1. The specific steps are as follows: Step 1.1, collect historical data of auxiliary variables; Step 1.2, performing data preprocessing on the collected auxiliary variable historical data, wherein the preprocessing includes abnormal data processing and normalization processing; Step 1.3, use the Gain mode in the LightGBM algorithm to evaluate the characteristic importance of the auxiliary variables to the wall temperature, and select the auxiliary variables with the top D characteristic importance as the auxiliary variables that have an important impact on the water-cooled wall temperature.

3. The method for predicting boiler water wall temperature according to claim 2, characterized in that: The characteristic importance of the auxiliary variables to the wall temperature is evaluated using the following formula: （1） （2） in, It is the reduction in the loss of the water-cooled wall tube metal temperature caused by the node splitting, measured by the mean square error; is the mean square error of the parent node, and is the mean square error between the left and right child nodes.

4. The method for predicting boiler water wall temperature according to claim 1, characterized in that: The auxiliary variables selected in step 1 include main steam pressure, fuel quantity, position of stationary and moving blades of the induced draft fan, load, superheat, B3 burner F baffle secondary air, B1 burner F baffle secondary air, A3 burner F baffle secondary air, F1 burner F baffle secondary air, and air preheater outlet flue gas temperature; the total number of auxiliary variables is D=10.

5. The method for predicting boiler water wall temperature according to claim 1, characterized in that: The auxiliary variable in step 1 is the auxiliary variable reconstructed after considering the delay time between the auxiliary variable and the water-cooled wall temperature; The i-th variable input to the water-cooled wall temperature prediction model is defined as ,i= is the number of samples; , the target variable is expressed as ; Since there is a delay between the target variable and the input variable, Indicates the dth input variable of the input With the target variable The delay time between The input variables are reconstructed as shown below: （3） （4） in, Represents the reconstructed value of the d-th input variable.

6. The method for predicting boiler water wall temperature according to claim 5, characterized in that: Delay time The calculation method is as follows: The average mutual information (AMI) method combined with the differential evolution (DE) algorithm is used to calculate the delay time between each auxiliary variable and the water-cooled wall temperature.

7. The method for predicting boiler water wall temperature according to claim 1, characterized in that: The steps for constructing the water-cooled wall temperature prediction model described in step 2 are as follows: Step 2.1, given the objective function , assuming that L-1 neural nodes of the hidden layer of SCN have been constructed, then the network output expression of the current hidden layer is: （5） in represents the output weight of hidden layer node j; represents the activation function; and are the input weight and bias of the jth node in the hidden layer, Represents the input of the current hidden layer; Step 2.2, calculate the input weight of the jth node in the hidden layer through the hiking algorithm HOA and the bias of the jth node in the hidden layer , and then get the optimal output weight , the specific steps are as follows: Step 2.2.1, Initialize the traveler position Initialize the traveler z's position , as shown in the following equation: （6） in is a uniformly distributed number in the range [0,1]; represents the current position of traveler z at time t; and Represents the upper and lower bounds of the j-th dimension of the decision variables of the optimization problem; Step 2.2.2, compare q with , q represents the qth traveler, , is the population size, i.e. the number of travelers; If q≤ When , calculate the fitness function , and proceed to step 2.2.3; If q> , determine the best position of the traveler, that is, the minimum fitness function , the last element in the minimum fitness function is the bias , and the transpose of the other elements is the input weight , and It>Maxlt, It is the current iteration number, Maxlt is the maximum iteration number, so the optimal output weights of all nodes are determined by formula (11) , go to step 2.2.6; If q> , and It≤Maxlt, go to step 2.2.3; (1) If q≤ Then calculate the fitness function for: （7） （8） （9） In the formula is the network residual of the Lth hidden layer node under the nth sample, is a function whose regularization parameter has a positive real-valued domain; is the output of the Lth hidden layer node, is the activation function of the Lth hidden layer node, , Represent the input weight and bias of the Lth hidden layer node respectively; is an introduced value and a measure, and its formula is as follows: （10） in is the regularization parameter; (2) If q> , determine the best position of the traveler, that is, the minimum fitness function ; The number of iterations It and the maximum number of iterations Maxlt are judged. If It>Maxlt, the best position is stored in each iteration, the input weight and bias are obtained, and the optimal output weight of all nodes is determined. : （11） in, express The Moore-Penrose generalized inverse matrix of represents the weight matrix of the input layer and the hidden layer, represents the optimal output weight of all nodes, = , represents the output weights of all nodes, is the Frobenius norm, O represents the target value of each input sample, K represents the identity matrix, represents the regularization strength; Step 2.2.3, the number of travelers z and the maximum number of travelers Make comparisons; If z> , then the minimum fitness function, i.e. the best position of the traveler, is assigned to the z+1th traveler, and the process returns to step 2.2.2 to determine the number of iterations It again; If z≤ , get the best position of traveler z at the Itth iteration , proceed to step 2.2.4 Step 2.2.4, calculate , is the speed of traveler z at time t, as shown in the following formula: （12） is the slope of the path or terrain; and It is expressed by formula (13): （13） in, and represent the elevation difference and the distance covered by the hiker, respectively; is the slope angle of the path or terrain; and ; and the speed of traveler z at time t+1 is: （14） in, is a uniformly distributed number in the range [0,1]; and denote the speed of the hiker at time t+1 and time t respectively; is the location of the lead hiker, is the scan factor of hiker z at time t; then, update the traveler's position And go to step 2.2.5, as shown in the following formula: （15） Step 2.2.5, set is the new fitness function, is the current fitness function, if ≤ , then store the best position , and return to step 2.2.3; if > , then directly return to step 2.2.3; Step 2.2.6, add new nodes until or Node ,in is the allowable error; is the maximum number of hidden neurons; Step 2.3, finally the predicted value y of the water-cooled wall temperature prediction model is: （16）。 8. A boiler water-cooled wall overtemperature early warning method, characterized in that: Based on the water-cooled wall temperature prediction model constructed in the prediction method according to claim 7, the water-cooled wall temperature is predicted to obtain the water-cooled wall temperature; The water-cooled wall temperature is compared with the maximum allowable temperature of the water-cooled wall material. If the water-cooled wall temperature is higher than the maximum allowable temperature of the water-cooled wall material, an over-temperature warning signal is issued.

9. Boiler water wall temperature prediction system, characterized in that: Based on the water-cooled wall temperature prediction model constructed in the water-cooled wall temperature prediction method according to claim 7, the water-cooled wall temperature prediction system includes: Data acquisition module, used to obtain auxiliary variable historical data from distributed control system DCS; A preprocessing module, used for performing abnormal data smoothing replacement and normalization processing on the auxiliary variable historical data to obtain preprocessed auxiliary variable historical data; The delay time optimization module is used to perform delay processing on the preprocessed auxiliary variable historical data to obtain the delayed auxiliary variable historical data; The prediction module uses the water-cooled wall temperature prediction model to predict the water-cooled wall temperature according to the delayed auxiliary variable historical data.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the program is executed by the processor, the method for predicting the water-cooled wall temperature of a boiler according to any one of claims 1 to 7 is implemented.

Images