On-line monitoring method and system for ash deposition and coking of heating surface of boiler

Abstract

The invention provides an online monitoring method and system for ash deposition and coking of a boiler heating surface. The method comprises the following steps that preprocessed historical operation data are obtained; a BP neural network is adopted to construct a plurality of advanced multi-step prediction models based on the preprocessed historical operation data; a genetic algorithm is adopted for optimization to obtain an optimal prediction model of the heat exchange amount of the boiler clean heating surface at the future moment; on the basis of the optimal prediction model, a boiler clean heating surface heat exchange amount prediction value at the future moment is obtained; the real-time operation heat exchange amount of the boiler heating surface is obtained; and according to the heat exchange amount predicted value of the clean heating surface of the boiler and the real-time operation heat exchange amount of the heating surface of the boiler, a cleaning coefficient reflecting the ash coking degree of the heating surface of the boiler is obtained. According to the method, the low-error prediction of the ash deposition and coking degree of the boiler heating surface under the boiler operation data of the multi-influence-factor data set is realized, and a theoretical basis is provided for guaranteeing the stable and safe operation of the boiler.

Description

Technical Field

[0001] The present application relates to the technical field of monitoring ash and coking on boiler heating surfaces, and in particular to an online monitoring method and system for ash and coking on boiler heating surfaces. Background Art

[0002] my country's overall energy structure is characterized by "more coal", "less oil" and "poor gas". Compared with other countries, my country's energy production and consumption rely to a large extent on coal resources. In the context of energy shortages such as fossil resources, in order to obtain greater economic benefits and ensure safe power generation, major domestic coal-fired power plants are actively using lower-priced low-grade coal. These coals generally have high ash content and low ash melting point, which will greatly increase the probability and degree of ash accumulation and coking on the heating surface of the boiler, reduce the heat transfer efficiency of the heating surface, affect the boiler load, and even cause serious accidents.

[0003] Therefore, it is of great significance to study a model for monitoring the degree of ash coking on the boiler heating surface. At the same time, changes in boiler operating parameters (such as coal feed, load, etc.) will not affect the change in boiler heat exchange at the same time, but will be reflected in the heat exchange at a certain time in the future. Therefore, parameters considering the time dimension are more practical.

[0004] A search of current technologies found that the technologies for monitoring the degree of ash and coking on the heating surface of boilers include:

[0005] (1) Using sensors to directly monitor the ash and coking conditions on the heating surface, but the sensors are prone to failure in high temperature environments and are expensive, making them unsuitable for large-scale installation on the heating surfaces of boilers.

[0006] (2) Use machine learning and other models for learning monitoring. For example, a boiler heating area dust monitoring method based on time series prediction is proposed using the LSTM (Long Short-Term Memory) model. Although the LSTM model has good processing capabilities for time series data, it has poor prediction capabilities for data sets with multiple influencing factors. Summary of the invention

[0007] The present application provides a method and system for online monitoring of ash and coking on the heating surface of a boiler, which can solve the technical problems that the sensors in the prior art for monitoring the degree of ash and coking on the heating surface of the boiler are prone to failure in a high temperature environment and are expensive, and are not suitable for large-scale installation and use on the heating surfaces of the boiler; the LSTM model has good processing capabilities for time series data, but has poor prediction capabilities for data sets with multiple influencing factors.

[0008] In a first aspect, the present application provides an online monitoring method for ash and coking on a heating surface of a boiler, comprising the following steps:

[0009] Obtain the historical operation data of the boiler within a preset time after the power plant DCS system obtains the soot blowing operation, perform data preprocessing on the historical operation data, and obtain the preprocessed historical operation data;

[0010] Based on the pre-processed historical operation data, BP neural network is used to build multiple multi-step-ahead prediction models;

[0011] Assign weights to multiple multi-step-ahead prediction models, use genetic algorithms to find the best weight combination, and obtain the optimal prediction model for the heat exchange of the clean heating surface of the boiler at future moments based on the best weight combination;

[0012] Based on the optimal prediction model, obtain the predicted value of the clean heating surface heat exchange of the boiler at a future time;

[0013] The real-time heat exchange of the boiler heating surface is calculated based on the real-time operation data of the boiler;

[0014] According to the predicted value of the heat exchange of the clean heating surface of the boiler and the real-time heat exchange of the heating surface of the boiler, the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler is obtained.

[0015] In combination with the first aspect, in one embodiment, the obtaining of the boiler historical operation data within a preset time after the soot blowing operation by the power plant DCS system, performing data preprocessing on the historical operation data, and obtaining the preprocessed historical operation data specifically includes the following steps:

[0016] Obtain the historical operation data of the boiler within the preset time of the power plant system after the soot blowing operation;

[0017] Delete discrete values ​​in the historical operation data, fill in missing data in the historical operation data, and perform data standardization to obtain preprocessed historical operation data.

[0018] In combination with the first aspect, in one implementation, the use of a BP neural network to construct multiple multi-step-ahead prediction models based on the preprocessed historical operation data specifically includes the following steps:

[0019] The enthalpy value is queried using the inlet and outlet working fluid temperature and pressure data of the boiler heating surface in the unstandardized historical operation data, and the heat exchange of the boiler's historical clean heating surface is calculated;

[0020] The preprocessed historical operation data is used as the model input, and the corresponding historical clean heating surface heat exchange of the boiler is used as the model output to construct multiple multi-step-ahead prediction models.

[0021] In combination with the first aspect, in one embodiment, the pre-processed historical operation data is used as the model input, and the corresponding boiler historical clean heating surface heat exchange rate is used as the model output to construct multiple multi-step-ahead prediction models, which specifically includes the following steps:

[0022] The pre-processed historical operation data and the corresponding historical clean heating surface heat exchange of the boiler are used as the data set;

[0023] Divide the dataset into training set and test set;

[0024] The pre-processed historical operation data in the training set is used as the model input, and the corresponding boiler historical clean heating surface heat exchange is used as the model output. The BP neural network is used to build multiple multi-step-ahead prediction models.

[0025] Based on the test set, the structural parameters and training parameters of the BP neural network model are adjusted to obtain the regression R value of each multi-step-ahead prediction model until the regression R value of the multi-step-ahead prediction model is not greater than the preset R value;

[0026] Obtain multiple multi-step-ahead forecasting models whose regression R values ​​are greater than the preset R values.

[0027] In combination with the first aspect, in one implementation, weights are assigned to a plurality of multi-step-ahead prediction models, a genetic algorithm is used to find the best weight combination, and an optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment is obtained based on the best weight combination, specifically comprising the following steps:

[0028] Assign weights to the multiple multi-step-ahead prediction models constructed, perform weighted summation, and obtain a hybrid model;

[0029] Construct the optimal objective function for weight combination in the hybrid model;

[0030] Optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model;

[0031] Based on the optimal weight combination of the hybrid model, the optimal prediction model for the heat exchange of the clean heating surface of the boiler at future moments is obtained.

[0032] In combination with the first aspect, in one implementation, constructing an optimization objective function for weight combinations in a hybrid model specifically includes the following steps:

[0033] The root mean square error of the heat transfer of the boiler heating surface output by the hybrid model is the objective function.

[0034] In combination with the first aspect, in one implementation, optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model specifically includes the following steps:

[0035] The root mean square error of the mixed model is taken as the dependent variable and the weight is the independent variable;

[0036] The non-dominated sorting genetic algorithm is used to find the optimal solution of the objective function. The optimization parameters include the non-dominated solution that is expected to be maintained in the population, the size of the population, the number of algorithm iterations, the number of overfitting verification failures and the objective function tolerance, and the best weight combination of the hybrid model is obtained.

[0037] In combination with the first aspect, in one embodiment, the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler is obtained according to the predicted heat exchange value of the clean heating surface of the boiler and the real-time operating heat exchange value of the heating surface of the boiler, as shown in the following formula:

[0038]

[0039] Where k is the cleanliness coefficient; ΔQ 实时运行 The real-time heat exchange rate of the boiler heating surface; ΔQ 洁净预测 It is the predicted value of heat exchange of clean heating surface of boiler.

[0040] In a second aspect, the present application provides an online monitoring system for ash and coking on a boiler heating surface, comprising:

[0041] A historical operation data acquisition module is used to acquire the historical operation data of the boiler within a preset time after the DCS system of the power plant obtains the soot blowing operation, preprocess the historical operation data, and acquire the preprocessed historical operation data;

[0042] A multi-step-ahead prediction model construction module is communicatively connected to the historical operation data acquisition module and is used to construct multiple multi-step-ahead prediction models using a BP neural network based on the preprocessed historical operation data;

[0043] An optimal prediction model acquisition module is connected to the multi-step-ahead prediction model construction module, and is used to assign weights to multiple multi-step-ahead prediction models, use a genetic algorithm to find the best weight combination, and obtain the optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment based on the best weight combination;

[0044] A boiler clean heating surface heat exchange amount acquisition module is connected to the optimal prediction model acquisition module and is used to obtain a predicted value of the boiler clean heating surface heat exchange amount at a future time based on the optimal prediction model;

[0045] The operating heat exchange amount acquisition module is used to calculate the real-time operating heat exchange amount of the boiler heating surface according to the real-time operating data of the boiler;

[0046] The cleanliness coefficient acquisition module is communicatively connected with the boiler clean heating surface heat exchange acquisition module and the operating heat exchange acquisition module, and is used to obtain the cleanliness coefficient of the reaction boiler heating surface ash and coking degree according to the predicted value of the boiler clean heating surface heat exchange and the real-time operating heat exchange of the boiler heating surface.

[0047] In conjunction with the second aspect, in one implementation, the optimal prediction model acquisition module includes:

[0048] A hybrid model acquisition unit is used to assign weights to the constructed multiple multi-step-ahead prediction models to perform weighted summation to obtain a hybrid model;

[0049] An objective function construction unit, which is in communication connection with the hybrid model acquisition unit and is used to construct an optimization objective function for weight combination in the hybrid model;

[0050] An optimization unit, which is in communication connection with the objective function construction unit, and is used to optimize the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model;

[0051] The optimal prediction model acquisition unit is connected to the optimization unit for obtaining the optimal prediction model of the heat exchange of the clean heating surface of the boiler at a future moment based on the optimal weight combination of the hybrid model.

[0052] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0053] By using BP neural network to construct a multi-step ahead prediction model based on time series, and then using genetic algorithm to find the weighted optimal solution of each multi-step ahead prediction model, the future heat exchange capacity of the clean heating surface of the boiler can be solved and calculated, thereby achieving low-error prediction of the ash accumulation and coking degree of the boiler heating surface under the boiler operation data of multiple influencing factor data sets, providing a theoretical basis for ensuring the stable and safe operation of the boiler. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of a process for online monitoring of ash and coking on a heating surface of a boiler provided in an embodiment of the present application;

[0055] Figure 2 A diagram illustrating the hybrid model construction process provided in the embodiment of the present application;

[0056] Figure 3 to Figure 7 An illustration of the regression R value corresponding to each multi-step-ahead prediction model provided in the embodiments of the present application;

[0057] Figure 8 to Figure 12 A comparison chart of the predicted values, actual values ​​and relative errors of each prediction model provided in the embodiments of the present application;

[0058] Fig.13 This is a graph showing how the cleanliness coefficient of the low-temperature reheater heat exchange surface changes over time under normal operating conditions of the boiler unit provided in the embodiment of the present application. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0060] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0061] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0062] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0063] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0064] First, some technical terms in the present application are explained to facilitate those skilled in the art to understand the present application.

[0065] R value: refers to the correlation coefficient;

[0066] DCS system: Distributed Control System, distributed control system;

[0067] RMSE: Root-Mean-Square Error, root mean square error;

[0068] NSGA: nondominated sorting genetic algorithm, nondominated sorting genetic algorithm.

[0069] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0070] First, as Figure 1 As shown, the present application provides an online monitoring method for ash and coking on the heating surface of a boiler, comprising the following steps:

[0071] Step S1: Obtain the historical operation data of the boiler within a preset time after the power plant DCS system obtains the soot blowing operation, perform data preprocessing on the historical operation data, and obtain the preprocessed historical operation data;

[0072] Step S2: construct multiple multi-step-ahead prediction models using a BP neural network based on the preprocessed historical operation data;

[0073] Step S3: assigning weights to multiple multi-step-ahead prediction models, using a genetic algorithm to find the best weight combination, and obtaining the optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment based on the best weight combination;

[0074] Step S4: based on the optimal prediction model, obtaining the predicted value of the clean heating surface heat exchange of the boiler at a future time;

[0075] Step S5: Calculate the real-time heat exchange capacity of the boiler heating surface according to the real-time operation data of the boiler;

[0076] Step S6: obtaining the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler according to the predicted value of the heat exchange of the clean heating surface of the boiler and the real-time operating heat exchange of the heating surface of the boiler.

[0077] This application uses a BP neural network to construct a multi-step-ahead prediction model based on a time series, and then uses a genetic algorithm to find the weighted optimal solution of each multi-step-ahead prediction model to achieve the solution and calculation of the future heat exchange capacity of the clean heating surface of the boiler, thereby achieving low-error prediction of the degree of ash accumulation and coking on the boiler heating surface under the boiler operation data of a multi-influencing factor data set, providing a theoretical basis for ensuring stable and safe operation of the boiler.

[0078] In one embodiment, the step S1: obtaining the historical operation data of the boiler within a preset time after the power plant DCS system obtains the soot blowing operation, performing data preprocessing on the historical operation data, and obtaining the preprocessed historical operation data, specifically includes the following steps:

[0079] Step S11: Obtain the historical operation data of the boiler within a preset time of the power plant system after the soot blowing operation; wherein the historical operation data includes the flow rate of the working medium of the heating surface, the inlet and outlet temperature and pressure of the working medium of the heating surface, the coal supply amount, the high calorific value of the coal, the carbon content of the coal, the air supply amount, the boiler load, and the O2, CO, SO2, NO, NO2 content of the tail flue gas, and the carbon content of the ash;

[0080] Step S12: pre-process the acquired boiler historical operation data as follows:

[0081] Preprocessing A: Deleting discrete values ​​in the historical operation data; more specifically, using the quartile difference to filter and delete discrete values ​​in the historical operation data group value;

[0082] Preprocessing B: filling in missing data in historical operation data; more specifically, using a K-nearest neighbor algorithm to fill in missing data in historical operation data;

[0083] Preprocessing C: data standardization;

[0084] After preprocessing A, preprocessing B and preprocessing C, the preprocessed historical operation data is obtained.

[0085] In one embodiment, the step S2: constructing multiple multi-step-ahead prediction models using a BP neural network based on the preprocessed historical operation data specifically includes the following steps:

[0086] Step S21: Calculate the historical clean heating surface heat exchange capacity of the boiler according to the following formula:

[0087] ΔQ 历史洁净 ={D wf ·(h out,wf -h in,wf )} 历史数据

[0088] Where ΔQ 历史洁净 The heat exchange of the boiler's historical clean heating surface; D wfRefers to the inlet flow rate of the boiler heating surface; h in,wf 、h out,wf They are respectively the steam inlet working medium enthalpy value and the steam outlet working medium enthalpy value; specifically, the steam inlet and outlet working medium enthalpy values ​​are queried using the boiler heating surface inlet and outlet working medium temperature and pressure data in the unstandardized historical operation data; more specifically, the X Steam ('h_pt', P, T) function in the MTALB toolbox is used to query the steam inlet and outlet working medium enthalpy values ​​at a pressure of P bar and a temperature of T ° C in the unstandardized historical operation data;

[0089] Step S22: Use the pre-processed historical operation data as the model input, and use the corresponding boiler historical clean heating surface heat exchange ΔQ 历史洁净 As model output, several multi-step-ahead forecast models are constructed.

[0090] In one embodiment, the step S22: using the pre-processed historical operation data as the model input, and using the corresponding boiler historical clean heating surface heat exchange ΔQ 历史洁净 For model output, multiple multi-step-ahead forecast models are constructed, which specifically include the following steps:

[0091] Step S221: taking the pre-processed historical operation data and the corresponding historical clean heating surface heat exchange amount of the boiler as a data set;

[0092] Step S222: Divide the data set into a training set and a test set; specifically, divide the data set into a training set and a test set according to a ratio of 80% and 20%;

[0093] Step S223: The pre-processed historical operation data in the training set is used as the model input, specifically, the pre-processed working medium side inlet temperature and pressure, coal feed rate, coal high calorific value, coal carbon content, air feed rate, boiler load, and tail flue gas O2, CO, SO2, NO, NO2 content, ash carbon content are used as the model input; the historical operation data corresponding to the historical clean heating surface heat exchange of the boiler is used as the advance multi-step prediction model output, and multiple advance multi-step prediction models are constructed using the BP neural network; specifically, the structure of the BP neural network model is a three-layer network model, including two hidden layers and one output layer, and the number of neuron nodes in the two hidden layers is 20 and 18 respectively; the training parameters set by the model include the number of training times, training target, learning rate, momentum factor and overfitting verification failure times, which are set to 10000, 10 and 10 respectively. -7 , 0.01, 0.9 and 10; the transfer functions of the model are 'tansig' and 'purelin', and the learning function is 'trainlm';

[0094] Step S224: Based on the training set, the structural parameters and training parameters of the BP neural network model are adjusted to obtain the regression R value of each multi-step-ahead prediction model, until the regression R value of the established multi-step-ahead prediction model is less than the preset R value; the model is trained using the BP neural network to construct a zero-step-ahead prediction model, a one-step-ahead prediction model, a two-step-ahead prediction model, a three-step-ahead prediction model and a four-step-ahead prediction model, respectively, and the corresponding regression R values ​​are as follows: Figure 3 to Figure 7 As shown, they are 0.962, 0.958, 0.922, 0.919 and 0.894 respectively; Figure 3 This is a legend for the regression R value corresponding to the 0-step-ahead forecast model; Figure 4 This is a legend for the regression R value corresponding to the one-step-ahead forecast model; Figure 5 This is a legend for the corresponding regression R value of the two-step-ahead forecast model; Figure 6 This is a legend for the regression R value corresponding to the three-step-ahead forecast model; Figure 7 This is a legend for the corresponding regression R value of the four-step-ahead forecast model;

[0095] Step S225: obtaining multiple multi-step-ahead prediction models whose regression R values ​​are greater than the preset R value; through the regression R values ​​corresponding to the multi-step-ahead prediction models, it can be known that the boiler operation data at time T (the inlet temperature and pressure on the working medium side, the coal supply, the high calorific value of coal, the carbon content of coal, the air supply, the boiler load, and the O2, CO, SO2, NO, NO2 content of the tail flue gas, and the carbon content of the ash) have the greatest correlation with the boiler clean heating surface heat exchange at time T+1, and the boiler operation data at time T has the second highest correlation with the boiler clean heating surface heat exchange at time T+2 and time T+3; and the regression R value of the four-step-ahead prediction model is less than 0.9, indicating that the boiler operation data at time T has a poor ability to predict the boiler clean heating surface heat exchange at time T+4 and later, therefore, the multi-step-ahead prediction models at and after T+4 are discarded, and a total of four multi-step-ahead prediction models whose regression R values ​​meet the requirements are obtained, and each multi-step-ahead prediction model corresponds to a BP neural network model based on influencing factors under a time gradient;

[0096] In one embodiment, the step S22: using the pre-processed historical operation data as the model input and the corresponding boiler historical clean heating surface heat exchange rate as the model output, constructing multiple multi-step-ahead prediction models is specifically implemented as follows:

[0097] Starting from 0 steps ahead, the data set required for the multi-step-ahead prediction model is constructed, the data set is divided into a training set and a test set, the structural parameters and training parameters of the BP neural network model are adjusted, and the regression R value of the multi-step-ahead prediction model is obtained to determine whether it is greater than 0.9;

[0098] Repeat the above operation until the regression R value of the established multi-step-ahead prediction model is less than 0.9, and obtain a total of n multi-step-ahead prediction models with regression R values ​​greater than 0.9.

[0099] In one embodiment, the step S3: assigning weights to multiple multi-step-ahead prediction models, and optimizing to obtain an optimal prediction model with an optimal weight combination, specifically includes the following steps:

[0100] Step S31: assigning weights to the constructed multiple multi-step-ahead prediction models and performing weighted summation to obtain a hybrid model;

[0101] Step S32: constructing an optimization objective function for weight combination in the hybrid model;

[0102] Step S33: optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model;

[0103] Step S34: Based on the optimal weight combination of the hybrid model, an optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment is obtained.

[0104] In one embodiment, if Figure 2 As shown, the step S31: assign weights to the constructed multiple multi-step-ahead prediction models and perform weighted summation, and the obtained hybrid model is shown in the following formula:

[0105] Y=w1Y1+w2Y2+…+w n Y n

[0106] Where, Y1, Y2, ..., Y n are the multi-step-ahead prediction models, n is the number of steps in the multi-step-ahead prediction model, w1, w2, ..., w n are the weights assigned to each multi-step-ahead prediction model; w1+w2+...+w n =1; the weight combination to be optimized is W = {w1, w2…w n}.

[0107] In one embodiment, the step S32: constructing an optimization objective function for weight combinations in a hybrid model specifically includes the following steps:

[0108] The input of the objective function is the weight combination W of the hybrid model = {w1, w2…w n}, the objective function is the root mean square error of the heat transfer of the boiler heating surface output by the hybrid model.

[0109] In one embodiment, the step S33: optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model specifically includes the following steps:

[0110] The root mean square error of the hybrid model is used as the dependent variable, and the weights of each multi-step-ahead prediction model are used as the independent variable. The non-dominated sorting genetic algorithm in the genetic algorithm is used to find the optimal solution of the objective function. The optimization parameters set by the genetic algorithm include the non-dominated solution expected to be maintained in the population, the size of the population, the number of algorithm iterations, the number of overfitting verification failures, and the objective function tolerance, which are set to 0.3, 200, 300, 200, and 10, respectively. -10 , we get the optimal weight combination W = {0.5185, 0.3429, 0.1300, 0.0094} of the hybrid model; Figure 4 It can be seen that the relative error of the hybrid model with the best weight combination is about 5%, which is lower than other models and is regarded as the optimal prediction model.

[0111] In one embodiment, step S4: based on the optimal prediction model, obtaining the predicted value of the clean heating surface heat exchange of the boiler at a future time, is specifically implemented as follows:

[0112] Input the boiler operation data at the current moment (the working medium side inlet temperature and pressure, coal feed rate, coal high calorific value, coal carbon content, air feed rate, boiler load, and tail flue gas O2, CO, SO2, NO, NO2 content, ash carbon content), and the boiler clean heating surface heat exchange amount predicted by each multi-step prediction model;

[0113] The heat exchange of the clean heating surface of the boiler predicted by each multi-step-ahead prediction model is respectively assigned corresponding weights in the optimal weight combination and added up to obtain the final predicted heat exchange of the clean heating surface of the boiler at the next moment.

[0114] For example, if you need to get T+3, then obtain the boiler operating data at time T, the boiler operating data at time T+1, and the boiler operating data at time T+2; use the one-step-ahead prediction model established with the boiler operating data at time T+2 to predict the boiler clean heating surface heat exchange Y1 at time T+3; use the two-step-ahead prediction model established with the boiler operating data at time T+1 to predict the boiler clean heating surface heat exchange Y2 at time T+3; use the three-step-ahead prediction model established with the boiler operating data at time T+0 to predict the boiler clean heating surface heat exchange Y3 at time T+3; then, multiply them by the corresponding weights in the optimal weight combination, and get the predicted value of the boiler clean heating surface heat exchange at time T+3 with the smallest prediction error.

[0115] In one embodiment, step S5: the real-time heat exchange amount of the heating surface of the boiler is calculated according to the real-time operation data of the boiler and the following formula:

[0116] ΔQ 实时运行 ={D wf ·(h out,wf -h in,wf)} 运行数据

[0117] Where ΔQ 实时运行 Real-time heat exchange for boiler heating surface; D wf Refers to the inlet flow rate of the boiler heating surface; h in,wf 、h out,wf are the enthalpy of the steam inlet and the enthalpy of the steam outlet respectively.

[0118] In one embodiment, step S6: according to the predicted heat exchange value of the clean heating surface of the boiler and the real-time heat exchange value of the heating surface of the boiler, the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler is obtained, as shown in the following formula:

[0119]

[0120] Where k is the cleanliness coefficient; ΔQ 实时运行 The real-time heat exchange rate of the boiler heating surface; ΔQ 预测洁净 It is the predicted value of heat exchange of clean heating surface of boiler.

[0121] The low temperature reheater of a power plant unit is used as the object, and the cleanliness of the heat exchanger surface is monitored online using the patent of this invention. The unit is equipped with two 660MW high-efficiency ultra-supercritical parameter coal-fired steam turbine generator sets, and the boiler is a high-efficiency ultra-supercritical parameter transformer operation DC furnace, single furnace, one reheat, balanced ventilation, solid slag discharge, all-steel frame, and full suspension structure; the boiler adopts a tight-fitting, closed, Π-shaped layout boiler.

[0122] A total of 5000 groups of operating parameters after a boiler sootblowing operation were screened out using the sootblowing signal of the power plant. It was considered that the surface of the low-temperature superheater was clean and the heat transfer performance was good under these operating conditions, and the cleaning data obtained an initial database. At the same time, 4000 groups of data of the unit after 10:15 on July 9, 2024 were obtained for testing the model monitoring performance.

[0123] The BP neural network is used to train the model, and the 0-step ahead prediction model, the one-step ahead prediction model, the two-step ahead prediction model, the three-step ahead prediction model and the four-step ahead prediction model are constructed respectively. The corresponding regression R values ​​are as follows: Figure 3 to Figure 7 As shown, according to Figures 3 to 7The order is 0.962, 0.958, 0.922, 0.919 and 0.894 respectively. Through the regression R value, it can be seen that the parameters at time T (working fluid side inlet temperature and pressure, coal supply, high calorific value of coal, coal carbon content, air supply, boiler load, and O2, CO, SO2, NO, NO2 content of tail flue gas, ash carbon content) are most correlated with the clean heating surface heat exchange at time T+1, followed by the heat exchange at time T+2 and time T+3. The regression R value of the four-step prediction model is less than 0.9, which means that the parameters at time T have poor prediction ability for the clean heating surface heat exchange at time T+4 and later, and the models of T+4 and later are discarded. A total of four BP neural network models are obtained, each of which corresponds to a BP neural network model based on influencing factors under a time dimension.

[0124] The data used for training is divided into training set and test set in the ratio of 80% and 20% respectively. The structure of BP model is a three-layer network model, including two hidden layers and one output layer. The number of neuron nodes in the two hidden layers is 20 and 18 respectively. The training parameters set by the model include the number of training times, training target, learning rate, momentum factor and the number of overfitting verification failures, which are set to 10000, 10 -7 , 0.01, 0.9 and 10; the transfer functions of the model are 'tansig' and 'purelin', and the learning function is 'trainlm'.

[0125] The weights W = {w1, w2, w3, w4} are assigned to the four multi-step-ahead prediction models, and a multi-parameter prediction model in the time dimension is constructed based on the four multi-step-ahead prediction models. The root mean square error of the hybrid model is used as the dependent variable, and the weight is used as the independent variable. The non-dominated sorting genetic algorithm in the genetic algorithm is used to find the optimal solution. The optimization parameters set by the genetic algorithm include the non-dominated solution expected to be maintained in the population, the size of the population, the number of algorithm iterations, the number of overfitting verification failures, and the objective function tolerance, which are set to 0.3, 200, 300, 200, and 10, respectively. -10 , and the optimal weight of the model is obtained as W = {0.5185, 0.3429, 0.1300, 0.0094}.

[0126] For comparison, we selected the LSTM model with the same time series and trained it with 5000 sets of data. The root mean square error of the model is as follows: Figure 8 to Figure 12 As shown; among them, Figure 8 It is a comparison chart of the predicted value, actual value and relative error of the 0-step-ahead prediction model; Fig. 9 A comparison chart of the predicted value, actual value and relative error of the one-step-ahead prediction model; Fig.10 It is a comparison chart of the predicted value, actual value and relative error of the two-step-ahead prediction model; Fig.11 This is a comparison chart of the predicted value, actual value and relative error of the three-step-ahead prediction model; Fig.12 The comparison chart of the predicted value and actual value and relative error of the hybrid prediction model; According to Table 1 and Figure 8 to Figure 12 It can be seen that the prediction accuracy of the hybrid model is higher than that of the LSTM model, followed by the 0-step-ahead prediction model, the one-step-ahead prediction model, the two-step-ahead prediction model and the three-step-ahead prediction model; the root mean square error of each model based on the test set is shown in Table 1.

[0127] The optimization algorithm of the LSTM model is the Adam optimization algorithm, and the training parameters set include the maximum number of iterations, the minimum batch size, the initial learning rate, the number of iterations at which the learning rate starts to decrease, the decrease factor, and the number of iterations for each verification, which are 500, 10, 0.005, 125, 0.2, and 5 respectively.

[0128] Table 1 Root mean square error corresponding to each model

[0129] Model 0 steps ahead One step ahead Two steps ahead Three steps ahead Hybrid Model LSTM RMSE <![CDATA[1.1120×10 4 ]]> <![CDATA[1.1761×10 4 ]]> <![CDATA[1.4432×10 4 ]]> <![CDATA[1.6338×10 4 ]]> <![CDATA[9.9267×10 3 ]]> <![CDATA[1.0111×10 4 ]]>

[0130] Finally, the cleanliness coefficient of the low-temperature reheater heat exchange surface under normal operating conditions is obtained using the hybrid model, such as Fig.13 As shown, its changing trend corresponds to the soot blowing operation, which verifies the robustness of the hybrid model of the optimal weight combination provided by the present application, that is, the optimal prediction model.

[0131] In a second aspect, the present application provides an online monitoring system for ash and coking on a boiler heating surface, comprising:

[0132] A historical operation data acquisition module is used to acquire the historical operation data of the boiler within a preset time after the DCS system of the power plant obtains the soot blowing operation, preprocess the historical operation data, and acquire the preprocessed historical operation data;

[0133] A multi-step-ahead prediction model construction module is communicatively connected to the historical operation data acquisition module and is used to construct multiple multi-step-ahead prediction models using a BP neural network based on the preprocessed historical operation data;

[0134] An optimal prediction model acquisition module is connected to the multi-step-ahead prediction model construction module, and is used to assign weights to multiple multi-step-ahead prediction models, use a genetic algorithm to find the best weight combination, and obtain the optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment based on the best weight combination;

[0135] A boiler clean heating surface heat exchange amount acquisition module is connected to the optimal prediction model acquisition module and is used to obtain a predicted value of the boiler clean heating surface heat exchange amount at a future time based on the optimal prediction model;

[0136] The operating heat exchange amount acquisition module is used to calculate the real-time operating heat exchange amount of the boiler heating surface according to the real-time operating data of the boiler;

[0137] The cleanliness coefficient acquisition module is communicatively connected with the boiler clean heating surface heat exchange acquisition module and the operating heat exchange acquisition module, and is used to obtain the cleanliness coefficient of the reaction boiler heating surface ash and coking degree according to the predicted value of the boiler clean heating surface heat exchange and the real-time operating heat exchange of the boiler heating surface.

[0138] In one embodiment, the optimal prediction model acquisition module includes:

[0139] A hybrid model acquisition unit is used to assign weights to the constructed multiple multi-step-ahead prediction models to perform weighted summation to obtain a hybrid model;

[0140] An objective function construction unit, which is in communication connection with the hybrid model acquisition unit and is used to construct an optimization objective function for weight combination in the hybrid model;

[0141] An optimization unit, which is in communication connection with the objective function construction unit, and is used to optimize the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model;

[0142] The optimal prediction model acquisition unit is connected to the optimization unit for obtaining the optimal prediction model of the heat exchange of the clean heating surface of the boiler at a future moment based on the optimal weight combination of the hybrid model.

[0143] Among them, the functional implementation of each module in the above-mentioned boiler heating area ash and coking online monitoring system corresponds to the steps in the above-mentioned boiler heating area ash and coking online monitoring method embodiment, and its functions and implementation processes are no longer repeated here.

[0144] In a third aspect, an embodiment of the present application provides an online monitoring device for ash and coking on the heating surface of a boiler. The online monitoring device for ash and coking on the heating surface of a boiler can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0145] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the boiler heating area ash coking online monitoring device, and the interface used to realize the interconnection of the boiler heating area ash coking online monitoring device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0146] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0147] The processor may be a general-purpose processor, which may call the boiler heating surface ash coking online monitoring program stored in the memory, and execute the boiler heating surface ash coking online monitoring method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). Among them, the method executed when the boiler heating surface ash coking online monitoring program is called may refer to the various embodiments of the boiler heating surface ash coking online monitoring method of the present application, which will not be repeated here.

[0148] In a fourth aspect, an embodiment of the present application also provides a readable storage medium.

[0149] The readable storage medium of the present application stores an online monitoring program for ash and coking on the heating surface of a boiler. When the online monitoring program for ash and coking on the heating surface of a boiler is executed by a processor, the steps of the online monitoring method for ash and coking on the heating surface of a boiler as described above are implemented.

[0150] Among them, the method implemented when the online monitoring program for ash and coking on the heating surface of the boiler is executed can refer to the various embodiments of the online monitoring method for ash and coking on the heating surface of the boiler of the present application, and will not be repeated here.

[0151] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0152] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0153] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for online monitoring of ash and coking on a boiler heating surface, characterized in that: The following steps are involved: Obtain the historical operation data of the boiler within a preset time after the power plant DCS system obtains the soot blowing operation, perform data preprocessing on the historical operation data, and obtain the preprocessed historical operation data; Based on the pre-processed historical operation data, BP neural network is used to build multiple multi-step-ahead prediction models; Assign weights to multiple multi-step-ahead prediction models, use genetic algorithms to find the best weight combination, and obtain the optimal prediction model for the heat exchange of the clean heating surface of the boiler at future moments based on the best weight combination; Based on the optimal prediction model, obtain the predicted value of the clean heating surface heat exchange of the boiler at a future time; The real-time heat exchange of the boiler heating surface is calculated based on the real-time operation data of the boiler; According to the predicted value of the heat exchange of the clean heating surface of the boiler and the real-time heat exchange of the heating surface of the boiler, the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler is obtained.

2. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 1, characterized in that: The method of obtaining the boiler historical operation data within a preset time after the soot blowing operation is performed by the power plant DCS system, performing data preprocessing on the historical operation data, and obtaining the preprocessed historical operation data specifically includes the following steps: Obtain the historical operation data of the boiler within the preset time of the power plant system after the soot blowing operation; Delete discrete values ​​in the historical operation data, fill in missing data in the historical operation data, and perform data standardization to obtain preprocessed historical operation data.

3. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 1, characterized in that: The method of constructing multiple multi-step-ahead prediction models based on the pre-processed historical operation data using a BP neural network specifically includes the following steps: The enthalpy value is queried using the inlet and outlet working fluid temperature and pressure data of the boiler heating surface in the unstandardized historical operation data, and the heat exchange of the boiler's historical clean heating surface is calculated; The preprocessed historical operation data is used as the model input, and the corresponding historical clean heating surface heat exchange of the boiler is used as the model output to construct multiple multi-step-ahead prediction models.

4. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 3, characterized in that: The pre-processed historical operation data is used as the model input, and the corresponding boiler historical clean heating surface heat exchange rate is used as the model output to construct multiple multi-step-ahead prediction models, which specifically includes the following steps: The pre-processed historical operation data and the corresponding historical clean heating surface heat exchange of the boiler are used as the data set; Divide the dataset into training set and test set; The pre-processed historical operation data in the training set is used as the model input, and the corresponding boiler historical clean heating surface heat exchange is used as the model output. The BP neural network is used to build multiple multi-step-ahead prediction models. Based on the test set, the structural parameters and training parameters of the BP neural network model are adjusted to obtain the regression R value of each multi-step-ahead prediction model until the regression R value of the multi-step-ahead prediction model is not greater than the preset R value; Obtain multiple multi-step-ahead forecasting models whose regression R values ​​are greater than the preset R values.

5. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 4, characterized in that: The method of assigning weights to the multiple multi-step-ahead prediction models, using a genetic algorithm to find the best weight combination, and obtaining the best prediction model for the heat exchange of the clean heating surface of the boiler at a future moment based on the best weight combination specifically includes the following steps: Assign weights to the multiple multi-step-ahead prediction models constructed, perform weighted summation, and obtain a hybrid model; Construct the optimal objective function for weight combination in the hybrid model; Optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model; Based on the optimal weight combination of the hybrid model, the optimal prediction model for the heat exchange of the clean heating surface of the boiler at future moments is obtained.

6. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 5, characterized in that: The optimization objective function is the root mean square error of the heat exchange of the boiler heating surface output by the hybrid model.

7. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 5, characterized in that: The step of optimizing the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model specifically includes the following steps: The root mean square error of the mixed model is taken as the dependent variable and the weight is the independent variable; The non-dominated sorting genetic algorithm is used to find the optimal solution of the objective function. The optimization parameters include the non-dominated solution that is expected to be maintained in the population, the size of the population, the number of algorithm iterations, the number of overfitting verification failures and the objective function tolerance, and the best weight combination of the hybrid model is obtained.

8. The method for online monitoring of ash and coking on the heating surface of a boiler according to claim 1, characterized in that: According to the predicted value of the heat exchange of the clean heating surface of the boiler and the real-time heat exchange of the heating surface of the boiler, the cleanliness coefficient of the ash and coking degree of the heating surface of the reaction boiler is obtained, as shown in the following formula: In the formula, k is the cleanliness coefficient; ΔQ 实时运行 The real-time heat exchange rate of the boiler heating surface; ΔQ 洁净预测 It is the predicted value of heat exchange of clean heating surface of boiler.

9. An online monitoring system for ash and coking on the heating surface of a boiler, characterized in that: include: A historical operation data acquisition module is used to acquire the historical operation data of the boiler within a preset time after the DCS system of the power plant obtains the soot blowing operation, preprocess the historical operation data, and acquire the preprocessed historical operation data; A multi-step-ahead prediction model construction module is communicatively connected to the historical operation data acquisition module and is used to construct multiple multi-step-ahead prediction models using a BP neural network based on the preprocessed historical operation data; An optimal prediction model acquisition module is connected to the multi-step-ahead prediction model construction module, and is used to assign weights to multiple multi-step-ahead prediction models, use a genetic algorithm to find the best weight combination, and obtain the optimal prediction model for the heat exchange of the clean heating surface of the boiler at a future moment based on the best weight combination; A boiler clean heating surface heat exchange amount acquisition module is connected to the optimal prediction model acquisition module and is used to obtain a predicted value of the boiler clean heating surface heat exchange amount at a future time based on the optimal prediction model; The operating heat exchange amount acquisition module is used to calculate the real-time operating heat exchange amount of the boiler heating surface according to the real-time operating data of the boiler; The cleanliness coefficient acquisition module is communicatively connected with the boiler clean heating surface heat exchange acquisition module and the operating heat exchange acquisition module, and is used to obtain the cleanliness coefficient of the reaction boiler heating surface ash and coking degree according to the predicted value of the boiler clean heating surface heat exchange and the real-time operating heat exchange of the boiler heating surface.

10. The boiler heating surface ash and coking online monitoring system according to claim 9, characterized in that: The optimal prediction model acquisition module includes: A hybrid model acquisition unit is used to assign weights to the constructed multiple multi-step-ahead prediction models to perform weighted summation to obtain a hybrid model; An objective function construction unit, which is in communication connection with the hybrid model acquisition unit and is used to construct an optimization objective function for weight combination in the hybrid model; An optimization unit, which is in communication connection with the objective function construction unit, and is used to optimize the weight combination of the hybrid model based on the established objective function to obtain the best weight combination of the hybrid model; The optimal prediction model acquisition unit is connected to the optimization unit for obtaining the optimal prediction model of the heat exchange of the clean heating surface of the boiler at a future moment based on the optimal weight combination of the hybrid model.

Images