Power station boiler heating surface ash pollution quantitative characterization method and system based on transfer learning

Abstract

The invention belongs to the technical field of ash pollution monitoring, and discloses a transfer learning-based boiler heating surface ash pollution quantitative characterization method, which comprises the following steps of: establishing a boiler-boiler coupling model of a research object boiler by virtue of analog simulation software, and adopting a method of combining simulation modeling with actual verification. And the transfer learning-based boiler heating surface ash dirt quantitative characterization method is researched. The model and the dirt label which are established in the simulation data and are used for verifying the accuracy are migrated to the same actual heating surface for dirt quantitative characterization modeling, so that the modeling difficulty is reduced, the modeling speed is increased, the required data volume is small, and the dust deposition trend of the actual heating surface can be obtained through effective fitting. The method effectively solves the problem that the actual operation data of the boiler unit lacks a characterization dirt label and cannot provide guidance for subsequent soot blowing optimization, and has a certain application prospect.

Description

Technical Field

[0001] The present invention belongs to the technical field of heating surface ash fouling monitoring, and in particular relates to a boiler heating surface ash fouling quantitative characterization modeling method and system based on transfer learning. Background Art

[0002] In the past decade, the proportion of new energy installed capacity has continued to increase, but its intermittent, random and volatile characteristics make the power support capacity of new energy still insufficient. Therefore, coal-fired thermal power generation, as the "ballast stone" to ensure the safety of my country's electricity, will still need to bear the "burden" of basic security for a long time in the future. At present, coal-fired thermal power generation is gradually changing from a basic power source that bears the basic load to a regulating power source that bears the fluctuation of new energy and ensures the safety of power supply. Flexible operation of coal-fired power is imperative. In addition, due to the complexity of the coal market, my country's coal-fired power plant boilers have been burning non-designed coal for a long time, and the frequent changes in the quality of the coal entering the furnace have further increased the complexity of boiler operation. Since there is a large amount of ash in coal, the flue gas generated during the combustion process will carry ash and contact the outer surface of different heating surfaces of the boiler, and gradually adhere to and deposit on its surface over time, so ash accumulation and slagging are inevitable during the operation of coal-fired power plant boilers. It is mainly related to the structure, load and coal quality of the boiler. For a specific boiler, there are differences in coal feed rate, boiler internal temperature level and flue gas flow rate under different loads, which affect the ash and slagging of the heating surface; the ash content and ash composition of different coal types are different, and their ash accumulation and slagging characteristics are quite different. The current situation of flexible coal quality and load of coal-fired power plant boilers makes the traditional timed and quantitative sootblowing strategy inconsistent with the actual sootblowing requirements of the boiler heating surface, resulting in "overblowing" or "underblowing". Therefore, it is urgent to develop a quantitative characterization technology for ash pollution on the heating surface to ensure the safety and economy of boiler operation. From the existing public information, the traditional mechanism-based or data-driven methods have problems such as complex mechanism simplification and lack of label data in ash pollution monitoring modeling. And most of them are only suitable for scenarios with basic load and fixed coal types. There are large errors in ash pollution monitoring under flexible operating conditions, which leads to blindness and randomness of heating surface sootblowing. Therefore, it is of great significance to break through the limitations of traditional mechanism modeling and data-driven modeling and establish a high-precision ash pollution monitoring model.

[0003] As important branches of machine learning, deep learning and transfer learning are widely used in anomaly detection and prediction of industrial equipment and monitoring systems. Their powerful learning ability can capture the dependencies between time series data and the nonlinear expressions between feature parameters. Deep learning algorithms often need strong and high-quality data to achieve good results. For industrial equipment operation and monitoring systems, the data collection process requires a lot of time and economic costs. If fixed tasks such as anomaly monitoring need to be completed, it will be found that anomalies and failures in actual situations are small sample events and lack data labels, which leads to certain limitations in the application field of deep learning algorithms. Although unsupervised learning, as a branch of deep learning, has been widely used in scenarios lacking labeled data, its hidden expressions are learned through data dimensionality reduction and feature extraction, and the effect is relatively good. However, when facing different research objects, it needs to re-model and optimize parameters to adapt to different learning tasks. In addition, massive data and complex and changeable working conditions limit the applicability and generalization of deep learning modeling methods. Transfer learning methods can use the idea of ​​knowledge transfer to apply feature relationships learned in one field to similar fields, further solving the actual industrial scenarios of small samples and data set deviations. According to the analysis of existing public data, the current application of anomaly detection and diagnosis in power plant boilers is mostly concentrated in equipment such as coal mills and pulverizing systems, and other equipment is rarely used. In actual power plant operation, after the combustion of furnace coal powder, the flue gas carries ash and slowly deposits on different heating surfaces, which will lead to problems such as reduced heat transfer efficiency and hidden dangers in safe operation. This is similar to the degradation anomalies and failures of industrial equipment. Therefore, we can use the advantages of transfer learning methods in solving practical problems such as lack of labels and data set offsets to explore new ideas for high-precision boiler heating surface ash pollution quantitative characterization and monitoring modeling. Summary of the invention

[0004] In order to solve the problems of complex mechanisms, difficulty in simplification, lack of labeled data, etc. in ash pollution monitoring modeling of traditional mechanism-based or data-driven methods, the present invention provides a boiler heating surface ash pollution quantitative characterization modeling method based on transfer learning. The dynamic simulation model outputs the dynamic operation data of the boiler heating surface and the label data characterizing the ash pollution degree. The pre-model construction and verification of the ash pollution quantitative characterization of the heating surface for simulation data are completed according to the transfer learning process of "model pre-training-model transfer-fine-tuning strategy-model verification". After that, the model is further migrated to the actual boiler heating surface, which proves that the transfer learning model can effectively fit the ash pollution trend of the boiler heating surface during the sootblowing cycle in the actual operation data.

[0005] The present invention is implemented as follows: a boiler heating surface ash pollution quantitative characterization modeling method based on transfer learning, comprising:

[0006] S1: Based on the boiler dynamic simulation model, obtain the full load and full operating condition simulation operation data when the ash pollution coefficient of each level of heating surface changes;

[0007] S2: Select the heating surface operation data set S1 in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain;

[0008] S3: Based on machine learning algorithms such as autoencoders and support vector regression, a quantitative characterization model for boiler heating surface ash pollution in the source domain is built and parameter optimization and training are performed to obtain a pre-trained model with the best heating surface ash pollution fitting result;

[0009] S4: Migrate the S3 pre-trained model to the target domain dataset, and use the method of fine-tuning the model parameters to make the model adapt to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface;

[0010] S5: Use the actual sootblowing execution action to determine the soot accumulation trend and soot contamination degree of the heated surface in different sootblowing cycles, and complete the verification and application of the transfer learning model.

[0011] Furthermore, the full-load and full-condition operation data acquisition under the ash pollution coefficient of each level of heating surface in S1 is studied with the help of the boiler dynamic simulation model. In the module property setting, the change of the ash pollution coefficient of the specific heating surface of the boiler is used as the indicator feature. The range is set according to the pollution situation of the specific object. Generally, it can be taken as 0.6-1. When the coefficient is 1 (100%), it means that the heating surface is clean. As the coefficient decreases, it means that the cleanliness of the surface of the heating surface begins to decrease due to ash accumulation; when the coefficient rises back to 1 again, it means that the heating surface has recovered to a clean state due to the execution of the soot blowing operation. The simulation data sampling interval is 10 seconds to 1 minute. For each level of heating surface, the original simulation data set is composed of multiple parameters such as load, coal feed, working fluid side flow, working fluid side temperature, working fluid side pressure, flue gas side temperature and pressure. The selection of feature data needs to be selected in combination with the actual boiler measurement point situation, which is used for the subsequent quantitative characterization of ash pollution of each level of heating surface of the boiler.

[0012] Furthermore, the division of the data set in S2 is based on the basic concept of transfer learning. Transfer learning uses the idea of ​​knowledge transfer to apply the feature relationship learned in one field to a similar field, and divides it into the source field D s (Source domain) and target domain D t(Target domain), the data in the source domain are generally annotated, and the data quality is high. Then, valuable knowledge such as feature expressions in the source domain data set is obtained through relevant algorithm models; while the data in the target domain are generally not annotated and the data distribution or data features are similar to those in the source domain. With the help of relevant algorithm models, similar feature expressions can be transferred from the source domain to the target domain, helping the target domain to complete data prediction or classification and improve the performance of the target task. In the above, the simulation operation data and the gray pollution coefficient label of the heating surface are selected as the source domain, and the actual operation data of the corresponding heating surface is selected as the target domain, and the input parameters of the model are relatively consistent.

[0013] Furthermore, the pre-trained model with the best fitting result of the heated surface dust pollution built in S3 has the following specific process: facing the source domain data, combining unsupervised learning algorithms such as autoencoder (AE) or traditional support vector regression (SVR) and other machine learning algorithms, using the encoder part in the AE model as a feature extractor, inputting data samples into the encoder part for data dimension reduction and deep feature extraction, and then inputting the feature vector and data label into models such as SVR to perform regression, and comparing the obtained predicted value of the heated surface dust pollution degree with the dust pollution coefficient label of the source domain, and obtaining the model framework and parameter setting with the best fitting result through parameter optimization, which is expressed as the pre-trained model AE-SVR for transfer learning, and subsequently used for model transfer learning for the target domain.

[0014] Among them, the autoencoder AE described in S3 includes only the encoding part:

[0015] The encoder E transforms the input feature vector x into a latent feature vector z, i =E(x i |θ E ).

[0016] In the formula, x is the input feature, θ E is the parameter of the encoder. The encoder part extracts features and reduces the dimension of the input data.

[0017] Among them, the basic idea of ​​the support vector regression SVR described in S3 is to map the sample data to a high-dimensional space through a nonlinear mapping function and perform linear regression in the high-dimensional space. By constructing a loss function that minimizes the training set data label and the model fitting value, an objective function f(x) is finally determined, which is expressed as:

[0018]

[0019] Where w is the weight coefficient vector; is the high-dimensional feature space obtained by mapping the input space x; b is the bias.

[0020] Furthermore, in S4, the pre-model in S3 is migrated to the target domain. Since the distribution differences of different data sets when migrating to the target domain will cause the model effect to deteriorate, different fine-tuning strategies need to be selected in deep transfer learning to adapt the parameters or models to the new data set and improve the model effect. For the AE-SVR model, a fine-tuning strategy is adopted to change the parameters of the AE encoder part to adapt to the target domain data.

[0021] Furthermore, in S5, in the verification of the migration results of the quantitative characterization model of the heating surface soot fouling for actual operating data, the actual boiler unit lacks label values ​​that characterize the degree of soot fouling on the heating surface, that is, the simulation data set in the source domain has labels, but the target domain data set does not have labels. Therefore, the actual sootblowing execution action is used to judge whether the AE-SVR model fitting results of each level of heating surface in different sootblowing cycles reflect the correct soot accumulation trend and soot fouling degree, thereby verifying the effectiveness and feasibility of the AE-SVR transfer learning method.

[0022] Another object of the present invention is to provide a boiler heating surface ash pollution quantitative characterization system based on transfer learning, comprising:

[0023] The simulation data acquisition module, based on a dynamic simulation model with detailed mechanisms, acquires full-load and full-operating simulation operation data when the dust pollution coefficient of each level of heating surface changes;

[0024] Run the data acquisition module for modeling, verification and application of quantitative characterization of heated surface contamination;

[0025] A selection module is used to select the heating surface operation data set in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain;

[0026] The training module is used to build a boiler heating surface ash pollution quantitative characterization model in the source domain based on the autoencoder and support vector regression algorithm, and perform parameter optimization and training to obtain a pre-trained model with the best heating surface ash pollution fitting result;

[0027] The migration module is used to migrate the pre-trained model to the target domain dataset and fine-tune the model parameters to adapt the model to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface;

[0028] The verification module is used to use the actual soot blowing execution action to determine the soot accumulation trend and soot contamination degree of the heated surface in different soot blowing cycles, and complete the verification and application of the transfer learning model.

[0029] Another object of the present invention is to provide a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantitative characterization of ash pollution on boiler heating surfaces based on transfer learning.

[0030] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to perform the steps of the method for quantitative characterization of boiler heating surface ash fouling based on transfer learning.

[0031] Another object of the present invention is to provide an information data processing terminal, which is used to implement the boiler heating surface ash pollution quantitative characterization system based on transfer learning.

[0032] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0033] First, although the method based on unsupervised learning has been widely used in scenarios where there is a lack of labeled data. However, it needs to be remodeled when facing different research objects. A large amount of data and changing working conditions limit its applicability and generalization. Transfer learning is to transfer the model trained and verified in the source domain to a target domain with similarity. Under the conditions of data distribution and feature dimension changes, the source domain knowledge is better used to assist the target domain in modeling, which is in line with the characteristics of similarity of data sets in the heating surfaces at all levels in this paper. Therefore, the present invention uses simulation software to build a dynamic simulation model of the research object boiler, and adopts a method of combining simulation modeling with actual verification to study the quantitative characterization method of ash pollution on the heating surface of the boiler based on transfer learning. By using the model and ash pollution labels that have been established and verified in the simulation data, they are transferred to the same actual heating surface for ash pollution quantitative characterization modeling, which reduces the difficulty of modeling, improves the modeling speed, requires less data, and can effectively fit the ash accumulation trend of the actual heating surface. It effectively solves the problem that there is a lack of ash pollution labels in the actual operation data of the current boiler unit, and it is impossible to provide guidance for subsequent soot blowing optimization, and has certain application prospects.

[0034] Second, traditional mechanism-based and data-driven modeling has problems such as complex mechanisms that are difficult to simplify, lack of labeled data, and is mostly applicable to scenarios that bear basic loads and burn fixed types of coal. For example, the calculation results of the currently commonly used clean factor will be affected by the rapid rise and fall of the load, and cannot correctly reflect the actual degree of ash pollution, and cannot provide guidance for subsequent soot blowing optimization. Data-driven modeling is a predictive model established on the basis of mechanism results, so it cannot effectively solve the problem of large deviations in monitoring results under flexible operation. Methods based on unsupervised learning can be applied to modeling scenarios that lack labeled data, but they need to be re-modeled when facing different research objects. Large amounts of data and changing working conditions limit their applicability and generalization.

[0035] The present invention follows the transfer learning process of "model pre-training, model migration, fine-tuning strategy, and model verification" to migrate the model trained and verified on the source domain, i.e., the simulation data set, to a similar target domain, i.e., the actual same heating surface. When there are differences in the data distribution and feature dimensions of the two, the source domain knowledge is used to assist the target domain in modeling. The present invention reduces the difficulty of modeling in actual operating data that lacks labels to characterize dust and dirt, and uses the dust and dirt label expression and feature expression in the simulation data set to help the actual operating data effectively fit its dust accumulation trend and verify the effectiveness of the actual soot blowing action. In addition, only a small number of heating surface objects in the simulation data need to be modeled and relevant feature expressions obtained, while the heating surface objects in the actual data are modeled according to transfer learning, and adapted to the actual objects through local or global parameter fine-tuning, which further reduces the time and economic costs of modeling training and improves the applicability and generalization of the model.

[0036] Third, whether the technical solution of the present invention solves the technical problems that people have been eager to solve but have never been able to solve successfully:

[0037] Under the background of dual carbon, my country's energy system is accelerating towards the direction of "clean, intelligent, flexible and efficient". At present, coal-fired thermal power generation is gradually changing from basic power supply that bears basic load to regulating power supply. In addition, due to fluctuations in the coal market, most power plants will adopt coal blending to reduce energy costs, such as burning high-alkali coal with abundant prices and reserves, but it is very easy to be contaminated and slag. Therefore, under the current situation of flexible unit load and coal quality entering the furnace, the ash accumulation and slag characteristics inside the boiler are different from before, resulting in the traditional timed and quantitative soot blowing strategy often not matching the actual soot blowing needs of the heating surface, resulting in "under-blowing" or "over-blowing" of the heating surface locally or even globally. Therefore, it is necessary to develop high-precision heating surface ash pollution quantitative characterization technology to ensure the safety and economy of boiler operation. However, the current traditional mechanism-based and data-driven modeling has problems such as complex mechanisms that are difficult to simplify, lack of labeled data, and is mostly applicable to scenarios that bear basic loads and burn fixed types of coal. For example, the calculation results of the commonly used clean factors will be affected by the rapid rise and fall of loads, and cannot correctly reflect the actual degree of ash pollution, and cannot provide guidance for subsequent soot blowing optimization. Data-driven modeling is a predictive model established on the basis of mechanism results, so it cannot solve the problem of large deviations in monitoring results under flexible operation. The present invention first analyzes the advantages of transfer learning in solving practical problems such as lack of labels and data set offsets in the field of industrial system monitoring. From the existing literature, the current application of transfer learning in power plant boilers is mostly concentrated on equipment such as coal mills and pulverizing systems, and other equipment is rarely used. In actual power plant operation, due to the large number of heating surfaces of the boiler, it is necessary to re-model when facing different research objects. Large amounts of data and changing working conditions limit the applicability and generalization of the model. In addition, the actual operation data is affected by multiple factors such as frequent load changes and environmental noise, and the data quality is low, resulting in the low accuracy of the ash pollution monitoring model established based on the actual operation data and the high time cost of the parameter optimization process, which lacks more accurate guidance for soot blowing. Transfer learning is to transfer the model trained and verified in the source domain to a similar target domain, and better use the source domain knowledge to assist the target domain in modeling under the conditions of changes in data distribution and feature dimensions. Therefore, transfer learning can be used to explore new ideas for quantitative characterization of ash pollution, and the ash pollution monitoring model established in the simulation data set can be transferred to the actual operation data set to help the corresponding actual boiler heating surface to quantify the ash pollution. The modeling speed is fast, the amount of data required is small, and the ash pollution trend can be effectively fitted.

[0038] Fourth, existing technical problems:

[0039] 1. Traditional ash accumulation detection methods lack accuracy: Currently, ash accumulation monitoring on boiler heating surfaces mostly relies on ash pollution monitoring systems based on simplified convection heat transfer mechanisms. These methods have limited accuracy and cannot accurately quantify the degree of ash pollution. In addition, the monitoring results deviate seriously from the actual situation when the operating conditions (variable load, variable coal type) change, resulting in reduced equipment operating efficiency, increased fuel consumption, and high maintenance costs.

[0040] 2. Data-driven models lack labeled data: In boiler heating surface fouling characterization, the most difficult part is obtaining labeled data, that is, the actual degree of fouling of the heating surface under all operating conditions. Existing machine learning models often use traditional simplified convection heat transfer mechanism fouling monitoring to generate labeled data. Such models themselves have defects, and the data generated by these models have inherent deficiencies in modeling and low accuracy. In addition to slightly faster calculation speed, they have no practical application value.

[0041] 3. Soot accumulation is difficult to synchronize with soot blowing optimization: Traditional soot cleaning methods are difficult to flexibly adjust according to the soot accumulation on the boiler heating surface. Soot cleaning is often performed according to a fixed cycle, resulting in excessive soot cleaning or soot cleaning only when there is too much soot accumulation. This model cannot make real-time adjustments based on the soot accumulation in actual boiler operation, affecting the overall efficiency and economy of the boiler.

[0042] The technical solution and significant technical progress of the present invention:

[0043] 1. Combine transfer learning to achieve wide adaptability of the model:

[0044] The present invention pre-trains the boiler heating surface ash pollution quantitative characterization model by combining an autoencoder (AE) with support vector regression (SVR), and performs transfer learning on the target domain data, so that the model can adapt to different boiler operating conditions. Compared with traditional single model training, the transfer learning model of the present invention realizes the efficient application of the model in different power plants or boilers through data interaction between the source domain and the target domain, greatly improving the adaptability and generalization ability of the model. This innovation solves the problem of insufficient generalization of existing models, allowing the model to be used across devices and environments.

[0045] 2. Accurate dust characterization based on deep feature extraction:

[0046] The present invention proposes to use an autoencoder (AE) to perform deep feature extraction on boiler operation data, thereby achieving accurate characterization of the degree of ash pollution. Compared with the traditional machine learning model based on shallow feature extraction, the present invention can mine the deep features in the boiler operation data, and combine the SVR algorithm for nonlinear mapping and regression calculation, which greatly improves the accuracy of ash pollution quantification. This hybrid modeling method of deep learning combined with machine learning makes the characterization model of the present invention about 15%-20% more accurate than traditional methods.

[0047] 3. Dynamic characterization of dust pollution and optimization of dust cleaning:

[0048] The present invention can analyze the accumulation of ash and dirt on the boiler heating surface in real time and dynamically adjust the soot blowing strategy through the transfer learning model. Through the accurate prediction of the ash accumulation trend and the quantification of ash and dirt, the present invention provides a flexible cleaning optimization solution, which solves the problem that traditional cleaning is too fixed and lacks flexibility. Cleaning optimization can not only avoid excessive cleaning, but also effectively reduce the decline in boiler operating efficiency caused by ash accumulation. It is estimated that after the cleaning cycle is optimized, the operating efficiency of the boiler can be improved by about 5%-10%, and the fuel cost can be reduced by about 5%.

[0049] 4. Improve boiler operation efficiency and reduce maintenance costs:

[0050] Through accurate ash quantification and dynamic ash cleaning optimization strategy, the present invention can significantly improve the operating efficiency of the boiler and reduce the energy consumption during boiler operation. In addition, the maintenance cost of the equipment is also reduced. Boiler maintenance personnel can predict and judge the ash accumulation through the model to avoid frequent shutdowns for inspection and maintenance. It is expected that the annual boiler fuel and maintenance costs will be reduced by about 10%-15%.

[0051] Significant technological advancements:

[0052] 1. Innovative application of dynamic simulation model in boiler ash pollution quantification:

[0053] There are two major problems in boiler ash pollution quantification: one is the lack of a large amount of valid label data, and the other is the difficulty of reverse solution of detailed mechanism modeling. The present invention uses a detailed mechanism dynamic simulation model to generate simulation label data for the first time. Since the simulation model takes into account the detailed mechanism of the entire boiler process, the simulation accuracy is high, which solves the problem of lack of a large amount of valid label data.

[0054] 1. Innovative application of transfer learning in boiler ash pollution quantification:

[0055] The present invention introduces transfer learning for the first time in boiler heating surface ash fouling characterization, realizes hybrid modeling based on simulation data and actual data, and solves the problem of difficulty in reverse solution of detailed mechanism modeling and poor model transferability in the prior art. Through fine-tuning strategies, the model can adapt to the actual operation of different boilers, realize cross-platform model application, and fill the gap in boiler heating surface ash fouling quantitative characterization technology.

[0056] 2. High-precision dust quantification and real-time prediction capabilities:

[0057] By using the autoencoder to extract deep features of boiler operation data, the present invention can capture the dynamics of ash accumulation on the boiler heating surface and achieve high-precision ash characterization in combination with SVR. Compared with the existing machine learning methods based on shallow feature extraction, the present invention can more accurately predict the ash accumulation of the boiler heating surface under different operating conditions, solving the problem of low ash accumulation prediction accuracy in the prior art.

[0058] 3. Intelligent cleaning strategy optimization:

[0059] The intelligent cleaning strategy of the present invention can dynamically adjust the cleaning cycle and cleaning intensity based on the accurate prediction of the ash accumulation trend by the transfer learning model, which not only avoids the energy waste caused by excessive cleaning, but also avoids the impact of excessive ash accumulation on the heat transfer efficiency of the boiler. This intelligent cleaning optimization strategy realizes the automated management of boiler cleaning, making the boiler operation more efficient and economical.

[0060] In summary, the present invention not only solves the technical problems of accuracy, model migration, cleaning optimization, etc. in the existing boiler ash pollution quantitative characterization, but also significantly improves the boiler operation efficiency, reduces fuel and maintenance costs, and has extremely high industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0062] Figure 1 is a flow chart of a method for quantitatively characterizing boiler heating surface ash pollution based on transfer learning provided by an embodiment of the present invention;

[0063] Figure 2 It is a schematic diagram of gray pollution fitting results of the AE-SVR model for the final superheater in the source domain simulation data set provided by an embodiment of the present invention;

[0064] Figure 3 It is a schematic diagram of gray pollution fitting results of the AE-SVR model for the final superheater in the actual operation data set of the target domain provided by an embodiment of the present invention;

[0065] Figure 4 1. It is a schematic diagram comparing the gray pollution fitting results of the AE-SVR model of the final stage superheater provided by the embodiment of the present invention and the calculation results of the clean factor method;

[0066] Figure 5 It is a structural block diagram of a boiler heating surface ash pollution quantitative characterization system based on transfer learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0068] Industrial application example 1: Ash monitoring and cleaning optimization of the heating surface of a coal-fired power plant boiler

[0069] In large coal-fired power plants, ash accumulation on the boiler heating surface is a major issue affecting boiler efficiency and safety. Ash accumulation reduces heat transfer efficiency, increases fuel consumption, and causes overheating of the heating surface and equipment damage. Therefore, how to quantify the ash contamination of the boiler heating surface in real time and accurately, and optimize it through effective cleaning strategies, is the key to improving the economy and operating efficiency of power plants.

[0070] The ash pollution quantitative characterization modeling method based on transfer learning provided by the present invention can accurately characterize and predict the ash situation on the heating surface under the actual operating conditions of the coal-fired power plant boiler.

[0071] 1. Data acquisition: First, the boiler operation data under different loads and working conditions are obtained through the boiler simulation model, especially the change data of the ash pollution coefficient. The actual power plant boiler operation data is obtained through sensor monitoring, including temperature, pressure, flue gas composition and other related parameters.

[0072] 2. Modeling process: Use autoencoders and support vector regression models to model the ash pollution quantification representation of boiler simulation data (source domain) to obtain the best pre-trained model. Then, migrate the pre-trained model to the actual power plant boiler operation data (target domain), and fine-tune the model parameters to adapt the model to the ash pollution quantification task under actual boiler operating conditions.

[0073] 3. Soot cleaning optimization: By analyzing the soot accumulation trend and soot contamination situation under different soot blowing cycles, it helps power plants optimize soot blowing strategies to avoid reduced heat transfer efficiency or equipment damage caused by excessive soot blowing or untimely soot cleaning.

[0074] Effects and benefits:

[0075] Improve boiler efficiency: This method can quantify the ash accumulation in real time, guide power plants to formulate reasonable ash removal plans, and optimize boiler operating efficiency. It is expected that the boiler thermal efficiency will increase by 2% to 3%.

[0076] Reduce maintenance costs: By avoiding equipment damage caused by excessive cleaning or excessive dust accumulation, reducing maintenance frequency and downtime, annual maintenance costs can be reduced by 10% to 15%.

[0077] Save fuel costs: The ash cleaning optimization solution under real-time monitoring of ash accumulation effectively reduces the fuel consumption caused by the decrease in boiler heat transfer efficiency, saving more than 5% of fuel costs.

[0078] Industrial application example 2: Monitoring and prediction of dust on the heating surface of boilers in urban central heating systems

[0079] The heating surfaces of boilers in urban central heating systems are prone to dust accumulation, especially during high-load operation in winter, which can reduce boiler heating efficiency and even affect the heating effect of residents. Traditional dust accumulation detection relies on manual regular inspections, which is not real-time and can easily lead to unstable heating or energy waste.

[0080] By applying the transfer learning-based heating surface ash pollution quantitative characterization modeling method of the present invention in the urban centralized heating system, the ash accumulation on the boiler heating surface can be accurately predicted under complex working conditions, the cleaning frequency can be optimized, and the stable operation of the heating system can be ensured.

[0081] 1. Source domain and target domain data selection: Through the boiler simulation model of the centralized heating system, the heated surface dust simulation data of the source domain is obtained, and the real-time operation data of the boiler is obtained as the target domain data through the field sensor, including parameters such as temperature, pressure, steam production, and flue gas composition.

[0082] 2. Model construction and migration: The autoencoder and support vector regression algorithm are used to model the ash pollution representation of the source domain data, and the pre-trained model is applied to the actual boiler operation data through transfer learning technology. After fine-tuning, the model can accurately predict the ash accumulation trend of the heating surface under different operating conditions.

[0083] 3. Dust cleaning management: Based on the predicted dust accumulation trend, the system can intelligently generate a dust cleaning plan to avoid the decrease in heating efficiency caused by excessive dust accumulation, and also avoid energy waste caused by frequent dust cleaning.

[0084] Effects and benefits:

[0085] Improve heating efficiency: By real-time monitoring of ash accumulation, efficient operation of the boiler is ensured, and the heating efficiency is expected to increase by more than 5%.

[0086] Stable heating service: Accurately predict boiler ash accumulation to effectively avoid heating interruptions or instability caused by excessive ash accumulation, ensure heating service quality, and improve residents' satisfaction.

[0087] Reduce cleaning frequency: Through accurate characterization and prediction of dust accumulation, unnecessary cleaning operations are reduced. It is expected that the annual cleaning frequency will be reduced by 20%, while reducing operating and maintenance costs.

[0088] These two embodiments demonstrate the industrial application of the present invention in different types of boilers, especially how to improve boiler operating efficiency, reduce costs and achieve intelligent management of the system through transfer learning and ash pollution quantification characterization technology.

[0089] In view of the problems existing in the prior art, the present invention provides a quantitative characterization modeling method for ash contamination on boiler heating surfaces based on transfer learning. The present invention is described in detail below in conjunction with the accompanying drawings.

[0090] like Figure 1 As shown, the boiler heating surface ash pollution quantitative characterization modeling method based on transfer learning provided by the embodiment of the present invention includes:

[0091] S1: Based on the boiler dynamic simulation model, obtain the full load and full operating condition simulation operation data when the ash pollution coefficient of each level of heating surface changes;

[0092] S2: Select the heating surface operation data set S1 in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain;

[0093] S3: Based on the autoencoder and support vector regression algorithm, a quantitative characterization model of boiler heating surface ash pollution in the source domain is built and parameter optimization and training are performed to obtain a pre-trained model with the best heating surface ash pollution fitting result;

[0094] S4: Migrate the S3 pre-trained model to the target domain dataset, and use the method of fine-tuning the model parameters to make the model adapt to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface;

[0095] S5: Use the actual sootblowing execution action to determine the soot accumulation trend and soot contamination degree of the heated surface in different sootblowing cycles, and complete the verification and application of the transfer learning model.

[0096] The full-load and full-operating condition operation data acquisition under the ash pollution coefficient of each level of heating surface in S1 is obtained by using Apros to establish a boiler-furnace coupling dynamic simulation model for research. In the module property setting, the change of the ash pollution coefficient of the specific heating surface of the boiler is used as an indicator feature, ranging from 0.8-1. When the coefficient is 1 (100%), it means that the heating surface is clean. As the coefficient decreases, it means that the cleanliness of the surface of the heating surface begins to decrease due to ash accumulation; when the coefficient rises back to 1 again, it means that the heating surface has returned to a clean state due to the execution of the soot blowing operation. The simulation data sampling interval is 1 minute. For each level of heating surface, the original simulation data set composed of multiple parameters such as load, coal feed, working fluid side flow, working fluid side temperature, working fluid side pressure, flue gas side temperature and pressure is used for the subsequent quantitative characterization of ash pollution of each level of heating surface of the boiler.

[0097] The division of the data set in S2 is based on the basic concept of transfer learning. Transfer learning uses the idea of ​​knowledge transfer to apply the feature relationship learned in one field to a similar field, and divides it into the source field D s (Sourcedomain) and target domain D t (Target domain), the data in the source domain are generally annotated, and the data quality is high. Then, valuable knowledge such as feature expressions in the source domain data set is obtained through relevant algorithm models; while the data in the target domain are generally not annotated and the data distribution or data features are similar to those in the source domain. With the help of relevant algorithm models, similar feature expressions can be transferred from the source domain to the target domain, helping the target domain to complete data prediction or classification and improve the performance of the target task. In the above, the simulation operation data and the gray pollution coefficient label of the heating surface are selected as the source domain, and the actual operation data of the corresponding heating surface is selected as the target domain, and the input parameters of the model are relatively consistent.

[0098] The pre-trained model with the best fitting result of the heated surface dust pollution built in S3 has the following specific process: facing the source domain data, combining the unsupervised learning algorithm Autoencoder (AE) and the traditional machine learning algorithm Support Vector Regression (SVR), using the encoder part in the AE model as a feature extractor, inputting the data samples into the encoder part for data dimension reduction and deep feature extraction, and then inputting the feature vector and data label into the SVR model to perform regression, and comparing the obtained predicted value of the heated surface dust pollution degree with the dust pollution coefficient label of the source domain, and obtaining the model framework and parameter setting with the best fitting result through parameter optimization, which is represented as the pre-trained model AE-SVR for transfer learning, and subsequently used for model transfer learning for the target domain.

[0099] The autoencoder AE described in S3 includes only the encoding part:

[0100] The encoder E transforms the input feature vector x into a latent feature vector z, i =E(x i |θ E ).

[0101] In the formula, x is the input feature, θ E is the parameter of the encoder. The encoder part extracts features and reduces the dimension of the input data.

[0102] Among them, the basic idea of ​​the support vector regression SVR described in S3 is to map the sample data to a high-dimensional space through a nonlinear mapping function and perform linear regression in the high-dimensional space. By constructing a loss function that minimizes the training set data label and the model fitting value, an objective function f(x) is finally determined, which is expressed as:

[0103]

[0104] Where w is the weight coefficient vector; is the high-dimensional feature space obtained by mapping the input space x; b is the bias.

[0105] In S4, the pre-model in S3 is migrated to the target domain. Since the distribution differences of different data sets when migrating to the target domain will cause the model effect to deteriorate, different fine-tuning strategies need to be selected in deep transfer learning to adapt the parameters or models to the new data set and improve the model effect. For the AE-SVR model, a fine-tuning strategy is adopted to change the parameters of the AE encoder part to adapt to the target domain data.

[0106] In S5, in the verification of the migration results of the quantitative characterization model of the heating surface ash fouling for actual operating data, the actual boiler unit lacks label values ​​that characterize the degree of ash fouling on the heating surface, that is, the simulation data set in the source domain has labels, but the target domain data set does not have labels. Therefore, the actual sootblowing execution action is used to judge whether the AE-SVR model fitting results of each level of heating surface in different sootblowing cycles reflect the correct ash accumulation trend and ash fouling degree, thereby verifying the effectiveness and feasibility of the AE-SVR transfer learning method.

[0107] like Figure 5 As shown, a boiler heating surface ash pollution quantitative characterization system based on transfer learning provided by an embodiment of the present invention includes:

[0108] An operation data acquisition module is used to obtain full-load and full-operating condition simulation operation data when the ash pollution coefficient of each level of heating surface changes based on the boiler dynamic simulation model;

[0109] A selection module is used to select the heating surface operation data set in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain;

[0110] The training module is used to build a boiler heating surface ash pollution quantitative characterization model in the source domain based on the autoencoder and support vector regression algorithm, and perform parameter optimization and training to obtain a pre-trained model with the best heating surface ash pollution fitting result;

[0111] The migration module is used to migrate the pre-trained model to the target domain dataset and fine-tune the model parameters to adapt the model to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface;

[0112] The verification module is used to use the actual soot blowing execution action to determine the soot accumulation trend and soot contamination degree of the heated surface in different soot blowing cycles, and complete the verification and application of the transfer learning model.

[0113] A computer device is provided in an embodiment of the present invention, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantitative characterization of ash pollution on boiler heating surfaces based on transfer learning.

[0114] A computer-readable storage medium is provided in an embodiment of the present invention, storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the method for quantitative characterization of boiler heating surface ash fouling based on transfer learning.

[0115] An embodiment of the present invention provides an information data processing terminal, which is used to implement the boiler heating surface ash pollution quantitative characterization system based on transfer learning.

[0116] The research object of this example is a 1000MW ultra-supercritical secondary reheat unit. The boiler is a secondary intermediate reheat unit and an ultra-supercritical pressure transformer. The simulation model of the research object is built with the help of Apros dynamic simulation software for subsequent data collection. In the module property setting, the change of the ash pollution coefficient of the specific heating surface of the boiler is used as an indicator feature. When the coefficient is 1 (100%), it means that the heating surface is clean. As the coefficient decreases, it means that the cleanliness of the surface of the heating surface begins to decrease due to ash accumulation; when the coefficient rises back to 1 again, it means that the heating surface has returned to a clean state due to the soot blowing operation. The simulation data sampling interval is 1 minute. For each level of heating surface, the original simulation data set consisting of multiple parameters such as load, coal feed, working fluid side flow, working fluid side temperature, working fluid side pressure, flue gas side temperature and pressure is used as the source domain. At the same time, a total of 380,965 sets of actual operation data were collected from the power plant DCS system from March 9, 2021 to May 10, 2022 as the target domain. The sampling interval and related parameters were consistent with the simulation data. The field data had completed the preprocessing step of mean denoising before modeling.

[0117] With the help of the concept of transfer learning, the present invention aims to utilize the established model and gray pollution label expression in the simulation data, i.e., the source domain, and migrate them to the same actual heating surface, i.e., the target domain, to reduce the difficulty of modeling. Therefore, facing the source domain data, the unsupervised learning algorithm Autoencoder (AE) and the traditional machine learning algorithm Support Vector Regression (SVR) are combined, and the encoder part in the AE model is used as a feature extractor. The data samples are input into the encoder part for data dimension reduction and deep feature extraction. After that, the feature vector and the data label are input into the SVR model together to perform regression, and the obtained predicted value of the gray pollution degree of the heating surface is compared with the gray pollution coefficient label of the source domain. Through parameter optimization, it is verified that when the number of neurons in the input hidden layer is 32 and 16 respectively, the training batch is set to 16, and the number of training times is 200, the model can be fully pre-trained, and the label fitting has the highest accuracy. Finally, the model framework and parameter settings with the best fitting results are obtained, which are expressed as the pre-training model of transfer learning as follows Figure 2 It can be seen that the quantitative characterization result curve of the final superheater based on the AE-SVR pre-training model in the source domain is most consistent with the trend of the original ash pollution coefficient label, and it is believed that it can be used as a pre-training model for subsequent model migration and model fine-tuning in the target domain.

[0118] Since there is a lack of label values ​​that characterize the degree of dust fouling on the heating surface in the actual boiler unit, that is, the simulation data set in the source domain has labels, but the target domain data set does not have labels, the actual soot blowing action is used to determine whether the AE-SVR model fitting results of each level of heating surface in different soot blowing cycles reflect the correct dust accumulation trend and dust fouling degree. Figure 3 As shown in the figure, the fitting results of the AE-SVR model are close to the actual dust pollution range, the trend is correct, and the fitting results in the soot blowing action range have a rapid rise stage, indicating that the actual soot blowing action effectively removes the dust accumulation on the heating surface, and then the cleanliness of the heating surface gradually decreases with the dust deposition process. Therefore, it can be concluded that in the absence of dust quantitative characterization labels, the simulation data modeling results can be transferred to effectively fit the dust accumulation trend of the actual heating surface.

[0119] In summary, through the verification of this embodiment, it is shown that the quantitative characterization method of boiler heating surface ash fouling based on transfer learning can reduce the difficulty of modeling in actual operation data that lacks characterization ash fouling labels, and use the ash fouling label expression and feature expression in the simulation data set to help the actual operation data effectively fit its ash accumulation trend and verify the effectiveness of the actual soot blowing action. In addition, only a small number of heating surface objects in the simulation data need to be modeled to obtain relevant feature expressions, while the heating surface objects in the actual data are modeled according to transfer learning, and the parameters are fine-tuned locally or globally to adapt to the actual objects, which further reduces the time cost and economic cost of modeling training and improves the applicability and generalization of the model.

[0120] In order to verify the effectiveness of boiler heating surface fouling monitoring modeling based on transfer learning, the current mainstream method is selected, and the model results of the final superheater are used as an example to calculate the fouling degree / cleanliness factor based on the heat transfer mechanism for comparison and verification. Figure 4 As shown in the figure, it is found that the model and ash pollution label expression established in the final superheater of the simulation data are transferred to the same actual heating surface for ash pollution fitting. The result is close to the actual ash pollution range and the trend is correct, while the calculation result of the clean factor fluctuates greatly, and there are some time periods where the result remains basically unchanged. It is believed that it cannot reflect the correct ash accumulation trend and has poor accuracy and stability. In summary, the quantitative characterization method of ash pollution on the boiler heating surface based on transfer learning provides a new modeling idea for the field of boiler ash pollution monitoring. The data output by the simulation model and the label representing the ash pollution are used to help the corresponding boiler heating surface in the actual operation data to carry out ash pollution monitoring modeling, which solves the problem of lack of labels representing ash pollution in practice. The modeling speed is fast, the amount of data required is small, and the ash pollution trend is effectively fitted, which has high feasibility.

[0121] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A quantitative characterization method for boiler heating surface ash pollution based on transfer learning, characterized in that: The following steps are involved: S1: A boiler dynamic simulation model is established based on three transmissions and one reaction, and verified with actual operation data. Dynamic simulation of variable operating conditions and variable ash-pollution coefficient is carried out to obtain full-load and full-operating condition simulation operation data when the ash-pollution coefficient of each level of heating surface changes; S2: Select the heating surface operation data set S1 in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain; S3: Based on the autoencoder and support vector regression algorithm, a quantitative characterization model of boiler heating surface ash pollution in the source domain is built and parameter optimization and training are performed to obtain a pre-trained model with the best heating surface ash pollution fitting result; S4: Migrate the S3 pre-trained model to the target domain dataset, and use the method of fine-tuning the model parameters to make the model adapt to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface; S5: Use the actual soot blowing execution action to determine the soot accumulation trend and soot contamination degree of the heating surface in different soot blowing cycles, and complete the transfer learning model verification; S6: Solidify the verified model in the industrial computer, connect it to the DCS system through Modbus and OPC communication, obtain real-time operation data, and obtain the contamination status of each level of heating surface after model calculation. The calculation results are further used for intelligent soot blowing decision-making or sent back to the DCS for reference by operators.

2. The method for quantitative characterization of boiler heating surface ash pollution based on transfer learning as claimed in claim 1, characterized in that: The full-load and full-operating condition data acquisition under the ash pollution coefficient of each level of heating surface in S1 is obtained by establishing a boiler-furnace coupling dynamic simulation model with the help of Apros. In the module property setting, the change of the ash pollution coefficient of the specific heating surface of the boiler is used as the indicator feature, ranging from 0.6 to 1. When the coefficient is 1, it is indicated that the heating surface is clean. As the coefficient decreases, it indicates that the cleanliness of the heating surface begins to decrease due to ash accumulation; when the coefficient rises back to 1 again, it means that the heating surface has recovered to a clean state due to the soot blowing operation; the simulation data sampling interval can be 10 seconds to 1 minute, and the original simulation data set consisting of multiple parameters such as load, coal feed rate, working fluid side flow, working fluid side temperature, working fluid side pressure, flue gas side temperature and pressure for each level of heating surface is used for the subsequent quantitative characterization of ash pollution on the heating surfaces of each level of the boiler.

3. The method for quantitative characterization of boiler heating surface ash pollution based on transfer learning as claimed in claim 1, characterized in that: The division of the data set in S2 is based on the basic concept of transfer learning; Transfer learning uses the idea of ​​knowledge transfer to apply the feature relationship learned in one field to a similar field, which is divided into source field D s and the target domain D t The data in the source domain are generally labeled and have high data quality. Then, valuable knowledge such as feature expressions in the source domain data set are obtained through relevant algorithm models. The data in the target domain are generally not labeled and the data distribution or data features are similar to those in the source domain. With the help of relevant algorithm models, similar feature expressions can be transferred from the source domain to the target domain, helping the target domain to complete data prediction or classification.

4. The method for quantitative characterization of boiler heating surface ash pollution based on transfer learning as claimed in claim 1, characterized in that: The specific process of the pre-trained model with the best fitting result of the heated surface dust pollution built in S3 is as follows: For source domain data, the unsupervised learning algorithm autoencoder AE and the traditional machine learning algorithm support vector regression SVR are combined. The encoder part in the AE model is used as a feature extractor, and the data samples are input into the encoder part for data dimension reduction and deep feature extraction. Then, the feature vector and data label are input into the SVR model to perform regression. The obtained predicted value of the dust pollution degree of the heated surface is compared with the dust pollution coefficient label of the source domain. Through parameter optimization, the model framework and parameter setting with the best fitting result are obtained, which is represented as the pre-trained model AE-SVR for transfer learning, and then used for model transfer learning for the target domain; The autoencoder AE described in S3 includes only the encoding part: The encoder E transforms the input feature vector x into a latent feature vector z, i =E(x i |θ E ); In the formula, x is the input feature, θ E is the parameter of the encoder; the encoder part performs feature extraction and dimension reduction on the input data; The basic idea of ​​the support vector regression SVR described in S3 is to map the sample data to a high-dimensional space through a nonlinear mapping function and perform linear regression in the high-dimensional space; by constructing a loss function that minimizes the training set data label and the model fitting value, an objective function f(x) is finally determined, which is expressed as: Where w is the weight coefficient vector; is the high-dimensional feature space obtained by mapping the input space x; b is the bias.

5. The method for quantitative characterization of boiler heating surface ash pollution based on transfer learning as claimed in claim 1, characterized in that: In S4, the pre-model in S3 is migrated to the target domain. In deep transfer learning, different fine-tuning strategies need to be selected to adapt the parameters or models to the new data set and improve the model effect. For the AE-SVR model, a fine-tuning strategy of changing the parameters of the AE encoder part to adapt to the target domain data is adopted.

6. The method for quantitative characterization of boiler heating surface ash pollution based on transfer learning as claimed in claim 1, characterized in that: In S5, in the verification of the migration results of the quantitative characterization model of the heating surface ash fouling for actual operating data, the actual boiler unit lacks label values ​​that characterize the degree of ash fouling on the heating surface, that is, the simulation data set in the source domain has labels, and the target domain data set does not have labels. The actual sootblowing execution action is used to judge whether the AE-SVR model fitting results of the heating surfaces at all levels in different sootblowing cycles reflect the correct ash accumulation trend and ash fouling degree, thereby verifying the effectiveness and feasibility of the AE-SVR transfer learning method.

7. A transfer learning-based boiler heating surface ash pollution quantitative characterization system implementing the transfer learning-based boiler heating surface ash pollution quantitative characterization method according to any one of claims 1 to 6, characterized in that: The boiler heating surface ash pollution quantitative characterization system based on transfer learning includes: An operation data acquisition module is used to obtain full-load and full-operating condition simulation operation data when the ash pollution coefficient of each level of heating surface changes based on the boiler dynamic simulation model; A selection module is used to select the heating surface operation data set in the simulation model as the source domain, and select the operation data set corresponding to the same heating surface in the actual coal-fired power plant boiler as the target domain; The training module is used to build a boiler heating surface ash pollution quantitative characterization model in the source domain based on the autoencoder and support vector regression algorithm, and perform parameter optimization and training to obtain a pre-trained model with the best heating surface ash pollution fitting result; The migration module is used to migrate the pre-trained model to the target domain dataset and fine-tune the model parameters to adapt the model to the ash pollution quantification characterization task under the actual operation data of the same boiler heating surface; The verification module is used to use the actual soot blowing execution action to determine the soot accumulation trend and soot contamination degree of the heated surface in different soot blowing cycles, and complete the verification and application of the transfer learning model.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for quantitative characterization of ash pollution on boiler heating surfaces based on transfer learning as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method for quantitative characterization of boiler heating surface ash fouling based on transfer learning as claimed in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal is used to implement the boiler heating surface ash pollution quantitative characterization system based on transfer learning as described in claim 7.

Images