Fusion mechanism and data boiler heating surface soot quantification characterization method and system
The ash pollution quantitative characterization method, which combines dynamic simulation model and autoencoder model, solves the problems of insufficient accuracy and tag data in ash pollution monitoring in traditional methods. It realizes high-precision ash pollution monitoring of boiler heating surfaces, optimizes soot blowing strategy, and improves the operating efficiency and economic benefits of coal-fired power plants.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2024-01-11
- Publication Date
- 2026-06-19
AI Technical Summary
Traditional methods for monitoring ash and fouling on boiler heating surfaces suffer from low computational accuracy and a lack of labeled data, resulting in insufficient accuracy and generalization performance of ash and fouling monitoring systems, which are unable to effectively address the impact of coal quality changes and dynamic load variations.
By employing a dynamic simulation model and an unsupervised learning autoencoder model, combined with boiler operation data, and through the nonlinear expression and analysis of characteristic parameters, a quantitative characterization method for ash pollution is established to avoid the influence of coal quality and load changes and improve monitoring accuracy.
It enables high-precision monitoring of ash and dirt on boiler heating surfaces, reduces soot blowing steam consumption, improves unit operation safety, reduces operating costs, and enhances fuel utilization efficiency and environmental friendliness.
Smart Images
Figure CN118194103B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ash pollution monitoring technology, and in particular relates to a method and system for quantitative characterization of ash pollution on boiler heating surfaces that integrates mechanisms and data. Background Technology
[0002] Coal-fired power generation is the cornerstone of my country's power security. In 2022, coal-fired power, with less than 50% of the installed capacity, contributed 60% of the electricity and handled 70% of the peak load. Under the "dual carbon" background, coal-fired power is gradually transforming from a basic power source that bears the base load to a regulating power source that smooths out fluctuations in renewable energy and ensures the security 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 long used non-designed coal types, and the quality of the coal fed into the boiler changes frequently and to a great extent.
[0003] Due to the high ash content in coal, ash accumulation and slagging are inevitable during the operation of coal-fired power plant boilers. Ash accumulation and slagging are mainly related to boiler structure, load, and coal quality. For a specific boiler, differences in coal feed rate, internal boiler temperature level, and flue gas velocity under different loads affect ash accumulation and slagging on the heating surfaces. Different coal types have different ash content and composition, resulting in significant differences in their ash accumulation and slagging characteristics. With flexible coal quality and load, traditional timed and quantitative soot blowing strategies often fail to meet the actual soot blowing needs of the heating surfaces, leading to localized or even global under-blowing or over-blowing. Under-blowing causes increased flue gas temperature, reduces boiler efficiency, and in severe cases, may lead to increased flue resistance and severe slagging and shedding in the furnace. Over-blowing not only wastes soot blowing steam but also causes thinning of the heating surface tube walls, increasing the risk of tube rupture. Therefore, soot blowing based on the actual ash and fouling conditions of the heating surfaces is crucial, and the fundamental solution lies in developing effective and reliable real-time monitoring technology for heating surface ash and fouling.
[0004] Based on publicly available information, real-time monitoring technologies for ash contamination on heated surfaces can be categorized into two types according to their underlying principles: those based on the heat transfer mechanism of the heated surface and those based on data-driven approaches. The heat transfer mechanism-based approach primarily involves ash contamination thermal resistance, heat transfer efficiency ratio, and cleanliness / contamination level. However, these methods generally suffer from low computational accuracy. This is because the modeling methods based on the heat transfer mechanism significantly simplify the heat transfer process, typically considering only convective heat transfer while neglecting the more complex radiative heat transfer. Furthermore, the modeling process does not account for the impact of coal quality changes and dynamic load variations on the heat transfer process. In addition, inaccurate flue gas temperature data during implementation further complicates the calculation results. Data-driven modeling methods, lacking labeled data (i.e., the actual ash contamination status of the heated surface is unknown), suffer from insufficient model accuracy and generalization performance, thus affecting the accuracy of ash contamination monitoring systems in making relevant diagnoses and decisions. Therefore, it is crucial to overcome the limitations of traditional mechanism-based and data-driven modeling to establish high-precision ash contamination monitoring models.
[0005] Based on the above analysis, the existing technologies have the following problems and shortcomings: Traditional mechanism-based modeling methods have greatly simplified the heat transfer process formulas. On the one hand, mechanism modeling does not consider the influence of radiation heat transfer, nor does it consider the impact of coal quality changes on flue gas flow rate, thus affecting heat transfer efficiency. According to our research, changes in coal quality in power plants can cause flue gas flow rate changes of up to 10%, and it also does not consider the influence of steam flow and the thermal inertia of metal walls on heat transfer during dynamic load changes. On the other hand, the parameter measurement points on the flue gas side are affected by the installation location of the measurement points, making it difficult to guarantee the validity and reliability of the measurements. For example, when a unit is under low load, the flue gas is concentrated in the middle of the flue, while the flue gas temperature measurement point is installed near the wall, resulting in the unreasonable phenomenon that the flue gas temperature is lower than the steam temperature. In addition, the flue gas baffle will change the distribution of flue gas flow in the tail flue, thereby changing the heat transfer coefficient of the heated surface, which is not considered by traditional mechanism models. Data-driven modeling methods basically use on-site operating data directly for modeling. Depending on the method, they mainly include unsupervised learning, such as principal component analysis, and supervised learning, such as artificial neural networks. Both methods face the challenge that the ash and dirt condition of the heated surface is unknown during field operation. Due to the lack of labeled data, the effectiveness of the model can only be verified by observing the trend of ash and dirt changes on the heated surface during the soot blowing cycle, resulting in insufficient accuracy and generalization performance. Some patented technologies use mechanistic models to calculate the cleanliness / fouling degree of the heated surface as labeled data in supervised learning. The accuracy of these methods depends entirely on the mechanistic model, offering no advantage other than fast computation speed. However, the ash and dirt contamination process on the heated surface is relatively slow, making computational speed unimportant. Summary of the Invention
[0006] To overcome the limitations of traditional mechanism-based and data-driven modeling, this invention provides a method and system for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanism and data. By using dynamic simulation models and unsupervised learning modeling to learn the nonlinear expression of strongly correlated feature parameters, and by using feature parameter data analysis and new feature construction methods, the influence of coal quality and dynamic load operation on the characterization of ash and fouling on boiler heating surfaces is avoided to a certain extent. Finally, the established model is verified and applied through actual power plant data, thereby improving the accuracy of the boiler heating surface ash and fouling monitoring model.
[0007] This invention is implemented as follows: a method and system for quantitative characterization of ash fouling on boiler heating surfaces that integrates mechanisms and data, comprising:
[0008] S1: Establish a dynamic simulation model of the boiler, including the control system, based on first principles, and verify it with field data;
[0009] S2: Obtain full-load, full-condition operating data during the cleaning of the heated surface based on a dynamic simulation model;
[0010] S3: Build an autoencoder monitoring model for the boiler heating surface, and train the model using the data obtained in S2 to minimize the reconstruction error;
[0011] S4: Based on the dynamic simulation model, adjust the ash and slag coefficient of the heating surface to simulate the ash and slag formation process of the heating surface, and obtain full load and full working condition operation data when the ash and slag coefficient of the heating surface changes;
[0012] S5: Input the S4 data into the S3 trained autoencoder model, obtain the reconstruction error, and establish the correlation between the model reconstruction error and the gray pollution coefficient;
[0013] S6: Validate and apply the model using actual power plant operating data and soot blowing time.
[0014] Furthermore, the simulation model construction in S1 involves selecting a power plant boiler as the object, establishing an accurate dynamic simulation model of the boiler object that includes the control system based on first principles such as conservation of mass, conservation of energy, and conservation of momentum, acquiring continuous time series data with a sampling interval of 5-60 seconds from the dynamic simulation platform, and comparing and verifying the simulation model with the main parameters of the unit in actual operation to ensure the consistency of the static and dynamic characteristics of the model with the actual boiler.
[0015] Furthermore, the full-load, full-condition operating data for clean heating surfaces in S2 uses the change in the ash and dirt coefficient of the specific heating surface of the boiler as an indicator, ranging from 0 to 1. That is, the closer the ash and dirt coefficient is to 0, the higher the degree of ash and dirt on the heating surface; the closer it is to 1, the lower the degree of ash and dirt and the cleaner the heating surface. Setting the ash and dirt coefficient to 1, meaning no ash and dirt on the heating surface, and using the dynamic simulation model to output full-load, full-condition operating data for clean heating surfaces, is used for subsequent related parameter research and model training.
[0016] Furthermore, in S3, the boiler heating surface self-encoder monitoring model is constructed, including an encoder and a decoder:
[0017] Encoding part: Encoder E transforms the input feature vector x into a latent feature vector z, z i =E(x) i |θ E Decoding part: Decoder D reconstructs the latent feature vector z into the input feature vector x', x' i =D(z) i |θ D ).
[0018] Where x is the input feature, θ E and θ D These are the parameters for the encoder and decoder, respectively;
[0019] The input features are reconstructed by training an autoencoder, and the parameters of the encoder and decoder are adjusted to minimize the error between the reconstructed result and the input. The reconstruction error is evaluated using MAE (mean absolute error) or MSE (mean squared error). or The training dataset is input into the autoencoder for model learning, feature data is reconstructed, and the reconstruction error is minimized.
[0020] Furthermore, in S3, the selection of input features for the autoencoder model is based on the heat transfer mechanism and parameter trend analysis of the boiler heating surface. According to the quality of the field data, variables with high reliability are selected as input features from load, feedwater flow rate, inlet and outlet steam temperature of the heating surface, inlet and outlet flue gas temperature of the heating surface, and heat exchange. However, rapid load fluctuations will make the influence of ash on the heating surface on each parameter not prominent, that is, load factors will interfere with the modeling. In order to reduce the impact of load changes on other thermal parameters, the heat exchange and outlet steam temperature of each load section are calculated based on the stable operation data of each load section. A linear fitting curve of heat exchange and outlet steam temperature with load change is established, and new features of ideal heat exchange and ideal outlet steam temperature are constructed.
[0021] Furthermore, in S4, the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes is similar to that in S2. The dynamic simulation model outputs the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes, which serves as the test set for model learning.
[0022] Furthermore, in S5, full-load, full-condition operating data with varying ash and dirt coefficients of the heating surface are selected as the test set for verification. The reconstruction error is calculated, and the correlation between the reconstruction error and the ash and dirt coefficient is studied to achieve quantitative characterization of ash and dirt on the boiler heating surface.
[0023] Furthermore, in S6, after preprocessing the actual operating data of the power plant's heating surface, since the characteristics of the ash and dirt coefficient cannot be obtained in the actual operation of the power plant, the actual soot blowing action of the power plant can be used to assist in completing the training and test dataset division steps similar to the dynamic simulation model. The model is then learned in the established autoencoder model and compared with the results of the cleanliness / fouling degree characterization parameters calculated based on the traditional mechanism model to verify the effectiveness and accuracy of this method, thus enabling its practical application.
[0024] Another object of the present invention is to provide a quantitative characterization system for boiler heating surface ash fouling that applies the aforementioned fusion mechanism and data, comprising:
[0025] Boiler Dynamic Simulation Model Building Module: Used to build accurate dynamic simulation models of object boilers, including control systems;
[0026] Dataset construction module: used to collect full-load, full-condition operating data with the change of ash and dirt coefficient of the heated surface as an indicator feature;
[0027] Monitoring model building module: used to build an autoencoder monitoring model for the boiler heating surface and reconstruct feature data;
[0028] Ash and dirt trend monitoring module: Used to verify the test set using an autoencoder model, study the correlation between reconstruction error and ash and dirt coefficient, and realize the quantitative characterization of ash and dirt on boiler heating surfaces.
[0029] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, causing the processor to perform the steps of a method for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data.
[0030] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:
[0031] First, regarding the technical problems existing in the aforementioned prior art and the difficulty of solving these problems, the creative technical effects that arise after solving the problems are described in detail below:
[0032] This invention employs simulation software to establish a dynamic mechanism-compliant simulation model of the heating surface based on actual power plant objects. This effectively reduces the noise generated by the complex conditions inside the furnace in actual power plants, which affects the measurement data at the measurement points. Feature correlation analysis can be performed based on the simulation data to study its influence mechanism, and a high-quality dataset is provided for subsequent algorithm modeling. An unsupervised learning autoencoder is used to build a boiler heating surface ash pollution monitoring model, simplifying the computational workload. Furthermore, it does not rely entirely on the heat transfer mechanism model; ash pollution monitoring modeling can be completed using only a few thermal parameters, avoiding the shortcomings of mechanism-based modeling and data-driven approaches, thus ensuring accuracy. New feature quantities constructed based on data analysis can effectively avoid the impact of variable load operation on model learning, better reflecting the actual operating conditions of power plant boilers under the current flexible operation background. This enables the monitoring of ash pollution trends on different heating surfaces, and the model is validated using actual operating data, demonstrating promising application prospects.
[0033] Secondly, this invention effectively addresses the limitations of traditional mechanistic modeling methods. Traditional methods heavily simplify heat transfer process formulas, neglecting the effects of radiative heat transfer, coal quality variations on flue gas flow and heat transfer efficiency, and the influence of steam flow, metal wall thermal inertia, and flue gas side parameter measurement points on heat transfer during dynamic load changes. These shortcomings lead to significant deviations in monitoring results when the unit operates under varying coal quality or load conditions. Furthermore, data-based modeling methods, lacking labeled data, rely solely on the trend of ash and dirt changes on the heated surface during soot blowing cycles to verify model effectiveness, resulting in insufficient accuracy and generalization performance. This invention overcomes the limitations of both traditional modeling methods. This invention utilizes simulation software to build a dynamic simulation model of the heating surface, enabling in-depth research into the correlation and influence mechanisms among thermal parameters and providing a high-quality dataset for subsequent algorithms. Simultaneously, employing anomaly monitoring and fault diagnosis principles, it selects an unsupervised autoencoder model, allowing it to learn its latent feature representations without relying on labeled data, by inputting relevant parameters for dimensionality reduction and reconstruction. Addressing the impact of variable load processes on modeling accuracy, the heat exchange and outlet steam temperature of each load segment are calculated using stable operating data. Linear fitting curves of heat exchange and outlet steam temperature versus load variation are established, and new features of ideal heat exchange and ideal outlet steam temperature are constructed as model inputs, effectively mitigating the impact of variable load operation on the accuracy of model learning. To verify the relative effectiveness of this modeling method, actual operating data is used to compare and validate the model learning effect. The results show that the boiler heating surface ash and fouling quantitative characterization method and system, which integrates mechanisms and data, further improves monitoring accuracy.
[0034] Third, the expected benefits and commercial value of the technical solution of this invention after transformation are as follows: After the technology is applied to coal-fired power plant boilers, it can realize on-demand soot blowing, reducing soot blowing steam consumption while improving the operational safety of the unit. Taking a 350MW unit as an example, after applying this technology, soot blowing steam consumption is reduced to 876-1208 tons / month, averaging 1063 tons / month. According to the original soot blowing strategy, the boiler steam consumption is about 2014 tons / month, saving 47.3% of steam compared to the original soot blowing strategy. Since reducing soot blowing steam consumption can save about 1300 tons of standard coal / year, it can reduce carbon dioxide emissions by 3444 tons / year. At the same time, after the system is put into operation, it significantly reduces the phenomenon of overheating of the heating surface tube wall caused by "coke collapse". The average monthly overheating time is reduced from 47.3 minutes to 14 minutes, a reduction of 70%, effectively extending the service life of the tube wall and saving 1 million yuan in maintenance costs annually. In addition, it also increases the proportion of high-ash, low-ash melting point, low-priced inferior coal used by the unit, saving 42.09 million yuan in fuel costs. It has significant economic benefits and commercial value.
[0035] The technical solution of this invention solves a long-standing but unresolved technical problem: under the context of flexible coal quality and load, traditional timed and quantitative soot blowing strategies often fail to match the actual soot blowing needs of the heating surfaces, resulting in localized or even global under-blowing or over-blowing of the heating surfaces. Developing effective and reliable real-time monitoring technology for ash and fouling of heating surfaces is a key step in realizing on-demand soot blowing of boiler heating surfaces. Currently, methods for ash and fouling monitoring and modeling in power plant boilers, both domestically and internationally, mainly rely on simplified heat transfer mechanisms to calculate cleanliness / fouling levels to determine the cleanliness of each heating surface. Alternatively, they use mechanistic models to calculate the cleanliness / fouling levels of the heating surfaces, which are then used as labeled data in supervised learning for machine learning modeling to make judgments or predictions. However, the accuracy of these methods depends entirely on the mechanistic model, offering no advantage other than fast calculation speed. The fouling process of heating surfaces is relatively slow, making calculation speed less important. Furthermore, the accuracy of measurement parameters is affected by various complex factors, making it difficult to guarantee accuracy. Especially in power plant boilers operating under normal and flexible conditions, it becomes increasingly difficult to accurately assess the dynamic process of variable loads through soot blowing. Therefore, new ash and fouling characterization methods are needed to effectively address these issues. This invention applies widely used fault diagnosis and anomaly monitoring methods from industrial equipment to the quantification and monitoring of ash and fouling. It draws an analogy between the gradual ash accumulation trend on various heating surfaces during combustion and the gradual degradation and damage of certain components in industrial equipment. An unsupervised learning method is used to address the lack of labeled data, such as anomaly data, in practical applications for boiler ash and fouling monitoring and modeling research. It simplifies the computation, requiring only a small number of thermal parameters to complete the characterization and monitoring modeling of ash pollution, while ensuring a certain level of accuracy. Attached Figure Description
[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0037] Figure 1 This is a flowchart of the method and system for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data, provided in this embodiment of the invention.
[0038] Figure 2 This invention provides a method and system for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data.
[0039] Figure 3 This is a schematic diagram of the low-pressure high-temperature reheater ash pollution monitoring model provided in an embodiment of the present invention;
[0040] Figure 4 This is a comparison chart of the actual operating data of the AE autoencoder model and the original mechanism model provided in the embodiments of the present invention. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0042] To address the problems existing in the prior art, this invention provides a method and system for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data. The invention will be described in detail below with reference to the accompanying drawings.
[0043] This invention addresses the following problems and deficiencies in the prior art, achieving significant technological advancements:
[0044] Inaccuracy of ash pollution monitoring: Traditional methods for monitoring ash pollution on boiler heating surfaces often rely on indirect observation or experience-based judgment, which lacks accuracy.
[0045] Ignoring dynamic changes: Existing technologies often neglect the dynamic changes in ash and pollution coefficients during boiler operation, resulting in an incomplete assessment of ash and pollution.
[0046] Slow response and weak predictive ability: Existing methods are slow to respond to the occurrence and changes of ash pollution and lack effective predictive ability.
[0047] To address the problems existing in the prior art, the technical solution adopted in this invention is as follows:
[0048] Establish an accurate dynamic simulation model (S1): Based on first principles such as mass conservation and energy conservation, a dynamic simulation model of the boiler including the control system is constructed, which improves the accuracy and reliability of ash and pollution monitoring.
[0049] Full-load, full-condition data acquisition (S2): Collect full-load, full-condition operating data during the cleaning of the heated surface to provide comprehensive baseline data for the model.
[0050] Establishment and training of autoencoder monitoring model (S3): By building an autoencoder monitoring model and training it with actual operating data, the accurate identification and quantification of the degree of ash contamination on the heated surface can be achieved.
[0051] Data acquisition and analysis under changes in ash contamination coefficient (S4, S5): Simulate the ash contamination process of the heated surface, acquire the operating data when the ash contamination coefficient changes, and establish the correlation between reconstruction error and ash contamination coefficient.
[0052] Verification and application of actual operating data (S6): Use actual operating data from power plants to verify and apply the model in practice, thereby improving response speed and predictive ability.
[0053] Two specific application embodiments of the present invention are as follows:
[0054] Example 1: Boiler Efficiency Optimization in Thermal Power Plants
[0055] In thermal power plants, optimizing boiler efficiency is crucial for overall power generation efficiency and environmental impact. The degree of ash fouling on the boiler's heating surfaces directly affects its thermal efficiency.
[0056] Dynamic simulation model establishment (S1): Based on the actual boiler of the thermal power plant, a dynamic simulation model of the boiler including the control system is established using first principles, and verified with actual operating data to ensure the accuracy of the model.
[0057] Monitoring model training (S2, S3): Collect full-load, full-condition operating data during the cleaning of the heated surface to train the autoencoder monitoring model and minimize reconstruction error.
[0058] Ash and fouling simulation and monitoring (S4, S5): Simulate the ash and fouling process of the heating surface, analyze the impact of changes in the ash and fouling coefficient on boiler operation, and use an autoencoder model to monitor the degree of ash and fouling.
[0059] Precise dynamic simulation models can simulate and predict the operating status of boiler heating surfaces under different ash and fouling conditions. Autoencoder models, by analyzing operating data, can accurately monitor the degree of ash and fouling on the heating surfaces, providing timely suggestions for cleaning or maintenance, and optimizing boiler efficiency.
[0060] Example 2: Reduced maintenance and operating costs of industrial boilers
[0061] In industrial production, boilers are key energy equipment, and their operating efficiency and maintenance costs directly affect the economy and stability of the entire production process.
[0062] Boiler simulation model construction (S1): For a specific industrial boiler, establish a dynamic simulation model of the boiler including the control system to ensure that it can accurately reflect the actual operation of the boiler.
[0063] Data-driven model construction (S2, S3): Collect full-load, full-condition data on the cleaning of the heating surfaces during boiler operation to train the autoencoder model.
[0064] Ash pollution monitoring and application (S4, S5, S6): Use simulated ash pollution data to train a model, monitor the ash pollution status in actual operation, and adjust soot blowing actions and maintenance plans based on model feedback.
[0065] By combining dynamic simulation models with data-driven autoencoder models, the ash and fouling status of boiler heating surfaces can be monitored and predicted in real time. This monitoring facilitates timely cleaning and maintenance, thereby reducing operating costs and improving boiler efficiency and production process stability.
[0066] like Figure 1 As shown in the embodiments of the present invention, the method and system for quantitative characterization of boiler heating surface ash fouling that integrates mechanisms and data includes:
[0067] S1: Establish a dynamic simulation model of the boiler, including the control system, based on first principles, and verify it with field data;
[0068] S2: Obtain full-load, full-condition operating data during the cleaning of the heated surface based on a dynamic simulation model;
[0069] S3: Build an autoencoder monitoring model for the boiler heating surface, and train the model using the data obtained in S2 to minimize the reconstruction error;
[0070] S4: Based on the dynamic simulation model, adjust the ash and slag coefficient of the heating surface to simulate the ash and slag formation process of the heating surface, and obtain full load and full working condition operation data when the ash and slag coefficient of the heating surface changes;
[0071] S5: Input the S4 data into the S3 trained autoencoder model, obtain the reconstruction error, and establish the correlation between the model reconstruction error and the gray pollution coefficient;
[0072] S6: Validate and apply the model using actual power plant operating data and soot blowing time.
[0073] The simulation model building process in S1 involves selecting a power plant boiler as the object, establishing an accurate dynamic simulation model of the boiler object that includes the control system based on first principles such as conservation of mass, energy, and momentum, acquiring continuous time series data with a sampling interval of 5-60 seconds from the dynamic simulation platform, and comparing and verifying the simulation model with the main parameters of the unit in actual operation to ensure the consistency of the static and dynamic characteristics of the model with the actual boiler.
[0074] The full-load, full-condition operating data for clean heating surfaces in S2 uses the change in the ash and dirt coefficient of the specific heating surface of the boiler as an indicator, ranging from 0 to 1. A coefficient closer to 0 indicates a higher degree of ash and dirt on the heating surface, while a coefficient closer to 1 indicates a lower degree of ash and dirt and a cleaner surface. Setting the ash and dirt coefficient to 1, meaning no ash and dirt on the heating surface, and using the dynamic simulation model to output full-load, full-condition operating data for clean heating surfaces, is used for subsequent related parameter research and model training.
[0075] In S3, the boiler heating surface self-encoder monitoring model is built, including an encoder and a decoder:
[0076] Encoding part: Encoder E transforms the input feature vector x into a latent feature vector z, z i =E(x) i |θ E Decoding part: Decoder D reconstructs the latent feature vector z into the input feature vector x', x' i =D(z) i |θ D ).
[0077] Where x is the input feature, θ E and θ D These are the parameters for the encoder and decoder, respectively;
[0078] The input features are reconstructed by training an autoencoder, and the parameters of the encoder and decoder are adjusted to minimize the error between the reconstructed result and the input. The reconstruction error is evaluated using MAE (mean absolute error) or MSE (mean squared error). or The training dataset is input into the autoencoder for model learning, feature data is reconstructed, and the reconstruction error is minimized.
[0079] In S3, the selection of input features for the autoencoder model is based on the heat transfer mechanism and parameter trend analysis of the boiler heating surface. According to the quality of field data, variables with high reliability are selected as input features from load, feedwater flow rate, inlet and outlet steam temperature of the heating surface, inlet and outlet flue gas temperature of the heating surface, and heat exchange. However, rapid load fluctuations can make the influence of ash on the heating surface on each parameter not prominent, that is, load factors can interfere with the modeling. In order to reduce the impact of load changes on other thermal parameters, the heat exchange and outlet steam temperature of each load section are calculated based on the stable operation data of each load section. A linear fitting curve of heat exchange and outlet steam temperature with load change is established, and new features of ideal heat exchange and ideal outlet steam temperature are constructed.
[0080] In S4, the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes is used. Similarly to S2, the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes is output by the dynamic simulation model and used as the test set for model learning.
[0081] In S5, full-load, full-condition operating data with varying ash and dirt coefficients of the heating surface are selected as the test set for verification. The reconstruction error is calculated, and the correlation between the reconstruction error and the ash and dirt coefficient is studied to achieve quantitative characterization of ash and dirt on the boiler heating surface.
[0082] In S6, after preprocessing the actual operating data of the power plant's heating surface, since the characteristics of the ash and dirt coefficient cannot be obtained in the actual operation of the power plant, the actual soot blowing action of the power plant can be used to assist in completing the training and test dataset division steps similar to the dynamic simulation model. The model is then learned in the established autoencoder model and compared with the results of the cleanliness / fouling degree characterization parameters calculated based on the traditional mechanism model to verify the effectiveness and accuracy of this method, thus enabling its practical application.
[0083] like Figure 2 As shown in the embodiments of the present invention, the method and system for quantitative characterization of boiler heating surface ash fouling by integrating mechanisms and data include:
[0084] Boiler Dynamic Simulation Model Building Module: Used to build accurate dynamic simulation models of object boilers, including control systems;
[0085] Dataset construction module: used to collect full-load, full-condition operating data with the change of ash and dirt coefficient of the heated surface as an indicator feature;
[0086] Monitoring model building module: used to build an autoencoder monitoring model for the boiler heating surface and reconstruct feature data;
[0087] Ash and dirt trend monitoring module: Used to verify the test set using an autoencoder model, study the correlation between reconstruction error and ash and dirt coefficient, and realize the quantitative characterization of ash and dirt on boiler heating surfaces.
[0088] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0089] An application embodiment of the present invention provides a computer device, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the processor performs the steps of a method for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data.
[0090] An application embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of a method for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data.
[0091] An application embodiment of the present invention provides an information data processing terminal, which is used to realize a boiler heating surface ash and fouling quantitative characterization system that integrates fusion mechanism and data.
[0092] This simulation example uses a 1000MW ultra-supercritical double reheat unit as the simulation object. The boiler operates with double intermediate reheat and ultra-supercritical pressure variation. The high-precision thermal-hydraulic dynamic simulation software Apros was used to perform dynamic simulation modeling of the boiler unit, and the results were compared and verified with the main variables of the unit in actual operation to ensure the accuracy of relevant parameters. After verification, the low-pressure high-temperature reheater, located on the upper part of the horizontal flue, was selected as the research object. Dynamic simulation data of the reheater under full load and full operating conditions was output, with a sampling interval of 1 minute. The change in the ash fouling coefficient was used as the indicator feature, ranging from 0 to 1. A ash fouling coefficient closer to 0 indicates a higher degree of ash fouling on the reheater surface, while a coefficient closer to 1 indicates a lower degree of ash fouling and a cleaner surface. The collected full-load, full-operation simulation data (normal health data) with a clean reheater surface was used as the training set for subsequent models, and the collected simulation data with changes in the ash fouling coefficient of the reheater surface under full load and full operating conditions was used as the test set. The heat exchange rate Q was selected as the thermal parameter. heat Ideal heat exchange Q idealheat Heat exchange deviation Q biasheat Outlet steam temperature T out Ideal outlet steam temperature T idealout Outlet steam temperature deviation T biasout Inlet steam temperature T in The steam flow rate D is used as a feature input to form the original dataset with a dimension of 120000×8.
[0093] An autoencoder model was used for training, the reconstruction error was calculated, and the model was validated on a test set. The reconstruction error and gray contamination coefficient were compared and verified. Figure 3As shown, it is clear that the reconstruction error and the ash contamination coefficient have a basically opposite trend. When the ash contamination coefficient gradually decreases, the reconstruction error (MAE) gradually increases, indicating that the model trained on the cleaning data of the heated surface can reconstruct the cleaning data well. When the ash contamination coefficient begins to change, the reconstruction error will gradually increase as the ash contamination coefficient gradually decreases (the cleanliness of the heated surface decreases), proving that the AE model effectively learns the distribution law of the degree of ash contamination of the heated surface operation data within a certain period.
[0094] In summary, the verification in this embodiment demonstrates that establishing a dynamic simulation model using simulation software is beneficial for providing high-quality datasets and studying the correlation between thermal parameters. Furthermore, building a boiler heating surface ash and fouling monitoring model using an unsupervised learning autoencoder is effective, feasible, and accurate. It eliminates the need for real-time calculation of cleanliness factors, relying solely on a small number of thermal parameters to complete ash and fouling monitoring modeling with a certain level of accuracy. The new feature quantities constructed based on data analysis can effectively avoid the impact of variable load operation on model learning, better reflecting the actual operating conditions of power plant boilers under the current flexible operation context. This enables the monitoring of ash and fouling trends on different heating surfaces, optimizes subsequent soot blowing processes, and provides a new quantitative characterization method and system for ash and fouling monitoring modeling.
[0095] To verify the effectiveness of the AE autoencoder in monitoring ash contamination, actual operating data was selected for validation, and the results were compared with those of a model that calculates ash contamination level / cleanliness factor based on heat transfer mechanism. The results showed that the autoencoder ash contamination quantification method and system, trained using dynamic simulation model data analysis and unsupervised learning algorithm modeling, can be well applied to actual operating data of power plant boilers. The accuracy of this method is superior to the effect of traditional mechanism-based ash contamination level / cleanliness factor judgment on quantifying the ash contamination status of the heating surface. Figure 4 As shown, the cleaning factor parameters characterizing the degree of ash fouling on the heating surface, calculated through the mechanistic model, exhibit significant fluctuations with load changes, resulting in some false positives and affecting actual soot blowing operations. This invention, based on feature selection and construction, can effectively avoid the impact of load variations on the results. In summary, it can be seen that the method and system for quantitative characterizing boiler heating surface ash fouling by integrating mechanistic and data aspects improves the accuracy of boiler heating surface ash fouling monitoring and modeling.
[0096] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.
[0097] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for quantitative characterization of ash and fouling on boiler heating surfaces that integrates mechanisms and data, characterized in that, include: S1: Establish a dynamic simulation model of the boiler, including the control system, based on first principles, and verify it with field data; S2: Obtain full-load, full-condition operating data during the cleaning of the heated surface based on a dynamic simulation model; S3: Build an autoencoder monitoring model for the boiler heating surface, and use the data obtained in S2 to train the model and minimize the reconstruction error; S4: Based on the dynamic simulation model, adjust the ash and slagging coefficient of the heating surface to simulate the ash and slagging process of the heating surface, and obtain full load and full working condition operation data when the ash and slagging coefficient of the heating surface changes; S5: Input the S4 data into the S3 trained autoencoder model, obtain the reconstruction error, and establish the correlation between the model reconstruction error and the gray pollution coefficient; S6: Validate and apply the model using actual power plant operating data and soot blowing time; The full-load, full-condition operating data for cleaning the heating surface in S2 uses the change in the ash and dirt coefficient of the specific heating surface of the boiler as the indicator feature, with a range of 0-1. That is, the closer the ash and dirt coefficient is to 0, the higher the degree of ash and dirt on the heating surface, and the closer it is to 1, the lower the degree of ash and dirt on the heating surface and the cleaner it is. The ash and dirt coefficient is set to 1, meaning that the heating surface is free of ash and dirt. The dynamic simulation model is used to output the full load and full working condition operation data when the heating surface is clean, which will be used for subsequent related parameter research and model training. In S3, the selection of input features for the autoencoder model is based on the heat transfer mechanism and parameter trend analysis of the boiler heating surface. According to the quality of the field data, variables with high reliability are selected as input features from load, feedwater flow rate, steam temperature at the inlet and outlet of the heating surface, flue gas temperature at the inlet and outlet of the heating surface, and heat exchange. However, rapid load fluctuations can obscure the impact of surface dust on various parameters, meaning that load factors can interfere with modeling. To reduce the impact of load changes on other thermal parameters, the heat exchange and outlet steam temperature of each load section are calculated based on stable operating data. Linear fitting curves of heat exchange and outlet steam temperature with load changes are established, and new characteristics of ideal heat exchange and ideal outlet steam temperature are constructed.
2. The method for quantitatively characterizing the amount of ash deposition on the heating surface of a boiler according to claim 1, wherein The simulation model construction in S1 involves selecting a power plant boiler as the object, establishing an accurate dynamic simulation model of the boiler object that includes the control system based on the first principles of mass conservation, energy conservation, and momentum conservation, acquiring continuous time series data with a sampling interval of 5-60 seconds from the dynamic simulation platform, and comparing and verifying the simulation model with the main parameters of the unit in actual operation to ensure the consistency of the static and dynamic characteristics of the model with the actual boiler.
3. The method of claim 1, wherein the fusion mechanism and data of the boiler heating surface fouling amount quantification characterization method is characterized by, In S3, the boiler heating surface self-encoder monitoring model is built, including an encoder and a decoder: Encoding section: Encoder E takes the input feature vector Transform into latent feature vectors , Decoding part: Decoder D will decode the latent feature vectors. Reconstructed into input feature vectors , ; wherein, are input features, and are parameters of the encoder and the decoder, respectively; The input features are reconstructed by training an autoencoder, and the parameters of the encoder and decoder are adjusted to minimize the error between the reconstructed result and the input. The reconstruction error is evaluated using MAE (mean absolute error) or MSE (mean squared error). or The training dataset is input into the autoencoder for model learning, feature data reconstruction is performed, and the reconstruction error is minimized.
4. The method of claim 1, wherein the fusion mechanism and data of the boiler heating surface fouling amount quantification characterization method is characterized by, In S4, the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes is used. Similarly to S2, the full-load, full-condition operating data when the ash and dirt coefficient of the heated surface changes is output by the dynamic simulation model and used as the test set for model learning.
5. The method of claim 1, wherein the fusion mechanism and data of the boiler heating surface fouling amount quantification characterization method is characterized by, In S5, full-load, full-condition operating data with varying ash and dirt coefficients of the heating surface are selected as the test set for verification. The reconstruction error is calculated, and the correlation between the reconstruction error and the ash and dirt coefficient is studied to achieve quantitative characterization of ash and dirt on the boiler heating surface.
6. The method of claim 1, wherein the fusion mechanism and data of the boiler heating surface fouling amount quantification characterization method is characterized by, In S6, after preprocessing the actual operating data of the power plant's heating surface, since the characteristics of the ash and dirt coefficient cannot be obtained in the actual operation of the power plant, the actual soot blowing action of the power plant can be used to assist in completing the training and test dataset division steps similar to the dynamic simulation model. The model is then learned in the established autoencoder model and compared with the results of the cleanliness / fouling degree characterization parameters calculated based on the traditional mechanism model to verify the effectiveness and accuracy of this method, thus enabling its practical application.
7. A system for quantitatively characterizing the fouling of a boiler heating surface by fusion of the mechanism and data of the method of claim 1, characterized by, include: Boiler Dynamic Simulation Model Building Module: Used to build an accurate dynamic simulation model of an object boiler that includes a control system; Dataset construction module: used to collect full-load, full-condition operating data with the change of ash and dirt coefficient of the heated surface as an indicator feature; Monitoring model building module: used to build an autoencoder monitoring model for the boiler heating surface and reconstruct feature data; Ash and dirt trend monitoring module: Used to verify the test set using an autoencoder model, study the correlation between reconstruction error and ash and dirt coefficient, and realize the quantitative characterization of ash and dirt on boiler heating surfaces.
8. A computer device, comprising a memory and a processor, the memory storing a computer program, which, when executed by the processor, causes the processor to perform the steps of the boiler heating surface ash and fouling quantitative characterization method as described in claim 1.