A multi-algorithm fusion furnace coking risk assessment system and method thereof

Abstract

This invention belongs to the field of coal-fired boiler coking prevention technology, and provides a multi-algorithm fusion furnace coking risk assessment system and method. The multi-algorithm fusion furnace coking risk assessment system includes a multi-source heterogeneous data fusion module, a multi-algorithm integrated prediction module, and a dynamic weight adaptive decision module. The multi-source heterogeneous data fusion module processes collected coal quality data, boiler operation data, and heat flux density monitoring data in real time, and integrates coking mechanism feature library data constructed offline by CFD to extract coking-related features. The multi-algorithm integrated prediction module integrates three algorithms: random forest, XGBoost, and support vector regression, predicting coking risk from different perspectives. The dynamic weight adaptive decision module dynamically adjusts the weights based on the historical prediction performance of each algorithm. This invention improves the accuracy and robustness of coking risk assessment through multi-algorithm fusion, enabling precise identification of early coking characteristics and providing a scientific decision-making basis for furnace coking prevention control.

Description

Technical Field

[0001] This invention belongs to the field of anti-coking technology for coal-fired boilers, and particularly relates to a multi-algorithm fusion furnace coking risk assessment system and method. Background Technology

[0002] Furnace coking is a common and thorny problem in the operation of coal-fired power plant boilers, especially in Xinjiang where high-sodium and high-calcium coal is used. Due to its low ash melting temperature and high alkali metal content, coking easily occurs on the water-cooled walls of the furnace during combustion. Coking leads to decreased heat transfer efficiency of heating surfaces, increased flue gas temperature at the furnace outlet, increased flue gas resistance, and may even cause corrosion of heating surfaces, insufficient boiler output, or unplanned shutdowns, seriously affecting the safety, economy, and environmental performance of the unit.

[0003] Accurate furnace coking risk assessment is crucial for achieving "early detection and early treatment" of coking. Currently, furnace coking risk assessment mainly relies on the experience and judgment of operators and single monitoring data, which has the following core shortcomings: The assessment is inaccurate and easily influenced by subjective factors: Traditional coking risk assessment mainly relies on operators' experience-based judgments based on changes in a few parameters such as furnace outlet flue gas temperature and wall temperature, lacking scientific quantitative assessment methods. The assessment results are greatly affected by subjective factors such as the experience level and sense of responsibility of the operators, which can easily lead to misjudgments or omissions, and make it difficult to detect early coking risks in a timely manner. Insufficient data utilization and weak feature extraction capabilities: Existing coking risk assessment methods mostly use a single data source or a few parameters for evaluation, failing to fully utilize multi-source information such as coal quality data, boiler operation data, heat flux density monitoring data, and CFD simulation data. Furthermore, the lack of effective feature extraction methods makes it impossible to mine hidden features closely related to coking from massive amounts of data, resulting in insufficient input information for the assessment model. Single algorithms have significant limitations and poor robustness: Existing coking risk assessment models mostly employ a single machine learning algorithm, such as neural networks or support vector machines. Different machine learning algorithms have different advantages, disadvantages, and applicable scenarios. A single algorithm cannot fully capture the complex nonlinear relationships of coking in the furnace, and may exhibit large prediction errors under certain operating conditions, resulting in poor robustness and generalization ability of the model. Lack of adaptability and inability to adapt to changing operating conditions: Existing coking risk assessment models mostly use offline training methods, and the model parameters remain unchanged once they are determined. However, factors such as coal quality, load, and equipment status constantly change during boiler operation, causing the model's predictive performance to gradually decline. The lack of online updates and adaptive adjustment mechanisms makes it unable to adapt to the dynamic changes in boiler operating characteristics.

[0004] Therefore, a multi-algorithm fusion furnace coking risk assessment system and method are needed to solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-algorithm fusion furnace coking risk assessment system and method to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-algorithm fusion furnace coking risk assessment system, comprising: The multi-source heterogeneous data fusion module is used to collect coal quality data, boiler operation data, and heat flux density monitoring data, fuse CFD offline coking mechanism feature library data, and extract multi-dimensional feature vectors related to furnace coking. The multi-algorithm integrated prediction module communicates with the output of the multi-source heterogeneous data fusion module, and integrates random forest algorithm, XGBoost algorithm and support vector regression algorithm to predict the coking risk value of furnace based on multi-dimensional feature vectors respectively. The dynamic weighted adaptive decision-making module communicates with the output of the multi-algorithm integrated prediction module. It dynamically adjusts the weight coefficients based on the historical prediction performance of each algorithm, performs weighted fusion of the prediction results of each algorithm, and outputs the final furnace coking risk assessment result and risk level.

[0007] Furthermore, the multi-source heterogeneous data fusion module includes a data acquisition unit, a data preprocessing unit, and a feature extraction unit. The data acquisition unit is communicatively connected to the online coal quality monitoring system, the boiler DCS system, and the heat flux density monitoring system, respectively. The CFD numerical simulation system is used to construct an offline coking mechanism feature library, providing mechanistic support for feature extraction. The data preprocessing unit cleans, denoises, normalizes, and performs spatiotemporal registration on the acquired real-time multi-source data, and only performs normalization and operating condition matching on the CFD offline feature library data. The feature extraction unit extracts coking-related statistical features, time-domain features, and frequency-domain features from the preprocessed data. This module can fully utilize multi-source information to comprehensively capture the influencing factors of furnace coking, providing rich input features for risk assessment.

[0008] Furthermore, the coking-related features extracted by the feature extraction unit include coal quality characteristics, combustion condition characteristics, thermal parameter characteristics, and CFD simulation characteristics. Coal quality characteristics include ash fusion temperature, alkali metal content, and ash content. Combustion condition characteristics include excess air coefficient, secondary air ratio, and burner swivel angle. Thermal parameter characteristics include water-cooled wall heat flux density, wall temperature, and furnace outlet flue gas temperature. CFD simulation characteristics include furnace temperature field distribution, wall heat load distribution, and reducing gas concentration distribution. These multi-dimensional features cover all the major influencing factors of furnace coking, effectively improving the prediction accuracy of the evaluation model.

[0009] Furthermore, the multi-algorithm integrated prediction module includes a random forest prediction unit, an XGBoost prediction unit, and a support vector regression prediction unit; the three prediction units work in parallel, each receiving a multi-dimensional feature vector output from the multi-source heterogeneous data fusion module, and independently predicting the furnace coking risk value. By integrating multiple algorithms with different advantages, the complex nonlinear relationship of furnace coking can be captured from different perspectives, overcoming the limitations of a single algorithm.

[0010] Furthermore, the random forest prediction unit obtains the coma risk value by constructing multiple decision trees and averaging their prediction results; the XGBoost prediction unit obtains the coma risk value by gradually fitting the residuals using the gradient boosting method; and the support vector regression prediction unit obtains the coma risk value by constructing the optimal hyperplane in a high-dimensional feature space. Each of the three algorithms has its advantages: random forest has strong anti-overfitting ability, XGBoost has high prediction accuracy, and support vector regression has good generalization ability. Combining the three can achieve complementary advantages.

[0011] Furthermore, the dynamic weighted adaptive decision-making module includes an algorithm performance evaluation unit, a weight calculation unit, and a risk fusion unit. The algorithm performance evaluation unit calculates the prediction accuracy and recall of each algorithm based on the deviation between its historical prediction results and the actual coking state. The weight calculation unit dynamically adjusts the weight coefficients based on the performance indicators of each algorithm. The risk fusion unit performs weighted fusion of the prediction results from each algorithm to obtain the final coking risk assessment result. This dynamic weight adjustment mechanism automatically allocates weights based on the performance of each algorithm under different operating conditions, ensuring the system always maintains optimal prediction performance.

[0012] Furthermore, the weight calculation unit updates the weight coefficients of each algorithm using an exponentially weighted moving average method, assigning higher weights to algorithms with good recent prediction performance and lower weights to algorithms with poor recent prediction performance. The exponentially weighted moving average method can effectively reflect the dynamic changes in algorithm performance, making weight adjustments smoother and more stable.

[0013] Furthermore, it also includes a coking risk early warning module, which communicates with the output of the dynamic weighted adaptive decision-making module. This module outputs corresponding early warning information based on the final coking risk level. The coking risk level is divided into three levels: mild, moderate, and severe, each corresponding to different early warning methods and control strategies. This tiered early warning mechanism can provide differentiated early warning information and control suggestions based on the severity of the coking risk, helping operators to take timely and effective measures.

[0014] A multi-algorithm fusion furnace coking risk assessment method, applied to any of the multi-algorithm fusion furnace coking risk assessment systems described above, includes the following steps: S1. Collect real-time multi-source data, including coal quality data, boiler operation data, and heat flux density monitoring data; call up the CFD offline coking mechanism feature library data for the corresponding operating conditions; S2. Perform preprocessing and spatiotemporal registration on real-time multi-source data, normalize and match CFD feature library data under operating conditions, and generate multi-dimensional feature vectors related to coking in the furnace. S3. Input the multidimensional feature vectors into three prediction models: Random Forest, XGBoost and Support Vector Regression, respectively, to obtain three independent predicted values ​​of coke risk. S4. Based on the historical prediction performance of each algorithm, the weight coefficients are dynamically adjusted, and the three prediction values ​​are weighted and fused to obtain the final coking risk assessment result. S5. Based on the final coking risk assessment results, classify the risk levels and output corresponding early warning information and prevention and control recommendations.

[0015] Furthermore, it also includes the following steps: S6. Online Model Update Steps: Regularly collect data on heat flux density changes, wall temperature fluctuations, and soot blowing effects as indirect representations of the actual coking state. Incremental learning is performed on the three prediction models to update their parameters. Simultaneously, the weighting coefficients are adjusted based on the latest predictive performance of each model to achieve adaptive optimization of the system. This online update mechanism enables the system to adapt to dynamic changes in boiler operating characteristics and maintain long-term stable predictive performance.

[0016] Compared with the prior art, the present invention has the following beneficial effects: High accuracy in assessment, capable of precisely identifying early coking: This invention utilizes multi-source heterogeneous data fusion to fully leverage information from coal quality, operation, monitoring, and simulation, extracting comprehensive coking-related features. Simultaneously, it integrates the advantages of three algorithms—random forest, XGBoost, and support vector regression—to capture the complex nonlinear relationships of furnace coking from different perspectives, significantly improving the accuracy of coking risk assessment and enabling timely detection of potential problems in the early stages of coking. Robust and adaptable to different working conditions: This invention adopts a multi-algorithm integration approach, avoiding the problem of large prediction errors of a single algorithm under certain working conditions. The dynamic weight adaptive decision module can dynamically adjust the weights according to the historical prediction performance of each algorithm, so that the system can maintain optimal prediction performance under different working conditions, and has strong robustness and generalization ability; The data utilization is comprehensive and the feature extraction capability is strong: the multi-source heterogeneous data fusion module of this invention can uniformly process and fuse data of different types and sources, and extract statistical features, time-domain features, and frequency-domain features closely related to coking from massive amounts of data. It makes full use of existing data resources, providing rich input information for risk assessment. Strong adaptability and stable long-term operation: This invention designs an online model update mechanism that can periodically use actual coking state data to incrementally learn the prediction model and update the model parameters. At the same time, the weight coefficients are dynamically adjusted according to the latest prediction performance of each model, enabling the system to adapt to changes in boiler operating characteristics and maintain high prediction accuracy throughout long-term operation; Highly practical and easy to apply in engineering: The system architecture of this invention is clear, with well-defined modules, making it easy to implement and deploy. It can seamlessly interface with existing online coal quality monitoring systems, DCS systems, heat flux density monitoring systems, and CFD numerical simulation systems in power plants without requiring large-scale modifications to existing systems, demonstrating excellent engineering practicality.

[0017] To more clearly illustrate the structural features and effects of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. Attached Figure Description

[0018] Figure 1 This is a block diagram of the overall architecture of the present invention; Figure 2 This is a schematic diagram of the overall process of the present invention.

[0019] In the diagram: 1. Multi-source heterogeneous data fusion module; 11. Data acquisition unit; 12. Data preprocessing unit; 13. Feature extraction unit; 2. Multi-algorithm integrated prediction module; 21. Random forest prediction unit; 22. XGBoost prediction unit; 23. Support vector regression prediction unit; 3. Dynamic weight adaptive decision-making module; 31. Algorithm performance evaluation unit; 32. Weight calculation unit; 33. Risk fusion unit; 4. Online coal quality monitoring system; 5. Boiler DCS system; 6. Heat flux density monitoring system; 7. CFD numerical simulation system; 8. Coking risk early warning module. Detailed Implementation

[0020] The present invention will be further described below with reference to embodiments.

[0021] The following embodiments are used to illustrate the present invention, but should not be used to limit the scope of protection of the present invention. The conditions in the embodiments can be further adjusted according to specific conditions, and simple improvements to the method of the present invention under the premise of the concept of the present invention are all within the scope of protection claimed by the present invention. Example 1

[0022] like Figure 1-2 As shown, this embodiment provides a multi-algorithm fusion furnace coking risk assessment system, including: Multi-source heterogeneous data fusion module 1: The multi-source heterogeneous data fusion module 1 includes a data acquisition unit 11, a data preprocessing unit 12, and a feature extraction unit 13.

[0023] Data acquisition unit 11 is connected to the following systems for communication: Coal quality online detection system 4: Collects industrial analysis data (moisture, ash, volatile matter, fixed carbon), elemental analysis data (C, H, O, N, S), ash fusion characteristics data (DT, ST, FT), and alkali metal content data (Na2O, CaO) of coal entering the furnace.

[0024] Boiler DCS System 5: Collects operating parameters such as boiler load, coal feed rate, secondary air damper opening at each level, excess air coefficient, burner tilt angle, flue gas temperature at furnace outlet, main steam temperature, and main steam pressure.

[0025] Heat flux density monitoring system 6: Collects heat flux density and wall temperature data in different areas of the furnace water-cooled wall, with a total of 30 monitoring points.

[0026] CFD numerical simulation system 7: Offline construction of a coking mechanism feature library covering furnace temperature field distribution, wall heat load distribution, velocity field distribution, CO concentration distribution, O2 concentration distribution, etc., under 20%-100% full load conditions, classified and stored according to load and coal quality, for online system access.

[0027] The data preprocessing unit 12 performs the following processing on the collected multi-source data: 1. Data cleaning: Remove outliers and missing values, and use interpolation to fill in missing data.

[0028] 2. Data Denoising: Wavelet transform is used to denoise real-time time-series data such as heat flux density and wall temperature; the CFD offline feature library data is regular data from simulation output and does not require denoising.

[0029] 3. Data Normalization: The Z-score standardization method is used to convert data of different dimensions to a uniform scale. The formula is as follows: in, The original data, The mean of the data. The standard deviation of the data. This is the normalized data.

[0030] 4. Spatiotemporal registration: Real-time multi-source data with different sampling frequencies are aligned in time and space to ensure that all real-time data are consistent in time and space; CFD offline feature library data is automatically matched with corresponding feature vectors according to the current operating conditions (load, coal quality).

[0031] Feature extraction unit 13 extracts the following four types of coking-related features from the preprocessed data: 1. Coal quality characteristics: ash softening temperature ST, Na2O content in ash, CaO content in ash, ash content Aad, volatile matter content Vad.

[0032] 2. Combustion operating characteristics: boiler load, excess air coefficient, total secondary air volume, secondary air ratio of each layer, burner sway angle, and pulverized coal fineness.

[0033] 3. Thermal parameters: average heat flux density, standard deviation of heat flux density, average wall temperature, standard deviation of wall temperature, flue gas temperature at furnace outlet, and oxygen content at furnace outlet for each monitoring point.

[0034] 4. CFD simulation characteristics: maximum furnace temperature, average temperature, maximum wall heat load, average wall heat load, average CO concentration in the near-wall zone, and average O2 concentration in the near-wall zone.

[0035] Finally, a feature vector containing 50 dimensions is generated, which serves as the input to the multi-algorithm integrated prediction module 2.

[0036] Multi-algorithm integrated prediction module 2: The multi-algorithm integrated prediction module 2 includes a random forest prediction unit 21, an XGBoost prediction unit 22, and a support vector regression prediction unit 23, which work in parallel.

[0037] Random Forest Prediction Unit 21: Random forest is a Bagging-based ensemble learning method that improves prediction accuracy and resistance to overfitting by constructing multiple decision trees and averaging their predictions. This example constructs a random forest model containing 100 decision trees, with a maximum depth of 10 for each tree, and randomly selects 10 features at each split.

[0038] The prediction process of a random forest can be represented as: in, For the number of decision trees, For the first The predicted values ​​of each decision tree. This represents the final prediction value from the random forest.

[0039] XGBoost Prediction Unit 22: XGBoost is an optimized gradient boosting decision tree algorithm that iteratively trains multiple weak learners and sums their predictions to obtain the final prediction. This embodiment constructs an XGBoost model with 50 trees, setting the learning rate to 0.1, the maximum depth to 6, the L1 regularization coefficient to 0.1, and the L2 regularization coefficient to 1.0.

[0040] The objective function of XGBoost is: in, For loss function, For the front The predicted value of the trees, For the first The predicted value of the trees, This is a regularization term used to control model complexity.

[0041] Support Vector Regression Prediction Unit 23: Support Vector Regression (SVR) is an application of Support Vector Machines (SVMs) to regression problems. It constructs an optimal hyperplane in a high-dimensional feature space such that the distance from the training samples to the hyperplane is less than or equal to... This embodiment uses the RBF kernel function, and the kernel function parameters are... Set to 0.1, penalty coefficient Set to 10, Set to 0.01.

[0042] The prediction function for SVR is: in, and For Lagrange multipliers, For kernel function, This is a bias term.

[0043] The three prediction units each receive a 50-dimensional feature vector output from the multi-source heterogeneous data fusion module 1, and independently predict the coking risk value in the furnace, thus obtaining... , and Three prediction results.

[0044] Dynamic weighted adaptive decision module 3: The dynamic weight adaptive decision module 3 includes an algorithm performance evaluation unit 31, a weight calculation unit 32, and a risk fusion unit 33.

[0045] The algorithm performance evaluation unit 31 periodically (e.g., once a day) compares the prediction results of each algorithm with the actual coking state, and calculates the prediction accuracy and mean square error (MSE) of each algorithm. The actual coking state is determined by the changing trend of furnace heat flux density monitoring data and the results of periodic furnace inspections.

[0046] Weight calculation unit 32 updates the weight coefficients of each algorithm using the Exponentially Weighted Moving Average (EWMA) method. First, the performance score of each algorithm is calculated: in, For the first The algorithm at time... The mean square error, The attenuation coefficient is set to 0.5 in this embodiment.

[0047] Then, the performance scores are normalized to obtain the weight coefficients for each algorithm: in, , , The weights of Random Forest, XGBoost, and Support Vector Regression are respectively, satisfying... .

[0048] Risk fusion unit 33 performs weighted fusion of the prediction results from the three algorithms to obtain the final coking risk assessment result: Final coking risk value The value ranges from 0 to 1, with a higher value indicating a higher risk of coking.

[0049] Coking Risk Warning Module 8: The coking risk warning module 8 classifies the risk level into three levels based on the final coking risk value: Mild risk: A blue alert is issued to remind operators to pay attention to the coking trend.

[0050] Moderate risk: A yellow alert has been issued, and it is recommended to activate the acoustic soot blowing system.

[0051] Severe risk: A red alert has been issued, and it is recommended to take immediate and targeted anti-coking measures, such as adjusting combustion parameters and starting steam soot blowing.

[0052] Warning information is displayed via audible and visual alarms in the central control room and computer pop-ups, and is also recorded in the system log. Example 2

[0053] Based on Example 1, this embodiment further optimizes the feature extraction unit 13 and adds a feature selection function based on SHAP values ​​to improve the prediction efficiency and interpretability of the model.

[0054] SHAP (SHapley Additive Explanations) is a game theory-based model interpretation method that quantifies the contribution of each feature to the model's prediction results. In this embodiment, after the model training is completed, the SHAP value of each feature is calculated, the features are ranked according to the size of the SHAP value, and the top 30 features that contribute the most to the prediction of coking risk are selected as the final input features.

[0055] Feature selection removes redundant and unimportant features, reducing the model's input dimensionality and improving prediction speed. Meanwhile, SHAP value analysis reveals the degree and direction of each feature's impact on coking risk, helping operators understand the main causes of coking and providing guidance for combustion adjustments.

[0056] For example, SHAP value analysis revealed that the ash softening temperature (ST), Na₂O content in the ash, excess air coefficient, and CO concentration in the near-wall region are the four most important characteristics affecting the risk of coking in the furnace. Among them, the lower the ash softening temperature (ST), the higher the Na₂O content in the ash, the smaller the excess air coefficient, and the higher the CO concentration in the near-wall region, the greater the risk of coking. Example 3

[0057] This embodiment, based on embodiment 2, further adds an online model update mechanism to adapt to the dynamic changes in boiler operating characteristics.

[0058] The online model update employs an incremental learning strategy, with the following specific steps: 1. Data Acquisition and Labeling: Boiler operation data, heat flux density monitoring data, and soot blowing feedback data are collected weekly. The actual coking status is determined indirectly through the following methods: the trend of heat flux density monitoring data (if the heat flux density in a certain area continues to decrease and then rebounds significantly after soot blowing, it indicates that coking exists in that area); and the model is periodically calibrated in conjunction with the results of annual boiler shutdown maintenance or furnace endoscope inspection.

[0059] 2. Dataset update: Add newly collected data to the training dataset, while removing older data from earlier times, keeping the size of the training dataset unchanged (e.g., maintaining 2000 samples).

[0060] 3. Incremental Model Training: The Random Forest, XGBoost, and Support Vector Regression models are incrementally trained using the updated training dataset. During incremental training, only some parameters of the model are adjusted, while retaining the previously learned knowledge.

[0061] 4. Weight Update: Based on the prediction performance of each model on the new dataset, recalculate the weight coefficients and update the dynamic weight adaptive decision module 3.

[0062] 5. Model Evaluation and Replacement: Evaluate the performance of the updated model on the validation set. If the updated model outperforms the original model, replace the original model; otherwise, retain the original model for continued use.

[0063] Through the online model update mechanism, the system can continuously learn new operating data, adapt to changes in boiler operating characteristics, and maintain long-term stable predictive performance.

[0064] The multi-algorithm fusion furnace coking risk assessment system of this invention is based on the concepts of multi-source data fusion and ensemble learning. By integrating the advantages of multiple data sources and algorithms, it achieves accurate assessment of furnace coking risk. Its core working principle is as follows: Use Case 1: Daily Coking Monitoring During daily boiler operation, the system collects coal quality data, boiler operation data, and heat flux density monitoring data in real time. Combined with CFD simulation data, it extracts multi-dimensional feature vectors related to coking. The three models in the multi-algorithm integrated prediction module 2 work in parallel to predict coking risk values. The dynamic weight adaptive decision module 3 dynamically adjusts the weights based on the historical prediction performance of each algorithm, outputting the final coking risk assessment result and risk level. When the coking risk reaches moderate or severe, the system automatically issues an early warning and provides corresponding prevention and control suggestions, such as adjusting the secondary air ratio and initiating soot blowing. Operators can take timely measures based on the system's prompts to prevent the occurrence and development of coking.

[0065] Use Case 2: Optimization of Coal Blending Scheme When blending coal for combustion, the system can predict the corresponding coking risk based on different blending schemes. By inputting the coal quality characteristics under different blending ratios into the system, it can quickly predict the coking risk level for each blending scheme. Operators can then select the optimal blending scheme based on the prediction results, maximizing the utilization of high-alkali coal resources and reducing power generation costs while ensuring safe boiler operation.

[0066] This invention fully utilizes various information resources through multi-source heterogeneous data fusion, achieves complementary advantages through multi-algorithm integration, ensures optimal system performance under different operating conditions through dynamic weight adaptive decision-making, and realizes adaptive optimization of the system through an online update mechanism. This system can accurately and reliably assess the risk of coking in the furnace, providing a scientific basis for furnace anti-coking control, effectively reducing the risk of unplanned boiler shutdowns, and improving the safety and economy of unit operation.

[0067] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-algorithm fusion furnace coking risk assessment system, characterized in that, include: The multi-source heterogeneous data fusion module (1) is used to collect coal quality data, boiler operation data, heat flux density monitoring data, fuse CFD offline coking mechanism feature library data, and extract multi-dimensional feature vectors related to furnace coking. The multi-algorithm integrated prediction module (2) is connected to the output end of the multi-source heterogeneous data fusion module (1) and integrates the random forest algorithm, XGBoost algorithm and support vector regression algorithm to predict the coking risk value of the furnace based on multi-dimensional feature vectors respectively. The dynamic weight adaptive decision module (3) communicates with the output of the multi-algorithm integrated prediction module (2). It dynamically adjusts the weight coefficients according to the historical prediction performance of each algorithm, performs weighted fusion of the prediction results of each algorithm, and outputs the final furnace coking risk assessment result and risk level.

2. The multi-algorithm fusion furnace coking risk assessment system according to claim 1, characterized in that, The multi-source heterogeneous data fusion module (1) includes a data acquisition unit (11), a data preprocessing unit (12), and a feature extraction unit (13). The data acquisition unit (11) is connected to the coal quality online detection system (4), the boiler DCS system (5), and the heat flux density monitoring system (6), respectively. The CFD numerical simulation system (7) is used to construct a coking mechanism feature library offline, providing mechanistic support for feature extraction. The data preprocessing unit (12) cleans, denoises, normalizes, and performs spatiotemporal registration processing on the acquired real-time multi-source data, and only performs normalization and operating condition matching processing on the CFD offline feature library data. The feature extraction unit (13) extracts coking-related statistical features, time-domain features, and frequency-domain features from the preprocessed data.

3. The multi-algorithm fusion furnace coking risk assessment system according to claim 2, characterized in that, The coking-related features extracted by the feature extraction unit (13) include coal quality features, combustion condition features, thermal parameter features and CFD simulation features; coal quality features include ash fusion temperature, alkali metal content and ash content; combustion condition features include excess air coefficient, secondary air ratio and burner sway angle; thermal parameter features include water-cooled wall heat flux density, wall temperature and furnace outlet flue gas temperature; CFD simulation features include furnace temperature field distribution, wall heat load distribution and reducing gas concentration distribution.

4. The multi-algorithm fusion furnace coking risk assessment system according to claim 1, characterized in that, The multi-algorithm integrated prediction module (2) includes a random forest prediction unit (21), an XGBoost prediction unit (22), and a support vector regression prediction unit (23). The three prediction units work in parallel and receive the multi-dimensional feature vectors output by the multi-source heterogeneous data fusion module (1) respectively, and independently predict the coking risk value of the furnace.

5. The multi-algorithm fusion furnace coking risk assessment system according to claim 4, characterized in that, The random forest prediction unit (21) obtains the coking risk value by constructing multiple decision trees and averaging their prediction results; the XGBoost prediction unit (22) obtains the coking risk value by gradually fitting the residuals through the gradient boosting method; and the support vector regression prediction unit (23) obtains the coking risk value by constructing the optimal hyperplane in the high-dimensional feature space.

6. The multi-algorithm fusion furnace coking risk assessment system according to claim 1, characterized in that, The dynamic weight adaptive decision module (3) includes an algorithm performance evaluation unit (31), a weight calculation unit (32), and a risk fusion unit (33); the algorithm performance evaluation unit (31) calculates the prediction accuracy and recall of each algorithm based on the deviation between the historical prediction results of each algorithm and the actual coking state; the weight calculation unit (32) dynamically adjusts the weight coefficients based on the performance indicators of each algorithm. The risk fusion unit (33) performs weighted fusion of the prediction results of each algorithm to obtain the final coking risk assessment result.

7. The multi-algorithm fusion furnace coking risk assessment system according to claim 6, characterized in that, The weight calculation unit (32) updates the weight coefficients of each algorithm using the exponential weighted moving average method, assigning higher weights to algorithms with good recent prediction performance and lower weights to algorithms with poor recent prediction performance.

8. The multi-algorithm fusion furnace coking risk assessment system according to claim 1, characterized in that, It also includes a coking risk early warning module (8), which is connected to the output end of the dynamic weight adaptive decision module (3) and outputs the corresponding early warning information according to the final coking risk level; the coking risk level is divided into three levels: mild, moderate and severe, which correspond to different early warning methods and prevention and control strategies.

9. A multi-algorithm fusion furnace coking risk assessment method, applied to the multi-algorithm fusion furnace coking risk assessment system according to any one of claims 1-8, characterized in that, Includes the following steps: S1. Collect real-time multi-source data, including coal quality data, boiler operation data, and heat flux density monitoring data; Call the CFD offline coking mechanism feature library data for the corresponding working condition; S2. Perform preprocessing and spatiotemporal registration on real-time multi-source data, normalize and match CFD feature library data under operating conditions, and generate multi-dimensional feature vectors related to coking in the furnace. S3. Input the multidimensional feature vectors into three prediction models: Random Forest, XGBoost and Support Vector Regression, respectively, to obtain three independent predicted values ​​of coke risk. S4. Based on the historical prediction performance of each algorithm, the weight coefficients are dynamically adjusted, and the three prediction values ​​are weighted and fused to obtain the final coking risk assessment result. S5. Based on the final coking risk assessment results, classify the risk levels and output corresponding early warning information and prevention and control recommendations.

10. The multi-algorithm fusion furnace coking risk assessment method according to claim 9, characterized in that, It also includes the following steps: S6. Online Model Update Steps: Periodically collect data on heat flux density change trends, wall temperature fluctuation characteristics, and soot blowing effect feedback as indirect representations of the actual coking state. Perform incremental learning on the three prediction models and update the model parameters. At the same time, adjust the weight coefficients according to the latest prediction performance of each model to achieve adaptive optimization of the system.

Images