Dry quenching boiler tube explosion intelligent prediction and management method based on multi-parameter cooperation

Abstract

The invention relates to the technical field of intelligent operation and maintenance and optimization control of a dry quenching boiler, and discloses an intelligent prediction and management method for tube explosion of a dry quenching boiler based on multi-parameter collaboration. According to the method, priority abnormal parameter characteristics of different pipe explosion types are found for the first time, header light pipe explosion takes boiler inlet and outlet gas temperature, boiler pressure and air conduction flow as priority abnormal parameters, and economizer pipe explosion takes hydrogen concentration, air introduction amount and boiler feed pump temperature as priority abnormal parameters; and a double-layer feature system of'priority abnormal parameter-associated collaborative parameter 'is constructed, and precise identification and early warning of pipe explosion types are realized in combination with a machine learning algorithm. And a decision basis is provided for optimized operation and safety guarantee of a coke dry quenching boiler control system. The problems that in the prior art, early warning blindness is high due to the fact that early warning only depends on single parameter early warning, pipe explosion types cannot be distinguished, early recognition precision is low, and therefore accurate and differentiated emergency disposal and maintenance decision bases cannot be provided for field personnel are solved.

Description

Technical Field

[0001] This application relates to the field of intelligent equipment operation and maintenance and industrial decision-making technology, and in particular to an intelligent prediction and management method for tube rupture in dry quenching coke boilers based on multi-parameter collaboration. Background Technology

[0002] As a core thermal energy equipment in the coking industry, the safe and stable operation of dry quenching coke boilers directly determines the continuity of production and personnel safety. Under high temperature and high pressure conditions, dry quenching coke boilers are prone to tube rupture accidents due to factors such as material fatigue, corrosion, and overheating. Among these, rupture of header tubes and economizer tubes are two common types of failures. Header tubes, being in direct contact with high-temperature flue gas, are susceptible to sudden temperature changes and pressure fluctuations; economizers, on the other hand, are at high risk of corrosion and tube rupture due to long-term contact with feedwater containing corrosive media.

[0003] Existing pipe burst monitoring technologies have several shortcomings: First, they rely heavily on threshold judgments for single parameters (such as temperature and pressure), failing to identify differences in "priority abnormal parameters" for different types of pipe bursts, leading to delayed early warnings. Alarms are typically triggered only when a pipe burst is imminent (e.g., 10-30 minutes before), leaving insufficient time for response. Second, they cannot distinguish between pipe burst types, only providing a general assessment of "pipe burst risk," requiring on-site personnel to blindly and urgently inspect all areas, which is both time-consuming and increases the risk of working in high-temperature environments. Third, the fixed baseline is susceptible to equipment aging (annual baseline drift rate of approximately 5%) and load fluctuations (60%-110% of design load), resulting in a false alarm rate as high as 10%-15%, severely disrupting normal production operations.

[0004] Furthermore, Chinese patent publication CN106802646A, with international classification G05B23 / 02, discloses a boiler tube rupture fault early warning method based on a decision tree system. However, it only provides simple alarm signals, and its output information (such as alarm type and risk level) lacks structured and interpretable decision support, resulting in an "information gap" between the early warning and subsequent operation and maintenance processes (such as inspection, control, and repair). This makes it difficult for on-site personnel to obtain accurate and differentiated handling guidance in a timely manner, and the efficiency and quality of the early warning response heavily rely on personal experience, failing to form a complete management closed loop from "risk perception" to "scientific decision-making" and then to "effective handling."

[0005] Therefore, there is an urgent need for an intelligent predictive management method for dry quenching coke boiler tube rupture that can accurately identify the priority abnormal parameters of different tube rupture types, achieve early warning, and provide accurate decision-making basis for the optimized control and intelligent operation and maintenance of the production system, so as to fundamentally solve the pain points of existing technologies.

[0006] The information in this background section is intended only to enhance the understanding of the overall background of this application and is not necessarily to be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] To address the aforementioned issues, this application aims to provide a multi-parameter collaborative intelligent prediction and management method for tube rupture in dry quenching boilers. By first discovering the "priority abnormal parameter characteristics" of tube rupture in header tubes and economizer tubes, a multi-parameter collaborative prediction system is constructed to achieve accurate identification of rupture types, early warning, and output of structured decision information, providing a clear decision-making basis for the optimized control and intelligent operation and maintenance of dry quenching boilers.

[0008] To achieve the above objectives, this application adopts the following technical solution: A method for intelligent prediction and management of tube rupture in dry quenching boilers based on multi-parameter synergy includes the following steps: S1. Construct a pipeline burst type-priority anomaly parameter mapping library: The types of tube ruptures include at least header tube ruptures and economizer tube ruptures. The priority abnormal parameters are the core parameters that first show abnormal fluctuations before the tube rupture occurs. Among them, the priority abnormal parameters for header tube ruptures include boiler outlet gas temperature, boiler inlet gas temperature, boiler pressure, and air conduction flow rate; the priority abnormal parameters for economizer tube ruptures include hydrogen concentration, air introduction rate, and boiler feedwater pump temperature. S2. Collection of multi-dimensional operating parameters: The sensor network of the dry quenching coke boiler collects multi-dimensional operating parameters in real time. The multi-dimensional operating parameters include at least the priority anomaly parameters and the associated coordination parameters. The associated coordination parameters are auxiliary parameters that are strongly correlated with the priority anomaly parameters, i.e., the correlation coefficient |R|≥0.7. S3. Preprocess the collected multi-dimensional operating parameters: The preprocessing includes distinguishing between short-term and long-term missing data and filling them with linear interpolation and sensor offline warning respectively, filtering out transient interference anomalies with a duration of less than 1 minute, and deleting data with a single timestamp null value ratio greater than 50%; where short-term missing data is defined as missing data with a sampling period of Δt and a duration of less than or equal to 5Δt, and long-term missing data is defined as missing data with a duration of more than 5Δt. S4. Priority Anomaly Parameter Identification: Based on the burst pipe type-priority anomaly parameter mapping library, priority anomaly parameters are identified for the preprocessed multi-dimensional operating parameters to determine whether at least one priority anomaly parameter exceeds the dynamic baseline range. S5. Determine if there is any priority abnormal data that exceeds the dynamic baseline range: If there is a priority abnormal parameter that exceeds the dynamic baseline range, extract the collaborative features of the priority abnormal parameter and the corresponding associated collaborative parameter. The collaborative features include the parameter change rate, the duration of the abnormality, and the degree of lag correlation between parameters. S6. Input the collaborative features into the pre-trained pipe burst prediction model and output the pipe burst type and predicted pipe burst time; S7. Risk Classification and Early Warning Processing: Based on the output results, risk classification is performed, a structured risk decision output is generated, an early warning is triggered according to the risk level, and the decision output is released. S8. The generated structured risk decision output is sent to the central control platform of the dry quenching coke production system as an auxiliary decision-making basis for the system to optimize and adjust operating parameters or perform safety interlocks.

[0009] In some embodiments of this application, the method for constructing the burst pipe type-priority anomaly parameter mapping library in step S1 includes: A1. Collect historical tube rupture case data from multiple sets of dry quenching coke boilers: The case data shall include at least the full-cycle parameter change records of header tube rupture and economizer tube rupture; A2. Time Series Backtracking Analysis: The first abnormal parameter before the occurrence of each type of tube rupture was determined as the priority abnormal parameter. The priority abnormal parameters for header tube rupture are: boiler outlet gas temperature fluctuation > ±5℃ / 10min, boiler inlet gas temperature fluctuation > ±3℃ / 10min, boiler pressure fluctuation > ±0.2MPa / 10min, and air conduction fluctuation > ±8m. The priority abnormal parameters for economizer tube rupture are: hydrogen concentration gradient change ≥0.02% / min, air introduction fluctuation > ±5m, and boiler feedwater pump temperature fluctuation > ±4℃ / 10min. 3 / h / 10min 3 / h / 10min A3. Constructing a mapping relationship between different burst pipe types and priority anomaly parameters: Constructing a one-to-one mapping relationship between different burst pipe types and priority anomaly parameters to form a burst pipe type-priority anomaly parameter mapping library; where the formula for calculating the change in hydrogen concentration gradient is ΔH2 / Δt=t2-t1H2(t2)-H2(t1), where ΔH2 / Δt is the change in hydrogen concentration gradient, in units of % / min; H2(t1) is the hydrogen concentration value at time t1; H2(t2) is the hydrogen concentration value at time t2; t2-t1 is the time interval.

[0010] In some embodiments of this application, in step A1, 120 sets of historical tube rupture case data of dry quenching coke boilers are collected; in step A3, t2-t1 is a time interval, which can be 1 minute.

[0011] In some embodiments of this application, the dynamic baseline range in step S4 is constructed using an adaptive baseline model, which includes: B1. The Long Short-Term Memory (LSTM) network was used to learn the normal operation data of the dry quenching coke boiler over the past 365 days to initially construct the baseline range of each parameter; B2. The deviation rate between the current operating parameters and the existing baseline is automatically evaluated every 24 hours. If the deviation rate is ≥5% or the boiler load change is >20%, the baseline is dynamically updated. B3. When dynamically updating the baseline, a gradient compensation algorithm is used to correct the upper and lower limits of the baseline to ensure that the baseline range is adapted to the current equipment status and operating conditions.

[0012] In some embodiments of this application, in step B3, the parameter deviation rate is calculated using the formula R. dev =P base |P curr -P base |×100%, where: R dev P is the parameter deviation rate. curr This is the current running parameter value; P base The baseline parameter values ​​are given; the gradient compensation correction formula is: P base-new =P base-old +k×t, where P base-new The updated baseline value; P base-old t represents the baseline value before the update; k is the gradient compensation coefficient; t is the number of days the baseline has been running.

[0013] In some embodiments of this application, in step B3, k is the gradient compensation coefficient, and the boiler outlet gas temperature gradient compensation coefficient is 0.0055℃ / day.

[0014] In some embodiments of this application, in step S5, the extraction rules for collaborative features include: for a burst header tube: the ratio of the rate of change of boiler outlet gas temperature to the rate of change of boiler pressure is R. T-P =V P V T V T =ΔtΔT is the rate of change of boiler outlet gas temperature, in °C / min, V P =ΔtΔP is the rate of change of boiler pressure, in MPa / min, R T-P The normal range is 0.8-1.2; The lag correlation between air conductance and boiler inlet gas temperature was calculated using the Pearson correlation coefficient, with a normal range of 0.6-0.9; where Xi is the air conductance value, X is the mean air conductance, Yi is the boiler inlet gas temperature at lag τ, Y is the mean temperature, and n is the number of sampling points.

[0015] In some embodiments of this application, τ is taken as 5 min.

[0016] In some embodiments of this application, the extraction rules for the cooperative features in step S5 include: for an economizer tube rupture, the cooperative change rate R of hydrogen concentration and air introduction amount. H-A =ΔA / ΔtΔH2 / Δt, the normal range of value is 0.5-0.8, where the rate of change of air intake is ΔA / Δt=t2-t1A(t2)-A(t1), the unit is m・min, and A(t1) and A(t2) are the air intake values ​​at time t1 and t2 respectively; 3 / h The lag correlation between boiler feedwater pump temperature and boiler feedwater flow rate is calculated using the Pearson correlation coefficient. The formula is the same as the lag correlation between air conductance and boiler inlet gas temperature, with a normal range of 0.7-0.95. Where Xi corresponds to the boiler feedwater pump temperature value, and Yi corresponds to the boiler feedwater flow rate value at the lag time τ, where τ is taken as 3 min.

[0017] In some embodiments of this application, in step S6, the pre-trained burst pipe prediction model is constructed using Gradient Boosting Decision Tree (GBDT), and the model training process includes: C1. Using the historical pipe burst case data and corresponding collaborative features collected in step A1 as training samples, the pipe burst type and pipe burst time are labeled as tags; C2. Bayesian optimization is used to determine the model hyperparameters, including the learning rate, maximum depth of the decision tree, and number of base evaluators; C3. The model is trained using 5-fold cross-validation. Training is complete when the model's accuracy in identifying burst pipe types is ≥96% and the burst pipe time prediction error is ≤15%. The formula for calculating the burst pipe time prediction error is: δ t =t true |t pred -t true |×100%, where: δ t The error in predicting the tube burst time; t pred The model predicts the pipe burst time, in hours (h); t true The actual burst time is expressed in hours. The formula for calculating the burst type identification accuracy is: Acc = TP + TN + FP + FNTP + TN × 100%, where: Acc is the type identification accuracy; TP is the number of true positive samples, i.e., correctly identifying burst tubes in the header or economizer; TN is the number of true negative samples, i.e., correctly identifying tubes that are not burst; FP is the number of false positive samples, i.e., misjudging tubes that are not burst; FN is the number of false negative samples, i.e., tubes that are not identified.

[0018] In some embodiments of this application, in step C1, the learning rate adopts a Gaussian distribution with a mean of 0.01 and a standard deviation of 0.01; the maximum depth of the decision tree is 3-10; and the number of base estimators follows a geometric distribution with a mean of 10.

[0019] In some embodiments of this application, in step S7, the following is determined based on the output results: when the predicted burst time is less than 2 hours, it is determined to be a level 1 risk and an immediate warning is triggered; when 2 hours < predicted burst time < 12 hours, it is determined to be a level 2 risk and a warning is triggered; when the predicted burst time is greater than 12 hours, it is determined to be a level 3 risk and only further detection is required.

[0020] In some embodiments of this application, in step S7, after the early warning is triggered, differentiated handling is carried out, specifically: when the header tube rupture is a level one risk, the load is reduced to 80% of the design load, the monitoring of boiler inlet and outlet gas temperature and pressure is strengthened, and parameter changes are recorded every 30 minutes; when the header tube rupture is a level two risk, the boiler and induced draft fan are interlocked and shut down, the header inlet and outlet valves are closed, and the pressure relief device is started. When the economizer tube rupture is classified as a Level 1 risk, reduce the air intake to 90% of the normal range, check the operating status of the boiler feedwater pump, and record the hydrogen concentration change every 20 minutes. When the economizer tube rupture is classified as a Level 2 risk, reduce the load to 50% of the design load, switch to the standby feedwater pump, and isolate and repair the economizer area.

[0021] In some embodiments of this application, in step S8, the generated structured risk decision output is sent to the dry quenching coke production system control platform through a standard industrial communication interface.

[0022] Compared with the prior art, this application has at least the following advantages: 1. Outstanding early warning capability: By discovering the priority abnormal parameters of different pipe burst types and combining them with quantitative formulas, early anomalies can be accurately captured, with an early warning lead time of 1.5-3 hours, which is 120%-360% higher than the existing technology (10-30 minutes), allowing sufficient time for on-site disposal; 2. Accurate identification of burst tube type: Based on priority anomaly parameters and collaborative feature quantitative analysis, the accuracy rate of burst tube type identification in header tubes reaches 97.2%, and the accuracy rate of burst tube type identification in economizers reaches 96.1%, avoiding blind investigation; 3. Strong adaptability to dynamic operating conditions: The adaptive dynamic baseline combined with the gradient compensation algorithm can adapt to equipment aging and load fluctuations in real time, and the false alarm rate is controlled within 3%, which is far lower than the industry average of 10%-15%. 4. Closed-loop full lifecycle management: Construct a four-dimensional status tracking system for pipe burst events, combined with quantitative indicators such as recovery rate and accuracy, to achieve full-process traceability from early warning and handling to archiving (and support pushing decision information to the upper-level management platform). After model iteration, the classification accuracy rate improves by 2%-3% every quarter. 5. Economical deployment cost: Based on the existing sensor network, no new dedicated equipment is required, and the deployment cost per set is reduced by more than 60% compared with existing technologies, making it suitable for the needs of coking enterprises of different sizes.

[0023] In summary, this application, by first discovering priority anomaly parameters for different tube rupture types and constructing a multi-parameter collaborative prediction system using multi-dimensional quantitative formulas, has formed a tube rupture early warning method and system integrating intelligent prediction, decision support, and industrial data services. This method and system can achieve early and accurate early warning and full lifecycle management of tube rupture in dry-quenching coke boilers, possessing extremely high industrial application value and market promotion prospects. Attached Figure Description

[0024] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an undue limitation of this application.

[0025] Figure 1 This is a schematic diagram of the overall process of the intelligent prediction and management method for tube rupture of dry quenching coke boiler based on multi-parameter collaboration in some embodiments of this application; Figure 2 This is a schematic diagram illustrating the construction process of the pipe burst type-priority anomaly parameter mapping library in some embodiments of this application; Figure 3 This is a schematic diagram illustrating the construction process of the adaptive baseline model in some embodiments of this application; Figure 4 This is a schematic diagram illustrating the training process of the pipe burst prediction model in some embodiments of this application; Figure 5 This is a schematic diagram of the main interface of the central control platform in some embodiments of this application; Figure 6 This is a schematic diagram illustrating the intelligent decision-making, alarm control, management, and distribution operation of the central control platform in some embodiments of this application. Detailed Implementation

[0026] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings, but these are not intended to limit the scope of this application. To enable those skilled in the art to better understand the technical solutions of this disclosure, the present disclosure will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments of this disclosure will be further described in detail below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of this disclosure.

[0027] All terms used in this disclosure (including technical or scientific terms) have the same meaning as understood by one of ordinary skill in the art to which this disclosure pertains, unless otherwise specifically defined. It should also be understood that terms defined in a general dictionary, such as a dictionary, should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and not as having an idealized or highly formalized meaning, unless expressly defined herein.

[0028] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0029] The following is in conjunction with the appendix Figure 1-6 The scheme of this application will be further explained.

[0030] The purpose of this application is to provide an intelligent prediction and management method for tube rupture in dry quenching coke boilers based on multi-parameter collaboration. By discovering for the first time the "priority abnormal parameter characteristics" of tube rupture in header tubes and economizer tubes, a multi-parameter collaborative prediction system is constructed to achieve accurate identification of tube rupture types, early warning and scientific handling.

[0031] In some embodiments of this application, a method for intelligent prediction and management of tube rupture in dry quenching boilers based on multi-parameter collaboration includes the following steps: S1. Construct a pipeline burst type-priority anomaly parameter mapping library: The types of tube ruptures include at least header tube ruptures and economizer tube ruptures. The priority abnormal parameters are the core parameters that first show abnormal fluctuations before the tube rupture occurs. Among them, the priority abnormal parameters for header tube ruptures include boiler outlet gas temperature, boiler inlet gas temperature, boiler pressure, and air conduction flow rate; the priority abnormal parameters for economizer tube ruptures include hydrogen concentration, air introduction rate, and boiler feedwater pump temperature. S2. Collection of multi-dimensional operating parameters: The sensor network of the dry quenching coke boiler collects multi-dimensional operating parameters in real time. The multi-dimensional operating parameters include at least the priority anomaly parameters and the associated coordination parameters. The associated coordination parameters are auxiliary parameters that are strongly correlated with the priority anomaly parameters, i.e., the correlation coefficient |R|≥0.7. S3. Preprocess the collected multi-dimensional operating parameters: The preprocessing includes distinguishing between short-term and long-term missing data and filling them with linear interpolation and sensor offline warning respectively, filtering out transient interference anomalies with a duration of less than 1 minute, and deleting data with a single timestamp null value ratio greater than 50%; where short-term missing data is defined as missing data with a sampling period of Δt and a duration of less than or equal to 5Δt, and long-term missing data is defined as missing data with a duration of more than 5Δt. S4. Priority Anomaly Parameter Identification: Based on the burst pipe type-priority anomaly parameter mapping library, priority anomaly parameters are identified for the preprocessed multi-dimensional operating parameters to determine whether at least one priority anomaly parameter exceeds the dynamic baseline range. S5. Determine if there is any priority abnormal data that exceeds the dynamic baseline range: If there is a priority abnormal parameter that exceeds the dynamic baseline range, extract the collaborative features of the priority abnormal parameter and the corresponding associated collaborative parameter. The collaborative features include the parameter change rate, the duration of the abnormality, and the degree of lag correlation between parameters. S6. Input the collaborative features into the pre-trained pipe burst prediction model and output the pipe burst risk level, pipe burst type and predicted pipe burst time; S7. Risk Classification and Early Warning Handling: Based on the output results, risk is classified, a structured risk decision output is generated, an early warning is triggered according to the risk level, and the decision output is released. S8. The generated structured risk decision output is sent to the central control platform of the dry quenching coke production system as an auxiliary decision-making basis for the system to optimize and adjust operating parameters or perform safety interlocks.

[0032] In some embodiments of this application, the method for constructing the burst pipe type-priority anomaly parameter mapping library in step S1 includes: A1. Collect historical tube rupture case data from multiple sets of dry quenching coke boilers: The case data shall include at least the full-cycle parameter change records of header tube rupture and economizer tube rupture; A2. Time Series Backtracking Analysis: Time series backtracking analysis was performed on the data of each case to determine the first abnormal parameter before the occurrence of each type of tube rupture as the priority abnormal parameter. The priority abnormal parameters for header tube rupture are as follows: boiler outlet gas temperature fluctuation > ±5℃ / 10min, boiler inlet gas temperature fluctuation > ±3℃ / 10min, boiler pressure fluctuation > ±0.2MPa / 10min, and air conduction flow fluctuation > ±8m. The priority abnormal parameters for economizer tube rupture are as follows: hydrogen concentration gradient change ≥0.02% / min, air introduction flow fluctuation > ±5m, and boiler feedwater pump temperature fluctuation > ±4℃ / 10min. 3 / h / 10min 3 / h / 10min A3. Constructing a mapping relationship between different burst pipe types and priority anomaly parameters: Based on the anomaly threshold and variation law of priority anomaly parameters, construct a one-to-one mapping relationship between different burst pipe types and priority anomaly parameters, forming a burst pipe type-priority anomaly parameter mapping library; where the formula for calculating the change in hydrogen concentration gradient is ΔH2 / Δt=t2-t1H2(t2)-H2(t1), where ΔH2 / Δt is the change in hydrogen concentration gradient, in units of % / min; H2(t1) is the hydrogen concentration value at time t1; H2(t2) is the hydrogen concentration value at time t2; t2-t1 is the time interval.

[0033] In some embodiments of this application, in step A1, 120 sets of historical tube rupture case data of dry quenching coke boilers are collected; in step A3, t2-t1 is a time interval, which can be 1 minute.

[0034] In some embodiments of this application, in step A1, a full-cycle parameter retrospective analysis is performed on 120 historical tube rupture cases of dry quenching coke boilers, including 68 cases of header tube rupture and 52 cases of economizer tube rupture.

[0035] In some embodiments of this application, in step A2, the header tube directly bears the scouring and pressure load of high-temperature flue gas, and the priority abnormal parameters before the tube burst are: • Boiler outlet gas temperature: Normal operating range 380-420℃, fluctuates 1.5-2.5 hours before tube rupture, fluctuation range >±5℃ / 10min; • Boiler inlet gas temperature: Normal operating range 920-960℃, fluctuation range >±3℃ / 10min 1.8-2.8 hours before tube rupture; • Boiler pressure: Normal operating range 8.5-9.5MPa, fluctuation range >±0.2MPa / 10min 1.2-2.2 hours before tube rupture; • Air conduction capacity: Normal operating range 120-150m 3 / h, fluctuation range > ±8m 1.0-2.0 hours before pipe burst 3 / h / 10min.

[0036] In some embodiments of this application, in step A2, the economizer is prone to pipe wall corrosion due to the presence of trace amounts of corrosive substances in the feedwater. The priority abnormal parameters before pipe rupture are: • Hydrogen concentration: Normal operating range 0.3%-0.5%, gradient change ≥0.02% / min 2.0-3.0 hours before tube rupture (i.e., concentration increase ≥0.01% every 30 seconds); • Air intake: Normal operating range 80-100m 3 / h, fluctuation range > ±5m 1.8-2.8 hours before pipe burst. 3 / h / 10min; • Boiler feed water pump temperature: Normal operating range 40-50℃, fluctuation range > ±4℃ / 10min 1.5-2.5 hours before tube rupture.

[0037] In some embodiments of this application, in step S3, missing value processing is performed as follows: short-term missing values ​​(≤5 sampling points) are filled by linear interpolation, and long-term missing values ​​(>5 sampling points) are marked as "sensor offline" and a device status warning is triggered to avoid feature bias caused by missing data; outlier filtering: based on time series analysis, only outliers with a duration of ≥1 minute are retained to filter out instantaneous sensor drift (such as 0.5-second jumps) and reduce invalid interference; integrity verification: data with a single timestamp null value ratio >50% are deleted to ensure the integrity of parameters at a single time point and avoid collaborative analysis bias caused by "partial parameter missingness"; wherein, the formula for calculating the single timestamp null value ratio is: R null =MN null ×100% Where: R null Percentage of null values ​​for a single timestamp (%); N null M represents the number of null parameters at this timestamp; M represents the total number of monitored parameters.

[0038] In some embodiments of this application, the dynamic baseline range in step S4 is constructed using an adaptive baseline model, which includes: B1. The Long Short-Term Memory (LSTM) network was used to learn the normal operation data of the dry quenching coke boiler over the past 365 days to initially construct the baseline range of each parameter; B2. The deviation rate between the current operating parameters and the existing baseline is automatically evaluated every 24 hours. If the deviation rate is ≥5% or the boiler load change is >20%, the baseline is dynamically updated. B3. When dynamically updating the baseline, a gradient compensation algorithm is used to correct the upper and lower limits of the baseline to ensure that the baseline range is adapted to the current equipment status and operating conditions.

[0039] In some embodiments of this application, in step B3, the parameter deviation rate is calculated using the formula R. dev =P base |P curr -P base |×100%, where: R dev P is the parameter deviation rate. curr This is the current running parameter value; P base The baseline parameter values ​​are given; the gradient compensation correction formula is: P base-new =P base-old +k×t, where P base-new The updated baseline value; P base-old t represents the baseline value before the update; k is the gradient compensation coefficient; t is the number of days the baseline has been running.

[0040] In some embodiments of this application, in step B3, k is the gradient compensation coefficient, and the boiler outlet gas temperature gradient compensation coefficient is 0.0055℃ / day.

[0041] In some embodiments of this application, in step S4, if at least one of the priority abnormal parameters related to the header light tube (such as boiler outlet gas temperature) exceeds the dynamic baseline range and the abnormality lasts for ≥1 minute, it is determined that there is a precursor to the header light tube rupture. • If at least one of the priority abnormal parameters related to the economizer (such as hydrogen concentration) exceeds the dynamic baseline range, and the duration of the abnormality is ≥1 minute, then a precursor to economizer tube rupture is determined; the formula for calculating the proportion of a single parameter exceeding the limit is: r=x upper -x lower max(0,xx upper )+max(0,x lower In the formula -x), r is the over-limit ratio of a single parameter; x is the real-time parameter value; x upper The upper limit of the baseline; x lower This is the lower limit of the baseline.

[0042] In some embodiments of this application, in step S5, the extraction rules for collaborative features include: for a burst header tube: the ratio of the rate of change of boiler outlet gas temperature to the rate of change of boiler pressure is R. T-P =V P V T V T =ΔtΔT is the rate of change of boiler outlet gas temperature, in °C / min, V P =ΔtΔP is the rate of change of boiler pressure, in MPa / min, R T-P The normal range is 0.8-1.2; The hysteresis correlation between air conductance and boiler inlet gas temperature was calculated using the Pearson correlation coefficient, with a normal range of 0.6-0.9. Where Xi is the air conductance value, X is the mean air conductance, Yi is the boiler inlet gas temperature at hysteresis τ, Y is the mean temperature, n is the number of sampling points, and τ is 5 min.

[0043] In some embodiments of this application, the extraction rules for the cooperative features in step S5 include: for an economizer tube rupture, the cooperative change rate R of hydrogen concentration and air introduction amount. H-A =ΔA / ΔtΔH2 / Δt, the normal range of value is 0.5-0.8, where the rate of change of air intake is ΔA / Δt=t2-t1A(t2)-A(t1), the unit is m・min, and A(t1) and A(t2) are the air intake values ​​at time t1 and t2 respectively; 3 / h The lag correlation between boiler feedwater pump temperature and boiler feedwater flow rate is calculated using the Pearson correlation coefficient. The formula is the same as the lag correlation between air conductance and boiler inlet gas temperature, with a normal range of 0.7-0.95. Where Xi corresponds to the boiler feedwater pump temperature value, and Yi corresponds to the boiler feedwater flow rate value at the lag time τ, where τ is taken as 3 min. A correlation of <0.7 indicates a decrease in economizer heat exchange efficiency.

[0044] In some embodiments of this application, principal component analysis (PCA) is used to compress the aforementioned collaborative features from 15 dimensions to 8 dimensions, retaining 98.5% of the key information and reducing the model training complexity; wherein, the formula for calculating collaborative anomaly quantification is: In the formula: C group,t S represents the quantized value of the collaborative anomaly at time t; m represents the number of anomaly parameters at time t; S abn W represents the set of abnormal parameters at time t. a,b The correlation strength weight between outlier parameters a and b (values ​​range from 0 to 1).

[0045] In some embodiments of this application, in step S6, the pre-trained burst pipe prediction model is constructed using Gradient Boosting Decision Tree (GBDT), and the model training process includes: C1. Using the historical pipe burst case data and corresponding collaborative features collected in step A1 as training samples, the pipe burst type and pipe burst time are labeled as tags; C2. Bayesian optimization is used to determine the model hyperparameters, including the learning rate, maximum depth of the decision tree, and number of base evaluators; C3. The model is trained using 5-fold cross-validation. Training is complete when the model's accuracy in identifying burst pipe types is ≥96% and the burst pipe time prediction error is ≤15%. The formula for calculating the burst pipe time prediction error is: δ t =t true |t pred -t true |×100%, where: δ t The error in predicting the tube burst time; t pred The model predicts the pipe burst time, in hours (h); t true The actual burst time is expressed in hours. The formula for calculating the burst type identification accuracy is: Acc = TP + TN + FP + FNTP + TN × 100%, where: Acc is the type identification accuracy; TP is the number of true positive samples, i.e., correctly identifying burst tubes in the header or economizer; TN is the number of true negative samples, i.e., correctly identifying tubes that are not burst; FP is the number of false positive samples, i.e., misjudging tubes that are not burst; FN is the number of false negative samples, i.e., tubes that are not identified.

[0046] In some embodiments of this application, in step C1, the learning rate adopts a Gaussian distribution with a mean of 0.01 and a standard deviation of 0.01; the maximum depth of the decision tree is 3-10; and the number of base estimators follows a geometric distribution with a mean of 10.

[0047] In some embodiments of this application, in step S7, the following is determined based on the output results: when the predicted burst time is less than 2 hours, it is determined to be a level 1 risk and an immediate warning is triggered; when 2 hours < predicted burst time < 12 hours, it is determined to be a level 2 risk and a warning is triggered; when the predicted burst time is greater than 12 hours, it is determined to be a level 3 risk and only further detection is required.

[0048] In some embodiments of this application, in step S7, after the warning is triggered, differentiated processing is performed, specifically as follows: When the header tube rupture is classified as a Level 1 risk, the boiler and induced draft fan interlock shutdown is triggered, the header inlet and outlet valves are closed, and the pressure relief device is activated. When the header tube rupture is classified as a Level 2 risk, the load is reduced to 80% of the design load, and the monitoring of boiler inlet and outlet gas temperature and pressure is strengthened, with parameter changes recorded every 30 minutes. When the economizer tube rupture is classified as a Level 1 risk, reduce the load to 50% of the design load, switch to the standby feedwater pump, and isolate and repair the economizer area. When the economizer tube rupture is classified as a Level 2 risk, reduce the air intake to 90% of the normal range, check the operating status of the boiler feedwater pump, and record the hydrogen concentration change every 20 minutes.

[0049] In some embodiments of this application, when the header tube rupture is at level two risk: reduce the load to 80% of the design load, close some flue gas inlet valves to stabilize the temperature; record the boiler inlet and outlet gas temperature, pressure and air conduction flow every 30 minutes; if the parameter fluctuation range decreases, the load operation can be maintained and continuous monitoring can be carried out. When a header tube bursts and is at Level 1 risk: immediately trigger the boiler and induced draft fan interlock shutdown, close the header inlet and outlet valves, and start the steam pressure relief device (pressure relief rate ≤ 0.5 MPa / min) to prevent secondary damage to the pipeline caused by a sudden pressure drop; When the economizer tube rupture is at level 2 risk: reduce the air intake to 90% of the normal range to reduce the oxidation rate of corrosive substances; switch the boiler feedwater to the standby feedwater pump and check if the temperature sensor of the original feedwater pump is abnormal; collect the hydrogen concentration every 20 minutes, and if the concentration gradient change drops to <0.01% / min, the normal air intake can be restored. When the economizer tube rupture is at Level 1 risk: reduce the load to 50% of the design load, isolate the economizer area (close the economizer inlet and outlet valves), and start the chemical cleaning device to descale and prevent corrosion of the economizer pipeline; after the parameters return to normal, gradually increase the load to 80%, observe for 2 hours if there are no abnormalities, and then restore the full load.

[0050] In some embodiments of this application, in step S8, the generated structured risk decision output is sent to the dry quenching coke production system control platform through a standard industrial communication interface.

[0051] In some embodiments of this application, step S8 further includes classifying the pipe burst event status into four categories: "active, ended, not properly ended, and stabilized," with the status switching rules as follows: • Active: After the alert is triggered, the abnormal parameters have not returned to the baseline range; • Completed: Abnormal parameter recovery rate ≥ 90%, calculated using the formula: R recover =N abn N recover ×100% Where: R recover The percentage of abnormal parameters recovered (%); N recover The number of outlier parameters to be restored to the baseline range; N abn This represents the total number of abnormal parameters; • Not ending normally: If the recovery rate of abnormal parameters is less than 90% more than 2 hours after the warning is triggered, a second warning will be triggered; • Stable status restored: Abnormal parameters remained normal for 1 hour, and the event was archived and saved as a five-dimensional visualization chart (gas, pressure, temperature, liquid level, others) to the historical case library.

[0052] In other embodiments of this application, a multi-parameter collaborative intelligent prediction and management device for tube rupture in dry quenching boilers is also provided, which uses the prediction and management method of one of the above embodiments, comprising: Mapping library construction module: used to build a mapping library of burst tube types and priority anomaly parameters, and to determine the priority anomaly parameters and anomaly thresholds for burst tubes in headers and economizers; Parameter acquisition module: used to acquire multi-dimensional operating parameters in real time through sensor network, covering priority abnormal parameters and related collaborative parameters; Data preprocessing module: used to handle missing values, filter outliers, and verify data integrity of the collected parameters; Priority anomaly identification module: used to identify priority anomaly parameters based on the mapping library and determine whether any parameters exceed the dynamic baseline range; Feature extraction module: used to extract collaborative features between priority anomaly parameters and associated collaborative parameters; Prediction output module: Used to input collaborative features into the pre-trained model and output the risk level, type, and prediction time of pipe bursts; Early warning and response module: used to trigger different levels of early warnings based on the prediction results.

[0053] Industrial Data Interface Module: This module is used to write or synchronize the structured risk decision output generated by the early warning and decision support module to the OPCUA server deployed on the lower-level machine through a predefined internal interface or message queue, thereby providing standardized data access services to the external industrial environment through the server.

[0054] In some embodiments of this application, a model iteration module is also included: used to automatically update the pipe burst type-priority anomaly parameter mapping library and pipe burst prediction model when a new pipe burst case or process adjustment is added, so as to ensure continuous optimization of model performance.

[0055] This application, by first discovering the priority abnormal parameters for different tube rupture types, and combining them with multi-dimensional quantitative formulas to construct a multi-parameter collaborative prediction system, enables early and accurate warning and full life-cycle management of tube rupture in dry quenching coke boilers. It has extremely high industrial application value and market promotion prospects.

[0056] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent prediction and management of tube rupture in dry quenching coke boilers based on multi-parameter synergy, characterized in that, Includes the following steps: S1. Construct a pipeline burst type-priority anomaly parameter mapping library: The types of tube ruptures include at least header tube ruptures and economizer tube ruptures. The priority abnormal parameters are the core parameters that first show abnormal fluctuations before the tube rupture occurs. Among them, the priority abnormal parameters for header tube ruptures include boiler outlet gas temperature, boiler inlet gas temperature, boiler pressure, and air conduction flow rate; the priority abnormal parameters for economizer tube ruptures include hydrogen concentration, air introduction rate, and boiler feedwater pump temperature. S2. Collection of multi-dimensional operating parameters: The sensor network of the dry quenching coke boiler collects multi-dimensional operating parameters in real time. The multi-dimensional operating parameters include at least the priority anomaly parameters and the associated coordination parameters. The associated coordination parameters are auxiliary parameters that are strongly correlated with the priority anomaly parameters, i.e., the correlation coefficient |R|≥0.

7. S3. Preprocess the collected multi-dimensional operating parameters: The preprocessing includes distinguishing between short-term and long-term missing data and filling them with linear interpolation and sensor offline warning respectively, filtering out transient interference anomalies with a duration of less than 1 minute, and deleting data with a single timestamp null value ratio greater than 50%; where short-term missing data is defined as missing data with a sampling period of Δt and a missing duration of less than or equal to 5Δt, and long-term missing data is defined as missing data with a missing duration greater than 5Δt. S4. Priority Anomaly Parameter Identification: Based on the burst pipe type-priority anomaly parameter mapping library, priority anomaly parameters are identified for the preprocessed multi-dimensional operating parameters to determine whether at least one priority anomaly parameter exceeds the dynamic baseline range. S5. Determine if there is any priority abnormal data that exceeds the dynamic baseline range: If there is a priority abnormal parameter that exceeds the dynamic baseline range, extract the collaborative features of the priority abnormal parameter and the corresponding associated collaborative parameter. The collaborative features include parameter change rate, abnormal duration, and lag correlation degree between parameters. S6. Input the collaborative features into the pre-trained pipe burst prediction model and output the pipe burst type and predicted pipe burst time; S7. Risk Classification and Early Warning Processing: Based on the output results, risk classification is performed, a structured risk decision output is generated, an early warning is triggered according to the risk level, and the decision output is released. S8. The generated structured risk decision output is sent to the central control platform of the dry quenching coke production system as an auxiliary decision-making basis for the system to optimize and adjust operating parameters or perform safety interlocks.

2. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 1, is characterized in that... The method for constructing the burst pipe type-priority anomaly parameter mapping library in step S1 includes: A1. Collect historical tube rupture case data from multiple sets of dry quenching coke boilers: The case data shall include at least the full-cycle parameter change records of header tube rupture and economizer tube rupture; A2. Time Series Backtracking Analysis: The first abnormal parameter before the occurrence of each type of tube rupture was determined as the priority abnormal parameter. The priority abnormal parameters for header tube rupture met the following conditions: boiler outlet gas temperature fluctuation > ±5℃ / 10min, boiler inlet gas temperature fluctuation > ±3℃ / 10min, boiler pressure fluctuation > ±0.2MPa / 10min, and air conduction flow fluctuation > ±8m³ / min. 3 / h / 10min; The priority abnormal parameters for economizer tube rupture must meet the following conditions: hydrogen concentration gradient change ≥ 0.02% / min, air introduction rate fluctuation > ±5m 3 / h / 10min, boiler feedwater pump temperature fluctuation range >±4℃ / 10min; A3. Constructing a mapping relationship between different burst pipe types and priority anomaly parameters: Constructing a one-to-one mapping relationship between different burst pipe types and priority anomaly parameters to form a burst pipe type-priority anomaly parameter mapping library; where the formula for calculating the change in hydrogen concentration gradient is ΔH2 / Δt=t2-t1H2(t2)-H2(t1), where ΔH2 / Δt is the change in hydrogen concentration gradient, in units of % / min; H2(t1) is the hydrogen concentration value at time t1; H2(t2) is the hydrogen concentration value at time t2; t2-t1 is the time interval.

3. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 1, is characterized in that... The dynamic baseline range in step S4 is constructed using an adaptive baseline model, which includes: B1. The Long Short-Term Memory (LSTM) network was used to learn the normal operation data of the dry quenching coke boiler over the past 365 days to initially construct the baseline range of each parameter; B2. The deviation rate between the current operating parameters and the existing baseline is automatically evaluated every 24 hours. If the deviation rate is ≥5% or the boiler load change is >20%, the baseline is dynamically updated. B3. When dynamically updating the baseline, a gradient compensation algorithm is used to correct the upper and lower limits of the baseline to ensure that the baseline range is adapted to the current equipment status and operating conditions.

4. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 3, is characterized in that... In step B3, the formula for calculating the parameter deviation rate is R. dev =P base |P curr -P base |×100%, where: R dev P is the parameter deviation rate. curr This is the current running parameter value; P base The baseline parameter values ​​are given; the gradient compensation correction formula is: P base-new =P base-old +k×t, where P base-new The updated baseline value; P base-old The value is the baseline value before the update; k is the gradient compensation coefficient; t is the number of days the baseline has been running.

5. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy as described in claim 1, characterized in that, In step S5, the rules for extracting collaborative features include: for a burst header tube: the ratio of the rate of change of boiler outlet gas temperature to the rate of change of boiler pressure is R. T-P =V P V T V T =ΔtΔT is the rate of change of boiler outlet gas temperature, in °C / min, V P =ΔtΔP is the rate of change of boiler pressure, in MPa / min, R T-P The normal range is 0.8-1.2; The hysteresis correlation between air conductance and boiler inlet gas temperature was calculated using the Pearson correlation coefficient. The normal range is 0.6-0.9; where Xi is the air conductance value, X is the mean air conductance value, Yi is the boiler inlet gas temperature value at time lag τ, Y is the corresponding mean temperature, and n is the number of sampling points.

6. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 5, is characterized in that... In step S5, the rules for extracting the cooperative features include: for economizer tube rupture, the cooperative change rate R of hydrogen concentration and air introduction rate. H-A =ΔA / ΔtΔH2 / Δt, with a normal range of 0.5-0.

8. The rate of change of the air intake is ΔA / Δt = t2-t1A(t2)-A(t1), in meters. 3 / h・min, A(t1) and A(t2) are the air inlet values ​​at times t1 and t2, respectively; The lag correlation between boiler feedwater pump temperature and boiler feedwater flow rate is calculated using the Pearson correlation coefficient. The formula is the same as the lag correlation between air conductance and boiler inlet gas temperature, with a normal range of 0.7-0.

95. Where Xi corresponds to the boiler feedwater pump temperature value, and Yi corresponds to the boiler feedwater flow rate value at the lag time τ, where τ is taken as 3 min.

7. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy as described in claim 1, characterized in that, In step S6, the pre-trained burst pipe prediction model is constructed using Gradient Boosting Decision Tree (GBDT). The model training process includes: C1. Using the historical pipe burst case data and corresponding collaborative features collected in step A1 as training samples, the pipe burst type and pipe burst time are labeled as tags; C2. Bayesian optimization is used to determine the model hyperparameters, including the learning rate, maximum depth of the decision tree, and number of base evaluators; C3. The model is trained using 5-fold cross-validation. Training is complete when the model's accuracy in identifying burst pipe types is ≥96% and the burst pipe time prediction error is ≤15%. The formula for calculating the burst pipe time prediction error is: δ t =t true |t pred -t true |×100%, where: δ t The error in predicting the tube burst time; t pred The model predicts the pipe burst time, in hours (h); t true The actual burst time is expressed in hours. The formula for calculating the burst type identification accuracy is: Acc = TP + TN + FP + FNTP + TN × 100%, where: Acc is the type identification accuracy; TP is the number of true positive samples, i.e., correctly identifying burst tubes in the header or economizer; TN is the number of true negative samples, i.e., correctly identifying tubes that are not burst; FP is the number of false positive samples, i.e., incorrectly identifying tubes that are not burst as burst tubes; FN is the number of false negative samples, i.e., tubes that are not identified as burst tubes.

8. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 7, is characterized in that... In step C1, the learning rate follows a Gaussian distribution with a mean of 0.01 and a standard deviation of 0.01; the maximum depth of the decision tree is 3-10; and the number of base estimators follows a geometric distribution with a mean of 10.

9. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy as described in claim 1, characterized in that, In step S7, based on the output results, the following judgments are made: if the predicted burst time is less than 2 hours, it is judged as a level 1 risk and an immediate warning is triggered; if 2 hours < predicted burst time < 12 hours, it is judged as a level 2 risk and a warning is triggered; if the predicted burst time is greater than 12 hours, it is judged as a level 3 risk and only further detection is required.

10. The intelligent prediction and management method for tube rupture in a dry quenching boiler based on multi-parameter synergy, as described in claim 9, is characterized in that... In step S7, after the early warning is triggered, differentiated handling is carried out, specifically: when the header tube rupture is a level 1 risk, the load is reduced to 80% of the design load, the monitoring of boiler inlet and outlet gas temperature and pressure is strengthened, and parameter changes are recorded every 30 minutes; when the header tube rupture is a level 2 risk, the boiler and induced draft fan are interlocked and shut down, the header inlet and outlet valves are closed, and the pressure relief device is activated. When the economizer tube rupture is classified as a Level 1 risk, reduce the air intake to 90% of the normal range, check the operating status of the boiler feedwater pump, and record the hydrogen concentration change every 20 minutes. When the economizer tube rupture is classified as a Level 2 risk, reduce the load to 50% of the design load, switch to the standby feedwater pump, and isolate and repair the economizer area.

Images