Coal-fired power plant boiler low-load hydrodynamic early warning method and device

Abstract

The invention discloses a coal-fired power plant boiler low-load hydrodynamic early warning method and device, the device comprises a boiler end and a central control terminal, the boiler end is internally provided with a sensor assembly, and the sensor assembly comprises a water level sensor, a pressure sensor, a steam flow sensor, a temperature sensor and a flue gas analyzer; a control unit is arranged in the central control end, an information processing end is arranged in the central control end, and the central control end is connected with the sensor assembly through wireless / wired data transmission; according to the coal-fired power plant boiler low-load hydrodynamic early warning method and device, through real-time monitoring and intelligent early warning, accidents caused by hydrodynamic unbalance can be effectively prevented, and safe operation of boiler equipment is guaranteed; by accurately predicting and adjusting the hydrodynamic characteristics in the low-load state, the combustion process of the boiler can be optimized, the fuel utilization rate and the heat efficiency can be improved, and by means of intelligent automatic adjustment and early warning, the human input is reduced, and the reaction speed and the accuracy are improved.

Description

Technical Field

[0001] The invention relates to the technical field of boiler operation status early warning, and in particular to a low-load hydraulic early warning method and device for a boiler in a coal-fired power plant. Background Art

[0002] In the daily operation of coal-fired power plants, low-load operation of boilers often brings a series of hydrodynamic problems, such as water level fluctuations, steam flow fluctuations, and uneven temperatures. These problems not only affect the thermal efficiency of the boiler, but may also threaten the safety of the boiler equipment. Therefore, a scientific and reliable early warning method is developed to address the hydrodynamic problems of boilers under low-load conditions, which can provide timely early warnings before hydrodynamic anomalies occur and ensure the safe operation of the boiler.

[0003] The prior art has the following defects or problems: The prior art has a publication number of CN117823882B, in which a method and device for early warning of low-load hydrodynamics of a coal-fired power plant boiler includes marking key position nodes of a hydrodynamic medium pipeline of a coal-fired power plant boiler; deploying data acquisition sensors at the nodes marked as key positions, and presetting the acquisition time nodes of the data acquisition sensors; the acquisition sensors include flow sensors, pressure sensors, and temperature sensors; according to the pre-set acquisition time nodes, controlling the data acquisition sensors to acquire hydrodynamic operation parameters, and performing data processing on the acquired hydrodynamic operation parameters to obtain a hydrodynamic three-dimensional operation characteristic matrix;

[0004] During use, the above equipment can more comprehensively and accurately evaluate the low-load hydraulic operation status of the boiler, detect potential problems in advance and issue early warnings, which helps to ensure the safe and stable operation of the boiler system;

[0005] In the above content, the boiler operating status is detected through matrix modeling, which does not have the ability of optimization and iterative upgrading. It is unable to automatically adjust and warn the boiler operating status, and the manpower investment is large.

[0006] It should be noted that the above contents belong to the technical knowledge of the inventor and do not necessarily constitute prior art. Summary of the invention

[0007] The purpose of the present invention is to address the deficiencies of the prior art and provide a method and device for early warning of low-load hydraulic power of a boiler in a coal-fired power plant, so as to solve the problems raised in the background technology.

[0008] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a low-load hydraulic early warning device for a boiler in a coal-fired power plant, comprising:

[0009] A boiler end, wherein the boiler end is equipped with a sensor assembly, wherein the sensor assembly includes a water level sensor, a pressure sensor, a steam flow sensor, a temperature sensor and a flue gas analyzer;

[0010] A central control terminal, wherein a control unit is provided inside the central control terminal, an information processing terminal is provided inside the central control terminal, and the central control terminal is connected to the sensor assembly via wireless / wired data transmission;

[0011] The control unit is data connected to an external data storage cloud platform support.

[0012] Another technical problem to be solved by the present invention is a low-load hydraulic early warning method for a coal-fired power plant boiler, comprising the following steps:

[0013] Step 1: Determine monitoring objectives, key parameters and monitoring requirements;

[0014] Step 2: Complete the establishment of a real-time monitoring system according to demand; select various sensors and install them in different parts of the boiler to comprehensively obtain the boiler's operating data;

[0015] Step 3: Transmit the signals collected by the sensor to the central control terminal via wireless / wired data to ensure that the data can be transmitted in real time;

[0016] Step 4: Display the real-time data of various parameters through the visual interface, and pre-process the data to facilitate the establishment of subsequent early warning models;

[0017] Step 5: Build an early warning model for real-time monitoring data. You can choose between statistical analysis and machine learning methods. After completing system testing and verification, optimize and regularly maintain the system.

[0018] As a preferred technical solution of the present invention, the monitoring objectives are specifically: by real-time monitoring of the boiler operation status, water dynamic anomalies can be discovered in advance to ensure stable operation of the boiler;

[0019] The key parameters are specifically: selecting key operating parameters related to water dynamics in the boiler, including but not limited to water level, steam flow, steam pressure, temperature, flue gas composition, water feed rate and heat conversion efficiency;

[0020] The monitoring requirements specifically include: the number, location and type of monitoring points, the reasonable arrangement of sensors to cover key parts of the boiler and the data frequency.

[0021] As a preferred technical solution of the present invention, the water level sensor is selected from one of the float type, water pressure type and radar type and is arranged inside the boiler water-cooled wall. The pressure sensor is selected to be adaptable to high temperature environment and is arranged inside the furnace and the superheater. The steam flow sensor is specifically an electromagnetic flowmeter, which is arranged inside the boiler feed water system. The temperature sensor is specifically a resistance temperature detector, which is arranged inside the boiler feed water system. The flue gas analyzer is used to detect flue gas components and is arranged inside the exhaust end of the boiler.

[0022] As a preferred technical solution of the present invention, the data processing includes filtering, denoising and data cleaning, detecting and marking abnormal data to ensure the reliability of subsequent analysis.

[0023] As a preferred technical solution of the present invention, the specific operation steps of the statistical analysis method are as follows:

[0024] 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data;

[0025] 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range;

[0026] 3) Conduct basic statistical analysis on each monitoring, such as mean, standard deviation, minimum, maximum and quartiles, analyze the distribution of each parameter and understand its normal fluctuation range;

[0027] 4) Use histograms, box plots and other methods to view the distribution of data, determine whether the parameters present normal distribution, skewed distribution or other forms of distribution, and analyze the changing trends of each parameter over time, especially the changing rules under different loads and different working conditions;

[0028] 5) Use correlation coefficients, such as the Pearson correlation coefficient, to analyze the relationship between various parameters. For example, the relationship between water level and steam flow, and the relationship between feed water flow and boiler load;

[0029] By analyzing the correlation between parameters, the key indicators that have the greatest impact on hydrodynamic changes are screened out;

[0030] 6) Analyze the abnormal situations in historical data and identify the characteristic patterns when the abnormalities occur. You can draw time series graphs, trend graphs, etc. to observe the changing trends of related parameters before and after the abnormal events;

[0031] 7) Based on the results of descriptive statistical analysis, set a reasonable threshold. For example, the mean ± n times the standard deviation of the parameter can be selected as the normal range;

[0032] 8) For important key parameters, upper and lower limits can be set according to the extreme values ​​of historical data as the trigger points of alarms;

[0033] 9) Using the quantile method to set the threshold, such as setting it to the 95% quantile, 99% quantile, etc., means that the probability of the data exceeding the threshold is 5% or 1%;

[0034] 10) For complex situations involving multiple parameters, multi-dimensional joint thresholds can be set, such as the joint thresholds of steam pressure and water level. If both exceed the normal range at the same time, an alarm is triggered, and a weighted or comprehensive scoring method is used to determine the weighted thresholds according to the importance of different parameters;

[0035] 11) Establish early warning rules, as follows:

[0036] Single threshold warning rule: When a parameter exceeds the set upper or lower limit, the system triggers an alarm. For example, when the boiler water level is lower than the set value, the system issues an alarm;

[0037] Combined threshold rules: Set the combined rules of multiple parameters. For example, if the water level is lower than the normal value and the steam flow is unstable, the system will consider that a hydrodynamic anomaly has occurred and trigger an early warning.

[0038] Time window rule: Set a time window and observe the fluctuation of data within the window; if the data continues to exceed the threshold within a certain period of time (such as 30 minutes, 1 hour, etc.), an alarm will be issued;

[0039] 12) Use test data to verify the effectiveness of the early warning model. The test data should include data when the boiler is operating normally and data when an abnormality occurs;

[0040] 13) Calculate the accuracy, recall and F1 value of the model to evaluate the performance of the early warning system;

[0041] 14) Analyze false alarms and missed alarms, and optimize threshold settings and rules;

[0042] 15) According to the verification results, adjust the threshold and warning rules to reduce false alarms and missed alarms;

[0043] The weighting coefficient of the parameter can be adjusted by increasing and decreasing the threshold range;

[0044] If the system frequently gives false alarms, you need to further adjust the normal range settings;

[0045] 16) Configure a real-time data acquisition system to continuously monitor the key operating parameters of the boiler and compare them with the set thresholds. The data can be transmitted to the central monitoring terminal in real time through SCADA and PLC;

[0046] 17) Once a parameter is detected to exceed the set threshold, the system should promptly trigger an alarm and notify the operator. The warning can be conveyed in different ways, such as visual alarm, sound alarm and SMS notification;

[0047] 18) The system sends the alarm information to the operator and provides detailed alarm data, so the operator can take quick measures based on the alarm information;

[0048] 19) Keep detailed records of each alarm, produce alarm logs, and regularly analyze alarm data to identify possible operating problems. For high-frequency alarm problems, in-depth analysis can be performed to improve boiler operating efficiency;

[0049] 20) Regularly evaluate the performance of the early warning model, and adjust and optimize the thresholds, rules and data models according to the actual operation situation, integrate boiler operation experience and new data, and continuously update and improve the early warning model;

[0050] 21) Regularly check the accuracy of sensor data collection and its working status to ensure the stability of the system and the real-time nature of the data, update software and hardware, and ensure that the system can adapt to new technologies and requirements.

[0051] As a preferred technical solution of the present invention, the specific operation steps of the machine learning method are as follows:

[0052] 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data;

[0053] 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range;

[0054] 3) Extract features from the raw data that are helpful for prediction, as follows:

[0055] Time window characteristics, such as average pressure and temperature over the past 30 minutes;

[0056] Statistical characteristics, such as mean, standard deviation, maximum, minimum, skewness, and kurtosis;

[0057] Lag features: based on the delay relationship of the time series, the data features of the production lag time are generated;

[0058] Periodic features, extract features based on the boiler’s operating cycle;

[0059] 4) If there are marked faults and abnormal records in the historical data, supervised learning methods can be used for training. Common supervised learning methods can be selected:

[0060] Classification models, such as logistic regression, support vector machine, decision tree, random forest, XGBoost, and LightGBM, are used to determine whether a fault or anomaly occurs;

[0061] Regression models, such as linear regression, support-linked regression, random forest regression, and gradient boosting regression, are used to predict future boiler operating parameters;

[0062] 5) If there are no anomaly records in the historical data, unsupervised learning methods can be used to detect anomalies based on the intrinsic structure of the data. Common supervised learning methods are available:

[0063] Clustering algorithms, such as K-means and DBSCAN, are used to discover different patterns in the data;

[0064] Isolation Forest, suitable for large-scale datasets;

[0065] Principal component analysis, to find outliers by reducing dimensionality;

[0066] Autoencoder, an unsupervised anomaly detection method based on neural networks, is used for anomaly identification in high-dimensional data;

[0067] 6) For high-dimensional data with complex patterns, deep learning models can be used, which can be selected:

[0068] Long short-term memory networks, which are used to process time series data and capture temporal dependencies;

[0069] Convolutional neural networks are used to extract features from multidimensional data and are suitable for processing complex spatial models;

[0070] Deep autoencoders, which are used for anomaly detection, especially in high-dimensional data;

[0071] 7) Divide the dataset into training set, validation set and test set, with the ratio of 70%-15%-15% divided. The training set is used to train the model, the validation set is used to tune the model parameters, and the test set is used to evaluate the model performance.

[0072] 8) Use the training set to train the selected machine learning model. For supervised learning, the model will learn the patterns in the data by minimizing the error and the error function.

[0073] 9) For unsupervised learning, the model will learn how to identify unusual patterns in the data;

[0074] 10) If it is a regression task, optimize the prediction error; if it is a classification task, optimize the classification accuracy;

[0075] 11) Use the validation set to evaluate the performance of the model, check the accuracy, precision, recall and F1 value of the model, and for regression problems, use the mean square error, root mean square error and R 2 Evaluation indicators: For classification problems, confusion matrix, ROC curve and AUC evaluation indicators are used;

[0076] 12) Use cross-validation methods to evaluate the performance of the model on different data subsets, reduce the risk of overfitting, and ensure the robustness of the model;

[0077] 13) Use grid search to tune the hyperparameters of the machine learning model, adjust the model's learning rate, tree depth, and regularization parameters to improve the model's performance;

[0078] 14) Optimize features based on model performance, remove irrelevant features, add new features, or use dimensionality reduction techniques to reduce redundant features;

[0079] 15) To improve the accuracy and robustness of prediction, ensemble learning methods can be used;

[0080] 16) Once the model predicts that the boiler's operating parameters are outside the normal range, an alarm is triggered. The alarm is triggered based on:

[0081] Probability threshold, when the model's anomaly probability exceeds the threshold;

[0082] Confidence threshold: when the confidence of the model is lower than a certain threshold, the prediction result is considered unreliable;

[0083] 17) Once the warning is triggered, the system will notify the operator through screen display, SMS, email and voice alarm, and the operator can make appropriate intervention and processing according to the alarm information;

[0084] 18) Deploy the trained model to the production environment and use containerization technology to directly integrate it into the monitoring system for real-time prediction;

[0085] 19) Continuously monitor the performance of the model in actual operation. If the model shows an increase in prediction error, the model can be retrained;

[0086] 20) The model should be updated regularly to maintain its accuracy as the boiler operating environment changes.

[0087] Compared with the prior art, the present invention provides a method and device for early warning of low-load hydraulic power of a coal-fired power plant boiler, which has the following beneficial effects:

[0088] Compared with the traditional boiler low-load hydrodynamic warning method, the coal-fired power plant boiler low-load hydrodynamic warning method and device can effectively prevent accidents caused by hydrodynamic imbalance and ensure the safe operation of boiler equipment through real-time monitoring and intelligent warning. By accurately predicting and adjusting the hydrodynamic characteristics under low-load conditions, the boiler combustion process can be optimized, fuel utilization and thermal efficiency can be improved, and intelligent automatic adjustment and warning can be used to reduce manpower input and improve response speed and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The drawings described herein are used to provide further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0090] Figure 1 It is a schematic diagram of the connection state of the low-load hydraulic early warning device of a coal-fired power plant boiler of the present invention;

[0091] Figure 2 The present invention is a schematic diagram of the implementation process of the low-load hydraulic early warning method for a coal-fired power plant boiler. DETAILED DESCRIPTION

[0092] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0093] In the first embodiment, refer to Figure 1 In this implementation scheme: the low-load hydraulic early warning device for boilers in coal-fired power plants includes:

[0094] Boiler end: The boiler end is equipped with a built-in sensor assembly, which includes a water level sensor, a pressure sensor, a steam flow sensor, a temperature sensor and a flue gas analyzer;

[0095] A central control terminal is provided with a control unit inside, which has an information processing terminal inside and is connected to the sensor assembly through wireless / wired data transmission;

[0096] The control unit is data-connected to the external data storage cloud platform bracket.

[0097] See also Figure 1-2 Based on the above-mentioned coal-fired power plant boiler low-load water dynamic early warning device, a coal-fired power plant boiler low-load water dynamic early warning method is proposed:

[0098] Step 1: Determine monitoring objectives, key parameters and monitoring requirements;

[0099] The specific monitoring objectives are: to detect water dynamic anomalies in advance and ensure the stable operation of the boiler by real-time monitoring of the boiler operation status;

[0100] The key parameters are specifically: select the key operating parameters related to water dynamics in the boiler, including but not limited to water level, steam flow, steam pressure, temperature, flue gas composition, feed water quantity and heat conversion efficiency;

[0101] The monitoring requirements are as follows: the number, location and type of monitoring points, the rational arrangement of sensors to cover all key parts of the boiler and the data frequency;

[0102] Step 2: Complete the establishment of a real-time monitoring system according to demand; select various sensors and install them in different parts of the boiler to comprehensively obtain the boiler's operating data;

[0103] The water level sensor is selected from one of the float type, water pressure type and radar type and is set inside the boiler water-cooled wall. The pressure sensor is selected to be adaptable to high temperature environment and is set inside the furnace and superheater. The steam flow sensor is specifically an electromagnetic flowmeter, which is set inside the boiler feed water system. The temperature sensor is specifically a resistance temperature detector, which is set inside the boiler feed water system. The flue gas analyzer is used to detect the flue gas composition and is set inside the boiler exhaust end.

[0104] Step 3: Transmit the signals collected by the sensor to the central control terminal via wireless / wired data to ensure that the data can be transmitted in real time;

[0105] Step 4: Display the real-time data of various parameters through the visual interface, and pre-process the data to facilitate the establishment of subsequent early warning models;

[0106] Data processing includes filtering, denoising and data cleaning, detecting and marking abnormal data to ensure the reliability of subsequent analysis.

[0107] Step 5: Build an early warning model for real-time monitoring data. You can use statistical analysis methods. After completing the testing and verification of the system, optimize and regularly maintain the system.

[0108] The specific steps of statistical analysis are as follows:

[0109] 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data;

[0110] 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range;

[0111] 3) Conduct basic statistical analysis on each monitoring, such as mean, standard deviation, minimum, maximum and quartiles, analyze the distribution of each parameter and understand its normal fluctuation range;

[0112] 4) Use histograms, box plots and other methods to view the distribution of data, determine whether the parameters present normal distribution, skewed distribution or other forms of distribution, and analyze the changing trends of each parameter over time, especially the changing rules under different loads and different working conditions;

[0113] 5) Use correlation coefficients, such as the Pearson correlation coefficient, to analyze the relationship between various parameters. For example, the relationship between water level and steam flow, and the relationship between feed water flow and boiler load;

[0114] By analyzing the correlation between parameters, the key indicators that have the greatest impact on hydrodynamic changes are screened out;

[0115] 6) Analyze the abnormal situations in historical data and identify the characteristic patterns when the abnormalities occur. You can draw time series graphs, trend graphs, etc. to observe the changing trends of related parameters before and after the abnormal events;

[0116] 7) Based on the results of descriptive statistical analysis, set a reasonable threshold. For example, the mean ± n times the standard deviation of the parameter can be selected as the normal range;

[0117] 8) For important key parameters, upper and lower limits can be set according to the extreme values ​​of historical data as the trigger points of alarms;

[0118] 9) Using the quantile method to set the threshold, such as setting it to the 95% quantile, 99% quantile, etc., means that the probability of the data exceeding the threshold is 5% or 1%;

[0119] 10) For complex situations involving multiple parameters, multi-dimensional joint thresholds can be set, such as the joint thresholds of steam pressure and water level. If both exceed the normal range at the same time, an alarm is triggered, and a weighted or comprehensive scoring method is used to determine the weighted thresholds according to the importance of different parameters;

[0120] 11) Establish early warning rules, as follows:

[0121] Single threshold warning rule: When a parameter exceeds the set upper or lower limit, the system triggers an alarm. For example, when the boiler water level is lower than the set value, the system issues an alarm;

[0122] Combined threshold rules: Set the combined rules of multiple parameters. For example, if the water level is lower than the normal value and the steam flow is unstable, the system will consider that a hydrodynamic anomaly has occurred and trigger an early warning.

[0123] Time window rule: Set a time window and observe the fluctuation of data within the window; if the data continues to exceed the threshold within a certain period of time (such as 30 minutes, 1 hour, etc.), an alarm will be issued;

[0124] 12) Use test data to verify the effectiveness of the early warning model. The test data should include data when the boiler is operating normally and data when an abnormality occurs;

[0125] 13) Calculate the accuracy, recall and F1 value of the model to evaluate the performance of the early warning system;

[0126] 14) Analyze false alarms and missed alarms, and optimize threshold settings and rules;

[0127] 15) According to the verification results, adjust the threshold and warning rules to reduce false alarms and missed alarms;

[0128] The weighting coefficient of the parameter can be adjusted by increasing and decreasing the threshold range;

[0129] If the system frequently gives false alarms, you need to further adjust the normal range settings;

[0130] 16) Configure a real-time data acquisition system to continuously monitor the key operating parameters of the boiler and compare them with the set thresholds. The data can be transmitted to the central monitoring terminal in real time through SCADA and PLC;

[0131] 17) Once a parameter is detected to exceed the set threshold, the system should promptly trigger an alarm and notify the operator. The warning can be conveyed in different ways, such as visual alarm, sound alarm and SMS notification;

[0132] 18) The system sends the alarm information to the operator and provides detailed alarm data, so the operator can take quick measures based on the alarm information;

[0133] 19) Keep detailed records of each alarm, produce alarm logs, and regularly analyze alarm data to identify possible operating problems. For high-frequency alarm problems, in-depth analysis can be performed to improve boiler operating efficiency;

[0134] 20) Regularly evaluate the performance of the early warning model, and adjust and optimize the thresholds, rules and data models according to the actual operation situation, integrate boiler operation experience and new data, and continuously update and improve the early warning model;

[0135] 21) Regularly check the accuracy of sensor data collection and its working status to ensure the stability of the system and the real-time nature of the data, update software and hardware, and ensure that the system can adapt to new technologies and requirements.

[0136] The core of the early warning model based on statistical analysis is to identify normal and abnormal patterns in boiler operation through the analysis of historical data, and to help operators detect potential hydrodynamic anomalies in a timely manner by setting reasonable thresholds and early warning rules; through real-time data monitoring and alarms, boiler accidents can be effectively prevented and the safety and stability of operation can be improved; at the same time, the model also needs to be continuously tuned and optimized according to actual data to ensure the accuracy and reliability of the early warning system.

[0137] In the second embodiment, refer to Figure 1 In this implementation scheme: the low-load hydraulic early warning device for boilers in coal-fired power plants includes:

[0138] Boiler end: The boiler end is equipped with a built-in sensor assembly, which includes a water level sensor, a pressure sensor, a steam flow sensor, a temperature sensor and a flue gas analyzer;

[0139] A central control terminal is provided with a control unit inside, which has an information processing terminal inside and is connected to the sensor assembly through wireless / wired data transmission;

[0140] The control unit is data-connected to the external data storage cloud platform bracket.

[0141] See also Figure 1-2 Based on the above-mentioned coal-fired power plant boiler low-load water dynamic early warning device, a coal-fired power plant boiler low-load water dynamic early warning method is proposed:

[0142] Step 1: Determine monitoring objectives, key parameters and monitoring requirements;

[0143] The specific monitoring objectives are: to detect water dynamic anomalies in advance and ensure the stable operation of the boiler by real-time monitoring of the boiler operation status;

[0144] The key parameters are specifically: select the key operating parameters related to water dynamics in the boiler, including but not limited to water level, steam flow, steam pressure, temperature, flue gas composition, feed water quantity and heat conversion efficiency;

[0145] The monitoring requirements are as follows: the number, location and type of monitoring points, the rational arrangement of sensors to cover all key parts of the boiler and the data frequency;

[0146] Step 2: Complete the establishment of a real-time monitoring system according to demand; select various sensors and install them in different parts of the boiler to comprehensively obtain the boiler's operating data;

[0147] The water level sensor is selected from one of the float type, water pressure type and radar type and is set inside the boiler water-cooled wall. The pressure sensor is selected to be adaptable to high temperature environment and is set inside the furnace and superheater. The steam flow sensor is specifically an electromagnetic flowmeter, which is set inside the boiler feed water system. The temperature sensor is specifically a resistance temperature detector, which is set inside the boiler feed water system. The flue gas analyzer is used to detect the flue gas composition and is set inside the boiler exhaust end.

[0148] Step 3: Transmit the signals collected by the sensor to the central control terminal via wireless / wired data to ensure that the data can be transmitted in real time;

[0149] Step 4: Display the real-time data of various parameters through the visual interface, and pre-process the data to facilitate the establishment of subsequent early warning models;

[0150] Data processing includes filtering, denoising and data cleaning, detecting and marking abnormal data to ensure the reliability of subsequent analysis.

[0151] Step 5: Build an early warning model for real-time monitoring data. You can use statistical analysis methods. After completing the testing and verification of the system, optimize and regularly maintain the system.

[0152] The specific steps of the machine learning method are as follows:

[0153] 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data;

[0154] 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range;

[0155] 3) Extract features from the raw data that are helpful for prediction, as follows:

[0156] Time window characteristics, such as average pressure and temperature over the past 30 minutes;

[0157] Statistical characteristics, such as mean, standard deviation, maximum, minimum, skewness, and kurtosis;

[0158] Lag features: based on the delay relationship of the time series, the data features of the production lag time are generated;

[0159] Periodic features, extract features based on the boiler’s operating cycle;

[0160] 4) If there are marked faults and abnormal records in the historical data, supervised learning methods can be used for training. Common supervised learning methods can be selected:

[0161] Classification models, such as logistic regression, support vector machine, decision tree, random forest, XGBoost, and LightGBM, are used to determine whether a fault or anomaly occurs;

[0162] Regression models, such as linear regression, support-linked regression, random forest regression, and gradient boosting regression, are used to predict future boiler operating parameters;

[0163] 5) If there are no anomaly records in the historical data, unsupervised learning methods can be used to detect anomalies based on the intrinsic structure of the data. Common supervised learning methods are available:

[0164] Clustering algorithms, such as K-means and DBSCAN, are used to discover different patterns in the data;

[0165] Isolation Forest, suitable for large-scale datasets;

[0166] Principal component analysis, to find outliers by reducing dimensionality;

[0167] Autoencoder, an unsupervised anomaly detection method based on neural networks, is used for anomaly identification in high-dimensional data;

[0168] 6) For high-dimensional data with complex patterns, deep learning models can be used, which can be selected:

[0169] Long short-term memory networks, which are used to process time series data and capture temporal dependencies;

[0170] Convolutional neural networks are used to extract features from multidimensional data and are suitable for processing complex spatial models;

[0171] Deep autoencoders, which are used for anomaly detection, especially in high-dimensional data;

[0172] 7) Divide the dataset into training set, validation set and test set, with the ratio of 70%-15%-15% divided. The training set is used to train the model, the validation set is used to tune the model parameters, and the test set is used to evaluate the model performance.

[0173] 8) Use the training set to train the selected machine learning model. For supervised learning, the model will learn the patterns in the data by minimizing the error and the error function.

[0174] 9) For unsupervised learning, the model will learn how to identify unusual patterns in the data;

[0175] 10) If it is a regression task, optimize the prediction error; if it is a classification task, optimize the classification accuracy;

[0176] 11) Use the validation set to evaluate the performance of the model, check the accuracy, precision, recall and F1 value of the model, and for regression problems, use the mean square error, root mean square error and R 2 Evaluation indicators: For classification problems, confusion matrix, ROC curve and AUC evaluation indicators are used;

[0177] 12) Use cross-validation methods to evaluate the performance of the model on different data subsets, reduce the risk of overfitting, and ensure the robustness of the model;

[0178] 13) Use grid search to tune the hyperparameters of the machine learning model, adjust the model's learning rate, tree depth, and regularization parameters to improve the model's performance;

[0179] 14) Optimize features based on model performance, remove irrelevant features, add new features, or use dimensionality reduction techniques to reduce redundant features;

[0180] 15) To improve the accuracy and robustness of prediction, ensemble learning methods can be used;

[0181] 16) Once the model predicts that the boiler's operating parameters are outside the normal range, an alarm is triggered. The alarm is triggered based on:

[0182] Probability threshold, when the model's anomaly probability exceeds the threshold;

[0183] Confidence threshold: when the confidence of the model is lower than a certain threshold, the prediction result is considered unreliable;

[0184] 17) Once the warning is triggered, the system will notify the operator through screen display, SMS, email and voice alarm, and the operator can make appropriate intervention and processing according to the alarm information;

[0185] 18) Deploy the trained model to the production environment and use containerization technology to directly integrate it into the monitoring system for real-time prediction;

[0186] 19) Continuously monitor the performance of the model in actual operation. If the model shows an increase in prediction error, the model can be retrained;

[0187] 20) The model should be updated regularly to maintain its accuracy as the boiler operating environment changes.

[0188] The early warning model based on machine learning can identify potential anomalies and faults in the boiler system by learning from historical data, and give early warnings. Compared with traditional statistical methods, machine learning models have stronger adaptability and prediction accuracy. Through reasonable data processing, feature engineering, model selection and tuning, an efficient and accurate boiler early warning system can be implemented to improve the safety and reliability of the equipment. At the same time, with the increase in data volume and continuous optimization of the model, the early warning capability of the system will continue to improve.

[0189] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0190] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A low-load hydraulic early warning device for boilers in coal-fired power plants, characterized in that: include: A boiler end, wherein the boiler end is equipped with a sensor assembly, wherein the sensor assembly includes a water level sensor, a pressure sensor, a steam flow sensor, a temperature sensor and a flue gas analyzer; A central control terminal, wherein a control unit is provided inside the central control terminal, an information processing terminal is provided inside the central control terminal, and the central control terminal is connected to the sensor assembly via wireless / wired data transmission; The control unit is data connected to an external data storage cloud platform support.

2. A low-load hydraulic early warning method for a coal-fired power plant boiler, characterized in that: The steps include: Step 1: Determine monitoring objectives, key parameters and monitoring requirements; Step 2: Complete the establishment of a real-time monitoring system according to demand; select various sensors and install them in different parts of the boiler to fully obtain the boiler's operating data; Step 3: Transmit the signals collected by the sensor to the central control terminal via wireless / wired data to ensure that the data can be transmitted in real time; Step 4: Display the real-time data of various parameters through the visual interface, and pre-process the data to facilitate the establishment of subsequent early warning models; Step 5: Build an early warning model for real-time monitoring data. You can choose between statistical analysis and machine learning methods. After completing system testing and verification, optimize and regularly maintain the system.

3. The method for early warning of low-load water dynamics of a coal-fired power plant boiler according to claim 1, characterized in that: The monitoring objectives described in step 1 are as follows: by real-time monitoring of the boiler operation status, water dynamic anomalies can be discovered in advance to ensure stable operation of the boiler; The key parameters are specifically: selecting key operating parameters related to water dynamics in the boiler, including but not limited to water level, steam flow, steam pressure, temperature, flue gas composition, water feed rate and heat conversion efficiency; The monitoring requirements specifically include: the number, location and type of monitoring points, the reasonable arrangement of sensors to cover key parts of the boiler and the data frequency.

4. The method for early warning of low-load water dynamics of a coal-fired power plant boiler according to claim 1, characterized in that: The water level sensor in step 2 is selected from one of float type, water pressure type and radar type and is arranged inside the boiler water-cooled wall. The pressure sensor is selected to be adaptable to high temperature environment and is arranged inside the furnace and superheater. The steam flow sensor is specifically an electromagnetic flowmeter, which is arranged inside the boiler feed water system. The temperature sensor is specifically a resistance temperature detector, which is arranged inside the boiler feed water system. The flue gas analyzer is used to detect flue gas components and is arranged inside the exhaust end of the boiler.

5. The method for early warning of low-load water dynamics of a coal-fired power plant boiler according to claim 1, characterized in that: The data processing described in step 4 includes filtering, denoising and data cleaning, detecting and marking abnormal data to ensure the reliability of subsequent analysis.

6. The method for early warning of low-load water dynamics of a coal-fired power plant boiler according to claim 1, characterized in that: The specific operation steps of the statistical analysis method described in step 5 are as follows: 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data; 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range; 3) Conduct basic statistical analysis on each monitoring, such as mean, standard deviation, minimum, maximum and quartiles, analyze the distribution of each parameter and understand its normal fluctuation range; 4) Use histograms, box plots and other methods to view the distribution of data, determine whether the parameters present normal distribution, skewed distribution or other forms of distribution, and analyze the changing trends of each parameter over time, especially the changing rules under different loads and different working conditions; 5) Use correlation coefficients, such as the Pearson correlation coefficient, to analyze the relationship between various parameters. For example, the relationship between water level and steam flow, and the relationship between feed water flow and boiler load; By analyzing the correlation between parameters, the key indicators that have the greatest impact on hydrodynamic changes are screened out; 6) Analyze the abnormal situations in historical data and identify the characteristic patterns when the abnormalities occur. You can draw time series graphs, trend graphs, etc. to observe the changing trends of related parameters before and after the abnormal events; 7) Based on the results of descriptive statistical analysis, set a reasonable threshold. For example, the mean ± n times the standard deviation of the parameter can be selected as the normal range; 8) For important key parameters, upper and lower limits can be set according to the extreme values ​​of historical data as the trigger points of alarms; 9) Using the quantile method to set the threshold, such as setting it to the 95% quantile, 99% quantile, etc., means that the probability of the data exceeding the threshold is 5% or 1%; 10) For complex situations involving multiple parameters, multi-dimensional joint thresholds can be set, such as the joint thresholds of steam pressure and water level. If both exceed the normal range at the same time, an alarm is triggered, and a weighted or comprehensive scoring method is used to determine the weighted thresholds according to the importance of different parameters; 11) Establish early warning rules, as follows: Single threshold warning rule: When a parameter exceeds the set upper or lower limit, the system triggers an alarm. For example, when the boiler water level is lower than the set value, the system issues an alarm; Combined threshold rules: Set the combined rules of multiple parameters. For example, if the water level is lower than the normal value and the steam flow is unstable, the system will consider that a hydrodynamic anomaly has occurred and trigger an early warning. Time window rule: Set a time window and observe the fluctuation of data within the window; if the data continues to exceed the threshold within a certain period of time (such as 30 minutes, 1 hour, etc.), an alarm will be issued; 12) Use test data to verify the effectiveness of the early warning model. The test data should include data when the boiler is operating normally and data when an abnormality occurs; 13) Calculate the accuracy, recall and F1 value of the model to evaluate the performance of the early warning system; 14) Analyze false alarms and missed alarms, and optimize threshold settings and rules; 15) According to the verification results, adjust the threshold and warning rules to reduce false alarms and missed alarms; The weighting coefficient of the parameter can be adjusted by increasing and decreasing the threshold range; If the system frequently gives false alarms, you need to further adjust the normal range settings; 16) Configure a real-time data acquisition system to continuously monitor the key operating parameters of the boiler and compare them with the set thresholds. The data can be transmitted to the central monitoring terminal in real time through SCADA and PLC; 17) Once a parameter is detected to exceed the set threshold, the system should promptly trigger an alarm and notify the operator. The warning can be conveyed in different ways, such as visual alarm, sound alarm and SMS notification; 18) The system sends the alarm information to the operator and provides detailed alarm data, so the operator can take quick measures based on the alarm information; 19) Keep detailed records of each alarm, produce alarm logs, and regularly analyze alarm data to identify possible operating problems. For high-frequency alarm problems, in-depth analysis can be performed to improve boiler operating efficiency; 20) Regularly evaluate the performance of the early warning model, and adjust and optimize the thresholds, rules and data models according to the actual operation situation, integrate boiler operation experience and new data, and continuously update and improve the early warning model; 21) Regularly check the accuracy of sensor data collection and its working status to ensure the stability of the system and the real-time nature of the data, update software and hardware, and ensure that the system can adapt to new technologies and requirements.

7. The method for early warning of low-load water dynamics of a coal-fired power plant boiler according to claim 1, characterized in that: The specific operation steps of the machine learning method described in step 5 are as follows: 1) Clean the collected historical data, remove invalid data, ensure data integrity, process outliers, and repair errors and outliers in the data; 2) Standardize data from different sources and dimensions to ensure that all parameters are within the same scale range; 3) Extract features that are helpful for prediction from the original data, as follows: Time window characteristics, such as average pressure and temperature over the past 30 minutes; Statistical characteristics, such as mean, standard deviation, maximum, minimum, skewness, and kurtosis; Lag features: based on the delay relationship of the time series, the data features of the production lag time are generated; Periodic features, extract features based on the boiler’s operating cycle; 4) If there are marked faults and abnormal records in the historical data, supervised learning methods can be used for training. Common supervised learning methods can be selected: Classification models, such as logistic regression, support vector machine, decision tree, random forest, XGBoost, and LightGBM, are used to determine whether a fault or anomaly has occurred; Regression models, such as linear regression, support-linked regression, random forest regression, and gradient boosting regression, are used to predict future boiler operating parameters; 5) If there are no anomaly records in the historical data, unsupervised learning methods can be used to detect anomalies based on the intrinsic structure of the data. Common supervised learning methods are available: Clustering algorithms, such as K-means and DBSCAN, are used to discover different patterns in the data; Isolation Forest, suitable for large-scale datasets; Principal component analysis, to find outliers by reducing dimensionality; Autoencoder, an unsupervised anomaly detection method based on neural networks, is used for anomaly identification in high-dimensional data; 6) For high-dimensional data with complex patterns, deep learning models can be used, which can be selected: Long short-term memory networks, which are used to process time series data and capture temporal dependencies; Convolutional neural networks are used to extract features from multidimensional data and are suitable for processing complex spatial models; Deep autoencoders, which are used for anomaly detection, especially in high-dimensional data; 7) Divide the dataset into training set, validation set and test set, with the ratio of 70%-15%-15% divided. The training set is used to train the model, the validation set is used to tune the model parameters, and the test set is used to evaluate the model performance. 8) Use the training set to train the selected machine learning model. For supervised learning, the model will learn the patterns in the data by minimizing the error and the error function. 9) For unsupervised learning, the model will learn how to identify unusual patterns in the data; 10) If it is a regression task, optimize the prediction error; If it is a classification task, optimize the classification accuracy; 11) Use the validation set to evaluate the performance of the model, check the accuracy, precision, recall and F1 value of the model, and for regression problems, use the mean square error, root mean square error and R 2 Evaluation indicators: For classification problems, confusion matrix, ROC curve and AUC evaluation indicators are used; 12) Use cross-validation methods to evaluate the performance of the model on different data subsets, reduce the risk of overfitting, and ensure the robustness of the model; 13) Use grid search to tune the hyperparameters of the machine learning model, adjust the model's learning rate, tree depth, and regularization parameters to improve the model's performance; 14) Optimize features based on model performance, remove irrelevant features, add new features, or use dimensionality reduction techniques to reduce redundant features; 15) To improve the accuracy and robustness of prediction, ensemble learning methods can be used; 16) Once the model predicts that the boiler's operating parameters are outside the normal range, an alarm is triggered. The alarm is triggered based on: Probability threshold, when the model's anomaly probability exceeds the threshold; Confidence threshold: when the confidence of the model is lower than a certain threshold, the prediction result is considered unreliable; 17) Once the warning is triggered, the system will notify the operator through screen display, SMS, email and voice alarm, and the operator can make appropriate intervention and processing according to the alarm information; 18) Deploy the trained model to the production environment and use containerization technology to directly integrate it into the monitoring system for real-time prediction; 19) Continuously monitor the performance of the model in actual operation. If the model shows an increase in prediction error, the model can be retrained; 20) The model should be updated regularly to maintain its accuracy as the boiler operating environment changes.

Images