A method and system for warning of combustion power and gas emissions of a water-cooled vibrating grate boiler

Abstract

The present invention relates to the field of power digitization technology, and provides a method and system for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler. The method of the present invention includes: obtaining real-time combustion data and storing the obtained real-time combustion data in a mysql database; exporting the real-time combustion data from the mysql database, and obtaining training feature data through data processing; calculating the rise and fall value of the target variable 5 minutes later in the training feature data, judging the rise and fall situation, and labeling the real-time combustion data according to the rise and fall situation; dividing the labeled real-time combustion data to obtain a training set and a test set, training, establishing and screening to obtain an AI three-class rise and fall prediction model; predicting the rise and fall situation of the target variable 5 minutes later at the current moment, and triggering a warning according to the prediction result. The method and system of the present invention can predict the rise and fall situation of combustion variables 5 minutes later, enabling the crew to issue instructions to adjust combustion variables in advance.

Description

Technical Field

[0001] The present invention relates to the technical field of power digitization, and particularly to a method and system for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler. Background Art

[0002] China is a large agricultural country with rich biomass resources. Using biomass fuel for power generation and implementing coal substitution can significantly reduce carbon emissions and bring huge economic and environmental benefits. Agricultural and forestry biomass fuels usually include various straws, wood chips, etc., which have the characteristics of low energy density, high chlorine, sodium, and potassium element content, and high volatile matter. However, compared with traditional coal power generation, agricultural and forestry biomass fuels are more likely to cause problems such as unstable combustion, frequent fluctuations in unit power, and unstable emissions of polluting gases.

[0003] Water-cooled vibrating grate boilers are currently widely used in the field of agricultural and forestry biomass power generation. Nowadays, the main means to realize the operation of biomass power generation units is to manually control the combustion variables of the boiler (such as air volume, fuel volume, etc.). The unit personnel set the fuel quantity command according to past boiler combustion experience and the target unit power of the day. When the unit load is lower than the target set power, the fuel quantity is appropriately increased; when the unit power is close to the target load, the current fuel quantity is maintained; when the unit power is greater than the target load, the fuel quantity is reduced. For the emissions of sulfur dioxide and nitrogen oxides after combustion, the unit personnel need to analyze the current gas concentration and the numerical values of combustion variables, predict the gas concentration 5 minutes later, and then adjust the two polluting gases by adding slaked lime and urea. In practical applications, there are the following problems with manual control of combustion variables:

[0004] 1) The real-time performance and accuracy of unit power adjustment are poor. During the process of unit power adjustment, when a manual instruction to change the fuel quantity is issued, there will be an obvious time delay from the fuel leaving the silo, entering the boiler for combustion, to the change in unit power. The adjustment real-time performance and accuracy are not high, and extremely high requirements are placed on the experience of operators.

[0005] 2) The real-time performance and accuracy of polluting gas emission control are poor. During the process of polluting gas emission control, there is a problem of instruction delay, and the relevant combustion variables are complex and difficult to analyze.

[0006] Therefore, how to provide a method for warning the power and gas emissions applicable to a water-cooled vibrating grate boiler has become an urgent technical problem to be solved. Summary of the Invention

[0007] In view of this, the present invention provides a model method and device for power warning and pollution gas (sulfur dioxide, nitrogen oxides) concentration warning of a water-cooled vibrating grate boiler unit, aiming to predict the rise and fall of the power of the water-cooled vibrating grate boiler unit and the concentration of pollution gases, and further solve the problem of delayed response in manually adjusting combustion variables.

[0008] On the one hand, the present invention provides a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler, including:

[0009] Step S1: Obtain real-time combustion data and store the obtained real-time combustion data in a mysql database. The real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp.

[0010] Step S2: Export the real-time combustion data from the mysql database, delete invalid data, and obtain training feature data by calculating the minute mean and minute difference of each combustion variable.

[0011] Step S3: Judge the rise and fall situation by calculating the rise and fall value of the target variable 5 minutes later in the training feature data obtained in Step S2, and label the real-time combustion data according to the rise and fall situation to obtain the label of each piece of data.

[0012] Step S4: Divide the real-time combustion data labeled in Step S3 to obtain a training set and a test set, and train, establish, and screen to obtain an AI three-class rise and fall prediction model.

[0013] Step S5: Use the rise and fall prediction model screened in Step S4 to predict the rise and fall situation of the target variable 5 minutes later at the current moment, and trigger a warning according to the prediction result.

[0014] Further, Step S1 of the method of the present invention includes: By installing sensors at various positions of the boiler, obtain real-time data of each combustion index, that is, real-time combustion data. The sensors monitor and record the corresponding combustion variables every second. The combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration.

[0015] Further, in Step S2 of the method of the present invention, calculating the minute mean and minute difference of each combustion variable includes: calculating the 1-minute data mean of each combustion variable and calculating the means of each combustion variable at 2 minutes, 5 minutes, and 8 minutes respectively.

[0016] Further, in step S3 of the method of the present invention, the rise and fall situation is judged, and the real-time combustion data is labeled according to the rise and fall situation, including: setting a rise and fall threshold a. If the rise and fall value is greater than a, it is judged as a rise, and the data label is set to 1; if the rise and fall value is less than -a, it is judged as a fall, and the data label is set to -1; if the rise and fall value is between -a and a, it is judged as flat, and the data label is set to 0.

[0017] Further, in step S3 of the method of the present invention, when judging the rise and fall situation and labeling the real-time combustion data according to the rise and fall situation, it further includes: sorting the rise and fall values calculated after 5 minutes in descending order. The data with a value in the top 30% and greater than 0 is judged as a rise, and the data label is set to 1; the data with a value in the bottom 30% and less than 0 is judged as a fall, and the data label is set to -1; the remaining data is judged as flat, and the data label is set to 0.

[0018] Further, step S4 of the method of the present invention includes:

[0019] Step S41: Obtain data within a certain time range from the real-time combustion data labeled in step S3 with days as the time unit;

[0020] Step S42: Randomly select data for a certain number of days as the training set from the data within the certain time range in step S41, and the data for the remaining days as the test set;

[0021] Step S43: Train, establish, and screen an AI three-classification prediction model according to the training set and test set obtained in step S42.

[0022] Further, step S43 of the method of the present invention includes:

[0023] Step S431: Based on the training set and test set in step S42, establish Random Forest, Adaboost, Gradient Boosting, and Bagging ensemble models:

[0024] Step S432: Use the same batch of training set and test set to train the 4 models in step S431 at the same time;

[0025] Step S433: Select the model with the highest accuracy from the 4 trained models as the final prediction model for deployment.

[0026] Further, step S5 of the method of the present invention includes: obtaining the importance factor of each combustion variable by calculating the average contribution degree of all combustion variables to the prediction result, and outputting the variables in descending order according to the importance factor.

[0027] On the other hand, the present invention provides a water-cooled vibrating grate boiler combustion power and gas emission warning system, including:

[0028] A data collection module, which is used to obtain real-time combustion data in the sensor and store the obtained real-time combustion data in a MySQL database. The real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp. The combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration;

[0029] A data processing module, which is used to export real-time combustion data from the MySQL database, delete invalid data, obtain training feature data by calculating the minute average value and minute difference value of each combustion variable; judge the increase or decrease situation by calculating the increase or decrease value of the target variable 5 minutes later in the training feature data, and label the real-time combustion data according to the increase or decrease situation to obtain the label of each piece of data;

[0030] A model training, establishment and screening module, which is used to divide the labeled real-time combustion data to obtain a training set and a test set, train, establish and screen an AI three-class increase or decrease prediction model, and package the screened AI three-class increase or decrease prediction model into an increase or decrease prediction model service interface;

[0031] An increase or decrease prediction model service interface, which is used to predict the increase or decrease situation of the target variable 5 minutes after the current moment, and send the increase or decrease prediction result of the unit power or gas concentration 5 minutes after the current moment and the top 20 combustion variables ranked by importance factors to the early warning module;

[0032] An early warning module, which is used to make an instruction judgment according to the sending result of the increase or decrease prediction model service interface, convert "rise" and "fall" into corresponding alarm instructions, and instruct the device to send out an early warning signal.

[0033] Further, in the combustion power and gas emission early warning system of the water-cooled vibrating grate boiler of the present invention, the data collection module first opens up a section of cache when collecting data. Each section of cache only stores 60 seconds of data. If the timestamp of the current data is the start moment of each minute, that is, 0 seconds, this piece of data is put into the cache area; if the timestamp of the current data is the last moment of each minute, that is, 59 seconds, then the 60 pieces of data within 1 minute in the cache area are sent to the data processing module to calculate the minute average value and minute difference value.

[0034] The combustion power and gas emission early warning method and system of the water-cooled vibrating grate boiler of the present invention have the following beneficial effects: it can predict the increase or decrease of the power and the increase or decrease of the pollution gas concentration of the boiler unit 5 minutes later, enable the unit operators to make instructions to adjust the combustion variables in advance, avoid delayed response, and thus reduce the labor cost and improve the safety and economy of the boiler. Brief Description of the Drawings

[0035] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0036] Figure 1 It is a flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to an exemplary first embodiment of the present invention.

[0037] Figure 2 It is a flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to an exemplary third embodiment of the present invention.

[0038] Figure 3 It is a flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to an exemplary fourth embodiment of the present invention.

[0039] Figure 4 It is an architecture diagram of a system for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to an exemplary sixth embodiment of the present invention. Detailed implementation manners

[0040] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0041] It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other; and, based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present disclosure.

[0042] It should be noted that the following describes various aspects of embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on the present disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. Additionally, this device and / or practice this method can be implemented using other structures and / or functionality in addition to one or more of the aspects described herein.

[0043] The following are the explanations of the terms involved in each of the following embodiments:

[0044] Random Forest Ensemble Model: Random Forest is an important model in the field of machine learning. Due to its high accuracy and interpretability, this model is widely used in classification and regression problems.

[0045] Adaboost Ensemble Model: Adaptive Boosting Ensemble Algorithm (Adaboost), whose core idea is to train different weak classifiers through an iterative method and then integrate them to form a strong classifier with better performance.

[0046] Gradient Boosting Ensemble Model: Gradient Boosting Algorithm. This method uses decision trees as weak classifiers, and each subtree is independent; this method uses the method of fitting gradients to obtain the optimal parameters.

[0047] Bagging Ensemble Model: The Bagging algorithm is similar to the Random Forest method. It is based on the bootstrap sampling method and trains models on different datasets to reduce variance. This method is applicable to classification problems with relatively small data bias.

[0048] Figure 1 It is a flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to an exemplary first embodiment of the present invention. As Figure 1 shown, the method of this embodiment includes:

[0049] Step S1: Obtain real-time combustion data and store the obtained real-time combustion data in the mysql database. The real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp.

[0050] Step S2: Export the real-time combustion data from the mysql database, delete invalid data, and obtain training feature data by calculating the minute mean and minute difference of each combustion variable.

[0051] Step S3: Judge the rise and fall situation by calculating the rise and fall value of the target variable 5 minutes later in the training feature data obtained in Step S2, and label the real-time combustion data according to the rise and fall situation to obtain the label of each data.

[0052] Step S4: Divide the real-time combustion data labeled in Step S3 to obtain a training set and a test set, and train, establish, and screen to obtain an AI three-class rise and fall prediction model.

[0053] Step S5: Use the rise and fall prediction model screened in Step S4 to predict the rise and fall situation of the target variable 5 minutes later at the current moment, and trigger a warning according to the prediction result.

[0054] In the method of this embodiment, sensors are installed at various positions of the boiler to obtain real-time data of various combustion indicators, namely real-time combustion data. The sensors monitor and record the corresponding combustion variables every second. The combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration.

[0055] In the method of this embodiment, the real-time combustion data collected by the sensors is stored in the mysql database, and the mysql data table is automatically exported as a csv file through a script program. All calculations on the data are performed on the csv file, which makes the subsequent data processing process safer and more convenient.

[0056] In the method of this embodiment, the reason for setting the deletion of invalid data is that the reason for generating invalid data is the instability of the boiler sensors, which may result in monitoring failures or abnormal acquisitions, etc., causing the recorded values to be null or infinite (small). Such data will affect the training of the AI model and needs to be deleted.

[0057] In the method of this embodiment, calculating the minute average value and minute difference value of each combustion variable includes: calculating the data average value of each combustion variable for 1 minute, and calculating the average values of each combustion variable for 2 minutes, 5 minutes, and 8 minutes respectively.

[0058] In practical applications, the time unit for the sensors to collect data is seconds. For most combustion variables, the change range between each second is very small. In order to more accurately capture their data characteristics, the method of this embodiment calculates the data average value of each combustion variable for 1 minute. In order to better capture the change law of the combustion data, the method of this embodiment calculates the average values of each combustion variable for 2 minutes, 5 minutes, and 8 minutes respectively, and uses them together with the 1-minute average value data as the training features of the AI model.

[0059] The exemplary second embodiment of the present invention provides a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler. This embodiment is Figure 1 a preferred embodiment of the method shown.

[0060] In step S3 of the method of this embodiment, judging the rise and fall situation and labeling the real-time combustion data according to the rise and fall situation includes: setting a rise and fall threshold a. If the rise and fall value is greater than a, it is judged as a rise, and the data label is set to 1; if the rise and fall value is less than -a, it is judged as a fall, and the data label is set to -1; if the rise and fall value is between -a and a, it is judged as flat, and the data label is set to 0.

[0061] In step S3 of the method of this embodiment, the rise and fall situation is judged, and the real-time combustion data is labeled according to the rise and fall situation, which further includes: sorting the calculated rise and fall values after 5 minutes in descending order, and judging the data in the top 30% and greater than 0 as rising, and setting the data label to 1; judging the data in the bottom 30% and less than 0 as falling, and setting the data label to -1; judging the remaining data as flat, and setting the data label to 0.

[0062] In the method of this embodiment, the target variables are the unit power and the concentration of harmful gases; the rise and fall value of the target variable after 5 minutes is the difference between the value after 5 minutes and the current moment.

[0063] The method of this embodiment can set different rise and fall thresholds or percentages according to the characteristics of different combustion variables.

[0064] Figure 2 As shown in the flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to the exemplary third embodiment of the present invention, this embodiment is Figure 1 a preferred embodiment of the method shown. As Figure 2 shown, step S4 of the method of this embodiment includes:

[0065] Step S41: Obtain data within a certain time range from the real-time combustion data labeled in step S3 in units of days;

[0066] Step S42: Randomly select data for a certain number of days as the training set from the data within a certain time range in step S41, and the data for the remaining days as the test set;

[0067] Step S43: Train, establish and screen an AI three-classification prediction model according to the training set and test set obtained in step S42.

[0068] In practical applications, the method of this embodiment divides data in units of days, and the unit personnel can set the number of days for division by themselves. For example, set the time range to 30 days, with 25 days for the training set and 5 days for the test set. Then obtain 30 days of data, and randomly select 25 days of data as the training set and 5 days as the test set according to the time stamp. Then train the AI three-classification rise and fall prediction model to predict the rise and fall situation of the target variable 5 minutes after the current moment.

[0069] Figure 3 As shown in the flowchart of a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to the exemplary fourth embodiment of the present invention, this embodiment is Figure 2 a preferred embodiment of the method shown. As Figure 3 shown, step S43 of the method of this embodiment includes:

[0070] Step S431: Based on the training set and test set in Step S42, establish Random Forest, Adaboost, Gradient Boosting, and Bagging ensemble models:

[0071] Step S432: Use the same batch of training set and test set to train the 4 models in Step S431 simultaneously;

[0072] Step S433: Select the model with the highest accuracy from the 4 trained models as the final prediction model for deployment.

[0073] The fifth embodiment of the present invention provides a method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler. This embodiment is Figure 1 a preferred embodiment of the method shown. Step S5 of the method in this embodiment includes: obtaining the importance factors of each combustion variable by calculating the average contribution of all combustion variables to the prediction result, and outputting them after sorting the variables in descending order according to the importance factors.

[0074] In practical applications, taking the unit power as an example, when the prediction result of the rise and fall prediction model is a rise, it means that the unit power 5 minutes later will be higher than the safety value. At this time, an over-load alarm will be triggered to prompt the unit personnel to reduce the boiler fuel quantity, air volume, oxygen content, etc. in advance; when the model prediction result is a fall, it means that the unit power is too low 5 minutes later. At this time, an under-load alarm will be triggered to prompt the unit personnel to increase the fuel quantity, air volume, and oxygen content in advance. Through the above warning mechanism, the negative impact caused by delayed response can be greatly reduced, the unit operation can be made stable, and the human cost of boiler operation and maintenance can be reduced.

[0075] The feature importance ranking can intuitively show the unit personnel the combustion variables most relevant to the current target variable.

[0076] Figure 4 The architecture diagram of a water-cooled vibrating grate boiler combustion power and gas emission warning system according to the sixth exemplary embodiment of the present invention is as Figure 4 shown. The system of this embodiment includes:

[0077] A data collection module, configured to obtain real-time combustion data in the sensor and store the obtained real-time combustion data in a mysql database, where the real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp, and the combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration;

[0078] A data processing module, which is used to export real-time combustion data from a MySQL database, delete invalid data, calculate the minute average value and minute difference of each combustion variable to obtain training feature data; calculate the increase or decrease value of the target variable 5 minutes later in the training feature data, judge the increase or decrease situation, and label the real-time combustion data according to the increase or decrease situation to obtain the label of each piece of data.

[0079] A model training, establishment and screening module, which is used to divide the labeled real-time combustion data to obtain a training set and a test set, train, establish and screen to obtain an AI three-class increase or decrease prediction model, and encapsulate the screened AI three-class increase or decrease prediction model into an increase or decrease prediction model service interface.

[0080] The increase or decrease prediction model service interface is used to predict the increase or decrease situation of the target variable 5 minutes later at the current moment, and send the increase or decrease prediction result of the unit power or gas concentration 5 minutes later at the current moment and the top 20 combustion variables in terms of importance factor ranking to the warning module.

[0081] The warning module is used to make an instruction judgment according to the sending result of the increase or decrease prediction model service interface, convert "rise" and "fall" into corresponding alarm instructions, and instruct the device to send out a warning signal.

[0082] In the system of this embodiment, when the data collection module collects data, it first opens up a section of cache, and each section of cache only stores 60 seconds of data. If the timestamp of the current data is the start moment of each minute, that is, 0 second, this piece of data is put into the cache area; if the timestamp of the current data is the last moment of each minute, that is, 59 seconds, then the 60 pieces of data within 1 minute in the cache area are sent to the data processing module to calculate the minute average value and minute difference.

[0083] As mentioned above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler, characterized in that, The method includes: Step S1: Obtain real-time combustion data and store the obtained real-time combustion data in a MySQL database. The real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp. Step S2: Export the real-time combustion data from the MySQL database, delete invalid data, and obtain training feature data by calculating the minute mean and minute difference of each combustion variable. Step S3: Determine the increase or decrease situation by calculating the increase or decrease value of the target variable 5 minutes later in the training feature data obtained in Step S2, and label the real-time combustion data according to the increase or decrease situation to obtain the label of each piece of data. Step S4: Divide the real-time combustion data labeled in Step S3 to obtain a training set and a test set, and train, establish, and screen to obtain an AI three-class increase or decrease prediction model. Step S5: Use the increase or decrease prediction model screened in Step S4 to predict the increase or decrease situation of the target variable 5 minutes later at the current moment, and trigger an alarm according to the prediction result.

2. The combustion power and gas emission warning method for a water-cooled vibrating grate boiler according to claim 1, characterized in that Step S1 includes: Obtain real-time data of each combustion index, i.e., real-time combustion data, by installing sensors at various positions of the boiler. The sensors monitor and record the corresponding combustion variables every second. The combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration.

3. The method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to claim 1, characterized in that, In Step S2, calculating the minute mean and minute difference of each combustion variable includes: Calculating the data mean of each combustion variable for 1 minute, and respectively calculating the means of each combustion variable for 2 minutes, 5 minutes, and 8 minutes.

4. The combustion power and gas emission warning method for a water-cooled vibrating grate boiler according to claim 1, characterized in that In Step S3, determining the increase or decrease situation and labeling the real-time combustion data according to the increase or decrease situation includes: Set an increase or decrease threshold a. If the increase or decrease value is greater than a, it is determined to be an increase, and the data label is set to 1; if the increase or decrease value is less than -a, it is determined to be a decrease, and the data label is set to -1; if the increase or decrease value is between -a and a, it is determined to be flat, and the data label is set to 0.

5. The method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to claim 1, characterized in that, In Step S3, determining the increase or decrease situation and labeling the real-time combustion data according to the increase or decrease situation further includes: Sort the calculated increase or decrease values 5 minutes later in descending order. The data in the top 30% and greater than 0 are determined to be an increase, and the data label is set to 1; the data in the bottom 30% and less than 0 are determined to be a decrease, and the data label is set to -1; the remaining data are determined to be flat, and the data label is set to 0.

6. The method for warning of combustion power and gas emissions of a water-cooled vibrating grate boiler according to claim 1, characterized in that Step S4 includes: Step S41: Obtain data within a certain time range from the real-time combustion data labeled in Step S3 in units of days. Step S42: Randomly select data for a certain number of days as the training set from the data within the certain time range in Step S41, and the data for the remaining days as the test set. Step S43: Train, establish, and screen to obtain an AI three-class prediction model according to the training set and test set obtained in Step S42.

7. The method for warning of combustion power and gas emissions of a water-cooled vibrating grate boiler according to claim 6, characterized in that, Step S43 of the method in this embodiment includes: Step S431: Based on the training set and test set in Step S42, establish Random Forest, Adaboost, GradientBoosting, and Bagging ensemble models: Step S432: Simultaneously train the four models in Step S431 using the training set and test set of the same batch; Step S433: Select the model with the highest accuracy from the four trained models as the final prediction model for deployment.

8. The method for warning the combustion power and gas emissions of a water-cooled vibrating grate boiler according to claim 1, characterized in that, Step S5 includes: obtaining the importance factors of each combustion variable by calculating the average contribution of all combustion variables to the prediction result, and outputting them after sorting the variables in descending order according to the importance factors.

9. A combustion power and gas emission warning system for a water-cooled vibrating grate boiler, characterized in that, The system includes: A data collection module, which is used to obtain real-time combustion data from sensors and store the obtained real-time combustion data in a mysql database. The real-time combustion data includes a timestamp and the values of all combustion variables at the current timestamp. The combustion variables include unit power, current fuel quantity, air volume, main steam pressure, and current harmful gas concentration; A data processing module, which is used to export real-time combustion data from the mysql database, delete invalid data, obtain training feature data by calculating the minute mean and minute difference of each combustion variable; judge the rise and fall situation by calculating the rise and fall value of the target variable 5 minutes later in the training feature data, and label the real-time combustion data according to the rise and fall situation to obtain the label of each piece of data; A model training establishment and screening module, which is used to divide the labeled real-time combustion data to obtain a training set and a test set, train, establish and screen an AI three-class rise and fall prediction model, and package the screened AI three-class rise and fall prediction model into a rise and fall prediction model service interface; A rise and fall prediction model service interface, which is used to predict the rise and fall situation of the target variable 5 minutes after the current moment, and send the rise and fall prediction results of the unit power or gas concentration 5 minutes after the current moment and the top 20 combustion variables ranked by importance factors to the warning module; A warning module, which is used to make an instruction judgment according to the sending result of the rise and fall prediction model service interface, convert "rise" and "fall" into corresponding alarm instructions, and instruct the device to send a warning signal.

10. The combustion power and gas emission warning system for a water-cooled vibrating grate boiler according to claim 9, characterized in that, When collecting data, the data collection module first opens up a section of cache. Each section of cache only stores 60 seconds of data. If the timestamp of the current data is the start moment of each minute, that is, 0 seconds, this piece of data is put into the cache area; if the timestamp of the current data is the last moment of each minute, that is, 59 seconds, then the 60 pieces of data within 1 minute in the cache area are sent to the data processing module to calculate the minute mean and minute difference.

Images