Combustor swing angle control method based on combination of large language model and expert knowledge

Abstract

The present application relates to the technical field of thermal power generating units, and specifically discloses a burner swing angle control method based on a large language model combined with expert knowledge, which comprises the following steps: obtaining unit historical data, establishing an optimal working condition data set according to the unit historical data; detecting non-steady working condition data in the optimal working condition data set, cleaning the optimal working condition data set according to the non-steady working condition data, and obtaining a steady working condition data set; obtaining historical coal mill start-stop conditions, establishing a complete data set according to the historical coal mill start-stop conditions; establishing a burner swing angle model according to the complete data set, and outputting a burner swing angle value under a steady working condition according to the burner swing angle model; obtaining unit real-time data, correcting the burner swing angle value under the steady working condition according to the unit real-time data combined with expert knowledge, and obtaining a final burner swing angle value. The present application can realize accurate control of the burner swing angle, and make the combustion process more efficient and stable.

Description

Technical Field

[0001] This application relates to the field of thermal power unit technology, and more specifically, to a burner tilt angle control method based on a combination of large language models and expert knowledge. Background Technology

[0002] Burner tilt angle adjustment is a common method for steam temperature control in coal-fired boilers, especially widely used in tangential combustion boilers. Its basic principle is to change the height of the flame center in the furnace by adjusting the tilt angle of the burner nozzles, thereby changing the flue gas temperature at the furnace outlet and regulating the superheated and reheated steam temperatures. With the continuous development of industrial automation, burner tilt angle control, as a key link in industrial production, directly determines the energy efficiency and safety stability of the unit due to its level of intelligence.

[0003] Against this backdrop, a burner tilt angle decision control project based on the combination of big data and expert knowledge has emerged. This project, with its advanced technology, cost-effectiveness, and superior performance parameters, demonstrates a strong competitive advantage in the market. Summary of the Invention

[0004] This invention provides a burner tilt angle control method based on a large language model and expert knowledge, to solve the problem of precise control of burner tilt angle in existing technologies, including: The process involves: acquiring historical data from the generating unit and establishing an optimal operating condition dataset; detecting unsteady operating condition data within the optimal operating condition dataset and cleaning the dataset to obtain a steady-state operating condition dataset; acquiring historical pulverizer start-up and shutdown data and establishing a complete dataset; establishing a burner tilt angle model based on the complete dataset and outputting the burner tilt angle value under steady-state conditions; acquiring real-time data from the generating unit and correcting the burner tilt angle value under steady-state conditions based on the real-time data and expert knowledge to obtain the final burner tilt angle value.

[0005] Furthermore, an optimal operating condition dataset is established based on historical unit data, including: determining the main steam temperature and main steam pressure corresponding to each operating condition during the unit's historical operation based on historical unit data; establishing multiple historical operating condition datasets based on the main steam temperature and main steam pressure corresponding to each operating condition; obtaining the unit load corresponding to each operating condition; clustering each historical operating condition dataset based on the unit load; and determining the optimal operating condition dataset corresponding to each unit load based on the clustering results.

[0006] Furthermore, the historical operating condition datasets are clustered based on the unit load, including: establishing sample datasets based on the unit load of each operating condition; randomly selecting k initial cluster centers for the sample datasets; calculating the Manhattan distance from each sample data point to the initial cluster centers; dividing the historical operating condition datasets of the sample data into corresponding clusters based on the Manhattan distance from each sample data point to the initial cluster centers; calculating the mean of the sample data within each cluster; updating the cluster centers based on the mean of the sample data within each cluster; and repeating the clustering steps iteratively based on the new cluster centers until the cluster centers no longer change or the preset maximum number of iterations is reached, thereby obtaining the clusters of k historical operating condition datasets.

[0007] Furthermore, the optimal operating condition dataset corresponding to the load of each unit is determined based on the clustering results, including: determining the clusters of each historical operating condition dataset based on the clustering results, and calculating the combustion efficiency corresponding to each historical operating condition dataset in the clusters; selecting the historical operating condition dataset with the highest combustion efficiency in each cluster to obtain the optimal operating condition dataset corresponding to each cluster.

[0008] Furthermore, detecting unstable operating condition data in the optimal operating condition dataset includes: establishing a time window centered on any target optimal operating condition data in the optimal operating condition dataset, calculating the mean of the optimal operating condition data within the time window, calculating the instability coefficient based on the mean of the optimal operating condition data; obtaining a preset allowable threshold, filtering out target optimal operating condition data with an instability coefficient greater than the preset allowable threshold, and obtaining the unstable operating condition data.

[0009] Furthermore, the instability coefficient is calculated based on the mean of the optimal working condition data, including: calculating the absolute value of the difference between the target optimal working condition data and the mean of the corresponding optimal working condition data, and standardizing the absolute value of the difference to obtain the instability coefficient.

[0010] Furthermore, a complete dataset is established based on historical coal mill start-up and shutdown data, including: obtaining historical coal mill start-up and shutdown data and corresponding load intervals; establishing a coal mill combination library based on historical coal mill start-up and shutdown data and corresponding load intervals; obtaining the unit loads corresponding to the steady-state operating condition dataset; determining the coal mill start-up and shutdown data of the steady-state operating condition dataset based on the load intervals of the unit loads corresponding to the steady-state operating condition dataset, according to the coal mill combination library; and adding the coal mill start-up and shutdown data to the steady-state operating condition dataset to obtain the complete dataset.

[0011] Furthermore, a burner tilt angle model is established based on the complete dataset, including: establishing a training sample set based on the complete dataset; establishing a burner tilt angle model based on the training sample set and training the burner tilt angle model to obtain a trained burner tilt angle model; inputting the current unit's main steam temperature, main steam pressure, and coal mill start-up and shutdown status into the trained burner tilt angle model, and outputting the burner tilt angle value under steady-state conditions.

[0012] Furthermore, the burner tilt angle value under steady-state operating conditions is corrected based on real-time unit data and expert knowledge to obtain the final burner tilt angle value. This includes: establishing an expert knowledge base and determining the burner tilt angle correction value corresponding to the real-time data of each unit based on the expert knowledge base; matching the current real-time data of the unit with the expert knowledge in the expert knowledge base and determining the tilt angle correction value corresponding to the current real-time data of each unit based on the matching result; determining the comprehensive tilt angle correction value based on the tilt angle correction value corresponding to the current real-time data of each unit; and correcting the burner tilt angle value under steady-state operating conditions based on the comprehensive tilt angle correction value to obtain the final burner tilt angle value.

[0013] Furthermore, the comprehensive swing angle correction value is determined based on the swing angle correction values ​​corresponding to the real-time data of each unit. This includes: determining the historical changes in main steam temperature and main steam pressure and the corresponding unit efficiency based on historical data of the units; calculating the correlation coefficients between main steam temperature and main steam pressure and unit efficiency; setting the weights of main steam temperature and main steam pressure based on the correlation coefficients between main steam temperature and main steam pressure and unit efficiency; and weighting and summing the corresponding swing angle correction values ​​based on the weights of main steam temperature and main steam pressure to obtain the comprehensive swing angle correction value.

[0014] The beneficial effects of this invention are as follows: By applying the above technical solutions, this invention can achieve precise control of the burner swing angle, making the combustion process more efficient and stable. It effectively solves the technical problems in existing burner swing angle control technologies, such as coarse division of operating conditions, insufficient data cleaning, inadequate multi-factor coupling processing, and separation of data-driven and expert knowledge. It realizes intelligent optimization control of the burner swing angle under all operating conditions, and significantly improves the combustion efficiency of the boiler and the overall operating economy of the unit while ensuring the stability of the main steam temperature. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 The diagram shows a schematic flow chart of a burner tilt angle control method based on a combination of large language model and expert knowledge proposed in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] This application provides a burner tilt angle control method based on a combination of large language models and expert knowledge, such as... Figure 1 As shown, it includes: S101, acquire historical data of the unit, and establish an optimal operating condition dataset based on the historical data of the unit; In some embodiments of this application, establishing an optimal operating condition dataset based on historical unit data includes: determining the main steam temperature and main steam pressure corresponding to each operating condition during the historical operation of the unit based on historical unit data; establishing multiple historical operating condition datasets based on the main steam temperature and main steam pressure corresponding to each operating condition; obtaining the unit load corresponding to each operating condition; clustering each historical operating condition dataset based on the unit load; and determining the optimal operating condition dataset corresponding to each unit load based on the clustering results.

[0019] In this embodiment, multiple historical operating condition datasets of the unit's historical data are clustered according to load, and the historical optimal value under each load condition is obtained based on the clustering results.

[0020] In some embodiments of this application, clustering of historical operating condition datasets based on unit load includes: establishing a sample dataset based on the unit load of each operating condition; randomly selecting k initial cluster centers for the sample dataset; calculating the Manhattan distance from each sample data point to the initial cluster center; dividing the historical operating condition dataset of the sample data into corresponding clusters based on the Manhattan distance from each sample data point to the initial cluster center; calculating the mean of the sample data within each cluster; updating the cluster centers based on the mean of the sample data within each cluster; and repeating the clustering steps iteratively based on the new cluster centers until the cluster centers no longer change or the preset maximum number of iterations is reached, thereby obtaining clusters of k historical operating condition datasets.

[0021] In this embodiment, the value of k is set to 5, and K-means clustering is performed on each historical operating condition dataset according to the unit load to divide the historical operating condition dataset into 5 operating condition clusters.

[0022] In some embodiments of this application, determining the optimal operating condition dataset corresponding to the load of each unit based on the clustering results includes: determining the clusters of each historical operating condition dataset based on the clustering results, and calculating the combustion efficiency corresponding to each historical operating condition dataset in the clusters; selecting the historical operating condition dataset with the highest combustion efficiency in each cluster to obtain the optimal operating condition dataset corresponding to each cluster.

[0023] In this embodiment, the historical operating data within each cluster of operating conditions is sorted according to combustion efficiency, and the one with the highest efficiency is selected as the benchmark operating condition for that cluster, providing high-quality training samples for subsequent model training.

[0024] S102, detect the non-steady working condition data in the optimal working condition dataset, clean the optimal working condition dataset based on the non-steady working condition data, and obtain the steady-state working condition dataset. In some embodiments of this application, detecting unstable operating condition data in the optimal operating condition dataset includes: establishing a time window centered on any target optimal operating condition data in the optimal operating condition dataset, calculating the mean of the optimal operating condition data within the time window, calculating the instability coefficient based on the mean of the optimal operating condition data, obtaining a preset allowable threshold, filtering out target optimal operating condition data whose instability coefficient is greater than the preset allowable threshold, and obtaining the unstable operating condition data.

[0025] In some embodiments of this application, the calculation of the non-stability coefficient based on the mean of the optimal working condition data includes: calculating the absolute value of the difference between the target optimal working condition data and the mean of the corresponding optimal working condition data, and standardizing the absolute value of the difference to obtain the non-stability coefficient.

[0026] In this embodiment, by establishing a time window to calculate the non-steady-state coefficients of each optimal operating condition data, the non-steady-state data segments of unit operation are removed from the optimal operating condition dataset, ensuring that the operating condition data used for model training are all steady-state operating data, thereby avoiding interference from non-steady-state data to the model. S103, Obtain historical coal mill start-up and shutdown information, and establish a complete dataset based on the historical coal mill start-up and shutdown information; In some embodiments of this application, a complete dataset is established based on historical coal mill start-up and shutdown data, including: obtaining historical coal mill start-up and shutdown data and corresponding load intervals; establishing a coal mill combination library based on historical coal mill start-up and shutdown data and corresponding load intervals; obtaining the unit load corresponding to the steady-state operating condition dataset; determining the coal mill start-up and shutdown data of the steady-state operating condition dataset based on the load interval of the unit load corresponding to the steady-state operating condition dataset according to the coal mill combination library; and adding the coal mill start-up and shutdown data to the steady-state operating condition dataset to obtain a complete dataset.

[0027] In this embodiment, different coal mills will be activated under different load segments, and the processing conditions of different coal mill combinations are also different. Moreover, the start-up and shutdown of coal mills are key factors affecting the position of the flame center in the furnace and the adjustment effect of the burner swing angle. By associating and fusing coal mill combination information with steady-state operating condition data, the training dataset contains more comprehensive operating condition feature dimensions.

[0028] S104, Establish a burner tilt angle model based on the complete dataset, and output the burner tilt angle value under steady-state conditions based on the burner tilt angle model; In some embodiments of this application, establishing a burner tilt angle model based on a complete dataset includes: establishing a training sample set based on the complete dataset; establishing a burner tilt angle model based on the training sample set and training the burner tilt angle model to obtain a trained burner tilt angle model; inputting the current unit's main steam temperature, main steam pressure, and coal mill start-up and shutdown status into the trained burner tilt angle model, and outputting the burner tilt angle value under steady-state conditions.

[0029] In this embodiment, the burner swing angle model adopts a large language model architecture. Its input layer receives feature vectors of three dimensions: main steam temperature, main steam pressure, and coal mill start-up and shutdown status. After processing by the model's multi-layer self-attention mechanism and feedforward neural network, the output layer outputs the burner swing angle value that best matches the current operating conditions.

[0030] S105: Obtain real-time data from the unit, and correct the burner tilt angle value under steady-state conditions based on the real-time data and expert knowledge to obtain the final burner tilt angle value.

[0031] In some embodiments of this application, the burner tilt angle value under steady-state operating conditions is corrected based on real-time unit data and expert knowledge to obtain the final burner tilt angle value. This includes: establishing an expert knowledge base; determining the burner tilt angle correction value corresponding to the real-time data of each unit based on the expert knowledge base; matching the current real-time data of the unit with the expert knowledge in the expert knowledge base; determining the tilt angle correction value corresponding to the current real-time data of each unit based on the matching result; determining a comprehensive tilt angle correction value based on the tilt angle correction value corresponding to the current real-time data of each unit; and correcting the burner tilt angle value under steady-state operating conditions based on the comprehensive tilt angle correction value to obtain the final burner tilt angle value.

[0032] In this embodiment, the expert knowledge specifically refers to the following: Regarding main steam temperature, when the steam temperature fluctuates, it is divided into four segments based on the actual operating pattern of the unit, representing excessively high, relatively high, relatively low, and excessively low main steam temperatures, respectively. Regarding main steam pressure, fluctuations are divided into three segments, representing excessively high, relatively high, and relatively low main steam pressure, respectively. When the main steam temperature is excessively high, the swivel angle moves downward by a certain angle, and vice versa; when the main steam pressure is excessively high, the swivel angle moves upward by a certain angle, and vice versa. By introducing expert knowledge, a comprehensive swivel angle correction value based on the current real-time data of the unit is determined to correct the current swivel angle value and obtain the optimal burner swivel angle value.

[0033] In some embodiments of this application, determining the comprehensive swing angle correction value based on the swing angle correction value corresponding to the real-time data of each unit includes: determining the historical changes in main steam temperature and main steam pressure and the corresponding unit efficiency based on historical data of the units, calculating the correlation coefficients of main steam temperature and main steam pressure with unit efficiency respectively; setting the weight of main steam temperature and main steam pressure based on the correlation coefficients of main steam temperature and main steam pressure with unit efficiency, and performing a weighted summation of the corresponding swing angle correction values ​​based on the weight of main steam temperature and main steam pressure to obtain the comprehensive swing angle correction value.

[0034] In this embodiment, the weight allocation of each correction factor is determined by analyzing the correlation coefficients between the main steam temperature and the main steam pressure and the unit efficiency, making the calculation of the comprehensive swing angle correction value more scientific and reasonable, and improving the final unit efficiency.

[0035] By applying the above technical solutions, this invention acquires historical data from the generating unit and establishes an optimal operating condition dataset based on this data; detects unsteady operating condition data in the optimal operating condition dataset, cleans the dataset based on this data, and obtains a steady-state operating condition dataset; acquires historical pulverizer start-up and shutdown data, and establishes a complete dataset based on this data; establishes a burner tilt angle model based on the complete dataset, and outputs the burner tilt angle value under steady-state conditions based on the model; acquires real-time data from the generating unit, and corrects the burner tilt angle value under steady-state conditions based on this data and expert knowledge, obtaining the final burner tilt angle value. This enables precise control of the burner tilt angle, making the combustion process more efficient and stable.

[0036] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented in hardware or by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A burner tilt angle control method based on a combination of large language models and expert knowledge, characterized in that, include: Acquire historical data of the generating unit and establish an optimal operating condition dataset based on the historical data of the generating unit; Detect the non-steady-state data in the optimal working condition dataset, clean the optimal working condition dataset based on the non-steady-state data, and obtain the steady-state working condition dataset. Obtain historical coal mill start-up and shutdown information, and build a complete dataset based on the historical coal mill start-up and shutdown information; A burner swing angle model is established based on the complete dataset, and the burner swing angle value under steady-state conditions is output based on the burner swing angle model. The unit acquires real-time data, and based on this data and expert knowledge, corrects the burner tilt angle value under steady-state conditions to obtain the final burner tilt angle value.

2. The burner tilt angle control method based on a large language model and expert knowledge as described in claim 1, characterized in that, An optimal operating condition dataset is established based on historical unit data, including: Based on the historical data of the unit, determine the main steam temperature and main steam pressure corresponding to each operating condition during the historical operation of the unit, and establish multiple historical operating condition datasets based on the main steam temperature and main steam pressure corresponding to each operating condition. Obtain the unit load corresponding to each operating condition, and cluster the historical operating condition datasets based on the unit load; The optimal operating condition dataset corresponding to the load of each unit is determined based on the clustering results.

3. The burner tilt angle control method based on a combination of large language models and expert knowledge according to claim 2, characterized in that, Clustering of historical operating condition datasets based on unit load includes: A sample dataset is established based on the unit load under various operating conditions, and k initial cluster centers of the sample dataset are randomly selected. Calculate the Manhattan distance from each sample data point to the initial cluster center, and divide the historical working condition dataset of the sample data into the corresponding clusters based on the Manhattan distance from each sample data point to the initial cluster center; Calculate the mean of the sample data within each cluster, and update the cluster centers based on the mean of the sample data within each cluster; Repeat the clustering steps based on the new cluster centers until the cluster centers no longer change or the preset maximum number of iterations is reached, to obtain clusters of k historical working condition datasets.

4. The burner tilt angle control method based on a combination of large language models and expert knowledge according to claim 2, characterized in that, The optimal operating condition dataset corresponding to the load of each unit is determined based on the clustering results, including: Based on the clustering results, determine the clusters of each historical operating condition dataset, and calculate the combustion efficiency corresponding to each historical operating condition dataset in the cluster. The historical operating condition datasets with the highest combustion efficiency in each cluster were selected to obtain the optimal operating condition datasets for each cluster.

5. The burner tilt angle control method based on a combination of large language models and expert knowledge according to claim 1, characterized in that, Detecting non-steady-state data in the optimal operating condition dataset, including: Establish a time window centered on any target optimal working condition data within the optimal working condition dataset, calculate the mean of the optimal working condition data within the time window, and calculate the instability coefficient based on the mean of the optimal working condition data; Obtain a preset allowable threshold, filter out the target optimal working condition data whose instability coefficient is greater than the preset allowable threshold, and obtain the instability working condition data.

6. The burner tilt angle control method based on a combination of large language models and expert knowledge according to claim 5, characterized in that, The instability coefficient is calculated based on the mean of the optimal operating condition data, including: Calculate the absolute value of the difference between the target optimal working condition data and the mean of the corresponding optimal working condition data, and standardize the absolute value of the difference to obtain the instability coefficient.

7. The burner tilt angle control method based on a large language model and expert knowledge as described in claim 1, characterized in that, A complete dataset was created based on historical coal mill start-up and shutdown data, including: Obtain historical coal mill start-up and shutdown information and corresponding load ranges, and establish a coal mill combination library based on historical coal mill start-up and shutdown information and corresponding load ranges; Obtain the unit load corresponding to the steady-state operating condition dataset, and determine the start-up and shutdown status of the coal mill based on the load range of the unit load corresponding to the steady-state operating condition dataset according to the coal mill combination library; Add the start-up and shutdown status of the coal mill to the steady-state operating condition dataset to obtain the complete dataset.

8. The burner tilt angle control method based on the combination of large language model and expert knowledge according to claim 1, characterized in that, A burner sway angle model is built based on the complete dataset, including: A training sample set is established based on the complete dataset. A burner sway angle model is established based on the training sample set and trained to obtain a trained burner sway angle model. Input the current main steam temperature, main steam pressure, and coal mill start-up and shutdown status of the unit into the trained burner tilt angle model, and output the burner tilt angle value under steady-state conditions.

9. The burner tilt angle control method based on a large language model and expert knowledge as described in claim 1, characterized in that, Based on real-time unit data and expert knowledge, the burner tilt angle value under steady-state operating conditions is corrected to obtain the final burner tilt angle value, including: Establish an expert knowledge base and determine the burner tilt angle correction value corresponding to the real-time data of each unit based on the expert knowledge base; The current real-time data of the generating units is matched with the expert knowledge in the expert knowledge base, and the swing angle correction value corresponding to the current real-time data of each generating unit is determined based on the matching results. The comprehensive swing angle correction value is determined based on the swing angle correction value corresponding to the real-time data of each unit. The burner swing angle value under steady-state conditions is then corrected based on the comprehensive swing angle correction value to obtain the final burner swing angle value.

10. The burner tilt angle control method based on the combination of large language model and expert knowledge according to claim 9, characterized in that, The comprehensive swing angle correction value is determined based on the swing angle correction values ​​corresponding to the current real-time data of each unit, including: Based on historical data of the unit, determine the historical changes in main steam temperature and main steam pressure and the corresponding unit efficiency, and calculate the correlation coefficients between main steam temperature and main steam pressure and unit efficiency respectively. The main steam temperature weight and main steam pressure weight are set according to the correlation coefficient between the main steam temperature and main steam pressure and the unit efficiency. The corresponding swing angle correction values ​​are weighted and summed according to the main steam temperature weight and main steam pressure weight to obtain the comprehensive swing angle correction value.

Images