CN-122014425-A - Online regulation and control method for combustion stability of gas turbine based on deep learning

CN122014425ACN 122014425 ACN122014425 ACN 122014425ACN-122014425-A

Abstract

The invention discloses a deep learning-based on-line regulation and control method for combustion stability of a gas turbine, and aims to solve the problems of combustion instability prevention and control under a scene of variable fuel components/heat values and emission constraint. The method comprises the steps of firstly fusing unit load, fuel characteristics and combustion chamber dynamic pressure to construct a ten-dimensional fuel characteristic package and a twenty-zero forty-eight-dimensional pressure sequence, then outputting self-adaptive stable allowance aligned across fuels by combining an adaptive stable allowance estimation network comprising a stable judging trunk and a fuel condition calibration layer and on-line calibration of a comparison event, then matching a risk grade and a threshold value group, screening regulation and control actions in a discharge qualified adjustable range, calculating amplitude, generating a regulation and control instruction draft, and finally outputting a final regulation and control instruction or a rollback instruction according to stable response quantity through closed loop verification. The invention realizes the on-line self-adaptive evaluation of the stability margin and closed-loop regulation and control under the emission constraint, effectively inhibits the risk of combustion instability, and improves the running safety and continuity of the unit.

Inventors

Zong Jiqi
SONG WEIYI
Chen Xuexinnan
Ai Rongshen
WEI JIAZHENG
SUN MINGFENG
YU HAIYANG
YU WENLONG
Xu baichuan

Assignees

华能南京燃机发电有限公司

Dates

Publication Date: 20260512
Application Date: 20260206

Claims (10)

1. The on-line regulation and control method for the combustion stability of the gas turbine based on deep learning is characterized by comprising the following steps of: s1, acquiring unit load, fuel components, heat value and dynamic pressure of a combustion chamber, retrieving a fuel component, heat value database and dynamic pressure historic library, and fusing to generate a fuel characteristic package under the current working condition; S2, searching a stability/instability comparison event based on a fuel feature packet under the current working condition, inputting an adaptive stability margin evaluation network during operation, calibrating a fuel condition calibration layer on line, and outputting an adaptive stability margin; S3, determining a risk level and generating a risk threshold group containing triggering, releasing and upgrading thresholds according to event labels of the self-adaptive stability allowance matched combustion stability event database; S4, defining an adjustable range of qualified emission in an emission-working condition mapping matrix, adjusting a corresponding adjusting action group according to a risk level to obtain an adjusting action, calculating an adjusting amplitude by combining an adaptive stability margin and a risk threshold value group, and combining to generate an adjusting instruction draft; S5, executing a regulation instruction draft, collecting dynamic pressure of a combustion chamber after execution, inputting the dynamic pressure of the combustion chamber after execution and the fuel feature pack of the current working condition into an evaluation network, and calculating and outputting stable response; and S6, when the stable response quantity reaches a release threshold value, outputting the regulation instruction draft as an online regulation instruction, and when the stable response quantity reaches an upgrade threshold value, generating and outputting a rollback regulation instruction as the online regulation instruction under the constraint of the acceptable adjustable range of emission.
2. The method of claim 1, wherein S1 comprises: collecting unit load and determining unit load sampling time, and simultaneously collecting dynamic pressure of a combustion chamber and taking the unit load sampling time as a working condition time reference; accessing a fuel component and heating value database according to a working condition time standard to obtain low-level heating value of the fuel and the volume fraction of a main component of the fuel, wherein the volume fraction of the main component of the fuel comprises methane, ethane, propane, hydrogen, carbon monoxide, carbon dioxide, nitrogen and inert gas; And splicing the unit load, the low-level heating value of the fuel and the volume fraction of the main fuel component into a ten-dimensional fuel characteristic package under the current working condition according to a fixed dimension sequence, and intercepting a continuous sampling sequence with the length of twenty-zero forty-eight from the dynamic pressure of the combustion chamber to form a twenty-zero forty-eight-dimensional pressure sequence for input in the subsequent step.
3. The method of claim 1, wherein S2 comprises: determining event retrieval conditions according to the fuel characteristic package under the current working condition, wherein the event retrieval conditions comprise that the unit load difference value does not exceed a first load threshold value, the fuel low-order heating value difference value does not exceed a first heating value threshold value, and the difference value of each dimension of the volume fraction of the main fuel component does not exceed a first component threshold value; inputting the event retrieval conditions into a combustion stability event database, retrieving to obtain a stable event and a destabilizing event which meet the event retrieval conditions, and forming a comparison event by the stable event and the destabilizing event according to time sequence; And carrying out on-line calibration on the fuel condition calibration layer based on the comparison event, wherein on-line calibration is defined as updating parameters of the fuel condition calibration layer and keeping parameters of a stable judging trunk unchanged, iteratively updating the parameters of the fuel condition calibration layer until the parameters of the fuel condition calibration layer meet calibration judging conditions or reach a preset iteration number upper limit, and outputting self-adaptive stability allowance after on-line calibration.
4. The method of claim 1, wherein S3 comprises: Inputting the self-adaptive stability margin and the fuel characteristic package under the current working condition into a combustion stability event database, and screening candidate event records according to the condition that the unit load difference value does not exceed a second load threshold value, the fuel low-order heating value difference value does not exceed a second heating value threshold value, and the difference value of each dimension of the volume fraction of the main component of the fuel does not exceed a second component threshold value; and determining event labels corresponding to the margin intervals of the self-adaptive stable margins in the candidate event records, and selecting the event label with the minimum self-adaptive stable margin difference value as a matching event label when a plurality of event labels exist.
5. The method of claim 1, wherein S4 comprises: Performing feasibility screening on the adjustment action candidate set under the constraint of the adjustable range of the qualified emission, and selecting the adjustment action with the residual adjustment space of the adjustable initial value corresponding to the dimension of the target control quantity as the adjustment action; Inputting an adaptive stable margin and risk threshold value group into an amplitude calculation rule, setting the regulation and control amplitude to be zero when the adaptive stable margin does not meet a trigger threshold value, determining the regulation and control amplitude in a linear interpolation mode when the adaptive stable margin meets the trigger threshold value and does not meet an upgrade threshold value, and setting the regulation and control amplitude to be a preset maximum amplitude when the adaptive stable margin meets the upgrade threshold value; Applying the regulation and control amplitude to the dimension of the target control quantity of the regulation and control action and performing boundary cutting, cutting the target control quantity to a corresponding boundary and synchronously correcting the regulation and control amplitude when the applied target control quantity exceeds the lower boundary or the upper boundary of the control quantity; and outputting the regulation action, the corrected regulation amplitude and the target control quantity as a regulation instruction draft.
6. The method of claim 1, wherein S5 comprises: Sending the regulation instruction draft to a unit control system and executing regulation action to obtain an executed working condition; Collecting and executing the dynamic pressure of the post-combustion chamber after the preset response delay, and intercepting a continuous sampling sequence with the length of two thousand zeros and forty-eight seconds to form a pressure sequence with four eighty seconds after execution; Inputting the executed twenty-four forty-eight dimensional pressure sequence and the fuel feature package of the current working condition into a self-adaptive stability margin estimation network during operation, and calculating and outputting the executed self-adaptive stability margin by adopting fuel condition calibration layer parameters and stability discrimination trunk parameters which are calibrated on line; and obtaining stable response quantity by differentiating the self-adaptive stable margin after execution and the self-adaptive stable margin before execution of the regulation instruction draft.
7. The method of claim 1, wherein S6 comprises: Comparing the stable response quantity with the upgrading threshold value in the risk threshold value group, and entering a rollback generation flow when the stable response quantity meets the upgrading threshold value; comparing the stable response quantity with a release threshold value in the risk threshold value group when the rollback generation flow is not entered, and outputting a regulation instruction draft as an online regulation instruction when the stable response quantity meets the release threshold value; in the rollback generation flow, determining a rollback control amount dimension according to a target control amount dimension of the regulation instruction draft, setting a rollback adjustment direction to be opposite to the target control amount adjustment direction of the regulation instruction draft, and setting a rollback amplitude to be a preset rollback amplitude; And under the constraint of the adjustable range of qualified emission, the rollback amplitude is acted on the dimension of the rollback control quantity to obtain a rollback target control quantity, boundary cutting is carried out on the rollback target control quantity to generate a rollback regulation instruction, and the rollback regulation instruction is output as an online regulation instruction.
8. On-line regulation and control device of combustion stability of gas turbine based on degree of depth study, characterized by comprising: the fuel characteristic package fusion generation module is used for acquiring unit load, fuel components, heat value and dynamic pressure of a combustion chamber, retrieving a fuel component, heat value database and dynamic pressure historic library, and fusing to generate a fuel characteristic package under the current working condition; The self-adaptive stability margin assessment and calibration module is used for searching a stability/instability comparison event based on a fuel feature packet under the current working condition, inputting a self-adaptive stability margin assessment network during operation, calibrating a fuel condition calibration layer on line and outputting self-adaptive stability margin; The risk level judging and threshold generating module is used for determining the risk level and generating a risk threshold group containing triggering, releasing and upgrading thresholds according to the event labels of the self-adaptive stability allowance matched combustion stability event database; The regulation and control instruction draft generation module is used for defining a qualified regulation range of emission in the emission-working condition mapping matrix, regulating a corresponding regulation and control action group according to the risk level to obtain regulation and control actions, calculating regulation and control amplitude by combining the self-adaptive stability margin and the risk threshold value group, and generating a regulation and control instruction draft in a combined mode; the stable response quantity calculation module is used for executing a regulation instruction draft, collecting dynamic pressure of the combustion chamber after execution, inputting the dynamic pressure of the combustion chamber after execution and the fuel characteristic package of the current working condition into an evaluation network, and calculating and outputting stable response quantity; The stable response quantity evaluation and instruction output module is used for outputting the regulation instruction draft as an online regulation instruction when the stable response quantity reaches a release threshold value, and generating and outputting a rollback regulation instruction as an online regulation instruction under the constraint of the acceptable adjustable range of emission when the stable response quantity reaches an upgrade threshold value.
9. A computer device comprising a processor and a memory; Wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory for implementing a deep learning-based on-line regulation method for combustion stability of a gas turbine according to any one of claims 1 to 7.
10. A non-transitory computer readable storage medium having a computer program stored thereon, wherein the program when executed by a processor implements a deep learning based on-line regulation method of combustion stability of a gas turbine as claimed in any one of claims 1 to 7.

Description

Online regulation and control method for combustion stability of gas turbine based on deep learning Technical Field The invention relates to the field of flow monitoring and energy efficiency analysis of gas turbine generator sets, in particular to a gas turbine combustion stability online regulation and control method based on deep learning. Background The gas turbine is used as core equipment of an energy power system and is widely applied to the fields of electric power, industrial heat supply and distributed energy. Along with the popularization of novel fuel applications such as hydrogen-doped fuel, renewable energy coupling and the like, combustion stability control faces complex working condition challenges such as fuel component fluctuation, heat value drift and the like. In the related art, a combustion dynamic pressure monitoring system is matched with a signal processing algorithm to construct a combustion instability early warning system. In the control strategy level, the existing system mostly adopts a combustion control map based on test calibration, load segmentation PID control and an expert system, and realizes combustion organization adjustment through regular association of event type-treatment action. The combustion stability event database is used as a core supporting technology, stores the stability/instability event records under the history working conditions, forms a mapping relation with the control parameters, and provides decision references for operators. However, the prior art has systematic drawbacks in fuel-variant scenarios. Specifically, a fixed threshold system is prone to discrimination caliber drift when fuel composition changes, representing a conclusion that the same pressure amplitude may correspond to diametrically opposed stability under different fuel conditions. While the empirical based control strategy enables basic regulation, the lack of a feasibility verification mechanism within the acceptable limits of NOx/CO emissions presents a risk of control volume out-of-limits. In addition, the conventional method does not establish a closed loop verification system for dynamic pressure feedback after execution, and when the stability boundary is deviated due to fuel switching, the regulation and control instruction may cause combustion oscillation to be aggravated or flame to fall off. The lack of the fuel condition calibration layer results in that the stable discrimination trunk feature cannot adapt to the fuel characteristic change, and the separation of the event-driven regulation and control frame and the emission constraint makes the control action difficult to simultaneously meet the requirements of environmental protection and stability. The technologies are limited in practical scenes such as fluctuation of hydrogen-doped fuel, switching of pipe network air sources and the like, unplanned shutdown of a unit can be caused, annual operation and maintenance cost is increased, and dual requirements of novel fuel application on instantaneity and reliability are difficult to meet. Disclosure of Invention The present invention aims to solve at least one of the technical problems in the related art to some extent. Therefore, a first object of the present invention is to provide an online regulation method for combustion stability of a gas turbine based on deep learning. Another object of the present invention is to provide an on-line regulation device for combustion stability of a gas turbine based on deep learning. A third object of the invention is to propose a computer device. A fourth object of the present invention is to propose a non-transitory computer readable storage medium. In order to achieve the above object, an embodiment of a first aspect of the present invention provides an online regulation method for combustion stability of a gas turbine based on deep learning, including: s1, acquiring unit load, fuel components, heat value and dynamic pressure of a combustion chamber, retrieving a fuel component, heat value database and dynamic pressure historic library, and fusing to generate a fuel characteristic package under the current working condition; S2, searching a stability/instability comparison event based on a fuel feature packet under the current working condition, inputting an adaptive stability margin evaluation network during operation, calibrating a fuel condition calibration layer on line, and outputting an adaptive stability margin; S3, determining a risk level and generating a risk threshold group containing triggering, releasing and upgrading thresholds according to event labels of the self-adaptive stability allowance matched combustion stability event database; S4, defining an adjustable range of qualified emission in an emission-working condition mapping matrix, adjusting a corresponding adjusting action group according to a risk level to obtain an adjusting action, calculating an adjusting amplitude by combining an adaptive stability mar