CN-122020447-A - Boiler combustion early warning method based on improved deep forest algorithm

CN122020447ACN 122020447 ACN122020447 ACN 122020447ACN-122020447-A

Abstract

The invention discloses a boiler combustion early warning method based on an improved deep forest algorithm, which comprises the steps of obtaining an initial quantification index of a combustion state of combustion equipment, determining a dynamic fluctuation mode of the combustion process of the combustion equipment under a multi-parameter coupling condition, adopting the improved deep forest algorithm to construct a multi-level characteristic evaluation structure to obtain a reliability score of combustion characteristics, obtaining an adjusted intermediate layer risk characteristic if the reliability score of the combustion characteristics is lower than a preset threshold value, judging a triggering condition of potential coking risk, determining a root position of the combustion abnormality of the combustion equipment, and generating a real-time warning signal according to the root position of the combustion abnormality of the combustion equipment. The invention effectively solves the problems of insufficient combustion abnormality early warning capability and response lag of the traditional scheme under the multi-source complex working condition.

Inventors

WANG HAOYONG
TIAN YUHUA
YAN XIAORUI

Assignees

陕西华电榆横煤电有限责任公司榆横发电厂
清贺蓝(上海)环保科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251224

Claims (8)

1. The early warning method for boiler combustion based on the improved deep forest algorithm is characterized by comprising the following steps of: Collecting temperature data, oxygen amount data and flame image data in a combustion device hearth, extracting bottom layer feature vectors of the temperature data, the oxygen amount data and the flame image data, and obtaining an initial quantization index of a combustion state of the combustion device; Adopting a time sequence analysis model to process the variation trend of the initial quantization index and determining the dynamic fluctuation mode of the combustion process of the combustion equipment under the multi-parameter coupling condition; Constructing a multi-level characteristic evaluation structure by adopting an improved deep forest algorithm, and analyzing the combustion characteristics of each layer through initial quantization indexes to obtain the credibility score of the combustion characteristics; if the reliability score of the combustion feature is lower than a preset threshold value, the influence factors are calibrated based on the weight of the bottom layer feature vector through a weighted fusion mechanism, and an adjusted middle layer risk feature is obtained; Aiming at the adjusted middle layer risk characteristics, acquiring high-layer abnormal probability distribution of the combustion process of the combustion equipment, and judging triggering conditions of potential coking risks; Tracing the interaction path among multiple parameters by adopting a causal analysis model according to the triggering condition of the potential coking risk, and determining the root position of the combustion abnormality of the combustion equipment; and generating a real-time early warning signal according to the source position of the abnormal combustion of the combustion equipment.
2. The method for early warning of boiler combustion based on an improved deep forest algorithm according to claim 1, wherein the extracting of the bottom layer feature vector of temperature data, oxygen amount data and flame image data comprises: acquiring temperature data and oxygen data in the hearth in real time through a temperature sensor and an oxygen sensor, and acquiring flame image data through an image sensor; performing feature extraction processing on the acquired temperature data, oxygen amount data and flame image data to generate corresponding feature vectors, and determining bottom layer characterization information of the combustion state; analyzing the change trend of the temperature data and the oxygen data according to the generated feature vector, and triggering an abnormal state mark if the change trend exceeds a preset threshold range to judge whether a potential unstable combustion condition exists or not; The method comprises the steps of performing mode comparison on feature vectors extracted from flame image data, acquiring a form category which is most matched with a current combustion state by combining a pre-established flame form database, and determining a visual characterization result of a combustion process; Comprehensively analyzing the bottom layer characterization information and the visual characterization result of the combustion state, and if the relevance of the temperature data and the oxygen data is lower than a preset threshold value, carrying out weighted adjustment on the feature vector to obtain a combustion state evaluation result; And according to the combustion state evaluation result, analyzing the running stability in the hearth by combining with the historical record data of the equipment state, acquiring the state change characteristics under the long-term running trend, and constructing a complete initial quantization index.
3. The early warning method for boiler combustion based on the improved deep forest algorithm according to claim 1, wherein the determining the dynamic fluctuation mode of the combustion process of the combustion device under the multi-parameter coupling condition comprises: the method comprises the steps of carrying out segmentation processing on initial data according to a combustion state evaluation result of combustion equipment, extracting variation trends in different time windows, and obtaining staged fluctuation information of a combustion process; According to the staged fluctuation information, carrying out association analysis on a plurality of variables in the combustion process by combining a coupling scene of a plurality of parameter groups, and determining main driving factors of dynamic fluctuation; the method comprises the steps of constructing a rule recognition frame of a combustion process through extracting main driving factors of dynamic fluctuation, and acquiring response modes of a combustion state in different coupling scenes; if the response mode is within the preset threshold range, continuously tracking the dynamic fluctuation of the combustion process, and judging whether an abnormal deviation phenomenon exists or not; if the current time sequence exceeds the threshold range, recording fluctuation data of the current time sequence to obtain an abnormal marking result; According to the abnormal marking result, carrying out localized analysis on the combustion state evaluation result of the combustion equipment, comparing the current dynamic fluctuation by adopting a pre-established reference database, and determining a potential fluctuation source; Generating a dynamic adjustment basis of a combustion process by combining the analysis result of the potential fluctuation source and the change trend of the time sequence, and acquiring optimized reference information aiming at a multi-parameter set coupling scene; And continuously collecting and comparing the operation data of the combustion equipment according to the optimized reference information, and judging whether the combustion state tends to be stable or not to obtain a final dynamic fluctuation mode.
4. The early warning method for boiler combustion based on the improved deep forest algorithm according to claim 1, wherein the construction of the multi-level feature evaluation structure by adopting the improved deep forest algorithm comprises the following steps: acquiring an initial quantization index of the combustion state of the combustion equipment, acquiring in real time aiming at the dynamic fluctuation condition, and recording various parameters in the combustion process by a sensor to obtain an original fluctuation data set; Denoising and standardizing an original fluctuation data set, extracting dynamic fluctuation fragments in a segmented mode, and determining a basic form of a fluctuation mode; Constructing a multi-layer feature extraction structure according to the basic form of the fluctuation mode, and decomposing and characterizing each layer of combustion features by adopting a deep forest algorithm to obtain a layered feature set; Aiming at the layered feature set, implementing a layer-by-layer analysis flow, and if the significance of a certain layer of features is lower than a preset threshold value, carrying out weight adjustment on the feature, and judging the preliminary significance ranking of each layer of features; calculating the credibility score of each layer of combustion characteristics according to the preliminary importance ranking, integrating the scores of the layers of characteristics in a weighted accumulation mode, and determining a final credibility score; and aiming at the final credibility score, if the final credibility score deviates from a preset range, triggering a feature re-evaluation mechanism, and carrying out local correction by backtracking the layered feature set to obtain an adjusted credibility result.
5. The method for early warning of boiler combustion based on an improved deep forest algorithm of claim 1, wherein the adjusted intermediate layer risk features comprise: If the reliability score of the combustion characteristics is lower than a preset threshold value, determining a primary structure of weight distribution by analyzing the weights of the bottom-layer characteristic vectors one by one; if the initial structure of the weight distribution has the condition of deviating from the preset threshold value, calibrating the weight of the weight through a weighted fusion mechanism to obtain a calibrated weight combination; Calculating risk characteristics of the middle level according to the calibrated weight combination, and analyzing the relevance between the risk characteristics and the combustion characteristics to obtain quantized representation of the risk characteristics; constructing feature adjustment mapping logic through quantitative representation of risk features, judging whether the adjusted features accord with expectations or not, and obtaining a final correction result; and generating an intermediate layer risk characteristic of the combustion characteristic according to the final correction result, and determining potential risk points of the intermediate layer risk characteristic in the combustion process.
6. The early warning method for boiler combustion based on the improved deep forest algorithm according to claim 1, wherein the step of obtaining the high-level abnormal probability distribution of the combustion process of the combustion equipment and judging the triggering condition of the potential coking risk comprises the following steps: Aiming at the adjusted middle layer risk characteristics, acquiring high-level abnormal probability distribution through real-time data acquisition of combustion equipment in the operation period, and determining the possibility range of occurrence of the abnormality; screening by adopting a preset threshold according to the high-level abnormal probability distribution, and layering abnormal data if the probability distribution exceeds the threshold range to obtain a key attention area of high-level abnormality; Aiming at a focus attention area, extracting risk feature data related to intermediate layer risk features, judging the association degree of potential coking and high-level abnormality, and obtaining priority ranking of the risk features; And analyzing triggering conditions of potential coking risks through priority sequencing, and determining concrete expression forms of the triggering conditions by combining with real-time changes of equipment states.
7. The early warning method for boiler combustion based on the improved deep forest algorithm of claim 1, wherein the determining the root position of the combustion abnormality of the combustion device by tracing back the interaction path between the multiple parameters by using the causal analysis model comprises: Through the triggering condition of the potential coking risk, a causal analysis method is adopted to comb logic association among parameter changes, and an interaction path among multiple parameters is obtained; According to the interaction path among the multiple parameters, analyzing the direct association between the combustion abnormality and the change of each parameter, and determining the primary root position of the combustion abnormality; Aiming at the primary root position, acquiring real-time monitoring data of equipment state, and judging key influence factors of abnormality judgment by combining dynamic information of parameter change; Constructing an assessment framework of coking risk through key influence factors, and if the parameter change exceeds a preset threshold range, sequencing the priority of the abnormal judgment to obtain the key direction of risk assessment; According to the key direction of risk assessment, extracting concrete performance data of triggering conditions of potential coking risks, analyzing abnormal probabilities of combustion equipment in different operation scenes, determining distribution intervals of the triggering conditions of the potential coking risks, and determining the root positions of combustion abnormality of the combustion equipment.
8. The early warning method for boiler combustion based on the improved deep forest algorithm according to claim 1, wherein the generating a real-time early warning signal according to the root position of the combustion abnormality of the combustion device comprises: aiming at the source position of the combustion abnormality of the combustion equipment, acquiring a historical operation record related to the source position of the combustion abnormality of the combustion equipment, and judging the specific category of the source of the abnormality by comparing the deviation of the current data and the historical data; Acquiring corresponding influence range data from specific categories of abnormal sources, classifying the influence ranges, and determining the priority and signal generation mode of real-time early warning; Generating a corresponding early warning signal according to the priority of real-time early warning, and transmitting the early warning signal to a control module through an internal channel of the system to obtain a signal triggered response mechanism; aiming at a response mechanism, an adjusting instruction of the combustion process is generated by combining an optimization method, and is issued to combustion equipment through an instruction execution module, so that process monitoring and state updating are completed.

Description

Boiler combustion early warning method based on improved deep forest algorithm Technical Field The invention relates to the technical field of boilers, in particular to a boiler combustion early warning method based on an improved deep forest algorithm. Background In the operation process of the industrial boiler, the real-time monitoring and early warning of the combustion state are important to the safety and the production efficiency of equipment. The boiler combustion relates to complex interaction of various parameters, and once abnormality occurs, equipment damage and even safety accidents can be caused, so that research on how to accurately identify combustion abnormality and trace root causes through an intelligent means becomes a key subject to be solved urgently in the industrial field. Many current monitoring methods often have difficulty accurately quantifying the confidence level of the determination in the face of complex changes in combustion conditions. The prior art relies on data analysis of a single level more, lacks dynamic evaluation capability for information interaction of different levels, and is difficult to effectively distinguish normal fluctuation from potential risks especially under the condition of multi-parameter coupling. The limitation makes the system often have misjudgment or missed judgment when facing to the fuzzy state, and can not provide clear decision basis for operation and maintenance personnel. Meanwhile, the evaluation of the combustion state needs to be gradually deepened from the bottom sensor data to the high-layer risk characteristics, and the judgment reliability of each layer is affected by the information quality of the previous layer. If the reliability evaluation of the bottom layer data is inaccurate, analysis deviation of the subsequent layers can be caused, and the reliability of the whole early warning is further affected. Disclosure of Invention The invention aims to provide a boiler combustion early warning method based on an improved deep forest algorithm, which effectively solves the problems of insufficient combustion abnormality warning capability and response lag of the traditional scheme under the multi-source complex working condition. According to the embodiment of the invention, the early warning method for boiler combustion based on the improved deep forest algorithm comprises the following steps: Collecting temperature data, oxygen amount data and flame image data in a combustion device hearth, extracting bottom layer feature vectors of the temperature data, the oxygen amount data and the flame image data, and obtaining an initial quantization index of a combustion state of the combustion device; Adopting a time sequence analysis model to process the variation trend of the initial quantization index and determining the dynamic fluctuation mode of the combustion process of the combustion equipment under the multi-parameter coupling condition; Constructing a multi-level characteristic evaluation structure by adopting an improved deep forest algorithm, and analyzing the combustion characteristics of each layer through initial quantization indexes to obtain the credibility score of the combustion characteristics; if the reliability score of the combustion feature is lower than a preset threshold value, the influence factors are calibrated based on the weight of the bottom layer feature vector through a weighted fusion mechanism, and an adjusted middle layer risk feature is obtained; Aiming at the adjusted middle layer risk characteristics, acquiring high-layer abnormal probability distribution of the combustion process of the combustion equipment, and judging triggering conditions of potential coking risks; Tracing the interaction path among multiple parameters by adopting a causal analysis model according to the triggering condition of the potential coking risk, and determining the root position of the combustion abnormality of the combustion equipment; and generating a real-time early warning signal according to the source position of the abnormal combustion of the combustion equipment. Optionally, the extracting the bottom layer feature vector of the temperature data, the oxygen amount data and the flame image data includes: acquiring temperature data and oxygen data in the hearth in real time through a temperature sensor and an oxygen sensor, and acquiring flame image data through an image sensor; performing feature extraction processing on the acquired temperature data, oxygen amount data and flame image data to generate corresponding feature vectors, and determining bottom layer characterization information of the combustion state; analyzing the change trend of the temperature data and the oxygen data according to the generated feature vector, and triggering an abnormal state mark if the change trend exceeds a preset threshold range to judge whether a potential unstable combustion condition exists or not; The method comprises the step