A combustion stability evaluation method and system

Abstract

The present invention provides a combustion stability evaluation method for performing combustion stability evaluation and analyzing causes of combustion instability, comprising: obtaining a first data set consisting of historical data records of boiler combustion operation; performing data analysis on historical records of all indicators related to combustion performance in the first data set, and dividing each indicator related to combustion performance into a normal indicator value interval and an abnormal indicator value interval; determining whether each indicator related to combustion performance in each historical data record is abnormal; determining a combustion stability evaluation result of the historical data record according to the determination result of all indicators related to combustion performance in each data record; based on the stability evaluation result of each determined data record, performing combustion stability evaluation on newly added data records using a K nearest neighbor model based on random projection; and analyzing operating environment related indicators corresponding to the causes of instability for newly added data records whose stability evaluation results are unstable.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power production, in particular to the field of power plant boiler combustion monitoring, and more particularly to a boiler combustion stability evaluation method and system. Background Art

[0002] Boilers are important equipment in thermal power plants. The operation of boilers largely determines the safety of the entire unit. Due to the complex structure of the boiler system and the poor operating conditions, boilers have always been the equipment with the most concentrated problems, the highest accident rate, and the greatest impact on the availability of the unit in power plant operations. The stability of combustion directly affects the safety of the boiler, that is, whether the boiler can burn stably and continuously is the most important factor related to combustion safety. Boiler combustion is a complex and changeable process. The quality of the combustion state directly affects the economic benefits of the power plant.

[0003] At present, there have been many research works on the stability diagnosis of boiler combustion in thermal power plants at home and abroad, which can be roughly divided into two categories: diagnostic methods based on measurement point parameters such as fire detection signals, flame spectrum or furnace pressure, and diagnostic methods based on flame digital image analysis. The former method uses the existing measurement points of the control system for mathematical analysis and processing, and the latter method designs a special image acquisition, transmission and processing system. Relatively speaking, the former method has lower construction and maintenance costs, and is easier to implement engineering applications in the context of power generation groups generally paying attention to cost control. The fire detection signal is a direct reflection of the combustion intensity. The latter method uses fuzzy C-means clustering algorithm, neural network algorithm, etc. to process the fire detection signal, which is one of the effective means to diagnose combustion safety and stability.

[0004] However, most of the existing research work focuses on the evaluation of combustion status, and most of them are aimed at relatively single data objects. There is a lack of further analysis on the causes and positioning of abnormal combustion status, which makes the fault analysis and optimization of combustion lack of practical guidance. When abnormal combustion occurs, it is impossible to accurately locate the problem points and causes that lead to abnormal combustion, resulting in the inability to take accurate and targeted improvement measures for abnormal combustion. Summary of the invention

[0005] Therefore, the purpose of the present invention is to overcome the defects of the prior art and provide a new combustion stability evaluation method and system.

[0006] The objective of the present invention is achieved through the following technical solutions:

[0007] According to a first aspect of the present invention, a combustion stability evaluation method is provided, which is used to evaluate combustion stability and analyze the causes of combustion instability according to any newly added data records corresponding to indicators in a boiler combustion process, and the method comprises the following steps: S1, obtaining a first data set consisting of historical data records of boiler combustion operation, wherein each data record contains historical records of all indicators related to the operating environment and historical records of all indicators related to combustion performance; S2, performing data analysis on the historical records of all indicators related to combustion performance in the first data set obtained in step S1, dividing each indicator related to combustion performance into a normal indicator value interval and an abnormal indicator value interval, and providing an online learning mechanism to dynamically update the normal indicator value interval and the abnormal indicator value interval of each indicator related to combustion performance; S3, determining each historical data according to the interval divided in step S2. Whether each indicator related to the combustion performance in the record is abnormal; S4, determine the combustion stability evaluation result of the historical data record according to the judgment results of all indicators related to the combustion performance in each data record, wherein the combustion stability evaluation result corresponding to the historical data record whose judgment results of all indicators related to the combustion performance are normal is stable, otherwise, the combustion stability evaluation result is unstable; S5, based on the stability evaluation result of each data record determined in step S4, construct a mapping relationship between all indicators related to the operating environment in each historical data record in the first data set and the combustion stability evaluation result, and use the K nearest neighbor model based on random projection to evaluate the combustion stability of the newly added data records, wherein the newly added data records contain all indicators related to the operating environment; S6, analyze the operating environment related indicators corresponding to the causes of instability for the newly added data records whose stability evaluation results are unstable.

[0008] Wherein, the step S1 comprises: S11, obtaining historical data records at least three months ago to form a first data set; S12, performing data cleaning on each historical data record in the first data set to remove invalid values.

[0009] In the step S2, the following steps are performed for each indicator related to the combustion performance: S21, extracting the historical record of the current indicator from each historical data record of the first data set to form a first data subset of the indicator; S22, using the skewness and kurtosis method based on the first data subset of the current indicator to detect whether the historical record of the indicator conforms to the normal distribution; S23, dividing the normal indicator value interval and the abnormal indicator value interval of the current indicator, wherein the 3σ criterion is used for the historical records of the indicator that meet the normal distribution, and the value interval corresponding to [μ-3σ,μ+3σ] is divided into the normal indicator value interval, and the others are the abnormal indicator value intervals, wherein μ is the mean of the historical records of the indicator, and σ is the standard deviation; the Tukey test method is used for the historical records of the indicator that do not meet the normal distribution, and [Q1-1.5*IQR,Q3+1.5*IQR] is divided into the normal indicator value interval, and the others are the abnormal indicator value intervals, wherein Q1 is the first quartile corresponding to the historical data record of the indicator when the Tukey test method is used, and Q3 is the Tukey test method. The third quartile of the historical data record of the indicator in the base test method, IQR is the interquartile range and IQR = Q3-Q1; S24, providing an online learning mechanism to dynamically update the normal value range of the indicator, mainly including two updating strategies: the first one, for newly added data records, for indicators that do not conform to the normal distribution and the newly added records are less than Q1-1.5*IQR (or), the lower limit of the original normal value range of the indicator is dynamically updated to Q1-1.5*IQR-ε, where ε is a smaller value. In some embodiments of the present invention, In the embodiment, ε is set to 0.01*IQR. For indicators that conform to the normal distribution and have new records less than μ-3σ, the lower limit of the original normal value interval of the indicator is dynamically updated to μ-3σ–ε; for indicators that do not conform to the normal distribution and have new records greater than Q1–1.5*IQR, the upper limit of the original normal value interval of the indicator is dynamically updated to Q1–1.5*IQR+ε, and for indicators that conform to the normal distribution and have new records greater than μ+3σ, the upper limit of the original normal value interval of the indicator is dynamically updated to μ+3σ+ε. The second method is to re-update the normal indicator value interval and abnormal indicator value interval of each combustion performance-related indicator through the process described in steps S1 and S2 after a preset fixed time (such as 1 month).

[0010] In step S3, the indicators that meet the normal distribution are judged according to the normal and abnormal indicator value intervals divided by the normal distribution, and the indicators that do not meet the normal distribution are judged according to the normal and abnormal indicator value intervals divided by the Tukey test method.

[0011] The step S5 comprises: S51, extracting the historical records of all indicators related to the operating environment in each historical data record in the first data set and vectorizing them, all the vectors constitute a training set, wherein all the indicators related to the operating environment in each historical data record correspond to a vector, and each vector is associated and mapped with the combustion stability evaluation result of the corresponding historical data record; S52, vectorizing the newly added records of all indicators related to the operating environment in the newly added data records as test vectors; S53, uniformly projecting the points in the high-dimensional space into the low-dimensional space by randomly projecting the test vector and each vector in the training set, and obtaining the test vector in the low-dimensional space and the training vector in the low-dimensional space. training set; S54, calculating the Euclidean distance between the test vector in the low-dimensional space obtained in step S53 and each vector in the training set in the low-dimensional space, and screening out a preset number of first neighbor vectors with the closest test trace distance from the training set; S55, calculating the average value of all indicators related to combustion performance in the historical data records corresponding to all the first neighbor vectors screened out in step S54 as the combustion performance related indicator value corresponding to the test vector; S56, judging whether each indicator related to combustion performance corresponding to the test vector is normal according to the normal indicator value range and the abnormal indicator value range of each combustion performance related indicator, and determining the combustion stability evaluation result corresponding to the test vector according to the judgment result of each indicator.

[0012] The step S6 includes: S61, selecting all vectors with stable combustion stability evaluation results from the training set to form a training sample set; S62, mapping the test vector and each vector in the training sample set to the low-dimensional space according to the random projection method; S63, calculating the Euclidean distance between the test vector in the low-dimensional space and each vector in the training sample set in the low-dimensional space, and selecting a preset number of second neighbor vectors closest to the test vector from the training sample set; S64, calculating the average value of all indicators related to the operating environment in the historical data records corresponding to all the second neighbor vectors selected in step S63 as the normal value of the operating environment related indicators corresponding to the test vector; S65, comparing all indicators in the test vector with the corresponding normal values ​​of the operating environment related indicators, and the indicators in the test vector that deviate from the preset threshold relative to the normal value are the indicators that cause combustion instability.

[0013] According to a second aspect of the present invention, there is provided a combustion stability evaluation system for the combustion stability evaluation method according to the first aspect of the present invention, comprising: a plurality of sensors, each of which is arranged at a different position of the boiler, and is used to collect various index data in real time; a storage module, which is used to store the data collected by the plurality of sensors to form data records, each of which contains records of all indicators; a data cleaning module, which is used to clean the historical data records stored in the storage module and remove invalid values; a data analysis module, which is used to analyze the characteristics of each combustion performance-related index in the historical data records processed by the data cleaning module, and to perform a normality test to detect whether the historical records of the index conform to the normal distribution, and to divide the historical records into different types according to the characteristics of each combustion performance-related index. The corresponding normal indicator numerical interval and abnormal indicator numerical interval, among which, for the historical records of indicators that meet the normal distribution, the 3σ criterion is adopted, and the numerical interval corresponding to [μ-3σ,μ+3σ] is divided into the normal indicator numerical interval, and the others are the abnormal indicator numerical intervals, among which, μ is the mean of the historical records of the indicator, and σ is the standard deviation; for the historical records of indicators that do not meet the normal distribution, the Tukey test method is adopted, and [Q1–1.5*IQR,Q3+1.5*IQR] is divided into the normal indicator numerical interval, and the others are the abnormal indicator numerical intervals, among which, Q1 is the first quartile corresponding to the historical data record of the indicator when the Tukey test method is adopted, and Q3 is the third quartile corresponding to the historical data record of the indicator when the Tukey test method is adopted, and IQR=Q3–Q1. An online learning mechanism is also provided. The dynamic update mainly includes two update strategies, namely, the update strategy for the newly added data set and the readjustment strategy at a fixed time, so as to ensure the accuracy of the abnormal indicator numerical range; the stability evaluation module is used to determine whether each record of the indicator related to the combustion performance in the data record is normal according to the normal indicator numerical range and the abnormal indicator numerical range of the indicators related to the combustion performance divided by the data analysis module, and determine the combustion stability evaluation result of the data record according to the judgment results of all the indicators related to the combustion performance in each data record, wherein the combustion stability evaluation result corresponding to the data record whose judgment results of all the indicators related to the combustion performance are normal is stable, otherwise, the combustion stability evaluation result is unstable; the cause analysis module is used to analyze the operating environment related indicators corresponding to the causes of instability of the data record whose combustion stability evaluation results are unstable.

[0014] Compared with the prior art, the advantages of the present invention are: based on the real measurement point data of a boiler equipment in a large power plant, the present invention comprehensively considers a total of 205 indicators in multiple categories such as operating environment related (operating conditions, grinding group air powder, boiler air distribution), combustion performance related (furnace negative pressure, flue gas oxygen content, fire detection signal strength), etc. to establish a combustion stability evaluation and cause analysis method and device, by integrating the 3σ criterion and Tukey test, dynamically updating the abnormal indicator value interval in the data analysis process in an online learning manner, and proposing a K-nearest neighbor model based on random projection to analyze the causes of instability, which can evaluate the combustion stability more comprehensively and accurately, and can give the causes of combustion instability caused by the current boiler, and give a precursory warning, so as to facilitate the early adoption of preventive measures, reduce or avoid the occurrence of faults, and provide a combustion state online evaluation and abnormal combustion state cause analysis method that is practical in engineering practice, which is of great significance for finding the crux of combustion instability and guiding on-site operators to perform correct operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The embodiments of the present invention are further described below with reference to the accompanying drawings, in which:

[0016] Figure 1 Schematic diagram of the structure of a combustion stability evaluation system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] As described in the background technology, the existing methods cannot comprehensively evaluate the combustion stability of the boiler, and cannot further analyze the causes and positioning of the abnormal combustion state. Therefore, the present invention provides a method and system for comprehensively analyzing the complete data record of the boiler combustion to evaluate the combustion stability, and using the actual operating indicators of the boiler over a period of time, including: operating environment related indicators and combustion performance related indicators, to evaluate the combustion stability and analyze the causes of abnormal combustion states. The operating abnormalities of the combustion performance related indicators are checked online, and the stability of the furnace combustion is comprehensively evaluated based on this, and the stable or unstable evaluation results are output, where the instability is divided into three levels: general, poor, and poor, with a total of four levels of stability evaluation. For boilers with unstable combustion stability evaluation status, through the cause analysis of the abnormal combustion state, the combustion performance related indicators with abnormal status and the operating environment related indicators that cause the abnormality are output, and an early warning is issued, thereby helping business personnel to determine the cause of the abnormal combustion state, reduce the possibility of safety accidents in the boiler, and avoid large economic losses.

[0019] According to one embodiment of the present invention, the present invention provides a combustion stability evaluation method, comprising the following steps:

[0020] Step S1, obtaining a first data set consisting of historical data records of boiler combustion operation, wherein each data record contains historical records of all indicators related to the operating environment and historical records of all indicators related to combustion performance; according to one embodiment of the present invention, multiple acquisition sensors are used to collect different indicator data in real time, and multiple indicators of the same type are installed at different positions to collect data of indicators related to combustion performance, so that the same type of indicators related to combustion performance correspond to multiple data, wherein the indicators related to the operating environment (for example, there are 176 indicators such as working conditions, grinding group air powder, boiler air distribution, etc.) and the indicators related to combustion performance (for example, furnace negative pressure, flue gas oxygen content, fire detection signal strength, etc. 29 indicators, wherein the fire detection signal strength includes 24 indicators, the furnace negative pressure includes 4 indicators and the flue gas oxygen content includes 1 indicator) total 205 indicators, and each data record contains all indicator records. Preferably, in order to better adapt to the influence of different seasons and temperature and humidity changes on the combustion stability of the boiler, the historical data records of the last three months are selected as the first data set, and the data in the first data set are cleaned to remove invalid values ​​such as null values ​​and NaN.

[0021] Step S2: Analyze the historical records of all indicators related to combustion performance in the first data set obtained in step S1, and divide each indicator related to combustion performance into a normal indicator value interval and an abnormal indicator value interval. According to one embodiment of the present invention, in step S2, steps S21, S22, and S23 are performed for each indicator related to combustion performance:

[0022] S21. Extract the historical records of the current indicator from each historical data record of the first data set to form a first data subset of the indicator. For example, there are 29 indicators related to combustion performance. Then, for each indicator related to combustion performance, extract the historical records of the indicator from each historical data record of the first data set to form its corresponding first data subset, for a total of 29 first data subsets.

[0023] S22. Based on the first data subset of the current indicator, a skewness and kurtosis method is used to detect whether the historical records of the indicator conform to the normal distribution.

[0024] S23. Divide the current indicator into normal indicator value intervals and abnormal indicator value intervals. For the historical records of indicators that meet the normal distribution, calculate their mean μ and standard deviation σ. According to the 3σ criterion, the probability of data being in (μ-3σ, μ+3σ) is 99.7%. Data beyond this normal fluctuation range is abnormal data. Therefore, the 3σ criterion is adopted to divide the value interval corresponding to [μ-3σ, μ+3σ] into normal indicator value intervals, and the others into abnormal indicator value intervals, that is, data less than μ-3σ or greater than μ+3σ is judged to be abnormal data; Tukey test method is used for historical records of indicators that do not meet the normal distribution, because there is a high probability that the 29 combustion performance-related indicators of the three categories of fire detection signal intensity, furnace negative pressure and flue gas oxygen content do not meet the assumption of normal distribution. Therefore, Tukey test method is used for indicators that do not meet the assumption of normal distribution. The test method, that is, the box plot method, is used for further anomaly identification. The standard for judging outliers by the box plot is based on quartiles (first quartile Q1, third quartile Q3) and interquartile range (IQR = Q3-Q1). Quartiles refer to the values ​​at three dividing points in statistics when all values ​​are arranged from small to large and divided into four equal parts. The first quartile (Q1) is equal to the 25th% number of all values ​​in the sample arranged from small to large; the second quartile (Q2), also known as the "median", is equal to the 50th% number of all values ​​in the sample arranged from small to large; the third quartile (Q3) is equal to the 75th% number of all values ​​in the sample arranged from small to large; the gap between the third quartile and the first quartile is also called the interquartile range (InterQuartileRange, IQR). Quartiles have a certain degree of resistance. Up to 25% of the data can become arbitrarily far away without greatly disturbing the quartiles, so outliers will not affect the data shape of the box plot. The results of the box plot identifying outliers are relatively objective. The lower limit of the normal fluctuation corresponding to each indicator Q1-1.5*IQR and the upper limit of the normal fluctuation corresponding to Q3+1.5*IQR are found. The data within this range accounts for 99.3% of the total data. Data beyond this normal fluctuation range is abnormal data. Therefore, [Q1-1.5*IQR, Q3+1.5*IQR] is divided into the normal indicator value interval, and the others are the abnormal indicator value interval, that is, when the data is less than Q1-1.5*IQR or greater than Q3+1.5*IQR, it is judged to be abnormal data.

[0025] S24. Provide an online learning mechanism to dynamically update the normal numerical range of indicators, mainly including two updating strategies: the first is that for indicators in the newly added data records that do not conform to the normal distribution and the newly added records are less than Q1-1.5*IQR, the lower limit of the original normal numerical range of the indicator is dynamically updated to Q1-1.5*IQR-ε, where ε is a smaller value. In some embodiments of the present invention, ε is set to 0.01*IQR, and for indicators that conform to the normal distribution and the newly added records are less than μ-3σ, the lower limit of the original normal numerical range of the indicator is dynamically updated to μ-3σ-ε; for indicators that do not conform to the normal distribution and the newly added records are greater than Q1-1.5*IQR, the upper limit of the original normal numerical range of the indicator is dynamically updated to Q1-1.5*IQR+ε, and for indicators that conform to the normal distribution and the newly added records are greater than μ+3σ, the upper limit of the original normal numerical range of the indicator is dynamically updated to μ+3σ+ε. The second method is to re-update the normal index value interval and abnormal index value interval of each combustion performance-related index through the process described in steps S1 and S2 after a preset fixed time (such as 1 month). S3. Determine whether each combustion performance-related index in each historical data record is abnormal according to the interval divided in step S2; wherein, the indexes that meet the normal distribution are determined according to the normal and abnormal index value intervals divided by the normal distribution, and the indexes that do not meet the normal distribution are determined according to the normal and abnormal index value intervals divided by the Tukey test method.

[0026] S4. Determine the combustion stability evaluation result of the historical data record according to the judgment results of all indicators related to the combustion performance in each data record. Since there are many indicators related to the combustion performance, some indicators may be normal and some may be abnormal. Therefore, when judging the combustion stability, the combustion stability evaluation result corresponding to the historical data record whose judgment results of all indicators related to the combustion performance are normal is stable; otherwise, the combustion stability evaluation result is unstable.

[0027] S5. Based on the stability evaluation result of each data record determined in step S4, a mapping relationship between all indicators related to the operating environment and the combustion stability evaluation result in each historical data record in the first data set is constructed, and a K-nearest neighbor model based on random projection is proposed to evaluate the combustion stability of the newly added data records, wherein the newly added data records contain all indicators related to the operating environment; according to one embodiment of the present invention, the random projection uses a Gaussian random projection method to perform the analysis task, and the combustion stability of the newly added data records is evaluated. In the regression analysis, indicators related to the operating environment are used as features (for example, 176 indicators such as operating conditions, grinding group air powder, and boiler air supply), and indicators related to combustion performance (for example, 29 indicators such as furnace negative pressure, flue gas oxygen content, and fire detection signal strength) are used as labels to construct a functional relationship from indicators related to the operating environment to indicators related to combustion performance, and a mapping relationship is established between indicators related to the operating environment and the combustion stability evaluation result. For any newly added data record, the predicted value of the corresponding combustion performance-related indicator can be obtained according to the regression analysis, and the combustion stability evaluation result corresponding to the newly added data record can be determined according to the normal indicator value range and the abnormal indicator value range of the indicator. Taking the K value of 3 (the preset number of neighbors) when calculating the nearest neighbors as an example, the regression analysis process is mainly composed of the following steps:

[0028] S51. Extract the historical records of all indicators related to the operating environment in each historical data record in the first data set and vectorize them. All vectors constitute a training set, wherein all indicators related to the operating environment in each historical data record correspond to a vector, and associate and map each vector with the combustion stability evaluation result of the corresponding historical data record; taking the aforementioned embodiment as an example, each historical data record in the first data set contains 205 indicator records, and 176 indicators related to the operating environment are extracted and vectorized, represented as a 176-dimensional vector, denoted as X1, X2...Xn, where n is the size of the training set.

[0029] S52, vectorizing all operating environment related indicators of the newly added data record as a test vector; taking the above embodiment as an example, the newly added data record only includes 176 records of operating environment related indicators, which are vectorized and represented as a 176-dimensional vector, denoted as T;

[0030] S53. Set the dimension K of the target space (K<176) and generate a random matrix R. According to the Johnson-Lindenstrauss lemma, a small number of points in a high-dimensional space can be embedded in a lower-dimensional space, so that the distance between the points is almost maintained. Multiply the test vector T and the training set (X1, X2...Xn) by R respectively to obtain the test vector T' in the low-dimensional space and the training set (X1', X2'...Xn') in the low-dimensional space.

[0031] S54, calculate the Euclidean distance between the test vector T' in the low-dimensional space and each vector in the training set (X1', X2'...Xn') in the low-dimensional space, and select the three first neighbor vectors closest to the test vector from the training set to find three historical data records.

[0032] S55. Calculate the average value of all combustion performance-related indicators in the historical data records corresponding to all the first neighbor vectors screened out in step S54 as the combustion performance-related indicator value corresponding to the test vector T.

[0033] S56. Determine whether each combustion performance-related indicator corresponding to the test vector T is normal based on the normal indicator value interval and the abnormal indicator value interval of each combustion performance-related indicator, and determine the combustion stability evaluation result corresponding to the test vector T based on the judgment result of each indicator. Among them, among the 29 combustion performance-related indicators in three categories, namely, fire detection signal intensity, furnace negative pressure and flue gas oxygen content, if only one of the indicators in each category is abnormal, it is considered that the indicator in this category is abnormal. If all three categories of indicators are normal, the stability rating is stable; if any one of them is abnormal, the stability rating is general; if two categories are abnormal, it is poor; if all three categories are abnormal, it is poor.

[0034] Step S6: Analyze the operating environment related indicators corresponding to the causes of instability for the newly added data records whose stability evaluation results are unstable. Still taking the K nearest neighbor model based on random projection in step S5, where K takes a value of 3 (preset number of neighbors) as an example to illustrate the instability cause analysis process, it mainly includes the following steps:

[0035] S61, selecting all vectors with stable combustion stability evaluation results from the training set to form a training sample set;

[0036] S62, set the dimension K of the target space (K<176), and generate a random matrix R. Multiply the test vector T, the training set (X1, X2...Xn) and R respectively to obtain the test vector T' in the low-dimensional space and the training set (X1', X2'...Xn') in the low-dimensional space;

[0037] S63, calculating the Euclidean distance between the test vector T' in the low-dimensional space and each vector in the low-dimensional space training sample set, and selecting three second neighbor vectors with the closest distance to the test vector from the training sample set;

[0038] S64, calculating the average value of all indicators related to the operating environment in the historical data records corresponding to all the second neighbor vectors screened out in step S63 as the normal value T" of the indicators related to the operating environment corresponding to the test vector;

[0039] S64. Compare all the indicators in the test vector T with the corresponding normal values ​​T” of the operating environment related indicators, and compare whether each dimension of T and T” exceeds the deviation threshold. The operating environment indicator corresponding to the dimension that exceeds the deviation threshold is the indicator that causes unstable combustion. Output all causes of unstable combustion, that is, the indicator in the test vector T that deviates from the preset threshold relative to the normal value T” is the indicator that causes unstable combustion. The threshold is set according to the specific working environment of the boiler.

[0040] According to one embodiment of the present invention, the present invention also provides a combustion stability evaluation system, comprising: a sensor set, a storage module, a data cleaning module, a data analysis module, a stability evaluation module, and a cause analysis module. The sensor set includes a plurality of sensors, each of which is configured at a different position of the boiler and is used to collect various index data in real time. The storage module is used to store the data collected by the sensor set to form data records, each of which contains records of all indicators. The data cleaning module is used to clean the historical data records stored in the storage module and remove invalid values. The data analysis module is used to analyze the characteristics of each combustion performance-related indicator in the historical data records processed by the data cleaning module, and perform normal detection to detect whether the historical records of the indicator conform to the normal distribution, and divide the corresponding normal indicator value interval and abnormal indicator value interval according to the characteristics of each combustion performance-related indicator, wherein the 3σ criterion is used for the historical records of indicators that meet the normal distribution, and the value interval corresponding to [μ-3σ, μ+3σ] is divided into the normal indicator value interval, and the others are abnormal indicator value intervals, wherein μ is the mean of the historical records of the indicator, and σ is the standard deviation; the Tukey test method is used for the historical records of indicators that do not meet the normal distribution, and [Q1-1.5*QR, Q3+1.5*IQR] is divided into the normal indicator value interval, and the others are abnormal indicator value intervals, wherein Q1 is the first quartile corresponding to the historical data record of the indicator when the Tukey test method is used, and Q3 is the third quartile corresponding to the historical data record of the indicator when the Tukey test method is used, and IQR=Q3-Q1. Among them, it is said that the data analysis module adopts an online learning mechanism to dynamically update the normal indicator numerical range and the abnormal indicator numerical range by combining two updating strategies: first, for newly added data records, for indicators that do not conform to the normal distribution and the newly added records are less than Q1-1.5*IQR, the lower limit of the original normal numerical range of the indicator is dynamically updated to Q1-1.5*IQR-ε, where ε is a smaller value. Preferably, in some embodiments of the present invention, ε is set to 0.01*IQR, and for indicators that conform to the normal distribution and the newly added records are less than μ-3σ, the lower limit of the original normal numerical range of the indicator is dynamically updated to μ-3σ-ε; for indicators that do not conform to the normal distribution and the newly added records are greater than Q1-1.5*IQR, the upper limit of the original normal numerical range of the indicator is dynamically updated to Q1-1.5*IQR+ε, and for indicators that conform to the normal distribution and the newly added records are greater than μ+3σ, the upper limit of the original normal numerical range of the indicator is dynamically updated to μ+3σ+ε. The second method is to re-update the normal index value range and the abnormal index value range of each combustion performance-related index through the process described in steps S1 and S2 after a preset fixed time (such as 1 month).The stability evaluation module is used to determine whether each combustion performance-related index record in the data record is normal according to the normal index value interval and abnormal index value interval of the combustion performance-related index divided by the data analysis module, and determine the combustion stability evaluation result of each data record according to the judgment results of all combustion performance-related indexes in each data record, wherein the combustion stability evaluation result corresponding to the data record whose judgment results of all combustion performance-related indexes are normal is stable, otherwise, the combustion stability evaluation result is unstable. The cause analysis module is used to analyze the operating environment-related indicators corresponding to the causes of instability of the data record whose combustion stability evaluation results are unstable, and can be analyzed by model construction regression analysis tasks, such as K nearest neighbor models based on random projection, neural network models, etc.

[0041] The boiler combustion system is the most important part of the boiler system. The quality of the combustion state directly affects the economic benefits of the power plant. However, the factors affecting boiler combustion are very complex, including the combustion environment, coal quality and other factors, which have an important impact on the stability of combustion. Existing research work lacks comprehensive consideration in the evaluation of the combustion state, and lacks further analysis of the causes of unstable combustion state, and lacks guidance for real-time optimization and fault analysis of combustion. The present invention is based on the real measurement point data of a boiler equipment in a large power plant, comprehensively considers a total of 205 indicators in multiple categories such as operating environment related (operating conditions, grinding group air powder, boiler air distribution), combustion performance related (furnace negative pressure, flue gas oxygen content, fire detection signal strength), etc. to establish a combustion stability evaluation and cause analysis method and device, by integrating the 3σ criterion and Tukey test, dynamically updating the abnormal indicator value interval in the data analysis process in an online learning manner, and proposing a K-nearest neighbor model based on random projection to analyze the causes of instability, which can evaluate the combustion stability more comprehensively and accurately, and can give the causes of combustion instability caused by the current boiler, and give a precursory warning, so as to facilitate the adoption of preventive measures in advance, reduce or avoid the occurrence of faults, and provide a combustion state online evaluation method and a combustion abnormal state cause analysis method that is practical in engineering practice, which is of great significance for finding the crux of combustion instability and guiding on-site operators to perform correct operations.

[0042] It should be noted that although the above describes the various steps in a specific order, it does not mean that the steps must be executed in the above specific order. In fact, some of these steps can be executed concurrently or even in a different order as long as the required functions can be achieved.

[0043] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0044] Computer readable storage medium can be a tangible device that holds and stores instructions used by an instruction execution device. Computer readable storage medium can include, for example, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a protruding structure in a groove on which instructions are stored, and any suitable combination thereof.

[0045] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A combustion stability evaluation method, used for evaluating combustion stability and analyzing the causes of combustion instability according to any newly added data records corresponding to indicators in the boiler combustion process, characterized in that: The method comprises the following steps: S1. Obtain a first data set consisting of historical data records of boiler combustion operation, wherein each data record contains historical records of all indicators related to the operating environment and historical records of all indicators related to combustion performance; wherein step S1 includes: S11, obtaining historical data records of at least three months ago to form a first data set; S12, performing data cleaning on each historical data record in the first data set to remove invalid values; S2. Analyze the historical records of all indicators related to combustion performance in the first data set obtained in step S1, divide the normal indicator value range and abnormal indicator value range of each indicator related to combustion performance; dynamically update the normal indicator value range and abnormal indicator value range by using an online learning mechanism; wherein, the following steps are performed for each indicator related to combustion performance: S21, extracting the historical record of the current indicator from each historical data record of the first data set to form a first data subset of the indicator; S22, based on the first data subset of the current indicator, using skewness and style methods to respectively detect whether the historical records of the indicator conform to normal distribution; S23. Divide the current indicator into normal indicator value intervals and abnormal indicator value intervals. For the historical records of indicators that meet the normal distribution, the 3σ criterion is used to divide the value intervals corresponding to [μ-3σ,μ+3σ] into normal indicator value intervals, and the others into abnormal indicator value intervals, where μ is the mean of the historical records of the indicator and σ is the standard deviation. For the historical records of indicators that do not meet the normal distribution, the Tukey test method is used to divide the value intervals corresponding to [Q1–1.5*IQR, Q3+1.5*IQR] is divided into normal index value intervals, and the rest are abnormal index value intervals. Among them, Q1 is the first quartile corresponding to the historical data record of the indicator when the Tukey test method is used, Q3 is the third quartile corresponding to the historical data record of the indicator when the Tukey test method is used, IQR is the interquartile range and IQR=Q3–Q1; S24, adopting an online learning mechanism, dynamically updating the normal value range and abnormal value range of the indicator in a combination of two updating strategies, including: first, for the indicator in the newly added data record that does not conform to the normal distribution and the newly added record is less than Q1-1.5*IQR, dynamically updating the lower limit of the original normal value range of the indicator to Q1-1.5*IQR-ε, where ε is set according to the data characteristics, for the indicator that conforms to the normal distribution and the newly added record is less than μ-3σ, the lower limit of the original normal value range of the indicator is dynamically updated to μ-3σ-ε; for the indicator that does not conform to the normal distribution and the newly added record is greater than Q1-1.5*IQR, dynamically updating the upper limit of the original normal value range of the indicator to Q1-1.5*IQR+ε, for the indicator that conforms to the normal distribution and the newly added record is greater than μ+3σ, dynamically updating the upper limit of the original normal value range of the indicator to μ+3σ+ε; second, after a preset fixed time, re-updating the normal indicator value range and abnormal indicator value range of each combustion performance-related indicator through the process described in steps S1 and S2; S3, judging whether each indicator related to combustion performance in each historical data record is abnormal according to the interval divided in step S2; wherein, the indicators that meet the normal distribution are judged according to the normal and abnormal indicator value intervals divided by the normal distribution, and the indicators that do not meet the normal distribution are judged according to the normal and abnormal indicator value intervals divided by the Tukey test method; S4. Determine the combustion stability evaluation result of each historical data record according to the determination results of all indicators related to the combustion performance in each data record, wherein the combustion stability evaluation result corresponding to the historical data record for which the determination results of all indicators related to the combustion performance are normal is stable; otherwise, the combustion stability evaluation result is unstable; S5. Based on the stability evaluation result of each data record determined in step S4, a mapping relationship between all indicators related to the operating environment in each historical data record in the first data set and the combustion stability evaluation result is constructed, and a K-nearest neighbor model based on random projection is used to evaluate the combustion stability of the newly added data record, wherein the newly added data record contains all indicators related to the operating environment; wherein step S5 includes: S51, extracting the historical records of all indicators related to the operating environment in each historical data record in the first data set and performing vectorized representation, wherein all the vectors constitute a training set, wherein all the indicators related to the operating environment in each historical data record correspond to one vector, and each vector is associated and mapped with the combustion stability evaluation result of the historical data record corresponding to it; S52, vectorizing the newly added records of all operating environment related indicators in the newly added data records as test vectors; S53, uniformly projecting the points of the high-dimensional space into the low-dimensional space by randomly projecting the test vector and each vector in the training set, to obtain the test vector in the low-dimensional space and the training set in the low-dimensional space; S54, calculating the Euclidean distance between the test vector in the low-dimensional space and each vector in the training set in the low-dimensional space, and selecting a preset number of first neighbor vectors that are closest to the test vector from the training set; S55, calculating the average value of all combustion performance-related indicators in the historical data records corresponding to all the first neighbor vectors screened out in step S54 as the combustion performance-related indicator value corresponding to the test vector; S56. Determine whether each combustion performance-related indicator corresponding to the test vector is normal based on the normal indicator value range and the abnormal indicator value range of each combustion performance-related indicator, and determine the combustion stability evaluation result corresponding to the test vector based on the judgment result of each indicator.

2. A combustion stability evaluation method according to claim 1, characterized in that: The method further comprises: S6. For newly added data records with unstable stability evaluation results, analyze the operating environment related indicators corresponding to the causes of instability.

3. A combustion stability evaluation method according to claim 2, characterized in that: The step S6 comprises: S61, selecting all vectors with stable combustion stability evaluation results from the training set to form a training sample set; S62, mapping the test vector and each vector in the training sample set to a low-dimensional space respectively according to the random projection method described in step 5; S63, calculating the Euclidean distance between the test vector in the low-dimensional space and each vector in the training sample set in the low-dimensional space, and selecting a preset number of second neighbor vectors closest to the test vector from the training sample set; S64, calculating the average value of all indicators related to the operating environment in the historical data records corresponding to all the second neighbor vectors screened out in step S63 as the normal value of the indicators related to the operating environment corresponding to the test vector; S65. Compare all the indicators in the test vector with their corresponding normal values ​​of the operating environment-related indicators. The indicators in the test vector that deviate from the preset threshold value relative to the normal value are indicators that cause unstable combustion.

4. A combustion stability evaluation system for use in the combustion stability evaluation method according to any one of claims 1 to 3, characterized in that: include: Multiple sensors, each of which is configured at a different location on the boiler to collect various indicator data in real time; A storage module, used to store the data collected by the multiple sensors to form data records, each data record containing records of all indicators; A data cleaning module is used to clean the historical data records stored in the storage module and remove invalid values; The data analysis module is used to analyze the characteristics of each combustion performance-related indicator in the historical data records processed by the data cleaning module, and perform normal detection to detect whether the historical records of the indicator conform to the normal distribution. According to the characteristics of each combustion performance-related indicator, the corresponding normal indicator value interval and abnormal indicator value interval are divided. For the historical records of indicators that meet the normal distribution, the 3σ criterion is adopted, and the value interval corresponding to [μ—3σ, μ+3σ] is divided into the normal indicator value interval, and the others are abnormal indicator value intervals, wherein μ is the mean of the historical records of the indicator, and σ is the standard deviation; for the historical records of indicators that do not meet the normal distribution, the Tukey test method is adopted, and [Q1–1.5*IQR, Q3+1.5*IQR] is divided into the normal indicator value interval, and the others are abnormal indicator value intervals, wherein Q1 is the first quartile corresponding to the historical data record of the indicator when the Tukey test method is adopted, and Q3 is the third quartile corresponding to the historical data record of the indicator when the Tukey test method is adopted, and IQR=Q3–Q1; the normal value interval and abnormal value interval of the indicator are dynamically updated through the online learning mechanism; A stability evaluation module, used to determine whether each record of the index related to combustion performance in the data record is normal according to the normal index value interval and the abnormal index value interval of the index related to combustion performance divided by the data analysis module, and determine the combustion stability evaluation result of the data record according to the determination results of all the indexes related to combustion performance in each data record, wherein the combustion stability evaluation result corresponding to the data record whose determination results of all the indexes related to combustion performance are normal is stable, otherwise, the combustion stability evaluation result is unstable; The cause analysis module is used to record and analyze the operating environment related indicators corresponding to the causes of instability of data with unstable combustion stability evaluation results.

5. A computer-readable storage medium, characterized in that: A computer program is included thereon, and the computer program can be executed by a processor to implement the steps of any one of the methods of claims 1 to 3.

6. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the electronic device to implement the steps of the method as claimed in any one of claims 1 to 3.

Images