Exhaust dispersity early warning model, early warning method, equipment and storage medium

Abstract

The invention discloses an exhaust gas dispersion degree early warning model which is obtained through combination of multiple models and is used for monitoring and early warning exhaust gas temperature dispersion degree when combustion parts and a heat channel in a gas turbine are abnormal and combustion is abnormal. An Isolation Forest and an OCSVM are adopted as a model constructed by an anomaly detection algorithm; in a machine starting stage and a machine stopping stage of gas turbine operation, adopting a PCA-based algorithm as a model constructed by an anomaly detection algorithm; and constructing a regression prediction model based on core measuring points in a stable operation stage, a machine starting stage and a machine stopping stage. According to the invention, various types of abnormalities of signals at each monitoring point of operation of the gas turbine in the sensor monitoring system are detected in real time, and an alarm is given according to the early warning model and the specified early warning rule, so that a user can find signal abnormalities related to equipment faults in time.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and early warning of gas generator sets, and in particular to an exhaust dispersion early warning model, an early warning method, a device and a storage medium. Background Art

[0002] Exhaust temperature dispersion is one of the signals used to determine whether combustion is abnormal in gas-fired generator sets in the prior art. When combustion components and heat channels are abnormal or combustion is abnormal, the exhaust temperature flow field will be seriously uneven and the exhaust dispersion will be large. Therefore, the exhaust temperature dispersion needs to be monitored at all times during operation.

[0003] Exhaust temperature dispersion relies on conventional sensors, which are generally not convenient for timely detection of various signal abnormalities related to equipment failures, and are not convenient for early warning monitoring to timely detect and solve problems.

[0004] At the same time, since the signal change rules in different sensor monitoring methods are different, the anomalies that users are concerned about and the corresponding warning rules may also be different, so different models are constructed for different monitoring methods. The main differences between these models are reflected in the choice of anomaly detection algorithms. The similarity is that the business logic of the models for users is basically similar.

[0005] To this end, the present application specifically proposes an exhaust dispersion warning method to solve the above technical problems. Summary of the invention

[0006] The main purpose of the present invention is to provide an exhaust dispersion warning model to solve the technical problems raised in the background technology.

[0007] The present invention adopts the following technical solutions to solve the above technical problems:

[0008] An exhaust gas dispersion warning model is obtained by combining multiple models. Modeling is performed in the stable operation stage, the start-up stage and the stop-up stage of the engine respectively. It is used to monitor and warn the exhaust gas temperature dispersion when the combustion components and heat channels in the engine are abnormal and the combustion is abnormal. The model includes:

[0009] In the stable operation stage of the gas turbine, the model constructed by using Isolation Forest and OCSVM as the anomaly detection algorithm;

[0010] In the start-up and stop-up stages of the gas turbine operation, the PCA-based algorithm is used as the model constructed as the anomaly detection algorithm;

[0011] A regression prediction model is constructed based on core measurement points in the stable operation stage, the machine startup stage, and the machine shutdown stage.

[0012] Preferably, the stable operation stage is set to the stable power generation stage of the gas turbine under a load greater than 250KW, the startup stage is set to the time period when the load is lower than 250KW, the ignition signal increases from 0 to 1, and the gas turbine speed continues to rise, and the shutdown stage is set to the time period when the load is lower than 250KW, the ignition signal decreases from 1 to 0, and the gas turbine speed continues to decrease.

[0013] Preferably, the monitoring and early warning content specifically includes:

[0014] (a) For all measurement points:

[0015] Sensor abnormality that exceeds the normal range of the sensor: According to the upper and lower limits of the measurement point abnormality, a single-point alarm will generate an early warning form;

[0016] Sensor dead point abnormality: if there is no change in the specified number of digits after the decimal point for 10 consecutive data, a comprehensive alarm will be generated, that is, an early warning sheet will be generated;

[0017] (b) For core measurement points and general measurement points:

[0018] Abnormality beyond the upper and lower limits of the control system alarm: According to the upper and lower limits of the control system, a single point alarm will generate an early warning form;

[0019] (c) Core measurement points for building machine learning models to detect anomalies in real time:

[0020] When the core measuring point is under stable working conditions, the signal value is larger or smaller than the historical normal state, and this abnormal state appears continuously;

[0021] Under start-stop conditions, the shape of the signal value change over time at the core measurement point is inconsistent with the historical normal state, and the difference exceeds the preset limit.

[0022] Preferably, the method for constructing the model comprises:

[0023] S1. Data collection: The real operating data of the internal combustion engine unit in the specified historical time period is exported from the exhaust temperature dispersion monitoring system. The data set contains time series data from a total of 45 sensor measurement points, of which the sampling frequency of the time series data of the stable operating condition is 2 minutes; the sampling frequency of the time series data of the start-stop stage is 1 second;

[0024] S2. Perform preprocessing operations on the collected data set;

[0025] S3. Based on the data after the preprocessing operation, modeling is performed in the stable operation stage of the gas turbine, the start-up stage of the machine, and the stop-up stage of the machine;

[0026] S4. Evaluate the indicators of the model data;

[0027] S5. Based on the qualified indicator data, test the model effect through abnormal or normal data of core measurement points;

[0028] S6. Optimize the model algorithm based on the test results.

[0029] Preferably, the specific operation flow of preprocessing the data set in step S2 includes:

[0030] S21. Remove rows or columns of non-numeric type elements including null values ​​and character strings from the time series data corresponding to each measurement point;

[0031] S22. Divide the data into working conditions, obtain data corresponding to three working conditions, and only retain the time series data of the pre-selected measurement points of interest.

[0032] Preferably, the indicator evaluation method in step S4 includes:

[0033] S41. Output an outlier score using an outlier detection algorithm, and convert the score into an outlier label using a threshold of the score;

[0034] S42. Determine whether the threshold selection of the score meets the preset conditions based on precision and recall to measure the effectiveness of the set-based retrieval: If the threshold selection cannot minimize the number of declared outliers, the threshold output is a false negative, indicating that the algorithm misses the real outliers;

[0035] S43. For any given threshold t on the outlier score, the set of declared outliers is denoted by S(t), and for a dataset D with true positives G, there exists:

[0036]

[0037]

[0038] Where TPR(t) is the true positive rate based on a given threshold t, FPR(t) is the false positive rate based on a given threshold t, Recall(t) is the recall rate, and BadRecall(t) is the negative recall rate.

[0039] Preferably, the specific operation steps of optimizing the model algorithm in step S6 include:

[0040] S61. Add the test results that meet the preset requirements to the data set for re-preprocessing, and re-execute the regression prediction model building operation of step S3 to update the regression prediction model;

[0041] S62. Based on the regression prediction result of the regression prediction model, if the regression prediction result is larger than the true value, the root mean square error during training is subtracted; if the regression prediction result is smaller than the true value, the root mean square error during training is added to make the regression prediction result closer to the true value.

[0042] An exhaust dispersion warning method uses any of the exhaust dispersion warning models to perform a warning operation, including: setting the relevant parameters passed in when calling the exhaust dispersion warning model through a front-end interface, and then calling the exhaust dispersion warning model of the back-end at a specified time interval, and each time the back-end is called, the result of the model calculation a period of time before the calling time is returned, and if the abnormality that occurred in the previous period meets the preset conditions, a warning work order is returned, wherein the back-end includes:

[0043] Module 1: Used to access real-time data at the second level and store it in the database to be used as data for model inference;

[0044] Module 2: It is used to call the early warning model at the specified time interval to infer the current data in real time, and then store the inference results in the result data table in chronological order. The result data in the table is used for early warning statistics in module 3;

[0045] Module 3: used to respond to front-end calls, return the model inference results at a specified time point before the current moment according to the parameters sent by the front-end, and count the reverse model inference results according to the early warning rules. If the preset threshold is reached, an early warning ticket is generated and returned to the front-end for display.

[0046] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0047] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0048] It can be seen from the above technical solution that the present invention provides an exhaust dispersion warning model and a warning method. Compared with the prior art, the present invention has the following advantages:

[0049] 1. The present invention detects various types of abnormalities in the signals at various monitoring points of the gas turbine operation in the sensor monitoring system in real time, and issues an alarm according to the specified early warning rules based on the early warning model, so that users can promptly discover signal abnormalities related to equipment failures, and then promptly discover and solve problems through diagnostic analysis and equipment fault troubleshooting.

[0050] 2. The present invention adopts different abnormality detection algorithm modeling at different stages of gas turbine operation, which can accurately identify the abnormal exhaust temperature dispersion under different working conditions, thereby achieving the effect of improving the accuracy of early warning, and finally being able to timely discover potential problems of equipment and take measures.

[0051] 3. The present invention evaluates the model by introducing an outlier detection algorithm and selects a suitable threshold based on the true positive rate and the false positive rate, which can effectively control the balance between false alarms and missed alarms, thereby achieving the effect of optimizing the performance of the early warning system, and finally ensuring that the information received by the user is neither too sensitive nor too insensitive.

[0052] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become easy to understand through the following description. Of course, it is not necessary to achieve all of the advantages described above simultaneously for any product implementing the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0054] Figure 1 It is a schematic diagram of the modularization of the model structure of the present invention;

[0055] Figure 2 It is a schematic diagram of the overall process of the model building method of the present invention;

[0056] Figure 3 It is a schematic diagram of the back-end structure of the method of the present invention;

[0057] Figure 4 It is a schematic diagram of the back-end response flow of the method of the present invention;

[0058] Figure 5 It is a line graph of signal value changes of the core measuring point in the stable operation stage of the present invention;

[0059] Figure 6 It is a line graph of signal value changes of the core measuring point in the stage of starting the machine according to the present invention;

[0060] Figure 7 This is a line graph of signal value changes at the core measuring point in the machine stopping stage of the present invention. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. In the absence of conflict, the embodiments in this application and the features in the embodiments can be combined with each other. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] In the embodiment, see Figures 1 to 7 .

[0063] like Figure 1 As shown, an exhaust dispersion warning model proposed in an embodiment of the present invention is an equipment fault warning model designed for DLN2.0 model gas turbines. The exhaust dispersion warning model is obtained by combining multiple models, and is modeled respectively in the stable operation stage of the gas turbine, the machine start-up stage and the machine stop stage. It is used to monitor and warn the exhaust temperature dispersion when abnormalities occur in the combustion components and heat channels of the gas turbine and abnormal combustion occurs.

[0064] The stable operation stage refers to the stage of stable power generation of the gas turbine under a load greater than 250KW, and this operating condition lasts the longest; the startup stage refers to the process in which the load is lower than 250KW, the ignition signal increases from 0 to 1, and the gas turbine speed continues to rise. The average duration of this operating condition is one hour each time; the shutdown stage refers to the process in which the load is lower than 250KW, the ignition signal decreases from 1 to 0, and the gas turbine speed continues to decrease. The average duration of this operating condition is 20 minutes each time.

[0065] It should be noted that the prediction model constructed in the present application is constructed for the exhaust temperature dispersion monitoring system. The exhaust temperature dispersion is a signal used to determine whether the combustion is abnormal. When the combustion components and heat channels are abnormal and the combustion is abnormal, the exhaust temperature flow field will be seriously uneven and the exhaust dispersion will be large. Therefore, the exhaust temperature dispersion needs to be monitored at all times during operation.

[0066] Furthermore, the exhaust dispersion warning model includes:

[0067] In the stable operation stage of the gas turbine, the model constructed by using Isolation Forest and OCSVM as the anomaly detection algorithm;

[0068] In the start-up and stop-up stages of the gas turbine operation, the PCA-based algorithm is used as the model constructed as the anomaly detection algorithm;

[0069] A regression prediction model is constructed based on core measurement points in the stable operation stage, the machine startup stage, and the machine shutdown stage.

[0070] In summary, by building a machine learning model, we can learn from historical data the data distribution patterns of several key signals in the exhaust dispersion monitoring system when the gas turbine is working normally at most times, and then achieve the ability to detect abnormal signals in real time and automatically during use, and issue an alarm.

[0071] At the same time, by using different anomaly detection algorithms (Isolation Forest and OCSVM, PCA-based algorithms) in different stages of gas turbine operation (stable operation, starting the machine, and stopping the machine), it can accurately identify the abnormal exhaust temperature dispersion under different operating conditions, thereby improving the accuracy of early warning, and finally timely discover potential equipment problems and take measures.

[0072] Furthermore, the monitoring and warning contents of the exhaust dispersion warning model specifically include:

[0073] (a) For all measurement points:

[0074] Sensor abnormality that exceeds the normal range of the sensor: According to the upper and lower limits of the measurement point abnormality, a single-point alarm will generate an early warning form;

[0075] Sensor dead point abnormality: if there is no change in the specified number of digits after the decimal point for 10 consecutive data, a comprehensive alarm will be generated, that is, an early warning sheet will be generated;

[0076] (b) For core measurement points and general measurement points:

[0077] Abnormality beyond the upper and lower limits of the control system alarm: According to the upper and lower limits of the control system, a single point alarm will generate an early warning form;

[0078] (c) The signals of several measurement points are used to build a machine learning model to detect abnormal situations in real time. These measurement points are called core measurement points. At this time, for the core measurement points used to build a machine learning model to detect abnormal situations in real time:

[0079] When the core measuring points are under stable working conditions, the signal values ​​are larger or smaller than the historical normal state, and these abnormal states / abnormal points generally appear continuously;

[0080] Under start-stop conditions, the shape of the signal value change over time at the core measurement point is inconsistent with the historical normal state, and the difference exceeds the preset limit and is generally large.

[0081] At this time, the signal value of the core measuring point is as follows under three working conditions: Figure 5 , Figure 6 and Figure 7 As shown, it can be seen that the signal value changes for each start and shutdown are regular. Therefore, a machine learning model is built for each core measurement point under each working condition.

[0082] Each model usually contains two types of machine learning algorithms, unsupervised anomaly detection algorithm and regression prediction algorithm. The anomaly detection algorithm is used to detect abnormal values ​​of core measurement points, and the regression prediction algorithm is used to fit the "normal state" signal curve, which mainly serves a convenient display purpose.

[0083] Since the signal change rules in different sensor monitoring systems are different, the anomalies that users are concerned about and the corresponding warning rules may also be different, so different models are built for different monitoring systems. The main difference between these models is in the choice of anomaly detection algorithms, and the similarity is that the business logic of the models for users is basically similar.

[0084] At this time, the model detects various anomalies in the signals in the sensor monitoring system in real time, and issues alarms according to certain early warning rules, so that users can promptly discover signal anomalies related to equipment failures, and then promptly discover and solve problems through diagnostic analysis and equipment troubleshooting.

[0085] In summary, by setting detailed monitoring and early warning rules for core measuring points and general measuring points, including situations where the sensor exceeds the normal range and there is no change in continuous data, it can play a role in comprehensively monitoring the signal status, thereby preventing false alarms caused by single or multiple sensor failures, and finally ensuring the reliability and stability of the system.

[0086] Furthermore, the model construction method is as follows Figure 2 As shown, specifically including:

[0087] S1. Data collection: Use the real operating data of the combustion engine unit within the specified historical time period, derived from the exhaust temperature dispersion monitoring system;

[0088] In a specific embodiment, the data used to construct the exhaust temperature dispersion anomaly detection model is the real operation data of a unit in the power plant over a period of time in the past, which is derived from the exhaust temperature dispersion monitoring system. The data set contains time series data from a total of 45 sensor measurement points;

[0089] The sampling frequency of the time series data of the stable operation condition is one line per 2 minutes; the sampling frequency of the time series data of the start-stop stage is one line per 1 second;

[0090] S2. Perform preprocessing operations on the collected data set. The specific operation process includes:

[0091] S21. Remove rows or columns containing non-numeric elements (such as null values, strings, etc.) in the time series data corresponding to each measurement point;

[0092] S22. Divide the data into working conditions, obtain data corresponding to three working conditions, and retain only the time series data of the pre-selected measurement points of interest;

[0093] In a specific embodiment, there are at least four core measurement points used to construct the machine learning model, namely, the average exhaust temperature and three groups of exhaust dispersion;

[0094] S3. Based on the data after the pre-processing operation, modeling is performed in the stable operation stage of the gas turbine, the start-up stage and the stop-up stage, including:

[0095] Under stable working conditions, Isolation Forest and OCSVM are used as anomaly detection algorithms to build the model, and during the startup and shutdown phase, the PCA-based algorithm is used as an anomaly detection algorithm to build the model;

[0096] A regression prediction model is built for each core measurement point under each working condition. The purpose is to provide a rough "normal" recommended value through the regression model when the timing signal of the measurement point is abnormal, which can help users understand the deviation degree of the abnormal value at the moment.

[0097] The regression prediction model is implemented using a multi-layer perceptron;

[0098] S4. Evaluate the indicators of the model data. The specific operation process includes:

[0099] S41. Output an outlier score using a conventional outlier detection algorithm, and convert the score into an outlier label using a threshold of the score;

[0100] S42. Determine whether the threshold selection of the score meets the preset conditions based on precision and recall to measure the effectiveness of set-based retrieval: If the threshold selection is too strict to minimize the number of declared outliers, then the threshold output is a false negative, indicating that the algorithm misses the real outliers. If the algorithm declares too many data points as outliers, it will cause too many false positives. This trade-off can be measured by precision and recall, which are two indicators commonly used to measure the effectiveness of set-based retrieval;

[0101] S43. For any given threshold t on the outlier score, the set of declared outliers is denoted by S(t), and for a dataset D with true positives G, there exists:

[0102]

[0103]

[0104] Where TPR(t) is the true positive rate based on a given threshold t, FPR(t) is the false positive rate based on a given threshold t, Recall(t) is the recall rate, and BadRecall(t) is the negative recall rate;

[0105] At this time, by introducing the outlier detection algorithm to evaluate the model and selecting the appropriate threshold based on the true positive rate and false positive rate, it can effectively control the balance between false alarms and missed alarms, thereby achieving the effect of optimizing the performance of the early warning system, and finally ensuring that the information received by the user is neither too sensitive nor too slow:

[0106] S5. Based on the qualified indicator data, test the model effect through abnormal or normal data of core measurement points;

[0107] The model effect is tested using abnormal or normal data from the core measurement points of the exhaust dispersion system provided by the power plant. Because the test data is not labeled with real labels, the model effect is manually evaluated by power plant professionals from the perspectives of missed detection and false detection by drawing a reasoning result graph.

[0108] In a specific embodiment, the conclusion of the evaluation is that more than 90% of the test results of the current test data are in line with expectations, and among them: (1) three operating conditions for detecting abnormalities: stable operation stage, startup stage, and shutdown stage; (2) core measurement points: exhaust dispersion 1, exhaust dispersion 2, exhaust dispersion 3, and exhaust average temperature; (3) model construction auxiliary measurement points: IGV opening, unit load, and compressor exhaust pressure; finally, the detection effect of the model on normal numbers and abnormal data under each operating condition is listed respectively;

[0109] S6. Optimize the model algorithm based on the test results. The specific operation process includes:

[0110] S61. Select a better regression prediction algorithm, add the test results that meet the preset requirements to the data set for re-preprocessing, and re-execute the regression prediction model building operation of step S3 to update the regression prediction model;

[0111] S62. According to the regression prediction result of the regression prediction model, if the regression prediction result is larger than the true value, the root mean square error during training is subtracted; if the regression prediction result is smaller than the true value, the root mean square error during training is added to make the regression prediction result closer to the true value;

[0112] At this time, by constructing a regression prediction model and continuously adjusting and optimizing the model according to the actual test results, the prediction can be made closer to the actual situation, thereby achieving the effect of improving the prediction accuracy of the model. Finally, it can more accurately determine whether the combustion components and heat channels are in an abnormal state.

[0113] In summary, the model detects various anomalies of signals in the sensor monitoring system in real time and issues alarms according to certain early warning rules, so that users can promptly discover signal anomalies related to equipment failures, and then promptly discover and solve problems through diagnostic analysis and equipment troubleshooting.

[0114] Since the signal change patterns in different sensor monitoring systems are different, the anomalies that users are concerned about and the corresponding warning rules may also be different, so different models are constructed for different monitoring systems.

[0115] The main difference between these models lies in the choice of anomaly detection algorithms, but the similarity is that the business logic of the models for users is basically similar. The model is built for the exhaust temperature dispersion monitoring system. The exhaust temperature dispersion is introduced as a signal to determine whether the combustion is abnormal. When the combustion components and hot channels are abnormal and the combustion is abnormal, the exhaust temperature flow field will be seriously uneven and the exhaust dispersion will be large. Therefore, the exhaust temperature dispersion needs to be monitored at all times during operation.

[0116] At this time, by developing customized anomaly detection models for specific monitoring systems, although there are differences between the models, they all follow similar business logic processes and can adapt to different types of monitoring needs, thereby achieving the effect of being widely used in various scenarios, and finally providing users with flexible and diverse solutions.

[0117] In summary, by building a machine learning model, we can learn from historical data the data distribution patterns of several key signals in the exhaust dispersion monitoring system when the gas turbine is working normally at most times, and then achieve the ability to detect abnormal signals in real time and automatically during use, and issue an alarm.

[0118] On the other hand, the present invention also discloses an exhaust dispersion warning method, which uses the exhaust dispersion warning model in the above embodiment to perform a warning operation, such as Figure 4 As shown in the figure, the user first sets the relevant parameters passed in when calling the model through the front-end interface, such as the alarm threshold and the window length for statistical abnormal results, and then calls the back-end early warning model at a certain time interval, such as once per minute. Each call to the back-end will return the result of the model calculation some time before the call, and if the abnormality that occurred some time ago meets the preset conditions, it will also return an early warning ticket;

[0119] The backend is Figure 3 As shown, including:

[0120] Module 1: Used to access real-time data at the second level and store it in the database to be used as data for model inference;

[0121] Module 2: It is used to call the model to call the early warning model at a certain time interval, for example, once every 5 seconds, to infer the current data in real time, and then store the inference results in the result data table in chronological order. These results will be used for early warning statistics in module 3;

[0122] Module 3: Used to respond to front-end calls, return the model inference results of the period before the current moment according to the parameters sent by the front-end, and count these inverse model inference results according to the early warning rules. If they reach the set threshold, an early warning ticket will be generated and returned to the front-end for display.

[0123] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0124] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0125] In another embodiment provided in the present application, a computer program product including instructions is also provided, which, when executed on a computer, enables the computer to execute any of the exhaust dispersion warning methods in the above embodiments.

[0126] It is understandable that the method provided in the embodiment of the present invention corresponds to the method provided in the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above-mentioned method.

[0127] The embodiment of the present application also provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus.

[0128] Memory, used to store computer programs;

[0129] The processor is used to implement the above-mentioned exhaust dispersion warning method when executing the program stored in the memory.

[0130] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0131] The communication interface is used for communication between the above electronic device and other devices.

[0132] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0133] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.

[0134] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0135] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

[0136] In addition, it should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components in a certain specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0137] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "and / or" appearing in the full text includes three parallel schemes. Taking "A and / or B" as an example, it includes scheme A, or scheme B, or schemes that A and B meet at the same time. In addition, in the embodiments of the present invention, "multiple" refers to more than two. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

Claims

1. An exhaust dispersion warning model, characterized in that: Through the combination of multiple models, models are respectively built in the stable operation stage of the gas turbine, the start-up stage and the stop-down stage, which are used to monitor and warn the exhaust temperature dispersion when abnormalities occur in the combustion components and heat channels in the gas turbine and when the combustion is abnormal. The model includes: In the stable operation stage of the gas turbine, the model constructed by using Isolation Forest and OCSVM as the anomaly detection algorithm; In the start-up and stop-down stages of the gas turbine operation, the PCA-based algorithm is used as the model constructed as the anomaly detection algorithm; A regression prediction model is constructed based on core measurement points in the stable operation stage, the machine startup stage, and the machine shutdown stage.

2. The exhaust dispersion warning model according to claim 1, characterized in that: The stable operation stage is set as the stable power generation stage of the gas engine under a load greater than 250KW, the startup stage is set as the time period when the load is lower than 250KW, the ignition signal increases from 0 to 1, and the gas engine speed continues to rise, and the shutdown stage is set as the time period when the load is lower than 250KW, the ignition signal decreases from 1 to 0, and the gas engine speed continues to decrease.

3. The exhaust dispersion warning model according to claim 1, characterized in that: The monitoring and early warning contents specifically include: (a) For all measurement points: Sensor abnormality that exceeds the normal range of the sensor: According to the upper and lower limits of the measurement point abnormality, a single-point alarm will generate an early warning form; Sensor dead point abnormality: if there is no change in the specified number of digits after the decimal point for 10 consecutive data, a comprehensive alarm will be generated, that is, an early warning sheet will be generated; (b) For core measurement points and general measurement points: Abnormality beyond the upper and lower limits of the control system alarm: According to the upper and lower limits of the control system, a single point alarm will generate an early warning form; (c) Core measurement points for building machine learning models to detect anomalies in real time: When the core measuring point is under stable working conditions, the signal value is larger or smaller than the historical normal state, and this abnormal state appears continuously; Under start-stop conditions, the shape of the signal value change over time at the core measurement point is inconsistent with the historical normal state, and the difference exceeds the preset limit.

4. The exhaust dispersion warning model according to claim 1, characterized in that: The method for constructing the model includes: S1. Data collection: The real operating data of the internal combustion engine unit in the specified historical time period is exported from the exhaust temperature dispersion monitoring system. The data set contains time series data from a total of 45 sensor measurement points, of which the sampling frequency of the time series data of the stable operating condition is 2 minutes; the sampling frequency of the time series data of the start-stop stage is 1 second; S2. Perform preprocessing operations on the collected data set; S3. Based on the data after the preprocessing operation, modeling is performed in the stable operation stage of the gas turbine, the start-up stage of the machine, and the stop-up stage of the machine; S4. Evaluate the indicators of the model data; S5. Based on the qualified indicator data, test the model effect through abnormal or normal data of core measurement points; S6. Optimize the model algorithm based on the test results.

5. The exhaust dispersion warning model according to claim 4, characterized in that: The specific operation flow of preprocessing the data set in step S2 includes: S21. Remove rows or columns of non-numeric type elements including null values ​​and character strings from the time series data corresponding to each measurement point; S22. Divide the data into working conditions, obtain data corresponding to three working conditions, and only retain the time series data of the pre-selected measurement points of interest.

6. The exhaust dispersion warning model according to claim 4, characterized in that: The indicator evaluation method in step S4 includes: S41. Output an outlier score using an outlier detection algorithm, and convert the score into an outlier label using a threshold of the score; S42. Determine whether the threshold selection of the score meets the preset conditions based on precision and recall to measure the effectiveness of the set-based retrieval: If the threshold selection cannot minimize the number of declared outliers, the threshold output is a false negative, indicating that the algorithm misses the real outliers; S43. For any given threshold t on the outlier score, the set of declared outliers is denoted by S(t), and for a dataset D with true positives G, there exists: Where TPR(t) is the true positive rate based on a given threshold t, FPR(t) is the false positive rate based on a given threshold t, Recall(t) is the recall rate, and BadRecall(t) is the negative recall rate.

7. The exhaust dispersion warning model according to claim 4, characterized in that: The specific operation steps of optimizing the model algorithm in step S6 include: S61. Add the test results that meet the preset requirements to the data set for re-preprocessing, and re-execute the regression prediction model building operation of step S3 to update the regression prediction model; S62. Based on the regression prediction result of the regression prediction model, if the regression prediction result is larger than the true value, the root mean square error during training is subtracted; if the regression prediction result is smaller than the true value, the root mean square error during training is added to make the regression prediction result closer to the true value.

8. An exhaust dispersion warning method, characterized in that: Using the exhaust dispersion warning model described in any one of claims 1 to 7 to perform a warning operation, comprising: setting the relevant parameters of the calling model through the front-end interface, and then calling the exhaust dispersion warning model of the back-end at a specified time interval, and each time the back-end is called, the result of the model calculation a period of time before the calling time is returned, and if the abnormality that occurred in the previous period meets the preset conditions, a warning work order is returned, wherein the back-end includes: Module 1: Used to access real-time data at the second level and store it in the database to be used as data for model inference; Module 2: It is used to call the early warning model at the specified time interval to infer the current data in real time, and then store the inference results in the result data table in chronological order. The result data in the table is used for early warning statistics in module 3; Module 3: used to respond to front-end calls, return the model inference results at a specified time point before the current moment according to the parameters sent by the front-end, and count the reverse model inference results according to the early warning rules. If the preset threshold is reached, an early warning ticket is generated and returned to the front-end for display.

9. A computer-readable storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the processor is caused to execute the steps of the method according to claim 8.

10. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method as claimed in claim 8.

Images