CN117390496A

CN117390496A - Operation information identification method and system for industrial gas generator set system

Info

Publication number: CN117390496A
Application number: CN202311682459.XA
Authority: CN
Inventors: 孙明达
Original assignee: Liyang Guangdong Energy Saving Technology Co ltd
Current assignee: Liyang Guangdong Energy Saving Technology Co ltd
Priority date: 2023-12-08
Filing date: 2023-12-08
Publication date: 2024-01-12
Anticipated expiration: 2043-12-08
Also published as: CN117390496B

Abstract

The invention relates to the technical field of data processing, in particular to an operation information identification method and an operation information identification system for an industrial gas generator set system. In addition, through identifying each operation record text block to be identified, local operation data log identification can be realized, operation data log identification processing of a text block layer can be met, and operation record text blocks with fault risks in the operation data logs to be identified of the selected industrial gas generator set system can be accurately and reasonably captured, so that fault risk positioning identification of the selected industrial gas generator set system is accurately and efficiently realized.

Description

Operation information identification method and system for industrial gas generator set system

Technical Field

The invention relates to the technical field of data processing, in particular to an operation information identification method and system for an industrial gas generator set system.

Background

In modern industrial production, gas generator systems are one of the important devices, the operating state of which is critical to ensure the stability of the entire production process. However, due to the complex industrial environment and large data volume, it is very challenging to accurately identify and predict faults for the operating information of the gas generating set system.

The traditional method usually analyzes and judges the operation data log manually, and is low in efficiency and easy to influence by individual experience, so that a larger deviation exists in the result. With the development of big data and machine learning technologies, operation information identification methods based on these technologies have attracted a great deal of attention. However, existing methods based on these techniques tend to ignore the fault-risk-free text description vectors contained in the past running data log, resulting in poor recognition accuracy. In addition, these methods generally can only identify the whole running data log, and cannot realize local running data log identification.

Disclosure of Invention

In order to improve the technical problems in the related art, the invention provides an operation information identification method and system for an industrial gas generator set system.

In a first aspect, an embodiment of the present invention provides an operation information identification method for an industrial gas generator set system, which is applied to an operation information identification system, and the method includes:

acquiring an operation data log to be identified of a selected industrial gas generator set system and a past operation data log corresponding to the operation data log to be identified;

performing performance state element mining on a plurality of to-be-recognized operation record text blocks in the to-be-recognized operation data log and a plurality of past operation record text blocks in the past operation data log through a target operation state monitoring network to obtain a plurality of to-be-recognized performance state element vectors and a plurality of past performance state element vectors; the target running state monitoring network is obtained by identifying and debugging fault record text blocks of two running data log learning examples and feature linkage debugging of text description vectors of the two running data log learning examples in non-fault record text blocks;

and determining a fault record text block of an operation data log to be identified of the selected industrial gas generating set system based on the performance state distinguishing variable between the plurality of to-be-identified performance state element vectors and the plurality of past performance state element vectors through the target operation state monitoring network.

Alternatively, the method further comprises:

acquiring a basic operation state monitoring network, wherein the basic operation state monitoring network comprises a first feature mining branch, a second feature mining branch and a fault judging branch, the first feature mining branch is identical to the second feature mining branch, the first feature mining branch is used for mining performance state elements of one operation data log learning example in the two operation data log learning examples, the second feature mining branch is used for mining performance state elements of the other operation data log learning example in the two operation data log learning examples, and the fault judging branch is used for determining fault record text blocks of the two operation data log learning examples based on the performance state elements of the first feature mining branch and the second feature mining branch;

feature linkage debugging of text description vectors of the two operation data log learning examples in a non-fault record text block is carried out through the performance state elements of the first feature mining branch and the second feature mining branch so as to adjust network parameters of the basic operation state monitoring network;

Performing recognition and debugging of fault record text blocks of the two operation data log learning examples through the performance state elements of the first feature mining branch and the second feature mining branch so as to adjust network parameters of the basic operation state monitoring network;

and taking the debugged basic running state monitoring network as the target running state monitoring network.

Alternatively, the feature linkage debugging of the text description vectors of the two operation data log learning examples in the non-fault record text blocks is performed through the performance state elements of the first feature mining branch and the second feature mining branch so as to adjust network parameters of the basic operation state monitoring network, including:

according to a plurality of performance state element vectors in a first operation data log learning example of a target industrial gas generator set system, a plurality of target feature clusters corresponding to the plurality of performance state element vectors are obtained;

acquiring a plurality of performance state element vectors in a second operation data log learning example of the target industrial gas generating set system through the first feature mining branch;

Acquiring pairing views of a plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the second operation data log learning example;

acquiring a plurality of performance state element vectors in a third operation data log learning example of the target industrial gas generating set system through the second feature mining branch, wherein the second operation data log learning example and the third operation data log learning example belong to the two operation data log learning examples;

acquiring pairing views of a plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the third operation data log learning example;

the pairing view points of the performance state element vectors of the non-fault record text blocks and the target feature clusters in the second operation data log learning example are used as adjustable notes of the pairing view points of the performance state element vectors of the non-fault record text blocks and the target feature clusters in the third operation data log learning example, and feature linkage debugging is conducted on the basic operation state monitoring network so as to adjust network parameters of the basic operation state monitoring network;

And taking the paired points of the performance state element vectors of the non-fault recorded text blocks and the target feature clusters in the third operation data log learning example as adjustable notes of the paired points of the performance state element vectors of the non-fault recorded text blocks and the target feature clusters in the second operation data log learning example, and performing feature linkage debugging on the basic operation state monitoring network so as to adjust network parameters of the basic operation state monitoring network.

Alternatively, the first running data log learning example includes a first running data log and a second running data log; the obtaining a plurality of target feature clusters corresponding to the plurality of performance state element vectors according to the plurality of performance state element vectors in the first operation data log learning example of the target industrial gas generator set system comprises:

acquiring a plurality of first performance state element vectors corresponding to the first operation data log through the first feature mining branch;

acquiring a plurality of second performance state element vectors corresponding to the second operation data log through the second feature mining branch;

Clustering operation is carried out on the first performance state element vectors and the second performance state element vectors to obtain target feature clusters.

Alternatively, the clustering operation is performed on the plurality of first performance state element vectors and the plurality of second performance state element vectors to obtain a plurality of target feature clusters, including:

adjusting a plurality of original feature clusters based on the plurality of first performance state element vectors to obtain a plurality of feature clusters to be processed;

and adjusting the plurality of feature clusters to be processed based on the plurality of second performance state element vectors to obtain a plurality of target feature clusters.

Alternatively, the adjusting the plurality of original feature clusters based on the plurality of first performance state element vectors to obtain a plurality of feature clusters to be processed includes:

determining a first quantization difference between the first performance state element vector and a feature cluster to which the first performance state element vector is currently paired, and determining a second quantization difference between the first performance state element vector and a feature cluster to which the first performance state element vector is not currently paired;

Determining a first vector clustering cost variable based on the first quantization difference and the second quantization difference;

and adjusting the feature clusters according to the first vector clustering cost variable until the first vector clustering cost variable meets a set requirement to obtain a plurality of feature clusters to be processed.

Alternatively, before adjusting the feature clustering in accordance with the first vector clustering cost variable, the method further comprises: determining a second vector clustering cost variable based on a first quantization difference between the first performance state element vector and a feature cluster to which the first performance state element vector is currently paired;

adjusting the feature clustering according to the first vector clustering cost variable until the first vector clustering cost variable meets a set requirement, including: and adjusting the characteristic clustering based on the first vector clustering cost variable and the first vector clustering cost variable until the first vector clustering cost variable and the first vector clustering cost variable meet a set requirement.

Alternatively, adjusting the feature clustering according to the first vector clustering cost variable includes:

Performing interval numerical mapping processing on the first performance state element vector to obtain an interval numerical mapping value;

and determining an adjusted feature cluster based on the interval numerical mapping value, the set weight and the current feature cluster so as to adjust the feature cluster.

Alternatively, the obtaining the pairing view of the plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the second running data log learning example includes:

determining quantization differences between the performance state element vectors of the non-fault recorded text blocks and each target feature cluster in the second operation data log learning example, and taking the target feature cluster corresponding to the minimum quantization difference as the target feature cluster matched with the performance state element vectors of the non-fault recorded text blocks in the second operation data log learning example;

and taking the target feature clusters in the second operation data log learning example, which are paired with the plurality of performance state element vectors of the non-fault record text block, as the paired points of the plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the second operation data log learning example.

and taking the thermodynamic statistical condition that a plurality of performance state element vectors of the non-fault recorded text blocks in the second operation data log learning example respectively belong to the target feature clusters as a pairing view of the performance state element vectors of the non-fault recorded text blocks in the second operation data log learning example and the target feature clusters.

Alternatively, the method further comprises:

determining quantization differences between the performance state element vectors of the non-fault record text blocks and each target feature cluster in the second running data log learning example;

and dividing the quantized difference of the performance state element vector of the non-fault recorded text block and each target feature cluster in the second operation data log learning example by the sum of the quantized differences of the performance state element vector of the non-fault recorded text block and each target feature cluster in the second operation data log learning example, wherein the performance state element vector of the non-fault recorded text block belongs to the thermal statistics condition of the target feature clusters in the second operation data log learning example.

Alternatively, adjusting the network parameters of the base operating condition monitoring network includes:

determining a first network training error based on identification information of text description vectors of the non-fault-recorded text blocks for the two operation data log learning examples generated in the characteristic linkage debugging process and adjustable notes corresponding to the two operation data log learning examples;

determining a second network training error based on fault record text block identification information for the two operation data log learning examples generated in the fault record text block identification process;

and adjusting network parameters of the basic operation state monitoring network based on the first network training error and the second network training error.

In a second aspect, the present invention also provides an operation information identification system, including a processor and a memory; the processor is in communication with the memory, and the processor is configured to read and execute a computer program from the memory to implement the method described above.

In a third aspect, the present invention also provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the method described above.

In the technical scheme provided by the embodiment of the invention, performance state element mining is carried out on a plurality of to-be-identified operation record text blocks in to-be-identified operation data logs and a plurality of past operation record text blocks in past operation data logs of a selected industrial gas generating set system respectively to obtain a plurality of to-be-identified performance state element vectors and a plurality of past performance state element vectors; and then, based on the performance state distinguishing variable between each to-be-identified performance state element vector and the corresponding past performance state element vector, determining whether the to-be-identified operation record text block corresponding to each to-be-identified performance state element vector has a fault risk, comprehensively considering the text description vector without the fault risk contained in the past operation data log to carry out operation data log identification, and improving the accuracy of operation data log identification. In addition, through identifying each operation record text block to be identified, local operation data log identification can be realized, operation data log identification processing of a text block layer can be met, and operation record text blocks with fault risks in the operation data logs to be identified of the selected industrial gas generator set system can be accurately and reasonably captured, so that fault risk positioning identification of the selected industrial gas generator set system is accurately and efficiently realized.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a schematic flow chart of an operation information identification method for an industrial gas generator set system according to an embodiment of the present invention.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with aspects of the invention.

It should be noted that the terms "first," "second," and the like in the description of the present invention and the above-described drawings are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order.

The method embodiments provided by the embodiments of the present invention may be performed in an operation information identification system, a computer device, or a similar computing device. Taking the example of operation on an operation information identification system, the operation information identification system may comprise one or more processors (the processor may include, but is not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA) and a memory for storing data, and optionally the operation information identification system may further include a transmission device for communication functions. It will be appreciated by those of ordinary skill in the art that the above-described configuration is merely illustrative and is not intended to limit the configuration of the operational information identification system described above. For example, the operational information identification system may also include more or fewer components than shown above, or have a different configuration than shown above.

The memory may be used to store a computer program, for example, a software program of application software and a module, for example, a computer program corresponding to an operation information identification method for an industrial gas generator set system in an embodiment of the present invention, and the processor executes various functional applications and data processing by running the computer program stored in the memory, that is, implements the method described above. The memory may include high speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid state memory. In some examples, the memory may further include memory remotely located with respect to the processor, the remote memory being connectable to the operation information identification system through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The transmission means is used for receiving or transmitting data via a network. The specific example of the network described above may include a wireless network provided by a communication provider operating the information recognition system. In one example, the transmission means comprises a network adapter (Network Interface Controller, simply referred to as NIC) that can be connected to other network devices via a base station to communicate with the internet. In one example, the transmission device may be a Radio Frequency (RF) module, which is used to communicate with the internet wirelessly.

Referring to fig. 1, fig. 1 is a flow chart of an operation information identification method for an industrial gas generator set system according to an embodiment of the present invention, where the method is applied to the operation information identification system, and further may include steps 110 to 130.

Step 110, acquiring an operation data log to be identified of the selected industrial gas generator set system and a past operation data log corresponding to the operation data log to be identified.

The industrial gas generator set system is equipment for converting gas energy into electric energy by using gas as a power source. Industrial grade gas power plants are commonly used in large-scale plants or power stations to provide a continuous and stable supply of electricity.

The operation data log to be identified refers to various operation data generated in the operation process of the generator set, including but not limited to the changing conditions of parameters such as temperature, pressure, speed and the like, various warnings and error messages which can occur, and the like. These data are recorded for later failure analysis and optimization improvement.

Past operational data logs refer to previously collected and stored genset operational data that may be used as historical references to compare and analyze current operational conditions to help better understand and diagnose possible problems.

The operation of step 110 is described in a specific example: such as monitoring the operation of an industrial gas power generator unit. On a certain day, some abnormal operation data such as a sudden rise in temperature, or excessive pressure fluctuation, etc. are noted. The data are recorded to form a log of operational data to be identified. At the same time, the previous log of operational data was reviewed and it was found that a similar anomaly did not occur over a period of time. This historical data may be used as a reference to analyze and determine what the current problem may be. By comparing the log of operational data to be identified with the log of past operational data, clues to problems may be found, such as finding significant differences in the changes of certain parameters from the historical data, to preliminarily infer which faults may have occurred.

And 120, performing performance state element mining on a plurality of to-be-identified operation record text blocks in the to-be-identified operation data log and a plurality of past operation record text blocks in the past operation data log through a target operation state monitoring network to obtain a plurality of to-be-identified performance state element vectors and a plurality of past performance state element vectors.

The target running state monitoring network is obtained by identifying and debugging fault record text blocks of two running data log learning examples and feature linkage debugging of text description vectors of the non-fault record text blocks of the two running data log learning examples.

The target operation state monitoring network is a model constructed by using machine learning or deep learning technology, and aims to monitor the operation state of the industrial gas generator set system in real time and predict possible faults. The text block of the operation record to be identified is the current operation data log that needs to be processed and analyzed by the target operation state monitoring network. The past run record text block is a run data log generated prior to the run record to be identified for comparison with the current run record to be identified. Performance state element mining is a process of extracting key elements reflecting the performance state of a system by analyzing a running record text block. The to-be-identified performance state element vector is a performance state element mined from the text block of the to-be-identified operation record, and is usually expressed as a vector. The past performance state element vector is a performance state element mined from a past operation record text block, and is also expressed as a vector. An example of a log of operational data learning is a learning sample that typically contains a set of operational data logs and corresponding results (e.g., whether a fault has occurred) for training a target operational state monitoring network. The fault record text block is information about a system fault recorded in the operation data log. Identification and debugging is an optimization process for monitoring a network for a target running state, and the accuracy is improved by continuously identifying and correcting errors. The non-fault record text block is information about the normal operation of the system recorded in the operation data log. Text description vectors are vectors that convert non-faulty recorded text blocks into numerical form by some method (e.g., word embedding), facilitating machine learning or deep learning. Feature linkage debugging is an optimization process of monitoring a network aiming at a target running state, and prediction performance is improved by adjusting feature weights.

For example, there is a gas turbine generator system that has collected a large number of operational data logs. Some of the data logs fail during operation, and these are fault log text blocks. There are also data logs that do not fail during run-time, which are non-failed log text blocks. These data logs are examples of running data log learning.

Firstly, the operation data log learning examples are input into a target operation state monitoring network for identification debugging and feature linkage debugging, so that the network can accurately judge whether faults occur according to the input operation record text blocks.

Then, when a new operation data log to be identified exists, the operation data log to be identified can be decomposed into a plurality of operation record text blocks to be identified, and each text block is subjected to performance state element mining to obtain a performance state element vector to be identified. Meanwhile, a corresponding past operation record text block is acquired, and the performance state element is mined to obtain a past performance state element vector.

Finally, the performance state element vectors to be identified and the past performance state element vectors are input into a target running state monitoring network trained before, if the network predicts that the fault possibly occurs, the possible problems in the running data log to be identified are known, and the possible problems can be processed in time.

In a practical industrial gas generator set system, it may be necessary to monitor a number of parameters, such as rotational speed, temperature, pressure, etc. Each parameter variation may affect the operating state of the system. The values of these parameters thus constitute "performance state elements".

When referring to the "to-be-identified performance state element vector," it refers to the values of various parameters collected in the genset system currently in operation. For example, assuming that the parameters of interest are rotational speed, temperature, and pressure, a particular to-be-identified performance state element vector may look like this: [3500rpm,85 ℃,15bar ]. This means that the current rotational speed is 3500rpm, the temperature is 85 degrees celsius and the pressure is 15 bar.

Similarly, a "past performance state element vector" refers to the value of the same parameter in the genset system at some point in the past. For example, if it is known that the rotational speed, temperature and pressure of the same genset are 3000rpm,75 degrees celsius and 14bar, respectively, one month ago, then this past performance state element vector can be expressed as: [3000rpm,75 ℃,14bar ].

By comparing the 'to-be-identified performance state element vector' with the 'past performance state element vector', whether the operation state of the generator set is changed can be found. For example, if the rotational speed suddenly increases, the temperature increases significantly and the pressure increases, this may mean that the generator set has problems and requires further inspection and maintenance.

In practical applications, the performance state element vector typically contains more parameters and may need to be processed and analyzed by complex mathematical models to accurately determine the operational state of the system.

In other examples, the to-be-identified performance state element vector and the past performance state element vector are explained in detail by way of a more specific example.

For example, an industrial gas generator set is being monitored, focusing on the following four operating parameters: exhaust temperature (T), rotational speed (R), compression ratio (C) and gas flow (F). Each parameter is recorded in real time, forming a large number of operational data logs. Among these parameters, they are considered to be key elements affecting the performance state of the unit.

Now, there is a log of operational data to be identified, which contains a record of operational parameters at a particular point in time (e.g., 2022, 1, 12: 00), as follows:

exhaust temperature: 650 ℃;

rotational speed: 15000rpm;

compression ratio: 10;

gas flow rate: 20m 3/s.

These four parameters can be considered as one to-be-identified performance state element vector: [650, 15000, 10, 20].

At the same time, the record of the operation parameters of the unit in the past normal operation (such as at the same time point in 2021) is also consulted, as follows:

Exhaust temperature: 600 ℃;

rotational speed: 14500rpm;

compression ratio: 9.8;

gas flow rate: 18m 3/s.

These four parameters form a past performance state element vector: [600, 14500,9.8, 18].

Next, it is possible to analyze whether there is an abnormality in the operation state of the unit by comparing the two vectors. For example, if one or some elements in the performance state element vector to be identified change too much as compared to the past performance state element vector, this may mean that the operation state of the unit has changed, requiring further inspection and analysis.

And 130, determining a fault record text block of the to-be-identified operation data log of the selected industrial gas generator set system based on the performance state distinguishing variable between the to-be-identified performance state element vectors and the past performance state element vectors through the target operation state monitoring network.

Wherein the performance state distinguishing variable in this context refers to a difference value obtained by comparing the to-be-identified performance state element vector with the past performance state element vector. This difference value can help understand that the current operating state of the gas generator set is different from that in the past, and thus determine possible problems.

For example, assume that the performance state element vector to be identified is [3000rpm, 85 ℃,15bar ], and the past performance state element vector is [3000rpm,75 ℃,14bar ]. The difference values of the various parameters can be obtained by simple subtraction calculations, i.e. [500rpm,10 ℃,1bar ]. This is the performance state discrimination variable.

Further, this performance state discrimination variable may be input into the target operating state monitoring network. If the network predicts a possible failure, then the possible problems in the log of operational data to be identified are known and can be further processed.

For the solution recorded in step 130, for example, when the target operating condition monitoring network receives the performance condition distinguishing variable, it may tell that an increase in rotational speed, an increase in temperature, and an increase in pressure may mean that some components of the generator set are overheating, which may cause serious failure. Thus, it is necessary to immediately stop operation and check the relevant components of the generator set.

Therefore, the possible problems are found out by comparing the to-be-identified performance state element vector with the past performance state element vector, and the possible problems are processed in advance so as to ensure the safe and stable operation of the gas generator set.

Steps 110-130 are described below by way of a complete example.

First, an operating industrial gas generator system is selected, and an operation data log to be identified is collected. This log is assumed to include all operational data for day 2022, 1. At the same time, a corresponding log of past operational data is also obtained, which contains all operational data for the same day as year 2021, month 1, day 1 (i.e., one year ago).

Next, the two logs are processed using a pre-trained target operating condition monitoring network. When the network is trained, a large number of operation data log learning examples are used, and the network can accurately extract performance state elements from operation record text blocks through recognition debugging and feature linkage debugging.

Specifically, each operation record (such as parameters of rotation speed, temperature and the like) in the operation data log to be identified is regarded as a text block of the operation record to be identified, and then the text block is processed by a network to obtain a performance state element vector to be identified. Similarly, each running record in the past running data log is regarded as a text block of the past running record, and the text block is processed by a network to obtain a past performance state element vector.

For example, assume that the network, when processing a block of run-record text to be identified, gets the following performance state element vector: [650, 15000, 10, 20] (corresponding to temperature, rotation speed, compression ratio, and gas flow rate). When the past operation record text block is processed, the following performance state element vectors are obtained: [600, 14500,9.8, 18].

And finally, inputting the to-be-identified performance state element vectors and the past performance state element vectors into a target running state monitoring network. The network calculates a performance state distinguishing variable between the two vectors, and if the difference exceeds a certain threshold value, the network judges that a fault record exists in the operation data log to be identified.

In this example, because there is a significant difference in the to-be-identified performance state element vector from the past performance state element vector (e.g., the temperature rises from 600 ℃ to 650 ℃), the network may predict that there is a fault record in the running data log for the day 2022, 1. Therefore, the prediction result can be further checked and maintained, and the safe and stable operation of the generator set is ensured.

By applying the technical scheme, performance state element mining is carried out on a plurality of to-be-identified operation record text blocks in to-be-identified operation data logs and a plurality of past operation record text blocks in past operation data logs of the selected industrial gas generator set system respectively to obtain a plurality of to-be-identified performance state element vectors and a plurality of past performance state element vectors; and then, based on the performance state distinguishing variable between each to-be-identified performance state element vector and the corresponding past performance state element vector, determining whether the to-be-identified operation record text block corresponding to each to-be-identified performance state element vector has a fault risk, comprehensively considering the text description vector without the fault risk contained in the past operation data log to carry out operation data log identification, and improving the accuracy of operation data log identification. In addition, through identifying each operation record text block to be identified, local operation data log identification can be realized, operation data log identification processing of a text block layer can be met, and operation record text blocks with fault risks in the operation data logs to be identified of the selected industrial gas generator set system can be accurately and reasonably captured, so that fault risk positioning identification of the selected industrial gas generator set system is accurately and efficiently realized.

In detail, by comparing the performance state element vector to be identified with the past performance state element vector, the operation record text block possibly having a fault risk can be accurately found. The method avoids simply relying on threshold judgment or manual inspection, thereby greatly improving the accuracy of identification. Furthermore, individual recognition of each of the to-be-recognized running record text blocks is possible, rather than merely a global determination of the entire running data log. This means that the operating state of the system can be analyzed more carefully and even specific problem areas or equipment can be located, providing valuable information for subsequent maintenance and repair. Further, the past operation data logs, especially those text description vectors without fault risk, are fully utilized. Therefore, the accuracy of identification can be improved, the performance parameter range in the normal running state can be better understood, and references are provided for future prediction and optimization. Finally, by accurately identifying the text blocks of the operation records with the fault risk, potential problems can be discovered earlier, and early warning and intervention can be performed even before the fault occurs. The method has very important effects on ensuring the stable operation of the industrial gas generator set system, preventing serious faults and reducing the downtime.

In conclusion, the technical scheme provides an efficient, accurate and real-time operation state monitoring and fault early warning method for the industrial gas generator set system, and has remarkable help for improving the safety, stability and operation efficiency of the system.

In some alternative embodiments, the method further comprises steps 210-240.

Step 210, obtaining a basic operation state monitoring network, where the basic operation state monitoring network includes a first feature mining branch, a second feature mining branch and a fault discrimination branch, where the first feature mining branch is the same as the second feature mining branch, the first feature mining branch is used to mine a performance state element of one operation data log learning example of the two operation data log learning examples, the second feature mining branch is used to mine a performance state element of another operation data log learning example of the two operation data log learning examples, and the fault discrimination branch is used to determine fault record text blocks of the two operation data log learning examples based on the performance state elements mined by the first feature mining branch and the second feature mining branch.

Step 220, feature linkage debugging of text description vectors of the non-fault record text blocks of the two operation data log learning examples is performed through the performance state elements of the first feature mining branch and the second feature mining branch so as to adjust network parameters of the basic operation state monitoring network.

Step 230, performing recognition and debugging of fault record text blocks of the two operation data log learning examples through the performance state elements of the first feature mining branch and the second feature mining branch so as to adjust network parameters of the basic operation state monitoring network.

Step 240, using the debugged basic operation state monitoring network as the target operation state monitoring network.

In this alternative embodiment, a new concept is introduced: the basic operation state monitoring network is a pre-built network structure and comprises two characteristic mining branches and a fault distinguishing branch. Two feature mining branches are used to handle different running data log learning examples, and a fault discrimination branch is used to predict faults based on these feature mining results.

First, this basic operation state monitoring network is acquired. In order for it to be better able to adapt to the problem, some adjustments need to be made to it. This involves the following steps.

Processing two running data log learning examples using the first feature mining branch and the second feature mining branch, and obtaining respective performance state elements. And then, performing feature linkage debugging on the text description vector of the non-fault record text block according to the performance state elements. This process can be seen as an optimization of network parameters so that the network can better extract valuable information from non-faulty recordings.

Similarly, the fault log text block is also identified and debugged to further optimize network parameters. This process enables the network to more accurately identify and predict faults.

And finally, taking the debugged basic running state monitoring network as a target running state monitoring network. The network is optimized, so that the operation data log of the industrial gas generator set system can be processed more effectively, and fault prediction is performed.

Through the design, the basic operation state monitoring network is optimized through feature linkage debugging and identification debugging, so that the basic operation state monitoring network can extract performance state elements more accurately and conduct fault prediction. The trained target running state monitoring network has better generalization capability by processing and debugging different types of running data log learning examples (including fault records and non-fault records), and can adapt to various complex running conditions.

In some exemplary embodiments, feature coordinated debugging of text description vectors of the two running data log learning examples in non-fault-recorded text blocks is performed through the performance state elements of the first feature mining branch and the second feature mining branch in step 220 to adjust network parameters of the basic running state monitoring network, including steps 221-227.

Step 221, obtaining a plurality of target feature clusters corresponding to the plurality of performance state element vectors according to the plurality of performance state element vectors in the first operation data log learning example of the target industrial gas generator set system.

Step 222, obtaining a plurality of performance state element vectors in a second operation data log learning example of the target industrial gas generator set system through the first feature mining branch.

Step 223, obtaining pairing views of the plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the second operation data log learning example.

Step 224, obtaining a plurality of performance state element vectors in a third operation data log learning example of the target industrial gas generating set system through the second feature mining branch, wherein the second operation data log learning example and the third operation data log learning example belong to the two operation data log learning examples.

Step 225, obtaining pairing views of the plurality of performance state element vectors of the non-fault record text block and the plurality of target feature clusters in the third operation data log learning example.

Step 226, taking the paired points of the performance state element vectors of the non-fault record text block and the target feature clusters in the second operation data log learning example as adjustable notes of the paired points of the performance state element vectors of the non-fault record text block and the target feature clusters in the third operation data log learning example, and performing feature linkage debugging on the basic operation state monitoring network to adjust network parameters of the basic operation state monitoring network.

Step 227, using the paired points of the performance state element vectors of the non-fault record text block and the target feature clusters in the third operation data log learning example as adjustable notes of the paired points of the performance state element vectors of the non-fault record text block and the target feature clusters in the second operation data log learning example, and performing feature linkage debugging on the basic operation state monitoring network to adjust network parameters of the basic operation state monitoring network.

In the above embodiment, the first feature mining branch and the second feature mining branch are components of the target running state monitoring network and are responsible for processing different running data log learning examples and mining out the performance state element vector. The target feature clustering is a data clustering method, and can help better understand and describe the operation state of the industrial gas generator set system by classifying similar performance state element vectors into one type. Pairing perspectives are a method for training a monitoring network by comparing performance state element vectors of the same feature cluster in different running data log learning examples, key factors that may lead to failure can be found. The tunable annotation is a strategy for optimizing the monitoring network, and can improve the recognition accuracy of the network by dynamically adjusting the weight of the pairing viewpoint.

Further, steps 221-227 generally describe how to use the first and second feature mining branches and target feature clusters for feature linkage debugging to adjust parameters of the underlying operating condition monitoring network. Specifically, a first feature mining branch is used for processing a first operation data log learning example, a performance state element vector is obtained, and clustering is carried out, so that a target feature cluster is obtained. Then, the second and third running data log learning examples are processed with a second feature mining branch, obtaining respective performance state element vectors. Next, paired perspectives of the performance state element vector and the target feature cluster in the non-fault-recorded text blocks in the second and third run data log learning examples are found. These pairing perspectives may help understand performance parameter variations under different operating conditions. And finally, taking the paired views as adjustable notes, and inputting the adjustable notes into a basic running state monitoring network to carry out feature linkage debugging. By the method, the parameters of the network can be further optimized, and the identification accuracy of the network can be improved.

In this way, by using the feature mining branch and feature linkage debugging, the operation record text block which may have fault risk can be more accurately identified. By using the target feature clustering and pairing viewpoints, the past operation data can be fully utilized, and the accuracy and reliability of identification are improved. By using the adjustable annotation, the parameters of the network can be dynamically optimized to adapt to the continuously changing running state, and the accuracy and stability of identification are further improved.

In some examples, the first running data log learning example includes a first running data log and a second running data log. Based on this, in step 221, a plurality of target feature clusters corresponding to the plurality of performance state element vectors are obtained according to the plurality of performance state element vectors in the first operation data log learning example of the target industrial gas generating set system, including steps 2211-2213.

Step 2211, obtaining a plurality of first performance state element vectors corresponding to the first operation data log through the first feature mining branch.

Step 2212, obtaining a plurality of second performance state element vectors corresponding to the second operation data log through the second feature mining branch.

Step 2213, clustering the first performance state element vectors and the second performance state element vectors to obtain target feature clusters.

In the above example, two running data logs (a first running data log and a second running data log) are combined into one learning example, and a performance state element vector is acquired by using a feature mining branch, and then a clustering operation is performed to obtain a target feature cluster.

First, a first operational data log is processed using a first feature mining branch from which a number of first performance state element vectors are obtained. For example, if the first operation data log records parameters such as temperature, pressure, etc. of the generator set, the first feature mining branch outputs a performance state element vector corresponding to the parameters.

Similarly, a second operational data log is processed using a second feature mining branch to obtain a number of second performance state element vectors.

Next, a clustering operation is performed on the first performance state element vector and the second performance state element vector. Clustering can be understood as a clustering method that classifies similar performance state element vectors into a class, forming a feature cluster. In this way, several clusters of target features are obtained, each cluster representing a particular operating state.

Thus, the performance state elements can be more comprehensively and accurately mined by integrating the first operation data log and the second operation data log. This is very helpful for subsequent fault prediction and localization. Clustering operations may classify similar performance state element vectors into a class, forming a feature cluster. Therefore, the complexity of data can be reduced, and various different running states can be intuitively understood, so that the interpretation of the model is improved.

In some preferred embodiments, the clustering operation is performed on the first performance state element vectors and the second performance state element vectors in step 2213 to obtain target feature clusters, which includes steps 22131-22132.

Step 22131, adjusting the plurality of original feature clusters based on the plurality of first performance state element vectors to obtain a plurality of feature clusters to be processed.

And 22132, adjusting the plurality of feature clusters to be processed based on the plurality of second performance state element vectors to obtain a plurality of target feature clusters.

In this embodiment, the following new concepts are introduced: the first performance state element vector and the second performance state element vector are performance state parameters extracted from the run data log learning example by the feature mining branch. The original feature clusters are a group of feature clusters predefined based on historical data, and the feature clusters to be processed are results obtained after the original feature clusters are adjusted.

First, the original feature clusters are adjusted according to the first performance state element vector. This process may involve changing the number, size, or shape, etc. of clusters to better match the current performance state. And then, further adjusting the feature cluster to be processed according to the second performance state element vector to finally obtain the target feature cluster. The process ensures that the characteristic clustering can fully reflect the actual running state of the industrial gas generator set system.

Therefore, through dynamic adjustment of the feature clusters, the running state of the industrial gas generator set system can be more accurately represented and identified, and therefore the accuracy of fault prediction is improved. The method allows the feature clustering to be dynamically adjusted according to actual conditions, so that the model can be better adapted to different running conditions, and has stronger generalization capability. By adjusting only the key feature clusters, the computing resources can be utilized more effectively, and the processing efficiency is improved.

In other preferred embodiments, the adjusting the number of original feature clusters based on the number of first performance state element vectors in step 22131 results in a number of feature clusters to be processed, including steps 221311-221313.

Step 221311, determining a first quantization difference between the first performance state element vector and the feature cluster to which the first performance state element vector is currently paired, and determining a second quantization difference between the first performance state element vector and the feature cluster to which the first performance state element vector is not currently paired.

Step 221312 determines a first vector clustering cost variable based on the first quantized difference and the second quantized difference.

And 221313, adjusting the feature clusters according to the first vector clustering cost variable until the first vector clustering cost variable meets a set requirement to obtain a plurality of feature clusters to be processed.

In this embodiment, a new optimization method is introduced: and dynamically adjusting the original feature clusters based on the performance state element vector to obtain the feature clusters to be processed. This process involves calculating quantization differences and vector clustering cost variables to determine how best to make the feature clustering adjustments.

First, a first quantization difference between a first performance state element vector and a feature cluster to which it is currently paired, and a second quantization difference between a feature cluster to which it is not currently paired, need to be determined. These two quantization differences can help understand whether the performance state element vector is properly categorized into a certain feature cluster.

Then, a first vector clustering cost variable is determined from the first quantization difference and the second quantization difference. The vector clustering cost variable is a metric representing the "cost" required to move the performance state element vector from the currently paired feature cluster to another feature cluster.

And then, adjusting the feature clusters according to the first vector clustering cost variable until the cost variable meets the set requirement, thereby obtaining a series of feature clusters to be processed. This process can be seen as an optimization problem with the goal of making the classification of all performance state element vectors in the feature clusters as accurate as possible.

By the method, the original characteristic clustering can be dynamically adjusted to be better matched with actual operation data, so that the accuracy of fault prediction and positioning is improved. By dynamically adjusting the feature clusters, the model can be better adapted to actual operation data, so that the accuracy of fault prediction and positioning is improved. By introducing a vector clustering cost variable and setting requirements, the feature clustering can be flexibly adjusted according to actual requirements, so that the model has better adaptability. By dynamically adjusting the feature clustering, the model has better recognition performance for different running states, so that the robustness of the model is improved.

Under some optional design considerations, the method further includes step 310 before adjusting the feature clusters according to the first vector cluster cost variable as described in step 221313.

Step 310, determining a second vector clustering cost variable based on a first quantization difference between the first performance state element vector and a feature cluster to which the first performance state element vector is currently paired.

Based on step 310, the adjusting the feature clustering according to the first vector clustering cost variable described in step 221313 until the first vector clustering cost variable meets a set requirement includes: and adjusting the characteristic clustering based on the first vector clustering cost variable and the first vector clustering cost variable until the first vector clustering cost variable and the first vector clustering cost variable meet a set requirement.

In this embodiment, some new concepts are introduced: the first vector clustering cost variable is a parameter that evaluates the effect of feature clustering adjustment. It reflects the degree of difference between the performance state element vector and the feature cluster it is currently paired with. The second vector clustering cost variable is a parameter determined based on the quantization difference between the first performance state element vector and the feature cluster to which it is currently paired. It is used to guide the further adjustment of feature clustering.

First, a second vector clustering cost variable is determined based on the quantization difference between the first performance state element vector and the feature cluster to which it is currently paired. This variable reflects the adjustment requirements of the current feature cluster.

And then, according to the first vector clustering cost variable and the second vector clustering cost variable, the characteristic clustering is adjusted until the two cost variables meet the set requirements. In this way, a cluster of features is obtained that more accurately reflects the actual operating conditions.

By the design, the adjustment requirement of the characteristic clustering can be more accurately measured by introducing the second vector clustering cost variable, so that the adjustment can be more accurately performed. By simultaneously considering the first vector clustering cost variable and the second vector clustering cost variable in the characteristic clustering adjustment process, the problem of over-fitting or under-fitting caused by a single parameter can be avoided, and the stability of the model is improved.

In still other embodiments, adjusting the feature clustering in accordance with the first vector clustering cost variable as described in step 221313 comprises: performing interval numerical mapping processing on the first performance state element vector to obtain an interval numerical mapping value; and determining an adjusted feature cluster based on the interval numerical mapping value, the set weight and the current feature cluster so as to adjust the feature cluster.

In this embodiment, a feature clustering adjustment method based on interval value mapping and set weights is introduced.

Interval value mapping is a data preprocessing technique, and by mapping the original data to a specific interval (such as between 0 and 1), the data processing process can be simplified, and the data in different ranges can be effectively compared. The set weight is an optimization strategy, and the influence of important features can be highlighted by giving different weights to different features or feature clusters, so that the prediction performance of the model is improved.

First, a section value mapping process is performed on the first performance state element vector to obtain a section value mapping value. For example, if the original performance state element vector has a range of values from 0-100, it can be mapped to between 0-1 by interval value mapping. Then, the adjusted feature clusters are determined based on the interval value mapping values, the set weights, and the current feature clusters. The set weights may be determined based on business needs and empirical knowledge. For example, if a feature is very important for fault prediction, it may be given a greater weight. Finally, the characteristic clusters are adjusted, so that the adjusted characteristic clusters can better reflect the running state of the industrial gas generator set system.

In this way, through interval numerical mapping and weight setting, the model can more accurately identify possible faults, and therefore the accuracy of fault prediction and positioning is improved. The method can reduce the noise influence of the data and improve the recognition performance of the model for different running states, thereby enhancing the robustness of the model. The contribution of each feature or feature cluster to the final result can be clearly seen through interval numerical mapping and weight setting, so that the interpretation of the model is improved.

In some preferred embodiments, the step 223 of obtaining the paired points of view of the plurality of performance state element vectors and the plurality of target feature clusters of the non-fault record text block in the second running data log learning example includes step 2231-step 2232.

Step 2231, determining a quantization difference between the performance state element vector of the non-fault-recorded text block and each target feature cluster in the second running data log learning example, and using the target feature cluster corresponding to the minimum quantization difference as the target feature cluster paired with the performance state element vector of the non-fault-recorded text block in the second running data log learning example.

Step 2232, using the target feature clusters in the second running data log learning example, where the target feature clusters are paired with the plurality of feature state element vectors of the non-fault record text block, as the paired points of the plurality of feature state element vectors of the non-fault record text block in the second running data log learning example.

In this embodiment, a pairing point of how to acquire the performance state element vector of the non-faulty recorded text block and the target feature cluster in the second running data log learning example is described.

First, it is necessary to determine the quantization difference between the performance state element vector of the non-fault-recorded text block and each target feature cluster in the second running data log learning example. The quantization discrepancy may be understood as a metric that reflects the similarity or distance between the performance state element vector and the feature cluster. Then, find out the minimum quantization difference, and regard its correspondent goal characteristic cluster as the goal characteristic cluster paired with element vector of the state of performance.

Next, this paired target feature cluster is taken as a paired point of view of the performance state element vector of the non-faulty recorded text block and the target feature cluster in the second running data log learning example. This means that it is considered that the performance state element vector most likely represents the operational state represented by the target feature cluster for this pairing.

In this way, each performance state element vector can be mapped onto one target feature cluster, thereby obtaining more visual and more accurate running state description. This is very helpful for subsequent fault prediction and localization work. By pairing the performance state element vector with the best matching target feature cluster, the operational record text blocks that may be at risk of failure can be more accurately identified. The pairing view provides an intuitive way to understand the relationship between the performance state element vector and the target feature cluster, enabling a better understanding of the model's predictions. Because the quantized differences are used to find the best matching clusters of target features, the model makes reasonable predictions even in the face of unseen operating conditions.

In other preferred embodiments, the obtaining the pairing point of the performance state element vectors of the non-fault record text block and the target feature clusters in the second running data log learning example in step 223 includes: and taking the thermodynamic statistical condition that a plurality of performance state element vectors of the non-fault recorded text blocks in the second operation data log learning example respectively belong to the target feature clusters as a pairing view of the performance state element vectors of the non-fault recorded text blocks in the second operation data log learning example and the target feature clusters.

In this embodiment, a pairing point of how to acquire the performance state element vector of the non-faulty recorded text block and the target feature cluster in the second running data log learning example is described. The new concept is referred to herein: thermodynamic statistics conditions.

The thermodynamic statistical condition is a measure of the likelihood or probability that a performance state element vector belongs to a cluster of individual target features. Such metrics may reflect the degree of similarity or proximity between the performance state element vector and the respective feature clusters.

First, a plurality of performance state element vectors of non-fault record text blocks in a second operation data log learning example are obtained. Then, the thermodynamic statistical conditions of the performance state element vectors respectively belonging to the target feature clusters are calculated. For example, if the thermodynamic statistics of a certain performance state element vector clustered with a feature are large, then the performance state element vector may be considered more likely to belong to the feature cluster. And finally, taking the calculated thermodynamic statistical condition as a pairing view of the performance state element vector of the non-fault record text block and the target feature cluster in the second operation data log learning example. In this way, an inference or a determination is made as to which feature cluster each performance state element vector should belong to.

Therefore, by introducing the thermodynamic statistics condition, the feature clusters to which the performance state element vector should belong can be judged more accurately, so that the accuracy of fault prediction and positioning is improved. The thermodynamic statistics condition intuitively reflects the similarity or the proximity degree between the element vector of the performance state and each characteristic cluster, and is helpful for understanding and explaining the prediction result of the model.

In some alternative embodiments, the method further comprises steps 410-420.

Step 410, determining a quantization difference between the performance state element vector of the non-fault record text block and each target feature cluster in the second operation data log learning example.

Step 420, dividing the quantized difference between the performance state element vector of the non-fault-recorded text block and each target feature cluster in the second operation data log learning example by the sum of the quantized differences between the performance state element vector of the non-fault-recorded text block and each target feature cluster in the second operation data log learning example, and using the result as a thermodynamic statistical condition that the performance state element vector of the non-fault-recorded text block belongs to the target feature clusters in the second operation data log learning example.

In this embodiment, first, it is necessary to determine the quantization difference between the performance state element vector of the non-faulty recorded text block and each target feature cluster in the second running data log learning example. This difference may be calculated by various distance measurement methods (e.g., euclidean distance, mahalanobis distance, etc.) or similarity measurement methods (e.g., cosine similarity, pearson correlation coefficient, etc.). Then, dividing the quantized difference between each performance state element vector and each target feature cluster by the sum of all quantized differences to obtain a result that the performance state element vector belongs to the thermodynamic statistical condition of each target feature cluster. This process corresponds to normalizing the quantized differences such that the sum of all the thermodynamic statistical conditions is 1.

By the design, the association degree between the performance state element vector and each target feature cluster can be described more accurately by calculating the thermodynamic statistics condition, so that the prediction performance of the model is improved. The thermodynamic statistical condition provides an intuitive probability of each performance state element vector to help understand the relationship between the performance state element vector and each target feature cluster. By carrying out normalization processing on the quantization difference, the influence of the data scale can be eliminated, so that the model has better robustness.

In other examples, adjusting network parameters of the base operating condition monitoring network includes: determining a first network training error based on identification information of text description vectors of the non-fault-recorded text blocks for the two operation data log learning examples generated in the characteristic linkage debugging process and adjustable notes corresponding to the two operation data log learning examples; determining a second network training error based on fault record text block identification information for the two operation data log learning examples generated in the fault record text block identification process; and adjusting network parameters of the basic operation state monitoring network based on the first network training error and the second network training error.

In this embodiment, it is described in detail how to adjust the network parameters of the underlying operating condition monitoring network. The newly introduced concept includes a first network training error, a second network training error.

The first network training error is an error determined according to the identification information of text description vectors of the non-fault-recorded text blocks for two running data log learning examples generated in the characteristic linkage debugging process and adjustable notes corresponding to the examples. The second network training error is an error determined from fault log text block identification information for two running data log learning examples generated in the fault log text block identification process and setting notes corresponding to these examples.

First, it is necessary to determine a first network training error based on identification information of text description vectors of examples in non-fault-recorded text blocks for two running data log learning examples generated in a feature linkage debugging process, and adjustable annotations corresponding to these examples. This error reflects the accuracy of the model in identifying the non-fault condition. Then, it is necessary to determine the second network training error based on the fault-recorded text block identification information for the two running-data log learning examples generated in the identification process of the fault-recorded text block, and the setting notes corresponding to these examples. This error reflects the accuracy of the model in identifying the fault condition. Finally, network parameters of the base operating state monitoring network are adjusted based on the first network training error and the second network training error. This procedure is an optimization problem with the goal of finding a set of parameters such that both errors can be minimized.

Thus, by continuously adjusting the network parameters, the model can be better fitted to the training data, thereby improving the accuracy of the model in identifying non-fault conditions and fault conditions. By considering both the first network training error and the second network training error, the model can exhibit good predictive performance in the face of different types of data, thereby enhancing the robustness of the model. By observing the changes of the first network training error and the second network training error, the learning process and the performance of the model can be better understood, so that the interpretability of the model is improved.

Further, there is also provided a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the above-described method.

In the embodiments provided in the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are merely illustrative, for example, flow diagrams and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

In addition, functional modules in the embodiments of the present invention may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.

The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a network device, or the like) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An operation information identification method for an industrial gas generator set system, which is characterized by being applied to an operation information identification system, comprising the following steps:

2. The method for identifying operational information for an industrial gas genset system of claim 1 further comprising:

3. The method for identifying operation information of an industrial gas generator set system according to claim 2, wherein the performing feature linkage debugging of text description vectors of the two operation data log learning examples in a non-fault record text block through performance state elements of the first feature mining branch and the second feature mining branch to adjust network parameters of the basic operation state monitoring network comprises:

4. The method for identifying operational information for an industrial gas genset system of claim 3 wherein the first operational data log learning example comprises a first operational data log and a second operational data log; the obtaining a plurality of target feature clusters corresponding to the plurality of performance state element vectors according to the plurality of performance state element vectors in the first operation data log learning example of the target industrial gas generator set system comprises:

clustering the first performance state element vectors and the second performance state element vectors to obtain target feature clusters;

the clustering operation is performed on the plurality of first performance state element vectors and the plurality of second performance state element vectors to obtain a plurality of target feature clusters, including:

5. The method for identifying operation information of an industrial gas generator set system according to claim 4, wherein the adjusting the plurality of original feature clusters based on the plurality of first performance state element vectors to obtain a plurality of feature clusters to be processed includes:

6. The method for identifying operational information for an industrial gas genset system of claim 5 further comprising, prior to adjusting the feature clustering based on the first vector clustering cost variable: determining a second vector clustering cost variable based on a first quantization difference between the first performance state element vector and a feature cluster to which the first performance state element vector is currently paired;

7. The method for identifying operational information for an industrial gas generating set system according to claim 5, wherein adjusting the feature clustering based on the first vector clustering cost variable comprises:

8. The method for identifying operation information of an industrial gas generating set system according to claim 3, wherein the obtaining a pairing view of a plurality of performance state element vectors of a non-fault-recorded text block and the plurality of target feature clusters in the second operation data log learning example includes:

9. The method for identifying operation information of an industrial gas generating set system according to claim 3, wherein the obtaining a pairing view of a plurality of performance state element vectors of a non-fault-recorded text block and the plurality of target feature clusters in the second operation data log learning example includes:

taking the thermodynamic statistical condition that a plurality of performance state element vectors of a non-fault record text block in the second operation data log learning example respectively belong to the target feature clusters as a pairing view of the performance state element vectors of the non-fault record text block in the second operation data log learning example and the target feature clusters;

wherein the method further comprises: determining quantization differences between the performance state element vectors of the non-fault record text blocks and each target feature cluster in the second running data log learning example; dividing the quantized difference between the performance state element vector of the non-fault recorded text block and each target feature cluster in the second operation data log learning example by the sum of the quantized differences between the performance state element vector of the non-fault recorded text block and each target feature cluster in the second operation data log learning example, and taking the result as a thermodynamic statistical condition that the performance state element vector of the non-fault recorded text block belongs to the target feature clusters in the second operation data log learning example;

Wherein adjusting the network parameters of the basic operation state monitoring network comprises: determining a first network training error based on identification information of text description vectors of the non-fault-recorded text blocks for the two operation data log learning examples generated in the characteristic linkage debugging process and adjustable notes corresponding to the two operation data log learning examples; determining a second network training error based on fault record text block identification information for the two operation data log learning examples generated in the fault record text block identification process; and adjusting network parameters of the basic operation state monitoring network based on the first network training error and the second network training error.

10. An operation information identification system, comprising a processor and a memory; the processor is communicatively connected to the memory, the processor being configured to read a computer program from the memory and execute the computer program to implement the method of any of claims 1-9.