An alarm data processing method and system based on dynamic labels and neural networks

Abstract

The application discloses an alarm data processing method based on dynamic labels and neural networks, comprising: collecting alarm data; cleaning and repairing the collected alarm data, wherein the cleaning comprises cleaning of invalid alarm data and cleaning of missing alarm data; the repairing comprises repairing of the alarm data; abstracting the cleaned and repaired alarm data, wherein the abstracting comprises: extracting the same feature information from alarm data with the same feature, so as to unify the alarm data into one type of data; performing dynamic label calculation by using a dynamic label calculation model; labeling the alarm data by using the dynamic label; performing mathematical and normalized processing on the labeled power station alarm data set to obtain a training data set; training a neural network model based on the training data set, optimizing the neural network model, saving model parameters, and outputting an intelligent alarm data recognition and classification algorithm based on the dynamic label. Corresponding systems, electronic devices and computer readable storage media are also disclosed.

Description

Technical Field

[0001] This invention belongs to the fields of computer, artificial intelligence and smart power generation technology, and in particular relates to an alarm data processing method and system based on dynamic tags and neural networks. Background Technology

[0002] With the advancement of the national new energy strategy, large-scale photovoltaic, wind power, and other new energy power plants have been put into operation. Due to the large number of devices in new energy power plants, especially photovoltaic power plants (a single plant can have hundreds of thousands of modules), even if the equipment quality meets standards, the sheer number of devices inevitably leads to a large amount of real-time alarm data due to issues with the quality of a few devices, sensors, signal transmission lines, or signal acquisition equipment. Furthermore, the poor classification of alarm data by monitoring and alarm platforms results in a large amount of redundant alarm data, significantly reducing the efficiency of management personnel in processing alarm data and making it very easy to overlook important alarm information. This causes critical alarms to be buried in the massive daily alarm records, thus jeopardizing the safe operation of the power plant.

[0003] Therefore, existing technologies require a method to assist power generation operators in quickly identifying high-priority alarms from a large volume of daily alarm information, and to reclassify large amounts of alarm data to improve its usability. While neural networks are very effective for processing data classification, there are currently no existing technologies that apply them to processing and classifying large amounts of power plant alarm data. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide an alarm data processing method and system based on dynamic tags and neural networks. This belongs to the category of power plant alarm data classification technology and intelligent algorithms based on neural networks. It can quickly identify key alarm information from tens of thousands of alarm messages generated daily at new energy power plants or new energy control centers. This allows for rapid and efficient classification of massive amounts of power plant alarm data, thus assisting power generation operators in their monitoring work. Simultaneously, it can be used to reduce the priority of large amounts of redundant, repetitive, and low-impact alarm data, enabling power generation operators to quickly identify important, high-priority alarm information and ensure timely processing of critical equipment alarms, thereby maintaining power plant safety.

[0005] This invention provides a method for processing alarm data based on dynamic tags and neural networks, comprising:

[0006] S1, collect alarm data;

[0007] S2. Clean and repair the collected alarm data. The cleaning includes cleaning invalid alarm data and missing alarm data; the repair includes repairing alarm data.

[0008] S3. Abstract the alarm data after cleaning and repair. The abstraction includes: extracting the same feature information from alarm data with the same features, so as to unify them into one type of data.

[0009] S4. Perform dynamic label calculation using a dynamic label calculation model.

[0010] S5. Label the alarm data using dynamic labels.

[0011] S6. Perform mathematization and normalization processing on the labeled power station alarm data set to obtain a training data set.

[0012] S7. Train a neural network model based on the training data set, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels.

[0013] Preferably, the alarm data in S1 comes from a power station and includes the following fields: "Alarm Time", "Power Station Code", "Power Station Name", "Equipment Name", "Equipment KKS Code", "Alarm Content", and "Alarm Level".

[0014] Preferably, the dynamic label calculation in S4 using a dynamic label calculation model includes:

[0015] S41. Calculate the probability label calculation value P, including: perform alarm classification statistics according to whether the power station name or power station code, equipment name or equipment KKS code, and alarm content are the same; then calculate the occurrence probability p(x i ) of each type of alarm data. The occurrence probability p(x i ) represents the proportion of the i-th type of alarm data volume to the total alarm data volume, 0 ≤ p(x i ) ≤ 1; according to the size of the occurrence probability p(x i ), assign the corresponding probability label calculation value P, 0 < P ≤ 16.

[0016] S42. Calculate the static label calculation value S, including: assign static label calculation values to each field content of the alarm data, accumulate the static label calculation values of each field, and then perform a 16-point conversion to obtain the static label calculation value S, 0 < S ≤ 16. The calculation formula is:

[0017]

[0018] S43. Calculate the dynamic label calculation value K, including: based on the probability label calculation value P of the alarm data and the static label value S, perform weighted calculation to obtain the dynamic label calculation value K. The calculation formula is K = PR + S(2 - R), where 0 < K ≤ 32, and R is the weight coefficient. The value of the weight coefficient R is adjusted and determined according to the size of the average daily alarm quantity n of the power station.

[0019] Preferably, the S5 includes: annotate the alarm data according to the dynamic label calculation value K, mark the corresponding dynamic label T for the alarm data, and obtain the labeled power station alarm data set; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases.

[0020] Preferably, the mathematization in the S6 includes: converting the text information into mathematized data that can be recognized by the neural network; the normalization process includes: performing normalization on the mathematized data so that the mathematized data is evenly distributed in the range of 0 - 1 and conforms to the normal distribution.

[0021] Preferably, the neural network model is a neural network model with three hidden layers, and sigmoid is used as the activation function.

[0022] Preferably, the training of the neural network model based on the training data set includes the following steps:

[0023] S71. Import the formed alarm data into the neural network model;

[0024] S72. Divide the collected alarm data into multiple data blocks on average;

[0025] S73. Set the learning rate, and calculate the current loss function after training block by block according to the data blocks;

[0026] S74. Adjust the learning rate through the discrete descent method;

[0027] S75. Repeat S73 and S74 to optimize the model until the loss function converges;

[0028] S76. Save the power station alarm data classification model parameters based on the neural network model. After training is completed, obtain the intelligent alarm data recognition and classification algorithm based on dynamic labels. [[ID=3,2]]

[0029] The second aspect of the present invention provides an alarm data processing system based on dynamic labels and neural networks, including:

[0030] A data acquisition module for acquiring alarm data;

[0031] The data processing module is used to clean and repair the collected alarm data. The cleaning includes cleaning invalid alarm data and cleaning missing alarm data; the repair includes repairing alarm data.

[0032] The abstraction module is used to abstract the cleaned and repaired alarm data. The abstraction includes: extracting the same feature information from alarm data with the same characteristics, thereby unifying them into one type of data.

[0033] The dynamic label calculation module is used to perform dynamic label calculation using a dynamic label calculation model.

[0034] The annotation module is used to annotate alarm data using dynamic labels, including: annotating alarm data according to the dynamic label calculation value K, annotating the alarm data with the corresponding dynamic label T, and obtaining the annotated power plant alarm dataset; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases;

[0035] The dataset processing module is used to perform mathematical and normalization processing on the labeled power plant alarm dataset to obtain a training dataset. The mathematical processing includes converting text information into mathematically processed data that can be recognized by the neural network. The normalization processing includes normalizing the mathematically processed data to avoid excessively large distribution differences in the power plant alarm data, so that the mathematically processed data is evenly distributed in the 0-1 interval and conforms to a normal distribution.

[0036] The neural network model training and optimization module is used to train the neural network model based on the training dataset, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels.

[0037] A third aspect of the present invention provides an electronic device including a processor and a memory, the memory storing a plurality of instructions, the processor being configured to read the instructions and execute the method as described in the first aspect.

[0038] A fourth aspect of the present invention provides a computer-readable storage medium storing a plurality of instructions which can be read by a processor and executed as described in the first aspect.

[0039] The method, apparatus, electronic device, and computer-readable storage medium provided by this invention have the following beneficial technical effects:

[0040] A dynamic data labeling method is introduced, which solves the problem that traditional static labels cannot effectively label the massive alarm data of new energy power plants or control centers (i.e., traditional static labeling technology cannot effectively label the key alarm information with the highest priority that urgently needs to be handled by monitoring personnel from tens of thousands of alarm messages per day). It also solves the problem that artificial intelligence algorithms such as neural networks are difficult to apply effectively to alarm data processing of new energy power plants, and provides key technologies for smart monitoring and smart power plant construction. Attached Figure Description

[0041] Figure 1 This is an overall flowchart illustrating an alarm data processing method based on dynamic tags and neural networks according to a preferred embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of a dynamic tag calculation model for power plant alarm data, based on a preferred embodiment of the present invention, which illustrates an alarm data preprocessing method using dynamic tags and neural networks.

[0043] Figure 3 This is a flowchart illustrating the mathematical and normalization process of power plant alarm data according to a preferred embodiment of the present invention, which describes an alarm data processing method based on dynamic tags and neural networks.

[0044] Figure 4 This is a schematic diagram of a neural network model for an alarm data preprocessing method based on dynamic tags and neural networks, according to a preferred embodiment of the present invention.

[0045] Figure 5 This is a schematic diagram of the structure of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0046] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0047] Example 1

[0048] like Figure 1 As shown, an alarm data processing method based on dynamic tags and neural networks includes:

[0049] S1, Collect alarm data; In this embodiment, the alarm data comes from the power plant and includes the following fields: "Alarm Time", "Power Plant Code", "Power Plant Name", "Equipment Name", "Equipment KKS Code", "Alarm Content", and "Alarm Level".

[0050] S2. Clean and repair the collected alarm data. The cleaning includes cleaning invalid power station alarm data and cleaning missing power station alarm data. The repair includes repairing power station alarm data.

[0051] S3. Abstract the alarm data after cleaning and repair. The abstraction includes: extracting the same feature information for alarm data with the same features, so as to unify them into a certain type of data.

[0052] S4. Perform dynamic label calculation using a dynamic label calculation model.

[0053] Traditional labels use static labels. Static labels rely on manual judgment, making it difficult to effectively label the massive alarm information of new energy power stations, and unable to change accordingly based on the amount of alarm data and the occurrence frequency of redundant alarms. Based on this, the present invention proposes a dynamic label calculation model. The basic principle is: calculate dynamic labels for alarm data according to the probability of power station alarm data appearance, the level of power station alarm data, the content of power station alarm data, and other alarm field information.

[0054] See Figure 2 , as a preferred implementation, the performing dynamic label calculation using the dynamic label calculation model includes:

[0055] S41. Calculate the probability label calculation value P, including: classify and count alarms according to whether the power station name or power station code, equipment name or equipment KKS code, and alarm content are the same; then calculate the occurrence probability p(x i ), where the occurrence probability p(x i ) represents the proportion of the i-th type of alarm data volume in the total alarm data volume, 0 ≤ p(x i ) ≤ 1; according to the size of the occurrence probability p(x i ), assign the corresponding probability label calculation value P, 0 < P ≤ 16.

[0056] S42. Calculate the static label calculation value S, including: assign static label calculation values to each field content of the alarm data respectively, accumulate the static label calculation values of each field, and then perform a 16-point conversion to obtain the static label calculation value S, 0 < S ≤ 16. The calculation formula is:

[0057]

[0058] S43. Calculate the dynamic label calculation value K, including: perform weighted calculation based on the probability label calculation value P and the static label value S of the alarm data to obtain the dynamic label calculation value K. The calculation formula is K = PR + S(2 - R), 0 < K ≤ 32, where R is the weight coefficient, and the value of the weight coefficient R is adjusted and determined according to the size of the average daily alarm quantity n of the power station.

[0059] S5, labeling alarm data using dynamic labels, including: labeling alarm data according to the dynamic label calculation value K, labeling alarm data with corresponding dynamic labels T, and obtaining the labeled power plant alarm dataset; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases;

[0060] See Figure 3 S6, the labeled power plant alarm dataset is mathematically and normally processed to obtain a training dataset. The mathematical processing includes converting text information into mathematically processed data that can be recognized by the neural network. The normalization process includes normalizing the mathematically processed data to avoid excessively large distribution differences in the power plant alarm data, so that the mathematically processed data is evenly distributed in the 0-1 interval and conforms to a normal distribution.

[0061] S7. Train the neural network model based on the training dataset, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels.

[0062] See Figure 4 In a preferred embodiment, the neural network model is a neural network model containing three hidden layers, using sigmoid as the activation function. Before using the power plant alarm data classification model, this classification model needs to be trained first. The training process includes the following steps:

[0063] S71, Import the alarm data generated into the neural network model;

[0064] S72 divides the collected alarm data into multiple data blocks on an even basis; for example, 20,000 alarms are divided into 200 blocks of 100 alarms each.

[0065] S73, Set the learning rate, and calculate the current loss function after training block by block according to the data block; In this embodiment, the learning rate is set to 0.001, one data block is input for training each time, and the current loss function is calculated.

[0066] S74 adjusts the learning rate using a discrete descent method;

[0067] S75, repeat S73 and S74 to optimize the model until the loss function converges;

[0068] S76, save the parameters of the power plant alarm data classification model based on the neural network model. After training, the intelligent alarm data recognition and classification algorithm based on dynamic tags is obtained.

[0069] The alarm data processing flow and specific implementation steps in a power plant scenario are as follows:

[0070] Step 1: Collect power plant alarm data. The power plant alarm data includes the following fields: "Alarm Time", "Power Plant Code", "Power Plant Name", "Equipment KKS Code", "Alarm Content", "Alarm Level" ... (n fields).

[0071] Step 2: Clean the collected power plant alarm data. The cleaned data includes invalid power plant alarm data, missing power plant alarm data, and power plant alarm data repair.

[0072] Step 3: Abstract the cleaned power plant alarm data, that is, extract the same characteristic information from power plant alarm data with the same characteristics, and unify them into a certain type of data.

[0073] Step 4: Perform dynamic tag calculation.

[0074] Traditional labeling uses static labels, which rely on manual judgment and are difficult to effectively label the massive amounts of alarm information from renewable energy power plants. Furthermore, static labels cannot be adjusted based on the size of the alarm data or the frequency of redundant alarms. Therefore, we propose a dynamic labeling calculation model. We calculate dynamic labels for alarm data based on the probability of alarm occurrence, alarm level, alarm content, and other alarm field information.

[0075] The specific method for calculating dynamic tags is as follows:

[0076] 1) Based on whether the "power station name (or power station code), equipment name (or equipment KKS code), and alarm content" are the same, classify and statistically analyze the alarms, and then calculate the probability p(x) of each type of alarm. i ), p(x i ) represents the proportion of the i-th type of alarm data to the total alarm data, 0≤p(x) i )≤1, according to p(x) i The probability label calculation value P is generated, and the assignment method is as follows:

[0077]

[0078]

[0079] 2) Based on the content of the "Alarm Level" field, the alarm level calculation value G = {g1, g2, ... g} is obtained by assigning values. i}, g i The label value representing the i-th alarm level is calculated and assigned using the following method:

[0080] Alarm Level <![CDATA[Alarm level field label calculated value g i > slight 1 generally 3 serious 6 … …

[0081] 3) Based on the content of the "Alarm Content" field, obtain the calculated alarm content value Q = {q1, q2, ..., q} by assigning values. i}, where q i The tag value for the alarm content of the i-th type of power station is calculated, and the value is assigned as follows:

[0082]

[0083]

[0084] 4) As needed, assign values ​​to other fields using the above method to obtain the corresponding calculated values ​​for each field (G, Q, ..., F). n );

[0085] 5) Sum the calculated values ​​of each field, then convert them to a 16-point scale to obtain the static label calculated value.

[0086] 6) By weighting the probability label value and the static label value, the dynamic label calculation value K = PR + S(2-R) is obtained, where the weighting coefficient R is determined according to the magnitude of the average daily alarm quantity n of the power station. The method for determining the value of R is as follows:

[0087]

[0088] Step 5: Based on the calculated value K of the dynamic label, label the alarm data with the corresponding dynamic label T to obtain the labeled power plant alarm dataset. The dynamic labels correspond to alarm level priorities, which increase from l1, with l5 having the highest priority, as shown in the table below:

[0089]

[0090]

[0091] Step 6: Perform mathematical and normalization processing on the dataset obtained in Step 5, that is: convert the text information into data that can be recognized by the neural network and perform normalization processing; in order to avoid excessively large distribution differences in the power plant alarm data, distribute it evenly in the 0-1 interval so that it conforms to a normal distribution.

[0092] Step 7: Train the neural network model algorithm using the dataset obtained in Step 6 to obtain an intelligent recognition algorithm based on dynamic labels. Here, we establish a neural network model with three hidden layers, using sigmoid as the activation function. Before using the power plant alarm data classification model, this classification model needs to be trained first. The training process includes the following steps:

[0093] (1) Power plant alarm data generated by importing neural network models;

[0094] (2) The collected power station alarm data, for example, 20,000 records, is divided into 200 blocks of 100 records each;

[0095] (3) Set the learning rate to 0.001, input one data block at a time for training, and calculate the current loss;

[0096] (4) Adjust the learning rate using the discrete descent method;

[0097] (5) Repeat (3) and (4), and optimize the model until the loss function converges;

[0098] (6) Save the parameters of the power plant alarm data classification model based on neural network. After training, a power plant alarm data identification and classification algorithm based on dynamic label is obtained.

[0099] Example 2

[0100] An alarm data processing system based on dynamic tags and neural networks includes:

[0101] The data acquisition module is used to collect alarm data. In this embodiment, the alarm data comes from the power plant and includes the following fields: "alarm time", "power plant code", "power plant name", "equipment name", "equipment KKS code", "alarm content" and "alarm level".

[0102] The data processing module is used to clean and repair the collected alarm data. The cleaning includes cleaning invalid power plant alarm data and cleaning missing power plant alarm data; the repair includes repairing the power plant alarm data.

[0103] The abstraction module is used to abstract the cleaned and repaired alarm data. The abstraction includes: extracting the same feature information from alarm data with the same characteristics, thereby unifying them into a certain type of data.

[0104] The dynamic label calculation module is used to perform dynamic label calculation using a dynamic label calculation model.

[0105] Traditional labels use static labels, which rely on manual judgment and are difficult to effectively label the massive amounts of alarm information from new energy power plants. Furthermore, they cannot be adjusted according to the size of the alarm data or the frequency of redundant alarms. Therefore, this invention proposes a dynamic label calculation model. The basic principle is to calculate dynamic labels for alarm data based on the probability of alarm occurrence, alarm level, alarm content, and other alarm field information.

[0106] The annotation module is used to annotate alarm data using dynamic labels, including: annotating alarm data according to the dynamic label calculation value K, annotating the alarm data with the corresponding dynamic label T, and obtaining the annotated power plant alarm dataset; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases;

[0107] The dataset processing module is used to perform mathematical and normalization processing on the labeled power plant alarm dataset to obtain a training dataset. The mathematical processing includes converting text information into mathematically processed data that can be recognized by the neural network. The normalization processing includes normalizing the mathematically processed data to avoid excessively large distribution differences in the power plant alarm data, so that the mathematically processed data is evenly distributed in the 0-1 interval and conforms to a normal distribution.

[0108] The neural network model training and optimization module is used to train the neural network model based on the training dataset, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels.

[0109] The present invention also provides a memory that stores a plurality of instructions for implementing the method as described in Embodiment 1.

[0110] like Figure 5 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301. The memory 302 stores a plurality of instructions, which can be loaded and executed by the processor to enable the processor to perform the method described in Embodiment 2.

[0111] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for processing alarm data based on dynamic tags and neural networks, characterized in that, include: S1, collect alarm data; S2, clean and repair the collected alarm data, the cleaning including cleaning invalid alarm data and cleaning missing alarm data; The repair includes repairing alarm data; S3, Abstract the cleaned and repaired alarm data. The abstraction includes: extracting the same feature information from alarm data with the same characteristics, thereby unifying them into one type of data. S4, using a dynamic label calculation model to perform dynamic label calculation; S5 uses dynamic labels to annotate alarm data; S6. Mathematicalize and normalize the labeled power plant alarm dataset to obtain the training dataset. S7. Train the neural network model based on the training dataset, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels. The alarm data in S1 comes from the power plant and includes the following fields: "alarm time", "power plant code", "power plant name", "equipment name", "equipment KKS code", "alarm content" and "alarm level"; The S4 method for dynamic tag calculation using a dynamic tag calculation model includes: S41, Calculate the probability label value P, including: classifying alarms statistically based on whether the power station name or power station code, equipment name or equipment KKS code, and alarm content are the same; then calculate the probability of occurrence of each type of alarm data. The probability of occurrence This represents the proportion of alarm data of type i to the total alarm data. Based on the probability of occurrence The magnitude of is assigned a corresponding probability label, and a calculated value P is obtained. ; S42, Calculate the static label calculation value S, including: assigning a static label calculation value to each field of the alarm data, summing the static label calculation values ​​of each field, and then performing a 16-point conversion to obtain the static label calculation value S. The calculation formula is: ； S43, Calculate the dynamic label calculation value K, including: weighting the probability label calculation value P and the static label value S of the alarm data to obtain the dynamic label calculation value K, the calculation formula is as follows: , R is the weighting coefficient, and the value of the weighting coefficient R is determined by adjusting the value of the average daily number of alarms n of the power station.

2. The alarm data processing method based on dynamic tags and neural networks according to claim 1, characterized in that, S5 includes: labeling alarm data according to the dynamic label calculation value K, labeling the alarm data with the corresponding dynamic label T, and obtaining the labeled power plant alarm dataset; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases.

3. The alarm data processing method based on dynamic tags and neural networks according to claim 2, characterized in that, The mathematization in S6 includes: converting text information into mathematized data that can be recognized by a neural network; the normalization process includes: normalizing the mathematized data so that the mathematized data is evenly distributed in the 0-1 interval and conforms to a normal distribution.

4. The alarm data processing method based on dynamic tags and neural networks according to claim 1, characterized in that, The neural network model is a neural network model containing three hidden layers, using sigmoid as the activation function.

5. The alarm data processing method based on dynamic tags and neural networks according to claim 1, characterized in that, The training of the neural network model based on the training dataset includes the following steps: S71, Import the alarm data generated into the neural network model; S72 divides the collected alarm data into multiple data blocks on an average basis; S73, Set the learning rate, and calculate the current loss function after training block by block according to the data blocks; S74 adjusts the learning rate using a discrete descent method; S75, repeat S73 and S74 to optimize the model until the loss function converges; S76, save the parameters of the power plant alarm data classification model based on the neural network model. After training, the intelligent alarm data recognition and classification algorithm based on dynamic tags is obtained.

6. An alarm data processing system based on dynamic tags and neural networks, used to implement the data processing method according to any one of claims 1-5, characterized in that, include: The data acquisition module is used to collect alarm data; The data processing module is used to clean and repair the collected alarm data, including cleaning invalid alarm data and cleaning missing alarm data. The repair includes repairing alarm data; The abstraction module is used to abstract the cleaned and repaired alarm data. The abstraction includes: extracting the same feature information from alarm data with the same characteristics, thereby unifying them into one type of data. The dynamic label calculation module is used to perform dynamic label calculation using a dynamic label calculation model. The annotation module is used to annotate alarm data using dynamic labels, including: annotating alarm data according to the dynamic label calculation value K, annotating the alarm data with the corresponding dynamic label T, and obtaining the annotated power plant alarm dataset; the dynamic label T corresponds to the alarm level, and as the dynamic label calculation value K increases, the alarm level corresponding to the dynamic label T increases; The dataset processing module is used to perform mathematical and normalization processing on the labeled power plant alarm dataset to obtain a training dataset. The mathematical processing includes converting text information into mathematically processed data that can be recognized by the neural network. The normalization processing includes normalizing the mathematically processed data to avoid excessively large distribution differences in the power plant alarm data, so that the mathematically processed data is evenly distributed in the 0-1 interval and conforms to a normal distribution. The neural network model training and optimization module is used to train the neural network model based on the training dataset, optimize the neural network model, save the model parameters, and output an intelligent alarm data recognition and classification algorithm based on dynamic labels.

7. An electronic device, characterized in that, It includes a processor and a memory, the memory storing multiple instructions, and the processor being used to read the instructions and execute the method as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions, which can be read by a processor and executed as described in any one of claims 1-5.

Images