Power grid monitoring information intelligent alarm method and system based on pre-training language model, computer equipment and storage medium

Abstract

The invention discloses a power grid monitoring information intelligent alarm method and system based on a pre-training language model, and the method comprises the steps: carrying out the preprocessing of a power grid monitoring information text, and constructing a standardized power grid monitoring information text corpus; performing incremental training on the language model by using the standardized power grid monitoring information text corpus to obtain an incremental training model; performing fine-tuning training on the incremental training model to obtain a fine-tuning training model, and extracting text data features through the fine-tuning training model; based on the text data features, a bidirectional gating loop unit deep learning model is used for analysis processing, and text classification is obtained; and obtaining an alarm result of the power grid monitoring information based on text classification. According to the method, the text corpus can be constructed in a standardized mode from mass monitoring information, the alarm result can be efficiently obtained, model fine adjustment is carried out by comparing the learning loss function, the problems of monitoring pressure and low information processing efficiency are effectively solved, missing monitoring and mistaken monitoring are prevented, and the intelligent level of equipment monitoring is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and in particular to a power grid monitoring information intelligent alarm method and system based on a pre-trained language model. Background Art

[0002] The substation control station is responsible for monitoring the main and auxiliary equipment of the substation within its jurisdiction. With the continuous growth of the scale of the power grid, the types and quantities of primary and secondary equipment connected to the substation control station continue to increase, and the amount of monitoring information has also increased exponentially. The massive abnormal operation signals have led to a sudden increase in monitoring pressure. The information query, analysis, decision-making and implementation of the monitoring information processing process completely rely on the experience and knowledge of the operating personnel, and there are many types of abnormal alarm information. The processing process is easily affected by invalid or low-value fault information. How to avoid missed monitoring and mismonitoring has become one of the urgent problems to be solved in the monitoring business. Monitoring information processing is an important part of daily monitoring. The improvement of its intelligent level is of great significance to realize the integrated and comprehensive monitoring of main and auxiliary equipment, solve the problems of insufficient equipment monitoring intensity, insufficient equipment management granularity, insufficient production informatization, insufficient intelligent support, and improve the comprehensive perception of substation equipment status and rapid response to abnormal grid alarms. Summary of the invention

[0003] In view of the problems existing in the prior art, the purpose of the present invention is to provide a method and system for intelligent alarm of power grid monitoring information based on a pre-trained language model, analyze and infer the monitoring information, standardize and construct a text corpus from massive monitoring information and efficiently obtain alarm results, fine-tune the model through a specific comparative learning loss function, effectively solve the problems of monitoring pressure and low information processing efficiency, prevent missed monitoring and false monitoring, and improve the level of intelligent equipment monitoring.

[0004] In order to achieve the above technical objectives, the present invention adopts the following technical solutions to achieve them.

[0005] In a first aspect, the present invention provides a method for intelligent warning of power grid monitoring information based on a pre-trained language model, comprising: Step S1, preprocessing the power grid monitoring information text, and constructing a standardized power grid monitoring information text corpus; the power grid monitoring information text includes a monitoring information point table and historical alarm information; the construction of the standardized power grid monitoring information text corpus includes corpus collection, data processing, entity labeling and manual verification, the corpus collection is to collect the monitoring information point table of the substation under the jurisdiction of the centralized control station and the historical alarm information of the centralized control system; the data processing is to remove the serial number, space, symbol, date, stop word in the monitoring information point table of the substation under the jurisdiction and the historical alarm information of the centralized control system, and divide the processed data into sentence levels; the entity labeling is to label the sentence as the centralized control station name, substation name, voltage level, dual name of equipment, and alarm category label; the manual verification is to manually verify and correct the entity labeling result; Step S2, using the standardized power grid monitoring information text corpus to perform incremental training on a pre-established language model to enhance the knowledge scope of the language model and obtain an incremental training model; Step S3, fine-tuning the incremental training model on the power grid monitoring information intelligent identification and alarm task to obtain a fine-tuning training model, and extracting the text data features of the power grid monitoring information intelligent identification and alarm task through the fine-tuning training model; the strategy of the fine-tuning training is based on the incremental training model, by constructing positive samples and negative samples of the power grid monitoring information intelligent identification and alarm task, based on contrastive learning, using the differences between data for learning, so that the representations of different data types are distinguished, and the representations of the same type of data are brought closer, so as to extract the text data features of the power grid monitoring information intelligent identification and alarm task; in the process of the fine-tuning training, the contrastive learning loss function is used to continuously narrow the distance between the positive sample pairs, and at the same time, the negative sample pairs are kept away from each other, so as to maximize the similarity between all positive sample pairs and minimize the similarity between all negative sample pairs, and the contrastive learning loss function is calculated as follows: ; Where: is a positive sample, is a negative sample, is the cosine similarity, is a hyperparameter, J is the positive sample size, L is the negative sample size; Step S4: Based on the text data features, a bidirectional gated recurrent unit deep learning model is used to perform analysis and processing to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks; and Step S5: Based on the text classification, obtain the alarm result of the power grid monitoring information.

[0006] Furthermore, in step S2, the pre-established language model consists of a static word embedding layer, a first bidirectional LSTM layer, a second bidirectional LSTM layer and a softmax output layer; the static word embedding layer converts the encoding and position information of the input corpus into a vector form as the input of the first bidirectional LSTM layer; the first bidirectional LSTM layer is used to extract syntactic information; the second bidirectional LSTM layer is used to extract semantic information; the softmax output layer is the encoding matrix of the standardized power grid monitoring information text corpus.

[0007] Furthermore, in step S2, the language model is based on LSTM and trained using bidirectional LSTM, including a forward LSTM model and a backward LSTM model; The calculation formula of the forward LSTM model is: ; Where: Represents a sequence of N words, the forward LSTM model models the k-1 words in the preceding context of the k-th word and predicts the probability of the k-th word; The calculation formula of the backward LSTM model is: ; The backward LSTM model models the k-1 words in the context after the k-th word and predicts the probability of the k-th word; The goal of the incremental training is to maximize the log-likelihood function of the forward LSTM model and the backward LSTM model, and the calculation formula is: ; Where: is the input sequence, is the initial word vector parameter of the input, are the parameters of the softmaxs output layer, and are the parameters of the forward and backward LSTM networks; The language model enhances the knowledge scope of the language model through the above-mentioned incremental training to obtain an incremental training model.

[0008] Furthermore, in step S4, the bidirectional gated recurrent unit deep learning model includes a bidirectional gated recurrent neural network layer and a classification layer.

[0009] Furthermore, in step S4, the calculation formula of the bidirectional gated recurrent neural network layer is as follows: ; ; ; Where: is the sigmoid activation function, , , , , , are model parameters; is the output vector of step i-1, is the output vector of step i, is the input vector; To update the gate vector; is the reset gate vector; The forward and reverse encodings through the bidirectional gated recurrent neural network layer are: ; ; and are the parameters of the forward and backward GRU networks. After bidirectional encoding, the encoding of the i-th word is , concatenate the codes of the two directions of the bidirectional gated recurrent neural network layer to obtain a code vector sequence ; The classification layer uses the softmax function to calculate the probability that any input belongs to each type of alarm. The calculation formula is: ; Where: is the classification layer weight matrix, is the bias parameter of the classification layer.

[0010] Furthermore, in step S5, the category with the highest probability is selected as the precise intelligent alarm result of the power grid monitoring information, and the alarm result includes accident, abnormality, over-limit, displacement and notification.

[0011] In a second aspect, a power grid monitoring information intelligent alarm system based on a pre-trained language model is provided, and a power grid monitoring information intelligent alarm method based on the pre-trained language model comprises: The corpus construction module removes blanks, removes symbols and performs standardization processing on the power grid monitoring information text to obtain a standardized power grid monitoring information text corpus; The incremental training module is used to enhance the knowledge scope of the language model, which consists of a static word embedding layer, a first bidirectional LSTM layer, a second bidirectional LSTM layer, and a softmax output layer; A fine-tuning training module is used to extract text data features of power grid monitoring information intelligent identification and alarm tasks. The fine-tuning training strategy is based on an incremental training model. By constructing positive and negative samples of power grid monitoring information intelligent identification and alarm tasks, based on comparative learning, the differences between data are used for learning, so that representations of different data types are distinguished, and representations of the same type of data are brought closer, so as to extract text data features of the power grid monitoring information intelligent identification and alarm tasks; A text classification module is used to analyze and process the text data features using a bidirectional gated recurrent unit deep learning model to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks; the bidirectional gated recurrent unit deep learning model includes a bidirectional gated recurrent neural network layer and a classification layer architecture; The result output module selects the category with the highest output probability of the classification layer of the text classification module as the alarm result of the power grid monitoring information.

[0012] In a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an intelligent alarm method for power grid monitoring information based on a pre-trained language model is implemented.

[0013] In a fourth aspect, a computer-readable storage medium is provided, storing computer-executable instructions, wherein the computer-executable instructions are used to implement the intelligent alarm method for power grid monitoring information based on a pre-trained language model.

[0014] In view of the problems of monitoring pressure and low information processing efficiency in existing substation monitoring, the present invention proposes a power grid monitoring information intelligent alarm method and system based on a pre-trained language model, analyzes and infers the monitoring information, builds a standardized corpus, realizes intelligent and efficient identification and alarm of monitoring information, and fine-tunes the model through a specific comparison learning loss function and a contrastive learning method, thereby solving the problems of insufficient equipment monitoring intensity, insufficient equipment management detail, insufficient production informatization, insufficient intelligent support, etc., prevents missed monitoring and false monitoring, and improves the intelligent level of equipment monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a flow chart of a method for intelligent warning of power grid monitoring information based on a pre-trained language model of the present invention; Figure 2 It is a structural diagram of an Emlo pre-trained language model in a power grid monitoring information intelligent alarm method based on a pre-trained language model of the present invention; Figure 3A structural diagram of a deep learning classification model in a power grid monitoring information intelligent alarm method based on a pre-trained language model of the present invention; Figure 4 This is a structural diagram of a power grid monitoring information intelligent alarm system based on a pre-trained language model of the present invention. DETAILED DESCRIPTION

[0016] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods of the technical scheme proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments of the specification. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them.

[0017] Example 1 In one embodiment of the present invention, as shown in the attached Figure 1 As shown, a method for intelligent alarm of power grid monitoring information based on a pre-trained language model is provided, comprising the following steps: S1, preprocess the power grid monitoring information text and construct a standardized power grid monitoring information text corpus. Specifically, it can be to collect and process the text corpus required by the power grid monitoring information intelligent alarm method based on the pre-trained language model. The power grid monitoring information text includes a monitoring information point table and historical alarm information; the construction of the standardized power grid monitoring information text corpus is divided into four steps: corpus collection, data processing, entity labeling, and manual verification, so as to obtain the standardized power grid monitoring information text corpus.

[0018] (1) Corpus collection: Select the monitoring information point table of the substation under the jurisdiction of a certain centralized control station, as well as the historical alarm information of the centralized control system. Take the historical alarm information of the centralized control system as an example. The historical alarm information of the centralized control system is as follows: 4708 2024-12-10 12:02:11 XX.AA changes to BBⅡ line 404 protection timing abnormal action; 4710 On December 10, 2024 at 12:02:20, the protection timing of 404 of XX.AA transformer BBⅡ line was abnormally restored.

[0019] (2) Data processing: remove the serial numbers, spaces, symbols, dates, and stop words (action, return) in the monitoring information point table and historical alarm information; divide the processed data into sentence levels. After data processing, the historical alarm information of the centralized control system is as follows: XXAA to BBⅡ line 404 protection timing abnormality; The 404 protection timing of XXAA to BBⅡ line is abnormal.

[0020] (3) Entity labeling: Label the sentence with the name of the centralized control station, the name of the substation, the voltage level, the dual name of the equipment, and the alarm category label (accident, abnormality, over-limit, displacement, and notification). The entity labeling results are as follows: XX (name of centralized control station) AA substation (name of substation) BBⅡ line 404 (double name of equipment) protection timing abnormality (abnormality); XX (name of centralized control station) AA transformer (name of transformer substation) BBⅡ line 404 (double name of equipment) protection timing abnormality (abnormality).

[0021] (4) Manual verification: Manually verify and correct the entity annotation results and remove duplicate text. The corrected results are as follows: XX (name of centralized control station) AA transformer (name of transformer substation) BBⅡ line 404 (double name of equipment) protection timing abnormality (abnormality).

[0022] S2, using the standardized power grid monitoring information text corpus to perform incremental training on a pre-established language model to enhance the knowledge scope of the language model and obtain an incremental training model.

[0023] In this embodiment, as shown in the attached Figure 2 As shown in the figure, E1, E2, ..., EN are the input of the ELMO pre-trained language model, that is, the standardized text corpus; T1, T2, ..., TN are the output word vectors; the pre-established language model (embeddings from language models, Elmo) described in the second step is composed of a static word embedding layer, a first bidirectional LSTM (Long Short-Term Memory, LSTM) layer, a second bidirectional LSTM layer and a softmax output layer; the static word embedding layer converts the encoding and position information of the input corpus into a vector form as the input of the first bidirectional LSTM layer. The first bidirectional LSTM mainly extracts syntactic information, and the second bidirectional LSTM mainly extracts semantic information. The softmax output layer is the encoding matrix of the standardized power grid monitoring information text corpus.

[0024] The Elmo language model is based on the LSTM structure and uses bidirectional LSTM for training, including a forward LSTM model and a backward LSTM model.

[0025] The calculation formula of the forward LSTM model is: ; Where: Represents a sequence of N words. The forward LSTM model models the k-1 words in the context before the k-th word and predicts the probability of the k-th word.

[0026] The calculation formula of the backward LSTM model is: ; The backward LSTM model models the k-1 words in the context after the k-th word and predicts the probability of the k-th word; The goal of model training is to maximize the log-likelihood function of the LSTM language model in these two directions. The calculation formula is: ; Where: is the input sequence, is the initial word vector parameter of the input, are the parameters of the softmaxs output layer, and are the parameters of the forward and backward LSTM networks. The Emlo language model is incrementally trained to obtain an incremental training model, thereby enhancing the knowledge scope of the language model.

[0027] S3, fine-tuning the incremental training model on the power grid monitoring information intelligent identification and alarm task to obtain a fine-tuning training model, and extracting text data features of the power grid monitoring information intelligent identification and alarm task through the fine-tuning training model.

[0028] The fine-tuning training strategy is based on an incremental training model. By constructing positive and negative samples for specific tasks, based on contrastive learning, the differences between data are used for learning, so that the representations of different data types are distinguished, while the representations of the same type of data are brought closer, and the text data features of specific tasks are extracted. The specific task requirements of the present invention refer to intelligent identification and alarm of power grid monitoring information.

[0029] The fine-tuning training process uses the contrastive learning loss function to continuously narrow the distance between positive sample pairs and keep the negative sample pairs away from each other, that is, to maximize the similarity between all positive sample pairs and minimize the similarity between all negative sample pairs, so as to achieve the purpose of separating different alarm categories. The contrastive learning loss function calculation formula is: ; Where: is a positive sample, is a negative sample, is the cosine similarity, is a hyperparameter, J is the positive sample size, L is the negative sample size.

[0030] S4, based on the text data features, a bidirectional gated recurrent unit deep learning model is used to perform analysis and processing to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks.

[0031] As attached Figure 3 As shown in the figure, the deep learning classification model consists of a bidirectional gated recurrent neural network layer (Bi-GRU) and a classification layer. The deep learning model analyzes and processes text classification for specific tasks.

[0032] In this embodiment, the calculation formula of the deep learning classification model Bi-GRU in the fourth step is as follows: ; ; ; Where: is the sigmoid activation function, , , , , , are model parameters; is the output vector of step i-1, is the output vector of step i, is the input vector; To update the gate vector; is the reset gate vector. Through Bi-GRU forward and reverse encoding: ; ; and are the parameters of the forward and backward GRU networks. After bidirectional encoding, the encoding of the i-th word is , concatenate the codes of the two directions of Bi-GRU to get the code vector sequence .

[0033] The classification layer uses the softmax function to calculate the probability that any input belongs to each type of alarm. The calculation formula is: ; Where: is the classification layer weight matrix, is the bias parameter of the classification layer.

[0034] S5, obtaining an alarm result of the power grid monitoring information based on the text classification.

[0035] The category with the highest probability is selected as the accurate intelligent alarm result of the power grid monitoring information. The alarm categories include accident, abnormality, over-limit, displacement, and notification.

[0036] Example 2 In one embodiment of the present invention, as shown in the attached Figure 4 As shown, a power grid monitoring information intelligent alarm system 1 based on a pre-trained language model is provided, and a power grid monitoring information intelligent alarm method based on the pre-trained language model comprises: The corpus construction module 11 removes blanks, removes symbols and performs standardization processing on the power grid monitoring information text to obtain a standardized power grid monitoring information text corpus; An incremental training module 12, for enhancing the knowledge scope of the language model, is composed of a static word embedding layer, a first bidirectional LSTM layer, a second bidirectional LSTM layer, and a softmax output layer; A fine-tuning training module 13 is used to extract text data features of the intelligent identification and alarm tasks of power grid monitoring information. The fine-tuning training strategy is based on an incremental training model. By constructing positive and negative samples of the intelligent identification and alarm tasks of power grid monitoring information, based on comparative learning, the differences between data are used for learning, so that the representations of different data types are distinguished, and the representations of the same type of data are brought closer, so as to extract the text data features of the intelligent identification and alarm tasks of the power grid monitoring information; A text classification module 14 is used to analyze and process the text data features using a bidirectional gated recurrent unit deep learning model to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks; the bidirectional gated recurrent unit deep learning model includes a bidirectional gated recurrent neural network layer and a classification layer architecture; The result output module 15 selects the category with the highest probability of being output by the classification layer of the text classification module 14 as the alarm result of the power grid monitoring information.

[0037] In an embodiment of the present invention, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the program, any one of the above-described methods for intelligent warning of power grid monitoring information based on a pre-trained language model is implemented. The memory may be various types of memory, such as a random access memory, a read-only memory, a flash memory, etc. The processor may be various types of processors, such as a central processing unit, a microprocessor, a digital signal processor, or an image processor, etc.

[0038] In an embodiment of the present invention, a computer-readable storage medium is also provided, which stores computer-executable instructions, and the computer-executable instructions are used to implement a method for intelligent warning of power grid monitoring information based on a pre-trained language model. The storage medium includes: a U disk, a mobile hard disk, a ROM, a RAM, a magnetic disk or an optical disk, etc., which can store program codes.

[0039] The present invention provides a method and system for intelligent alarm of power grid monitoring information based on a pre-trained language model, which analyzes and infers the monitoring information, realizes intelligent identification and alarm of the monitoring information, effectively solves the problems of monitoring pressure and low information processing efficiency in substation monitoring, prevents missed monitoring and false monitoring, and improves the intelligent level of equipment monitoring.

[0040] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for intelligent warning of power grid monitoring information based on a pre-trained language model, characterized in that: include: Step S1, preprocessing the power grid monitoring information text to construct a standardized power grid monitoring information text corpus; the power grid monitoring information text includes a monitoring information point table and historical alarm information; The construction of a standardized power grid monitoring information text corpus includes corpus collection, data processing, entity labeling and manual verification; the corpus collection is to collect the monitoring information point table of the substation under the jurisdiction of the centralized control station and the historical alarm information of the centralized control system; the data processing is to remove the serial number, space, symbol, date, stop words in the monitoring information point table of the substation under the jurisdiction and the historical alarm information of the centralized control system, and divide the processed data into sentence levels; the entity labeling is to label the sentence as the name of the centralized control station, the name of the substation, the voltage level, the dual name of the equipment, and the alarm category label; the manual verification is to manually verify and correct the entity labeling results; Step S2, using the standardized power grid monitoring information text corpus to perform incremental training on a pre-established language model to enhance the knowledge scope of the language model and obtain an incremental training model; Step S3, fine-tuning the incremental training model on the power grid monitoring information intelligent identification and alarm task to obtain a fine-tuning training model, and extracting the text data features of the power grid monitoring information intelligent identification and alarm task through the fine-tuning training model; the strategy of the fine-tuning training is based on the incremental training model, by constructing positive samples and negative samples of the power grid monitoring information intelligent identification and alarm task, based on comparative learning, using the differences between data for learning, so that the representations of different data types are distinguished, and at the same time, the representations of the same type of data are brought closer, so as to extract the text data features of the power grid monitoring information intelligent identification and alarm task; During the fine-tuning training process, the contrastive learning loss function is used to continuously narrow the distance between the positive sample pairs, while moving the negative sample pairs away from each other, so as to maximize the similarity between all positive sample pairs and minimize the similarity between all negative sample pairs. The contrastive learning loss function calculation formula is: ； Where: is a positive sample, is a negative sample, is the cosine similarity, is a hyperparameter, J is the positive sample size, L is the negative sample size; Step S4: Based on the text data features, a bidirectional gated recurrent unit deep learning model is used to perform analysis and processing to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks; as well as Step S5: Based on the text classification, obtain the alarm result of the power grid monitoring information.

2. According to claim 1, a method for intelligent warning of power grid monitoring information based on a pre-trained language model is characterized in that: In step S2, the pre-established language model consists of a static word embedding layer, a first bidirectional LSTM layer, a second bidirectional LSTM layer and a softmax output layer; the static word embedding layer converts the encoding and position information of the input corpus into a vector form as the input of the first bidirectional LSTM layer; the first bidirectional LSTM layer is used to extract syntactic information; the second bidirectional LSTM layer is used to extract semantic information; the softmax output layer is the encoding matrix of the standardized power grid monitoring information text corpus.

3. The method for intelligent warning of power grid monitoring information based on a pre-trained language model according to claim 2 is characterized in that: In step S2, the language model is based on LSTM and trained using bidirectional LSTM, including a forward LSTM model and a backward LSTM model; The calculation formula of the forward LSTM model is: ； Where: Represents a sequence of N words, the forward LSTM model models the k-1 words in the preceding context of the k-th word and predicts the probability of the k-th word; The calculation formula of the backward LSTM model is: ； The backward LSTM model models the k-1 words in the context after the k-th word and predicts the probability of the k-th word; The goal of the incremental training is to maximize the log-likelihood function of the forward LSTM model and the backward LSTM model, and the calculation formula is: ； Where: is the input sequence, is the initial word vector parameter of the input, are the parameters of the softmaxs output layer, and are the parameters of the forward and backward LSTM networks; The language model enhances the knowledge scope of the language model through the above-mentioned incremental training to obtain an incremental training model.

4. The method for intelligent warning of power grid monitoring information based on a pre-trained language model according to claim 1, characterized in that: In step S4, the bidirectional gated recurrent unit deep learning model includes a bidirectional gated recurrent neural network layer and a classification layer.

5. The method for intelligent warning of power grid monitoring information based on a pre-trained language model according to claim 4 is characterized in that: In step S4, the calculation formula of the bidirectional gated recurrent neural network layer is as follows: ； ； ； Where: is the sigmoid activation function, , , , , , are model parameters; is the output vector of step i-1, is the output vector of step i, is the input vector; To update the gate vector; is the reset gate vector; The forward and reverse encodings through the bidirectional gated recurrent neural network layer are: ； ； and are the parameters of the forward and backward GRU networks. After bidirectional encoding, the encoding of the i-th word is , concatenate the codes of the two directions of the bidirectional gated recurrent neural network layer to obtain a code vector sequence ; The classification layer uses the softmax function to calculate the probability that any input belongs to each type of alarm. The calculation formula is: ； Where: is the classification layer weight matrix, is the bias parameter of the classification layer.

6. The method for intelligent warning of power grid monitoring information based on a pre-trained language model according to claim 5 is characterized in that: In the step S5, the category with the highest probability is selected as the precise intelligent alarm result of the power grid monitoring information, and the alarm result includes accident, abnormality, over-limit, displacement and notification.

7. A power grid monitoring information intelligent alarm system based on a pre-trained language model, a power grid monitoring information intelligent alarm method based on a pre-trained language model according to any one of claims 1 to 6, characterized in that: include: The corpus construction module removes blanks, removes symbols and performs standardization processing on the power grid monitoring information text to obtain a standardized power grid monitoring information text corpus; The incremental training module is used to enhance the knowledge scope of the language model, which consists of a static word embedding layer, a first bidirectional LSTM layer, a second bidirectional LSTM layer, and a softmax output layer; A fine-tuning training module is used to extract text data features of power grid monitoring information intelligent identification and alarm tasks. The fine-tuning training strategy is based on an incremental training model. By constructing positive and negative samples of power grid monitoring information intelligent identification and alarm tasks, based on comparative learning, the differences between data are used for learning, so that representations of different data types are distinguished, and representations of the same type of data are brought closer, so as to extract text data features of the power grid monitoring information intelligent identification and alarm tasks; A text classification module is used to analyze and process the text data features using a bidirectional gated recurrent unit deep learning model to obtain text classification for the power grid monitoring information intelligent identification and alarm tasks; the bidirectional gated recurrent unit deep learning model includes a bidirectional gated recurrent neural network layer and a classification layer architecture; The result output module selects the category with the highest output probability of the classification layer of the text classification module as the alarm result of the power grid monitoring information.

8. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, an intelligent alarm method for power grid monitoring information based on a pre-trained language model as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to implement a power grid monitoring information intelligent alarm method based on a pre-trained language model as described in any one of claims 1-6.

Images