Method and system for studying and judging power transformation monitoring alarm information

Through natural language processing technology based on BERT language model, substation monitoring and alarm information is pre-trained and fine-tuned, combined with expert system event rules, independent research and judgment is achieved, and the analysis strategy in the existing technology depends on manual rules and low fault tolerance. The problem has improved the accuracy and fault tolerance of monitoring information processing.

CN118965177AInactive Publication Date: 2024-11-15STATE GRID HUNAN ELECTRIC POWER CO LTD ELECTRIC POWER SCI RES INST +3

Patent Information

Application Number
CN202411428165.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-14
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing substation monitoring information analysis strategy is highly dependent on manual rules and has low fault tolerance. It is prone to fail to analyze alarm events due to insufficient information naming specifications or missed original signals, and it is impossible to adapt to the current demand for digital and intelligent development of the power grid.

Method used

Using the BERT language model based on the bidirectional Transformer encoder and the fully connected layer pre-training model, the tagless historical monitoring alarm information is pre-trained and fine-tuned, combined with the expert system event rules for annotation and verification, and natural language processing technology is used. Realize independent research and judgment on substation monitoring and alarm events.

Benefits of technology

It improves the accuracy and fault tolerance of incident analysis, reduces the pressure on monitoring personnel to monitor the market, and enhances the ability to process massive monitoring alarm information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965177A_ABST
    Figure CN118965177A_ABST
Patent Text Reader

Abstract

The invention discloses a power transformation monitoring alarm information studying and judging method and system. The method comprises the following steps: acquiring historical monitoring alarm information without labels; inputting the historical monitoring alarm information without labels into the initial model for pre-training to obtain a pre-training model; labeling the historical monitoring alarm information without labels by utilizing an expert system event rule, and checking the labeled data according to the scheduling log to obtain the historical monitoring alarm information with labels; inputting the historical monitoring alarm information with the label into a pre-training model, performing fine tuning on the pre-training model, and updating model parameters of the pre-training model through back propagation to obtain a trained network model; and obtaining monitoring alarm data during the monitoring operation period of the power transformation centralized control station, inputting the monitoring alarm data into the trained network model, and obtaining a research and judgment result of the power transformation monitoring alarm event. According to the scheme of the invention, autonomous research and judgment of the power transformation monitoring alarm event can be realized, and the accuracy and the error-tolerant rate of an event research and judgment strategy are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of substation monitoring, and in particular to a substation monitoring alarm information analysis method and system. Background Art

[0002] The substation control station of the substation company is responsible for monitoring the main and auxiliary equipment of the substation within its jurisdiction. With the growth of the scale of the power grid and the continuous improvement of the level of intelligence 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. At present, the research on the processing methods of monitoring alarm information usually adopts the monitoring information event analysis strategy based on artificial rules. However, the current substation monitoring information analysis strategy is highly dependent on the perfection of the alarm event knowledge base, and the event reasoning is strictly based on established rules, and the analysis fault tolerance is low. When the quality of monitoring information is not high, such as a signal missing or data abnormality, it will directly lead to the failure of event analysis, which can no longer meet the needs of the current digital and intelligent development of the power grid. Summary of the invention

[0003] In order to solve the technical problems that the existing power grid event analysis strategy still has the following problems in actual monitoring business analysis: the alarm event analysis coverage is highly dependent on manual rules, the event analysis fault tolerance is low, and the alarm event analysis failure is easily caused by insufficient information naming specifications and the mis-sending or missing of original signals, the embodiments of the present invention provide a substation monitoring alarm information analysis method and system.

[0004] The technical solution of the embodiment of the present invention is achieved as follows: The embodiment of the present invention provides a method for analyzing substation monitoring alarm information, the method comprising: obtaining unlabeled historical monitoring alarm information; inputting the unlabeled historical monitoring alarm information into an initial model for pre-training to obtain a pre-trained model; the initial model is a BERT language model based on a bidirectional Transformer encoder and a fully connected layer; the initial model can learn the semantic information and contextual relationship of the input text; the fully connected layer is used to perform text classification tasks; using expert system event rules to annotate the unlabeled historical monitoring alarm information, and checking the annotated data against the scheduling log to obtain labeled historical monitoring alarm information; inputting the labeled historical monitoring alarm information into the pre-trained model, fine-tuning the pre-trained model, and updating the model parameters of the pre-trained model through back propagation to obtain a trained network model; obtaining monitoring alarm data during the monitoring operation of the substation control station, inputting the monitoring alarm data into the trained network model, and obtaining the substation monitoring alarm event analysis result.

[0005] In one embodiment, the unlabeled historical monitoring alarm information is input into an initial model for pre-training to obtain a pre-trained model, including: performing a pre-processing operation on the unlabeled historical monitoring alarm information to obtain the pre-processed historical monitoring alarm information; the pre-processing operation includes text cleaning, text standardization, word segmentation, and removal of stop words; the pre-processed historical monitoring alarm information is input into the initial model so that the initial model performs feature vector conversion on the pre-processed historical monitoring alarm information to obtain a word feature vector, and the word feature vector is used to perform a masked language model task and a next sentence prediction task, and the masked language model task is maximized. The probability of the word feature vector task and the next sentence prediction task is calculated to complete the pre-training; the word feature vector includes a semantic embedding vector, a segment embedding vector and a position embedding vector; the semantic embedding vector is a word embedding vector that represents each word fragment mapped in a fixed dimension; the position embedding vector is a vector that represents the position of a word in an input sequence; the segment embedding vector is a vector used to distinguish different fragments to complete the classification task; the masked language model task is to randomly select words in the input substation monitoring alarm signal and replace them with masks, and predict the correct words corresponding to the masks; the next sentence prediction task is to determine the logical relationship between two substation monitoring alarm signals.

[0006] In one embodiment, maximizing the probabilities of the masked language model task and the next sentence prediction task includes: iteratively updating the parameters of the initial model using an optimization algorithm until the probabilities of the masked language model task and the next sentence prediction task reach a maximum or until the number of iterations of the initial model reaches a preset number of iterations.

[0007] In one embodiment, the loss function when using the optimization algorithm for iteration is: Formula (1) Formula (2) Formula (3) in, are the parameters of the encoder part; are the parameters of the output layer of the MLM task connected to the encoder; It is the classifier parameter connected to the encoder in the NSP task; is a hyperparameter used to balance the loss weights of the two tasks; is the loss function for the masked language model task; is the loss function for the next sentence prediction task; Represents the total number of masked words in the masked language model task; Represents the predicted word output by the masked language model task; Indicates that given the context, the initial model predicts the word at position i The probability of Indicates the size of the full Chinese dictionary that comes with the power monitoring text dictionary BERT; Represents the total number of samples in the next sentence prediction task; Indicates the predicted category of a monitoring alarm signal in the jth sample in the next sentence prediction task; Represents the classification probability of the relationship between two monitoring alarm signals in the jth sample.

[0008] In one embodiment, inputting the labeled historical monitoring alarm information into the pre-trained model includes: processing the labeled historical monitoring alarm information using data enhancement means to obtain enhanced samples; inputting the labeled historical monitoring alarm information and the enhanced samples into the pre-trained model; the data enhancement means include synonym replacement, random changes and sentence-level perturbations.

[0009] In one embodiment, the pre-trained model is fine-tuned, and the model parameters of the pre-trained model are updated through back propagation to obtain a trained network model, including: calculating the gradient through the back propagation algorithm and updating the weight parameters and bias vectors during probability calculation, so that the predicted probability of the pre-trained model for various monitoring alarm event labels is higher than the set threshold, thereby completing the fine-tuning of the pre-trained model.

[0010] In one embodiment, the output of the trained network model includes: Z = XW + b Formula (4) Among them, Z is the output of the trained network model, W is the weight matrix, b is the bias vector, and X is the vector output after performing the masked language model task and the next sentence prediction task.

[0011] In one embodiment, the method for calculating the predicted probability of the output of the trained network model includes: Formula (5) in, Represents the i-th element of the output Z of the trained network model After being converted into a probability distribution by the Sigmoid activation function, the obtained input corresponds to the probability of the i-th category.

[0012] An embodiment of the present invention also provides a substation monitoring alarm information analysis system, comprising: a processor and a memory for storing a computer program that can be run on the processor; wherein the processor executes the steps of any of the above methods when running the computer program.

[0013] An embodiment of the present invention further provides a storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0014] The substation monitoring alarm information analysis method and system provided in the embodiment of the present invention obtain unlabeled historical monitoring alarm information; input the unlabeled historical monitoring alarm information into an initial model for pre-training to obtain a pre-trained model; the initial model is a BERT language model based on a bidirectional Transformer encoder and a fully connected layer; the initial model can learn the semantic information and contextual relationship of the input text; the fully connected layer is used to perform text classification tasks; the unlabeled historical monitoring alarm information is annotated using expert system event rules, and the annotated data is checked against the scheduling log to obtain labeled historical monitoring alarm information; the labeled historical monitoring alarm information is input into the pre-trained model, the pre-trained model is fine-tuned, and the model parameters of the pre-trained model are updated through back propagation to obtain a trained network model; the monitoring alarm data during the monitoring operation of the substation control station is obtained, and the monitoring alarm data is input into the trained network model to obtain the substation monitoring alarm event analysis result. The solution provided by the present invention utilizes natural language processing technology based on a pre-trained language model to achieve autonomous analysis and judgment of substation monitoring alarm events, improve the accuracy and fault tolerance of event analysis strategies, and reduce the monitoring pressure of monitoring personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a flow chart of a method for analyzing and judging substation monitoring alarm information according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the overall framework of the initial model of an embodiment of the present invention; Figure 3 This is a schematic diagram of the input representation of the monitoring signal text in the BERT model according to an embodiment of the present invention; Figure 4 The figure is a diagram showing the internal structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

[0017] The embodiment of the present invention provides a method for analyzing and judging substation monitoring alarm information, such as Figure 1 As shown, the method includes: Step 101: Obtain unlabeled historical monitoring alarm information; Step 102: inputting the unlabeled historical monitoring alarm information into an initial model for pre-training to obtain a pre-trained model; the initial model is a BERT language model based on a bidirectional Transformer encoder and a fully connected layer; the initial model can learn the semantic information and contextual relationship of the input text; the fully connected layer is used for text classification tasks; Step 103: using the event rules of the expert system to mark the unlabeled historical monitoring alarm information, and checking the marked data against the scheduling log to obtain the labeled historical monitoring alarm information; Step 104: inputting the labeled historical monitoring alarm information into the pre-trained model, fine-tuning the pre-trained model, and updating the model parameters of the pre-trained model through back propagation to obtain a trained network model; Step 105: Acquire monitoring alarm data during the monitoring operation of the substation centralized control station, input the monitoring alarm data into the trained network model, and obtain the substation monitoring alarm event analysis result.

[0018] The initial model in this embodiment is a monitoring alarm information event analysis model based on natural language processing, which is used to identify and classify accident abnormal information in the power grid. The initial model in this embodiment is specifically a BERT language model based on a bidirectional Transformer encoder and a fully connected layer. The BERT language model can capture the context-related representation of each position and obtain rich semantic information. Through pre-training, the BERT language model can learn the semantic information and contextual relationships of the monitoring alarm text, so that the model can better understand and process the monitoring alarm text data in downstream tasks.

[0019] The historical monitoring alarm information in this embodiment is the data required for the intelligent analysis of substation monitoring alarm information, mainly including the historical monitoring alarm signals with time stamps collected by the SCADA system, the name information of substations and transmission lines, and other data. Among them, the historical monitoring alarm signal should contain five types of alarm information, namely accidents, abnormalities, displacements, over-limits, and notification information. This information contains all the factors (time, location, equipment name, and action status) possessed by a power grid monitoring alarm event. For example, when a tripping accident occurs, it will trigger signals such as the total accident (accident), protection outlet (accident), and switch displacement (displacement), and may be accompanied by abnormal signals such as the switch spring not storing energy, the switch control circuit being disconnected, and the protection device alarm. These signals can be used to construct an event sample library for model training, so as to pre-train the analysis and judgment model.

[0020] Specifically, the pre-training process is: performing a preprocessing operation on the unlabeled historical monitoring alarm information to obtain the preprocessed historical monitoring alarm information; the preprocessing operation includes text cleaning, text standardization, word segmentation and removal of stop words; inputting the preprocessed historical monitoring alarm information into the initial model so that the initial model performs feature vector conversion on the preprocessed historical monitoring alarm information to obtain a word feature vector, using the word feature vector to perform a masked language model task and a next sentence prediction task, and maximizing the probability of the masked language model task and the next sentence prediction task , to complete pre-training; the word feature vector includes a semantic embedding vector, a segment embedding vector and a position embedding vector; the semantic embedding vector is a word embedding vector that characterizes the mapping of each word fragment in a fixed dimension; the position embedding vector is a vector that characterizes the position of a word in an input sequence; the segment embedding vector is a vector used to distinguish different fragments to complete the classification task; the masked language model task is to randomly select words in the input substation monitoring alarm signal and replace them with masks, and predict the correct words corresponding to the masks; the next sentence prediction task is to determine the logical relationship between two substation monitoring alarm signals.

[0021] In this embodiment, the historical monitoring alarm information is pre-processed and then input into the initial model for pre-training, so that the model has the ability to recognize the substation monitoring information text. The pre-processing operation here includes text cleaning, text standardization, word segmentation and removal of stop words.

[0022] Text cleaning refers to removing interference items such as special characters, punctuation marks, HTML tags, etc. in the text to retain key text information. Text standardization refers to standardizing the text, such as converting uppercase letters to lowercase letters, removing extra spaces or line breaks, etc. Word segmentation refers to splitting the text into a sequence of words or subwords. In this embodiment, the word segmentation tool provided by the BERT language model can be used to segment the text into meaningful word units. Stop word removal refers to filtering out words that appear frequently in the text but lack actual meaning, such as prepositions, conjunctions, etc.

[0023] After preprocessing the historical monitoring alarm information, the preprocessed historical monitoring alarm information can be input into the initial model for pre-training. Figure 2 , Figure 2This is a schematic diagram of the overall framework of the initial model of an embodiment of the present invention. The BERT language model in the initial model adopts a multi-layer Transformer encoder structure. The input embedding passes through a multi-head self-attention mechanism (Multi-head Attention) and a feed-forward neural network (Feed-forward Neural Network), and gradually transmits information through the encoder layer. These encoder layers enable the BERT language model to capture the context-related representation of each position, thereby obtaining rich semantic information. Since the BERT language model only retains the encoder part in the Transformer, the BERT model only learns the text semantics during training, and does not perform decoding to complete specific tasks.

[0024] After the preprocessed historical monitoring alarm information text is input into the BERT language model, the BERT language model will use WordPiece or similar word segmentation methods to divide the historical monitoring alarm information text into word fragments and add special tags to each fragment. Among them, the "[CLS]" tag is used to indicate the beginning of the text sequence, and the "[SEP]" tag is used to separate different fragments. These word fragments will be converted into three feature vectors: semantic embedding vectors (Token Embeddings), segment embedding vectors (segmentEmbeddings), and position embedding vectors (Position Embeddings). Among them, the semantic embedding vector is a word embedding vector that represents the mapping of each word fragment in a fixed dimension; the position embedding vector is a vector that represents the position of the word in the input sequence; the segment embedding vector is a vector used to distinguish different fragments to complete the classification task. For example, the input representation corresponding to "Limu Substation Accident Total Signal Action" is as follows Figure 3 shown.

[0025] After converting the text into vector representation, the BERT model will perform two parallel tasks: Masked Language Modeling (MLM) and Next Sentence Prediction (NSP) to learn language representation, thereby improving the generalization ability of the model. Among them, MLM refers to randomly selecting some words in the input substation monitoring alarm signal text and replacing them with masks, allowing the model to predict the correct words of these masks, so as to train the model's ability to understand the context. NSP is a continuity prediction task at the sentence level, that is, to determine the logical relationship between two substation monitoring alarm signals. This task can help the model understand the semantic connection between various alarm signals, thereby improving the model's reasoning ability.

[0026] The pre-training goal of BERT is to maximize the probability of the masked language model and the next sentence prediction task. The entire pre-training process is completed through iterative training, using an optimization algorithm (such as batch stochastic gradient descent) to iteratively update the parameters of the initial model until the probability of the masked language model task and the next sentence prediction task reaches the maximum or until the number of iterations of the initial model reaches a preset number of iterations.

[0027] Here, the loss function of the pre-trained model can be expressed as follows: Formula (1) Formula (2) Formula (3) in, are the parameters of the encoder part; are the parameters of the output layer of the MLM task connected to the encoder; It is the classifier parameter connected to the encoder in the NSP task; is a hyperparameter used to balance the loss weights of the two tasks. It is a hyperparameter used to balance the loss weights of the two tasks. By adjusting The value of can control the emphasis of the model on MLM and NSP tasks during pre-training; is the loss function for the masked language model task; is the loss function for the next sentence prediction task; Represents the total number of masked words in the masked language model task; Represents the predicted word output by the masked language model task; Indicates that given the context, the initial model predicts the word at position i The probability of Indicates the size of the full Chinese dictionary that comes with the power monitoring text dictionary BERT, which covers all Chinese words related to the power monitoring field; Represents the total number of samples in the next sentence prediction task; Indicates the prediction category of a monitoring alarm signal in the jth sample in the next sentence prediction task, including IsNext (a monitoring alarm signal has a contextual association with the previous monitoring alarm signal) and NotNext (a monitoring alarm signal is independent of the previous monitoring alarm signal); Represents the classification probability of the relationship between two monitoring alarm signals in the jth sample.

[0028] Through the pre-training process, the BERT model can learn the semantic information and contextual relationships of monitoring warning texts. This enables the model to better understand and process monitoring warning text data in downstream tasks.

[0029] After pre-training is completed, samples related to event synthesis can be extracted from historical monitoring alarm information and labeled according to the existing expert system event rules to construct various types of labeled monitoring alarm events. For positive samples, after labeling, they are checked against the dispatch log and entered into the event sample library for model training after verification. In the case of a lack of power monitoring alarm information samples corresponding to a certain type of alarm event, data enhancement methods can be used to improve the training effect of the event analysis model.

[0030] That is, in one embodiment, inputting the labeled historical monitoring alarm information into the pre-trained model includes: using data enhancement means to process the labeled historical monitoring alarm information to obtain enhanced samples; inputting the labeled historical monitoring alarm information and the enhanced samples into the pre-trained model; the data enhancement means include synonym replacement, random changes and sentence-level perturbations.

[0031] Synonym Replacement: Replace some words in the input text with their synonyms to increase sample diversity. For example, the power grid in a certain region uniformly names the signals related to line overcurrent protection as "xx line protection action". When performing data enhancement, the signal can be modified into various forms such as "xx line protection outlet", "xx line overcurrent outlet", "xx line I section action" and labeled separately, thereby enhancing the applicability of the model for accident analysis in power grids in different regions.

[0032] Random Modification: Randomly select a position in the input text and change a randomly selected word to simulate the situation where individual signals are not named in a standardized way. This helps force the model to pay more attention to the contextual information of the remaining words, improves the robustness of the model, and increases the diversity of samples.

[0033] Sentence-level perturbation: Perform sentence-level operations on the input text, such as random deletion or random insertion of a warning message, to simulate the situation where samples are missing or confused due to the mis-sending or missed sending of the warning signal, and improve the model's ability to handle abnormal inputs.

[0034] After obtaining the labeled historical monitoring alarm information, the labeled historical monitoring alarm information can be input into the pre-trained model for fine-tuning. Here, specifically, the gradient is calculated by the back propagation algorithm and the weight parameters and bias vectors in the probability calculation are updated, so that the predicted probability of the pre-trained model for various monitoring alarm event labels is higher than the set threshold, and the pre-trained model is fine-tuned.

[0035] Here, it should be noted that: during the pre-training process, the main goal of updating the model parameters is to learn the basic characteristics of professional terms in the power field based on general language understanding; while during the fine-tuning process, the main goal of updating the model parameters is to conduct supervised learning for specific downstream tasks (classifying alarm information according to alarm event labels). That is, in the fine-tuning stage, the model parameter adjustment is aimed at adapting to specific data distribution and task requirements, and optimizing the performance of the alarm signal classification task.

[0036] This embodiment will input the sentence vector extracted by the BERT language model into the fully connected layer to obtain the intelligent judgment result. The fully connected layer is a commonly used neural network layer for text classification tasks. Its basic calculation process is as follows: The feature representation in the form of a vector or matrix output by the BERT language model is recorded as X. Each output of the fully connected layer can be regarded as each node of the previous layer multiplied by a weight coefficient W, and finally a bias value b is added to obtain: Z = XW + b Formula (4) Among them, Z is the output of the trained network model (that is, the output of the fully connected layer, representing the score or original output of the sample), W is the weight matrix, b is the bias vector, and X is the vector output after performing the masked language model task and the next sentence prediction task.

[0037] In order to introduce nonlinear properties, a nonlinear activation function is usually applied after the fully connected layer. Common activation functions include ReLU (Rectified Linear Unit), Sigmoid, and Softmax. Since the eventization of substation monitoring alarm information involves multi-label classification problems (a monitoring alarm information sequence may contain multiple events at the same time), this embodiment uses the Sigmoid function applicable to multi-classification problems to map features to the predicted probability of event categories. The calculation formula of the Sigmoid function is as follows: Formula (5) in, Represents the i-th element of the output Z of the trained network model After being converted into a probability distribution by the Sigmoid activation function, the obtained input corresponds to the probability of the i-th category, that is, the probability that a certain monitoring alarm information sequence input belongs to the i-th category power grid alarm event.

[0038] During the fine-tuning process, it is necessary to define a loss function suitable for the task. For classification tasks, the cross entropy loss function is generally used to measure the difference between the model's prediction results and the true labels. By minimizing the loss function, the model can learn to adjust parameters to better fit the training data, making the prediction results closer to the true labels. At the same time, the back propagation algorithm is used to calculate the gradient and update the weight parameters W and b to optimize the classification performance of the model, so that the model can learn to better classify and discriminate the input features. When the model's predicted probability for various monitoring alarm event labels is higher than the set threshold, the model fine-tuning is completed.

[0039] After completing the fine-tuning of the model, the trained network model can be used to perform intelligent analysis of substation monitoring alarm information. This embodiment will batch export the monitoring alarm signals collected by the SCADA system during the monitoring operation duty, and input them into the trained model as a verification set to generate event-based analysis results. The monitoring personnel will check the analysis results. If there is an error in the analysis or the analysis result cannot be generated, the monitoring personnel will modify it to the correct label and include the alarm signal sequence in the sample library to further optimize the model; when the accuracy of the analysis reaches the expected standard, the model can be deployed in the field.

[0040] This embodiment has the following advantages: (1) The method provided in this embodiment uses the BERT pre-trained language model for text processing. Compared with the traditional text vectorization method, the BERT model can learn richer semantic feature representations by pre-training on massive text data, thereby better capturing the meaning and semantic relationships in the text.

[0041] (2) Data enhancement methods are used to simulate various scenarios that may occur in actual operation and construct corresponding alarm signal sequences. On the one hand, this improves the applicability of the model and the error tolerance of judgment. On the other hand, it makes up for the problem of insufficient number of original alarm signal samples corresponding to some event types, thereby improving the training quality of the alarm judgment model.

[0042] The substation monitoring alarm information analysis method provided in the embodiment of the present invention obtains unlabeled historical monitoring alarm information; inputs the unlabeled historical monitoring alarm information into an initial model for pre-training to obtain a pre-trained model; the initial model is a BERT language model based on a bidirectional Transformer encoder and a fully connected layer; the initial model can learn the semantic information and contextual relationship of the input text; the fully connected layer is used to perform text classification tasks; the unlabeled historical monitoring alarm information is annotated using expert system event rules, and the annotated data is checked against the scheduling log to obtain labeled historical monitoring alarm information; the labeled historical monitoring alarm information is input into the pre-trained model, the pre-trained model is fine-tuned, and the model parameters of the pre-trained model are updated through back propagation to obtain a trained network model; the monitoring alarm data during the monitoring operation of the substation control station is obtained, and the monitoring alarm data is input into the trained network model to obtain the substation monitoring alarm event analysis result. The solution provided by the present invention utilizes natural language processing technology based on a pre-trained language model to achieve autonomous analysis and judgment of substation monitoring alarm events, improve the accuracy and fault tolerance of event analysis strategies, and reduce the monitoring pressure of monitoring personnel.

[0043] In order to implement the method of an embodiment of the present invention, an embodiment of the present invention also provides a substation monitoring alarm information analysis system, including: a processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, the steps of any one of the above methods are executed.

[0044] The above-mentioned system provided in this embodiment belongs to the same concept as the above-mentioned method embodiment. The specific implementation process thereof is detailed in the method embodiment and will not be repeated here.

[0045] In order to implement the method of the embodiment of the present invention, the embodiment of the present invention also provides a computer program product, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps of the above method.

[0046] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides an electronic device (computer device). Specifically, in one embodiment, the computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05 and a memory (not shown in the figure) connected through a system bus. Among them, the processor A01 of the computer device is used to provide computing and control capabilities. The memory of the computer device includes an internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 in the non-volatile storage medium A06. The network interface A02 of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor A01, the method of any one of the above embodiments is implemented. The display screen A04 of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device A05 of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0047] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0048] The device provided by the embodiment of the present invention includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the method of any one of the above embodiments is implemented.

[0049] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0050] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0051] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0053] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0054] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0055] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0056] It can be understood that the memory of the embodiment of the present invention can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM, SyncLink Dynamic Random Access Memory), and direct RAMbus random access memory (DRRAM, Direct Rambus Random Access Memory).The memories described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memories.

[0057] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0058] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for analyzing and judging substation monitoring alarm information, characterized in that: The method comprises: Obtain unlabeled historical monitoring alarm information; Inputting the unlabeled historical monitoring alarm information into an initial model for pre-training to obtain a pre-trained model; the initial model is a BERT language model based on a bidirectional Transformer encoder and a fully connected layer; the initial model can learn the semantic information and contextual relationship of the input text; the fully connected layer is used for text classification tasks; The unlabeled historical monitoring alarm information is labeled by using the event rules of the expert system, and the labeled data is checked against the scheduling log to obtain the labeled historical monitoring alarm information; Inputting the labeled historical monitoring alarm information into the pre-trained model, fine-tuning the pre-trained model, and updating the model parameters of the pre-trained model through back propagation to obtain a trained network model; The monitoring alarm data during the monitoring operation of the substation centralized control station is obtained, and the monitoring alarm data is input into the trained network model to obtain the substation monitoring alarm event analysis result.

2. The method for analyzing and judging substation monitoring alarm information according to claim 1 is characterized in that: Inputting the unlabeled historical monitoring alarm information into the initial model for pre-training to obtain a pre-trained model includes: Performing a preprocessing operation on the unlabeled historical monitoring alarm information to obtain preprocessed historical monitoring alarm information; the preprocessing operation includes text cleaning, text standardization, word segmentation, and stop word removal; The preprocessed historical monitoring alarm information is input into the initial model so that the initial model performs feature vector conversion on the preprocessed historical monitoring alarm information to obtain a word feature vector, and uses the word feature vector to perform a masked language model task and a next sentence prediction task, and maximizes the probability of the masked language model task and the next sentence prediction task to complete pre-training; the word feature vector includes a semantic embedding vector, a segmented embedding vector and a position embedding vector; the semantic embedding vector is a word embedding vector that represents the mapping of each word fragment in a fixed dimension; the position embedding vector is a vector that represents the position of a word in an input sequence; the segmented embedding vector is a vector used to distinguish different fragments to complete a classification task; the masked language model task is to randomly select words in the input substation monitoring alarm signal and replace them with masks, and predict the correct words corresponding to the masks; the next sentence prediction task is to determine the logical relationship between two substation monitoring alarm signals.

3. The method for analyzing and judging substation monitoring alarm information according to claim 2 is characterized in that: Maximizing the probability of the masked language model task and the next sentence prediction task, including: The parameters of the initial model are iteratively updated using an optimization algorithm until the probabilities of the masked language model task and the next sentence prediction task reach a maximum or until the number of iterations of the initial model reaches a preset number of iterations.

4. The method for analyzing and judging substation monitoring alarm information according to claim 3 is characterized in that: The loss function when using the optimization algorithm to iterate is: Formula (1) Formula (2) Formula (3) in, are the parameters of the encoder part; are the parameters of the output layer of the MLM task connected to the encoder; It is the classifier parameter connected to the encoder in the NSP task; is a hyperparameter used to balance the loss weights of the two tasks; is the loss function for the masked language model task; is the loss function for the next sentence prediction task; Represents the total number of masked words in the masked language model task; Represents the predicted word output by the masked language model task; Indicates that given the context, the initial model predicts the position Words in The probability of Indicates the size of the full Chinese dictionary that comes with the power monitoring text dictionary BERT; Represents the total number of samples in the next sentence prediction task; Indicates the next sentence prediction task The predicted category of a monitoring alarm signal in samples; Indicates The classification probability of the relationship between two monitoring alarm signals in samples.

5. The method for analyzing and judging substation monitoring alarm information according to claim 1 is characterized in that: Inputting the labeled historical monitoring alarm information into the pre-trained model includes: Processing the labeled historical monitoring alarm information using data enhancement means to obtain enhanced samples; The labeled historical monitoring alarm information and the enhanced samples are input into the pre-trained model; the data enhancement means include synonym replacement, random change and sentence-level disturbance.

6. The method for analyzing and judging substation monitoring alarm information according to claim 5 is characterized in that: Fine-tuning the pre-trained model and updating the model parameters of the pre-trained model by back-propagation to obtain a trained network model, including: The gradient is calculated through the back propagation algorithm and the weight parameters and bias vectors in the probability calculation are updated so that the predicted probability of the pre-trained model for various monitoring alarm event labels is higher than the set threshold, thereby completing the fine-tuning of the pre-trained model.

7. The method for analyzing and judging substation monitoring alarm information according to claim 6 is characterized in that: The output of the trained network model includes: Z = XW + b Formula (4) Among them, Z is the output of the trained network model, W is the weight matrix, b is the bias vector, and X is the vector output after performing the masked language model task and the next sentence prediction task.

8. The method for analyzing and judging substation monitoring alarm information according to claim 7 is characterized in that: The prediction probability calculation method of the output of the trained network model includes: Formula (5) in, Represents the i-th element of the output Z of the trained network model After being converted into a probability distribution by the Sigmoid activation function, the obtained input corresponds to the probability of the i-th category.

9. A substation monitoring alarm information analysis system, characterized in that: include: A processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, the steps of the method according to any one of claims 1 to 8 are executed.

10. A storage medium storing a computer program, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Citation Information

Patent Citations

  • Model training method, electronic equipment and storage medium

    CN114611692A

  • BERT-based power grid fault diagnosis method

    CN114661905A

  • Automobile technology planning and decision-making method and system based on GPT network model

    CN116468298A

  • Improved molecule-protein reaction prediction classification method based on Transform network

    CN116469470A

  • Early prediction method for acute kidney injury based on Bert model

    CN117743578A

Cited By

  • Transform-based network security alarm noise reduction method

    CN119449581A

  • Method and device for detecting abnormal operation behavior of software system and medium

    CN121456684A