Power grid intelligent anti-misoperation method and system based on BiLSTM-CRF model

Abstract

The invention discloses an intelligent power grid anti-misoperation method and system based on a BiLSTM-CRF model, and relates to the field of intelligent power grid dispatching, and the method comprises the steps: obtaining a structured data set from a power grid dispatching system; based on the structured data set and a BiLSTM-CRF model, constructing a first model; training the first model to obtain a second model; performing named entity identification on the to-be-processed power grid dispatching record by using the second model; extracting related attribute information from the identified named entities, and constructing an attribute dictionary; based on the attribute dictionary, receiving and analyzing an operation order input by the user, and performing operation according to an analysis result; the intelligent level of the power grid dispatching system is improved; the accuracy of named entity recognition is improved; recognition errors are reduced; the requirement of manual intervention is reduced; real-time decision support is provided for dispatchers, and potential misoperation risks are helped to be recognized and avoided; and the safety and the reliability of power grid dispatching operation are improved.

Description

Technical Field

[0001] The present invention relates to the field of smart grid dispatching, and in particular to a method and system for preventing smart misoperation of a power grid based on a BiLSTM-CRF model. Background Art

[0002] With the continuous growth of global electricity demand and the continuous expansion of power grid scale, the structure and operation mode of power system are becoming more and more complex. Modern power system not only has to face multi-source heterogeneous energy access, but also has to ensure efficient, safe and stable operation. This puts higher requirements on power grid dispatching operation. Power grid dispatching operation is the core link of power system operation. Its main task is to reasonably allocate power resources according to real-time load and power generation conditions to ensure the stable operation of power system. However, in the complex and changeable power system environment, the complexity and difficulty of dispatching operation are also increasing. Traditional power grid dispatching operation relies on manual experience and manual operation, which is not only inefficient but also prone to human errors. Dispatchers need to process a lot of information, including operation monitoring, mode adjustment, power dispatching, etc. Any carelessness may lead to misoperation, which in turn causes power accidents. In order to reduce the occurrence of misoperation and improve the accuracy and efficiency of dispatching operation, intelligent power grid dispatching system came into being. In recent years, with the rapid development of artificial intelligence technology, especially the breakthrough of deep learning technology in the field of natural language processing, it has provided new ideas for the intelligent dispatching of power grid. Through learning and training of large amounts of data, deep learning technology can automatically extract and identify complex features, achieve efficient pattern recognition and decision support. In power systems, deep learning technology can be used in equipment status monitoring, fault diagnosis, load forecasting and other aspects. Named entity recognition (NER) is an important task in natural language processing, which aims to identify specific entities (such as names of people, places, and organizations) from unstructured text. In power systems, named entity recognition can be used to identify information such as equipment names, operation steps, and parameters in dispatch operation records. Traditional NER methods mainly rely on rules and dictionaries, which are difficult to cope with diverse and complex power dispatch data. The BiLSTM-CRF model in deep learning combines the advantages of bidirectional long short-term memory network (BiLSTM) and conditional random field (CRF), can effectively handle sequence labeling problems, and performs well in named entity recognition tasks. In addition, the construction of a knowledge base for power grid dispatch operations is also an important research direction. By deeply analyzing and extracting dispatch operation records and establishing a complete knowledge base, real-time decision support can be provided to dispatchers to reduce the occurrence of misoperation. The construction of a knowledge base requires efficient and accurate entity recognition and attribute extraction methods, and deep learning technology has significant advantages in this regard.

[0003] The prior art discloses a method for preventing misoperation of power grids based on depth-first search. By establishing a power grid topology model and using a depth-first search algorithm to perform topology disconnection analysis, risk points (such as cutting points and cutting edges) in the power grid are quickly identified, and misjudgment of operation commands is performed based on the risk points to ensure the safety of the operation. The system also includes a device status library to update the device status in real time to ensure the accuracy and real-time nature of the operation. Although the method uses depth-first search to reduce complexity, the processing capacity and performance of the system may still be limited as the scale and complexity of the power grid further increase. In addition, the establishment and simplification of the power grid topology model depends on the accuracy of the initial data. Any data deviation may affect the accuracy and reliability of the overall system, making it difficult to accurately prevent misoperation and lacking intelligent misjudgment prevention. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: how to improve the intelligence and safety of the power grid dispatching system, reduce human error operations, and improve the power grid operation efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for intelligently preventing misoperation of a power grid based on a BiLSTM-CRF model, comprising:

[0008] Collect historical operation records, equipment information, and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set;

[0009] Based on the structured data set and the BiLSTM-CRF model, a first model is constructed; the first model is trained to obtain a second model;

[0010] Using the second model to perform named entity recognition on the power grid dispatch records to be processed;

[0011] Extract relevant attribute information from the identified named entities and construct an attribute dictionary;

[0012] Based on the attribute dictionary, receive and analyze the operation commands input by the user, and perform operations based on the analysis results.

[0013] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0014] The data cleaning and preprocessing include: denoising, word segmentation, and labeling; the construction of the structured data set includes ensuring the uniformity of field names and data types.

[0015] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0016] The constructing of the first model comprises:

[0017] Input the structured data set into the BiLSTM-CRF model. The data input process includes word embedding and input format.

[0018] Word embedding: convert each word in the original text data into a vector of fixed dimension;

[0019] Input format: The text sequence of each operation record is converted into a word embedding matrix. Each row of the matrix represents the vector representation of a word. The entire matrix is ​​used as the input of BiLSTM.

[0020] The output of the first model includes the feature vector output by the BiLSTM layer, which represents the context information of each word; the CRF layer performs label prediction based on the output of the BiLSTM and outputs the label of each word.

[0021] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0022] The training of the first model to obtain the second model includes:

[0023] 80% of the preprocessed structured data set is used as a training set, 10% as a validation set, and 10% as a test set; the outputs of the BiLSTM and CRF layers are trained; the optimization algorithm is used to minimize the loss function and update the parameters of the first model.

[0024] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0025] The training of the first model to obtain the second model further comprises:

[0026] Forward propagation: The input sequence is forward propagated through the BiLSTM layer and the CRF layer to calculate the prediction results of the first model;

[0027] Loss calculation: Calculate the loss between the first model prediction result and the true label;

[0028] Back propagation: Calculate the gradient through the back propagation algorithm and update the first model parameters;

[0029] Iterative training: Repeat forward propagation, loss calculation, and back propagation until the first model reaches the preset performance index on the training set and validation set to obtain the second model.

[0030] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0031] The using the second model to perform named entity recognition on the power grid dispatch record to be processed includes:

[0032] The power grid dispatch record to be processed is passed into the second model for forward propagation to obtain output, including the label prediction result of each word;

[0033] The label prediction results are mapped back to the corresponding actual entity names to form the final named entity recognition results and stored in memory or database.

[0034] As a preferred solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model, the following is the solution:

[0035] The receiving and analyzing the operation command input by the user, and performing the operation according to the analysis result comprises:

[0036] The input operation command is analyzed and combined with the information in the attribute dictionary to identify possible risk points. If the equipment involved in the operation command is in a faulty state, it is identified as a risk point; if an operation is inconsistent with the historical record, it is identified as a potential erroneous operation; for the identified risk points, the system returns modification suggestions to prompt the user of possible erroneous operations.

[0037] In a second aspect, an embodiment of the present invention provides a power grid intelligent error-proof operation system based on a BiLSTM-CRF model, including:

[0038] The data set acquisition module is used to collect historical operation records, equipment information and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set;

[0039] A model building module, used for building a first model based on a structured data set and a BiLSTM-CRF model; training the first model to obtain a second model;

[0040] A named entity recognition module, used for performing named entity recognition on the power grid dispatch record to be processed using the second model;

[0041] The attribute dictionary building module is used to extract relevant attribute information from the identified named entities and build an attribute dictionary;

[0042] The analysis operation module is used to receive and analyze the operation command input by the user based on the attribute dictionary, and perform operations according to the analysis results.

[0043] In a third aspect, an embodiment of the present invention provides a computing device, including:

[0044] Memory and processor;

[0045] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent anti-misoperation method for the power grid based on the BiLSTM-CRF model as described in any embodiment of the present invention.

[0046] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the intelligent anti-misoperation method for a power grid based on a BiLSTM-CRF model.

[0047] Beneficial effects of the present invention: The present invention aims at the defects of the risk of misoperation, low efficiency of manual processing, insufficient recognition accuracy and the like in the existing power grid dispatching operation. By constructing a named entity recognition model based on deep learning dedicated to the power field, the present invention realizes efficient and accurate recognition and attribute extraction of equipment names, operation steps and parameters in power grid dispatching operation, thereby improving the intelligence level of the power grid dispatching system; the accuracy of named entity recognition is improved: the BiLSTM-CRF model is adopted to optimize the recognition of specific named entities in the power dispatching data to reduce recognition errors; automated attribute extraction is realized: through the automated attribute extraction method, richer contextual information is provided to support dispatching decisions and reduce the need for manual intervention; a dynamic error prevention knowledge base is constructed: the identified and extracted named entities and their attribute information are stored in the knowledge base to provide real-time decision support for dispatchers and help identify and avoid potential risks of misoperation; intelligent error prevention judgment: in the operation and dispatching order processing, risk points are identified through intelligent judgment, and modification suggestions are fed back in time to improve the safety and reliability of power grid dispatching operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0049] Figure 1 It is an overall flow chart of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model described in the first embodiment of the present invention;

[0050] Figure 2 It is a schematic diagram of a BiLSTM-CRF model of the intelligent anti-misoperation method for a power grid based on a BiLSTM-CRF model described in the first embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0054] Example 1

[0055] Reference Figure 1-Figure 2 , which is the first embodiment of the present invention, and provides a method for intelligently preventing misoperation of a power grid based on a BiLSTM-CRF model, comprising:

[0056] S1: Collect historical operation records, equipment information and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set;

[0057] S2: Based on the structured data set and the BiLSTM-CRF model, a first model is constructed; the first model is trained to obtain a second model;

[0058] S3: using the second model to perform named entity recognition on the power grid dispatch records to be processed;

[0059] S4: Extract relevant attribute information from the identified named entities and build an attribute dictionary;

[0060] S5: Based on the attribute dictionary, receive and analyze the operation command input by the user, and perform operations according to the analysis results.

[0061] In the above step S1, collecting historical operation records, equipment information and operating procedures from the power grid dispatching system includes:

[0062] Historical operation records: The dispatcher's operation records are exported from the power grid dispatching system. These records include the timestamp of each operation, the type of operation (such as switching equipment, adjusting loads, etc.), the operation object, the operator information and the operation results.

[0063] Equipment information: Collect detailed information about various types of equipment in the power grid, including equipment name, number, type, status, location, etc. This information helps to accurately identify and describe the equipment in subsequent analysis.

[0064] Operating procedures: Collect the operating procedures and specification documents of power grid dispatching, which usually describe in detail the operating steps, precautions and emergency response plans for each equipment.

[0065] The data cleaning and preprocessing in the above step S1 includes:

[0066] Denoising: Check for errors or invalid information in the data, such as redundant records, inconsistent formats, missing values, etc. Statistical methods or rules can be used for denoising, duplicate operation records can be deleted, and missing device information can be filled in or marked as missing based on the context.

[0067] Word segmentation: Perform word segmentation on text data (such as operation records and operating procedures). Word segmentation is to divide continuous text into meaningful words or phrases. In power dispatching data, professional word segmentation tools or dictionary-based word segmentation algorithms can be used to identify key terms such as equipment names and operating instructions.

[0068] Annotation: Annotate the segmented data to provide necessary label information for subsequent model training. Annotation can be divided into the following categories: Named entity annotation: Identify and annotate the device name, operation type, etc.

[0069] For example, "switch A" is labeled as "device name" and "off" is labeled as "operation instruction." Attribute labeling: Label the attributes related to the device, such as device model, operating parameters, etc.

[0070] Construction of structured data sets: The cleaned, segmented and annotated data is converted into a structured data set to facilitate subsequent model training and analysis. This is usually represented in the form of a table, where each row corresponds to an operation record, and the columns include time, operator, device name, operation type, attributes, etc.

[0071] Unified format: Ensure that all data is in a consistent format to facilitate subsequent processing.

[0072] In an embodiment of the present application, the data may be converted into a commonly used format (CSV) and the uniformity of field names and data types may be ensured.

[0073] In another possible implementation, in the process of data format conversion and unification, in addition to converting the data into the commonly used CSV format and ensuring the unification of field names and data types, the data can also be converted into other commonly used data formats, such as JSON, XML, Parquet, or a format supported by a database (such as an SQL database table). These formats each have their own advantages, for example, JSON is suitable for nested data structures, and Parquet is suitable for large data storage and has efficient compression performance.

[0074] Data segmentation: Divide the structured data set into training set, validation set and test set. Use 80% of the data for training and 20% for validation and testing.

[0075] It should be noted that data segmentation is to ensure the generalization ability and performance evaluation of the model.

[0076] The construction of the first model in the above step S2 includes:

[0077] It should be noted that if Figure 2 As shown in the figure, the BiLSTM-CRF model is a powerful structure that combines the bidirectional long short-term memory network (BiLSTM) and the conditional random field (CRF). It is mainly used for sequence labeling tasks such as named entity recognition (NER), part-of-speech tagging (POS), syntactic analysis, etc. The figure shows the BiLSTM-CRF model diagram. In the bidirectional LSTM (BiLSTM), LSTM (Long Short-Term Memory Network) is a special recurrent neural network (RNN) that can effectively capture long-term dependencies and control the flow of information by introducing forget gates, input gates, and output gates. BiLSTM consists of two LSTMs: one is a forward LSTM (processing sequences from left to right) and the other is a reverse LSTM (processing sequences from right to left). In this way, contextual information can be considered at the same time to improve the understanding of the sequence. The conditional random field (CRF) is a probabilistic model for labeling and segmenting sequence data. It can capture the dependency between labels and consider contextual information. CRF handles the dependency of label sequences by defining the transition probability between labels to ensure that the output label sequence is reasonable in the context.

[0078] BiLSTM processes the input sequence through two LSTM units (forward and backward). Given an input sequence X = (x1, x2, ..., x T ), the output of BiLSTM is represented as the hidden state h t :

[0079]

[0080]

[0081] The final BiLSTM output is:

[0082]

[0083] Among them, “;” means concatenating the forward and reverse hidden states.

[0084] Input the structured data set into the BiLSTM-CRF model. The data input process includes word embedding and input format.

[0085] Word embedding: Convert each word in the original text data into a vector of fixed dimension.

[0086] In an embodiment of the present application, a pre-trained word vector model (Word2Vec) may be used or initialized by training word embeddings.

[0087] In another possible implementation, GloVe (Global Vectors for Word Representation) can also be used for initialization:

[0088] GloVe is a word vector model based on global word frequency statistics. It generates word vectors by factoring the co-occurrence matrix. Unlike Word2Vec, GloVe pays more attention to the statistical information of the global corpus and can capture the global co-occurrence pattern of words.

[0089] In another possible implementation, FastText can also be used for initialization:

[0090] FastText not only considers word-level information, but also sub-word information (such as character n-gram). For some out-of-vocabulary words, FastText can generate more reasonable word vectors through sub-word information, which is particularly suitable for text data containing a large amount of professional terms or deformations.

[0091] In another possible implementation, BERT (Bidirectional Encoder Representations from Transformers) can also be used for initialization:

[0092] BERT is a pre-trained language model based on the Transformer network structure. By using the Masked Language Model (MLM) and Next Sentence Prediction (NSP) tasks for pre-training, BERT is able to capture rich contextual information. The output of BERT can be used as the initialization of word embedding to obtain a more context-sensitive representation.

[0093] Input format: The text sequence of each operation record is converted into a word embedding matrix. Each row of the matrix represents the vector representation of a word. The entire matrix is ​​used as the input of BiLSTM.

[0094] The output of the first model includes the feature vector output by the BiLSTM layer, which represents the context information of each word; the CRF layer performs label prediction based on the output of the BiLSTM and outputs the label of each word.

[0095] Exemplarily, the label of each word includes device name, operation instruction, attribute, etc.

[0096] In the above step S2, the first model is trained to obtain the second model including:

[0097] Use 80% of the preprocessed structured dataset as a training set, 10% as a validation set, and 10% as a test set;

[0098] It should be noted that the training set is used to train the model, the validation set is used to adjust hyperparameters and prevent overfitting, and the test set is used to evaluate the performance of the model.

[0099] Define the loss function: Use the cross entropy loss function to train the output of the BiLSTM and CRF layers; the loss function of the CRF layer is based on the conditional probability of the label sequence, including the log-likelihood loss, expressed as:

[0100]

[0101] Where N is the number of samples, C is the number of categories; y ij is the true label of sample i in category j is 0 or 1, is the predicted probability of sample i in category j.

[0102] Using an optimization algorithm to minimize the loss function, update the parameters of the first model;

[0103] In an embodiment of the present application, an Adam optimizer is used to minimize the loss function. The Adam optimizer can automatically adjust the learning rate and converge quickly during the training process.

[0104] The training steps include:

[0105] Forward propagation: The input sequence is forward propagated through the BiLSTM layer and the CRF layer to calculate the prediction results of the first model;

[0106] Loss calculation: Calculate the loss between the first model prediction result and the true label;

[0107] Back propagation: Calculate the gradient through the back propagation algorithm and update the first model parameters;

[0108] Iterative training: Repeat forward propagation, loss calculation, and back propagation until the first model reaches the preset performance index on the training set and validation set to obtain the second model.

[0109] It should be noted that when setting and verifying performance indicators, evaluation indicators: accuracy, precision, recall and F1 value are used to evaluate the performance of the model. These indicators measure the performance of the model in the entity recognition task.

[0110] Hyperparameter Tuning: Select the best hyperparameter settings by testing different hyperparameter combinations (such as learning rate, number of LSTM units, batch size, etc.) on the validation set.

[0111] Model validation: Perform model validation on the validation set and adjust hyperparameters during training to prevent overfitting.

[0112] It should be noted that the power grid dispatch records to be processed need to be preprocessed to be consistent with the data format during training. Including:

[0113] Word segmentation: Segment the dispatch record text into words to ensure that the model can identify the boundary of each word.

[0114] Vectorization: Convert the segmented text into corresponding word vectors, usually generated using a trained word embedding model (such as Word2Vec, GloVe).

[0115] Padding and masking: For variable-length input sequences, padding is used to unify them to a fixed length, and a mask is used to identify the padding part.

[0116] In the above step S3, using the second model to perform named entity recognition on the power grid dispatch record to be processed includes:

[0117] The power grid dispatch record to be processed is passed into the second model for forward propagation to obtain output, including the label prediction result of each word;

[0118] Exemplarily, the label prediction results may include device name, operation instructions, time, location, etc. The specific category depends on the label set defined during model training.

[0119] The label prediction results are mapped back to the corresponding actual entity names to form the final named entity recognition results and stored in memory or database.

[0120] Furthermore, in power grid dispatching operations, the system needs to process real-time incoming dispatching records to ensure the immediacy and efficiency of named entity recognition. The specific steps are as follows:

[0121] a. Real-time data access

[0122] Data stream access: The system receives new scheduling records in real time through an interface (API or message queue, etc.) to ensure timely arrival of data.

[0123] Asynchronous processing: To improve processing efficiency, you can use an asynchronous programming model to avoid blocking the main thread and ensure that the system remains responsive while receiving new data.

[0124] b. Real-time preprocessing

[0125] Fast word segmentation and vectorization: For the received scheduling records, word segmentation and vectorization are quickly performed to ensure that the model input is ready in the shortest time.

[0126] Concurrent processing: When multiple scheduling records arrive at the same time, a concurrent processing strategy is adopted, and a thread pool or process pool can be used to improve processing efficiency.

[0127] c. Real-time reasoning

[0128] Fast reasoning: The processed input data is quickly passed into the BiLSTM-CRF model for named entity recognition.

[0129] Batch processing: If you need to process a large number of scheduling records, you can use batch reasoning to improve the model reasoning speed.

[0130] d. Real-time feedback and storage

[0131] Feedback mechanism: Provide real-time feedback on the identified entities, such as displaying the results to the dispatcher or triggering subsequent processing procedures.

[0132] Dynamically update the knowledge base: The identified named entities are dynamically stored in the knowledge base to ensure that the information in the system remains up to date.

[0133] e. Monitoring and Optimization

[0134] Performance monitoring: Monitor the latency and accuracy of real-time processing to ensure that the system can complete processing tasks within the specified time.

[0135] Regular optimization: Based on the performance feedback from real-time processing, regular optimization of models and processing flows is performed to improve the system’s response speed and recognition accuracy.

[0136] The relevant attribute information extracted in the above step S4 includes the status, type, operation time, etc. of the device, and the attribute extraction can be performed by defining rules or training additional models; the extracted attribute information is organized into structured data to facilitate subsequent storage and retrieval.

[0137] It should be noted that the identified and extracted information can be stored in a knowledge base, and the database can be a relational database (MySQL) or a non-relational database (MongoDB), which can be selected according to application requirements. The knowledge base supports real-time query and can provide timely information support for power grid dispatching operations, including equipment status, historical operation records, operating procedures, etc.

[0138] In the above step S5, the operation command input by the user is received and analyzed, and the operation is performed according to the analysis result, including:

[0139] The input operation command is analyzed and combined with the information in the attribute dictionary to identify possible risk points. If the equipment involved in the operation command is in a faulty state, it is identified as a risk point; if an operation is inconsistent with the historical record, it is identified as a potential erroneous operation; for the identified risk points, the system returns modification suggestions to prompt the user of possible erroneous operations.

[0140] The above is a schematic scheme of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model of this embodiment. It should be noted that the technical solution of the intelligent anti-misoperation system of the power grid based on the BiLSTM-CRF model and the technical solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model belong to the same concept. The details of the technical solution of the intelligent anti-misoperation system of the power grid based on the BiLSTM-CRF model in this embodiment that are not described in detail can be referred to the description of the technical solution of the intelligent anti-misoperation method of the power grid based on the BiLSTM-CRF model.

[0141] The power grid intelligent anti-error operation system based on the BiLSTM-CRF model in this embodiment includes:

[0142] The data set acquisition module is used to collect historical operation records, equipment information and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set;

[0143] A model building module, used for building a first model based on a structured data set and a BiLSTM-CRF model; training the first model to obtain a second model;

[0144] A named entity recognition module, used for performing named entity recognition on the power grid dispatch record to be processed using the second model;

[0145] The attribute dictionary building module is used to extract relevant attribute information from the identified named entities and build an attribute dictionary;

[0146] The analysis operation module is used to receive and analyze the operation command input by the user based on the attribute dictionary, and perform operations according to the analysis results.

[0147] This embodiment further provides a computing device, which is applicable to the case of the intelligent anti-misoperation method for a power grid based on the BiLSTM-CRF model, including:

[0148] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent anti-misoperation method for the power grid based on the BiLSTM-CRF model proposed in the above embodiment.

[0149] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the intelligent anti-misoperation method for a power grid based on a BiLSTM-CRF model as proposed in the above embodiment is implemented.

[0150] The storage medium proposed in this embodiment and the intelligent anti-misoperation method for the power grid based on the BiLSTM-CRF model proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0151] Example 2

[0152] Referring to Table 1, an embodiment of the present invention provides a method for intelligently preventing misoperation of a power grid based on a BiLSTM-CRF model. In order to verify the beneficial effects of the present invention, a scientific demonstration is carried out through simulation experiments.

[0153] 1. Data collection and preprocessing

[0154] After data collection and preprocessing, the final structured data set is in the following format:

[0155] Table 1 Structured dataset

[0156]

[0157] 2. BiLSTM-CRF model construction and training

[0158] 3. Named Entity Recognition

[0159] Input example: Assume that the new grid dispatch record is "On August 2, 2024, substation A underwent equipment maintenance."

[0160] Output example: The model recognizes "August 2, 2024" as a time entity, "Substation A" as an equipment entity, and "Perform equipment maintenance" as an operation instruction.

[0161] 4. Attribute extraction

[0162] (1) Attribute Information Extraction

[0163] Extract relevant attribute information from the identified named entities, such as the status, type, operation time, etc. of the device, and build an attribute dictionary.

[0164] Combining rules and models: Attribute extraction can be performed by defining rules or training additional models.

[0165] (2) Data structuring

[0166] The extracted attribute information is organized into structured data to facilitate subsequent storage and retrieval.

[0167] 5. Knowledge base construction

[0168] (1) Information Storage

[0169] The identified and extracted information is stored in a knowledge base. The database can be a relational database (MySQL) or a non-relational database (MongoDB), depending on the application requirements.

[0170] (2) Real-time support

[0171] The knowledge base supports real-time query and can provide timely information support for power grid dispatching operations, including equipment status, historical operation records, operating procedures, etc.

[0172] 6. Prevent misjudgment

[0173] (1) Input operation command

[0174] Receive operation commands input by users and make intelligent judgments on them.

[0175] (2) Risk point identification

[0176] Analyze the input operation instructions and combine them with the information in the knowledge base to identify possible risk points, such as:

[0177] If the equipment involved in the operation order is in a faulty state, it is identified as a risk point.

[0178] If an operation is inconsistent with the historical records, it is identified as a potential erroneous operation.

[0179] (3) Feedback and modification suggestions

[0180] For the identified risk points, the system returns modification suggestions to prompt users of possible erroneous operations to ensure the safety of scheduling operations.

[0181] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for intelligently preventing misoperation of a power grid based on a BiLSTM-CRF model, characterized in that: include: Collect historical operation records, equipment information, and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set; Based on the structured data set and the BiLSTM-CRF model, a first model is constructed; the first model is trained to obtain a second model; Using the second model to perform named entity recognition on the power grid dispatch records to be processed; Extract relevant attribute information from the identified named entities and construct an attribute dictionary; Based on the attribute dictionary, receive and analyze the operation commands input by the user, and perform operations based on the analysis results.

2. The method for preventing intelligent misoperation of a power grid based on the BiLSTM-CRF model according to claim 1, characterized in that: The data cleaning and preprocessing include: denoising, word segmentation, and labeling; the construction of the structured data set includes ensuring the uniformity of field names and data types.

3. The intelligent anti-misoperation method for power grid based on BiLSTM-CRF model as claimed in claim 2, characterized in that: The constructing of the first model comprises: Input the structured data set into the BiLSTM-CRF model. The data input process includes word embedding and input format. Word embedding: convert each word in the original text data into a vector of fixed dimension; Input format: The text sequence of each operation record is converted into a word embedding matrix. Each row of the matrix represents the vector representation of a word. The entire matrix is ​​used as the input of BiLSTM. The output of the first model includes the feature vector output by the BiLSTM layer, which represents the context information of each word; the CRF layer performs label prediction based on the output of the BiLSTM and outputs the label of each word.

4. The method for preventing intelligent misoperation of a power grid based on a BiLSTM-CRF model according to claim 3, characterized in that: The training of the first model to obtain the second model includes: 80% of the preprocessed structured data set is used as a training set, 10% as a validation set, and 10% as a test set; the outputs of the BiLSTM and CRF layers are trained; the optimization algorithm is used to minimize the loss function and update the parameters of the first model.

5. The method for preventing intelligent misoperation of a power grid based on the BiLSTM-CRF model according to claim 4, characterized in that: The training of the first model to obtain the second model further comprises: Forward propagation: The input sequence is forward propagated through the BiLSTM layer and the CRF layer to calculate the prediction results of the first model; Loss calculation: Calculate the loss between the first model prediction result and the true label; Back propagation: Calculate the gradient through the back propagation algorithm and update the first model parameters; Iterative training: Repeat forward propagation, loss calculation, and back propagation until the first model reaches the preset performance index on the training set and validation set to obtain the second model.

6. The method for preventing intelligent misoperation of a power grid based on the BiLSTM-CRF model according to claim 5, characterized in that: The using the second model to perform named entity recognition on the power grid dispatch record to be processed includes: The power grid dispatch record to be processed is passed into the second model for forward propagation to obtain output, including the label prediction result of each word; The label prediction results are mapped back to the corresponding actual entity names to form the final named entity recognition results and stored in memory or database.

7. The method for preventing intelligent misoperation of a power grid based on the BiLSTM-CRF model according to claim 6, characterized in that: The receiving and analyzing the operation command input by the user, and performing the operation according to the analysis result comprises: The input operation command is analyzed and combined with the information in the attribute dictionary to identify possible risk points. If the equipment involved in the operation command is in a faulty state, it is identified as a risk point; if an operation is inconsistent with the historical record, it is identified as a potential erroneous operation; for the identified risk points, the system returns modification suggestions to prompt the user of possible erroneous operations.

8. A system using the BiLSTM-CRF model-based intelligent anti-misoperation method for a power grid as claimed in any one of claims 1 to 7, characterized in that: include: The data set acquisition module is used to collect historical operation records, equipment information and operating procedures from the power grid dispatching system, perform data cleaning and preprocessing, and obtain a structured data set; A model building module, used for building a first model based on a structured data set and a BiLSTM-CRF model; training the first model to obtain a second model; A named entity recognition module, used for performing named entity recognition on the power grid dispatch record to be processed using the second model; The attribute dictionary building module is used to extract relevant attribute information from the identified named entities and build an attribute dictionary; The analysis operation module is used to receive and analyze the operation command input by the user based on the attribute dictionary, and perform operations according to the analysis results.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the intelligent anti-misoperation method for power grid based on the BiLSTM-CRF model according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the intelligent anti-misoperation method for a power grid based on a BiLSTM-CRF model as described in any one of claims 1 to 7.

Images