Power grid dispatching knowledge extraction method and system

Abstract

The invention discloses a power grid dispatching knowledge extraction method and system, and the method comprises the steps: obtaining a first parameter of a target power dispatching system, and carrying out the first preprocessing of the first parameter, and obtaining a second parameter; establishing a first entity recognition model according to the second parameter; establishing a second joint model according to the first entity recognition model in combination with a preset integration rule; and performing target power grid dispatching knowledge extraction according to the second joint model. By constructing an efficient and intelligent power grid dispatching knowledge extraction method, the intelligent level of a power grid dispatching system is improved, and the safety and stability of power grid operation are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid dispatching knowledge extraction, and in particular to a power grid dispatching knowledge extraction method and system. Background Art

[0002] Grid dispatching is an important part of power system operation, and its core task is to ensure the safe, stable and efficient operation of the power system. In the process of grid dispatching, dispatchers need to rely on a lot of professional knowledge, including grid equipment status, fault handling procedures, dispatching operation procedures, etc. However, traditional grid dispatching knowledge mostly exists in the form of documents, and the efficiency of knowledge acquisition and updating is low, which makes it difficult to meet the real-time and accuracy requirements of modern smart grids.

[0003] The forward maximum matching and fuzzy comparison processes in the prior art may involve a lot of calculations, especially when processing long texts and large-scale data, which may lead to slow processing speeds and affect the overall performance of the system. In addition, fixed multi-label dictionaries and preset algorithms may lack sufficient adaptability and flexibility when faced with changing natural language expressions and complex contexts, and may find it difficult to cope with the diversity and complexity of language use. In some other prior arts, the integration of multiple advanced models and complex training processes may require a lot of computing resources, especially when processing large amounts of data, which may lead to slow processing speeds and high resource consumption. In addition, in the multi-level processing flow from dialogue state tracking to keyword extraction to summary generation, errors in the previous step may be amplified and passed to the next step, affecting the quality of the final summary.

[0004] In the existing technology, power grid dispatching knowledge mostly exists in the form of documents, and the efficiency of knowledge acquisition and updating is low, which makes it difficult to meet the real-time and accuracy requirements of modern smart grids. Summary of the invention

[0005] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the present invention provides a method and system for extracting knowledge from a power grid dispatching, which can solve the problems mentioned in the background technology.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a method for extracting knowledge of power grid dispatching, comprising:

[0010] Acquire a first parameter of a target power dispatching system, and perform a first preprocessing on the first parameter to obtain a second parameter;

[0011] Establishing a first entity recognition model according to the second parameter;

[0012] According to the first entity recognition model, combined with preset integration rules, a second joint model is established;

[0013] The target power grid dispatching knowledge is extracted according to the second joint model.

[0014] As a preferred solution of the power grid dispatching knowledge extraction method of the present invention, the first entity recognition model includes:

[0015] The first entity recognition model is any type of model whose input is the second parameter and whose output is the key entity or key entity related data;

[0016] The first entity recognition model is at least used to capture the second parameter context information and the relationship between the labels in the second parameter.

[0017] As a preferred solution of the power grid dispatching knowledge extraction method of the present invention, wherein: the second joint model includes an entity recognition task and a relationship extraction task;

[0018] The entity recognition task is recognized by the first entity recognition model;

[0019] The relationship extraction task is any operation of obtaining a relationship from the result of recognition by the first entity recognition model.

[0020] As a preferred solution of the power grid dispatching knowledge extraction method of the present invention, the preset integration rules include:

[0021] Construct a shared BERT encoding layer and task-specific layers for entity recognition and relation extraction tasks respectively;

[0022] The entity recognition task and the relation extraction task are integrated into a unified framework. By sharing the encoding layer of BERT, the feedback information of the entity recognition task is used to improve the performance of the relation extraction task.

[0023] As a preferred solution of the power grid dispatching knowledge extraction method described in the present invention, the second parameter is the concatenation of word vector and character embedding.

[0024] As a preferred solution of the power grid dispatching knowledge extraction method of the present invention, the first entity recognition model further includes:

[0025] Select bidirectional long short-term memory network and conditional random field to jointly construct the first entity recognition model;

[0026] In the first entity recognition model architecture, a bidirectional long short-term memory network is used to capture the contextual information of the second parameter, and a conditional random field is used to capture the relationship between labels in the second parameter.

[0027] As a preferred solution of the power grid dispatching knowledge extraction method described in the present invention, the second joint model includes adding a self-attention layer on the basis of the BERT encoding layer.

[0028] In a second aspect, the present invention provides a power grid dispatching knowledge extraction system, comprising:

[0029] A data acquisition and processing module, used for acquiring a first parameter of a target power dispatching system, and performing a first preprocessing on the first parameter to obtain a second parameter;

[0030] A first model building module, used to build a first entity recognition model according to the second parameter;

[0031] A second model building module, used to build a second joint model based on the first entity recognition model and in combination with preset integration rules;

[0032] A knowledge extraction module is used to extract target power grid scheduling knowledge according to the second joint model.

[0033] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0034] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method described above when executed by a processor.

[0035] Compared with the prior art, the invention has the following beneficial effects: the invention proposes a method and system for extracting knowledge from power grid dispatching, obtains a first parameter of a target power dispatching system, performs a first preprocessing on the first parameter, and obtains a second parameter; establishes a first entity recognition model according to the second parameter; establishes a second joint model according to the first entity recognition model, combined with a preset integration rule; and extracts knowledge from the target power grid dispatching according to the second joint model. By constructing a joint model architecture including a BERT encoding layer, a self-attention layer, and a relation classifier, and adopting a joint training method of NER and RE, the accuracy of named entity recognition and relation extraction is greatly improved. The adoption of the iLSTM-CRF entity recognition model enables the model to effectively capture contextual information and label dependencies, and improves the accuracy and robustness of named entity recognition. The introduction of the self-attention mechanism enables the model to focus on important characters and words, enhances the ability to capture fine-grained features, and thus improves the performance of the overall model. Through the joint training method of NER and RE, the model can simultaneously learn the information of entities and relations, and improves the generalization ability and training efficiency of the model. By building an intelligent power grid dispatching knowledge base, we can provide accurate decision support for power grid dispatchers and improve the accuracy and efficiency of power grid dispatching operations. By building an efficient and intelligent power grid dispatching knowledge extraction method, we can improve the intelligence level of the power grid dispatching system and ensure the safety and stability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0037] Figure 1 A method flow chart of a power grid dispatching knowledge extraction method and system provided by an embodiment of the present invention;

[0038] Figure 2 An internal structural diagram of a computer device of a power grid dispatching knowledge extraction method and system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0039] 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.

[0040] Example 1

[0041] Reference Figure 1-Figure 2 , which is the first embodiment of the present invention, and provides a method and system for extracting knowledge from power grid dispatching, including:

[0042] There are some problems in the existing related technologies, such as low accuracy of entity recognition and relationship extraction, insufficient model generalization ability, and lack of effective power grid dispatching knowledge base support.

[0043] This application provides a method that can effectively solve the above-mentioned problems. Next, we will combine multiple embodiments to explain in detail how to implement the power grid dispatching knowledge extraction method;

[0044] Figure 1 A method flow chart of a power grid dispatching knowledge extraction method and system is shown, including:

[0045] S101, obtaining a first parameter of a target power dispatching system, and performing a first preprocessing on the first parameter to obtain a second parameter;

[0046] In an optional embodiment, the first parameter of the target power dispatching system may include text data such as daily operation records, dispatching logs, and accident reports in the power dispatching system;

[0047] In an optional embodiment, the first parameter of the target power dispatching system may also include relevant technical documents, operation manuals, rules and regulations, etc. in the internal database of the power company;

[0048] In an optional embodiment, the first parameter of the target power dispatching system may also include relevant data such as public data such as public power industry reports, research papers, standards and specifications.

[0049] In the embodiment of the present application, in order to design an intelligent power grid dispatching knowledge base, the required data needs to be very comprehensive, so all the above-mentioned relevant data are selected for processing. The present application sacrifices time in exchange for sufficiently high accuracy.

[0050] In an optional embodiment, the first preprocessing may include steps such as text cleaning, word segmentation, part-of-speech tagging, and entity tagging. Text cleaning is used to remove irrelevant characters and noise data, word segmentation divides the text into separate words, part-of-speech tagging assigns a corresponding part-of-speech tag to each word, and entity tagging identifies specific entities in the text, such as device name, time, location, etc.

[0051] In an optional embodiment, the first preprocessing is to improve data quality and ensure the accuracy and efficiency of subsequent model training. The preprocessed data will provide more accurate input for the entity recognition model, thereby improving the performance of the entire system. During the preprocessing, natural language processing techniques can be employed, such as using specific algorithms to identify and process proper nouns, terms, and abbreviations in the text, ensuring that these key information will not be lost or misinterpreted during subsequent processing. In addition, the preprocessing steps can also include removing stop words, correcting spelling mistakes, etc., to further optimize data quality.

[0052] In the embodiment of the present application, the first preprocessing specifically includes:

[0053] Uniformly format the obtained various data into plain text files and remove the format information in the data.

[0054] Furthermore, use the text processing technology of regular expressions to remove the noise in the text, that is, remove redundant spaces, irrelevant symbols, and specific format characters.

[0055] Furthermore, use the natural language processing (NLP) tool NLTK to perform word segmentation and part-of-speech tagging on the text. Word segmentation divides the continuous text into independent words, and part-of-speech tagging tags the part of speech of each word.

[0056] Furthermore, remove common meaningless words such as "of", "is", "in", etc. that are not helpful for knowledge extraction.

[0057] Furthermore, use the NLP tool to perform named entity recognition on the text, label the key entities in the text, such as device names, fault types, operation actions, etc., and then normalize the terms of the labeled entities to ensure the consistency of the same entities in different texts.

[0058] Furthermore, store the processed and labeled data in a structured database.

[0059] It should be noted that the data stored in the structured database is the second parameter.

[0060] In the embodiment of the present application, the second parameter is the concatenation of word vectors and character embeddings.

[0061] It should be noted that obtaining the first parameter of the target power dispatching system and performing a first preprocessing on the first parameter to obtain the second parameter can provide more accurate and rich information for subsequent entity recognition and relationship extraction. The second parameter, as input data, will be used to train and optimize the power grid dispatching knowledge extraction model. In this way, the model can better understand the contextual information of the power dispatching system, thereby achieving more accurate entity recognition and relationship extraction in practical applications. In addition, by building an intelligent power grid dispatching knowledge base, dynamic updating and maintenance of power grid dispatching knowledge can be achieved to ensure the timeliness and accuracy of the knowledge base content. This not only helps to improve the efficiency and safety of power grid dispatching, but also provides more accurate decision-making support for power grid dispatchers, thereby improving the operating efficiency and reliability of the entire power grid system.

[0062] S102, establishing a first entity recognition model according to the second parameter;

[0063] In an embodiment of the present application, the first entity recognition model includes:

[0064] The first entity recognition model is any type of model whose input is the second parameter and whose output is the key entity or key entity related data;

[0065] The first entity recognition model is used to capture at least the context information of the second parameter and the relationship between the labels in the second parameter.

[0066] In an optional embodiment, the key entities include equipment name, fault type, operation action, etc. Text: Transformer T1 failed in substation S1 due to overload. Annotation: {Transformer T1: [equipment]}, {Substation S1: [location]}, {Overload: [fault type]};

[0067] In an optional embodiment, the key entity-related data may include the model of the equipment, the specific time when the fault occurred, the name of the operator, etc. For example, for the text "Transformer T1 failed at substation S1 due to overload", the annotation can be expanded to: {Transformer T1: [equipment, model is T1]}, {Substation S1: [location, specific location is S1]}, {Overload: [fault type, fault occurrence time is specific time]}. Such detailed annotations are helpful for subsequent fault analysis and processing, as well as enriching and updating the knowledge base of power grid dispatching.

[0068] In an optional embodiment, the first entity recognition model can be designed by different methods, for example, natural language processing technology based on deep learning, such as recurrent neural network (RNN), long short-term memory network (LSTM) or Transformer model. These models are able to process sequence data and can capture long-distance dependencies, which is crucial for understanding complex power dispatch texts. For example, the LSTM model can effectively learn time series information in the text through its gating mechanism, while the Transformer model uses a self-attention mechanism to capture the relationship between any two positions in the text, which is particularly effective for entity recognition and relationship extraction.

[0069] In the embodiment of the present application, the first entity recognition model further includes:

[0070] Select bidirectional long short-term memory network and conditional random field to jointly construct the first entity recognition model;

[0071] In the first entity recognition model architecture, a bidirectional long short-term memory network is used to capture the contextual information of the second parameter, and a conditional random field is used to capture the relationship between labels in the second parameter.

[0072] Specifically, a bidirectional long short-term memory network (BiLSTM) and a conditional random field (CRF) are selected to jointly construct a named entity recognition model (ENR). In the BiLSTM-CRF model architecture, BiLSTM is used to capture contextual information, and CRF is used to capture the dependencies between tags.

[0073] In the embodiment of the present application, the model training steps of BiLSTM-CRF are as follows:

[0074] 1) Feature extraction and input representation: Use the pre-trained word vector model GloVe to convert the text into word vectors for initialization, and then form the training set, validation set, and test set; the input representation is the concatenation of word vectors and character embeddings.

[0075] 2) Model initialization

[0076] Use the pre-trained word vectors to initialize the embedding layer; use Gaussian distribution to initialize the parameters of the BiLSTM layer; and then randomly initialize the parameters of the CRF layer, including the parameters of the state score and the transfer score.

[0077] 3) Model training: BiLSTM and CRF layers are trained using alternating optimization, and the loss function is used to update parameters. The loss function is defined as:

[0078]

[0079] Among them, θ is the model parameter, P(y i|x i ; θ) is the given input sequence x i and label sequence y i The conditional probability of .

[0080] 4) Model optimization and verification

[0081] The Adam algorithm is used to optimize the model, the performance of the model is evaluated on the validation set, and the model hyperparameters are adjusted until the best effect is achieved.

[0082] It should be noted that establishing the first entity recognition model according to the second parameter can significantly improve the accuracy of entity recognition. Through the BiLSTM-CRF model, the sequence data in the power dispatch text can be effectively processed, and the complex relationship between entities can be captured. During the model training process, by alternately optimizing the BiLSTM layer and the CRF layer, the model can better understand the contextual information of the text and the dependencies between labels during the learning process. In addition, by evaluating the model performance on the validation set, the hyperparameters of the model can be adjusted in time to ensure that the model can achieve the best recognition effect in practical applications. Ultimately, this optimization and verification process helps to build a more accurate and efficient knowledge extraction model for power grid dispatching, providing more accurate decision support for power grid dispatching, thereby improving the overall operation efficiency and reliability of the power grid system.

[0083] S103, establishing a second joint model based on the first entity recognition model and in combination with preset integration rules;

[0084] In the embodiment of the present application, the second joint model includes an entity recognition task and a relationship extraction task;

[0085] The entity recognition task is identified by the first entity recognition model;

[0086] The relationship extraction task is any operation to obtain the relationship of the result after the first entity recognition model is recognized.

[0087] In the embodiment of the present application, the preset integration rules include:

[0088] Construct a shared BERT encoding layer and task-specific layers for entity recognition and relation extraction tasks respectively;

[0089] The entity recognition task and the relation extraction task are integrated into a unified framework. By sharing the encoding layer of BERT, the feedback information of the entity recognition task is used to improve the performance of the relation extraction task.

[0090] In an embodiment of the present application, the second joint model includes adding a self-attention layer on the basis of the BERT encoding layer.

[0091] In an optional embodiment, the relationship extraction task is any operation of acquiring a relationship from the result of the recognition of the first entity recognition model. For example, the relationship extraction task may include but is not limited to the following steps: first, the key entities in the text are identified using the first entity recognition model; second, the relationship type between the entities is determined based on the semantic relationship and context information between the entities; then, the relationship between the entities is extracted through predefined relationship extraction rules or pattern matching; finally, the extracted relationship information is integrated into the power grid dispatching knowledge base to enrich the content of the knowledge base. For example, for the text "Transformer T1 failed at substation S1 due to overload", after entity recognition, the causal relationship between "transformer T1" and "overload" can be further identified, and this relationship information is stored in the knowledge base. Such relationship extraction not only helps to understand the background of the failure, but also provides data support for future fault prediction and prevention. In this way, the power grid dispatching knowledge base can continuously update and optimize itself, thereby providing more intelligent and automated decision support for power grid dispatching.

[0092] In an optional embodiment, the relationship extraction task can be implemented by designing a relationship extraction model or using a machine learning algorithm. For example, methods such as support vector machines (SVM), random forests, or deep learning models can be used to identify the relationship between entities. These models can predict the type of relationship that may exist between entities based on the results of entity recognition by learning a large amount of labeled data.

[0093] In an optional embodiment, the relationship extraction task can also improve the accuracy of relationship extraction through an ensemble learning method. Ensemble learning improves overall performance by combining the prediction results of multiple models. For example, a voting mechanism can be used, in which each model votes on the prediction results of the relationship between entities, and the final decision is based on the consensus of the majority of models. This method can reduce the possible deviations of a single model, thereby improving the robustness and accuracy of relationship extraction. In addition, ensemble learning can further optimize performance by adjusting the weights of different models to ensure that the relationship information stored in the power grid dispatching knowledge base is more reliable and useful. Through these optimization and verification steps, the power grid dispatching knowledge extraction system can more accurately identify and extract key information, providing more intelligent and efficient decision support for power grid dispatching.

[0094] In the embodiment of the present application, a BERT multi-task relation extraction learning model is established as the second joint model;

[0095] In an embodiment of the present application, the second joint model structure includes a shared BERT encoding layer, and task-specific layers for NER and RE, respectively.

[0096] The named entity recognition (NER) and relation extraction (RE) tasks are integrated into a unified framework. By sharing the BERT encoding layer, the feedback information of the NER task is used to improve the performance of the RE task. The model structure includes a shared BERT encoding layer and task-specific layers for NER and RE respectively. Through this shared architecture, the model can learn the features of entity recognition and relation extraction at the same time, thereby enhancing the overall performance of the model. During the training process of the model, the self-attention mechanism is introduced to dynamically assign different attention weights to each entity pair, enhancing the model's understanding and processing capabilities of contextual information.

[0097] In an embodiment of the present application, all possible entity pairs are selected from the identified entities as candidates for relationship extraction, and then input features are constructed for each entity pair, including context information and position encoding of the entity.

[0098] 1) Entity Recognition

[0099] Use the BiLSTM-CRF entity recognition model to identify all entities and classify the identified entities into categories, such as equipment, location, time, etc.

[0100] 2) Intelligent entity pair selection

[0101] A hybrid strategy based on rules and machine learning is used for entity pair selection, and entity pairs are screened according to predefined rules. Only entities in adjacent sentences can have a relationship. A classification model is then trained using the classic machine learning method of Random Forest to predict whether there is a relationship between entity pairs based on the contextual features of the entity pairs.

[0102] 3) Context feature construction

[0103] Context features are constructed for each entity pair, including the context information, location encoding and category information of the entity. For example, entity 1: power station A; entity 2: transformer B; context information: power station A is connected to transformer B. Location encoding: the location of power station A, the location of transformer B.

[0104] In the embodiment of the present application, an input sequence is constructed to concatenate entity pairs and their contexts. For example: "[CLS] Power station A [SEP] Power station A is connected to transformer B [SEP] Transformer B [SEP]". Then, special tags [CLS] and [SEP] are added to the input sequence for the BERT model to process.

[0105] Furthermore, the input sequence is fed into the BERT model to generate the context representation h ij

[0106]

[0107] Among them, h i and h j denote the context representation of entity i and entity j respectively, represents element-by-element multiplication, - represents element-by-element subtraction, and [·;·] represents vector concatenation.

[0108] Furthermore, based on the BERT encoding layer, a self-attention layer is added to further process the context vector h ij , improve the accuracy of relation extraction. The combination represents h ij Input the self-attention layer to generate weighted context representation. The attention weight is expressed as:

[0109]

[0110] Among them, e ij represents the attention score of entity pair i and j, and n represents the total number of entity pairs. It is expressed as:

[0111]

[0112] Furthermore, the present application constructs a Softmax classifier as an implementation model for the relationship extraction task, classifies the relationship of the weighted context representation, and then assigns a relationship label to each entity pair to indicate the type of relationship between the entity pairs.

[0113] In an embodiment of the present application, a training set including entity labels and relationship labels is constructed, and then training is performed on the constructed model architecture including a BERT encoding layer, a self-attention layer, and a relationship classifier.

[0114] In an optional embodiment, assume that the training data set contains N samples, each sample contains T words. For each word t, the true label is y t , the label probability distribution predicted by the model is P(y t |x t ), where x t represents the input feature of the tth word, then the NER loss function is specifically defined as,

[0115]

[0116] In an optional embodiment, assume that the training data set contains M entity pairs, the true relationship label of each entity pair is r, and the probability distribution of the relationship label predicted by the model is in represents the contextual representation of the entity pair (i, j), then the RE loss function is specifically defined as,

[0117]

[0118] Finally, the joint loss function of NER and RE tasks can be set:

[0119]

[0120] in, represents the loss function of the NER task, The loss function λ of the RE task is represented as a weight parameter.

[0121] It should be noted that the model training process includes joint training of the two tasks of NER and RE. Through joint training, the model can learn the features of entity recognition and relation extraction at the same time, thereby enhancing the overall performance of the model.

[0122] S104: Extract target power grid dispatching knowledge according to the second joint model.

[0123] In an optional embodiment, new information is collected from different sources, including new scheduling procedures, troubleshooting cases, equipment maintenance records, etc. The collected information is then verified to ensure its accuracy and reliability. The constructed named entity recognition and relationship extraction model is used to extract key knowledge points from the new information and update the entities and relationships in the knowledge base. The new knowledge is integrated with the existing knowledge base to eliminate redundancy and conflict and ensure the consistency and integrity of the knowledge base. Evaluation indicators for knowledge updates are set, such as accuracy, completeness, and practicality of knowledge.

[0124] In an optional embodiment, target power grid dispatching knowledge is extracted according to the second joint model, and an intelligent power grid dispatching knowledge base is constructed. By constructing an intelligent power grid dispatching knowledge base, accurate decision support can be provided to power grid dispatchers, and the accuracy and efficiency of power grid dispatching operations can be improved. By constructing an efficient and intelligent power grid dispatching knowledge extraction method, the intelligence level of the power grid dispatching system is improved, and the safety and stability of power grid operation are ensured.

[0125] In summary, the present invention proposes a method for extracting knowledge from power grid dispatching, obtaining a first parameter of a target power dispatching system, and performing a first preprocessing on the first parameter to obtain a second parameter; establishing a first entity recognition model according to the second parameter; establishing a second joint model according to the first entity recognition model, combined with a preset integration rule; and extracting knowledge from the target power grid dispatching according to the second joint model. By constructing a joint model architecture including a BERT encoding layer, a self-attention layer, and a relationship classifier, and adopting a joint training method of NER and RE, the accuracy of named entity recognition and relationship extraction is greatly improved. The adoption of the iLSTM-CRF entity recognition model enables the model to effectively capture contextual information and label dependencies, thereby improving the accuracy and robustness of named entity recognition. The introduction of the self-attention mechanism enables the model to focus on important characters and words, enhances the ability to capture fine-grained features, and thus improves the performance of the overall model. Through the joint training method of NER and RE, the model can simultaneously learn information about entities and relationships, thereby improving the generalization ability and training efficiency of the model. By constructing an intelligent power grid dispatching knowledge base, it is possible to provide accurate decision support for power grid dispatchers and improve the accuracy and efficiency of power grid dispatching operations. By building an efficient and intelligent power grid dispatching knowledge extraction method, the intelligence level of the power grid dispatching system can be improved to ensure the safety and stability of power grid operation.

[0126] This embodiment also provides a power grid dispatching knowledge extraction system, including:

[0127] A data acquisition and processing module, used for acquiring a first parameter of a target power dispatching system, and performing a first preprocessing on the first parameter to obtain a second parameter;

[0128] A first model building module, used to build a first entity recognition model according to the second parameter;

[0129] A second model building module, used to build a second joint model based on the first entity recognition model and in combination with preset integration rules;

[0130] The knowledge extraction module is used to extract the target power grid scheduling knowledge according to the second joint model.

[0131] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0132] This embodiment also provides a computer device, which may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 2As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for extracting knowledge from power grid dispatching is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0133] This embodiment further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0134] Acquire a first parameter of a target power dispatching system, and perform a first preprocessing on the first parameter to obtain a second parameter;

[0135] Establishing a first entity recognition model according to the second parameter;

[0136] According to the first entity recognition model, combined with the preset integration rules, a second joint model is established;

[0137] The target power grid dispatching knowledge is extracted according to the second joint model.

[0138] 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.

[0139] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can 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.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0140] 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.

[0141] 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.

[0142] 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 A step that specifies a function in one or more boxes.

[0143] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0144] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for extracting knowledge from power grid dispatching, characterized in that: include: Acquire a first parameter of a target power dispatching system, and perform a first preprocessing on the first parameter to obtain a second parameter; Establishing a first entity recognition model according to the second parameter; According to the first entity recognition model, combined with preset integration rules, a second joint model is established; The target power grid dispatching knowledge is extracted according to the second joint model.

2. The power grid dispatching knowledge extraction method according to claim 1, characterized in that: The first entity recognition model includes: The first entity recognition model is any type of model whose input is the second parameter and whose output is the key entity or key entity related data; The first entity recognition model is at least used to capture the second parameter context information and the relationship between the labels in the second parameter.

3. The power grid dispatching knowledge extraction method according to claim 2, characterized in that: The second joint model includes an entity recognition task and a relationship extraction task; The entity recognition task is recognized by the first entity recognition model; The relationship extraction task is any operation of obtaining a relationship from the result of recognition by the first entity recognition model.

4. The power grid dispatching knowledge extraction method according to claim 3, characterized in that: The preset integration rules include: Construct a shared BERT encoding layer and task-specific layers for entity recognition and relation extraction tasks respectively; The entity recognition task and the relation extraction task are integrated into a unified framework. By sharing the encoding layer of BERT, the feedback information of the entity recognition task is used to improve the performance of the relation extraction task.

5. The power grid dispatching knowledge extraction method according to claim 4, characterized in that: The second parameter is the concatenation of the word vector and the character embedding.

6. The method for extracting knowledge of power grid dispatching according to claim 5, characterized in that: The first entity recognition model also includes: Select bidirectional long short-term memory network and conditional random field to jointly construct the first entity recognition model; In the first entity recognition model architecture, a bidirectional long short-term memory network is used to capture the contextual information of the second parameter, and a conditional random field is used to capture the relationship between labels in the second parameter.

7. The method for extracting knowledge from power grid dispatching according to claim 6, characterized in that: The second joint model includes adding a self-attention layer on the basis of the BERT encoding layer.

8. A power grid dispatching knowledge extraction system, characterized in that: include: A data acquisition and processing module, used for acquiring a first parameter of a target power dispatching system, and performing a first preprocessing on the first parameter to obtain a second parameter; A first model building module, used to build a first entity recognition model according to the second parameter; A second model building module, used to build a second joint model based on the first entity recognition model and in combination with preset integration rules; A knowledge extraction module is used to extract target power grid scheduling knowledge according to the second joint model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images