Key process identification method, device and equipment of power text and storage medium

Abstract

The invention discloses a key process identification method and device for a power text, equipment and a storage medium, and the method comprises the steps: obtaining the power text of a to-be-identified key process, carrying out the semantic representation of the power text, generating a semantic vector, analyzing the depth meaning of a sentence through the context semantics of each sentence of the power text, and obtaining the key process of the power text; and performing semantic enhancement on the semantic vector to obtain an enhanced semantic vector, and identifying the key process of the power text based on the enhanced semantic vector. Therefore, the context semantic analysis is used for mining the depth meaning of the sentence, the deep semantics of the power operation word, the sequential logic of the operation steps and the entity equipment association information can be accurately captured, the semantic vector representation accuracy is remarkably improved, the key process identification accuracy is greatly improved, the method is adaptive to complex scenes such as polysemy and synonyms in the power text, and the method is suitable for popularization and application. And the operation specification and efficiency are ensured.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of power text recognition, in particular to a power text key procedure recognition method and device, equipment and a storage medium. BACKGROUND

[0002] With the continuous expansion of the power grid scale and the rapid improvement of the intelligent level, the complexity and safety control requirements of power grid field operation have significantly improved. Accurate identification of key procedures is the core of ensuring safety, standardizing processes and improving efficiency. In the existing key procedure identification method, manual annotation and traditional machine learning are used to manually annotate key procedures, extract bag-of-words or TF-IDF shallow features, and then use a support vector machine model for training and identification. This method has weak semantic understanding ability and cannot handle synonyms and polysemous words.

[0003] Therefore, experts have made targeted knowledge coverage of words with similar semantics, so that the model can understand synonyms and polysemous words. However, in power documents, there are not only one-word multiple-word expressions, but also one-word multiple-meaning situations in different scenarios. It can be seen that only isolated processing of single sentences or phrases cannot accurately understand the semantics of power documents, resulting in low accuracy of key procedure identification.

[0004] Therefore, how to enhance the semantic vector to accurately mine the deep semantic association of the key procedure and further improve the accuracy of the key procedure identification of the power grid field operation is a problem that needs attention. SUMMARY

[0005] In view of the above problems, the application provides a power text key procedure recognition method, device, equipment and storage medium to accurately mine the deep semantic association of the key procedure and further improve the accuracy of the key procedure identification of the power grid field operation.

[0006] In order to achieve the above purpose, the specific scheme is as follows:

[0007] A power text key procedure recognition method comprises the following steps:

[0008] Obtaining a power text to be identified for a key procedure;

[0009] Performing semantic representation on the power text to generate a semantic vector;

[0010] Performing semantic enhancement on the semantic vector by analyzing the deep meaning of each sentence in the power text based on the context semantic of the sentence to obtain an enhanced semantic vector;

[0011] Identifying the key procedure of the power text based on the enhanced semantic vector.

[0012] Optionally, the semantic enhancement on the semantic vector according to the deep meaning of the power terms in each sentence of the power text comprises:

[0013] According to the context of each sentence of the power text, the sentence is resolved to obtain an updated sentence.

[0014] According to the chapters of the power text, the deep meaning of the power terms in each sentence of the power text is determined.

[0015] For each target sentence of the power text, the mood verb in the target sentence is determined, and the importance weight of the operation step described by the target sentence is determined according to the coerciveness of the mood verb and the scene described by the target sentence.

[0016] For each target sentence of the power text, the entity associated with the target sentence is determined, and if the entity contains a high-risk device, the critical process determination weight of the operation step of the target sentence is increased.

[0017] According to the association prompt word, the association relationship between each operation step is determined, and the timing logic relationship feature is extracted based on the association relationship between each operation step.

[0018] The deep meaning of the power terms in each sentence, the importance weight of each operation step, the critical process determination weight of each operation step, and the timing logic relationship feature are fused with the semantic vector to obtain an enhanced semantic vector.

[0019] Optionally, according to the context of each sentence of the power text, the sentence is resolved to obtain an updated sentence, comprising:

[0020] For each target sentence of the power text, a sliding window of each target sentence is created, which contains a plurality of sentences before the target sentence, a plurality of sentences after the target sentence, and the target sentence.

[0021] Each pronoun of the target sentence is determined.

[0022] According to each sentence in the sliding window of the target sentence, the reference object of each pronoun of the target sentence is determined, and the reference object is replaced with the corresponding pronoun to obtain an updated target sentence.

[0023] Optionally, according to the chapters of the power text, the deep meaning of the power terms in each sentence of the power text is determined, comprising:

[0024] identifying chapter keywords of the power text, and dividing chapters according to the chapter keywords;

[0025] For each target sentence of the power text, determining target content associated with the power term in the target sentence according to semantics of the target sentence within its chapter and semantic dependency of the target sentence in other chapters;

[0026] Based on the power term in the target sentence and its associated target content, determining the deep meaning of the power term in the target sentence.

[0027] Optionally, determining target content associated with the power term in the target sentence according to semantics of the target sentence within its chapter and semantic dependency of the target sentence in other chapters comprises:

[0028] If the target sentence contains a first operation step, determining action information within the chapter to which the target sentence belongs, and determining each second operation step of other chapters that does not contain the target sentence;

[0029] From the action information, filtering target action information associated with the first operation step;

[0030] From each of the second operation steps, determining a target second operation step that is executed before and / or after the first operation step;

[0031] According to the target action information and the target second operation step, determining target content associated with the power term in the target sentence.

[0032] Optionally, before the power text is semantically represented to generate a semantic vector, the method further comprises:

[0033] Pretreating the power text to obtain a pretreated power text.

[0034] Optionally, after the key process of the power text is identified based on the enhanced semantic vector, the method further comprises:

[0035] Outputting the key process in the form of text or chart.

[0036] A key process identification device of a power text, comprising:

[0037] A power text acquisition unit configured to acquire a power text to be identified for a key process;

[0038] A semantic vector generation unit configured to semantically represent the power text to generate a semantic vector;

[0039] a semantic enhancement unit configured to enhance semantics of the semantic vector by parsing deep meanings of each sentence of the power text in the context of the semantics, to obtain an enhanced semantic vector;

[0040] a recognition unit configured to recognize a key procedure of the power text based on the enhanced semantic vector.

[0041] Optionally, the semantic enhancement unit comprises:

[0042] a coreference resolution unit configured to perform coreference resolution on each sentence of the power text according to the context of the sentence, to obtain an updated sentence;

[0043] a deep meaning mining unit configured to determine deep meanings of power terms of each sentence of the power text according to chapters of the power text;

[0044] an operation step weight determination unit configured to determine an emotive verb in each target sentence of the power text, and determine an importance weight of an operation step described by the target sentence according to the coerciveness of the emotive verb and a scene described by the target sentence;

[0045] a key procedure weight improvement unit configured to determine an entity associated with each target sentence of the power text, and improve a key procedure determination weight of an operation step of the target sentence if the entity comprises a high-risk device;

[0046] a temporal logic relationship feature extraction unit configured to determine an association relationship between each operation step according to an association cue, and extract a temporal logic relationship feature based on the association relationship between each operation step;

[0047] a weight feature fusion unit configured to fuse the deep meanings of power terms of each sentence, the importance weights of each operation step, the key procedure determination weights of each operation step, the temporal logic relationship feature, and the semantic vector, to obtain an enhanced semantic vector.

[0048] Optionally, the coreference resolution unit comprises:

[0049] a sliding window creation unit configured to create a sliding window of each target sentence of the power text, the sliding window comprising a plurality of sentences before the target sentence, a plurality of sentences after the target sentence, and the target sentence;

[0050] a pronoun determination unit configured to determine each pronoun of the target sentence;

[0051] The sentence updating unit is configured to determine the referent of each pronoun in the target sentence according to each sentence within the sliding window of the target sentence, and replace the pronoun with the referent to obtain an updated target sentence.

[0052] Optionally, the term deep meaning mining unit comprises:

[0053] The chapter dividing unit is configured to identify chapter keywords of the power text, and divide chapters according to the chapter keywords.

[0054] The associated content determining unit is configured to determine, for each target sentence of the power text, target content associated with the power term in the target sentence according to the semantics of the target sentence in the chapter where the target sentence is located and the semantic dependency of the target sentence in other chapters.

[0055] The deep meaning determining unit is configured to determine the deep meaning of the power term in the target sentence based on the power term in the target sentence and the target content associated with the power term.

[0056] Optionally, the associated content determining unit comprises:

[0057] The action step determining unit is configured to determine action information in the chapter where the target sentence is located and determine each second operation step of other chapters not containing the target sentence if the target sentence contains a first operation step.

[0058] The target action screening unit is configured to screen target action information associated with the first operation step from the action information.

[0059] The target operation step determining unit is configured to determine a target second operation step executed before and / or after the first operation step from each second operation step.

[0060] The target content determining unit is configured to determine target content associated with the power term in the target sentence according to the target action information and the target second operation step.

[0061] Optionally, the device further comprises:

[0062] The preprocessing unit is configured to preprocess the power text to obtain preprocessed power text before performing semantic representation on the power text to generate a semantic vector.

[0063] Optionally, the device further comprises:

[0064] The output unit is configured to output the key process in the form of text or chart after identifying the key process of the power text based on the enhanced semantic vector.

[0065] A key process identification device for electrical text, including a memory and a processor;

[0066] The memory is used to store programs;

[0067] The processor is used to execute the program to implement the various steps of the key process identification method for power text as described above.

[0068] A storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the various steps of the key process identification method for power text as described above.

[0069] By employing the aforementioned technical solution, this application acquires the power text containing the key processes to be identified, performs semantic representation on the power text, generates semantic vectors, and enhances the semantic vectors by analyzing the deep meaning of each sentence in the power text using contextual semantic analysis, resulting in enhanced semantic vectors. Based on these enhanced semantic vectors, the key processes in the power text are identified. Therefore, contextual semantic analysis, by mining the deep meaning of sentences, can accurately capture the deep semantics of power terminology, the temporal logic of operational steps, and the associated information of physical equipment, significantly improving the accuracy of semantic vector representation, greatly increasing the accuracy of key process identification, and adapting to complex scenarios such as polysemy and synonyms in power texts, ensuring standardized and efficient operations. Attached Figure Description

[0070] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0071] Figure 1 A flowchart illustrating the key process identification of power text provided in an embodiment of this application;

[0072] Figure 2 This is a schematic diagram of a process for implementing semantic enhancement provided in an embodiment of this application;

[0073] Figure 3 A schematic diagram of a device structure for identifying key processes in electrical text, provided in an embodiment of this application;

[0074] Figure 4 This is a schematic diagram of the structure of a device for identifying key processes in electrical text, provided as an embodiment of this application. Detailed Implementation

[0075] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0076] The proposed solution can be implemented using a terminal with data processing capabilities, such as a computer, cloud platform, or server. This terminal can contain a large DS language model.

[0077] Before training the DS large language model, data preparation is required. Specifically, historical textual data from power grid field operations can be collected, including work instructions, work orders, and operation records. Further, the data is manually labeled to identify key procedures, forming a training dataset. Finally, a power grid domain knowledge graph is constructed, which can include knowledge in areas such as equipment type, work steps, and safety regulations.

[0078] The training process of the DS large language model is as follows: pre-processed text data is input into the DS large language model for pre-training to enable it to have basic language understanding capabilities; the large language model is fine-tuned using the labeled training dataset DS to adapt it to the task of identifying key processes in power grid operations; during the fine-tuning process, a domain knowledge graph is introduced, and domain knowledge is integrated into the semantic representation of the model through knowledge fusion technology.

[0079] The trained DS large language model can be used to perform key process recognition schemes for power text.

[0080] Next, combined Figure 1 The key process identification method for power text in this application may include the following steps:

[0081] Step S110: Obtain the power text of the key process to be identified.

[0082] Specifically, power text can be core text data that carries relevant information about the operation in the power grid field operation scenario and from which key processes need to be identified. It includes on-site operation steps, equipment information (such as equipment number and voltage level), safety specifications, process association logic, etc., and contains power professional terms such as "circuit breaker" and "grounding switch", all of which come from the actual power grid field operation scenario.

[0083] Step S120: Perform semantic representation on the power text to generate semantic vectors.

[0084] Specifically, a basic semantic vector can be generated after semantic representation using the DS large language model. Its generation relies on the natural language understanding capabilities of the DS large language model, which can initially capture the literal information and basic semantics of power texts, such as the basic meanings of power terminology like circuit breakers and grounding switches, as well as simple semantic relationships between operational steps. However, this semantic vector may not resolve the semantic differences between sentences in different chapters, the temporal logic between operational steps, or connect to entity information in fields such as high-risk equipment. It struggles to handle complex semantic scenarios in power texts, such as polysemy and synonyms, resulting in insufficient depth and accuracy of semantic representation. It only serves as a basic vector for subsequent semantic enhancement and requires further supplementation with deeper semantic information to support accurate identification of key processes.

[0085] Step S130: By analyzing the contextual semantics of each sentence in the power text to understand the deep meaning of the sentences, semantic vectors are semantically enhanced to obtain enhanced semantic vectors.

[0086] Specifically, we can first create a sliding window containing the preceding and following context for each sentence in the power text to perform referential resolution, replacing pronouns with specific equipment names or operation content to eliminate semantic ambiguity; then, we can identify keywords in the text chapters to divide them into chapters, and combine the semantics of sentences in their respective chapters and the semantic dependencies across chapters to mine the deep meaning of power terminology; at the same time, we can determine the importance weight of operation steps based on the modal verbs in the text and the work scenario, adjust the weight of key processes by associating them with high-risk equipment, and extract the temporal logical relationship features between operation steps. Finally, we can fuse these deep semantic information with the original semantic vector to complete semantic enhancement and obtain an enhanced semantic vector.

[0087] Understandably, by enhancing semantic vectors, it is possible to accurately capture the differences between polysemous words and synonyms in power texts, as well as the logical connections between operational steps. This makes semantic vectors more consistent with the actual scenarios of power grid operations, reduces misidentification caused by semantic ambiguity, significantly improves the accuracy of semantic recognition, and adapts to the complex semantic environment of power grid field operations.

[0088] Step S140: Identify key processes in power text based on enhanced semantic vectors.

[0089] Specifically, key processes in power-related texts can be identified using classification algorithms based on enhanced semantic vectors.

[0090] Understandably, by identifying key processes in power industry text based on enhanced semantic vectors, the accuracy of key process identification can be improved by accurately utilizing information such as contextual semantics, deep meaning of power industry terms, operational step weights, and timing logic fused in the enhanced semantic vectors. This avoids misidentifying synonymous processes as different processes. Furthermore, it eliminates the need for extensive manual annotation and rule maintenance, automatically adapts to changes in power grid operation scenarios and the introduction of new processes, reduces labor costs, and provides real-time and reliable identification results for on-site power grid operation safety management and process standardization, offering core technical support for intelligent operation management.

[0091] The key process identification method for power text provided in this embodiment obtains the power text containing the key process to be identified, performs semantic representation on the power text to generate a semantic vector, and enhances the semantic vector by analyzing the deep meaning of each sentence in the power text using contextual semantic analysis, resulting in an enhanced semantic vector. Based on the enhanced semantic vector, the key process of the power text is identified. Therefore, contextual semantic analysis can accurately capture the deep semantics of power terminology, the temporal logic of operational steps, and the associated information of physical equipment, significantly improving the accuracy of semantic vector representation, greatly increasing the accuracy of key process identification, and adapting to complex scenarios such as polysemy and synonyms in power text, ensuring standardized and efficient operations.

[0092] In some embodiments of this application, the process of step S130 above, which involves semantically enhancing the semantic vector by parsing the deep meaning of sentences based on the contextual semantics of each sentence in the power text, is described. (Refer to...) Figure 2 The process may include:

[0093] Step S210: Based on the context of each sentence in the power text, perform referential resolution on that sentence to obtain the updated sentence.

[0094] Specifically, the process of resolving the referential insufficiency of each sentence in the power text based on its context to obtain the updated sentence may include:

[0095] S211. For each target sentence in the power text, create a sliding window for each target sentence.

[0096] The sliding window can contain several sentences preceding the target sentence, several sentences following the target sentence, and the target sentence itself.

[0097] Specifically, the sliding window can be centered on the target sentence, with the first three sentences and the last three sentences of the target sentence also included in the sliding window.

[0098] S212. Identify each pronoun in the target sentence.

[0099] Specifically, pronouns can represent referential words in a sentence where the semantics may be ambiguous, such as "it", "other", "its", "the device", etc.

[0100] S213. Determine the referent of each pronoun in the target sentence based on each sentence within the sliding window of the target sentence, and replace the pronoun with its corresponding referent to obtain the updated target sentence.

[0101] For example, if it is recognized that "its" in the target sentence "disconnect it" refers to the "10kV #1 main transformer high-voltage side circuit breaker" mentioned earlier in the power text, then "10kV #1 main transformer high-voltage side circuit breaker" can be used to replace "its".

[0102] It can be understood that the anaphora resolution operation can eliminate pronouns (such as "its", "the device") in the power text where the semantics may be ambiguous, replacing the vague referents with specific device names (such as "10kV #1 main transformer high-voltage side circuit breaker") or operation steps, making the sentence semantics more explicit. This can avoid the loss of key information caused by isolated understanding of sentences, ensure the accuracy of subsequent in-depth meaning mining of power technical terms, weight determination of operation steps, and chronological logic analysis, provide an accurate text basis for semantic vector enhancement, and thus improve the reliability of subsequent key process identification.

[0103] Step S220. Determine the in-depth meaning of each power technical term in each sentence of the power text according to each chapter of the power text.

[0104] Specifically, the process of determining the in-depth meaning of each power technical term in each sentence of the power text according to each chapter of the power text may include:

[0105] S221. Identify the chapter keywords of the power text and divide the chapters according to the chapter keywords.

[0106] It can be understood that different chapters of power documents are mostly divided into semantic parts by descriptions such as "operation tasks", "safety measures", "operation steps", "hazard point analysis", etc. Therefore, when these chapter keywords are identified, chapter division can be carried out. Each chapter is not necessarily just a paragraph division, but can be a division with different semantics.

[0107] S222. For each target sentence of the power text, determine the target content associated with the power technical term in the target sentence according to the semantics of the target sentence within its chapter and the semantic dependency of the target sentence in other chapters.

[0108] Specifically, for each target sentence in the power text, the first step is to identify the chapter category to which the target sentence belongs by recognizing chapter keywords. Then, the basic context of the target sentence is located by combining the core semantics of that chapter (e.g., if the target sentence is in the safety measures chapter, its semantics tend to be related to operational risk prevention and control; if it is in the operation steps chapter, it focuses on the sequence of process execution). Subsequently, the attention mechanism of the DS large language model or graph neural network can be used to mine the semantic dependencies between the target sentence and other chapters: if the target sentence contains the operation step of closing the grounding switch, it will be associated with the requirement to install the grounding wire in the safety measures chapter and the warnings to prevent electric shock in the hazard analysis chapter, thereby determining the action information associated with the grounding switch and the operation steps connected before and after it, such as the preceding voltage testing step and the subsequent grounding wire hanging step.

[0109] Taking the example of operational steps related to electricity contained in the target sentence, this paper introduces the process of determining the target content associated with the electricity terms in the target sentence based on the semantics of the target sentence within its own chapter and the semantic dependencies of the target sentence in other chapters. Specifically, this process may include:

[0110] S2221. If the target sentence contains the first operation step, then determine the action information within the chapter to which the target sentence belongs, and determine the second operation steps for each of the other chapters that do not contain the target sentence.

[0111] Understandably, since operational steps involve actions, it is necessary to analyze the action information within this chapter to uncover the deeper meaning of the first operational step in the target sentence. Furthermore, there may be temporal or logical connections between operational steps, often appearing in other chapters; therefore, it is also necessary to analyze them in conjunction with the operational steps in other chapters.

[0112] S2222: Filter the target action information associated with the first operation step from the action information.

[0113] It is understandable that a chapter may contain a large amount of action information, including action information related to the first operation step and action information unrelated to the first operation step. Therefore, it is necessary to filter the related action information.

[0114] S2223. From each of the second operation steps, determine the target second operation step that was executed before and / or after the first operation step.

[0115] It is understandable that the operation steps most strongly associated with the operation steps are the steps that are executed before it and the steps that are executed after it. For example, if the order of operation steps in the power document is specified as "SA→SB→SC", then if the target sentence contains the first operation step as SB, and steps SA and SC exist in other chapters respectively, then SA and SC can be identified as the target second operation steps.

[0116] S2224. Based on the target action information and the target second operation step, determine the target content associated with the electrical terminology in the target sentence.

[0117] Specifically, S2221-S2223 have roughly located the deep meaning of electrical terms, so the target content associated with electrical terms in the target sentence can be analyzed and extracted from the target action information and the target second operation step.

[0118] S223. Based on the electrical terms in the target sentence and their associated target content, determine the in-depth meaning of the electrical terms in the target sentence.

[0119] Specifically, based on the electrical terminology in the target sentence and its associated target content, we can first extract the basic literal meaning of the terminology, and then conduct in-depth analysis in conjunction with the associated target content. For example, if the electrical term is "voltage testing," the associated target content includes the context of its "safety measures" chapter, the preceding "disconnecting the circuit breaker" step, and the subsequent "installing the grounding wire" requirement. We can first clarify the basic operational definition of "voltage testing," and then determine its core function as risk prevention based on the chapter attributes. Based on the preceding and following steps, we can clarify that its operation timing is after power outage and before grounding. At the same time, we can further supplement the operational specifications and safety boundaries by associating it with the "preventing live voltage testing" prompt in the "hazard analysis" chapter. By integrating the basic meaning of the terminology with the associated chapter context, process logic, safety requirements, and other target content, we can break through the limitations of simply understanding the terminology literally and ultimately determine the deeper meaning of "voltage testing" in power grid field operations: it is a safety verification procedure performed after power outage and before grounding to avoid the risk of electric shock, and it must be strictly followed according to regulations.

[0120] Step S230: For each target sentence in the power text, identify the modal verbs in the target sentence, and determine the importance weight of the operation steps described in the target sentence based on the mandatory nature of the modal verbs and the scenario described in the target sentence.

[0121] Specifically, for each target sentence in the power industry text, part-of-speech tagging (POS) technology can be used to accurately extract modal verbs such as "must," "strictly prohibited," "should," and "may" from the sentence. Then, the mandatory nature of these modal verbs is graded based on the characteristics of the power grid operation scenario. For example, "strictly prohibited" and "must" are classified as the highest mandatory level, "should" as the medium mandatory level, and "may" as the low mandatory level. Further calibration is performed based on the scenario described by the target sentence. For instance, if a sentence containing the highest mandatory level modal verb is in the "Safety Measures" section, describing an operation scenario involving key equipment such as main transformers and circuit breakers, it indicates that the operation step is directly related to operational safety, and its importance weight should be increased to the highest level. If a sentence containing the low mandatory level modal verb is in the "Operation Records" section, describing an auxiliary recording scenario, its importance weight is correspondingly reduced. Through this dual determination combining the mandatory level of modal verbs and scenario attributes, the importance weight of the operation step described in each target sentence is ultimately accurately determined, providing a crucial weighting basis for subsequent semantic vector enhancement.

[0122] Step S240: For each target sentence in the power text, determine the entity associated with the target sentence. If the entity contains high-risk equipment, increase the weight of the key process determination in the operation steps of the target sentence.

[0123] Specifically, for each target sentence in the power industry text, we can first rely on the power grid domain knowledge graph to extract related entities such as equipment, tools, and work objects involved in the sentence through entity recognition technology. For example, from "checking the status of the 10kV #1 main transformer high-voltage side 101 circuit breaker", we can identify entities such as "10kV #1 main transformer" and "101 circuit breaker". By comparing these entities with the pre-set high-risk equipment list in the knowledge graph (such as main transformers, high-voltage circuit breakers, grounding switches, and other equipment that directly affects the safe and stable operation of the power grid), we can determine whether the extracted entities belong to the high-risk category. If the related entities of the target sentence include high-risk equipment such as "10kV #1 main transformer", it means that the operation steps described in the sentence are directly related to work safety and power grid operation risks. Based on the risk level matching mechanism, we need to increase the weight of its key procedures compared to the operation steps of related ordinary equipment, such as by 30%-50%, to highlight the importance of high-risk equipment-related operations in the identification of key procedures.

[0124] Step S250: Determine the relationship between each operation step based on the associated prompt words, and extract the temporal logical relationship features based on the relationship between each operation step.

[0125] Specifically, related prompts can include "first step," "second step," "after," "first," and "then." For example, the "first / then" in "first disconnect the circuit breaker, then test for voltage" clearly indicates the sequential relationship between disconnecting the circuit breaker and testing for voltage; in "only after confirming that the voltage test is qualified can the grounding wire be installed," it is clear that "qualified voltage test" is a prerequisite for installing the grounding wire. Based on these relationships, sequence labeling or logic graph construction methods are used to extract temporal logical relationship features such as sequential execution, conditional triggering, and causal deduction, transforming the implicit logic between operation steps into quantifiable feature data.

[0126] Step S260: The deep meaning of electrical terminology in each sentence, the importance weight of each operation step, the key process weight of each operation step, and the temporal logical relationship features are fused with the semantic vector to obtain the enhanced semantic vector.

[0127] Specifically, the deep meaning of electrical terminology, the importance weight of operation steps, the determination weight of key processes, and the time sequence logical relationship features of each sentence can be structured. Non-numerical semantic information can be transformed into a computable vector form through an embedding layer, unifying the data dimension with the quantified weight features and logical features.

[0128] The processed features are dynamically fused with the initial semantic vector through a fusion network. During the fusion process, a gating mechanism assigns appropriate weights to different features based on the requirements of the power grid operation scenario. For example, higher fusion priority is given to high-risk equipment directly related to safety and the importance of prohibited operations, while preconditions that affect the sequence of procedures in the temporal logical features are integrated in a focused manner. Finally, through multiple rounds of feature interaction and vector adjustment, an enhanced semantic vector is generated that retains the core of the initial semantics while incorporating deep semantics, weights, and logical information, meeting the semantic requirements for accurate identification of subsequent key procedures.

[0129] Furthermore, the fusion process can be a dynamic and adaptive adjustment process. For example, for sentences with complex referential relationships, the features provided by referential resolution will receive higher fusion weights; for sentences in the security measures section, the information features of that section will dominate the direction of adjustment of the semantic vector.

[0130] Understandably, after semantic enhancement, the final output is a new semantic vector enhanced with contextual information. This vector not only contains the literal meaning of the text itself, but also deeply integrates its precise semantics within a specific task scenario, document location, and logical process. When this enhanced vector is fed into the subsequent key process recognition classifier, it can greatly reduce ambiguity and improve classification accuracy.

[0131] To improve the quality of text data and reduce the space for semantic analysis errors, this application can also preprocess the power text before performing semantic representation on the power text and generating semantic vectors as mentioned in the foregoing embodiments, so as to obtain preprocessed power text.

[0132] Understandably, preprocessing before semantic representation and semantic vector generation of power data can improve the quality of text data from the source and lay a solid foundation for subsequent semantic analysis.

[0133] Specifically, firstly, data cleaning can remove noise from power industry texts, such as redundant formatting symbols in work records, repetitive operation descriptions, and irrelevant notes in work orders, preventing invalid information from interfering with semantic representation and ensuring the model focuses on core operational data. Secondly, word segmentation and part-of-speech tagging can accurately break down power industry terminology, clarifying the part-of-speech attributes of verbs and nouns, solving the problem of long and technical terms in power industry texts, making it easier for the model to capture key semantic units. Finally, preprocessing can standardize text format, unify the expression style of power industry texts from different sources, eliminate semantic understanding biases caused by format differences, and enable the DS large language model to stably process various types of power industry texts during semantic representation, generating basic semantic vectors that are more in line with the power grid field, providing a prerequisite guarantee for the accuracy of subsequent semantic enhancement and key process identification.

[0134] To more intuitively present the identified key process names, execution requirements, and timing relationships, and to further adapt to the needs of operators to quickly access information on-site via mobile terminals, this application, after identifying key processes in power text based on enhanced semantic vectors as mentioned in the aforementioned embodiments, can also output the key processes in the form of text or charts.

[0135] The charts can take the form of process flow diagrams, weighted heatmaps, etc.

[0136] Understandably, charts can visualize complex process relationships and their importance weights, helping managers quickly grasp the core links and risk points in the work process, and more efficiently carry out safety control and process optimization. Text or chart outputs improve the practicality and readability of the recognition results, reduce the information acquisition costs for operators and managers, and provide more flexible results support for intelligent management of power grid operations.

[0137] The apparatus for identifying key processes in power text provided in the embodiments of this application will be described below. The apparatus for identifying key processes in power text described below can be referred to in correspondence with the method for identifying key processes in power text described above.

[0138] See Figure 3 , Figure 3This is a schematic diagram of a device for identifying key processes in electrical text, as disclosed in an embodiment of this application.

[0139] like Figure 3 As shown, the device may include:

[0140] The power text acquisition unit 11 is used to acquire the power text of the key process to be identified;

[0141] Semantic vector generation unit 12 is used to perform semantic representation on the power text and generate semantic vectors;

[0142] The semantic enhancement unit 13 is used to enhance the semantic vector by parsing the deep meaning of sentences in the context of each sentence of the power text, and obtain an enhanced semantic vector.

[0143] The identification unit 14 is used to identify the key processes of the power text based on the enhanced semantic vector.

[0144] Optionally, the semantic enhancement unit includes:

[0145] The substitution resolution unit is used to perform substitution resolution on each sentence in the power text based on the context of the sentence, so as to obtain an updated sentence;

[0146] The terminology deep meaning mining unit is used to determine the deep meaning of electrical terms in each sentence of the electrical text based on each chapter of the electrical text.

[0147] The operation step weight determination unit is used to determine the modal verbs in each target sentence of the power text, and to determine the importance weight of the operation steps described in the target sentence based on the mandatory nature of the modal verbs and the scenario described in the target sentence.

[0148] The key process weight enhancement unit is used to determine the entity associated with each target sentence in the power text, and if the entity contains high-risk equipment, increase the key process determination weight of the operation steps of the target sentence.

[0149] The temporal logical relationship feature extraction unit is used to determine the relationship between each operation step based on the associated prompt words, and extract temporal logical relationship features based on the relationship between each operation step;

[0150] The weighted feature fusion unit is used to fuse the deep meaning of electrical terminology in each sentence, the importance weight of each operation step, the key process determination weight of each operation step, and the temporal logical relationship features with the semantic vector to obtain an enhanced semantic vector.

[0151] Optionally, the decontamination unit includes:

[0152] A sliding window creation unit is used to create a sliding window for each target sentence in the power text. The sliding window contains several sentences preceding the target sentence, several sentences following the target sentence, and the target sentence itself.

[0153] A pronoun determination unit is used to determine each pronoun in the target sentence;

[0154] The sentence update unit is used to determine the referent of each pronoun in the target sentence based on the sentences in the sliding window of the target sentence, and replace the corresponding pronoun with the referent to obtain the updated target sentence.

[0155] Optionally, the terminology deep meaning mining unit includes:

[0156] A chapter division unit is used to identify chapter keywords of the power text and divide it into chapters based on the chapter keywords;

[0157] The associated content determination unit is used to determine, for each target sentence in the power text, the target content associated with the power-related terms in the target sentence based on the semantics of the target sentence within its own chapter and the semantic dependencies of the target sentence in other chapters;

[0158] The deep meaning determination unit is used to determine the deep meaning of the electricity terms in the target sentence based on the electricity terms in the target sentence and their associated target content.

[0159] Optionally, the associated content determination unit includes:

[0160] An action step determination unit is used to determine the action information within the chapter to which the target sentence belongs if the target sentence contains a first operation step, and to determine each second operation step in other chapters that do not contain the target sentence.

[0161] The target action filtering unit is used to filter target action information associated with the first operation step from the action information;

[0162] The target operation step determination unit is used to determine, from each of the second operation steps, the target second operation step that is executed before and / or after the first operation step.

[0163] The target content determination unit is used to determine the target content associated with the electrical terminology in the target sentence based on the target action information and the target second operation step.

[0164] Optionally, the device may also include:

[0165] The preprocessing unit is used to preprocess the power text before semantic representation and semantic vector generation, so as to obtain preprocessed power text.

[0166] Optionally, the device may also include:

[0167] The output unit is used to output the key processes in the form of text or charts after identifying the key processes of the power text based on the enhanced semantic vector.

[0168] The device for identifying key processes in power text provided in this application embodiment can be applied to devices for identifying key processes in power text, such as terminals: computers, cloud computing, servers, etc. Optionally, Figure 4 The hardware structure block diagram of the device for identifying key processes in power text is shown, with reference to... Figure 4 The hardware structure of the device for identifying key processes in power text may include: at least one processor 1, at least one communication interface 2, at least one memory 3, and at least one communication bus 4.

[0169] In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4;

[0170] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0171] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0172] The memory stores a program, which the processor can call. The program is used for:

[0173] Obtain the power data of the key processes to be identified;

[0174] The electrical text is semantically represented to generate a semantic vector;

[0175] By parsing the deep meaning of sentences in the context of each sentence in the power text, the semantic vector is semantically enhanced to obtain an enhanced semantic vector;

[0176] Based on the enhanced semantic vector, the key processes of the power text are identified.

[0177] Optionally, the refined and extended functions of the program can be found in the description above.

[0178] This application embodiment also provides a storage medium that can store a program suitable for execution by a processor, the program being used for:

[0179] Obtain the power data of the key processes to be identified;

[0180] The electrical text is semantically represented to generate a semantic vector;

[0181] By parsing the deep meaning of sentences in the context of each sentence in the power text, the semantic vector is semantically enhanced to obtain an enhanced semantic vector;

[0182] Based on the enhanced semantic vector, the key processes of the power text are identified.

[0183] Optionally, the refined and extended functions of the program can be found in the description above.

[0184] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0185] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.

[0186] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying key processes of an electric power text, characterized by, The method comprises the following steps: acquiring an electric power text to be identified; performing semantic representation on the electric power text to generate a semantic vector; performing semantic enhancement on the semantic vector by analyzing deep meanings of sentences according to context semantics of each sentence of the electric power text to obtain an enhanced semantic vector; identifying a key process of the electric power text based on the enhanced semantic vector.

2. The method of claim 1, wherein, The method of performing semantic enhancement on the semantic vector by analyzing deep meanings of sentences according to context semantics of each sentence of the electric power text comprises the following steps: performing coreference resolution on each sentence of the electric power text according to context of the sentence to obtain an updated sentence; determining deep meanings of electric power terms of each sentence of the electric power text according to chapters of the electric power text; determining an emotive verb in each target sentence of the electric power text, determining importance weights of operation steps described by the target sentence according to the coerciveness of the emotive verb and a scene described by the target sentence; determining an entity associated with each target sentence of the electric power text, and increasing key process determination weights of operation steps of the target sentence if the entity contains a high-risk device; determining association relationships between operation steps according to association cues, and extracting time sequence logical relationship features based on the association relationships between the operation steps; fusing the deep meanings of the electric power terms of each sentence, the importance weights of each operation step, the key process determination weights of each operation step, the time sequence logical relationship features, and the semantic vector to obtain the enhanced semantic vector.

3. The method of claim 2, wherein, The method of performing coreference resolution on each sentence of the electric power text according to context of the sentence to obtain an updated sentence comprises the following steps: creating a sliding window of each target sentence of the electric power text, the sliding window containing a plurality of sentences before the target sentence, a plurality of sentences after the target sentence, and the target sentence; determining each pronoun of the target sentence; determining a referent of each pronoun of the target sentence according to each sentence in the sliding window of the target sentence, and replacing the pronoun with the referent to obtain an updated target sentence.

4. The method of claim 2, wherein, The method of determining deep meanings of electric power terms of each sentence of the electric power text according to chapters of the electric power text comprises the following steps: identifying chapter keywords of the electric power text, and dividing chapters according to the chapter keywords; determining target content associated with an electric power term in each target sentence of the electric power text according to semantics of the target sentence in a chapter where the target sentence is located and semantic dependency relationships of the target sentence in other chapters; determining deep meanings of the electric power term in the target sentence based on the electric power term in the target sentence and the target content associated with the electric power term.

5. The method of claim 4, wherein, The method of determining target content associated with an electric power term in each target sentence of the electric power text according to semantics of the target sentence in a chapter where the target sentence is located and semantic dependency relationships of the target sentence in other chapters comprises the following steps: If the target sentence contains a first operation step, action information in a chapter to which the target sentence belongs is determined, and each second operation step of other chapters not containing the target sentence is determined; From the action information, target action information associated with the first operation step is screened; From each of the second operation steps, a target second operation step executed before and / or after the first operation step is determined; According to the target action information and the target second operation step, target content associated with the power term in the target sentence is determined.

6. The method of claim 1, wherein, Before the power text is semantically represented to generate a semantic vector, it further includes: The power text is preprocessed to obtain a preprocessed power text.

7. The method according to any one of claims 1 to 6, characterized in that, After the key process of the power text is identified based on the enhanced semantic vector, it further includes: The key process is output in the form of text or chart.

8. A key process identification device for electrical engineering text, characterized in that, It includes: A power text acquisition unit is configured to acquire a power text to be identified for a key process; A semantic vector generation unit is configured to semantically represent the power text to generate a semantic vector; A semantic enhancement unit is configured to enhance the semantic vector by analyzing the deep meaning of each sentence in the context of the power text to obtain an enhanced semantic vector; An identification unit is configured to identify the key process of the power text based on the enhanced semantic vector.

9. An electric power text key process identification device characterized by comprising: It includes a memory and a processor; The memory is configured to store a program; The processor is configured to execute the program to implement each step of the key process identification method of the power text according to any one of claims 1-7.

10. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement each step of the key process identification method of the power text according to any one of claims 1-7. The computer program is executed by the processor to implement each step of the key process identification method of the power text according to any one of claims 1-7.

Images