Electric power dispatching instruction intelligent identification method, system and equipment based on semantic analysis and medium

Abstract

The invention discloses an electric power dispatching instruction intelligent identification method, system and device based on semantic analysis and a medium, and relates to the field of electric power dispatching instruction identification, and the method comprises the steps: obtaining an electric power dispatching instruction generated by a dispatching platform, and carrying out the preprocessing of the dispatching instruction, so as to form standardized word sequence input; based on standardized word sequence input, performing semantic coding on the scheduling instruction to generate semantic representation representing overall semantic features of the scheduling instruction; performing instruction type identification on the scheduling instruction based on the semantic representation, and determining an instruction type; under the constraint of the instruction type, performing sequence labeling processing on the scheduling instruction, and extracting element information in the scheduling instruction; and outputting a structured recognition result based on the instruction type and the element information. The scheduling instruction can be accurately understood and expressed in a standardized mode without depending on a fixed template and a keyword rule, and therefore the problems that in the prior art, the expression mode is sensitive, the generalization ability is insufficient, and element extraction is unstable are solved.

Description

Technical Field

[0001] This invention relates to the field of power dispatching instruction recognition, and in particular to a method, system, device, and medium for intelligent recognition of power dispatching instructions based on semantic parsing. Background Technology

[0002] With the continuous expansion of the power grid, the increasing complexity of its operation, and the diversification of dispatching scenarios, the number and complexity of dispatching orders have significantly increased. Currently, the power grid dispatching process generates a large number of dispatching instructions in natural language form, involving various types of information such as plant / station names, equipment numbers, voltage levels, action types, and operating statuses. However, existing dispatching instruction parsing systems still heavily rely on human experience, fixed template matching, or keyword retrieval, making it difficult to effectively adapt to the diverse writing habits, word order differences, and expression methods of different dispatchers. Against this backdrop, there is an urgent need for an intelligent parsing method capable of semantic understanding of natural language dispatching orders, automatic identification of instruction types, accurate extraction of equipment and action parameters, and output of unified structured results to support business scenarios such as intelligent ticketing, automated dispatching execution, and dispatching safety verification.

[0003] To address the limitations of existing technologies in parsing power dispatch instructions, this invention proposes an intelligent recognition method for power dispatch instructions based on semantic parsing and sequence labeling, which aims to improve the automation level and parsing accuracy of dispatch operations. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and medium for intelligent recognition of power dispatching instructions based on semantic parsing.

[0005] Therefore, the problem that this invention aims to solve is that the existing scheduling instruction parsing system still relies heavily on human experience, fixed template matching, or keyword retrieval, which makes it difficult to effectively adapt to the different writing habits, word order differences, and expression methods of different schedulers.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for intelligent recognition of power dispatch instructions based on semantic parsing, comprising: acquiring power dispatch instructions in natural language form generated by a dispatch platform, and performing text preprocessing on the dispatch instructions to form a standardized word sequence input; semantically encoding the dispatch instructions based on the standardized word sequence input to generate a semantic representation characterizing the overall semantic features of the dispatch instructions; identifying the instruction type of the dispatch instructions based on the semantic representation to determine the instruction type corresponding to the dispatch instructions; performing sequence labeling processing on the dispatch instructions under the constraint of the instruction type to extract element information from the dispatch instructions; and outputting the structured recognition result corresponding to the dispatch instructions based on the instruction type and element information.

[0007] As a preferred embodiment of the intelligent recognition method for power dispatch instructions based on semantic parsing described in this invention, the step of forming a standardized word sequence input includes: performing word segmentation processing on dispatch instructions in natural language form based on a pre-constructed power-specific dictionary, dividing the original character sequence into several word entries to form a standardized word sequence; assigning corresponding preliminary semantic tags to each word entry based on its semantic role in the dispatch instruction; and performing unified mapping processing on the word entries after word segmentation and semantic tagging.

[0008] As a preferred embodiment of the intelligent recognition method for power dispatch instructions based on semantic parsing described in this invention, the step of generating a semantic representation to characterize the overall semantic features of the dispatch instructions includes: converting a standardized word sequence input into a corresponding vectorized representation, making each word a word vector of a unified dimension; performing semantic feature fusion on the word vectors based on a context modeling structure, comprehensively depicting the contextual relationships between each word in the word sequence, and generating an intermediate representation that reflects the overall semantic information of the dispatch instructions; and performing convergence processing on the intermediate representation to obtain a semantic representation to characterize the overall semantic features of the dispatch instructions.

[0009] As a preferred embodiment of the intelligent identification method for power dispatch instructions based on semantic parsing described in this invention, the method for identifying the instruction type of the dispatch instructions includes: inputting the semantic representation into a multi-category classification model to map the semantic representation to a category space; determining the probability that the semantic representation belongs to each preset instruction type based on the category space to form an instruction type determination result; and determining the instruction type corresponding to the dispatch instruction based on the instruction type determination result.

[0010] The beneficial effects of this preferred technical solution are as follows: by introducing a multi-category classification model based on semantic representation, scheduling instructions are mapped to a category space and the probability of them belonging to each preset instruction type is determined, so that scheduling instructions can be accurately classified into the corresponding instruction type, thereby achieving clear category distinction at the semantic level of scheduling instructions.

[0011] As a preferred embodiment of the intelligent recognition method for power dispatch instructions based on semantic parsing described in this invention, the step of performing sequence labeling processing on dispatch instructions includes: constructing corresponding element labeling constraints based on the determined instruction type, and using the constraints as prior restrictions for sequence labeling processing; under the action of the element labeling constraints, performing word-by-word labeling processing on the word sequence corresponding to the dispatch instructions to generate labeling results reflecting the semantic roles of each word; and extracting element information in the dispatch instructions that matches the instruction type based on the labeling results.

[0012] As a preferred embodiment of the intelligent identification method for power dispatch instructions based on semantic parsing described in this invention, the structured identification result corresponding to the output dispatch instruction includes: organizing and associating element information to construct an element structure that matches the instruction type; standardizing the element structure to output structured data content as the structured identification result of the dispatch instruction.

[0013] The beneficial effects of this preferred technical solution are: it transforms the originally unstructured natural language scheduling instructions into standardized data content that can be directly used for subsequent system processing, thereby improving the consistency and manageability of scheduling instructions during transmission, storage, and invocation between systems.

[0014] As a preferred embodiment of the intelligent identification method for power dispatching instructions based on semantic parsing described in this invention, the structured identification result of the dispatching instruction includes: transmitting the structured identification result to the processing module corresponding to the power dispatching business; executing the dispatching business processing logic that matches the instruction type; and generating the corresponding dispatching processing result.

[0015] To address the aforementioned technical problems, this invention provides the following technical solution: a power dispatching instruction intelligent recognition system based on semantic parsing, comprising: a standardization module, a type recognition module, a recognition result module, and a processing module; the standardization module acquires power dispatching instructions in natural language form generated by the dispatching platform and performs text preprocessing on the dispatching instructions to form a standardized word sequence input; the type recognition module, based on the standardized word sequence input, performs semantic encoding on the dispatching instructions to generate a semantic representation characterizing the overall semantic features of the dispatching instructions, and based on the semantic representation, performs instruction type recognition on the dispatching instructions to determine the instruction type corresponding to the dispatching instructions; the recognition result module, under the constraint of the instruction type, performs sequence labeling processing on the dispatching instructions, extracts element information from the dispatching instructions, and outputs the structured recognition result corresponding to the dispatching instructions based on the instruction type and element information; the processing module executes dispatching business processing logic matching the instruction type to generate the corresponding dispatching processing result.

[0016] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the intelligent recognition method for power dispatch instructions based on semantic parsing as described above.

[0017] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the intelligent recognition method for power dispatch instructions based on semantic parsing as described above.

[0018] The beneficial effects of this invention are as follows: By constructing a semantic parsing and structured recognition method for natural language power dispatch instructions, this invention realizes a complete processing flow from dispatch instruction text preprocessing, semantic encoding, instruction type recognition, element extraction under type constraints, to structured result generation and business application. This enables dispatch instructions to be accurately understood and standardized without relying on fixed templates and keyword rules, thereby effectively overcoming the problems of sensitivity to expression methods, insufficient generalization ability, and unstable element extraction in existing technologies. It improves the accuracy and consistency of dispatch instruction recognition results and provides a reliable data foundation for the automatic processing, unified management, and intelligent support of dispatch business. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a method for intelligent recognition of power dispatch instructions based on semantic parsing in Example 1. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for intelligent recognition of power dispatching instructions based on semantic parsing, including: S1: Obtain power dispatch instructions in natural language form generated by the dispatch platform, and preprocess the dispatch instructions to form a standardized word sequence input.

[0024] S2: Based on standardized word sequence input, semantically encode the scheduling instructions to generate a semantic representation that characterizes the overall semantic features of the scheduling instructions.

[0025] S3: Based on semantic representation, identify the instruction type of the scheduling instruction and determine the instruction type corresponding to the scheduling instruction.

[0026] S4: Under the constraint of instruction type, perform sequence labeling on the scheduling instructions and extract the element information in the scheduling instructions.

[0027] S5: Based on the instruction type and element information, output the structured recognition result corresponding to the scheduling instruction.

[0028] It should be noted that in existing technologies, the parsing and processing of power dispatch instructions generally rely on human experience, fixed templates, or keyword matching rules, lacking the ability to understand the semantic layer of natural language dispatch instructions. This results in poor stability and accuracy of the parsing results when dispatchers express themselves in various ways, with different word orders or synonymous expressions. At the same time, existing solutions usually do not establish a unified instruction type modeling and element extraction mechanism. The instruction classification and element recognition processes are disconnected from each other, making it difficult to ensure the completeness and consistency of key information extraction in complex instruction scenarios.

[0029] Therefore, in response to the above problems, such as Figure 1 As shown, improving steps S1-S5 involves first standardizing and organizing the instruction content using text processing methods tailored to the power industry, transforming the instruction text into a word sequence with clear semantic boundaries. Based on this, the overall semantics of the instruction are modeled to form a unified representation reflecting its semantic features. Subsequently, the dispatch instruction is categorized according to the semantic representation, clarifying its instruction type. Under the constraints of the instruction type, the instruction content is analyzed in a fine-grained manner, gradually identifying and extracting key element information matching the instruction type. Finally, the identified instruction type and element information are organized and standardized to generate a structured dispatch instruction recognition result. This structured result is then used for subsequent dispatch business processing, thus realizing a complete transformation process from natural language expression of dispatch instructions to business-usable data.

[0030] Example 2, the second embodiment of the present invention, differs from the first embodiment in that: a method for intelligent recognition of power dispatch instructions based on semantic parsing further includes, in step S1, forming a standardized word sequence input, which includes the following steps A1-A3: A1: Based on a pre-built dictionary specifically for the power industry, the dispatch instructions in natural language form are segmented into words, dividing the original character sequence into several entries to form a standardized word sequence.

[0031] A2: For each word segmented entry, assign a corresponding preliminary semantic tag to each entry based on its semantic role in the scheduling instruction.

[0032] A3: Perform unified mapping processing on the entries after word segmentation and semantic tagging.

[0033] Specifically, the processing flow for dispatch instruction recognition begins with text preprocessing. Original dispatch instructions are typically composed of natural language, containing information such as plant name, voltage level, equipment number, bus identifier, action words, and operating status. However, due to differences in writing habits among dispatchers, the same type of entity may appear in different orders or expressions in the text. For example, "281 switch" and "switch 281" are semantically equivalent, but completely different at the string level. Without preprocessing, directly inputting such data into the subsequent model can easily lead to recognition confusion and errors in feature extraction. Therefore, the core task of text preprocessing is to transform the raw character stream into a word sequence with clear boundaries, and then add necessary prior features to ensure the input sequence has uniformity and parsability.

[0034] First, the system receives natural language dispatch orders from the dispatch platform and performs word segmentation based on a specialized dictionary built for the power industry. Assume the original dispatch order is a sequence of characters: The word sequence can then be obtained using the maximum matching algorithm: in, Represents a sequence of characters. Indicates the first One character, Indicates the total number of characters. Represents a word sequence, Indicates the first One entry, , Indicates the total number of entries. This represents a dictionary set in the field of electricity.

[0035] The goal of this word segmentation process is to ensure that the segmented word units are as consistent as possible with the semantic entities while covering all characters. In this way, key phrases containing terms such as "factory," "equipment," "action," "object," and "state" can be extracted as independent units, avoiding ambiguity caused by segmentation errors in subsequent steps.

[0036] In this embodiment of the application, step A3, the unified mapping processing adopts a unified mapping processing method based on standardized lexical normalization, including the following steps A311-A313: A311: After completing word segmentation and initial semantic tag allocation, a set of semantic unit standard representations is pre-constructed to describe the standard expression forms of various semantic units, taking into account the different writing orders, word form combinations, or expression methods that may exist for the same semantic unit in the scheduling instructions.

[0037] A312: Match each word segmented and semantically tagged entry with the set of semantic unit normative representations, identify the normative representation that is semantically equivalent to the current word, and determine the normative representation as the unified mapping result corresponding to the word.

[0038] A313: Replace the original word expression with the determined standard representation to generate a word sequence that is consistent at the semantic level and uniformly standardized at the representation level, which serves as the input for the standardized word sequence.

[0039] Specifically, after obtaining the word segmentation results, the system further assigns preliminary tags to each word. Let the tag set be... Then there exists a mapping function. : in, Indicates the first tag in the tag set Categories This includes categories such as plant name, voltage level, equipment number, action words, and operating status. Through this mapping process, each word segmentation unit is assigned preliminary semantic role information, becoming an important feature for subsequent semantic modeling and sequence labeling.

[0040] In an optional implementation, the unified mapping process may also employ a unified mapping process method based on encoded identifiers, including the following steps A321-A323: A321: After word segmentation and preliminary semantic tagging are determined, a unique internal encoding identifier is pre-assigned to semantic units of different categories to represent the unified identity of the semantic unit in the system.

[0041] A322: For each term with a labeled semantic role, query the corresponding encoding identifier based on its semantic tag and content features, and establish a mapping relationship between the term and the encoding identifier.

[0042] A323: Use encoding identifiers to replace or supplement the original terms, so that the representation of the same semantic unit in different scheduling instructions is unified into the same encoding expression, thereby forming a standardized word sequence input.

[0043] In another alternative implementation, the unified mapping process can also employ a unified mapping processing method based on a mapping rule table, including the following steps A331-A333: A331: After completing word segmentation and semantic tagging, a mapping rule table is pre-constructed to describe the mapping relationship between different word expressions and their corresponding unified semantic representations.

[0044] A332: Based on the mapping rule table, perform rule matching on each word after word segmentation and semantic tagging to determine the unified representation of each word under the current semantic role.

[0045] A333: The unified representation obtained by matching is applied to the word sequence, and the original words are replaced in a standardized manner to obtain a standardized word sequence input that is unified in representation and consistent in semantic roles.

[0046] It should be noted that by introducing word segmentation, semantic role labeling, and unified mapping based on a dedicated dictionary for the power industry during the text preprocessing stage, dispatch instructions in natural language form are converted into standardized word sequences with well-defined structures and semantic boundaries. This effectively eliminates the differences in writing order, expression methods, and word usage habits among different dispatchers, providing a consistent and stable input foundation for subsequent semantic encoding and type recognition, and improving the robustness and reusability of the overall recognition process.

[0047] Furthermore, in step S2, generating a semantic representation to characterize the overall semantic features of the scheduling instruction includes the following steps B1-B3: B1: Convert the standardized word sequence input into a corresponding vectorized representation, making each word a word vector of a uniform dimension.

[0048] B2: Based on the context modeling structure, semantic features of the word vectors are fused to comprehensively depict the contextual relationships between words in the word sequence and generate an intermediate representation that reflects the overall semantic information of the scheduling instructions.

[0049] B3: The intermediate representations are aggregated to obtain a semantic representation that characterizes the overall semantic features of the scheduling instructions.

[0050] Specifically, after text preprocessing, the original natural language dispatch orders are transformed into standardized word sequences with preliminary category labels. To further capture the overall semantic features of the dispatch orders and provide reliable input for subsequent classification and feature extraction, a semantic representation stage is introduced. In this stage, the system uses sentence vector modeling to encode the dispatch orders, mapping them to fixed-dimensional vector representations, thereby achieving the transition from discrete word sequences to a continuous semantic space.

[0051] Suppose that the word sequence obtained after preprocessing the scheduling order is... Each of them Corresponding to a word vector representation , Indicates by Composed of real numbers A real-valued vector space is used. To obtain the overall semantic representation of the dispatch order, a sentence vector model based on a bidirectional transformer structure is adopted, and the word sequence is modeled through a multi-layer attention mechanism. During computation, the model can simultaneously consider the contextual relationships and global dependencies of words, avoiding the gradient vanishing and limited dependency range problems existing in traditional sequence models in long text processing. In the sentence vector calculation process, each word vector... First, generate context-sensitive vector representations. (The middle part indicates).

[0052] In this embodiment of the application, step B3 employs a semantic representation generation method based on weighted aggregation, including the following steps B311-B313: B311: After completing context modeling and obtaining intermediate representations (context-related vectors) of word sequences. After that, weight parameters that reflect the degree of contribution of each semantic unit in the intermediate representation to the overall semantics are assigned to each semantic unit, which is used to characterize the differences in importance of different semantic units in the scheduling instructions.

[0053] B312: Based on the weight parameters, the intermediate representations corresponding to each semantic unit are weighted and summarized so that the semantic units that play a key role in the semantic understanding of scheduling instructions occupy a higher proportion in the overall representation.

[0054] B313: The weighted summation result is output as the overall semantic representation of the scheduling instruction, which is used to characterize the overall semantic features of the scheduling instruction.

[0055] Specifically, these vectors are weighted and combined using an attention mechanism to obtain the overall semantic vector of the dispatch order. For example, action words and device names typically determine the intent and execution goal of the dispatch order, and their weights should be higher than ordinary function words. Sentence vectors can be defined using the attention mechanism. for: in, Indicates the first Attention weight for each term, Represents the trainable context vector parameters. This represents the index variable used to traverse the entire sequence of scheduling instruction words. This weighted averaging process ensures that the model can assign a higher contribution to semantic key components when generating sentence vectors.

[0056] The obtained sentence vector A semantic vector is a fixed-dimensional vector used to characterize the overall intent and semantic features of a dispatch order. Compared to traditional representations based on keyword matching or statistical features, this semantic vector can better capture the diversity of language expressions used by different dispatchers. For example, "Switch 281 from bus I to bus II" and "Switch switch 281 from bus I to bus II for operation," although they differ in literal form, can exhibit high similarity in semantic space after sentence vector modeling, thus ensuring the accuracy of subsequent classification and rule retrieval.

[0057] In an optional implementation, the semantic representation can also employ a semantic representation generation method based on global statistical features, including the following steps B321-B323: B321: After obtaining the intermediate representation that reflects the contextual relationship of the word sequence, perform global statistical processing on the intermediate representation to extract statistical information that can reflect the overall semantic distribution characteristics.

[0058] B322: Based on statistical information, the intermediate representation is aggregated as a whole, so that the contextual features of each semantic unit in the word sequence are integrated in a unified space.

[0059] B323: The result after overall aggregation processing is output as the semantic representation of the scheduling instruction, which is used to describe the overall characteristics of the scheduling instruction at the semantic level.

[0060] In another alternative implementation, the semantic representation may also employ a semantic representation generation method based on the selection of key semantic units, including the following steps B331-B333: B331: After obtaining the intermediate representation, the importance of the semantic units in the intermediate representation is evaluated to identify key semantic units that are representative of the semantic expression of the scheduling instructions.

[0061] B332: For the identified key semantic units, their corresponding intermediate representations are centrally aggregated to form an aggregated result that highlights the core semantic content of the scheduling instructions.

[0062] B333: Output the converged results as the semantic representation of the scheduling instructions, which is used to characterize the overall semantic features of the scheduling instructions.

[0063] It should be noted that by converting standardized word sequences into word vectors of a unified dimension, and combining the semantic relationships between words with the context modeling structure, and then performing convergence processing on the intermediate representations to generate an overall semantic representation, the semantic features of scheduling instructions can be fully expressed in the vector space. This avoids relying solely on local words or keywords for judgment, and improves the accuracy and stability of semantic understanding of scheduling instructions in cases of complex sentence structures, word order changes, and synonymous expressions.

[0064] Furthermore, in step S3, the instruction type identification for the scheduling instruction includes the following steps C1-C3: C1: Input the semantic representation into the multi-class classification model so that the semantic representation is mapped to the class space.

[0065] C2: Based on the category space, determine the probability that the semantic representation belongs to each preset instruction type, and form an instruction type determination result.

[0066] C3: Determine the instruction type corresponding to the scheduling instruction based on the instruction type determination result.

[0067] Specifically, after obtaining the semantic vector representation of the scheduling order, it is necessary to further determine its operational intent and identify the corresponding instruction type. Instruction type identification plays a crucial role in the entire scheduling instruction identification process. Its function is not only to directly determine the category of the instruction but also to provide range constraints and prior conditions for subsequent parameter extraction and rule template matching. If the instruction type identification result is inaccurate, even with strong parameter extraction capabilities, subsequent steps may still match incorrect templates due to type bias, ultimately leading to incorrect operation ticket generation. Therefore, the accuracy of this step is crucial to the overall system stability.

[0068] A deterministic multi-class classification model is used to determine the semantic vector. Let the semantic representation be... This vector contains global context information about the scheduling order. First, a linear transformation is used to map this vector to the class space: in, The number of predefined instruction type categories, For the first The weight vector corresponding to the class, This is the bias term. The above mapping results... It can be viewed as scheduling instructions and categories Match scores between them.

[0069] In this embodiment of the application, in step C2, the determination result adopts an instruction type determination method based on probability distribution, including the following steps C211-C213: C211: After mapping the semantic representation to the category space, obtain the judgment value of the semantic representation in the category corresponding to each preset instruction type, and normalize the judgment value to form a probability distribution that reflects the relative probability of different instruction types.

[0070] C212: Based on probability distribution, compare the probability that the semantic representation belongs to each preset instruction type to obtain the corresponding instruction type determination result.

[0071] C213: Outputs the instruction type determination result, which is used to determine the instruction type corresponding to the scheduling instruction in the future.

[0072] Specifically, a Softmax function is applied to all category scores to normalize them into a probability distribution: By using probability distributions, the system can measure the likelihood that a scheduling order belongs to each instruction type. In the final determination, the category with the highest probability is selected as the prediction result. : During the training phase, the classifier is optimized using the cross-entropy loss function to maximize the probability of the true class, thereby continuously improving the classification ability. After sufficient training, the classification model can establish semantic equivalence between commands with different expressions. For example, "Switch switch 281 from bus section I to bus section II" and "Switch 281 is transferred from section I to section II for operation" differ in literal expression, but their sentence vector representations are highly similar in semantic space. Therefore, the classification model can correctly classify both as "bus section reversal" commands.

[0073] In an optional implementation, the determination result may also employ a threshold-based instruction type determination method, including the following steps C221-C223: C221: After obtaining the judgment value of the semantic representation in the category space, the judgment value corresponding to each preset instruction type is compared with the corresponding threshold condition according to the preset judgment rule.

[0074] C222: When an instruction type that meets the threshold condition exists, the instruction type that meets the threshold condition is used as the instruction type determination result; when there are multiple instruction types that meet the threshold condition, the instruction type determination result is determined according to the preset priority rule.

[0075] C223: Outputs the determined instruction type result for subsequent instruction type determination processing.

[0076] In another optional implementation, the determination result may also employ an instruction type determination method based on sorting selection, including the following steps C231-C233: C231: After the semantic representation is mapped to the category space, the judgment values ​​corresponding to each preset instruction type are sorted to reflect the relative matching degree of different instruction types.

[0077] C232: Based on the sorting results, select instruction types from the preset instruction types whose sorting results meet the preset conditions to form an instruction type determination result.

[0078] C233: Outputs the instruction type determination result, used to determine the instruction type corresponding to the scheduling instruction.

[0079] It should be noted that this step can classify all dispatch orders into a clear set of categories. Compared with traditional methods that rely on keyword triggering or manual rules, the method of this invention has stronger generalization ability. First, the model exhibits high fault tolerance when dealing with complex sentence structures and synonymous expressions, no longer limited to the predefined keyword matching range. Second, the classification results provide clear boundaries for subsequent parameter extraction, avoiding interference from incorrect categories on parameter identification. Third, the output format based on probability distribution allows the system to attach confidence to the judgment results, providing a quantitative basis for further verification and security control.

[0080] Furthermore, in step S4, the sequence labeling process for the scheduling instructions includes the following steps D1-D3: D1: Based on the determined instruction type, construct the corresponding feature labeling constraints and use the constraints as prior restrictions for sequence labeling processing.

[0081] D2: Under the constraints of feature annotation, the word sequence corresponding to the scheduling instruction is annotated word by word to generate annotation results that reflect the semantic role of each word.

[0082] D3: Based on the annotation results, extract the element information in the scheduling instructions that matches the instruction type.

[0083] Specifically, after completing instruction type identification, category determination alone is insufficient to support the automated execution of scheduling operations, as scheduling orders often involve crucial information such as specific operation objects, execution actions, and operational status. If these elements cannot be accurately extracted from natural language text, even with correct category identification, structured data that can directly drive the generation of rule templates and operation tickets cannot be formed. A deterministic sequence labeling method is employed to extract key elements of scheduling orders. It is assumed that a word sequence is obtained after text preprocessing and semantic modeling. Therefore, a label needs to be assigned to each term in the sequence to indicate its semantic role in the dispatch order. Let the label set be... To achieve this, a sequence labeling model based on Bidirectional Long Short-Term Memory (BiLSTM-CRF) is employed. BiLSTM can simultaneously capture both forward and backward contextual dependencies of words, thereby establishing long-distance cross-word associations at the semantic level.

[0084] In this embodiment of the application, step D1, the annotation constraints adopts an annotation constraint construction method based on the instruction type feature set, including the following steps D111-D113: D111: After determining the instruction type corresponding to the scheduling instruction, retrieve the set of elements pre-associated with that instruction type to limit the range of element categories allowed under that instruction type, thereby forming the corresponding element labeling constraints.

[0085] D112: Input the feature annotation constraints into the sequence annotation process to limit the semantic role types that can be assigned to each term in the word sequence.

[0086] D113: Under the constraints of feature annotation, perform word-by-word annotation processing on the word sequence corresponding to the scheduling instruction.

[0087] Specifically, the conditional random field layer imposes constraints on the entire label sequence to ensure that the transition relationships between labels conform to semantic logic. For example, the label "B-device" should be followed by "I-device", and "B-action" cannot appear directly.

[0088] In an optional implementation, the annotation constraints can also employ an annotation constraint construction method based on instruction type-feature relationship rules, including the following steps D121-D123: D121: After obtaining the instruction type of the scheduling instruction, generate the element labeling constraints that match the current instruction type based on the pre-built correspondence rules between instruction types and elements.

[0089] D122: Apply feature labeling constraints to the sequence labeling process to limit the order and combination of different semantic roles in the word sequence.

[0090] D123: Under the constraints of feature annotation, perform word-by-word annotation on the word sequence corresponding to the scheduling instruction.

[0091] In another alternative implementation, the annotation constraints can also employ a dynamic constraint construction method based on instruction type-driven candidate feature set pruning and consistency verification, including the following steps D131-D133: D131: After determining the instruction type of the scheduling instruction, the system reads the set of candidate elements allowed for the instruction type from the preset "instruction type-candidate element" correspondence, and further generates element consistency constraints based on the business semantic relationship of the instruction type. The consistency constraints include at least element mandatory constraints, element mutual exclusion constraints, and element combination legality constraints.

[0092] D132: When introducing feature labeling constraints, the system uses the candidate feature set to limit the label space range of word-by-word labeling, so that sequence labeling processing is only assigned within the semantic roles corresponding to the candidate feature set; at the same time, consistency constraints are applied to the label sequence generation and verification stage in the labeling process, which is used to filter or verify the candidate label sequences in real time, thereby eliminating label assignment results that do not meet the mandatory, mutually exclusive or combined legality constraints.

[0093] D133: Under the constraints, the system performs word-by-word annotation on the word sequence corresponding to the scheduling instruction, and continuously restricts and corrects the label allocation based on the candidate element set and consistency constraints during the annotation process, so that the final annotation result only contains the semantic role of the element that matches the instruction type.

[0094] It should be noted that by constructing corresponding feature annotation constraints under the already determined instruction type constraints, and performing sequence annotation processing under the action of these constraints, the semantic role allocation process in the word sequence is kept consistent with the specific instruction type. This avoids interference from irrelevant elements or unreasonable element combinations, and improves the accuracy, consistency, and business adaptability of feature extraction results in different instruction type scenarios.

[0095] Step D2 specifically includes: inputting the word vector representation of each word into a bidirectional long short-term memory network to obtain context-dependent hidden representations. Subsequently, a transition matrix is ​​defined in the conditional random field layer. ,in Indicates label To the target The transfer score of the tag. For a given tag sequence. Its conditional probability can be expressed as: in, Indicates hidden representation With tags Match score, Indicates the sequence with the current annotation. All different possible candidate label sequences.

[0096] The system can efficiently solve for the label sequence with the highest conditional probability during the inference phase, thereby obtaining the parameter labeling results of the dispatch order. Through this sequence labeling mechanism, the system can accurately extract key elements such as equipment, actions, objects, and states from natural language dispatch orders.

[0097] Furthermore, in step S5, the structured recognition result corresponding to the output scheduling instruction includes the following steps E1-E2: E1: Organize and associate element information to construct an element structure that matches the instruction type.

[0098] E2: Standardize the element structure and output structured data content as the structured recognition result of the scheduling command.

[0099] For example, for the instruction "281 switch is switched from bus section I to bus section II", the structured result output by the system is: Device = "281 switch", Action = "Switched to", Object = {"Bus section I", "Bus section II"}, Status = "Running".

[0100] To further explain, the structured recognition results are passed to the processing module corresponding to the power dispatching business, which executes the dispatching business processing logic that matches the instruction type and generates the corresponding dispatching processing results.

[0101] Example 3, the third embodiment of the present invention, differs from the previous two embodiments in that it provides a semantic parsing-based intelligent recognition system for power dispatch instructions, comprising a standardization module, a type recognition module, a recognition result module, and a processing module. The standardization module acquires power dispatch instructions in natural language form generated by the dispatch platform and performs text preprocessing on the instructions to form a standardized word sequence input. The type recognition module performs semantic encoding on the dispatch instructions based on the standardized word sequence input, generating a semantic representation to characterize the overall semantic features of the dispatch instructions. Based on the semantic representation, it performs instruction type recognition on the dispatch instructions to determine the corresponding instruction type. Under the constraint of the instruction type, the recognition result module performs sequence labeling on the dispatch instructions, extracts element information from the dispatch instructions, and outputs the structured recognition result corresponding to the dispatch instructions based on the instruction type and element information. The processing module executes dispatch business processing logic matching the instruction type to generate the corresponding dispatch processing result.

[0102] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0103] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0104] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0105] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented in combination with any of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent recognition of power dispatching instructions based on semantic parsing, characterized in that: include, The system acquires power dispatch instructions in natural language form generated by the dispatch platform and performs text preprocessing on the dispatch instructions to form a standardized word sequence input. Based on standardized word sequence input, the scheduling instructions are semantically encoded to generate a semantic representation that characterizes the overall semantic features of the scheduling instructions; Based on semantic representation, instruction type identification is performed on scheduling instructions to determine the instruction type corresponding to the scheduling instructions; Under the constraint of instruction type, the scheduling instructions are sequence labeled to extract the element information in the scheduling instructions; Based on the instruction type and element information, the structured recognition results corresponding to the scheduling instructions are output.

2. The intelligent identification method for power dispatching instructions based on semantic parsing as described in claim 1, characterized in that: The input for forming the standardized word sequence includes, Based on a pre-built dictionary specifically for the power industry, the dispatch instructions in natural language form are segmented into words, dividing the original character sequence into several words to form a standardized word sequence; For each word segmented entry, a corresponding preliminary semantic tag is assigned to each entry based on its semantic role in the scheduling instruction; A unified mapping process is performed on the entries after word segmentation and semantic tagging.

3. The intelligent identification method for power dispatching instructions based on semantic parsing as described in claim 2, characterized in that: The generation of the semantic representation used to characterize the overall semantic features of the scheduling instruction includes, The standardized word sequence input is converted into a corresponding vectorized representation, making each word a word vector of a uniform dimension; Based on the context modeling structure, semantic features of word vectors are fused to comprehensively depict the contextual relationships between words in the word sequence and generate an intermediate representation that reflects the overall semantic information of the scheduling instructions. The intermediate representations are aggregated to obtain a semantic representation that characterizes the overall semantic features of the scheduling instructions.

4. The intelligent recognition method for power dispatching instructions based on semantic parsing as described in claim 3, characterized in that: The instruction type identification of the scheduling instructions includes... Input the semantic representation into the multi-class classification model so that the semantic representation is mapped to the class space; Based on the category space, the probability of a semantic representation belonging to each preset instruction type is determined, and an instruction type determination result is formed. Based on the instruction type determination result, the instruction type corresponding to the scheduling instruction is determined.

5. The intelligent identification method for power dispatching instructions based on semantic parsing as described in claim 4, characterized in that: The sequence labeling process for scheduling instructions includes... Based on the determined instruction type, construct the corresponding feature labeling constraints, and use the constraints as prior restrictions for sequence labeling processing; Under the constraints of element labeling, the word sequence corresponding to the scheduling instruction is labeled word by word to generate labeling results that reflect the semantic role of each word; Based on the annotation results, extract the element information in the scheduling instructions that matches the instruction type.

6. The intelligent identification method for power dispatching instructions based on semantic parsing as described in claim 5, characterized in that: The structured recognition results corresponding to the output scheduling instructions include: Organize and associate element information to construct an element structure that matches the instruction type; The element structure is standardized, and the structured data content is output as the structured recognition result of the scheduling instructions.

7. The intelligent identification method for power dispatching instructions based on semantic parsing as described in claim 6, characterized in that: The structured recognition results of the scheduling instructions include, The structured recognition results are then transmitted to the processing module corresponding to the power dispatching business. Execute the scheduling business processing logic that matches the instruction type and generate the corresponding scheduling processing result.

8. A power dispatching instruction intelligent recognition system based on semantic parsing, employing the power dispatching instruction intelligent recognition method based on semantic parsing as described in any one of claims 1 to 7, characterized in that: It includes a standardization module, a type recognition module, a recognition result module, and a processing module; The standardization module acquires power dispatch instructions in natural language form generated by the dispatch platform and performs text preprocessing on the dispatch instructions to form a standardized word sequence input. The type recognition module performs semantic encoding on the scheduling instructions based on standardized word sequence input, generates a semantic representation to characterize the overall semantic features of the scheduling instructions, and performs instruction type recognition on the scheduling instructions based on the semantic representation to determine the instruction type corresponding to the scheduling instructions. The recognition result module performs sequence labeling processing on the scheduling instructions under the constraint of the instruction type, extracts the element information in the scheduling instructions, and outputs the structured recognition result corresponding to the scheduling instructions based on the instruction type and element information. The processing module executes scheduling business processing logic that matches the instruction type and generates the corresponding scheduling processing result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent recognition method for power dispatch instructions based on semantic parsing, as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent identification method for power dispatch instructions based on semantic parsing as described in any one of claims 1 to 7.

Images