Semantic clustering-based power starting scheme generation method and system

Abstract

The invention discloses an electric power starting scheme generation method and system based on semantic clustering, and relates to the technical field of electric power dispatch. A training starting scheme text is obtained, vector conversion processing is conducted on the training starting scheme text through a preset starting scheme model, and multiple starting sentence vectors are obtained; and performing semantic clustering processing on each starting sentence vector to obtain a corresponding clustering result, constructing a power knowledge graph based on the clustering result, and when information of the to-be-started equipment is received, inputting the information of the to-be-started equipment into the power knowledge graph to obtain a power starting scheme corresponding to the information of the to-be-started equipment. The technical problems that in an existing power grid equipment knowledge graph construction process, non-structured data in a text scheme cannot be accurately extracted, consequently, key operation information is lacked in generation of a power equipment starting scheme, and the operation reliability of a power system is reduced are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of power dispatching, and in particular to a method and system for generating a power startup plan based on semantic clustering. Background Art

[0002] In the context of the intelligent transformation of new power systems, the generation of power equipment startup plans has become a core link in ensuring the safe operation of the power grid. Traditional methods for generating power equipment startup plans rely mainly on manual experience to parse and extract knowledge from text plans. This not only has low data processing efficiency but also high extraction error rates. Therefore, it cannot meet the ever-changing scheduling needs of the power system.

[0003] At present, current technology mainly generates the power equipment startup plan required for power system scheduling by building a standardized power grid equipment knowledge graph. Although the relationship between equipment and startup plan is established in the power grid equipment knowledge graph, the unstructured data in the text plan cannot be accurately extracted during the construction of the power grid equipment knowledge graph, resulting in the lack of key operation information in the generated power equipment startup plan, which reduces the reliability of power system operation. Summary of the Invention

[0004] The present invention provides a method and system for generating an electric power startup plan based on semantic clustering, which solves the technical problem that current technology mainly generates the electric power equipment startup plan required for power system scheduling by constructing a standardized power grid equipment knowledge graph. Although the relationship between equipment and startup plan is established in the power grid equipment knowledge graph, the unstructured data in the text plan cannot be accurately extracted during the construction of the power grid equipment knowledge graph, resulting in the lack of key operation information in the generated power equipment startup plan, which reduces the reliability of the power system operation.

[0005] A first aspect of the present invention provides a method for generating a power startup plan based on semantic clustering, comprising:

[0006] Obtaining a training startup scenario text, and performing vector conversion processing on the training startup scenario text using a preset startup scenario model to obtain multiple startup sentence vectors;

[0007] Performing semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results;

[0008] Constructing an electric power knowledge graph based on the clustering results;

[0009] When information about a device to be started is received, the information about the device to be started is input into the power knowledge graph to obtain a power startup solution corresponding to the information about the device to be started.

[0010] Optionally, the step of performing vector conversion processing on the training startup scenario text using a preset startup scenario model to obtain a plurality of startup sentence vectors includes:

[0011] Segmenting the training startup program text to obtain a plurality of startup sentences;

[0012] Perform word segmentation processing on each of the startup sentences respectively to obtain word segmentation results corresponding to each of the startup sentences;

[0013] Input the word segmentation results corresponding to each of the startup sentences into a preset startup scheme model to obtain multiple word vectors;

[0014] The word vectors associated with each of the startup sentences are weighted averaged to obtain multiple startup sentence vectors.

[0015] Optionally, the step of performing semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results includes:

[0016] Calculating the cosine similarity between any two of the startup sentence vectors using a preset cosine similarity function;

[0017] Determine whether each cosine similarity is greater than or equal to a preset similarity threshold;

[0018] When the cosine similarity is greater than or equal to the similarity threshold, performing a deduplication operation on the startup sentence vector associated with the cosine similarity to obtain a deduplication sentence vector;

[0019] Clustering is performed on each of the deduplicated sentence vectors based on a preset clustering algorithm to obtain corresponding clustering results.

[0020] Optionally, the step of constructing a power knowledge graph based on the clustering results includes:

[0021] Acquire the power language entities in the clustering results according to a pre-trained entity recognition model;

[0022] Each type of equipment information of the clustering result is used as a node to construct an association graph with the relevant power language entity;

[0023] Combine all related graphs to generate the power knowledge graph.

[0024] Optionally, the training process of the entity recognition model is specifically as follows:

[0025] Obtaining a power startup training text, and classifying and labeling the power startup training text according to a preset classification order to obtain a startup training data set, wherein the classification order is step number, device name, operation step, device status, and operation requirements;

[0026] Inputting the startup training data set into a preset sequence labeling model for training to obtain a power language entity classification result;

[0027] Calculating the training loss value of the startup training data set according to the power language entity classification result;

[0028] When the training loss value is greater than or equal to a preset standard loss value, a preset optimizer is used to adjust the model parameters of the sequence labeling model until the training loss value is less than the standard loss value;

[0029] When the training loss value is less than the standard loss value, an entity recognition model is generated.

[0030] Optionally, it also includes:

[0031] Acquire a newly added entity in real time, and select nodes associated with the newly added entity as similar nodes from the power knowledge graph according to a phonetic code algorithm;

[0032] Calculating the character similarity between the newly added entity and the similar node, and determining whether the character similarity is greater than a preset character similarity threshold;

[0033] If the character similarity is greater than the character similarity threshold, the newly added entity is merged with similar nodes in the electric power knowledge graph to obtain an updated electric power knowledge graph;

[0034] If the character similarity is less than or equal to the character similarity threshold, the newly added entity is added as a new node to the electric power knowledge graph to obtain an updated electric power knowledge graph.

[0035] A second aspect of the present invention provides a system for generating a power startup plan based on semantic clustering, comprising:

[0036] An acquisition module is used to obtain a training startup plan text, and perform vector conversion processing on the training startup plan text through a preset startup plan model to obtain multiple startup sentence vectors;

[0037] A semantic clustering module, configured to perform semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results;

[0038] A construction module, configured to construct an electric power knowledge graph based on the clustering results;

[0039] The generation module is used to input the information of the device to be started into the power knowledge graph when receiving the information of the device to be started, and obtain the power startup plan corresponding to the information of the device to be started.

[0040] A third aspect of the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method for generating a power startup scheme based on semantic clustering as described in any one of the above items.

[0041] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the method for generating a power startup plan based on semantic clustering as described in any one of the above items.

[0042] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the power startup scheme generation method based on semantic clustering as described in any one of the above items.

[0043] It can be seen from the above technical solutions that the present invention has the following advantages:

[0044] The present invention performs semantic clustering processing on each startup sentence vector to obtain corresponding clustering results, and constructs an electric power knowledge graph based on the clustering results. When the information of the device to be started is received, the information of the device to be started is input into the electric power knowledge graph to obtain the electric power startup plan corresponding to the information of the device to be started. This overcomes the technical problem that the unstructured data in the text plan cannot be accurately extracted during the construction of the existing power grid equipment knowledge graph, resulting in a lack of key operating information in the generated electric power equipment startup plan, which reduces the reliability of the power system operation. Compared with the traditional method for generating electric power startup plans, the present invention performs semantic clustering processing on each startup sentence vector to obtain corresponding clustering results, optimizes the sample data quality for constructing the electric power knowledge graph, ensures the accuracy of the electric power equipment startup plan generated by the electric power knowledge graph, and improves the reliability of the power system operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0046] Figure 1 A flowchart of a method for generating a power startup plan based on semantic clustering provided in the first embodiment of the present invention;

[0047] Figure 2A flowchart of a method for generating a power startup plan based on semantic clustering provided in the second embodiment of the present invention;

[0048] Figure 3 A schematic diagram of the structure of the BiLSTM-CRF model provided in Example 2 of the present invention;

[0049] Figure 4 This is a training flowchart of the BiLSTM-CRF model provided in Example 2 of the present invention;

[0050] Figure 5 This is a structural block diagram of a power startup plan generation system based on semantic clustering provided in the third embodiment of the present invention;

[0051] Figure 6 This is a structural block diagram of a computer device provided in Example 4 of the present invention. DETAILED DESCRIPTION

[0052] An embodiment of the present invention provides a method and system for generating an electric power startup plan based on semantic clustering, which is used to solve the technical problem that current technology mainly generates the electric power equipment startup plan required for power system scheduling by constructing a standardized power grid equipment knowledge graph. Although the relationship between equipment and startup plan is established in the power grid equipment knowledge graph, the unstructured data in the text plan cannot be accurately extracted during the construction of the power grid equipment knowledge graph, resulting in the lack of key operation information in the generated power equipment startup plan, thereby reducing the reliability of the power system operation.

[0053] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] See also Figure 1 , Figure 1 This is a flowchart of the steps of a method for generating a power startup plan based on semantic clustering provided in the first embodiment of the present invention.

[0055] The present invention provides a method for generating a power startup plan based on semantic clustering, comprising:

[0056] Step 101: Obtain a training startup plan text, and perform vector conversion processing on the training startup plan text using a preset startup plan model to obtain multiple startup sentence vectors;

[0057] The training startup plan text refers to the massive documents and operation records of historical power grid startup plans.

[0058] The starting sentence vector refers to the vector corresponding to each sentence in the training starting solution text.

[0059] In an embodiment of the present invention, a training startup plan text is obtained, and the training startup plan text is segmented to obtain a plurality of startup sentences. A preset Chinese word segmentation tool is used to segment each startup sentence to obtain a word segmentation result for each startup sentence. The word segmentation result corresponding to each startup sentence is input into a preset startup plan model to obtain a plurality of word vectors. The word vectors associated with each startup sentence are weighted averaged to obtain a plurality of startup sentence vectors. For example, sentence 1 is "After putting main transformer #1 into operation, check the bus voltage." After being processed by the Chinese word segmentation tool and the startup plan model in turn, it is converted into a startup sentence vector of fixed dimension (i.e., [0.23, -0.45, 1.21, ...]). The startup sentence vector reflects the semantic information of the sentence, including but not limited to power equipment, operation steps, equipment status, etc.

[0060] It's important to note that the startup model is a pre-trained Word2Vec model. The Word2Vec model is a classic word embedding model that maps words into a low-dimensional vector space, capturing the semantic and grammatical relationships between words. The Word2Vec model generates word embeddings through training, which can be used in various natural language processing tasks, such as text classification, machine translation, and sentiment analysis.

[0061] Step 102: Perform semantic clustering on each startup sentence vector to obtain corresponding clustering results;

[0062] Clustering results refer to the results of dividing multiple data into several groups (i.e., "clusters" or "classes") through clustering algorithms.

[0063] In an embodiment of the present invention, any two startup sentence vectors are input into a preset cosine similarity function to obtain multiple cosine similarities. A determination is then made as to whether each cosine similarity is greater than or equal to a preset similarity threshold. If the cosine similarity is greater than or equal to the similarity threshold, the startup sentence vectors associated with the cosine similarity are deduplicated to obtain deduplicated sentence vectors. A preset clustering algorithm is then used to cluster the deduplicated sentence vectors to obtain corresponding clustering results.

[0064] Step 103: constructing an electric power knowledge graph based on the clustering results;

[0065] The power knowledge graph is a graph-based knowledge representation method specifically designed for knowledge management and application in the power sector. It models and stores entities (such as equipment, power grids, and users), attributes (such as voltage, current, and power), and relationships (such as connections, dependencies, and controls) within the power sector in a graph format, supporting the generation of startup plans for power system equipment.

[0066] In an embodiment of the present invention, the power language entities in the clustering results are obtained based on a pre-trained entity recognition model, and each type of equipment information in the clustering results is used as a node to construct an association graph with the relevant power language entities. All the association graphs are combined to generate a power knowledge graph.

[0067] Step 104: When the information of the device to be started is received, the information of the device to be started is input into the power knowledge graph to obtain a power startup plan corresponding to the information of the device to be started.

[0068] Information on equipment to be started refers to the relevant data and status information of power equipment in the power system that is planned to be started but has not yet been officially put into operation.

[0069] In an embodiment of the present invention, when information about a device to be started is received, the information about the device to be started is used as input to the power knowledge graph to obtain a power startup plan corresponding to the information about the device to be started.

[0070] In an embodiment of the present invention, semantic clustering is performed on each startup sentence vector to obtain corresponding clustering results, and an electric power knowledge graph is constructed based on the clustering results. When information about the device to be started is received, the information about the device to be started is input into the electric power knowledge graph to obtain an electric power startup plan corresponding to the information about the device to be started. This overcomes the technical problem that the unstructured data in the text plan cannot be accurately extracted during the construction of the existing knowledge graph for power grid equipment, resulting in a lack of key operating information in the generated electric power equipment startup plan, thereby reducing the reliability of the power system operation. Compared with the traditional method for generating electric power startup plans, the present invention performs semantic clustering on each startup sentence vector to obtain corresponding clustering results, thereby optimizing the sample data quality for constructing the electric power knowledge graph, ensuring the accuracy of the electric power equipment startup plan generated by the electric power knowledge graph, and improving the reliability of the power system operation.

[0071] See also Figure 2 , Figure 2 This is a flowchart of the steps of a method for generating a power startup plan based on semantic clustering provided in the second embodiment of the present invention.

[0072] The present invention provides a method for generating a power startup plan based on semantic clustering, comprising:

[0073] Step 201: Obtain a training startup plan text, and perform vector conversion processing on the training startup plan text using a preset startup plan model to obtain multiple startup sentence vectors;

[0074] Furthermore, step 201 includes the following sub-steps:

[0075] S11, performing sentence segmentation on the training startup program text to obtain multiple startup sentences;

[0076] In an embodiment of the present invention, sentence segmentation is performed on the training startup scenario text based on a preset sentence segmentation rule to obtain a plurality of startup sentences.

[0077] It should be noted that the specific sentence segmentation rule is: use the period, question mark, exclamation mark, and line break as the end of the sentence to segment the sentence. If the end mark is preceded by some abbreviations or decimal points, it will not be used as the basis for sentence segmentation.

[0078] S12, performing word segmentation processing on each startup sentence respectively to obtain the word segmentation results corresponding to each startup sentence;

[0079] In the embodiment of the present invention, a preset Chinese word segmentation tool is used to perform word segmentation processing on each startup sentence to obtain a word segmentation result for each startup sentence.

[0080] S13, inputting the word segmentation results corresponding to each starting sentence into a preset starting scheme model to obtain multiple word vectors;

[0081] In an embodiment of the present invention, the word segmentation results corresponding to each startup sentence are input into a pre-trained Word2Vec model to obtain multiple word vectors.

[0082] S14. Perform weighted averaging on the word vectors associated with each startup sentence to obtain multiple startup sentence vectors.

[0083] In an embodiment of the present invention, the word vectors associated with each startup sentence are respectively input into a preset weighted average function to obtain multiple startup sentence vectors.

[0084] It should be noted that the weighted average function is specifically:

[0085]

[0086] in, For the word vectors, For the The importance weight of each word vector can be manually set based on the experience of the power industry. For example, equipment names and action words ("commissioning" and "exit") are given higher weights. n is the total number of word vectors associated with the startup sentence, and y is the startup sentence vector.

[0087] Step 202: Perform semantic clustering on each startup sentence vector to obtain corresponding clustering results;

[0088] Furthermore, step 202 includes the following sub-steps:

[0089] S21, calculating the cosine similarity between any two starting sentence vectors using a preset cosine similarity function;

[0090] In an embodiment of the present invention, any two starting sentence vectors are respectively input into a preset cosine similarity function to obtain multiple cosine similarities.

[0091] It should be noted that the cosine similarity function is specifically:

[0092]

[0093] in, and They represent the starting sentence vectors respectively. The closer the similarity is to 1, the closer the semantics of the two starting sentence vectors are.

[0094] S22, determining whether each cosine similarity is greater than or equal to a preset similarity threshold;

[0095] The similarity threshold refers to the critical value for determining whether two startup sentence vectors are similar (the value range is 0.8-0.9).

[0096] In the embodiment of the present invention, it is determined whether each cosine similarity is greater than or equal to 0.8.

[0097] It is worth mentioning that the similarity threshold is set between 0.8 and 0.9, which can balance the similarity and difference between the startup sentence vectors, ensuring that only semantically similar and representative sentences are retained, while avoiding excessive deduplication and retaining sufficient key information.

[0098] S23, when the cosine similarity is greater than or equal to the similarity threshold, performing a deduplication operation on the starting sentence vector associated with the cosine similarity to obtain a deduplication sentence vector;

[0099] In an embodiment of the present invention, when the cosine similarity is greater than or equal to 0.8, the startup sentence vectors associated with the cosine similarity are deduplicated (two startup sentence vectors are merged or one startup sentence vector is removed) to obtain a deduplicated sentence vector.

[0100] It is worth mentioning that when the cosine similarity is greater than or equal to the similarity threshold, the starting sentence vectors associated with the cosine similarity are deduplicated to obtain the deduplicated sentence vectors. This prevents repeated sentences from being classified into the same category during clustering.

[0101] S24. Clustering is performed on each deduplicated sentence vector based on a preset clustering algorithm to obtain corresponding clustering results.

[0102] It should be noted that the clustering algorithm used is the K-means clustering algorithm, or hierarchical clustering algorithm. K-means is a distance-based partitioning clustering algorithm that aims to partition data into K mutually exclusive clusters, ensuring that data points within a cluster are as similar as possible and data points between clusters are as dissimilar as possible. Hierarchical clustering algorithms gradually merge or split clusters by constructing a tree-like structure. There are two types: agglomerative (bottom-up), where each data point is initially a cluster, and the closest clusters are gradually merged; and divisive (top-down), where all data points are initially a cluster, which is gradually split into smaller clusters.

[0103] In an embodiment of the present invention, a partitioning clustering algorithm or a hierarchical clustering algorithm is used to perform clustering processing on each deduplication sentence vector to obtain a corresponding clustering result.

[0104] In another embodiment, each deduplication sentence vector is used to construct a similarity matrix, and a preset clustering algorithm is used to perform clustering processing on the similarity matrix to obtain a corresponding clustering result. Represents the cosine similarity between the i-th sentence and the j-th sentence

[0105] It's worth noting that, based on the clustering results, the training startup scenario texts are categorized into different categories. For example, Category 1 includes "Equipment Startup Operations (Category 1 includes but is not limited to "Commissioning of Main Transformer #1"), Category 2 includes "Relay Protection Operations (Category 2 includes but is not limited to "Bus Differential Protection Startup"), and Category 3 includes "Fault Detection and Recovery (Category 3 includes but is not limited to "Checking for Equipment Overload")." Each text is assigned to the appropriate category, reducing text heterogeneity and improving model adaptability.

[0106] Step 203: Obtaining power language entities in the clustering results based on the pre-trained entity recognition model;

[0107] In an embodiment of the present invention, the power speech entities in the clustering results are obtained based on a pre-trained BiLSTM-CRF model (i.e., a bidirectional long short-term memory-conditional random field model).

[0108] It should be noted that the training process of the entity recognition model is as follows:

[0109] A1. Obtain power startup training text and classify and label it according to a preset classification order to obtain a startup training dataset. The classification order is step number, device name, operation step, device status, and operation requirements.

[0110] In this embodiment of the present invention, the power startup training text is obtained by referring to Table 1. The power startup training text is then annotated with BIO tags according to a preset classification order (i.e., annotating the entity start position, internal portion, and non-entity portion of the text) to obtain a startup training set. The classification order is step number, device name, operation step, device status, and operation requirement.

[0111] Table 1

[0112]

[0113] It should be noted that, before classification and labeling, the power startup training text needs to be processed in steps 201 - 202 to obtain a training result corresponding to the power startup training text.

[0114] It should be noted that during the classification and annotation process, the BIO annotation format can clearly semantically segment and categorize key information in the text, providing precise support for the subsequent structured processing of the model. In BIO annotation, each word is assigned a specific label to clarify its role in the sentence. For example, the starting point of the entity is marked as B, the internal part of the entity is marked as I, and the non-entity part is marked as O. In response to the actual needs of the power startup plan, the main entity categories annotated include the startup sequence (such as "Step 1"), the equipment body (such as "4V67 switch"), the operation steps (such as "cold standby to operation"), and the relay protection steps (such as "bus differential protection"). This annotation method can accurately capture the logical relationship between the key operation steps, equipment objects, and protection actions in the power startup plan.

[0115] It's worth noting that under the BIO annotation rules, 1. The label B-NAM (beginning of the equipment name) must be immediately followed by I-NAM (the internal portion of the equipment name). The B-OPE (operation procedure) label is often closely associated with the B-NAM (equipment name), for example, "Cold Standby to Operation Main Transformer #1." As shown in Table 2, after BIO annotation, the data has a more refined classification.

[0116] Table 2

[0117]

[0118] A2. Input the startup training dataset into the preset sequence labeling model for training to obtain the power language entity classification results;

[0119] In the embodiment of the present invention, see Figure 4 As shown, the startup training data set is input into the preset BiLSTM-CRF model for training to obtain the power speech entity classification results.

[0120] Notably, the BiLSTM-CRF model utilizes a multi-layered architecture consisting of an input layer, a BiLSTM layer (i.e., a bidirectional long short-term memory layer), and a CRF layer (i.e., a conditional random field layer). This layer achieves refined text processing and entity recognition through semantic feature extraction and sequence labeling optimization. At the input layer, the startup training dataset is fed into the model. Each word is converted into a word vector representation using an embedding model. This embedding model is domain-adapted and trained on a large number of power grid startup plan documents and operation records to capture the semantic relevance of power terminology. Semantic embeddings for terms such as "main transformer grid connection" and "cold standby" ensure accurate representation of their meaning in different contexts. At the BiLSTM layer, a bidirectional long short-term memory network (BiLSTM) extracts semantic features from the input sequence. The forward LSTM (i.e., forward long short-term memory network) in the BiLSTM learns forward dependencies between sentences, while the backward LSTM (backward long short-term memory network) learns backward dependencies. This bidirectional modeling approach captures the complex semantics and logic of power startup plans. For example, in the sentence "Step 1: Check the bus voltage after commissioning the main transformer", the forward LSTM can learn the association between "step" and "1", while the backward LSTM can recognize the backward logical relationship between "check" and "bus voltage".

[0121] The output hidden state of BiLSTM is expressed as:

[0122]

[0123] in, is the hidden state of the forward LSTM, is the hidden state of the backward LSTM, is the output hidden state of BiLSTM, which contains the complete semantic information of the context.

[0124] In the CRF layer, a conditional random field network (CRF) is used as the output layer for sequence labeling to optimize the global consistency of the output labels. CRF not only predicts each word independently based on the features output by the BiLSTM layer, but also learns the dependencies between labels.

[0125] The goal of CRF is to maximize the conditional probability of the labeled sequence:

[0126]

[0127] in, is the local characteristic function, is the i-th label, is the i-1th label, is the i-th context feature, is the normalization term of the i-th label sequence, is the normalization term of the i-1th label sequence. The Viterbi algorithm is used to find the optimal path that maximizes the conditional probability, thereby generating a logical entity labeling sequence.

[0128] A3. Calculate the training loss value of the startup training dataset based on the power language entity classification results;

[0129] In an embodiment of the present invention, based on a negative log-likelihood function, a training loss value of the startup training data set is calculated according to the power speech entity classification results.

[0130] A4. When the training loss value is greater than or equal to the preset standard loss value, the preset optimizer is used to adjust the model parameters of the sequence labeling model until the training loss value is less than the standard loss value;

[0131] In an embodiment of the present invention, it is determined whether the training loss value is less than a preset standard loss value. If the training loss value is greater than or equal to the standard loss value, a preset optimizer is used to adjust the model parameters of the sequence labeling model until the training loss value is less than the standard loss value.

[0132] A5. When the training loss value is less than the standard loss value, an entity recognition model is generated.

[0133] In an embodiment of the present invention, if the training loss value is less than the standard loss value, an entity recognition model is generated.

[0134] Step 204: Use each type of equipment information from the clustering results as a node and construct a correlation graph with the relevant power language entity;

[0135] In the embodiment of the present invention, each type of equipment information of the clustering result is used as a node and respectively associated with the power language entity to construct an association graph.

[0136] Step 205: Combine all associated graphs to generate a power knowledge graph.

[0137] In an embodiment of the present invention, all associated graphs are combined to obtain an electric power knowledge graph.

[0138] Step 206: When the information of the device to be started is received, the information of the device to be started is input into the power knowledge graph to obtain a power startup plan corresponding to the information of the device to be started.

[0139] In an embodiment of the present invention, when information about a device to be started is received, the information about the device to be started is used as input to the power knowledge graph to obtain a power startup plan corresponding to the information about the device to be started.

[0140] Furthermore, it also includes:

[0141] Step 207: Acquire the newly added entity in real time, and select nodes associated with the newly added entity as similar nodes from the power knowledge graph according to the phonetic code algorithm;

[0142] Calculate the character similarity between the newly added entity and the similar node, and determine whether the character similarity is greater than a preset character similarity threshold;

[0143] If the character similarity is greater than the character similarity threshold, the newly added entity is merged with the similar node in the electric power knowledge graph to obtain an updated electric power knowledge graph;

[0144] If the character similarity is less than or equal to the character similarity threshold, the new entity is added as a new node to the electric power knowledge graph to obtain an updated electric power knowledge graph.

[0145] New entities refer to power language entities extracted from new equipment or operation plans.

[0146] In an embodiment of the present invention, the presence of new entities is detected in real time. When a new entity is detected, nodes associated with the new entity are selected from the power knowledge graph as similar nodes based on a phonetic-graphic coding algorithm. The character similarity between the new entity and the similar node is calculated, and a determination is made as to whether the character similarity exceeds a preset character similarity threshold. If the character similarity exceeds the threshold, the new entity is merged with the similar node in the power knowledge graph to obtain an updated power knowledge graph. If the character similarity is less than or equal to the threshold, the new entity is added to the power knowledge graph as a new node to obtain an updated power knowledge graph. For example, when a new device "4V68 switch" is added, the system can determine its similarity to "4V67 switch" based on its pronunciation characteristics and character morphology, and quickly integrate it into the knowledge graph based on its topological role. Character similarity calculation not only solves the problem of spelling differences in device names but also supports the classification of synonyms, such as mapping "main transformer #1" and "main transformer 1" to the same device entity.

[0147] It is worth mentioning that when an abandoned entity is detected, the node corresponding to the abandoned entity in the power knowledge graph is marked as inactive, and irrelevant edges are deleted to ensure the simplicity and timeliness of the power knowledge graph.

[0148] In an embodiment of the present invention, semantic clustering is performed on each startup sentence vector to obtain corresponding clustering results, and an electric power knowledge graph is constructed based on the clustering results. When information about the device to be started is received, the information about the device to be started is input into the electric power knowledge graph to obtain an electric power startup plan corresponding to the information about the device to be started. This overcomes the technical problem that the unstructured data in the text plan cannot be accurately extracted during the construction of the existing knowledge graph for power grid equipment, resulting in a lack of key operating information in the generated electric power equipment startup plan, thereby reducing the reliability of the power system operation. Compared with the traditional method for generating electric power startup plans, the present invention performs semantic clustering on each startup sentence vector to obtain corresponding clustering results, thereby optimizing the sample data quality for constructing the electric power knowledge graph, ensuring the accuracy of the electric power equipment startup plan generated by the electric power knowledge graph, and improving the reliability of the power system operation.

[0149] See also Figure 5 , Figure 5 This is a structural block diagram of a power startup plan generation system based on semantic clustering provided in the third embodiment of the present invention.

[0150] The present invention provides a system for generating a power startup plan based on semantic clustering, comprising:

[0151] The acquisition module 301 is used to obtain the training startup plan text and perform vector conversion processing on the training startup plan text using a preset startup plan model to obtain multiple startup sentence vectors;

[0152] Semantic clustering module 302, used to perform semantic clustering processing on each startup sentence vector to obtain corresponding clustering results;

[0153] A construction module 303 is used to construct an electric power knowledge graph based on the clustering results;

[0154] The generation module 304 is used to input the information of the device to be started into the power knowledge graph when receiving the information of the device to be started, and obtain the power startup plan corresponding to the information of the device to be started.

[0155] Furthermore, the acquisition module 301 includes:

[0156] The segmentation submodule is used to segment the training startup program text into sentences to obtain multiple startup sentences;

[0157] The word segmentation submodule is used to perform word segmentation on each startup sentence to obtain the word segmentation results corresponding to each startup sentence;

[0158] The word vector submodule is used to input the word segmentation results corresponding to each startup sentence into the preset startup scheme model to obtain multiple word vectors;

[0159] The weighted submodule is used to perform weighted averaging on the word vectors associated with each startup sentence to obtain multiple startup sentence vectors.

[0160] Furthermore, the semantic clustering module 302 includes:

[0161] The cosine similarity submodule is used to calculate the cosine similarity between any two starting sentence vectors using a preset cosine similarity function;

[0162] A first analysis submodule is used to determine whether each cosine similarity is greater than or equal to a preset similarity threshold;

[0163] When the cosine similarity is greater than or equal to the similarity threshold, the starting sentence vector associated with the cosine similarity is deduplicated to obtain a deduplicated sentence vector;

[0164] Based on the preset clustering algorithm, each deduplication sentence vector is clustered to obtain the corresponding clustering results.

[0165] Furthermore, the construction module 303 includes:

[0166] The extraction submodule is used to obtain the power language entities in the clustering results based on the pre-trained entity recognition model;

[0167] The association submodule is used to construct an association graph with the relevant power language entities by taking each type of equipment information of the clustering results as a node;

[0168] The combination submodule is used to combine all related graphs to generate the power knowledge graph.

[0169] Furthermore, the training process of the entity recognition model is as follows:

[0170] Obtaining power startup training text, classifying and labeling the power startup training text according to a preset classification order to obtain a startup training data set, where the classification order is step number, device name, operation step, device status, and operation requirements;

[0171] Input the startup training data set into the preset sequence labeling model for training to obtain the power language entity classification results;

[0172] Calculate the training loss value of the startup training dataset based on the power language entity classification results;

[0173] When the training loss value is greater than or equal to the preset standard loss value, the preset optimizer is used to adjust the model parameters of the sequence labeling model until the training loss value is less than the standard loss value;

[0174] When the training loss value is less than the standard loss value, an entity recognition model is generated.

[0175] Furthermore, it also includes:

[0176] Dynamic update module, used to obtain new entities in real time, and select nodes associated with the new entities as similar nodes from the power knowledge graph based on the phonetic code algorithm;

[0177] Calculate the character similarity between the newly added entity and the similar node, and determine whether the character similarity is greater than a preset character similarity threshold;

[0178] If the character similarity is greater than the character similarity threshold, the newly added entity is merged with the similar node in the electric power knowledge graph to obtain an updated electric power knowledge graph;

[0179] If the character similarity is less than or equal to the character similarity threshold, the new entity is added as a new node to the electric power knowledge graph to obtain an updated electric power knowledge graph.

[0180] See also Figure 6 , Figure 6 This is a structural block diagram of a computer device provided in Example 4 of the present invention.

[0181] An electronic device according to an embodiment of the present invention includes: a memory 401 and a processor 402, wherein the memory 401 stores a computer program; when the computer program is executed by the processor 402, the processor 402 executes a power startup scheme generation method based on semantic clustering as in any of the above embodiments.

[0182] Memory 401 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, hard disk, or ROM. Memory 401 has storage space 403 for program code 413 for executing any of the method steps described above. For example, storage space 403 for program code may include individual program codes 413 for implementing various steps in the method described above. These program codes may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards, or floppy disks. The program codes may be compressed, for example, in a suitable format. When executed by a processing device, these codes cause the processing device to execute the various steps in the method described above. These program codes may be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards, or floppy disks. The program codes may be compressed, for example, in a suitable format. When these codes are executed by a computing and processing device, they cause the computing and processing device to execute the steps of the above-described method for generating a power-on plan based on semantic clustering.

[0183] The fifth embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for generating a power startup plan based on semantic clustering as described in any of the above embodiments is implemented.

[0184] Embodiment 6 of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the power startup scheme generation method based on semantic clustering as described in any of the above embodiments.

[0185] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0186] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0187] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0188] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0189] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part 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 for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0190] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for generating a power startup plan based on semantic clustering, characterized in that: include: Obtaining a training startup scenario text, and performing vector conversion processing on the training startup scenario text using a preset startup scenario model to obtain multiple startup sentence vectors; Performing semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results; Constructing an electric power knowledge graph based on the clustering results; When information about a device to be started is received, the information about the device to be started is input into the power knowledge graph to obtain a power startup solution corresponding to the information about the device to be started.

2. The method for generating a power startup plan based on semantic clustering according to claim 1, characterized in that: The step of performing vector conversion processing on the training startup scenario text using a preset startup scenario model to obtain multiple startup sentence vectors includes: Segmenting the training startup program text to obtain a plurality of startup sentences; Perform word segmentation processing on each of the startup sentences respectively to obtain word segmentation results corresponding to each of the startup sentences; Input the word segmentation results corresponding to each of the startup sentences into a preset startup scheme model to obtain multiple word vectors; The word vectors associated with each of the startup sentences are weighted averaged to obtain multiple startup sentence vectors.

3. The method for generating a power startup plan based on semantic clustering according to claim 1, characterized in that: The step of performing semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results includes: Calculating the cosine similarity between any two of the startup sentence vectors using a preset cosine similarity function; Determine whether each cosine similarity is greater than or equal to a preset similarity threshold; When the cosine similarity is greater than or equal to the similarity threshold, performing a deduplication operation on the startup sentence vector associated with the cosine similarity to obtain a deduplication sentence vector; Clustering is performed on each of the deduplicated sentence vectors based on a preset clustering algorithm to obtain corresponding clustering results.

4. The method for generating a power startup plan based on semantic clustering according to claim 1, characterized in that: The step of constructing a power knowledge graph based on the clustering results includes: Acquire the power language entities in the clustering results according to a pre-trained entity recognition model; Each type of equipment information of the clustering result is used as a node to construct an association graph with the relevant power language entity; Combine all related graphs to generate the power knowledge graph.

5. The method for generating a power startup plan based on semantic clustering according to claim 4, characterized in that: The training process of the entity recognition model is specifically as follows: Obtaining a power startup training text, and classifying and labeling the power startup training text according to a preset classification order to obtain a startup training data set, wherein the classification order is step number, device name, operation step, device status, and operation requirements; Inputting the startup training data set into a preset sequence labeling model for training to obtain a power language entity classification result; Calculating the training loss value of the startup training data set according to the power language entity classification result; When the training loss value is greater than or equal to a preset standard loss value, a preset optimizer is used to adjust the model parameters of the sequence labeling model until the training loss value is less than the standard loss value; When the training loss value is less than the standard loss value, an entity recognition model is generated.

6. The method for generating a power startup plan based on semantic clustering according to claim 1, characterized in that: Also includes: Acquire a newly added entity in real time, and select nodes associated with the newly added entity as similar nodes from the power knowledge graph according to a phonetic code algorithm; Calculating the character similarity between the newly added entity and the similar node, and determining whether the character similarity is greater than a preset character similarity threshold; If the character similarity is greater than the character similarity threshold, the newly added entity is merged with similar nodes in the electric power knowledge graph to obtain an updated electric power knowledge graph; If the character similarity is less than or equal to the character similarity threshold, the newly added entity is added as a new node to the electric power knowledge graph to obtain an updated electric power knowledge graph.

7. A power startup plan generation system based on semantic clustering, characterized in that: include: An acquisition module is used to obtain a training startup plan text, and perform vector conversion processing on the training startup plan text through a preset startup plan model to obtain multiple startup sentence vectors; A semantic clustering module, configured to perform semantic clustering processing on each of the startup sentence vectors to obtain corresponding clustering results; A construction module, configured to construct an electric power knowledge graph based on the clustering results; The generation module is used to input the information of the device to be started into the power knowledge graph when receiving the information of the device to be started, and obtain the power startup plan corresponding to the information of the device to be started.

8. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the method for generating a power startup plan based on semantic clustering according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for generating a power startup plan based on semantic clustering according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, wherein the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the power startup plan generation method based on semantic clustering according to any one of claims 1 to 6.

Images