A signal point table checking method and related device

Abstract

The application discloses a signal point table checking method and related device, and relates to the field of power automation. Signal text descriptions in a signal point table are obtained, a plurality of attribute classification models are used to respectively classify and identify the signal text descriptions to obtain attribute recognition results of a plurality of attribute dimensions, and then the attribute recognition results are combined into standard signal descriptions. An named entity recognition model is used to extract equipment entity names from the signal text descriptions, and the signal text descriptions are grouped based on the equipment entity names to obtain a plurality of entity groups. In each entity group, the standard signal descriptions are matched with a preset standard signal set, so that corresponding checking results are generated according to the matching results. The application combines multi-dimensional attribute classification models and named entity recognition technology, realizes automatic and high-precision checking of signal point tables, effectively solves problems such as difficulty of artificial rules in comprehensive coverage, easiness of conflict and strong subjectivity, and avoids signal loss or errors.

Description

Technical Field

[0001] This invention relates to the field of power automation technology, and more specifically, to a method and related apparatus for verifying signal point tables. Background Technology

[0002] Power dispatch automation systems are the core support for ensuring the safe and stable operation of the power grid. Telecontrol point tables, acting as a data bridge connecting substations and the master station, bear the responsibility of transmitting critical monitoring information such as telemetry and telesignaling. With the continuous expansion of the power grid and the in-depth application of integrated dispatching models, the number of controlled stations has increased dramatically, leading to an exponential increase in the number of signal points requiring maintenance and management. In current industry practice, the creation and maintenance of telecontrol point tables mainly rely on maintenance personnel manually selecting, entering, and verifying points according to relevant procedures and specifications. This process typically involves interpreting massive amounts of unstructured text descriptions, converting them into a standard format, and ensuring the integrity and accuracy of the information. Existing technical approaches often employ a combination of manual experience-based judgment and preset keyword matching or rule base configuration, comparing the consistency between the pre-existing point table and the dispatching baseline point table to complete the verification work, thereby maintaining the normal operation of the telecontrol system's three-remote functions.

[0003] However, faced with the ever-growing number of existing signal point tables and the frequent need for new site construction, existing technologies are gradually revealing their limitations in practical applications. Due to the diversity and complexity of signal description methods, rules summarized manually are insufficient to comprehensively cover all scenarios, and conflicts easily arise between rules, resulting in inadequate coverage during the verification process. Furthermore, manual operation is highly subjective, easily leading to omissions and errors in recording, making it difficult to fundamentally guarantee the integrity and standardization of the point tables. This situation not only increases the workload of maintenance personnel but also poses potential risks to the safe and stable operation of the power grid due to the lack of monitoring of important signals. How to achieve efficient, accurate, and automated verification of large-scale signal point tables has become an urgent technical challenge to be solved in this field. Summary of the Invention

[0004] In view of this, the present invention discloses a signal point table verification method and related apparatus to solve the technical problems in the prior art where rules summarized by humans are difficult to fully cover all scenarios, conflicts easily arise between rules, and the strong subjectivity of human operation makes it difficult to guarantee the integrity and standardization of the point table.

[0005] A method for verifying a signal point table, comprising:

[0006] Retrieve the text description of the signals in the signal point table;

[0007] The signal text description is classified and identified by multiple attribute classification models to obtain attribute identification results of multiple attribute dimensions, wherein each attribute classification model corresponds to one attribute dimension.

[0008] The multiple attribute recognition results are combined into a standard signal description corresponding to the signal text description;

[0009] The device entity name is extracted from the signal text description using a named entity recognition model;

[0010] The signal text description is grouped based on the device entity name to obtain multiple entity groups;

[0011] Within each entity group, the standard signal description is matched with a preset set of standard signals, and a corresponding verification result is generated based on the matching result.

[0012] Optionally, the multiple attribute dimensions include voltage level dimension, device type dimension, and signal type dimension;

[0013] Combining multiple attribute recognition results into a standard signal description corresponding to the signal text description includes:

[0014] The attribute identification results of the voltage level dimension, the device type dimension, and the signal type dimension are concatenated and mapped according to a preset naming rule to obtain the standard signal description.

[0015] Optionally, the attribute classification model performs the classification and recognition process including:

[0016] The input signal text description is processed by word embedding and position encoding to obtain the input feature vector;

[0017] The input feature vector is fed into a multilayer transformer network for feature extraction, and a classification feature vector is output.

[0018] The classification feature vector is input into the classifier, and the predicted probability of each candidate category under the attribute dimension is output.

[0019] The candidate category with the highest predicted probability is determined as the attribute recognition result.

[0020] Optionally, the step of extracting device entity names from the signal text description using a named entity recognition model includes:

[0021] The named entity recognition model is used to perform sequence annotation on the signal text description to obtain the annotation sequence;

[0022] The name of the device entity is determined based on the entity boundary information extracted from the labeled sequence.

[0023] Optionally, the step of performing sequence annotation on the signal text description using the named entity recognition model to obtain the annotation sequence includes:

[0024] The signal text description is sequentially labeled using a preset labeling system to generate a labeling sequence. The preset labeling system includes labels for identifying the boundaries of device entity names and labels for identifying device types.

[0025] Correspondingly, determining the device entity name based on the entity boundary information extracted from the labeled sequence includes:

[0026] In the labeled sequence, identify a tag segment consisting of tags used to identify the boundaries of device entity names;

[0027] Extract the character subsequence corresponding to the tag fragment, and concatenate the character subsequence to form the device entity name. At the same time, determine the device type corresponding to the device entity name based on the tag used to identify the device type.

[0028] Optionally, within each entity group, matching the standard signal description with a preset set of standardized signals and generating a corresponding verification result based on the matching result includes:

[0029] The standard signal description within the entity group is matched with the preset standard signal set, and the standard signal description that matches successfully is determined as the standard matching signal;

[0030] The unmatched standard signals in the preset standard signal set are identified as missing signals;

[0031] The standard signal descriptions within the entity group that do not match the preset standard signal set are identified as non-standard signals.

[0032] Optionally, before classifying and recognizing the signal text description using multiple attribute classification models, the signal point table verification method further includes:

[0033] The original signal text in the signal point table is subjected to format unification processing and / or special symbol standardization processing to obtain the initial text description;

[0034] Extract the global highest voltage level from the telemetry and telesignaling signals in the signal point table;

[0035] The highest global voltage level is appended as a prefix to the beginning of the initial text description to generate the signal text description.

[0036] A signal point table verification device, comprising:

[0037] The text description acquisition module is used to acquire the signal text descriptions from the signal point table.

[0038] The classification and recognition module is used to classify and recognize the signal text description through multiple attribute classification models to obtain attribute recognition results for multiple attribute dimensions, wherein each attribute classification model corresponds to one attribute dimension;

[0039] A signal combination module is used to combine multiple attribute recognition results into a standard signal description corresponding to the signal text description;

[0040] The extraction module is used to extract device entity names from the signal text description using a named entity recognition model;

[0041] The grouping module is used to group the signal text description based on the device entity name to obtain multiple entity groups;

[0042] The verification module is used to match the standard signal description with a preset set of standard signals within each entity group, and generate a corresponding verification result based on the matching result.

[0043] A computer-readable storage medium storing at least one instruction that, when executed by a processor, implements the signal point table verification method described above.

[0044] An electronic device, comprising: a memory and a processor;

[0045] The memory is used to store at least one instruction;

[0046] The processor is used to execute the at least one instruction to implement the signal point table verification method described above.

[0047] As can be seen from the above technical solution, the present invention discloses a signal point table verification method and related apparatus. The method involves obtaining signal text descriptions from a signal point table, classifying and recognizing the signal text descriptions using multiple attribute classification models to obtain attribute recognition results for multiple attribute dimensions, and then combining these multiple attribute recognition results into a standard signal description. Based on this, a named entity recognition model is used to extract device entity names from the signal text descriptions, and the signal text descriptions are grouped based on these device entity names to obtain multiple entity groups. Subsequently, within each entity group, the standard signal description is matched with a preset set of standard signals, thereby generating a corresponding verification result based on the matching result. This invention introduces a multi-dimensional attribute classification model to achieve fine-grained semantic parsing of unstructured signal text, decomposing the originally complex joint classification task into independent sub-tasks, reducing the difficulty of model training and improving generalization ability. Combined with named entity recognition technology, it automatically extracts and groups device entity names, ensuring that the verification process can be carried out on a per-device basis, avoiding omissions caused by differences in expression in traditional rule matching. By automating the comparison of standardized descriptions and specification sets within groups, it effectively solves the technical problems in existing technologies where manually summarized rules cannot fully cover all scenarios, rules are prone to conflict, and the subjectivity of manual operation makes it difficult to guarantee the integrity and standardization of the point table. Therefore, it avoids signal loss or errors caused by human verification oversights, improving the efficiency, accuracy, and standardization level of remote point table creation in power dispatch automation systems. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the published drawings without creative effort.

[0049] Figure 1 This is a flowchart of a signal point table verification method disclosed in an embodiment of the present invention;

[0050] Figure 2 This is a schematic diagram of the structure of a signal point table verification device disclosed in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] This invention addresses the challenges of creating remote control service point (RCS) tables for newly built substations and verifying the completeness and standardization of RCS tables for existing substations. It employs natural language processing technologies, including Transformer, BERT (Bidirectional Encoder Representations from Transformers), and TinyBERT, for text classification and named entity extraction. See also... Figure 1 The present invention discloses a flowchart of a signal point table verification method, which includes:

[0054] Step S101: Obtain the signal text description from the signal point table.

[0055] Signal text descriptions refer to strings of text used in substations or centralized control systems to characterize the semantic information of telemetry, remote signaling, and other signal points. A signal point refers to a measurable state or data point on a remotely monitored physical device or during its operation. A typical signal point table for a single substation includes telemetry signals, remote signaling signals, and remote control signals. Telemetry signals are analog quantities, while remote signaling and remote control signals are digital quantities. Each signal point contains a digitally sequentially encoded point number and a corresponding text description. The actual signal text descriptions vary across different substations, bays, and equipment.

[0056] Signal text descriptions typically originate from the configuration database of the substation telecontrol system or the point table file of the dispatch automation system. The data format can be a comma-separated values ​​(CSV) file, an extensible markup language (XML) file, or an Excel spreadsheet. For example, the signal name or description field column can be extracted from the telecontrol configuration table exported from the substation as the signal text description; alternatively, it can be obtained by parsing the common description information corresponding to the information body address in the IEC104 protocol messages transmitted through the dispatch data network. The obtained signal text description serves as the raw input data for subsequent semantic analysis and verification processing, and its content usually includes unstructured text information such as voltage level, equipment bay, equipment type, and signal attributes.

[0057] Step S102: Classify and identify the signal text description using multiple attribute classification models to obtain attribute identification results for multiple attribute dimensions.

[0058] Attribute classification models are algorithmic models used to map input text to a predefined set of categories. Their core function is to extract structured semantic features from unstructured text. Each attribute classification model corresponds to one attribute dimension.

[0059] In this embodiment, a multi-model decoupling strategy is adopted, that is, independent classification models are constructed for different semantic aspects of the signal description. Attribute dimensions can refer to specific semantic categories contained in the signal description. Multiple attribute dimensions can include: voltage level dimension, equipment type dimension, and signal type dimension. Specifically, the voltage level dimension characterizes the voltage level of the electrical location of the signal, covering typical voltage levels such as 380V, 10kV, 35kV, 66kV, 110kV, 220kV, or 500kV; the equipment type dimension characterizes the physical equipment category to which the signal belongs, including main transformers, busbars, lines, capacitor banks, reactor banks, bus branch / bus tie, station service transformers, or DC systems; the signal type dimension characterizes the specific measurement and control type or physical quantity attribute of the signal, such as phase A voltage, phase B current, and active power in telemetry, or protection action signals and switch position signals in remote signaling. By decomposing the complex signal description into the above three independent dimensions, the complexity of the classification task can be effectively reduced. Compared to directly classifying all signal names, this dimensionality partitioning method significantly reduces the number of categories required by a single classification model, thereby reducing the need for a large number of training samples and improving the model's generalization ability and robustness.

[0060] It should be noted that in signal standards and specifications, voltage level and equipment type can form various combinations, such as 500kV main transformer, 220kV main transformer, 110kV main transformer, 35kV main transformer, etc. By separating voltage level and equipment type into two independent classification dimensions, instead of setting separate categories for each "voltage level + equipment type" combination (e.g., 500kV main transformer as one category, 220kV main transformer as another category), it can now be uniformly classified into a single equipment type category of "main transformer." This significantly reduces the number of categories in the text classifier, lowers the difficulty of model training, and reduces the requirement for a large number of training samples.

[0061] The attribute classification model is constructed during the offline training phase: First, a large number of signal text description samples are collected from historical signal point tables, and each attribute dimension is manually labeled to form a training dataset for that dimension. Second, based on deep learning frameworks, such as Bidirectional Encoder Representations from Transformer (BERT) or Text Convolutional Neural Networks (TextCNN) as the basic network architecture, the training data is input into the network for parameter learning, and the model weights are optimized through backpropagation until the model's classification accuracy on the validation set reaches a preset standard. Finally, the trained model parameter file is saved and loaded for use during the inference phase. In actual operation, the attribute classification model is configured as follows: First, the input signal text description is segmented and vectorized, converting it into numerical feature vectors that the model can recognize. Then, forward computation is performed through the neural network layers within the model to output the probability distribution of each candidate category. Finally, the category with the highest probability is selected as the attribute recognition result for that attribute dimension. By processing multiple models in parallel, structured labels for signal text descriptions across multiple dimensions can be obtained simultaneously.

[0062] Step S103: Combine the multiple attribute recognition results into a standard signal description corresponding to the signal text description.

[0063] A standard signal description can refer to a standardized string generated by reconstructing the original signal text description according to preset naming conventions or formatting rules. This step aims to integrate the extracted discrete attribute tags into descriptive information that conforms to industry standards or a unified internal system format.

[0064] The combination process of standard signal descriptions is performed according to preset naming rules. For example, the attribute identification results of the voltage level dimension, equipment type dimension, and signal type dimension are concatenated and mapped according to preset naming rules to obtain standard signal descriptions. Among them, preset naming rules are string combination templates built based on the signal naming specifications commonly used in the power industry, which define the order and connection method of each attribute identification result in the final description.

[0065] Specifically, the attribute identification results of each attribute dimension are connected in series using connectors (such as underscores or hyphens) according to the order of voltage level, equipment type, and signal type. For example, if the voltage level identification result is "10kV", the equipment type identification result is "line", and the signal type identification result is "A-phase current", then the combined standard description is "10kV ×× line A-phase current". Another example is if the voltage level identification result is "10kV", the equipment type identification result is "switch", and the signal type identification result is "trolley working position", then the combined standard description is "10kV ×× switch trolley working position". Yet another example is if the voltage level identification result is "110kV", the equipment type identification result is "busbar", and the signal type identification result is "bus differential protection operation", then the combined standard description is "110kV ×M busbar bus differential protection operation".

[0066] This deterministic splicing mapping can transform non-standardized actual signal descriptions into semantically clear and formatted standardized signal identifiers, thereby eliminating writing differences and redundant information in the original text. This provides the necessary data foundation for subsequent accurate matching with a pre-defined set of standardized signals, ensuring the consistency and accuracy of the verification process.

[0067] Step S104: Extract the device entity name from the signal text description using the named entity recognition model.

[0068] Named Entity Recognition (NER) models can refer to sequence labeling models specifically designed to identify entities in text that have specific meanings, such as names of people, places, and organizations.

[0069] In this application, the named entity recognition model is specifically used to locate and extract entity names representing specific physical devices from signal text descriptions.

[0070] The process of building a named entity recognition model includes: collecting a corpus of signal text containing device names, and performing sequence labeling on each character in the corpus, with labels including B-DEV (start of device entity), I-DEV (inside device entity), and O (non-entity); subsequently, using a Conditional Random Field (CRF) or Long Short-Term Memory (LSTM) network combined with a CRF architecture to train the model and learn the dependency relationship between character context and labels.

[0071] During the operational phase, the named entity recognition model receives signal text descriptions as input, labels each character in the text, and outputs a tag sequence of the same length as the input text. The system then parses this tag sequence, determining the boundaries of character subsequences based on the tag type. For example, it extracts character segments consecutively labeled B-DEV and I-DEV and concatenates them into complete device entity names. For instance, a device entity name might be: {<actual signal description, standard signal description>}, where each device entity name contains a set of actual signals. As another example, given the input text "#1 main transformer high-voltage side switch temperature," the model can identify and extract "#1 main transformer" or "switch" as the device entity name. This process accurately identifies key equipment objects from free text, providing a crucial basis for subsequent signal grouping.

[0072] In practical applications, the name of the device entity to which the signal belongs can be inferred from the input signal text description, so that signals with the same device entity name can be grouped into the same group.

[0073] For example, a telemetry meter for a certain power plant contains the following 12 point descriptions: “10kV×Yun Line 036IA”, “10kV×Yun Line 036IB”, “10kV×Yun Line 036IC”, “10kV×Yun Line 036P”, “10kV×Yun Line 036Q”, “10kV×Yun Line 036COS”, “10kV×Route 037IA”, “10kV×Route 037IB”, “10kV×Route 037IC”, “10kV×Route 037P”, “10kV×Route 037Q”, and “10kV×Route 037COSΦ”. Using an interval entity extraction model, “×Yun Line” and “×Route” can be automatically extracted from the descriptions.

[0074] For example, the telemetry signals of a certain power plant include "the telemetry signal value of the non-electrical quantity protection action of the No. 1 main transformer", and the entity extraction result is "the No. 1 main transformer"; "the telemetry signal value of the 1121 disconnect switch on the high voltage side of the No. 2 main transformer", and the entity extraction result is "the No. 2 main transformer"; "the telemetry signal value of the zero-sequence overcurrent stage I action of the 10kV Tianshui 051 line", and the entity extraction result is "the Tianshui 051 line"; "the telemetry signal value of the protection action of the 10kV station service transformer 002", and the entity extraction result is "the station service transformer 002".

[0075] Named Entity Extraction (NER) is an automatic sequence labeling task that uses the BIOES tagging system to label sequences. Here, B (Begin) indicates the start, I (Intermediate) indicates the middle, E (End) indicates the end, S (Single) indicates a single character (not applicable in this scenario), and O (Other) indicates other irrelevant characters. B, I, and E tags are further subdivided according to equipment type, including busbars, bus couplers, main transformers, station service transformers, grounding transformers, lines, capacitors, reactors, etc.

[0076] For example, the sequence label for "Chengnan I Circuit Phase A Current Value of 110kV×× Substation" is [O, O, O, O, O, O, O, O, B-line, I-line, I-line, E-line, O, O, O, O, O]; the sequence label for "Measured Winding Temperature of No. 2 Main Transformer of 110kV×× Substation" is [O, O, O, O, O, O, O, O, B-main transformer, I-main transformer, I-main transformer, E-main transformer, O, O, O, O, O].

[0077] Step S105: Group the signal text description based on the device entity name to obtain multiple entity groups.

[0078] Entity grouping refers to the process of logically grouping signal text descriptions belonging to the same physical device or the same device interval. This step uses the device entity names extracted in the previous step as grouping keys. The specific execution logic includes: traversing all signal text descriptions and their corresponding device entity names to establish a mapping relationship between device names and signal lists; grouping signal text descriptions with the same or semantically equivalent device entity names into the same set; and assigning complex descriptions containing multiple device entity names to the group corresponding to the master device based on preset priority rules or master-slave relationships.

[0079] By grouping entities, the originally scattered point table data is reconstructed into a tree or network structure centered on the devices. For example, all signal descriptions containing the name of the #1 main transformer entity, regardless of their original text format differences, are grouped into the #1 main transformer entity group. This structured organization can simulate the topological relationships of actual physical devices, allowing subsequent verification operations to be performed within the same device context, avoiding incorrect matching across devices, thereby improving the accuracy and logical consistency of the verification.

[0080] Step S106: Within each entity group, the standard signal description is matched with a preset set of standard signals, and a corresponding verification result is generated based on the matching result.

[0081] In this embodiment, the verification process is performed independently within each independent entity group to ensure the specificity of the verification.

[0082] A pre-defined set of standard signals can refer to a library of standard signals predefined according to power grid dispatching regulations, equipment operation specifications, or centralized control system monitoring requirements. This library stores descriptions of the standard signals that each piece of equipment should possess under normal operating conditions, along with their corresponding attribute information. This set can be constructed in ways such as: compiling a list of mandatory monitoring signals stipulated in the dispatching regulations of various regions, standardizing it, and then entering it into a database; or generating a benchmark library based on a historically well-performing benchmark site's signal point list after cleaning and deduplication.

[0083] This embodiment matches standard signal descriptions with a preset set of standard signals to generate verification results that include signal point numbers, original descriptions, standard descriptions, matching status, and anomaly types. This refined comparison within groups allows for rapid identification of missing, incorrect, or multiple items in the point table, providing maintenance personnel with an intuitive basis for rectification.

[0084] In summary, this invention discloses a signal point table verification method. It obtains signal text descriptions from the signal point table, classifies and identifies the signal text descriptions using multiple attribute classification models to obtain attribute recognition results for multiple attribute dimensions, and then combines these multiple attribute recognition results into a standard signal description. Based on this, it extracts device entity names from the signal text descriptions using a named entity recognition model, and groups the signal text descriptions based on the device entity names to obtain multiple entity groups. Subsequently, within each entity group, the standard signal description is matched with a preset set of standard signals, thereby generating a corresponding verification result based on the matching results. This invention introduces a multi-dimensional attribute classification model to achieve fine-grained semantic parsing of unstructured signal text, decomposing the originally complex joint classification task into independent sub-tasks, reducing the difficulty of model training and improving generalization ability. Combined with named entity recognition technology, it automatically extracts and groups device entity names, ensuring that the verification process can be carried out on a per-device basis, avoiding omissions caused by differences in expression in traditional rule matching. By automating the comparison of standardized descriptions and specification sets within groups, it effectively solves the technical problems in existing technologies where manually summarized rules cannot fully cover all scenarios, rules are prone to conflict, and the subjectivity of manual operation makes it difficult to guarantee the integrity and standardization of the point table. Therefore, it avoids signal loss or errors caused by human verification oversights, improving the efficiency, accuracy, and standardization level of remote point table creation in power dispatch automation systems.

[0085] In practical applications, corresponding classification models can be set up for voltage level, equipment type, and signal type.

[0086] Voltage level classification models are used to infer the voltage level of a signal based on the input location text description, such as 380V, 10kV, 35kV, 66kV, 110kV, 220kV, 500kV, etc. Classification models can employ Transformer, BERT, TinyBERT, etc. The input tokens are embedded using embedding and positional encoding, transformed through multiple Transformer layers, and output as CLS features. These features are then passed through a classifier HEAD, which outputs the probability of each class. The class with the highest probability is selected as the predicted class.

[0087] Equipment type classification models are used to infer the equipment type of a signal based on the input location text description, such as main transformer, busbar, line, capacitor bank, reactor bank, bus branch / bus tie, station service transformer, DC system, etc. Classification models can also use BERT, etc. In signal standards and specifications, voltage level and equipment type can be combined, such as 500kV#×main transformer, 220kV#×main transformer, 110kV#×main transformer, 35kV#×main transformer. By separating voltage level and equipment type into two independent classification dimensions, the number of categories in the text classifier can be reduced. For example, although the above four combinations belong to different voltage levels, they all belong to the main transformer equipment type.

[0088] The signal type classification model is used to infer the corresponding standard type of the actual signal based on the input point location text description. For example, the point signal descriptions in telemetry, such as "220kV bus tie 210 measurement and control_Ua" and "220kV ××Ⅱ circuit 208Ua", both correspond to "A-phase voltage". The former belongs to the bus tie equipment type, while the latter belongs to the line equipment type. Similarly, the point signal descriptions, such as "110kV_×× transformer No. 2 capacitor A-phase current value" and "110kV_×× transformer East × line A-phase current amplitude", both correspond to "A-phase current". The former belongs to the capacitor bank equipment type, while the latter belongs to the line equipment type. After separating the voltage level and equipment type, there is no need to distinguish the differences between voltage level and equipment type, thus reducing the number of types corresponding to the actual signal. Common telemetry types include A-phase voltage, B-phase voltage, C-phase voltage, AB line voltage, BC line voltage, CA line voltage, zero-sequence voltage, A-phase current, B-phase current, C-phase current, zero-sequence current, neutral point current, active power, reactive power, and power factor. There are many more types of telemetry signaling.

[0089] The standard description of a real signal can be formed by combining the outputs of three classification models: voltage level, equipment type, and signal type. For example, "110kV_×× substation east × line A-phase current amplitude", the three classification models identify 110kV, line, and A-phase current, respectively, and the outputs can be combined to form the signal description "110kV ×× line A-phase current" in the standard specification. For example, "35kV No.1 capacitor 361 test_Ia", the three classification models identify 35kV, capacitor bank, and A-phase current, respectively, and the outputs can be combined to form the signal description "35kV ×× capacitor bank A-phase current" in the standard specification.

[0090] In one embodiment, the process of an attribute classification model performing classification and recognition may include:

[0091] The input signal text description is processed by word embedding and position encoding to obtain the input feature vector;

[0092] The input feature vector is fed into a multilayer transformer network for feature extraction, and a classification feature vector is output.

[0093] The classification feature vector is input into the classifier, and the predicted probability of each candidate category under the attribute dimension is output.

[0094] The candidate category with the highest predicted probability is determined as the attribute recognition result.

[0095] In this embodiment, word embedding aims to map discrete text tokens (lexical units) into continuous real-valued vector representations to capture semantic relationships between words. Using a pre-trained word vector matrix or a randomly initialized lookup table, each character or word in the signal text description is converted into a fixed-dimensional vector. Since the Transformer architecture itself lacks the ability to distinguish the order of token positions in a sequence, positional encoding is used to inject positional information into the word embedding vector, enabling the model to understand the order of keywords in the signal description. Based on a preset positional encoding function, positional vectors with the same dimension as the word embeddings are generated, and positional information is fused into the word embeddings through vector addition, forming an input feature vector that combines semantic and positional features. This input feature vector constitutes the basic data structure for subsequent deep neural network processing, representing the initial state of the signal text in the original semantic space.

[0096] The multilayer transformer network consists of stacked encoder modules, each primarily containing a self-attention mechanism and a feedforward neural network. The self-attention mechanism calculates the correlation weights between any two tokens in the input sequence and captures long-range dependencies in the signal text description through a weighted summation operation, such as the pairing pattern between voltage level and device type, thus achieving effective aggregation of global contextual information. The feedforward neural network performs a non-linear transformation on the vector representation at each position to further extract higher-order semantic features. Through layer-by-layer propagation and abstraction within the multilayer transformer network, the input feature vector is transformed into a classification feature vector containing deep semantic information. This classification feature vector is a highly condensed and abstracted representation of the input signal text description; its dimensionality is typically related to the length of the input sequence and the dimension of the hidden layers, providing rich feature basis for subsequent classification decisions.

[0097] The classifier is connected to the output of the multilayer transformer network, and its structure typically includes fully connected layers and activation functions. The classifier receives the classification feature vector output from the multilayer transformer network and maps it to a vector space of all predefined candidate categories under the current attribute dimension through a linear transformation. Subsequently, using activation functions such as the normalized exponential function or the sigmoid function, the mapped vector values ​​are converted into a probability distribution, i.e., the predicted probabilities. This set of predicted probabilities reflects the likelihood that the input signal text description belongs to each candidate category under the current attribute dimension, covering all potential options for specific classification targets such as voltage level, device type, or signal type.

[0098] After obtaining the predicted probabilities of all candidate categories, a maximum value selection algorithm is used to select the candidate category corresponding to the highest probability value from the probability distribution. This candidate category is the model's best prediction result for the current signal text description in a specific attribute dimension and is ultimately determined as the attribute identification result. Based on the probability maximization decision mechanism, the attribute classification model can output deterministic classification labels, such as identifying a signal description as 110kV, line, or A-phase current, thus completing the transformation from raw text to structured attributes and providing accurate dimensional attribute input for subsequent construction of standard signal descriptions.

[0099] As a preferred implementation, the attribute classification model performs word embedding and positional encoding on each word in the input signal text description to generate an input feature vector; the input feature vector is then fed into a multilayer transformer network for feature extraction, outputting a classification feature vector that represents global semantic information, such as CLS (Classification Token) features; the classification feature vector is then fed into the classification head, i.e., the classifier head, which, after processing by a fully connected layer and activation function, outputs the predicted probability of each candidate category under the current attribute dimension; the candidate category with the highest predicted probability is determined as the attribute recognition result.

[0100] This application utilizes deep neural networks to automatically learn the deep semantic features of power signal text, overcoming the problems of poor generalization and low coverage of traditional rule-based methods, and improving the classification accuracy and robustness in complex power scenarios.

[0101] In one embodiment, the process of extracting device entity names from the signal text description using a named entity recognition model may specifically include:

[0102] The named entity recognition model is used to perform sequence annotation on the signal text description to obtain an annotation sequence.

[0103] The name of the device entity is determined based on the entity boundary information extracted from the labeled sequence.

[0104] This embodiment inputs a signal text description into a named entity recognition model, which then performs sequence-level annotation on each character or term in the text, thereby generating a labeled sequence. Essentially, this process transforms unstructured text strings into structured label sequences, revealing hidden semantic units within the text through the distribution patterns of these labels. Upon receiving input text such as 10kV×Yunxian 036IA or the remote signaling value of non-electrical protection action of the No. 1 main transformer, the model assigns a specific label to each position in the text sequence based on pre-learned feature patterns. These labels characterize which semantic entity the character at that position belongs to, or whether it belongs to non-entity content.

[0105] In engineering implementation, named entity recognition models can use pre-trained language models as their infrastructure, such as bidirectional encoder representation transformers or their lightweight variant TinyBERT. These models capture deep features of textual context through multi-layer transformer networks. The method provided in this application embodiment can, for example, use the BIOES annotation system to annotate sequences, where the B (Begin) label indicates the start of an entity, the I (Intermediate) label indicates the middle part of an entity, the E (End) label indicates the end of an entity, the S (Single) label indicates a single-character entity (usually not involved in this scenario), and the O (Other) label indicates other characters not belonging to the target entity. For power equipment scenarios, the B, I, and E labels are further associated with specific equipment types, such as B-main transformer, I-main transformer, E-main transformer, or B-line, I-line, E-line, etc. Through this fine-grained annotation, the model can accurately distinguish and locate different types of equipment entities.

[0106] Taking "Measured winding temperature of 110kV ×× transformer No. 2" as an example, the labeling sequence output by the named entity recognition model is [O, O, O, O, O, O, O, O, B-transformer, I-transformer, I-transformer, E-transformer, O, O, O, O, O]. In this sequence, the first eight characters are labeled as O, indicating that they do not belong to the target entity name; the subsequent "No. 2 transformer" is labeled as B-transformer, I-transformer, I-transformer in sequence, and the word "transformer" is labeled as E-transformer. This continuous label sequence clearly defines the boundary and category of the entity segment "No. 2 transformer". For example, the labeling sequence for "110kV ×× Substation Chengnan I Circuit Phase A Current Value" is [O, O, O, O, O, O, O, O, B-line, I-line, I-line, E-line, O, O, O, O, O], thus accurately extracting the entity "Chengnan I Circuit". Similarly, the labeling sequence for "10kV ×Yun Line 036IA" is [O, O, O, O, B-line, I-line, I-line, E-line, O, O, O, O], which extracts the entity "×Yun Line". Through this sequence labeling mechanism, the originally messy text description is transformed into structured data with clear semantic boundary information, providing a direct algorithmic basis for subsequent entity name extraction.

[0107] The generated labeled sequence is decoded and parsed. Based on the entity boundary information contained in the labels, corresponding text fragments are extracted from the original signal text description to determine the final device entity name. This process is a crucial step in restoring the abstract label sequence output by the model into a human-readable entity string.

[0108] The specific parsing logic is as follows: A specific label pattern is scanned in the label sequence to locate the start and end positions of entities. When a B-category label is detected, the start of an entity is marked; subsequent positions are then checked, and if an I-category label appears, the entity is considered to be continuing; this continues until an E-category label is encountered, at which point the entity is marked as ending. All the original characters between the start and end positions are concatenated sequentially to form the entity's name. If an S-category label appears in the label sequence, the single character corresponding to that position directly constitutes the entity name.

[0109] Based on the example of the above steps, for the labeled sequence [O, O, O, O, O, O, O, O, B-main transformer, I-main transformer, I-main transformer, E-main transformer, O, O, O, O, O], first, the B-main transformer label with index 8 is located, determining the start of the entity; then, the I-main transformer labels with indices 9 and 10 confirm the continuation of the entity; finally, the E-main transformer label is encountered at index 11, determining the end of the entity. Accordingly, the character substring "2-main transformer" corresponding to indices 8 to 11 is extracted from the original text "110kV×× transformer No. 2 main transformer winding temperature measured value", and this is determined as the equipment entity name corresponding to the signal text description. Similarly, for “110kV×× Substation Chengnan I Circuit Phase A Current Value”, its labeling sequence is [O, O, O, O, O, O, O, O, B-line, I-line, I-line, E-line, O, O, O, O, O], and “Chengnan I Circuit” can be extracted as the equipment entity name; for “10kV×Yun Line 036IA”, the labeling sequence is [O, O, O, O, B-line, I-line, I-line, E-line, O, O, O, O], and “×Yun Line” can be extracted as the equipment entity name.

[0110] This embodiment, through tag boundary information-based extraction, can accurately locate device entities from non-standardized and diverse signal descriptions, regardless of whether the entity is located at the beginning, middle, or end of the text, or whether it contains interfering information such as voltage levels, numerical codes, or signal types. This mechanism ensures the accuracy and robustness of device entity name extraction, avoids misjudgments that may occur based on simple keyword matching, and thus provides high-quality grouping keys for subsequent signal grouping according to device entity names.

[0111] By combining sequence labeling and boundary extraction, this method enables the automatic identification and structured extraction of key equipment information from free text. Compared to methods relying on manual rules or regular expressions, this approach can adapt to complex variations and implicit patterns in substation equipment naming, significantly improving the automation level and accuracy of the pre-processing stage for signal point table verification.

[0112] In one embodiment, the process of performing sequence annotation on the signal text description using the named entity recognition model to obtain the annotation sequence specifically includes:

[0113] The signal text description is sequentially labeled using a preset labeling system to generate a labeling sequence. The preset labeling system includes labels for identifying the boundaries of device entity names and labels for identifying device types.

[0114] The construction of the pre-defined labeling system directly determines the accuracy and localization capability of the named entity recognition model in parsing semantic information. This system employs a composite labeling architecture based on location and category, such as the BIOES (Begin, Intermediate, End, Single) labeling pattern, rather than a simple binary classification. Labels used to identify the boundaries of device entity names (such as B, I, E, S) are responsible for defining the start and end range of the entity within the character sequence, ensuring the integrity of the entity fragment. Labels used to identify device types are embedded in location labels, forming combinations such as B-line, I-main transformer, E-capacitor. This design allows the model to simultaneously complete the dual decision-making process of entity boundary determination and type attribution when outputting the labeled sequence. In the specific construction process, historical data describing substation monitoring signals is first collected and manually labeled. Then, a set of equipment types is defined according to business needs, covering objects monitored by phasor measurement units (PMUs) and supervisory control and data acquisition (SCADA) systems, such as main transformers, buses, lines, capacitors, and reactors. Finally, location markers are combined with equipment type labels to generate a complete set of labels covering all scenarios for model training. Through this refined labeling system, each character in the output labeled sequence is assigned a label value with clear semantic meaning.

[0115] Correspondingly, the process of determining the device entity name based on the entity boundary information extracted from the labeled sequence specifically includes:

[0116] In the labeled sequence, identify a tag segment consisting of tags used to identify the boundaries of device entity names;

[0117] Extract the character subsequence corresponding to the tag fragment, and concatenate the character subsequence to form the device entity name. At the same time, determine the device type corresponding to the device entity name based on the tag used to identify the device type.

[0118] This embodiment reconstructs entity objects based on the generated annotation sequence by parsing the tag distribution pattern. Tag segments consist of consecutive tags with specific location attributes and the same device type attribute. For example, consecutive B-line, I-line, and E-line tags together constitute a complete tag segment. The process of determining tag segments is essentially pattern recognition based on state transition logic: scanning the annotation sequence, recording the segment start point when a starting tag (such as B-type) is detected, continuously tracking subsequent intermediate tags (I-type) until an ending tag (E-type) or a single-word tag (S-type) is encountered, at which point the segment end point is locked. This mechanism can accurately cut out all tag segments representing independent entities from a lengthy tag sequence, providing an accurate index range for subsequent extraction of corresponding text content.

[0119] Based on the position information of the tag fragments in the original signal text description, the entity name and its type are extracted simultaneously. For each determined tag fragment, its index position in the character-level sequence is obtained, and the characters at the corresponding position are extracted from the original signal text description to form a character subsequence. For example, if the tag fragment corresponds to the 5th to 8th position of the text index, then four characters within that interval are extracted. Subsequently, the extracted continuous character subsequences are concatenated in order to reconstruct the complete equipment entity name, such as Tianshui 051 line. At the same time, since each tag in the tag fragment carries equipment type semantic information, the type information contained in the tag is directly parsed and assigned to the corresponding equipment entity name, thereby determining its equipment type as a line. This tag fragment-based parsing method achieves synchronous locking of the equipment entity name and its type in a single processing step, effectively avoiding data inconsistency problems that may be caused by step-by-step extraction, and providing a high-precision data foundation for subsequent group verification based on equipment type.

[0120] In one embodiment, step S106 may specifically include:

[0121] The standard signal description within the entity group is matched with the preset standard signal set, and the standard signal description that matches successfully is determined as the standard matching signal;

[0122] The unmatched standard signals in the preset standard signal set are identified as missing signals;

[0123] The standard signal descriptions within the entity group that do not match the preset standard signal set are identified as non-standard signals.

[0124] In this embodiment, the preset standard signal set is pre-configured according to the power grid operation safety regulations and the access requirements of the dispatch automation master station, and includes a list of telemetry and remote signaling signal standard names that must be available under specific voltage levels and equipment types.

[0125] The specific matching logic includes: calculating the string similarity or performing exact matching between each standard signal description within an entity group and entries in a preset set of standardized signals; if a completely identical or similar entry with a similarity higher than a preset threshold is found in the preset set of standardized signals, the signal is considered a successful match and marked as a standardized matching signal; if a standard signal description within an entity group does not have a corresponding entry in the preset set of standardized signals, it is considered a non-standard signal, potentially indicating a naming error or redundant signal; conversely, if some standardized entries in the preset set of standardized signals do not have a corresponding standard signal description within the current entity group, the standardized entry is considered a missing signal, indicating a risk of missed monitoring. Finally, the system summarizes the above matching statuses and generates a verification result containing the signal point number, original description, standard description, matching status, and anomaly type. This refined comparison within groups allows for rapid location of missing, incorrect, or multiple entries in the point table, providing maintenance personnel with an intuitive basis for rectification and effectively solving the problems of low efficiency and high error rates associated with manual verification.

[0126] In one embodiment, before classifying and recognizing the signal text description using multiple attribute classification models, the signal point table verification method further includes:

[0127] The original signal text in the signal point table is subjected to format unification processing and / or special symbol standardization processing to obtain the initial text description;

[0128] Extract the global highest voltage level from the telemetry and telesignaling signals in the signal point table;

[0129] The highest global voltage level is appended as a prefix to the beginning of the initial text description to generate the signal text description.

[0130] For the original signal text in the signal point table, a format standardization process is first performed to eliminate text irregularities caused by differences in human input habits. This process includes removing leading and trailing spaces, standardizing full-width and half-width characters, converting English letters to uppercase or lowercase, truncating excessively long text descriptions, and performing word segmentation operations such as kV and GOOSE. Through these cleaning operations, the messy original text can be transformed into a standardized text format that meets the input requirements of subsequent models.

[0131] Building upon the unified format, further standardization of special symbols is performed. In power automation scenarios, the representation of the same physical quantity often exists in multiple heterogeneous forms, such as the mixing of Roman numerals and Arabic numerals (e.g., segment II, segment 2), or the mixing of special characters and Chinese characters (e.g., kV and kilovolt). The standardization of special symbols aims to map these heterogeneous symbols representing the same meaning into a unified internal standard form. Specifically, all Roman numerals or Chinese numerals representing voltage levels are uniformly converted to Arabic numeral format, and various special unit symbols are converted to unified standard characters. This process is not a simple character replacement, but rather semantic alignment based on a pre-defined power terminology dictionary, thereby reducing the processing difficulty for subsequent text classification and named entity recognition models.

[0132] After format standardization and special symbol normalization, noise in the original signal text was effectively filtered out, generating a semantically clear and formatted initial text description. This initial text description, as an intermediate data product, provides the data foundation for subsequent extraction of the global highest voltage level and generation of the final signal text description.

[0133] Using the generated initial text description, the text content corresponding to all telemetry and telesignaling signals in the signal point table is scanned to extract the global highest voltage level. In a power system, a single substation typically contains multiple voltage levels, such as 380V, 10kV, 35kV, 110kV, and 220kV. The global highest voltage level can refer to the highest voltage level involved in all telemetry and telesignaling signals of that substation; it represents the highest voltage level of that substation.

[0134] During the extraction process, a pre-defined database of voltage level keywords (such as 500kV, 220kV, 110kV, etc.) is used to match and identify the initial text description, or regular expressions are used to extract combinations of numbers and units from the text. After identifying the voltage levels corresponding to all signals, the numerical values ​​of each voltage level are compared, and the voltage level with the largest numerical value is determined as the global highest voltage level for the substation. For example, if a substation simultaneously has 10kV line signals and 220kV main transformer signals, then the extracted global highest voltage level is 220kV.

[0135] In this embodiment, after determining the global highest voltage level, it is appended as a prefix to the beginning of each initial text description to generate the final signal text description. Specifically, the extracted global highest voltage level string is concatenated at the beginning of the initial text description to form a new input sequence. For example, if an initial text description is ×Yun Line 036IA, and the global highest voltage level of the substation is 110kV, then the generated signal text description is 110kV×Yun Line 036IA.

[0136] In actual operation, signal descriptions of low-voltage equipment (such as 10kV lines) often lack voltage level prefixes, making it difficult for models to directly determine their voltage level. By forcibly placing the highest global voltage level at the beginning, it's equivalent to labeling all signals with the station's overall voltage level. For subsequent attribute classification models, this prefix helps them more accurately understand the signal's position within the station topology. Especially when distinguishing similar equipment names at different voltage levels (e.g., main transformer No. 1 might exist on both the 220kV and 35kV sides), this prefix provides crucial disambiguation, significantly improving the accuracy and robustness of model classification.

[0137] In summary, this invention achieves deep cleaning and enhancement of the original signal text, constructs a high-quality signal text description containing global voltage level context, and lays a solid data foundation for subsequent accurate classification and recognition using multiple attribute classification models.

[0138] In one embodiment, after obtaining the verification result, the signal point table verification method further includes:

[0139] Generate a verification report containing the standardized matching signal, the missing signal, and the non-standard signal. The verification report includes detailed statistical data by device type and global summary statistical data.

[0140] In response to receiving a manual modification instruction for the verification report, the erroneous entity extraction result or erroneous classification identification result indicated by the manual modification instruction is corrected;

[0141] Based on the corrected results, the device entity name and the standard signal description corresponding to the signal text description are automatically updated in real time.

[0142] When generating the verification report, the system categorizes and summarizes the matching results within each entity group according to three states, generating detailed lists of standardized matching signals, missing signals, and non-standard signals. Based on this, the system counts the quantity of each type of signal according to equipment type such as main transformers, lines, and capacitor banks, and then summarizes the data to generate global statistics for the entire plant, forming a complete verification report containing detailed and summarized data. Since attribute classification models and named entity recognition models cannot guarantee 100% accuracy in actual reasoning—for example, they might mistakenly extract "Main Transformer No. 2" as "Main Transformer No. 1" or misclassify "Phase A Current" as "Phase B Current"—the system allows maintenance personnel to manually correct erroneous entity extraction or classification results after the verification report is output. The system automatically updates the corresponding equipment entity name groups and standard signal descriptions in real time based on the user's corrections, re-establishing the mapping relationship between entity names and the original signal text descriptions. The system then regenerates the verification report with the updated data, automatically saves it after user confirmation, and completes the closed-loop process from verification report generation to manual correction and report update.

[0143] In summary, this invention provides maintenance personnel with an intuitive and comprehensive basis for verification by generating verification reports containing detailed and summary data. Furthermore, it allows maintenance personnel to manually correct erroneous entity extraction or classification results in the report. The system automatically updates the device entity name groupings and standard signal descriptions in real time based on the corrections, automatically regenerating and saving the updated verification report, thus forming a complete closed loop of machine initial screening, manual fine-tuning, and automatic updating. Compared to relying entirely on manual verification of each point table, this invention significantly reduces the workload of maintenance personnel through automated verification, compensates for the insufficient accuracy of automated model recognition by introducing manual inspection and modification, and avoids new errors that may be introduced by secondary manual data entry. This improves the efficiency, accuracy, and standardization of remote control point table verification.

[0144] Corresponding to the above method embodiments, the present invention also discloses a signal point table verification device.

[0145] See Figure 2 The present invention discloses a schematic diagram of a signal point table verification device, which includes:

[0146] The text description acquisition module 201 is used to acquire the signal text description in the signal point table.

[0147] The classification and recognition module 202 is used to classify and recognize the signal text description through multiple attribute classification models to obtain attribute recognition results of multiple attribute dimensions.

[0148] Each attribute classification model corresponds to one attribute dimension.

[0149] The signal combination module 203 is used to combine multiple attribute recognition results into a standard signal description corresponding to the signal text description.

[0150] Extraction module 204 is used to extract device entity names from the signal text description through a named entity recognition model.

[0151] Grouping module 205 is used to group the signal text description based on the device entity name to obtain multiple entity groups.

[0152] The verification module 206 is used to match the standard signal description with a preset set of standard signals within each entity group, and generate a corresponding verification result based on the matching result.

[0153] In summary, this invention discloses a signal point table verification device. It acquires signal text descriptions from a signal point table, uses multiple attribute classification models to classify and identify the signal text descriptions to obtain attribute recognition results for multiple attribute dimensions, and then combines these multiple attribute recognition results into a standard signal description. Based on this, it extracts device entity names from the signal text descriptions using a named entity recognition model, and groups the signal text descriptions based on the device entity names to obtain multiple entity groups. Subsequently, within each entity group, the standard signal description is matched with a preset set of standard signals, thereby generating a corresponding verification result based on the matching result. This invention introduces a multi-dimensional attribute classification model to achieve fine-grained semantic parsing of unstructured signal text, decomposing the originally complex joint classification task into independent sub-tasks, reducing the difficulty of model training and improving generalization ability. Combined with named entity recognition technology, it automatically extracts and groups device entity names, ensuring that the verification process can be carried out on a per-device basis, avoiding omissions caused by differences in expression in traditional rule matching. By automating the comparison of standardized descriptions and specification sets within groups, it effectively solves the technical problems in existing technologies where manually summarized rules cannot fully cover all scenarios, rules are prone to conflict, and the subjectivity of manual operation makes it difficult to guarantee the integrity and standardization of the point table. Therefore, it avoids signal loss or errors caused by human verification oversights, improving the efficiency, accuracy, and standardization level of remote point table creation in power dispatch automation systems.

[0154] In one embodiment, the multiple attribute dimensions include voltage level dimension, device type dimension, and signal type dimension.

[0155] Therefore, the signal combination module 203 can be specifically used for:

[0156] The attribute identification results of the voltage level dimension, the device type dimension, and the signal type dimension are concatenated and mapped according to a preset naming rule to obtain the standard signal description.

[0157] In one embodiment, the classification and recognition module 202 can be specifically used for:

[0158] The input signal text description is processed by word embedding and position encoding to obtain the input feature vector;

[0159] The input feature vector is fed into a multilayer transformer network for feature extraction, and a classification feature vector is output.

[0160] The classification feature vector is input into the classifier, and the predicted probability of each candidate category under the attribute dimension is output.

[0161] The candidate category with the highest predicted probability is determined as the attribute recognition result.

[0162] In one embodiment, the extraction module 204 may specifically include:

[0163] The annotation submodule is used to perform sequence annotation on the signal text description using the named entity recognition model to obtain an annotation sequence.

[0164] The name determination submodule is used to determine the name of the device entity based on the entity boundary information extracted from the annotation sequence.

[0165] In one embodiment, the annotation submodule can be specifically used for:

[0166] The signal text description is sequentially labeled using a preset labeling system to generate a labeling sequence. The preset labeling system includes labels for identifying the boundaries of device entity names and labels for identifying device types.

[0167] Correspondingly, the name determination submodule can be specifically used for:

[0168] In the labeled sequence, identify a tag segment consisting of tags used to identify the boundaries of device entity names;

[0169] Extract the character subsequence corresponding to the tag fragment, and concatenate the character subsequence to form the device entity name. At the same time, determine the device type corresponding to the device entity name based on the tag used to identify the device type.

[0170] In one embodiment, the verification module 206 can be specifically used for:

[0171] The standard signal description within the entity group is matched with the preset standard signal set, and the standard signal description that matches successfully is determined as the standard matching signal;

[0172] The unmatched standard signals in the preset standard signal set are identified as missing signals;

[0173] The standard signal descriptions within the entity group that do not match the preset standard signal set are identified as non-standard signals.

[0174] In one embodiment, the signal point table verification device may further include:

[0175] The processing module is used to perform format unification processing and / or special symbol standardization processing on the original signal text in the signal point table before classifying and recognizing the signal text description through multiple attribute classification models to obtain the initial text description.

[0176] The extraction module is used to extract the global highest voltage level from the telemetry signals and remote signaling signals of the signal point table;

[0177] The text description generation module is used to append the global highest voltage level as a prefix to the beginning of the initial text description to generate the signal text description.

[0178] It should be noted that for the specific working principles of each component in the device embodiment, please refer to the corresponding section of the method embodiment, which will not be repeated here.

[0179] Corresponding to the above embodiments, the present invention also discloses a computer-readable storage medium that stores at least one instruction, which, when executed by a processor, implements the steps shown in the embodiments of the signal point table verification method.

[0180] Computer-readable storage media can be tangible media that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. Computer-readable storage media can be machine-readable signal media or machine-readable storage media. Computer-readable storage media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0181] Corresponding to the above embodiments, such as Figure 3 As shown, the present invention also provides a schematic diagram of the structure of an electronic device, which may include: a processor 1 and a memory 2;

[0182] The processor 1 and memory 2 communicate with each other via communication bus 3.

[0183] Processor 1, for executing at least one instruction;

[0184] Memory 2 is used to store at least one instruction;

[0185] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0186] Memory 2 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0187] In this embodiment, processor 1 executes at least one instruction to implement the steps shown in the signal point table verification method.

[0188] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0189] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0190] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for verifying a signal point table, characterized in that, include: Retrieve the text description of the signals in the signal point table; The signal text description is classified and identified by multiple attribute classification models to obtain attribute identification results of multiple attribute dimensions, wherein each attribute classification model corresponds to one attribute dimension. The multiple attribute recognition results are combined into a standard signal description corresponding to the signal text description; The device entity name is extracted from the signal text description using a named entity recognition model; The signal text description is grouped based on the device entity name to obtain multiple entity groups; Within each entity group, the standard signal description is matched with a preset set of standard signals, and a corresponding verification result is generated based on the matching result.

2. The signal point table verification method according to claim 1, characterized in that, The multiple attribute dimensions include voltage level dimension, device type dimension, and signal type dimension; Combining multiple attribute recognition results into a standard signal description corresponding to the signal text description includes: The attribute identification results of the voltage level dimension, the device type dimension, and the signal type dimension are concatenated and mapped according to a preset naming rule to obtain the standard signal description.

3. The signal point table verification method according to claim 1 or 2, characterized in that, The attribute classification model performs the classification and recognition process as follows: The input signal text description is processed by word embedding and position encoding to obtain the input feature vector; The input feature vector is fed into a multilayer transformer network for feature extraction, and a classification feature vector is output. The classification feature vector is input into the classifier, and the predicted probability of each candidate category under the attribute dimension is output. The candidate category with the highest predicted probability is determined as the attribute recognition result.

4. The signal point table verification method according to claim 1, characterized in that, The step of extracting device entity names from the signal text description using a named entity recognition model includes: The named entity recognition model is used to perform sequence annotation on the signal text description to obtain the annotation sequence; The name of the device entity is determined based on the entity boundary information extracted from the labeled sequence.

5. The signal point table verification method according to claim 4, characterized in that, The step of performing sequence annotation on the signal text description using the named entity recognition model to obtain the annotation sequence includes: The signal text description is sequentially labeled using a preset labeling system to generate a labeling sequence. The preset labeling system includes labels for identifying the boundaries of device entity names and labels for identifying device types. Correspondingly, determining the device entity name based on the entity boundary information extracted from the labeled sequence includes: In the labeled sequence, identify a tag segment consisting of tags used to identify the boundaries of device entity names; Extract the character subsequence corresponding to the tag fragment, and concatenate the character subsequence to form the device entity name. At the same time, determine the device type corresponding to the device entity name based on the tag used to identify the device type.

6. The signal point table verification method according to claim 1, characterized in that, Within each entity group, the standard signal description is matched with a preset set of standardized signals, and a corresponding verification result is generated based on the matching result, including: The standard signal description within the entity group is matched with the preset standard signal set, and the standard signal description that matches successfully is determined as the standard matching signal; The unmatched standard signals in the preset standard signal set are identified as missing signals; The standard signal descriptions within the entity group that do not match the preset standard signal set are identified as non-standard signals.

7. The signal point table verification method according to claim 1, characterized in that, Before classifying and recognizing the signal text description using multiple attribute classification models, the signal point table verification method further includes: The original signal text in the signal point table is subjected to format unification and / or special symbol standardization to obtain the initial text description; Extract the global highest voltage level from the telemetry and telesignaling signals in the signal point table; The highest global voltage level is appended as a prefix to the beginning of the initial text description to generate the signal text description.

8. A signal point table verification device, characterized in that, include: The text description acquisition module is used to acquire the signal text descriptions from the signal point table. The classification and recognition module is used to classify and recognize the signal text description through multiple attribute classification models to obtain attribute recognition results for multiple attribute dimensions, wherein each attribute classification model corresponds to one attribute dimension; A signal combination module is used to combine multiple attribute recognition results into a standard signal description corresponding to the signal text description; The extraction module is used to extract device entity names from the signal text description using a named entity recognition model; The grouping module is used to group the signal text description based on the device entity name to obtain multiple entity groups; The verification module is used to match the standard signal description with a preset set of standard signals within each entity group, and generate a corresponding verification result based on the matching result.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one instruction, which, when executed by a processor, implements the signal point table verification method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: a memory and a processor; The memory is used to store at least one instruction; The processor is used to execute the at least one instruction to implement the signal point table verification method as described in any one of claims 1 to 7.

Images