Navigation notice grading processing method, system, equipment and medium
By preprocessing and splitting the text length of NOTAMs, and combining them with a dedicated multi-classification model and risk level assessment, the problem of insufficient accuracy and risk control in the existing NOTAM processing technology has been solved, and refined NOTAM processing and risk identification have been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-27
AI Technical Summary
Existing methods for processing NOTAM texts suffer from poor adaptability to heterogeneity, insufficient generalization ability of single models, and coarse classification granularity, resulting in low accuracy and risk control capabilities in NOTAM processing.
By preprocessing the NOTAM text, using state automata and ICAO abbreviation restoration technology, long and short texts are separated and adapted to dedicated multi-classification models, and combined with event category and risk level judgment, refined processing is achieved.
It has improved the accuracy of NOTAM text processing and risk control capabilities, especially the accuracy of short text classification, and achieved refined processing from event classification to risk identification, thereby enhancing the level of intelligence and flexibility.
Smart Images
Figure CN121745668A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of aviation data processing technology, and in particular to a method, system, device and medium for classifying and processing NOTAMs (Notifications to Airmen). Background Technology
[0002] NOTAMs are text messages used to inform aviation professionals about in-flight facilities, services, procedures, or hazardous conditions. Efficiently and accurately extracting key risk information from the vast amount of NOTAMs in various formats has become a major challenge for ensuring aviation safety and operational efficiency.
[0003] Existing methods for processing NOTAM texts mainly rely on automated processing based on a single model, employing a single pre-trained language model to uniformly classify all types of NOTAMs. However, this method suffers from poor adaptability to the heterogeneity of NOTAM texts, insufficient generalization ability of a single model, low classification accuracy for short texts, and coarse classification granularity, resulting in low accuracy and risk control capabilities in NOTAM processing. Summary of the Invention
[0004] The purpose of this invention is to provide a method, system, device, and medium for classifying and processing NOTAMs (Notifications to Airmen), thereby improving the accuracy of NOTAM text processing and risk control capabilities.
[0005] To achieve the above objectives, the present invention provides a method for classifying and processing NOTAMs (Notifications to Airmen), comprising: The original NOTAM text is preprocessed to obtain the preprocessed NOTAM text. Based on the text length of the preprocessed NOTAM text, the preprocessed NOTAM text is divided into long text NOTAM and short text NOTAM. The long text notice is used as input to the long text multi-classification model to classify the event. The short text notice is used as input to the short text multi-classification model to classify the event, thus obtaining the navigation notice event category. Based on the aforementioned NOTAM event category, a corresponding event category multi-classification model is selected to determine the risk level of the NOTAM event category, and the risk level determination result is obtained.
[0006] Optionally, after determining the risk level of the NOTAM event category by selecting the corresponding event category multi-classification model based on the NOTAM event category and obtaining the risk level determination result, the method further includes: A structured data object is generated for each original NOTAM (Notice to Airmen); wherein the fields of the data object include the original NOTAM identifier, the original NOTAM text, the preprocessed NOTAM text, the length type of the NOTAM text, the NOTAM event category label, the risk level judgment result, and the confidence score of the risk level judgment result; The data objects are then visualized.
[0007] Optionally, the preprocessing of the original NOTAM text to obtain the preprocessed NOTAM text includes: A state automaton is constructed based on the original NOTAM text; wherein the state automaton transitions between different states based on the original NOTAM text to perform clause segmentation on the original NOTAM text; The clauses segmented by the state automaton are processed by word segmentation to obtain a word-segmented text sequence; The text sequence is traversed based on a pre-defined abbreviation mapping dictionary, and the abbreviations in the text sequence are replaced with their corresponding full names to obtain the pre-processed Notice to Airmen text.
[0008] Optionally, before the clauses segmented by the state automaton are processed for word segmentation, the method further includes: Domain entities in clauses after splitting the state automaton are predefined as indivisible independent terms.
[0009] Optionally, the step of dividing the preprocessed NOTAM text into long text NOTAMs and short text NOTAMs based on the text length of the preprocessed NOTAM text includes: Calculate the text length of each preprocessed NOTAM text; The calculated text length is compared with a preset length threshold; wherein the length threshold is set by statistical analysis of the navigation notice corpus; If the text length is greater than the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a long text notice; if the text length is less than or equal to the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a short text notice.
[0010] Optionally, the long text multi-classification model is a pre-trained language model fine-tuned on a subset of long text announcement data, and the short text multi-classification model is a pre-trained language model fine-tuned on a subset of short text announcement data.
[0011] Optionally, the corresponding event category multi-classification model is a pre-trained language model that has been fine-tuned on historical data of the corresponding air traffic notice event category.
[0012] To achieve the above objectives, the present invention also provides a navigation notification classification processing system, comprising: The data preprocessing module is used to preprocess the original NOTAM text to obtain the preprocessed NOTAM text. The text splitting module is used to divide the preprocessed NOTAM text into long text NOTAMs and short text NOTAMs based on the text length of the preprocessed NOTAM text. The event classification module is used to take the long text notice as input to the long text multi-classification model to classify the long text notice as an event, and take the short text notice as input to the short text multi-classification model to classify the short text notice as an event, so as to obtain the navigation notice event category. The risk level identification module is used to select the corresponding event category multi-classification model to determine the risk level of the NOTAM event category based on the NOTAM event category, and obtain the risk level determination result.
[0013] To achieve the above objectives, the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the navigation notice classification processing method as described above.
[0014] To achieve the above objectives, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the NOTAM classification processing method as described above.
[0015] Compared with existing technologies, the present invention provides a method, system, device, and medium for classifying and processing NOTAMs (Notifications to Airmen). First, it preprocesses the original NOTAM text to lay a standardized foundation for subsequent model processing. Then, it performs classification processing based on text length, allowing long and short text NOTAMs to be adapted to dedicated multi-classification models. This enables the targeted capture of features from texts of different lengths, improving the accuracy of event classification, especially addressing the problem of insufficient accuracy in short text classification. Finally, it combines the event category with the corresponding model to determine the risk level, achieving refined processing from event classification to risk identification. Ultimately, it can efficiently and accurately identify important NOTAMs with potential risks, improving the intelligence level and risk control capabilities of NOTAM processing. Attached Figure Description
[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a flowchart of a navigation notice classification processing method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a navigation notification text status automaton provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a navigation notification classification processing system provided in an embodiment of the present invention; Figure 4 This is a structural block diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation
[0018] 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.
[0019] It should be noted that the traditional method of processing NOTAMs is manual interpretation, meaning that dispatchers, pilots, and other professionals manually read, screen, and analyze them. This method is labor-intensive, inefficient, and costly in terms of manpower. More seriously, under long hours and heavy workloads, human fatigue or negligence can easily lead to misjudgments or omissions of critical risk information, directly threatening flight safety.
[0020] Existing methods for processing NOTAMs (Notices to Airmen) use a single pre-trained language model, BERT (Bidirectional Encoder Representation of Transformer), to uniformly classify all types of NOTAMs. This approach has the following drawbacks: Limitation 1: Poor adaptability to the heterogeneity of NOTAM texts. NOTAM texts are highly unstructured and domain-specific, containing numerous specialized abbreviations, non-standard punctuation, and diverse encoding formats. Standard natural language processing preprocessing tools struggle to effectively clean and normalize such texts, resulting in low-quality data input to the model and limiting its performance ceiling from the outset.
[0021] Defect 2: Insufficient generalization ability of a single model leads to low classification accuracy for short texts. Notices to Airmen vary significantly in text length and information density, ranging from highly condensed phrases (such as "RWY 18L / 36R CLSD") to long paragraphs containing complex context. When using a single BERT model to process such mixed-length datasets, the model's attention weights inevitably favor learning the rich contextual features of long texts, becoming insensitive to the key but sparse lexical patterns in short texts. This "learning bias" results in significantly lower classification accuracy for short text notices compared to long texts.
[0022] Defect 3: Coarse classification granularity and limited decision support capabilities. Most existing processing methods remain at the level of "risk / non-risk" binary judgment, and cannot provide more refined event categories (such as "equipment failure", "runway maintenance", "airspace restrictions" etc.). This makes it difficult for downstream automated decision-making systems to respond accurately and distribute information based on specific event types.
[0023] In summary, the embodiments of the present invention provide a method for classifying and processing NOTAMs (Notifications to Airmen) to address the aforementioned deficiencies.
[0024] See Figure 1 , Figure 1 This is a flowchart of a NOTAM (Notification to Airmen) classification processing method provided by an embodiment of the present invention, the NOTAM classification processing method including steps S1 to S4: S1. Preprocess the original NOTAM text to obtain the preprocessed NOTAM text. It should be noted that the purpose of step S1 is to transform the original, unstructured NOTAM text into a semantically rich, standardized format that is easy for subsequent models to understand and process.
[0025] In an optional embodiment, step S1 includes steps S101 to S103: S101. Construct a state automaton based on the original NOTAM text; wherein, the state automaton transitions between different states based on the original NOTAM text to perform clause segmentation on the original NOTAM text; For example, to address the issue of non-standard use of line breaks and punctuation in NOTAM texts, a state automaton is constructed. The state automaton transitions between different states (e.g., "beginning of sentence," "middle of sentence," "candidate for end of sentence") based on the current character (e.g., line break, uppercase letter, period) and its contextual information (e.g., whether the preceding character is a sentence-ending symbol). This achieves accurate and fine-grained segmentation of a complete NOTAM text, ensuring that each clause is semantically complete.
[0026] See Figure 2 , Figure 2 This is a schematic diagram of a navigation notification text status automaton provided in an embodiment of the present invention. Figure 2 As shown, the state automaton is used for robust sentence segmentation. Starting from "start processing text", it first enters the "SentenceStart" state. If a letter, number or other regular character is received in this state, it jumps to "InSentence". In the "InSentence" state, if a letter, number or ordinary punctuation mark is received, it continues to maintain this state. If a potential sentence-ending character (such as a period or newline character '\n') is received, it enters "EndCandidate". In the "EndCandidate" state, sentence segmentation is performed when the subsequent character is an uppercase letter or a newline character. Sentence segmentation is also performed when the text ends. Then the process ends. This loop achieves robust sentence segmentation of the text, which can accurately handle different character situations and complete the text sentence segmentation task. S102. Perform word segmentation on the clauses segmented by the state automaton to obtain the word-segmented text sequence; In one alternative embodiment, prior to step S102, the method further includes: Domain entities in clauses after splitting the state automaton are predefined as indivisible independent terms.
[0027] It is worth noting that proprietary entities in the aviation field, such as airport four-letter codes (ZBAA), runway numbers (RWY18L / 36R), and waypoints (LADIX), are predefined as indivisible independent tokens to prevent clauses from being incorrectly segmented during word segmentation, thereby preserving the core semantics.
[0028] S103. Based on a preset abbreviation mapping dictionary, traverse the text sequence and replace the abbreviations in the text sequence with the corresponding full names to obtain the preprocessed navigation notice text.
[0029] For example, a comprehensive "ICAO abbreviation-full name" mapping dictionary can be constructed and maintained based on the abbreviations in the ICAO (International Civil Aviation Organization) standard dictionary (e.g., CLSD -> CLOSED, TWY -> TAXIWAY, WIP -> WORK INPROGRESS). By traversing the segmented text sequence, the identified abbreviations are replaced with their corresponding full words or phrases, thus obtaining the normalized notification text.
[0030] It is worth noting that this method can eliminate the ambiguity of text, thereby restoring the natural language form of the text sentences, enabling the pre-trained language model to more effectively utilize the pre-trained knowledge learned on massive general corpora.
[0031] S2. Based on the text length of the preprocessed NOTAM text, divide the preprocessed NOTAM text into long text NOTAM and short text NOTAM. In an optional embodiment, step S2 includes steps S201 to S203: S201. Calculate the text length of each preprocessed NOTAM text; S202. The calculated text length is compared with a preset length threshold; wherein the length threshold is set by statistical analysis of the navigation notice corpus; S203. If the text length is greater than the preset length threshold, the corresponding preprocessed Notice to Navigation text is marked as a long text notice; if the text length is less than or equal to the preset length threshold, the corresponding preprocessed Notice to Navigation text is marked as a short text notice.
[0032] For example, for each preprocessed NOTAM text, a text length index is calculated, preferably the number of tokens in the text. The calculated text length is then compared to a preset length threshold T. This threshold T can be set through statistical analysis of a large-scale NOTAM corpus (e.g., by taking the median, quartiles, or determining the optimal segmentation point through cross-validation experiments). If the text length is greater than the threshold T, it is marked as a "long text NOTAM"; if the text length is less than or equal to the threshold T, it is marked as a "short text NOTAM".
[0033] It is worth noting that sorting navigation notices based on text length can match the optimal processing model to texts with different characteristics.
[0034] S3. Use the long text notice as input to the long text multi-classification model to classify the long text notice into events, and use the short text notice as input to the short text multi-classification model to classify the short text notice into events, thereby obtaining the navigation notice event category. Specifically, the long text multi-classification model is a pre-trained language model fine-tuned on a subset of long text announcement data, and the short text multi-classification model is a pre-trained language model fine-tuned on a subset of short text announcement data.
[0035] For example, the long text multi-classification model (BERT-L) is a pre-trained language model, such as the BERT model, that is specifically fine-tuned on a subset of long text announcement data. By training specifically on such long text announcement data, the BERT model's self-attention mechanism is optimized to capture long-distance semantic dependencies and complex contextual logic, thereby accurately classifying announcements of highly descriptive event categories such as "complex weather" and "long-term equipment failure". The short text multi-classification model is also based on the BERT architecture, but it is only fine-tuned on a subset of short text announcement data. Since the training data consists entirely of short texts, the model is "forced" to focus on learning local core features such as keywords, fixed collocations, and key entity combinations in phrases. This effectively avoids interference from complex contexts in long texts and improves the classification accuracy of concise event category announcements such as "runway closed" and "navigation station service suspended".
[0036] It should be noted that the long text announcement data subset and the short text announcement data subset are training / validation / test datasets specifically used for training, which are divided and constructed after the splitting process in step S2.
[0037] It is worth noting that step S3 uses the text splitting results from step S2 to send long and short text announcements into their respective dedicated, independently fine-tuned BERT multi-classification models for refined event category classification, which can improve the classification accuracy of announcement event categories.
[0038] S4. Based on the notification-of-navigation event category, select the corresponding event category multi-classification model to determine the risk level of the notification-of-navigation event category, and obtain the risk level determination result.
[0039] Specifically, the corresponding event category multi-classification model is a pre-trained language model that has been fine-tuned on historical data of the corresponding air traffic notification event category.
[0040] For example, based on the event category output in step S3, a corresponding, dedicated BERT importance binary classification model is selected for each NOTAM. For instance, if a NOTAM is classified as "runway maintenance," the "Runway Maintenance - Importance Judgment Model" is invoked. The preprocessed text of the NOTAM is then input into the selected dedicated binary classification model, thereby outputting a binary classification result, namely "Important" (high risk) or "Normal" (low risk), and may include a confidence score.
[0041] It should be noted that each corresponding, dedicated BERT importance binary classification model has undergone deep fine-tuning on historical data for its corresponding event category, learning risk assessment knowledge under that specific event scenario. For example, the "Runway Maintenance - Importance Judgment Model" can learn to distinguish the risk difference between "temporary cleaning" and "indefinite closure"; the "Navigation Station Service Suspension - Importance Judgment Model" can distinguish the risk difference between "service suspension within 1 hour due to short-term equipment calibration" and "service suspension due to serious equipment failure that is expected to be unrecoverable for more than 72 hours".
[0042] In an optional embodiment, after step S4, the NOTAM classification processing method further includes: For each original NOTAM (Notice to Airmen), a structured data object is generated; wherein the fields of the data object include the original NOTAM identifier (ID), the original NOTAM text, the preprocessed NOTAM text, the length type of the NOTAM text (long / short), the NOTAM event category label, the risk level judgment result (importance label), and the confidence score of the risk level judgment result; The data objects are then visualized.
[0043] For example, by integrating the processing results of all the above steps, a structured data object is generated for each original NOTAM (Notice to Airmen), and then the structured data object is output, which can be used for database storage, downstream system calls, or visualization on a user interface.
[0044] In summary, the navigation notice classification processing method provided by this invention first preprocesses the original navigation notice text by combining sentence segmentation methods of state automata and ICAO abbreviation restoration technology. This effectively transforms the "machine-unfriendly" original notice text into "model-friendly" semantically rich data, laying a solid data foundation for the subsequent optimal performance of the BERT model. Then, it performs classification processing based on text length. Through a "segmentation first, classification later" strategy, it matches independently optimized BERT models (BERT-L and BERT-S) for long and short texts, enabling long and short text notices to be adapted to dedicated multi-classification models. This allows for the targeted capture of features of texts of different lengths, improving the accuracy of event classification, especially addressing the problem of insufficient classification accuracy for short texts. This fundamentally solves the performance bottleneck caused by learning bias when a single model processes heterogeneous data, and significantly improves the classification accuracy of a large number of short text notices. By combining event categories with corresponding models for risk level assessment, a hierarchical classification system of "multi-classification of events followed by binary classification of importance" is implemented. This achieves refined processing from event classification to risk identification, not only accurately identifying the nature of the event in the notification ("what is happening"), but also further determining its risk level ("how serious is it"), thereby improving the intelligence level and risk control capabilities of navigation notification processing. Furthermore, the modular design of this embodiment makes the system easily expandable. When a new notification category needs to be added, only a new multi-classification label needs to be added, and a dedicated binary classification model for importance needs to be trained separately for that category. There is no need to reconstruct the entire system architecture, demonstrating excellent flexibility and maintainability.
[0045] Based on the above method items, the present invention provides corresponding system items embodiments.
[0046] See Figure 3 , Figure 3 This is a structural block diagram of a NOTAM (Notification of Navigation) classification processing system provided in an embodiment of the present invention. The NOTAM classification processing system includes: Data preprocessing module 21 is used to preprocess the original NOTAM text to obtain the preprocessed NOTAM text. The text splitting module 22 is used to divide the preprocessed NOTAM text into long text NOTAM and short text NOTAM according to the text length of the preprocessed NOTAM text. The event classification module 23 is used to take the long text notice as input to the long text multi-classification model to classify the long text notice as an event, and take the short text notice as input to the short text multi-classification model to classify the short text notice as an event, so as to obtain the navigation notice event category. The risk level identification module 24 is used to select the corresponding event category multi-classification model to judge the risk level of the notification of navigation event category based on the notification of navigation event category, and obtain the risk level judgment result.
[0047] In one optional embodiment, the NOTAM (Notification of Navigation) classification processing system further includes a data visualization module for: A structured data object is generated for each original NOTAM (Notice to Airmen); wherein the fields of the data object include the original NOTAM identifier, the original NOTAM text, the preprocessed NOTAM text, the length type of the NOTAM text, the NOTAM event category label, the risk level judgment result, and the confidence score of the risk level judgment result; The data objects are then visualized.
[0048] In one optional embodiment, the data preprocessing module 21 includes: A state automaton construction unit is used to construct a state automaton based on the original NOTAM text; wherein the state automaton transitions between different states based on the original NOTAM text to perform clause segmentation on the original NOTAM text; The word segmentation unit is used to segment the clauses after they have been segmented by the state automaton to obtain a segmented text sequence. The abbreviation replacement unit is used to traverse the text sequence based on a preset abbreviation mapping dictionary, replace the abbreviations in the text sequence with the corresponding full names, and obtain the preprocessed navigation notice text.
[0049] In an optional embodiment, the data preprocessing module 21 further includes a domain entity definition unit, used for: Domain entities in clauses after splitting the state automaton are predefined as indivisible independent terms.
[0050] In one alternative embodiment, the text splitting module 22 is configured to: Calculate the text length of each preprocessed NOTAM text; The calculated text length is compared with a preset length threshold; wherein the length threshold is set by statistical analysis of the navigation notice corpus; If the text length is greater than the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a long text notice; if the text length is less than or equal to the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a short text notice.
[0051] It should be noted that the navigation notice classification processing system provided in this embodiment of the invention is used to execute all the process steps of the navigation notice classification processing method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.
[0052] This invention also provides a terminal device, such as... Figure 4 The diagram shown is a structural block diagram of a preferred embodiment of a terminal device provided by the present invention. The terminal device includes a processor 31, a memory 32, and a computer program stored in the memory 32 and configured to be executed by the processor 31. When the processor 31 executes the computer program, it implements the navigation notice classification processing method as described in any of the above embodiments.
[0053] In addition, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the navigation notice classification processing method as described in any of the above embodiments.
[0054] When the processor 31 executes the computer program, it implements the steps in the above-described embodiments of the NOTAM (Notification of Navigation) classification processing method, for example... Figure 1 All steps of the NOTAM classification processing method shown. Alternatively, when the processor 31 executes the computer program, it implements the functions of each module in the above-described NOTAM classification processing system embodiment, for example... Figure 3 The diagram illustrates the functions of each module in the NOTAM (Notification to Navigation) classification system.
[0055] Preferably, the computer program can be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.
[0056] The processor 31 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 31 can be any conventional processor. The processor 31 is the control center of the terminal device, connecting various parts of the terminal device through various interfaces and lines.
[0057] The memory 32 mainly includes a program storage area and a data storage area. The program storage area can store the operating system, applications required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory 32 can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, and a flash card, etc., or the memory 32 can also be other volatile solid-state storage devices.
[0058] It should be noted that the aforementioned terminal devices may include, but are not limited to, processors and memory, as will be understood by those skilled in the art. Figure 4 The structural block diagram shown is merely a structural example of the terminal device described above and does not constitute a limitation on the structure of the terminal device. The terminal device may include more or fewer components than shown, or combine certain components, or use different components.
[0059] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for classifying and processing NOTAMs (Notifications to Airmen), characterized in that, include: The original NOTAM text is preprocessed to obtain the preprocessed NOTAM text. Based on the text length of the preprocessed NOTAM text, the preprocessed NOTAM text is divided into long text NOTAM and short text NOTAM; The long text notice is used as input to the long text multi-classification model to classify the event. The short text notice is used as input to the short text multi-classification model to classify the event, thus obtaining the navigation notice event category. Based on the aforementioned NOTAM event category, a corresponding event category multi-classification model is selected to determine the risk level of the NOTAM event category, and the risk level determination result is obtained.
2. The navigation notice classification processing method as described in claim 1, characterized in that, After determining the risk level of the NOTAM event category by selecting the corresponding event category multi-classification model based on the NOTAM event category and obtaining the risk level determination result, the method further includes: A structured data object is generated for each original NOTAM (Notice to Airmen); wherein the fields of the data object include the original NOTAM identifier, the original NOTAM text, the preprocessed NOTAM text, the length type of the NOTAM text, the NOTAM event category label, the risk level judgment result, and the confidence score of the risk level judgment result; The data objects are then visualized.
3. The navigation notice classification processing method as described in claim 1, characterized in that, The preprocessing of the original NOTAM text to obtain the preprocessed NOTAM text includes: A state automaton is constructed based on the original NOTAM text; wherein the state automaton transitions between different states based on the original NOTAM text to perform clause segmentation on the original NOTAM text; The clauses segmented by the state automaton are processed by word segmentation to obtain a word-segmented text sequence; The text sequence is traversed based on a pre-defined abbreviation mapping dictionary, and the abbreviations in the text sequence are replaced with their corresponding full names to obtain the pre-processed Notice to Airmen text.
4. The navigation notice classification processing method as described in claim 2, characterized in that, Before the clauses segmented by the state automaton are processed for word segmentation, the following steps are also included: Domain entities in clauses after the state automaton is split are predefined as indivisible independent terms.
5. The navigation notice classification processing method as described in claim 1, characterized in that, The preprocessed NOTAM text is divided into long text NOTAMs and short text NOTAMs based on its length, including: Calculate the text length of each preprocessed NOTAM text; The calculated text length is compared with a preset length threshold; wherein the length threshold is set by statistical analysis of the navigation notice corpus; If the text length is greater than the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a long text notice; if the text length is less than or equal to the preset length threshold, the corresponding preprocessed Notice to Navigation text will be marked as a short text notice.
6. The navigation notice classification processing method as described in claim 1, characterized in that, The long text multi-classification model is a pre-trained language model fine-tuned on a subset of long text announcement data, and the short text multi-classification model is a pre-trained language model fine-tuned on a subset of short text announcement data.
7. The navigation notice classification processing method as described in claim 1, characterized in that, The corresponding event category multi-classification model is a pre-trained language model that has been fine-tuned on historical data of the corresponding air traffic notice event category.
8. A navigational notice classification processing system, characterized in that, include: The data preprocessing module is used to preprocess the original NOTAM text to obtain the preprocessed NOTAM text. The text splitting module is used to divide the preprocessed NOTAM text into long text NOTAMs and short text NOTAMs based on the text length of the preprocessed NOTAM text. The event classification module is used to take the long text notice as input to the long text multi-classification model to classify the long text notice as an event, and take the short text notice as input to the short text multi-classification model to classify the short text notice as an event, so as to obtain the navigation notice event category. The risk level identification module is used to select the corresponding event category multi-classification model to determine the risk level of the NOTAM event category based on the NOTAM event category, and obtain the risk level determination result.
9. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the NOTAM classification processing method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the NOTAM classification processing method as described in any one of claims 1 to 7.