Airport emergency processing method, device, equipment, storage medium and program product

Abstract

The invention provides an airport emergency processing method and device, equipment, a storage medium and a program product. The method comprises the steps of firstly obtaining a real-time information text of an airport, determining a risk event type according to the real-time information text, then extracting related elements from the real-time information text according to the risk event type and a risk event element template corresponding to the risk event type, and finally integrating the risk event type, the risk event elements, a preset emergency knowledge graph and a pre-arranged plan library, and determining a to-be-executed emergency plan. According to the method, the problems of low response speed, great influence of experience difference and the like during manual processing are effectively solved, rapid and accurate response to airport emergency situations is realized, the emergency processing efficiency and accuracy are improved, and the safety and high efficiency of airport operation are guaranteed.

Description

Technical Field

[0001] The present application relates to the technical field of airport emergency digital plans, and in particular to an airport emergency handling method, apparatus, equipment, storage medium and program product. Background Art

[0002] With the vigorous development of the air transport industry, airports, as crucial core nodes in the air transport network, have attracted much attention for their operational safety and efficiency.

[0003] When an emergency occurs at an airport, existing emergency plans rely heavily on manual judgment. This means aviation safety personnel must determine the nature of the incident based on the content of ground-to-air communications and initiate an emergency response.

[0004] However, this manual processing method has problems such as slow response speed and inaccurate judgment, which may affect the effectiveness of the airport's emergency response. Summary of the Invention

[0005] The embodiments of the present application provide an airport emergency handling method, apparatus, equipment, storage medium, and program product to solve the problem that emergency situations at existing airports mainly rely on manual judgment.

[0006] In a first aspect, an embodiment of the present application provides an airport emergency handling method, comprising:

[0007] Acquire real-time information text of the airport, and determine the risk event type based on the real-time information text, wherein the real-time information text at least includes ground-to-air communication text;

[0008] Extracting risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type;

[0009] Based on the risk event type, the risk event elements, the preset emergency knowledge graph and the plan library, the emergency plan to be executed is determined.

[0010] In a possible implementation, determining the risk event type based on the real-time information text includes:

[0011] After word segmentation processing is performed on the real-time information text, the word segmentation results are input into a pre-trained language model to obtain a word vector for each word;

[0012] Inputting the word segmentation results into a topic model to obtain a text-topic probability distribution, a topic-word probability distribution, and a topic vector for each topic corresponding to the real-time information text;

[0013] Determine a topic vector corresponding to each word according to the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic;

[0014] The word vector of each word and the topic vector corresponding to each word are spliced ​​into a text feature vector of each word, and the risk event type is determined based on the text feature vector of each word in the real-time information text.

[0015] In a possible implementation, determining the topic vector corresponding to each word according to the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic includes:

[0016] Determine the probability distribution of each word under each topic based on the topic-word probability distribution and the text-topic probability distribution;

[0017] The topic vector corresponding to each word is determined based on the probability distribution of each word under each topic and the topic vector of each topic.

[0018] In a possible implementation, obtaining the real-time information text of the airport includes:

[0019] A ground-to-air call recording of the airport is obtained, and the ground-to-air call recording is converted into text to obtain the ground-to-air call text.

[0020] In a possible implementation, obtaining the real-time information text of the airport further includes at least one of the following:

[0021] Acquiring flight dynamic data of the airport, and converting specific codes in the flight dynamic data into corresponding flight dynamic text, wherein the real-time information text includes the flight dynamic text;

[0022] Obtaining a weather warning text for the airport, wherein the real-time information text includes the weather warning text;

[0023] A security log text of the airport is obtained, where the real-time information text includes the security log text.

[0024] In a possible implementation, extracting risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type includes:

[0025] splicing the risk event type, the risk event element template, and the real-time information text into a first prompt word;

[0026] The first prompt word is input into a large language model to obtain the risk event element.

[0027] In a possible implementation, determining the emergency plan to be executed based on the risk event type, the risk event elements, a preset emergency knowledge graph, and a plan library includes:

[0028] Retrieving a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph;

[0029] The emergency plan output prompt information, the risk event type, the risk event elements and the plan group are spliced ​​into a second prompt word, and the second prompt word is input into a large language model to obtain the emergency plan to be executed.

[0030] In a possible implementation, the retrieving a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph includes:

[0031] Acquire related knowledge from the emergency knowledge graph according to the risk event type and the risk event elements;

[0032] A risk feature is extracted from the risk event type, the risk event elements and the associated knowledge, and a plan group matching the risk feature is determined from the plan library.

[0033] In a possible implementation, acquiring associated knowledge from the emergency knowledge graph according to the risk event type and the risk event elements includes:

[0034] Extracting at least one first entity from the risk event type and the risk event elements;

[0035] The knowledge corresponding to the first entity and / or the second entity associated with the first entity in the emergency knowledge graph is determined as the associated knowledge.

[0036] In one possible implementation, the step of concatenating the emergency plan output prompt information, the risk event type, the risk event elements, and the plan group into a second prompt word, and inputting the second prompt word into a large language model to obtain the emergency plan to be executed includes:

[0037] Inputting the second prompt word into a large language model, encoding the emergency plan output prompt information, the risk event type, and the risk event elements in the second prompt word using a prompt encoder in the large language model to obtain a prompt feature vector, and encoding the plan group in the second prompt word using a plan encoder in the large language model to obtain a plan feature vector;

[0038] The emergency plan to be executed is obtained based on the prompt feature vector and the plan feature vector.

[0039] In a possible implementation, the method further includes:

[0040] Obtain multiple emergency knowledge texts;

[0041] For each emergency knowledge text, segment the emergency knowledge text to obtain words in the emergency knowledge text, and extract the original feature vector of each word in the emergency knowledge text;

[0042] Identifying a head entity in the emergency knowledge text, and determining an attention weight of the head entity and each word in the emergency knowledge text, and determining an enhanced feature vector for each word in the emergency knowledge text based on the attention weight and the original feature vector of each word in the emergency knowledge text;

[0043] Extracting a tail entity from the emergency knowledge text based on the enhanced feature vector of each word in the emergency knowledge text, and determining a relationship between the head entity and the tail entity;

[0044] The emergency knowledge graph is constructed based on the head entity, the tail entity, and the relationship between the head entity and the tail entity in each of the emergency knowledge texts.

[0045] In a second aspect, an embodiment of the present application provides an airport emergency response device, comprising:

[0046] An acquisition module, configured to acquire real-time information text of the airport, wherein the real-time information text at least includes ground-to-air communication text;

[0047] A determination module, configured to determine a risk event type based on the real-time information text;

[0048] An extraction module, configured to extract risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type;

[0049] The determination module also determines the emergency plan to be executed based on the risk event type, the risk event elements, the preset emergency knowledge graph and the plan library.

[0050] In a possible implementation, the apparatus further includes: a processing module;

[0051] The processing module is configured to perform word segmentation on the real-time information text and input the word segmentation results into a pre-trained language model to obtain a word vector for each word;

[0052] The device further comprises: an input module;

[0053] The input module is used to input the word segmentation results into the topic model to obtain the text-topic probability distribution, topic-word probability distribution and topic vectors of each topic corresponding to the real-time information text;

[0054] The determination module is further configured to determine a topic vector corresponding to each word based on the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic;

[0055] The determination module is specifically used to splice the word vector of each word and the topic vector corresponding to each word into a text feature vector of each word, and determine the risk event type based on the text feature vector of each word in the real-time information text.

[0056] In a possible implementation, the determination module is further configured to determine the probability distribution of each word under each topic based on the topic-word probability distribution and the text-topic probability distribution;

[0057] The determination module is specifically configured to determine the topic vector corresponding to each word based on the probability distribution of each word under each topic and the topic vector of each topic.

[0058] In a possible implementation, the acquisition module is further configured to acquire ground-to-air call recordings at the airport;

[0059] The processing module is further configured to perform text processing on the ground-to-air call recording to obtain the ground-to-air call text.

[0060] In a possible implementation, the acquisition module is further configured to acquire flight dynamic data of the airport;

[0061] The device further comprises: a conversion module;

[0062] The conversion module is configured to convert a specific code in the flight dynamic data into a corresponding flight dynamic text, wherein the real-time information text includes the flight dynamic text;

[0063] The acquisition module is further configured to acquire a weather warning text of the airport, wherein the real-time information text includes the weather warning text;

[0064] The acquisition module is further used to acquire the security log text of the airport, and the real-time information text includes the security log text.

[0065] In a possible implementation, the device further includes: a splicing module;

[0066] The splicing module is used to splice the risk event type, the risk event element template and the real-time information text into a first prompt word;

[0067] The input module is further configured to input the first prompt word into a large language model to obtain the risk event element.

[0068] In a possible implementation, the apparatus further includes: a retrieval module;

[0069] The retrieval module is configured to retrieve a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph;

[0070] The splicing module is further used to splice the emergency plan output prompt information, the risk event type, the risk event element and the plan group into a second prompt word;

[0071] The input module is further configured to input the second prompt word into the large language model to obtain the emergency plan to be executed.

[0072] In a possible implementation, the acquisition module is further configured to acquire associated knowledge from the emergency knowledge graph according to the risk event type and the risk event elements;

[0073] The extraction module is further configured to extract risk features from the risk event type, the risk event elements, and the associated knowledge, and determine a plan group matching the risk features from the plan library.

[0074] In a possible implementation, the extraction module is further configured to extract at least one first entity from the risk event type and the risk event element;

[0075] The determination module is specifically used to determine the knowledge corresponding to the first entity and / or the second entity associated with the first entity in the emergency knowledge graph as the associated knowledge.

[0076] In one possible implementation, the input module is further configured to input the second prompt word into a large language model, encode the emergency plan output prompt information, the risk event type, and the risk event elements in the second prompt word using a prompt encoder in the large language model to obtain a prompt feature vector, and encode the plan group in the second prompt word using a plan encoder in the large language model to obtain a plan feature vector.

[0077] The determining module is specifically configured to obtain the emergency plan to be executed based on the prompt feature vector and the plan feature vector.

[0078] In a possible implementation, the acquisition module is further configured to acquire a plurality of emergency knowledge texts;

[0079] The device further comprises: a word segmentation module;

[0080] The word segmentation module is used to segment each emergency knowledge text to obtain words in the emergency knowledge text;

[0081] The extraction module is further used to extract the original feature vector of each word in the emergency knowledge text;

[0082] The device further includes: an identification module;

[0083] The recognition module is used to recognize the head entity in the emergency knowledge text;

[0084] The determination module is further configured to determine the attention weight of each word in the head entity and the emergency knowledge text, and determine an enhanced feature vector for each word in the emergency knowledge text based on the attention weight and the original feature vector of each word in the emergency knowledge text;

[0085] The determination module is further configured to extract a tail entity in the emergency knowledge text based on the enhanced feature vector of each word in the emergency knowledge text, and determine a relationship between the head entity and the tail entity;

[0086] The device further comprises: a construction module;

[0087] The construction module is used to construct the emergency knowledge graph based on the head entity, the tail entity and the relationship between the head entity and the tail entity in each emergency knowledge text.

[0088] In a third aspect, an embodiment of the present application provides an electronic device, including: a memory, a processor;

[0089] The memory stores computer-executable instructions;

[0090] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0091] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementation methods of the first aspect.

[0092] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0093] The airport emergency handling method, device, equipment, storage medium and program product provided by the embodiment of the present application obtains the real-time information text of the airport, first determines the risk event type based on the text content, and then extracts the relevant risk elements from the real-time information text based on the risk event type and its corresponding element template. Then, the emergency plan to be executed is quickly determined in combination with the risk event type, risk elements, preset emergency knowledge graph and plan library. This method can quickly extract key information from the ground-to-air call text, accurately determine the risk event type, and further extract risk event elements, and then quickly match the most appropriate emergency plan in combination with the preset emergency knowledge graph and plan library, effectively solving the common problems of slow response speed and inaccurate judgment in the manual processing process, realizing rapid response and accurate handling of airport emergencies, greatly shortening the emergency response time, improving the efficiency and accuracy of emergency handling, and providing a solid guarantee for the safe and efficient operation of the airport. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0095] Figure 1 The process of airport emergency handling method provided for this application Figure 1 ;

[0096] Figure 2 The process of airport emergency handling method provided for this application Figure 2 ;

[0097] Figure 3 The process of airport emergency handling method provided for this application Figure 3 ;

[0098] Figure 4 A schematic diagram of the structure of the airport emergency response device provided for this application;

[0099] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application.

[0100] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0101] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0102] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present invention described herein can, for example, be practiced in an order other than that illustrated or described herein.

[0103] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0104] With the vigorous development of the air transport industry, airports, as crucial core nodes in the air transport network, have attracted much attention for their operational safety and efficiency.

[0105] When an emergency occurs at an airport, existing emergency plans rely heavily on manual judgment. This means aviation safety personnel must determine the nature of the incident based on the content of ground-to-air communications and initiate an emergency response.

[0106] However, this manual processing method has problems such as slow response speed and inaccurate judgment, which may affect the effectiveness of the airport's emergency response.

[0107] This application provides an airport emergency response method. First, the text of airport ground-to-air communication is obtained to determine the risk event type. Then, based on the risk event type and its corresponding risk event element template, relevant elements are extracted from the text. Finally, the emergency plan to be implemented is determined by integrating the risk event type, elements, a preset emergency knowledge graph, and a plan library. This method effectively solves the problems of slow response speed and significant influence of experience differences in manual processing. It achieves a rapid and accurate response to airport emergencies, improves the efficiency and accuracy of emergency handling, and ensures the safety and efficiency of airport operations.

[0108] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0109] Figure 1 Schematic diagram of the airport emergency response process provided for this application Figure 1 ,like Figure 1 As shown, the method includes:

[0110] S101. Acquire real-time information text of an airport, and determine a risk event type based on the real-time information text, wherein the real-time information text at least includes ground-to-air communication text.

[0111] Real-time information text refers to a collection of relevant information presented in text form that reflects the current operating status of the airport, flight activities, and various dynamic situations. Real-time information text includes but is not limited to ground-to-air communication text, flight status text, weather warning text, and safety log text.

[0112] An air-ground communication transcript is a document that records the voice conversations between airport air traffic controllers and pilots during flight-related operations during air transportation. For example, an air-ground communication transcript might include an airport air traffic controller informing a pilot that there is a foreign object on the scheduled runway 1 and requesting the pilot to change to runway 3 for descent.

[0113] Risk event types are categorized into various potentially hazardous or disruptive events that may occur during airport operations and flights. These include, but are not limited to, emergency rescue, fire emergency response, aircraft failure, runway anomaly, and severe weather anomaly.

[0114] Understandably, the airport operating environment is complex and ever-changing, and various risk events may occur at any time, such as aircraft mechanical failure, runway incursion, severe weather, communication interference, etc. Real-time information text contains a large amount of information related to these risk events.

[0115] Therefore, by using the real-time information text as a reference factor, it is possible to determine the keywords, sentences or abnormal situations in the real-time information text that may indicate the occurrence of a risk event, and then determine the type of risk event.

[0116] The method for obtaining the airport's real-time information text in this step may be, for example, obtaining it by querying a flight data recorder, obtaining it by connecting to an air traffic management system, or obtaining it by querying an airline's internal system. This application does not impose any particular restrictions on this.

[0117] For example, if the real-time information text is an airport air traffic controller telling the pilot to suspend the plane's takeoff due to bad weather, it can be determined first that the real-time information text is a ground-to-air communication text, and then the risk event type can be determined to be a bad environment abnormal type.

[0118] S102: Extract risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type.

[0119] Different risk event types correspond to different risk event element templates. For example,

[0120] If the risk event type is emergency rescue for a house fire, the risk event element template corresponding to the emergency rescue for a house fire is: the rescue object is {fill in placeholder} + the implementer of the event is {fill in placeholder} + the victim of the event is {fill in placeholder}.

[0121] If the risk event type is severe environmental anomaly, the risk event element template corresponding to the severe environmental anomaly is: abnormal weather phenomenon is {fill in placeholder} + affected area is {fill in placeholder} + risk level is {fill in placeholder} + event implementer is {fill in placeholder} + affected object is {fill in placeholder}.

[0122] If the risk event type is runway safety risk, the risk event element template corresponding to the runway safety risk is: risk object is {fill in placeholder} + event location is {fill in placeholder} + event implementer is {fill in placeholder} + risk level is {fill in placeholder}.

[0123] Risk event elements refer to key information that can accurately describe and define the characteristics, nature, and impact of a risk event. Risk event elements include, but are not limited to: event type, event location, event perpetrator, affected parties, and event level.

[0124] The purpose of this step is to extract information related to risk events from real-time information text in order to accurately identify, analyze and handle possible risk events.

[0125] As you can understand, after determining the risk event type and its corresponding risk event element template, the real-time information text can be analyzed to accurately identify and extract information that matches the elements in the risk event element template. For example, if the risk event type is an aircraft failure, the risk event element template includes the model of the failed aircraft and the number of the failed engine. Based on this information, relevant sentences can be located in the real-time information text, and specific elements such as the aircraft model and engine number can be extracted.

[0126] S103. Determine the emergency plan to be executed based on the risk event type, risk event elements, the preset emergency knowledge graph, and the plan library.

[0127] The pre-set emergency knowledge graph is a graphically presented knowledge system. For example, it uses risk event types, emergency resources, and emergency organizations as nodes, connecting them through semantic relationships (such as causal relationships, sequential relationships, and affiliation) to form a large and orderly knowledge network. For example, the emergency knowledge graph displays the risk event type as "aircraft fire" and lists the relevant emergency organizations, including "fire trucks" and "firefighters."

[0128] The plan library includes emergency response plans for multiple different types of risk events.

[0129] The purpose of this step is to ensure that when a risk event occurs, the most appropriate plan can be selected quickly and accurately from a variety of emergency plans for implementation.

[0130] While risk event types clearly define the overall nature of an event, the same type of event may have different characteristics and response priorities depending on specific factors. Risk event factors further refine the key information of the event, providing a basis for selecting emergency plans. The pre-configured emergency knowledge graph provides comprehensive emergency knowledge, helping to understand the relationship between the event and other emergency resources and processes, and avoiding decision-making errors due to knowledge gaps. The emergency plan library stores various specific response plans.

[0131] Therefore, by comprehensively considering the risk event types, elements, emergency knowledge maps and plan libraries, we can ensure that the selection of emergency plans is both targeted and in line with the airport's overall emergency management system and resource conditions, so that the airport can carry out emergency response work quickly and orderly when facing various risk events, and ensure the normal operation of the airport and flight safety.

[0132] The airport emergency response method provided in the embodiments of this application first obtains the airport's real-time information text and determines the risk event type based on it. Then, based on the risk event type and corresponding element template, it extracts the risk event elements from the real-time information text. Finally, it determines the emergency plan to be executed by combining the risk event type, elements, and a preset emergency knowledge graph and plan library. This method effectively solves the common problems of slow response and inaccurate judgment in manual processing by automatically identifying risks and matching them with emergency plans. It significantly improves the efficiency of emergency response, thereby providing efficient technical support for aviation safety.

[0133] Figure 2 Schematic diagram of the airport emergency response process provided for this application Figure 2 .like Figure 2As shown, this embodiment Figure 1 Based on the embodiment, the airport emergency handling method is described in detail, and the method includes:

[0134] S201. Acquire real-time information text of the airport, wherein the real-time information text at least includes ground-to-air communication text.

[0135] Optionally, the present application provides an implementation method for determining the ground-to-air call text, including: obtaining the ground-to-air call recording at the airport, text-processing the ground-to-air call recording, and obtaining the ground-to-air call text.

[0136] Understandably, during air transportation, ground-to-air recordings fully capture the voice conversations between air traffic controllers and pilots during flight-related operations. However, this audio information is difficult for the system to effectively utilize directly, and the system cannot quickly identify and extract valuable information from the audio, such as flight status and changes in control instructions.

[0137] Therefore, by textually processing the ground-to-air call recordings, the audio information can be expressed in a textual manner, which helps the system to quickly perform semantic parsing and analysis on the ground-to-air call texts.

[0138] Optionally, the real-time information text provided by this application includes not only ground-to-air communication text, but also flight dynamic text, weather warning text, and security log text. This application provides a method for obtaining flight dynamic text, weather warning text, and security log text, respectively:

[0139] (1) For flight dynamic text: obtain the flight dynamic data of the airport and convert the specific codes in the flight dynamic data into the corresponding flight dynamic text.

[0140] Among them, flight dynamic data includes but is not limited to: flight take-off and landing time, flight status, flight location, airline information, and flight number.

[0141] The purpose of this step is to convert flight status data, which originally existed in the form of specific codes, into flight status text that is easy to understand and use.

[0142] Understandably, the specific codes in flight status data are typically designed to facilitate data storage, transmission, and processing, but these codes cannot directly convey the meaning of a flight within the air transport process. Therefore, by converting specific codes into corresponding flight status text, flight status information can be more intuitively obtained, thereby facilitating a more systematic and comprehensive analysis of the types of risk events facing airports.

[0143] The method of obtaining the airport's flight dynamic data in this step can be, for example, obtained by querying the flight data recorder or by querying the airline's internal system. This application does not impose any special restrictions on this.

[0144] (2) For weather warning text: obtain the weather warning text of the airport.

[0145] The meteorological warning text includes but is not limited to: the specific manifestations of abnormal weather, potential impacts, warning levels, and recommended preventive measures.

[0146] Weather warnings are, understandably, responses developed by airport management based on meteorological information. They help airports prepare for possible extreme weather conditions and ensure flight safety and smooth operations. Therefore, obtaining these warnings allows for more accurate analysis of the types of risk events facing airports.

[0147] The method of obtaining the airport's weather warning text in this step may be, for example, obtaining it from the airport's weather management platform, or manually inputting it after staff analyze the airport's weather.

[0148] (3) Regarding security log text: obtain the airport’s security log text.

[0149] The safety log text includes, but is not limited to: safety inspection and maintenance records of airport facilities and equipment, such as runway pavement condition inspections and navigation lighting system troubleshooting; records of personnel safety operations, such as security personnel's safety inspections of passengers and baggage, and maintenance personnel's aircraft maintenance operations; various types of safety training; and investigation, handling, and improvement measures taken in response to safety incidents.

[0150] As you can see, the security log clearly records the security risks and risk events that occurred at the airport during different time periods and business processes. Therefore, by obtaining the airport's security log text, you can use it as a reference to more comprehensively analyze the types of risk events the airport faces.

[0151] The method of obtaining the airport security log text in this step can be, for example, obtaining it from the airport security management platform. This application does not impose any special restrictions on this.

[0152] S202: After word segmentation processing is performed on the real-time information text, the word segmentation results are input into a pre-trained language model to obtain a word vector for each word.

[0153] Among them, word segmentation processing includes: Chinese word segmentation and removal of stop words.

[0154] Among them, Chinese word segmentation refers to the process of breaking down a continuous real-time information text into meaningful words. For example, Chinese word segmentation can be performed using the Jieba word segmentation technology to segment the real-time information text. For example, suppose the real-time information text is "The airport air traffic controller told the pilot to suspend the plane's takeoff due to bad weather." Based on the above information, the analysis results can be obtained as "airport", "air", "traffic", "controller", "told", "pilot", "due to", "weather", "bad", "," "suspend", "plane", and "takeoff".

[0155] Stop word removal refers to the process of removing words that have no practical significance for real-time information text analysis, such as punctuation, particles, and pronouns. These words, while frequently appearing in sentences, do not contribute to understanding the main message of the real-time information text. Removing them helps improve data processing efficiency and accuracy. For example, the Chinese word segmentation results are: "airport," "air," "traffic," "controller," "tell," "pilot," "due to," "weather," "bad," "pause," "airplane," and "takeoff." After stop word removal, the word segmentation results become: "airport," "air," "traffic," "controller," "tell," "pilot," "due to," "weather," "bad," "pause," "airplane," and "takeoff."

[0156] The pre-trained language model may be, for example, a bidirectional encoder representation model or a knowledge-enhanced semantic representation model, which is not specifically limited in this application.

[0157] This step inputs the word segmentation results into the pre-trained language model to obtain the word vector for each word. The purpose is to capture the semantic information of the vocabulary through the pre-trained model, thereby providing a richer contextual representation for subsequent text analysis or processing tasks.

[0158] As you can understand, pre-trained language models, trained on large-scale corpora, can understand the deep semantics of words and appropriately embed them in different contexts, generating corresponding word vectors. These word vectors contain semantic information about words in different contexts, helping the system better understand language structure and sentence meaning.

[0159] The result of word segmentation is to split the original real-time information text into independent words according to certain rules. These words after word segmentation are the basic components of the real-time information text, but the system cannot directly understand their semantics.

[0160] Therefore, by inputting the word segmentation results as input data into the pre-trained language model, the pre-trained language model can use the language knowledge and semantic features it has learned to generate a corresponding word vector for each word after segmentation.

[0161] S203 : Input the word segmentation result into the topic model to obtain the text-topic probability distribution, topic-word probability distribution and topic vector of each topic corresponding to the real-time information text.

[0162] The topic model may be, for example, a topic word vector model. The topic model includes multiple topics, such as a topic on harsh environment, a topic on emergency rescue, and a topic on runway anomaly.

[0163] The text-topic probability distribution represents the probability that a real-time information text belongs to each topic. The text-topic probability distribution ranges from [0, 1]. It represents the likelihood that a real-time information text belongs to a specific topic. A larger text-topic probability distribution indicates that the real-time information text is more inclined toward that topic; conversely, a smaller text-topic probability distribution indicates that the real-time information text is less relevant to that topic.

[0164] For example, consider a real-time information message from an airport air traffic controller informing a pilot that a flight is grounded due to bad weather. The topics include "Emergency Rescue" and "Bad Weather." Based on this information, we can determine that the probability of the real-time information message being in the "Emergency Rescue" topic is 0.6, and the probability of the real-time information message being in the "Bad Weather" topic is 1.0.

[0165] The topic-word probability distribution represents the probability of all words in a text appearing in each topic. The topic-word probability distribution ranges from [0 to 1]. It represents the likelihood that a word in a real-time information text belongs to a particular topic. A larger topic-word probability distribution indicates that a word in the real-time information text is more likely to be associated with that topic; conversely, a smaller topic-word probability distribution indicates that a word in the real-time information text is less relevant to that topic.

[0166] For example, if the word segmentation results for the real-time information text include "weather" and "bad," and the topics include "emergency rescue" and "bad environment," then based on this information, we can determine that the probability of "weather" in the "bad environment" topic is 0.9 and the probability of "emergency rescue" is 0. Also, the probability of "bad" in the "bad environment" topic is 0.9 and the probability of "emergency rescue" is 0.8.

[0167] The topic vector for a topic represents all the words in a topic in a vectorized form. For example, if the topic is "runway anomaly," the topic vector includes "runway" and "anomaly." If the topic is "abnormally harsh environment," the topic vector includes "environment," "harsh," and "weather."

[0168] S204 : Determine the topic vector corresponding to each word according to the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic.

[0169] As you can understand, the topic-word probability distribution reflects the probability of a word in different topics and clearly indicates the likelihood that a word belongs to a certain topic. The text-topic probability distribution reflects the relationship between a document and different topics and clearly indicates the likelihood that a document belongs to a specific topic. The topic vector, on the other hand, vectorizes the lexical features of each topic, reflecting the core characteristics of the topic.

[0170] Therefore, by comprehensively considering the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic, we can determine which topic a word in a document is most likely related to, and thus assign an accurate topic vector to the word.

[0171] Optionally, this application provides a possible implementation method, including:

[0172] The first step is to determine the probability distribution of each word under each topic based on the topic-word probability distribution and the text-topic probability distribution.

[0173] It is understandable that the topic-word probability distribution can only reflect the probability of a word appearing in different topics, but does not consider the impact of the text as a whole on the topic; the text-topic probability distribution can only reflect the association between the text and the topic, but cannot be specific to the distribution of each word in the topic.

[0174] Therefore, combining the topic-word probability distribution and the text-topic probability distribution allows the system to comprehensively consider the text as a whole and the occurrence of words in the topic, which helps the system to more comprehensively and accurately determine the probability distribution of each word under each topic.

[0175] Optionally, this application provides a possible implementation method, including:

[0176]

[0177] in, A text sequence representing emergency knowledge text { }; Indicates emergency knowledge text The word; Indicates the Themes Indicates emergency knowledge text Belong to the theme probability; Indicates emergency knowledge text The words, in the subject The probability of occurrence in Indicates emergency knowledge text The words belonging to the topic probability.

[0178] For example, suppose an emergency knowledge text Contains words = "fire extinguisher", and there are two topics: = "Fire First Aid", = "bad environment".

[0179] Known topic-word distributions include: ,as well as .

[0180] Text-topic distribution: ,as well as .

[0181] Based on the above information, it can be determined that the probability that the word "fire extinguisher" belongs to the "fire first aid" topic is 0.72, and the probability that the word "fire extinguisher" belongs to the "harsh environment" topic is 0.02.

[0182] In the second step, the topic vector corresponding to each word is determined based on the probability distribution of each word under each topic and the topic vector of each topic.

[0183] It's understandable that the probability distribution of each word under each topic reflects the degree of association between the word and each topic, but this degree of association alone cannot fully express the semantic characteristics of the topic. The topic vector of each topic, on the other hand, contains the semantic information of the topic and represents the core characteristics of the topic.

[0184] Therefore, by combining the probability distribution of a word under each topic with the topic vector, the semantic information of the topic can be incorporated into the word representation. In this way, the resulting topic vector corresponding to the word not only considers the strength of the word's association with each topic, but also integrates the semantic characteristics of the topic, making the word's semantic expression more accurate and comprehensive.

[0185] S205 , concatenating the word vector of each word and the topic vector corresponding to each word into a text feature vector of each word, and determining the risk event type based on the text feature vector of each word in the real-time information text.

[0186] It is understandable that word vectors can only represent the semantics of the word itself, but when considering the specific context of real-time information text, a single word vector may not be enough to accurately convey the multiple meanings of a word.

[0187] Therefore, by concatenating the word vector of each word with its corresponding topic vector, the system can obtain a text feature vector that integrates semantics and thematic background, making the expression of each word in the real-time information text richer, helping the system to improve the recognition accuracy of risk event types in real-time information text, and helping the system make more reasonable judgments in complex situations.

[0188] S206: Combine the risk event type, the risk event element template, and the real-time information text into a first prompt word.

[0189] The purpose of this step is to ensure that the system can obtain more comprehensive and multi-dimensional contextual information when analyzing real-time information text.

[0190] It is understandable that a risk event type or risk event element template alone may not be sufficient to fully describe the details of the event, while the real-time information text provides the actual context of the event.

[0191] Therefore, by integrating risk event types, risk event element templates and real-time information text, the system can better understand the background, nature and potential threats of the event and improve recognition accuracy.

[0192] For example, suppose the real-time information text is an airport air traffic controller informing the pilot that the flight is suspended due to bad weather. The risk event type corresponding to this real-time information text is the bad environment anomaly type. The risk event element template corresponding to the known bad environment anomaly type is: abnormal weather phenomenon is {fill in placeholder} + affected area is {fill in placeholder} + risk level is {fill in placeholder} + event implementer is {fill in placeholder}.

[0193] Based on the above information, it can be determined that the first prompt word is: the airport air traffic controller told the pilot to suspend the aircraft takeoff due to bad weather; the environment is bad and abnormal; the abnormal weather phenomenon is {fill in placeholder} + the affected area is {fill in placeholder} + the risk level is {fill in placeholder} + the perpetrator of the incident is {fill in placeholder}.

[0194] S207: Input the first prompt word into the large language model to obtain risk event elements.

[0195] The large language model may be, for example, a bidirectional transformer model.

[0196] The purpose of this step is to use the natural language processing capabilities of the large language model to automatically analyze and extract the core elements of risk events.

[0197] For example, suppose the first prompt word is that the airport air traffic controller tells the pilot that the plane is suspended from taking off due to bad weather; the environment is bad and abnormal; the abnormal weather phenomenon is {fill in placeholder} + the affected area is {fill in placeholder} + the risk level is {fill in placeholder} + the person who carried out the incident is {fill in placeholder}.

[0198] Based on the above information, the first prompt word can be input into the bidirectional converter model to obtain the risk event elements output by the bidirectional converter model, including:

[0199] Event type: Abnormal type due to severe environment;

[0200] Abnormal weather phenomena: heavy rain, lightning;

[0201] Affected areas: All areas of the airport;

[0202] Risk level: Level 2;

[0203] Incident implementer: Command Center;

[0204] Affected persons: Passengers

[0205] }.

[0206] S208. Based on the risk event type, risk event elements, and the preset emergency knowledge graph, retrieve the plan group related to the risk event from the plan library.

[0207] The purpose of this step is to accurately and efficiently screen out a set of plans that are closely related to the current risk event from the plan library.

[0208] As you can understand, risk event types are used to define the categories of risk events. Different types of events generally require different types of emergency response plans. Risk event elements further refine the specific circumstances of the event, such as the time of occurrence, the perpetrators, the affected parties, and the incident level.

[0209] The preset emergency knowledge graph is a graphical knowledge system that contains the relationships and rules between various entities in the emergency field.

[0210] Therefore, by comprehensively considering the risk event type, risk event elements and the preset emergency knowledge graph, the plans can be screened and matched from multiple dimensions to ensure that the retrieved plan group not only matches the risk event type, but also fully considers the specific elements of the event and conforms to the emergency logic and specifications contained in the emergency knowledge graph, thereby providing an emergency plan that is more in line with the actual situation.

[0211] Optionally, this application provides a possible implementation method, including:

[0212] The first step is to obtain related knowledge from the emergency knowledge graph based on the risk event type and risk event elements.

[0213] The purpose of this step is to obtain comprehensive knowledge and information that is closely related to the current risk event, so as to provide a solid basis for the subsequent formulation of effective emergency plans.

[0214] It is understandable that when facing risk events, relying solely on a single risk event type and risk event element is often unable to fully respond to complex emergency situations.

[0215] Therefore, it is also necessary to master various related knowledge, such as previous experience in handling similar incidents, relevant emergency resource information, and the responsibilities of different entities in emergency response.

[0216] By obtaining associated knowledge related to risk event types and risk event elements from the emergency knowledge graph, the system can understand risk events more comprehensively.

[0217] The second step is to extract risk characteristics from risk event types, risk event elements and related knowledge, and determine the plan group that matches the risk characteristics from the plan library.

[0218] Among them, risk characteristics refer to specific attributes or attribute combinations that can comprehensively and accurately describe the essential attributes, key elements and related influencing factors of risk events, and can be used to distinguish different risk events, measure the degree of risk, and guide the matching of emergency plans.

[0219] The purpose of this step is to select the set of plans that best suits the current risk event from the plan library.

[0220] Understandably, in actual risk scenario emergency response, the plan library stores numerous different types of emergency plans, each with its own applicable conditions and scope. By extracting risk characteristics and matching them within the plan library, we can avoid blindly searching through a large number of plans, greatly improving the efficiency and accuracy of plan retrieval. This ensures that the appropriate plan can be quickly deployed in an emergency, providing strong support for emergency decision-making and minimizing the losses caused by risk events.

[0221] Optionally, this application provides a possible implementation method for obtaining related knowledge from the emergency knowledge graph based on risk event types and risk event elements, including:

[0222] The first step is to extract at least one first entity from the risk event type and the risk event element.

[0223] Among them, the first entity refers to the core element that is most basic and directly determines the occurrence and development of a risk event. It is the key factor that triggers or promotes the risk event.

[0224] It's understandable that different risk event types and risk event elements may include multiple factors, but the most important entity usually determines the direction of the event. Extracting the first entity can help the system quickly understand the essence of the risk event from a macro perspective.

[0225] For example, when the risk event type is severe abnormal environment, and the risk event elements are severe abnormal environment, abnormal weather phenomenon, affected area, risk level, event implementer, and affected object, the first entity may include severe abnormal environment, abnormal weather phenomenon, and risk level;

[0226] When the risk event type is emergency rescue and the risk event elements are emergency rescue type, rescue object, event implementer and event victim, the first entity may include emergency rescue and rescue object.

[0227] In the second step, the knowledge corresponding to the first entity and / or the second entity associated with the first entity in the emergency knowledge graph is determined as associated knowledge.

[0228] The second entity refers to an element that is closely related to the first entity and plays a supporting or influencing role in the risk event. It may be a direct response or result of the first entity, or a factor that has a causal relationship or other close connection with the first entity. For example, if the first entity is "aircraft fire risk event," the associated second entities may be "airport runway," "aviation fuel," and "aircraft model."

[0229] Associated knowledge refers to the second entity and its related knowledge information associated with the first entity in the emergency knowledge graph.

[0230] The purpose of this step is to construct a complete knowledge system by identifying and extracting knowledge related to the first entity and its associated second entity, so as to ensure a better understanding of the interactive relationship between various factors when facing risk events, and provide comprehensive knowledge support for decision-making and emergency response.

[0231] It can be understood that the first entity only represents the most basic core elements in risk events that directly determine the occurrence and development of events, while the second entity expands the scope related to the first entity and involves more factors.

[0232] Therefore, by querying the second entity associated with the first entity in the emergency knowledge graph and determining the second entity and its corresponding knowledge information as associated knowledge, the system can be helped to more comprehensively understand and analyze the multi-dimensional impact of risk events.

[0233] For example, assuming that the first entity is an abnormally bad environment, the second entities associated with "abnormally bad environment" can be found in the emergency knowledge graph, namely "extreme weather conditions" and "meteorological conditions".

[0234] S209: Concatenate the emergency plan output prompt information, risk event type, risk event elements, and plan group into a second prompt word, input the second prompt word into the large language model, and obtain the emergency plan to be executed.

[0235] The large language model may be, for example, a convolutional neural network model.

[0236] The emergency plan output prompt information is used to guide the large language model to output the emergency plan.

[0237] Understandably, risk event handling often requires multiple pieces of information to support decision-making. However, individual prompt information, risk event types, risk event elements, and contingency plans may not be sufficient to form a complete solution.

[0238] Therefore, by integrating the emergency plan output prompt information, risk event types, risk event elements and plan groups, and inputting them into the large language model, the system can be helped to generate emergency plans that are more in line with actual needs based on specific risk situations, thereby reducing the probability of errors and improving processing efficiency.

[0239] For example, suppose the prompt message for the emergency plan output is: Please output the emergency plan for abnormal types of severe environments.

[0240] The risk event type is the abnormal type of severe environment;

[0241] The risk event elements are: {Event type: abnormal type of severe environment; abnormal weather phenomenon: heavy rain, lightning;

[0242] Affected area: the entire airport area; Risk level: Level 2; Incident implementer: Command Center; Affected objects: Passengers;

[0243] The contingency plan includes: {Plan 1: First, the command center notifies all relevant airport departments to halt all takeoffs and landings and suspend all ground services. Second, airport security personnel should organize the evacuation of passengers around the airport, ensuring that passages in all areas are clear and congestion is avoided.};

[0244] Plan 2: First, the command center notifies all relevant airport departments to halt all takeoffs and landings and suspend all ground services. Secondly, for passengers already at the airport, a dedicated service team should be deployed to provide them with drinking water and food, ensuring their basic needs are met.

[0245] Based on the above information, the emergency plan to be implemented is as follows: First, the command center notifies all relevant airport departments to halt all takeoffs and landings and suspend all ground services. Second, airport security personnel should organize the evacuation of passengers around the airport, ensuring that passageways in all areas are clear and congested. Finally, for passengers already at the airport, a dedicated service team should be deployed to provide them with drinking water and food, ensuring that their basic needs are met.

[0246] Optionally, this application provides a possible implementation method, including:

[0247] In the first step, the second prompt word is input into the large language model, and the prompt encoder in the large language model is used to encode the emergency plan output prompt information, risk event type and risk event elements in the second prompt word to obtain a prompt feature vector. The plan encoder in the large language model is used to encode the plan group in the second prompt word to obtain a plan feature vector.

[0248] Among them, the prompt encoder is used to convert the text information in the emergency plan output prompt information, risk event type and risk event elements into a structured feature vector.

[0249] The plan encoder is used to convert complex information such as text descriptions, operation steps, resource scheduling, etc. in the plan group into structured feature vectors.

[0250] The purpose of this step is to convert the text information in the second prompt word into a numerical feature vector that can be processed by the computer system.

[0251] Understandably, the original second prompt word information is typically presented in text form, making it difficult for computer systems to directly process and analyze it. By encoding the emergency plan output prompt information, risk event type, and risk event elements in the prompt word using a prompt encoder to generate a prompt feature vector, the system can understand the semantics and intent of this information, such as clarifying the type of plan currently required, the nature of the risk event, and key elements.

[0252] By encoding the plan group through the plan encoder and obtaining the plan feature vector, the system can grasp the specific content and response strategies in the plan group, thereby providing effective input for subsequent emergency decision-making, plan matching and other tasks, and improving the accuracy and efficiency of the system in handling emergency-related issues.

[0253] In the second step, the emergency plan to be executed is obtained based on the prompt feature vector and the plan feature vector.

[0254] The purpose of this step is to ensure that the emergency plan can be quickly and accurately matched to the actual situation and efficiently executed in complex emergency situations.

[0255] It can be understood that the prompt feature vector is used to provide background information and key features of the current risk event, such as event type, scope of impact, etc.; while the plan feature vector contains the detailed content and implementation steps of the emergency plan corresponding to the risk event, including specific response measures, required resources, etc.

[0256] Therefore, by combining the prompt feature vector and the plan feature vector, the system can accurately identify the current risk event and automatically select the most appropriate emergency plan based on the characteristics of the event to ensure that the correct response measures are taken, thereby improving the efficiency and effectiveness of the emergency response.

[0257] The airport emergency handling method provided in the embodiment of the present application first obtains the real-time information text of the airport and performs word segmentation processing on it, and then inputs the word segmentation results into the pre-trained language model to obtain the word vector of each word. Then, the word segmentation results are input into the topic model to generate the text-topic probability distribution, the topic-word probability distribution and the topic vector of each topic, and the topic vector of each word is determined based on this information. Then, the word vector and the topic vector are spliced ​​to form a text feature vector for each word, and these feature vectors are used to determine the type of risk event. By combining the risk event type, the risk event element template and the real-time information text, a first prompt word is generated and input into the large language model to extract the risk event element. Next, based on the preset emergency knowledge graph and plan library, the relevant emergency plan group is retrieved, and the plan information and the risk event element are spliced ​​into a second prompt word, which is further input into the large language model to generate the emergency plan to be executed.

[0258] This method combines natural language processing, topic models and large language models in an automated manner to comprehensively identify and analyze real-time information text, quickly determine the type of risk event, and accurately retrieve emergency plans. It effectively solves the tedious and inefficient problems faced by existing manual analysis, significantly improves processing speed and accuracy, and can provide scientific and reasonable plans for emergency response in a short period of time.

[0259] Figure 3 Schematic diagram of the airport emergency response process provided for this application Figure 3 .like Figure 3 As shown, this embodiment, based on the above embodiment, describes in detail the implementation process of the emergency knowledge graph. The method includes:

[0260] S301. Acquire multiple emergency knowledge texts.

[0261] Understandably, different types of risk events have different natures, characteristics, and response methods, and therefore, the corresponding emergency knowledge texts are also different. If only emergency knowledge texts for a single type are available, insufficient knowledge reserves may prevent comprehensive and effective responses to complex and changing risk events.

[0262] Therefore, by acquiring multiple emergency knowledge texts, not only can the system's emergency knowledge reserves be enriched, but the system's ability to respond to different types of risk events can also be enhanced, enabling it to respond more flexibly to various complex emergency situations.

[0263] The method of obtaining the multiple emergency knowledge texts in this step may be, for example, obtaining them from an emergency knowledge database, or manually inputting the emergency knowledge texts into the system by professional emergency management personnel. This application does not impose any special restrictions on this.

[0264] S302: For each emergency knowledge text, segment the emergency knowledge text to obtain words in the emergency knowledge text, and extract the original feature vector of each word in the emergency knowledge text.

[0265] Among them, the original feature vector refers to the initial digital representation of each word in the emergency knowledge text in the semantic space.

[0266] Understandably, emergency knowledge texts are often long, consisting of continuous character sequences, making it difficult for computer systems to directly understand the complex semantic information contained within them. Word segmentation can break emergency knowledge texts down into individual words, helping computer systems identify the basic semantic units within the text.

[0267] However, the split words still cannot be used directly for data analysis and processing. In order to enable the computer system to effectively process these words, it is necessary to extract the original feature vector of each word.

[0268] Optionally, the present application provides a possible implementation method for extracting the original feature vector of each word in the emergency knowledge text, including: using a Transformer encoder to extract each word in the emergency knowledge text to obtain the original feature vector of each word.

[0269] It is understandable that the process of Transformer encoder extracting each word is as follows:

[0270] The first step is the word embedding layer: each word in the emergency knowledge text is converted into a word vector as the input of the Transformer encoder.

[0271] The second step is the masked multi-head attention mechanism: multi-head attention processing is performed on the input word vector, allowing the Transformer encoder to focus on different parts in different subspaces and capture the dependencies between words.

[0272] The third step is residual connection and layer normalization: the output of each attention layer is added to the input (residual connection), and then layer normalization is performed to keep the value stable.

[0273] The fourth step is multi-head self-attention: through parallel processing of multiple independent attention heads, the Transformer encoder's understanding of different contexts is further enhanced.

[0274] Step 5: Residual Connection and Layer Normalization: Repeat the above steps to ensure the flow of information and the stability of the Transformer encoder.

[0275] Step 6: Feedforward neural network: Perform further nonlinear transformation on the features of each word to increase the expressive power of the Transformer encoder.

[0276] Step 7, residual connection and layer normalization: Residual connection and layer normalization are applied again to ensure that deep information does not disappear.

[0277] Step 8, linear layer: maps features to a dimension more suitable for the task.

[0278] Step 9: Output feature vector: Finally, output the original feature vector of each word.

[0279] S303. Identify the head entity in the emergency knowledge text, and determine the attention weight of the head entity and each word in the emergency knowledge text. According to the attention weight and original feature vector of each word in the emergency knowledge text, determine the enhanced feature vector of each word in the emergency knowledge text.

[0280] Among them, the head entity refers to the most critical words or phrases in the emergency knowledge text.

[0281] Attention weights are used to represent the degree of association between the head entity and each word in the emergency knowledge text. Attention weights range from 0 to 1. A larger attention weight indicates a greater association between the head entity and a word in the emergency knowledge text; conversely, a smaller attention weight indicates a lower association between the head entity and a word in the emergency knowledge text.

[0282] The purpose of this step in determining the enhanced feature vector of each word in the emergency knowledge text based on the attention weight and original feature vector of each word in the emergency knowledge text is to generate an enhanced feature vector that can more accurately express the importance and relevance of each word in the text by combining the attention weight and original feature vector of each word in the emergency knowledge text.

[0283] It is understandable that the original feature vector only contains the basic semantic information of each word, but it does not fully consider the semantic association between the word and the head entity in the emergency knowledge text. The attention weight can reflect the closeness of the association between each word and the head entity.

[0284] Therefore, by combining the attention weight of each word with the original feature vector, the semantic features of each word can be enriched, which helps the system capture key information from emergency knowledge text more accurately and quickly.

[0285] For example, assuming the head entity is "House Fire Rescue," and the emergency knowledge context includes the following: fire truck, fire extinguishing, aircraft, command center, and airplane. Based on this information, we can first determine that the attention weights of "House Fire Rescue" with fire trucks, fire extinguishing, aircraft, and command center are 0.8, 1, 0, and 0, respectively. We can then determine that "House Fire Rescue" has a strong association with fire trucks, the closest association with fire extinguishing, and a low association with aircraft and command center.

[0286] Optionally, the present application provides an implementation method for determining the attention weight of each word in the head entity and the emergency knowledge text, including:

[0287] In the first step, the head entity is projected according to the learnable first parameter matrix to obtain the head entity corresponding head entity mapping result.

[0288] For example, assuming the header entity is , the first parameter matrix that can be learned is Based on the above information, we can determine that the mapping result of the header entity to the header entity is .

[0289] In the second step, the text word vectors corresponding to the emergency knowledge text are projected through the learnable first parameter matrix and the learnable second parameter matrix to obtain the key vector and the value vector.

[0290] For example, suppose the text word vector corresponding to the emergency knowledge text is , the first parameter matrix that can be learned And the learnable second parameter matrix Based on the above information, we can determine the key vector as , and the value vector is .

[0291] In the third step, the attention weights of the head entity and each word in the emergency knowledge text are determined based on the emergency knowledge text, the head entity mapping result, the key vector and the scaling factor.

[0292] For example, suppose Expressed as a scaling factor, is represented as a key vector, Represented as the header entity mapping result, Represents emergency knowledge text, T={ Then, the attention weights of the head entity and each word in the emergency knowledge text can be determined based on the following formula.

[0293]

[0294] S304: Based on the enhanced feature vector of each word in the emergency knowledge text, extract the tail entity in the emergency knowledge text and determine the relationship between the head entity and the tail entity.

[0295] Among them, the tail entity refers to the target connected to the head entity through a certain relationship in the emergency knowledge text.

[0296] Relationships include: collaborative relationships, dependency relationships, and causal relationships.

[0297] The purpose of this step is to extract the tail entities that are strongly associated with the head entity from the emergency knowledge text and clarify the relationship between them, so as to construct a complete and accurate emergency knowledge triple, namely head entity-relationship-tail entity.

[0298] As you can understand, in emergency knowledge text, the head entity represents the core concept or key element, while the tail entity is other elements that are strongly semantically related to the head entity, and they are connected to each other through certain relationships. By extracting the relationship between these head and tail entities, we can construct a complete emergency knowledge triple—that is, an organic combination of the head entity, the relationship, and the tail entity, thus making emergency knowledge more clear and systematic.

[0299] S305. Construct an emergency knowledge graph based on the head entity, tail entity, and the relationship between the head entity and the tail entity in each emergency knowledge text.

[0300] Among them, the purpose of this step is to build a structured and visual emergency knowledge network to integrate and associate the information scattered in various emergency knowledge texts.

[0301] Understandably, emergency knowledge is often scattered in emergency knowledge texts, and the information is fragmented and difficult to quickly obtain and comprehensively utilize.

[0302] Therefore, by constructing an emergency knowledge graph, the head entity, tail entity and the relationship between them can be clearly presented, which helps the system to quickly and comprehensively grasp the emergency knowledge system, so that when facing sudden emergency events, the system can quickly locate key information based on the knowledge graph, make scientific and reasonable decisions, and improve the efficiency and accuracy of emergency response.

[0303] The airport emergency handling method provided in the embodiment of the present application first obtains multiple emergency knowledge texts, then performs word segmentation processing on each text to obtain the words therein, and extracts the original feature vector of each word; then identifies the head entity in the text, determines the attention weight of the head entity and each word, and obtains the enhanced feature vector of each word based on the weight and the original feature vector; then extracts the tail entity based on the enhanced feature vector, and determines the relationship between the head entity and the tail entity; finally, constructs an emergency knowledge graph based on the head entity, tail entity and their relationship in each text. This method can efficiently extract key information from a large amount of emergency knowledge text, construct an emergency knowledge graph with a clear structure and rich information, provide strong support for applications such as emergency decision-making and knowledge retrieval, and improve the efficiency and accuracy of emergency management.

[0304] Figure 4 The schematic diagram of the airport emergency response device provided for this application is as follows: Figure 4 As shown, the airport emergency handling device 400 provided in this embodiment includes:

[0305] An acquisition module 401 is configured to acquire real-time information text of an airport, wherein the real-time information text at least includes ground-to-air communication text;

[0306] A determination module 402 is configured to determine a risk event type based on the real-time information text;

[0307] An extraction module 403 is configured to extract risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type;

[0308] The determination module 402 further determines the emergency plan to be executed based on the risk event type, the risk event elements, the preset emergency knowledge graph and the plan library.

[0309] In a possible implementation, the apparatus further includes: a processing module 404;

[0310] The processing module 404 is configured to perform word segmentation on the real-time information text and input the word segmentation results into a pre-trained language model to obtain a word vector for each word;

[0311] The device further comprises: an input module 405;

[0312] The input module 405 is used to input the word segmentation results into the topic model to obtain the text-topic probability distribution, topic-word probability distribution and topic vector of each topic corresponding to the real-time information text;

[0313] The determining module 402 is further configured to determine a topic vector corresponding to each word based on the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic;

[0314] The determination module 402 is specifically configured to concatenate the word vector of each word and the topic vector corresponding to each word into a text feature vector of each word, and determine the risk event type based on the text feature vector of each word in the real-time information text.

[0315] In a possible implementation, the determining module 402 is further configured to determine the probability distribution of each word under each topic based on the topic-word probability distribution and the text-topic probability distribution;

[0316] The determining module 402 is specifically configured to determine a topic vector corresponding to each word according to the probability distribution of each word under each topic and the topic vector of each topic.

[0317] In a possible implementation, the acquisition module 401 is further configured to acquire a ground-to-air call recording at the airport;

[0318] The processing module 404 is further configured to perform text processing on the ground-to-air call recording to obtain the ground-to-air call text.

[0319] In a possible implementation, the acquisition module 401 is further configured to acquire flight dynamic data of the airport;

[0320] The device further includes: a conversion module 405;

[0321] The conversion module 405 is configured to convert a specific code in the flight dynamic data into a corresponding flight dynamic text, wherein the real-time information text includes the flight dynamic text;

[0322] The acquisition module 401 is further configured to acquire a weather warning text of the airport, wherein the real-time information text includes the weather warning text;

[0323] The acquisition module 401 is further configured to acquire the security log text of the airport, wherein the real-time information text includes the security log text.

[0324] In a possible implementation, the apparatus further includes: a splicing module 406;

[0325] The splicing module 406 is configured to splice the risk event type, the risk event element template, and the real-time information text into a first prompt word;

[0326] The input module 405 is further configured to input the first prompt word into a large language model to obtain the risk event element.

[0327] In a possible implementation, the apparatus further includes: a retrieval module 407;

[0328] The retrieval module 407 is configured to retrieve a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph;

[0329] The splicing module 406 is further configured to splice the emergency plan output prompt information, the risk event type, the risk event element, and the plan group into a second prompt word;

[0330] The input module 405 is further configured to input the second prompt word into the large language model to obtain the emergency plan to be executed.

[0331] In a possible implementation, the acquisition module 401 is further configured to acquire associated knowledge from the emergency knowledge graph according to the risk event type and the risk event elements;

[0332] The extraction module 403 is further configured to extract risk features from the risk event type, the risk event elements, and the associated knowledge, and determine a plan group matching the risk features from the plan library.

[0333] In a possible implementation, the extraction module 403 is further configured to extract at least one first entity from the risk event type and the risk event element;

[0334] The determination module 402 is specifically configured to determine the knowledge corresponding to the first entity and / or the second entity associated with the first entity in the emergency knowledge graph as the associated knowledge.

[0335] In one possible implementation, the input module 405 is further configured to input the second prompt word into a large language model, encode the emergency plan output prompt information, the risk event type, and the risk event elements in the second prompt word using a prompt encoder in the large language model to obtain a prompt feature vector, and encode the plan group in the second prompt word using a plan encoder in the large language model to obtain a plan feature vector.

[0336] The determining module 402 is specifically configured to obtain the emergency plan to be executed based on the prompt feature vector and the plan feature vector.

[0337] In a possible implementation, the acquisition module 401 is further configured to acquire a plurality of emergency knowledge texts;

[0338] The apparatus further comprises: a word segmentation module 408;

[0339] The word segmentation module 408 is used to segment each emergency knowledge text to obtain words in the emergency knowledge text;

[0340] The extraction module 403 is further used to extract the original feature vector of each word in the emergency knowledge text;

[0341] The apparatus further includes: an identification module 409;

[0342] The identification module 409 is used to identify the head entity in the emergency knowledge text;

[0343] The determination module 402 is further configured to determine the attention weight of each word in the head entity and the emergency knowledge text, and determine an enhanced feature vector for each word in the emergency knowledge text based on the attention weight and the original feature vector of each word in the emergency knowledge text;

[0344] The determining module 402 is further configured to extract a tail entity in the emergency knowledge text based on the enhanced feature vector of each word in the emergency knowledge text, and determine a relationship between the head entity and the tail entity;

[0345] The apparatus further includes: a construction module 410;

[0346] The construction module 410 is used to construct the emergency knowledge graph based on the head entity, the tail entity and the relationship between the head entity and the tail entity in each of the emergency knowledge texts.

[0347] The airport emergency handling device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effects are similar, and are not described in detail in this embodiment.

[0348] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 5 As shown, the electronic device 500 provided in this embodiment includes: at least one processor 501 and a memory 502. Optionally, the device 500 further includes a communication component 503. The processor 501, the memory 502 and the communication component 503 are connected via a bus 504.

[0349] In a specific implementation process, at least one processor 501 executes the computer-executable instructions stored in the memory 502, so that the at least one processor 501 performs the above method.

[0350] The specific implementation process of the processor 501 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0351] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0352] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage.

[0353] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0354] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0355] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0356] The readable storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0357] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0358] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, either through an interface, electrical, mechanical, or other means.

[0359] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0360] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0361] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0362] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0363] Finally, it should be noted that those skilled in the art will readily identify other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. The present invention is not limited to the precise structure described above and illustrated in the accompanying drawings, and various modifications and variations may be made without departing from the scope thereof. The scope of the present invention is limited solely by the appended claims.

Claims

1. An airport emergency handling method, characterized in that: include: Acquire real-time information text of the airport, and determine the risk event type based on the real-time information text, wherein the real-time information text at least includes ground-to-air communication text; Extracting risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type; Based on the risk event type, the risk event elements, the preset emergency knowledge graph and the plan library, the emergency plan to be executed is determined.

2. The method according to claim 1, characterized in that The determining of the risk event type based on the real-time information text includes: After word segmentation processing is performed on the real-time information text, the word segmentation results are input into a pre-trained language model to obtain a word vector for each word; Inputting the word segmentation results into a topic model to obtain a text-topic probability distribution, a topic-word probability distribution, and a topic vector for each topic corresponding to the real-time information text; Determine a topic vector corresponding to each word according to the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic; The word vector of each word and the topic vector corresponding to each word are spliced ​​into a text feature vector of each word, and the risk event type is determined based on the text feature vector of each word in the real-time information text.

3. The method according to claim 2, characterized in that Determining the topic vector corresponding to each word according to the topic-word probability distribution, the text-topic probability distribution, and the topic vectors of each topic includes: Determine the probability distribution of each word under each topic based on the topic-word probability distribution and the text-topic probability distribution; The topic vector corresponding to each word is determined based on the probability distribution of each word under each topic and the topic vector of each topic.

4. The method according to any one of claims 1 to 3, characterized in that The method of obtaining the real-time information text of the airport includes: A ground-to-air call recording of the airport is obtained, and the ground-to-air call recording is converted into text to obtain the ground-to-air call text.

5. The method according to claim 4, characterized in that The obtaining of the real-time information text of the airport further includes at least one of the following: Acquiring flight dynamic data of the airport, and converting specific codes in the flight dynamic data into corresponding flight dynamic text, wherein the real-time information text includes the flight dynamic text; Obtaining a weather warning text for the airport, wherein the real-time information text includes the weather warning text; A security log text of the airport is obtained, where the real-time information text includes the security log text.

6. The method according to claim 1, characterized in that The extracting risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type includes: splicing the risk event type, the risk event element template, and the real-time information text into a first prompt word; The first prompt word is input into a large language model to obtain the risk event element.

7. The method according to claim 1, characterized in that The step of determining the emergency plan to be executed based on the risk event type, the risk event elements, the preset emergency knowledge graph, and the emergency plan library includes: Retrieving a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph; The emergency plan output prompt information, the risk event type, the risk event elements and the plan group are spliced ​​into a second prompt word, and the second prompt word is input into a large language model to obtain the emergency plan to be executed.

8. The method according to claim 7, characterized in that The retrieving a plan group related to the risk event from the plan library based on the risk event type, the risk event elements, and a preset emergency knowledge graph includes: Acquire related knowledge from the emergency knowledge graph according to the risk event type and the risk event elements; A risk feature is extracted from the risk event type, the risk event elements and the associated knowledge, and a plan group matching the risk feature is determined from the plan library.

9. The method according to claim 8, characterized in that The acquiring of associated knowledge from the emergency knowledge graph according to the risk event type and the risk event elements includes: Extracting at least one first entity from the risk event type and the risk event elements; The knowledge corresponding to the first entity and / or the second entity associated with the first entity in the emergency knowledge graph is determined as the associated knowledge.

10. The method according to claim 7, characterized in that The step of splicing the emergency plan output prompt information, the risk event type, the risk event elements, and the plan group into a second prompt word, and inputting the second prompt word into a large language model to obtain the emergency plan to be executed includes: Inputting the second prompt word into a large language model, encoding the emergency plan output prompt information, the risk event type, and the risk event elements in the second prompt word using a prompt encoder in the large language model to obtain a prompt feature vector, and encoding the plan group in the second prompt word using a plan encoder in the large language model to obtain a plan feature vector; The emergency plan to be executed is obtained based on the prompt feature vector and the plan feature vector.

11. The method according to any one of claims 1 to 3, characterized in that Also includes: Obtain multiple emergency knowledge texts; For each emergency knowledge text, segment the emergency knowledge text to obtain words in the emergency knowledge text, and extract the original feature vector of each word in the emergency knowledge text; Identifying a head entity in the emergency knowledge text, and determining an attention weight of the head entity and each word in the emergency knowledge text, and determining an enhanced feature vector for each word in the emergency knowledge text based on the attention weight and the original feature vector of each word in the emergency knowledge text; Extracting a tail entity from the emergency knowledge text based on the enhanced feature vector of each word in the emergency knowledge text, and determining a relationship between the head entity and the tail entity; The emergency knowledge graph is constructed based on the head entity, the tail entity, and the relationship between the head entity and the tail entity in each of the emergency knowledge texts.

12. An airport emergency handling device, characterized in that: include: An acquisition module, configured to acquire real-time information text of the airport, wherein the real-time information text at least includes ground-to-air communication text; A determination module, configured to determine a risk event type based on the real-time information text; An extraction module, configured to extract risk event elements from the real-time information text based on the risk event type and the risk event element template corresponding to the risk event type; The determination module is further used to determine the emergency plan to be executed based on the risk event type, the risk event elements, the preset emergency knowledge graph and the plan library.

13. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the method according to any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 11 when executed by a processor.

15. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 11 when the computer program is executed by a processor.

Images