A method for air traffic control instruction recognition and aircraft behavior early warning for apron control

Abstract

The present application relates to a kind of air traffic control instruction recognition and aircraft behavior early warning method for apron control, belong to air traffic management technical field, first using speech recognition model, the air traffic control instruction voice of apron control is transcribed into text, the text identified is then recognized intention and slot extraction, accurately extract the behavior category and key parameters in instruction, form instruction analysis result, ensure the accurate understanding of instruction content;Then, through trajectory clustering and association rule, a standardized intention association path set is constructed, combined with airport map, accurate path generation is realized;Finally, the aircraft position monitored in real time is matched with intention path and verified, abnormal behavior is identified and early warning is triggered, the predictability and safety of air traffic control system are improved.The present application realizes accurate abnormal behavior early warning by efficient instruction recognition and intention understanding, combined with behavior knowledge base and real-time position matching.

Description

Technical Field

[0001] This invention relates to the field of air traffic management technology, specifically to a method for air traffic control instruction recognition and aircraft behavior early warning for apron control. Background Technology

[0002] Apron control refers to the management and control of aircraft pushback, start-up, taxiing, towing, and other activities by apron control units within their ground responsibility area at the airport, to ensure the safe and efficient operation of aircraft.

[0003] Traditional apron control still employs a "human-on-the-loop" air traffic control model, where controllers relay instructions to pilots via radio. However, with increasing flight density and runway layouts becoming more complex, incidents of aircraft deviating from their designated positions within the air traffic control system have been rising annually in recent years, indicating the limitations of the traditional model in complex airport environments. Therefore, introducing multimodal information-assisted technologies to achieve real-time monitoring and optimization of ground aircraft and control tasks, thereby improving the accuracy and efficiency of apron control, has become a critical requirement. Utilizing technological means to assist apron control in improving its accuracy and safety has become a key need.

[0004] Since air traffic control instructions are typically delivered via voice, the first step is to recognize the voice commands, convert the spoken audio into text, and then extract key information to determine the aircraft's expected behavior. This text is then dynamically compared to the actual trajectory, triggering an alert when the aircraft deviates from the instructions, thus achieving real-time verification of aircraft behavior. However, implementing this process faces several challenges. Air traffic control instructions contain specialized vocabulary and unique sentence structures, differing significantly from natural language, making semantic understanding difficult. Establishing a behavioral knowledge base is also challenging. Air traffic control instructions are complex and diverse, and airport surface conditions change frequently. Taxi path generation must consider the dynamic airport environment and ensure the path conforms to the aircraft's status and airport safety regulations. Aircraft behavior early warning based on air traffic control instructions requires the system to monitor deviations between aircraft operations and expected paths in real time, placing extremely high demands on the speed and accuracy of data acquisition and processing to ensure timely detection and response to sudden deviations.

[0005] Existing methods mostly rely on situational information transmitted by aircraft or ground situational background information for single feature description, which makes it difficult to achieve comprehensive security situation analysis, resulting in limited effectiveness of early warning systems in dynamic monitoring and alarm response. Summary of the Invention

[0006] In view of the above problems, this invention provides a method for air traffic control instruction recognition and aircraft behavior early warning for apron control. First, this invention uses a speech recognition model to transcribe apron control instructions from speech into text. The recognized text is then subjected to intent recognition and slot extraction to accurately extract behavioral categories and key parameters from the instructions, forming instruction parsing results and ensuring accurate understanding of the instruction content. Next, a standardized intent-related path set is constructed through trajectory clustering and association rules, combined with an airport map, to achieve accurate path generation. Finally, the real-time monitored aircraft position is matched and verified with the intent path to identify abnormal behavior and trigger early warnings, improving the predictability and safety of the air traffic control system. This invention achieves accurate abnormal behavior early warnings through efficient instruction recognition and intent understanding, combined with a behavioral knowledge base and real-time position matching.

[0007] This invention provides a method for air traffic control instruction recognition and aircraft behavior early warning for apron control, comprising:

[0008] Step S1: Establish the intent association path set corresponding to the air traffic control command;

[0009] Optionally, the specific steps for establishing the intent-associated path set in step S1 include:

[0010] Step S11: Obtain multiple historical air traffic control commands and multiple historical aircraft trajectories. Based on each historical air traffic control command, segment each historical aircraft trajectories to obtain multiple flight trajectories and corresponding command behavior categories.

[0011] Step S12: Perform cluster analysis on multiple flight trajectories and their corresponding command behavior categories to obtain multiple motion patterns;

[0012] Step S13: Establish an intent-related path set based on multiple motion patterns;

[0013] For example, the historical air traffic control instructions in step S11 include instruction behavior category, key parameters, timestamp, geographical location, and speed.

[0014] Exemplary categories of the command behavior include: control handover command, coasting command, release command, and rollout command;

[0015] Specifically, each flight path corresponds to an air traffic control instruction;

[0016] For example, the historical aircraft trajectory can be segmented according to the timestamp or geographical location of each historical air traffic control instruction, with each segment corresponding to an air traffic control instruction, resulting in multiple segments of flight trajectory.

[0017] For example, if a flight path contains both control handover instructions and taxiing instructions, then the path needs to be divided into two segments: the flight path corresponding to the control handover instructions and the flight path corresponding to the taxiing instructions.

[0018] It is understood that the behavior categories of the historical air traffic control instructions mentioned in step S11 include permission, instruction, report, confirmation, request, restriction, information, and status report.

[0019] It is understandable that the motion pattern described in step S12 reflects the motion of the aircraft under a certain type of air traffic control command.

[0020] Optionally, the feature vector of the flight trajectory in step S12 includes velocity, acceleration, heading angle, and turning radius, with the following expressions:

[0021]

[0022]

[0023]

[0024]

[0025] Among them, v t It is the velocity at time t, Δx t Δy represents the positional change along the x-axis at consecutive time points. t It represents the positional change along the y-axis at continuous intervals, a t It is the acceleration at time t, Δv t The velocity change at time t, θ t It is the heading angle at time t, r t It is the turning radius.

[0026] Step S2: Let n = 1. When n = 1, it represents the first empty control command. Let t = 1. When t = 1, it represents the initial time.

[0027] Step S3: Obtain the track point of the nth air traffic control command at time t;

[0028] Based on the high-precision map of the airport surface, candidate road segments are matched for the flight path points at time t of the nth air traffic control instruction, resulting in candidate road segment a at time t of the nth air traffic control instruction. n,t ;

[0029] For example, the candidate road segment at time t of the nth air traffic control instruction is the nearest road segment within a preset distance;

[0030] Step S4: Introduce the segment ID from the candidate segments at time t of the nth air traffic control instruction;

[0031] Step S5: Based on connectivity test and road segment ID, process the candidate road segments at time t of the nth air traffic control instruction to obtain the processed candidate road segments at time t of the nth air traffic control instruction.

[0032] Optionally, the specific steps in step S5 for obtaining the candidate road segment at the nth air traffic control command time t after processing include:

[0033] Determine whether the segment ID of the candidate road segment at time t of the nth air traffic control instruction is consistent with the segment ID of the candidate road segment at the previous time. If they are consistent, proceed to the next step. If they are inconsistent, update the candidate road segment at time t of the nth air traffic control instruction based on the connectivity test. The candidate road segment with the same segment ID as the candidate road segment at the previous time is used as the candidate road segment for the aircraft at time t corresponding to the nth air traffic control instruction, and proceed to the next step.

[0034] Step S6: Based on multimodal fusion classification, obtain the aircraft behavior at time t corresponding to the nth air traffic control instruction and the corresponding intent-related region;

[0035] Obtain key information at time t of the nth air traffic control instruction;

[0036] Furthermore, the key information includes aircraft number and / or control frequency;

[0037] Optionally, the specific steps for obtaining the key information at time t of the nth air traffic control command in step S6 include:

[0038] Step S61: Collect the nth air traffic control instruction information at time t and perform preprocessing to obtain the nth raw speech signal at time t;

[0039] Input the nth original speech signal at time t into the speech recognition model, and output the transcribed text of the nth speech sequence at time t;

[0040] Step S62: Decompose and label the transcribed text of the nth speech information at time t, and input it into a multi-layer Transformer encoder to extract the nth semantic feature at time t;

[0041] Obtain the prediction intent and prediction slot conditions corresponding to the nth semantic feature at time t, which are represented as key information of the n air traffic control instructions at time t;

[0042] It is understood that the stated intent refers to core instructions for aircraft operation, including taxiway path, takeoff clearance, and / or speed.

[0043] For example, a slot-filling-based natural language processing method is used to perform intent recognition and precise labeling of key slots on the transcribed text.

[0044] The slot conditions include aircraft number, taxiing path and / or motion parameters;

[0045] Furthermore, the specific steps in step S62 for obtaining the prediction intent and prediction slot conditions corresponding to the nth semantic feature include:

[0046] Step S621: Decompose the transcribed text of the nth speech information into the nth word sequence;

[0047] Add special markers to the beginning and end of each word in the nth word sequence to obtain the marked nth word sequence;

[0048] For example, the special markers in step S621 include: a marker CLS indicating the start position of the sequence, serving as a global representation of the intended classification; and a marker SEP separating multiple instructions.

[0049] The expressions for each word are:

[0050] h i =E token (w i )+E segment (w i )+E position (w i )

[0051] Among them, h i Let h = [h1, ..., hi, ..., hn] be the input vector of the i-th word, and E be the input vector of the i-th word. token (w i ) is word embedding, indicating word w i semantic information, E segment (w i () is a sentence embedding, distinguishing sentence A from sentence B, E position (w i ) represents the position of the i-th word, and w represents the word. i Position in the sequence.

[0052] Step S622: Input the labeled nth word sequence into a multi-layer Transformer encoder to extract the nth semantic feature;

[0053] It is understood that the multi-layer Transformer encoder includes 12 layers of Transformer encoders, and each layer of Transformer encoder uses a multi-head self-attention mechanism to enhance feature representation capabilities;

[0054] The expression for the nth semantic feature is:

[0055]

[0056] Where Q represents the query vector matrix, K represents the key vector matrix, V represents the value vector matrix, and d k It is the dimension of the key vector matrix, used to scale the attention score.

[0057] Step S623: The natural language processing method based on slot filling obtains the predicted intent and predicted slot conditions corresponding to the nth semantic feature, respectively;

[0058] Furthermore, the semantic features described in step S622 include sentence features and individual word features;

[0059] Furthermore, the specific steps in step S623 for obtaining the nth predicted intent and the nth predicted slot condition corresponding to the nth semantic feature include:

[0060] Input the sentence features into the intent classifier to obtain the probability of each predicted intent corresponding to the sentence features;

[0061] Fill the slots with the features of each word to obtain the probability of each predicted slot condition corresponding to the word;

[0062] Furthermore, the expression for the probability of the predicted intention is:

[0063] P intent =Softmax(W intent ·h CLS +b intent )

[0064] Among them, h CLS It is the feature representation of the CLS, W intent It is the weight matrix of the fully connected layer, b intent It is the bias term of the fully connected layer, P intent This is the probability of predicting an intention.

[0065] Furthermore, the specific steps for obtaining the probabilities of each predicted slot condition corresponding to the word include:

[0066] Each word is passed through a fully connected layer to generate its corresponding initial slot category, expressed as follows:

[0067] P slot =Softmax(W slot ·h t +b slot )

[0068] Among them, P slot It is the probability distribution of slot filling, h t W is the feature representation of the t-th word. slot It is the weight matrix of the fully connected layer, b slot It is the bias term of the fully connected layer.

[0069] The initial slot category corresponding to each word is optimized using a Conditional Random Field (CRF) to obtain the probability of each predicted slot condition for each word, expressed as:

[0070]

[0071] Where P(y|x) is the conditional probability of slot sequence labeling, and ψ(y t ,y t-1 y′ represents the relationship between the current label, the previous label, and the input features. t In conditional random fields, y represents all possible combinations of slot labels used to compute the normalization term. t y is the slot label at the current time t, x is the input sequence, and y is the output sequence.

[0072] Step S7: Based on the key information of the nth air traffic control instruction at time t, search and query the intent association path set to obtain the intent association path of the nth air traffic control instruction at time t;

[0073] Step S8: Verify the candidate road segment at time t of the nth air traffic control instruction after processing in step S5 with the intended target associated path at time t of the nth air traffic control instruction as described in step S7. If there is abnormal behavior, issue an early warning and directly output the alarm result; if there is no abnormality, proceed to the next step.

[0074] Step S9: Determine whether t is greater than or equal to T, where T represents the total number of time points. If yes, take the candidate road segment at time t as the final candidate road segment and proceed to the next step. If no, let t = t + 1 and return to step S2.

[0075] Step S10: Traverse N air traffic control instructions, where N represents the number of air traffic control instructions. Repeat steps S2-S9 to provide behavioral warnings for the candidate road segments corresponding to each air traffic control instruction.

[0076] Specifically, the segment ID of the candidate road segment at time t of the nth air traffic control instruction is obtained and the segment ID of the path associated with the intended target at time t of the nth air traffic control instruction is obtained respectively. When the two segment IDs do not match, an alarm is issued to realize the aircraft behavior verification and early warning.

[0077] This invention is based on auxiliary information such as source and target provided by high-precision maps. It detects the connectivity of taxiway segments by utilizing the preceding and following endpoints of the segment and the correlation between adjacent segments, thereby achieving the task of matching taxiway segments for aircraft.

[0078] Understandably, issuing an alert indicates that the aircraft's behavior has deviated from the controller's instructions;

[0079] Ultimately, this will enable air traffic control instruction recognition and aircraft behavior early warning, ensuring the safety and efficiency of apron operations and improving the safety, flexibility, predictability, and operational efficiency of apron control.

[0080] Compared with the prior art, the present invention has at least the following beneficial effects:

[0081] (1) This invention improves the accuracy of instruction recognition and intent understanding: Based on the intent recognition technology of speech recognition model and BERT-IRSF model, this invention effectively improves the accuracy of speech recognition and the stability of the system in the airport environment, ensuring accurate understanding of instructions;

[0082] (2) This invention organizes historical data into a behavioral knowledge base through trajectory clustering and association rule technology, and generates an intent-related path set by combining it with a high-precision airport map, ensuring the accurate description of the instruction intent in the knowledge base and providing a standardized path reference for real-time path verification.

[0083] (3) This invention achieves multi-source information fusion and real-time early warning: This invention combines the real-time monitored aircraft position with the generated intention path set for dynamic matching, quickly identifies deviation behavior and triggers early warning, improves the response speed and accuracy of early warning, helps controllers and pilots to deal with potential risks in a timely manner, and significantly improves the safety and controllability of apron control. Attached Figure Description

[0084] The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0085] Figure 1 This is a schematic diagram of the process of the air traffic control instruction recognition and aircraft behavior early warning method for apron control in an embodiment of the present invention;

[0086] Figure 2 This is a schematic diagram of the flowchart for the accurate extraction of air traffic control command information based on noise reduction recognition and semantic understanding in an embodiment of the present invention;

[0087] Figure 3 This is a schematic diagram of the process for constructing an aircraft surface behavior knowledge base under the field monitoring trajectory representation in an embodiment of the present invention;

[0088] Figure 4 This is a schematic diagram of the specific flowchart for aircraft behavior early warning in an embodiment of the present invention. Detailed Implementation

[0089] To better understand the above-described objectives, features, and advantages of the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other. Furthermore, the present invention can be implemented in other ways different from those described herein; therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0090] A specific embodiment of the present invention, such as Figure 1-4 This invention discloses a method for air traffic control instruction recognition and aircraft behavior early warning for apron control. To illustrate the effectiveness of the proposed method, a specific embodiment is provided below for detailed explanation of the above technical solution. The specific implementation steps are as follows:

[0091] Step S1: Establish the intent association path set corresponding to the air traffic control command;

[0092] Optionally, the specific steps for establishing the intent-associated path set in step S1 include:

[0093] Step S11: Obtain multiple historical air traffic control commands and multiple historical aircraft trajectories. Based on each historical air traffic control command, segment each historical aircraft trajectories to obtain multiple flight trajectories and corresponding command behavior categories.

[0094] Step S12: Perform cluster analysis on multiple flight trajectories and their corresponding command behavior categories to obtain multiple motion patterns;

[0095] Step S13: Establish an intent-related path set based on multiple motion patterns;

[0096] For example, the historical air traffic control instructions in step S11 include instruction behavior category, key parameters, timestamp, geographical location, and speed.

[0097] Exemplary instruction behavior categories include: control handover instructions, coasting instructions, release instructions, and rollout instructions; as shown in Table 1;

[0098] Specifically, each flight path corresponds to an air traffic control instruction;

[0099] For example, the historical aircraft trajectory can be segmented according to the timestamp or geographical location of each historical air traffic control instruction, with each segment corresponding to an air traffic control instruction, resulting in multiple segments of flight trajectory.

[0100] For example, if a flight path contains both control handover instructions and taxiing instructions, then the path needs to be divided into two segments: the flight path corresponding to the control handover instructions and the flight path corresponding to the taxiing instructions.

[0101] It is understood that the behavior categories of the historical air traffic control instructions mentioned in step S11 include permission, instruction, report, confirmation, request, restriction, information, and status report.

[0102] Table 1. Knowledge Base of Aircraft Surface Behavior Represented by Airport Monitoring Trajectory

[0103]

[0104] It is understandable that the motion pattern described in step S12 reflects the motion of the aircraft under a certain type of air traffic control command.

[0105] Optionally, the feature vector of the flight trajectory in step S12 includes velocity, acceleration, heading angle, and turning radius, with the following expressions:

[0106]

[0107]

[0108]

[0109]

[0110] Among them, v t It is the velocity at time t, Δx t Δy represents the positional change along the x-axis at consecutive time points. t It represents the positional change along the y-axis at continuous intervals, a t It is the acceleration at time t, Δv t The velocity change at time t, θ t It is the heading angle at time t, r t It is the turning radius.

[0111] Step S2: Let n = 1. When n = 1, it represents the first empty control command. Let t = 1. When t = 1, it represents the initial time.

[0112] Step S3: Obtain the track point of the nth air traffic control command at time t;

[0113] Based on the high-precision map of the airport surface, candidate road segments are matched for the flight path points at time t of the nth air traffic control instruction, resulting in candidate road segment a at time t of the nth air traffic control instruction. n,t ;

[0114] For example, the candidate road segment at time t of the nth air traffic control instruction is the nearest road segment within a preset distance;

[0115] Step S4: Introduce the segment ID from the candidate segments at time t of the nth air traffic control instruction;

[0116] Step S5: Based on connectivity test and road segment ID, process the candidate road segments at time t of the nth air traffic control instruction to obtain the processed candidate road segments at time t of the nth air traffic control instruction.

[0117] Optionally, the specific steps in step S5 for obtaining the candidate road segment at the nth air traffic control command time t after processing include:

[0118] Determine whether the segment ID of the candidate road segment at time t of the nth air traffic control instruction is consistent with the segment ID of the candidate road segment at the previous time. If they are consistent, proceed to the next step. If they are inconsistent, update the candidate road segment at time t of the nth air traffic control instruction based on the connectivity test. The candidate road segment with the same segment ID as the candidate road segment at the previous time is used as the candidate road segment for the aircraft at time t corresponding to the nth air traffic control instruction, and proceed to the next step.

[0119] Step S6: Based on multimodal fusion classification, obtain the aircraft behavior at time t corresponding to the nth air traffic control instruction and the corresponding intent-related region;

[0120] Obtain key information at time t of the nth air traffic control instruction;

[0121] Furthermore, the key information includes aircraft number and / or control frequency;

[0122] Optionally, the specific steps for obtaining the key information at time t of the nth air traffic control command in step S6 include:

[0123] Step S61: Collect the nth air traffic control instruction information at time t and perform preprocessing to obtain the nth raw speech signal at time t;

[0124] Input the nth original speech signal at time t into the speech recognition model, and output the transcribed text of the nth speech sequence at time t;

[0125] Step S62: Decompose and label the transcribed text of the nth speech information at time t, and input it into a multi-layer Transformer encoder to extract the nth semantic feature at time t;

[0126] Obtain the prediction intent and prediction slot conditions corresponding to the nth semantic feature at time t, which are represented as key information of the n air traffic control instructions at time t;

[0127] It is understood that the stated intent refers to core instructions for aircraft operation, including taxiway path, takeoff clearance, and / or speed.

[0128] For example, a slot-filling-based natural language processing method is used to perform intent recognition and precise labeling of key slots on the transcribed text.

[0129] The slot conditions include aircraft number, taxiing path and / or motion parameters;

[0130] Furthermore, the specific steps in step S62 for obtaining the prediction intent and prediction slot conditions corresponding to the nth semantic feature include:

[0131] Step S621: Decompose the transcribed text of the nth speech information into the nth word sequence;

[0132] Add special markers to the beginning and end of each word in the nth word sequence to obtain the marked nth word sequence;

[0133] For example, the special markers in step S621 include: a marker CLS indicating the start position of the sequence, serving as a global representation of the intended classification; and a marker SEP separating multiple instructions.

[0134] The expressions for each word are:

[0135] h i =E token (w i )+E segment (w i )+E position (w i )

[0136] Among them, h i Let h = [h1, ..., hi, ..., hn] be the input vector of the i-th word, and E be the input vector of the i-th word. token (w i ) is word embedding, indicating word w i semantic information, E segment (w i () is a sentence embedding, distinguishing sentence A from sentence B, E position (w i ) represents the position of the i-th word, and w represents the word. i Position in the sequence.

[0137] Step S622: Input the labeled nth word sequence into a multi-layer Transformer encoder to extract the nth semantic feature;

[0138] It is understood that the multi-layer Transformer encoder includes 12 layers of Transformer encoders, and each layer of Transformer encoder uses a multi-head self-attention mechanism to enhance feature representation capabilities;

[0139] The expression for the nth semantic feature is:

[0140]

[0141] Where Q = WQ ·h, K=W K ·h,V=W V ·h

[0142] It is a Query, Key, and Value matrix, obtained from the input features h through a linear transformation, d k It is the dimension of the Key matrix, used to scale the attention score.

[0143] Step S623: The natural language processing method based on slot filling obtains the predicted intent and predicted slot conditions corresponding to the nth semantic feature, respectively;

[0144] Furthermore, the semantic features described in step S622 include sentence features and individual word features;

[0145] Furthermore, the specific steps in step S623 for obtaining the nth predicted intent and the nth predicted slot condition corresponding to the nth semantic feature include:

[0146] Input the sentence features into the intent classifier to obtain the probability of each predicted intent corresponding to the sentence features;

[0147] Fill the slots with the features of each word to obtain the probability of each predicted slot condition corresponding to the word;

[0148] Furthermore, the expression for the probability of the predicted intention is:

[0149] P intent =Softmax(W intent ·h CLS +b intent )

[0150] Among them, h CLS It is the feature representation of the CLS, W intent It is the weight matrix of the fully connected layer, b intent It is the bias term of the fully connected layer, P inttnt This is the probability of predicting an intention.

[0151] Furthermore, the specific steps for obtaining the probabilities of each predicted slot condition corresponding to the word include:

[0152] Each word is passed through a fully connected layer to generate its corresponding initial slot category, expressed as follows:

[0153] P slot =Softmax(W slot ·h t +b slot )

[0154] Among them, P slot It is the probability distribution of slot filling, ht W is the feature representation of the t-th word. slot It is the weight matrix of the fully connected layer, b slot It is the bias term of the fully connected layer.

[0155] The initial slot category corresponding to each word is optimized using a Conditional Random Field (CRF) to obtain the probability of each predicted slot condition for each word, expressed as:

[0156]

[0157] Where P(y|x) is the conditional probability of slot sequence labeling, and ψ(y t ,y t-1 y′ represents the relationship between the current label, the previous label, and the input features. t In conditional random fields, y represents all possible combinations of slot labels used to compute the normalization term. t y is the slot label at the current time t, x is the input sequence, and y is the output sequence.

[0158] Step S7: Based on the key information of the nth air traffic control instruction at time t, search and query the intent association path set to obtain the intent association path of the nth air traffic control instruction at time t;

[0159] Step S8: Verify the candidate road segment at time t of the nth air traffic control instruction after processing in step S5 with the intended target associated path at time t of the nth air traffic control instruction as described in step S7. If there is abnormal behavior, issue an early warning and directly output the alarm result; if there is no abnormality, proceed to the next step.

[0160] Step S9: Determine whether t is greater than or equal to T, where T represents the total number of time points. If yes, take the candidate road segment at time t as the final candidate road segment and proceed to the next step. If no, let t = t + 1 and return to step S2.

[0161] Step S10: Traverse N air traffic control instructions, where N represents the number of air traffic control instructions. Repeat steps S2-S9 to provide behavioral warnings for the candidate road segments corresponding to each air traffic control instruction.

[0162] Specifically, the segment ID of the candidate road segment at time t of the nth air traffic control instruction is obtained and the segment ID of the path associated with the intended target at time t of the nth air traffic control instruction is obtained respectively. When the two segment IDs do not match, an alarm is issued to realize the aircraft behavior verification and early warning.

[0163] This invention is based on auxiliary information such as source and target provided by high-precision maps. It detects the connectivity of taxiway segments by utilizing the preceding and following endpoints of the segment and the correlation between adjacent segments, thereby achieving the task of matching taxiway segments for aircraft.

[0164] Understandably, issuing an alert indicates that the aircraft's behavior has deviated from the controller's instructions;

[0165] Ultimately, this will enable air traffic control instruction recognition and aircraft behavior early warning, ensuring the safety and efficiency of apron operations and improving the safety, flexibility, predictability, and operational efficiency of apron control.

[0166] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for air traffic control instruction recognition and aircraft behavior early warning for apron control, characterized in that, include: Step S1: Establish the intent association path set corresponding to the air traffic control command; Step S2: Let n = 1. When n = 1, it represents the first empty control command. Let t = 1. When t = 1, it represents the initial time. Step S3: Obtain the flight path of the aircraft corresponding to the nth air traffic control instruction at time t and perform candidate route matching to obtain the candidate route of the aircraft corresponding to the nth air traffic control instruction at time t. Step S4: Introduce a road segment ID from the candidate road segments at aircraft time t corresponding to the nth air traffic control instruction; Step S5: Based on connectivity test and road segment ID, process the candidate road segments at time t of the nth air traffic control instruction to obtain the processed candidate road segments at time t of the nth air traffic control instruction. Step S6: Based on multimodal fusion classification, obtain the aircraft behavior at time t corresponding to the nth air traffic control instruction and the corresponding intent-related region; Obtain key information at time t of the nth air traffic control instruction; Step S7: Based on the key information of the nth air traffic control instruction at time t, search and query the intent association path set to obtain the intent association path of the nth air traffic control instruction at time t; Step S8: Verify the candidate road segment at time t of the nth air traffic control instruction after processing in step S5 with the intended target associated path at time t of the nth air traffic control instruction as described in step S7. If there is abnormal behavior, issue a behavior warning and output the alarm result. If no abnormalities are found, proceed to the next step; Step S9: Determine whether t is greater than or equal to T, where T represents the total number of time points. If yes, take the candidate road segment at time t as the final candidate road segment and proceed to the next step. If no, let t = t + 1 and return to step S2. Step S10: Traverse N air traffic control instructions, where N represents the number of air traffic control instructions. Repeat steps S2-S9 to provide behavioral warnings for the candidate road segments corresponding to each air traffic control instruction.

2. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 1, characterized in that, The specific steps for establishing the intent-related path set in step S1 include: Step S11: Obtain multiple historical air traffic control commands and multiple historical aircraft trajectories. Based on each historical air traffic control command, segment each historical aircraft trajectories to obtain multiple flight trajectories and corresponding command behavior categories. Step S12: Perform cluster analysis on multiple flight trajectories and their corresponding command behavior categories to obtain multiple motion patterns; Step S13: Establish an intent-related path set based on multiple motion patterns.

3. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 2, characterized in that, The historical air traffic control instructions mentioned in step S11 include instruction behavior category, key parameters, timestamp, geographical location, and speed; The command behavior categories include: control handover command, coasting command, release command, and rollout command.

4. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 1, characterized in that, The specific steps for obtaining the key information at time t of the nth air traffic control command, as described in step S6, include: Step S61: Collect the nth air traffic control instruction information at time t and perform preprocessing to obtain the nth raw speech signal at time t; Input the nth original speech signal at time t into the speech recognition model, and output the transcribed text of the nth speech sequence at time t; Step S62: Decompose and label the transcribed text of the nth speech information at time t, and input it into a multi-layer Transformer encoder to extract the nth semantic feature at time t; Obtain the prediction intent and prediction slot conditions corresponding to the nth semantic feature at time t, which are represented as key information of the n air traffic control instructions at time t.

5. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 1, characterized in that, The intent refers to the core instructions for aircraft operation, including taxiing path, takeoff clearance, and / or speed.

6. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 4, characterized in that, The slot conditions include aircraft number, taxiing path and / or motion parameters.

7. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 4, characterized in that, Step S62, which describes obtaining the prediction intent and prediction slot conditions corresponding to the nth semantic feature, includes the following specific steps: Step S621: Decompose the transcribed text of the nth speech information into the nth word sequence; Add special markers to the beginning and end of each word in the nth word sequence to obtain the marked nth word sequence; Step S622: Input the labeled nth word sequence into a multi-layer Transformer encoder to extract the nth semantic feature; Step S623: The natural language processing method based on slot filling obtains the predicted intent and predicted slot conditions corresponding to the nth semantic feature.

8. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 7, characterized in that, The semantic features described in step S622 include sentence features and word features.

9. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 7, characterized in that, Step S623, which describes obtaining the nth predicted intent and the nth predicted slot condition corresponding to the nth semantic feature, includes the following specific steps: Input the sentence features into the intent classifier to obtain the probability of each predicted intent corresponding to the sentence features; Fill the slots with the features of each word to obtain the probability of each predicted slot condition corresponding to the word.

10. The air traffic control instruction recognition and aircraft behavior early warning method for apron control according to claim 1, characterized in that, The specific steps for conducting behavioral warnings as described in step S8 include: Obtain the segment ID of the candidate road segment at time t of the nth air traffic control instruction and the segment ID of the path associated with the intended target at time t of the nth air traffic control instruction. Issue an alarm when the two segment IDs do not match.

Images