A method and apparatus for predicting flight target trajectory based on Geohash algorithm

By combining the Geohash algorithm with a large-scale language model, the problems of accuracy and data processing in flight target trajectory prediction are solved, achieving high-precision flight path prediction and effective processing of complex trajectory data.

CN118779788BActive Publication Date: 2026-11-17CHINA ACADEMY OF ELECTRONICS AND INFORMATION TECHNOLOGY OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410795484.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-19
Publication Date
2026-11-17
Estimated Expiration
2044-06-19

AI Technical Summary

Technical Problem

Existing flight target trajectory prediction methods are not very accurate when faced with complex and ever-changing flight conditions. They are unable to accurately capture the details and long-term dependencies of complex dynamic systems. Furthermore, the raw latitude and longitude coordinate data is not suitable for direct use in training large models, which limits data processing and prediction performance.

Method used

The Geohash algorithm is used to convert the original trajectory data into character data. Combined with a pre-trained model of a large-scale language model such as the Transformer architecture, a high-precision flight target trajectory prediction map is generated through outlier detection and cleaning, and model fine-tuning.

Benefits of technology

It improves the accuracy of flight target trajectory prediction and the ability to process large-scale complex trajectory data, achieving efficient data processing and accurate prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118779788B_ABST
    Figure CN118779788B_ABST
Patent Text Reader

Abstract

This application discloses a method and apparatus for predicting flight target trajectories based on the Geohash algorithm. The method includes: performing outlier detection and cleaning on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data; using the Geohash algorithm to convert the original geographic coordinates into character-type data for large-scale model training, generating a character-type trajectory dataset; constructing a large-scale model training process based on the character-type trajectory dataset, and using a pre-trained model with a target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target; fine-tuning the large model based on a typical target trajectory prediction sample set to obtain the target model; generating character-type trajectory data corresponding to the flight target based on the target model, and using Geohash reverse decoding to convert the character-type trajectory data into target geographic coordinates, and generating a trajectory prediction map of the flight target. This method has the advantages of strong data processing capabilities, good real-time performance, high accuracy, and good generalization ability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of aviation information technology, and in particular to a method for predicting flight target trajectories based on the Geohash algorithm. Background Technology

[0002] Current predictions of flight target trajectories mainly rely on relatively traditional mathematical models and statistical analysis methods. However, due to their low accuracy, especially when facing complex and ever-changing flight conditions, such as flight derailments caused by weather, route adjustments under air traffic control orders, and the influence of aircraft performance, the prediction results based on traditional methods often deviate from the actual trajectories. Therefore, they have limitations when dealing with rapidly developing aviation information technology and massive amounts of flight data.

[0003] Therefore, how to provide a method for predicting flight target trajectories, improve the accuracy of trajectory prediction, and enhance the ability to process large-scale complex trajectory data has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a flight target trajectory prediction method based on the Geohash algorithm, a flight target trajectory prediction device based on the Geohash algorithm, an electronic device, and a computer storage medium, which are used to improve the accuracy of trajectory prediction and enhance the ability to process large-scale complex trajectory data.

[0005] In a first aspect of this application, a method for predicting the trajectory of a flight target based on the Geohash algorithm is provided, comprising: Anomaly detection and cleaning are performed on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data. The original trajectory data includes first trajectory data corresponding to the flight target collected in real time, and / or second trajectory data corresponding to the flight target stored in advance. The original geographic coordinates are converted into character-type data for large model training using the Geohash algorithm, generating a character-type trajectory dataset; A large model training process based on the aforementioned character-type trajectory dataset is constructed, and the spatiotemporal patterns of the flight trajectory corresponding to the flight target are mined using the pre-trained model of the target neural network architecture. Based on the typical target trajectory prediction sample set, the large model is fine-tuned to obtain the target model. The typical target trajectory prediction sample set is a sample set generated based on flight trajectories related to the flight target in historical data. Based on the target model, character-type trajectory data corresponding to the flight target is generated, and the character-type trajectory data is converted into target geographic coordinates through Geohash reverse decoding. Based on the target's geographic coordinates, a trajectory prediction map of the flight target is generated.

[0006] In a second aspect of this application, a flight target trajectory prediction device based on the Geohash algorithm is provided, comprising: The cleaning module is configured to perform outlier detection and cleaning on the raw trajectory data to obtain the original geographic coordinates corresponding to the raw trajectory data. The first conversion module is configured to use the Geohash algorithm to convert the original geographic coordinates into character data for large model training, generating a character trajectory dataset. The construction module is configured to build a large model training process based on the character-type trajectory dataset, and to use the pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target. The fine-tuning module is configured to fine-tune the large model based on a typical target trajectory prediction sample set to obtain the target model. The typical target trajectory prediction sample set is a sample set generated based on flight trajectories related to the flight target in historical data. The second conversion module is configured to generate character-type trajectory data corresponding to the flight target based on the target model, and convert the character-type trajectory data into target geographic coordinates through Geohash reverse decoding; The generation module is configured to generate a trajectory prediction map of the flight target based on the target's geographic coordinates.

[0007] In a third aspect of the embodiments of this application, a computer-readable storage medium is provided, which stores computer-executable instructions that, when executed by a processor, implement the steps of the above-described method for predicting flight target trajectories based on the Geohash algorithm.

[0008] In a fourth aspect of this application, a computer device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the flight target trajectory prediction method based on the Geohash algorithm described above.

[0009] This application provides a method for predicting flight target trajectories based on the Geohash algorithm. First, outlier detection and cleaning are performed on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data. The original trajectory data includes first trajectory data corresponding to the flight target collected in real time, and / or second trajectory data corresponding to the flight target pre-stored. Then, the Geohash algorithm is used to convert the original geographic coordinates into character-type data for large-scale model training, generating a character-type trajectory dataset. Next, a large-scale model training process is constructed based on the character-type trajectory dataset, using a pre-trained model with a target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target. Then, based on a typical target trajectory prediction sample set, the large model is fine-tuned to obtain a target model. The typical target trajectory prediction sample set is a sample set generated based on flight trajectories related to the flight target from historical data. Next, based on the target model, character-type trajectory data corresponding to the flight target is generated, and the character-type trajectory data is converted into target geographic coordinates through reverse decoding using Geohash. Finally, a trajectory prediction map of the flight target is generated based on the target geographic coordinates.

[0010] The flight target trajectory prediction method based on the Geohash algorithm provided in this application combines the Geohash algorithm with a large-scale language model to predict the flight target trajectory. It has the advantages of strong data processing capability, good real-time performance, high accuracy and good generalization ability.

[0011] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0012] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating a flight target trajectory prediction method based on the Geohash algorithm provided in an embodiment of this application; Figure 2 This is a block diagram of a flight target trajectory prediction device based on the Geohash algorithm provided in an embodiment of this application; Figure 3 This is a structural block diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0013] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0014] Existing technologies face numerous challenges and limitations in predicting flight target trajectories. Traditional methods often suffer from low accuracy when dealing with dynamic trajectory predictions of flight targets, especially when faced with complex and ever-changing flight conditions, such as flight course changes due to weather, route adjustments under air traffic control instructions, and the influence of aircraft performance factors. The predicted results often deviate from the actual trajectory.

[0015] Furthermore, traditional methods fall short when dealing with large-scale flight trajectory datasets. Their limitations in data processing and analysis restrict the effective mining and utilization of large-scale, high-frequency, and high-dimensional trajectory data. Traditional flight target trajectory prediction methods often fail to fully leverage and utilize the large-model techniques developed in the fields of artificial intelligence and big data processing in recent years.

[0016] Large models, such as neural network models in deep learning, possess powerful learning and generalization capabilities, demonstrating excellent performance when handling complex nonlinear relationships. However, existing flight trajectory prediction methods have not successfully integrated these advanced technologies, thus failing to fully leverage the advantages of large models in deep learning spatiotemporal sequence data.

[0017] Furthermore, the raw flight trajectory data is presented in latitude and longitude coordinates. This continuous numerical data structure is not directly applicable to the training framework of most large models. Large models typically expect the input data to undergo some preprocessing and feature transformation to facilitate the model's understanding and learning of the underlying patterns. However, in specific application scenarios of flight trajectory prediction, directly inputting raw latitude and longitude data into the model may result in the model's inability to effectively capture the spatial and temporal relationships between trajectories, further affecting the prediction results.

[0018] Therefore, existing methods for predicting flight target trajectories often suffer from the following technical drawbacks: insufficient accuracy, as existing methods often struggle to accurately capture the details and long-term dependencies of complex dynamic systems, resulting in limited prediction accuracy; data processing difficulties, as traditional methods struggle to efficiently process and extract useful information from large-scale, high-frequency flight trajectory data; and unsuitability for large model training, as the raw latitude and longitude coordinate data is not suitable for direct use in large model training and requires appropriate preprocessing and feature transformation.

[0019] This invention aims to solve this problem by proposing a novel flight target trajectory prediction method based on a combination of Geohash transformation and large model technology. By characterizing the trajectory data and training it with a large model, the accuracy of trajectory prediction and the ability to process large-scale complex trajectory data are effectively improved.

[0020] See Figure 1 , Figure 1 This is a flowchart illustrating a flight target trajectory prediction method based on the Geohash algorithm provided in an embodiment of this application. Figure 1 As shown, the specific steps include:

[0021] Step S102: Perform outlier detection and cleaning on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data. The original trajectory data includes the first trajectory data corresponding to the flight target collected in real time, and / or the second trajectory data corresponding to the flight target stored in advance.

[0022] Step S104: Use the Geohash algorithm to convert the original geographic coordinates into character data for large model training, and generate a character trajectory dataset.

[0023] Step S106: Construct a large model training process based on the character-type trajectory dataset, and use the pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target.

[0024] Step S108: Based on the typical target trajectory prediction sample set, fine-tune the large model to obtain the target model, wherein the typical target trajectory prediction sample set is a sample set generated based on the flight trajectories related to the flight target in historical data.

[0025] Step S110: Based on the target model, generate character-type trajectory data corresponding to the flight target, and convert the character-type trajectory data into target geographic coordinates through Geohash reverse decoding.

[0026] Step S112: Generate a trajectory prediction map of the flight target based on the target's geographical coordinates.

[0027] It should be noted that the original trajectory data includes first trajectory data and second trajectory data. The first trajectory data is the trajectory data of the flight target collected in real time. In practical applications, the position data of the flight target is obtained in real time by connecting to the air traffic control system, satellite communication or other sensors to ensure the timely updating of the data. The second trajectory data is the batch trajectory data of the flight target that is stored in advance. In practical applications, the flight trajectory data within a certain period of time is periodically extracted from the storage system or historical records for the training of analysis and prediction models.

[0028] After determining the original trajectory data, statistical methods are used to identify and remove abnormal coordinate points, thereby realizing outlier detection and cleaning. The specific implementation method is as follows.

[0029] The step of detecting and cleaning outliers in the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data includes: analyzing the original trajectory data using statistical methods to determine abnormal geographic coordinate points, wherein the abnormal geographic coordinate points include a first abnormal geographic coordinate point and a second abnormal geographic coordinate point. The first abnormal geographic coordinate point is the geographic coordinate point corresponding to the abnormal data in the original trajectory data corresponding to the flight target, and the second abnormal geographic coordinate point is the geographic coordinate point where the location area corresponding to the original trajectory data corresponding to the flight target is located in an illegal flight area; the original geographic coordinates are obtained by processing the abnormal geographic coordinate points.

[0030] It should be noted that the abnormal geographic coordinate points here include the first abnormal geographic coordinate point and the second abnormal geographic coordinate point. The first abnormal geographic coordinate point can be a coordinate value that jumps momentarily due to equipment failure or remains unchanged. The second abnormal geographic coordinate point can be a coordinate value that is outside the normal geographic range.

[0031] The illegal flight zone here refers to flight areas outside the legal flight zone. The legal flight zone is a geographical area clearly defined according to air traffic control rules or drone flight specifications, allowing flight targets to enter, excluding oceans, polar regions, and other non-flying areas. It should be noted that in this embodiment, coordinate points falling within the illegal flight zone are directly marked as outliers and removed.

[0032] In this embodiment of the application, the detection and cleaning of outliers can be achieved by filling in or deleting missing data points to ensure the continuity and consistency of the trajectory.

[0033] Specifically, the step of using statistical methods to analyze the original trajectory data and determine abnormal geographic coordinate points includes: calculating the standard deviation of each original geographic coordinate point relative to the sample mean based on the original geographic coordinate points corresponding to the original trajectory data, and calculating the standardized score corresponding to the standard deviation based on the target formula; in response to the standardized score being greater than a preset standardized score threshold, the original geographic coordinates are determined to be abnormal geographic coordinates.

[0034] In practical applications, for the latitude and longitude coordinates of the trajectory data formed by the flight target in one time period, the standard deviation (σ) of each coordinate point relative to the sample mean (μ) is calculated. Then, based on the target formula, its Z-score (i.e., standardized score) is calculated, where the target formula is: (Xi-μ) / σ, where Xi is a single latitude and longitude coordinate value.

[0035] Judgment threshold: Select an appropriate Z-score threshold (such as ±3) as the preset standardized score threshold. If the Z-score value of a point exceeds this threshold, it is considered a potential outlier, because this indicates that the point deviates significantly from most other points.

[0036] In this embodiment of the application, the processing of abnormal geographic coordinates includes: correcting a first abnormal geographic coordinate point; if the correction result of the first abnormal geographic coordinate point does not meet the reasonable interpolation, then removing the first geographic coordinate point; and removing a second abnormal geographic coordinate point.

[0037] The first abnormal geographic coordinate point is corrected based on time continuity checks and calculation of time intervals.

[0038] Specifically, the correction of the first abnormal geographic coordinate point includes: determining the data characteristics corresponding to the data sequence in which the first abnormal geographic coordinate point is located; determining a corresponding data correction method based on the data characteristics; and correcting the first abnormal geographic coordinate point based on the data correction method.

[0039] In practical applications, for consecutive pairs of coordinate points, the time difference Δt between them is calculated. A reasonable time interval threshold is set based on the target aircraft's maximum speed and acceleration information. If the time difference is too large or too small to meet physical conditions (such as supersonic instantaneous transfer or prolonged stationary motion), these coordinate points are considered outliers due to equipment malfunction or data errors. If jump points that do not conform to physical laws do exist, interpolation methods can be attempted to correct them based on the speed and direction information of the preceding and following points; if reasonable interpolation is not possible, these points are also discarded.

[0040] Specifically, missing data is filled using interpolation methods, including: for continuous and clearly trending trajectory segments, linear interpolation is used to calculate the location of missing points based on the position and time information of the preceding and following points; for complex trajectory changes, high-order polynomial interpolation is used to simulate the trajectory's changing trend; for flight behaviors with a certain periodicity and regularity, intelligent filling based on historical behavior is used, employing probability distribution models learned from previous trajectory patterns to predict the location of missing points. Furthermore, for missing points that cannot be effectively filled, deletion is only considered if they severely affect the trajectory's continuity or the accuracy of subsequent analysis; otherwise, the gaps are retained, marked as missing data, and processed accordingly in subsequent modeling. For outliers that are difficult to determine, they are retained and marked, and given special weights or considered as uncertainties during model training.

[0041] The process of outlier detection and cleaning also includes a normalization step, namely coordinate normalization.

[0042] To accommodate the requirements of Geohash encoding, the original latitude and longitude coordinates are divided by a preset base (e.g., longitude base is 100,000, latitude base is 100,000) to convert the latitude and longitude values ​​into integers. This ensures data scale consistency and facilitates subsequent encoding operations. For example, if the actual longitude of a flying target is 116.4074° and the latitude is 39.9042°, after normalization, they will be 1164,074 and 399,042 respectively.

[0043] In this embodiment of the application, the encoding of trajectory data based on Geohash specifically includes two steps: coordinate to integer conversion and Geohash encoding process.

[0044] That is, the step of using the Geohash algorithm to convert the original geographic coordinates into character-type data for large-scale model training and generate a character-type trajectory dataset includes: normalizing the original geographic coordinates to convert them into corresponding latitude and longitude integers; splitting the latitude and longitude integers bit by bit based on a preset precision level to obtain at least one set of latitude and longitude integers, and assigning the set of latitude and longitude integers to corresponding encoding intervals; encoding each encoding interval into characters, and concatenating the character codes corresponding to each encoding interval to obtain character-type data for large-scale model training; and obtaining a character-type trajectory dataset based on the character-type data.

[0045] The step of converting coordinates to integers is the same as the coordinate normalization process described above, which converts real latitude and longitude into integer form.

[0046] When encoding Geohash trajectory data, the first step is bit allocation, which involves splitting the latitude and longitude integers bit by bit according to a pre-set precision level (Geohash length) and allocating them to the corresponding binary bits or Base32-encoded intervals. Then, the encoding algorithm is performed, which involves encoding each interval corresponding to the latitude and longitude with 0 / 1 or Base32 characters respectively and concatenating them into a complete Geohash string.

[0047] Suppose a latitude and longitude pair (1164074, 399042) is encoded by Geohash and the resulting string is "u3ffcp". This encoding not only contains latitude and longitude information, but also implies the division and hierarchical relationship of geographical regions.

[0048] In this embodiment of the application, the implementation of obtaining a character-type trajectory dataset based on character-type data includes two parts: full trajectory encoding and dataset organization.

[0049] Specifically, full trajectory encoding involves Geohash encoding all historical trajectory points of the flight target to form an ordered sequence of strings.

[0050] The dataset is organized such that the trajectory dataset for each flight target is represented as a time series of Geohash strings, where each element in the series represents the geographical location of a point in time. Furthermore, in this embodiment, the entire trajectory dataset can be divided into a training set, a validation set, and a test set to facilitate the training, performance evaluation, and generalization ability testing of the machine learning model.

[0051] This application uses Geohash encoding to transform the trajectory of a flight target into a character-based dataset that is easy for computers to process and for model training. This process effectively integrates spatiotemporal information and provides an ideal input data format for subsequent trajectory prediction based on large models.

[0052] In the large model training process of this application embodiment, advanced machine learning techniques, especially pre-trained models of the Transformer architecture (such as BERT, GPT series, etc.), are used to train the character trajectory dataset.

[0053] The process of constructing a large model training process based on the character-type trajectory dataset, which utilizes a pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target, includes: inputting the character-type trajectory dataset into the target language model, iteratively learning the character-type trajectory dataset to output the spatiotemporal patterns corresponding to the flight trajectory of the flight target; and optimizing the key points in the spatiotemporal patterns through an attention mechanism to obtain the target spatiotemporal patterns.

[0054] First, the model is initialized by selecting or building a large-scale language model suitable for sequence prediction, with sufficient parameters and a complex self-attention mechanism to capture long-range dependencies between trajectory points. Then, the model learns spatiotemporal patterns: using a character-based trajectory dataset as input, the model learns iteratively to understand the behavioral patterns and potential rules of the flying target in the temporal and spatial dimensions. Finally, an attention mechanism is added: during training, the attention mechanism allows the model to focus on key points in the trajectory sequence, enhancing the model's ability to recognize the flying target's motion trends, speed changes, and turning angles.

[0055] This application provides a large model training method based on GPT-2, and the specific implementation is as follows.

[0056] First, initialize the model.

[0057] Model Framework Selection: Considering that flight trajectory prediction needs to capture temporal and spatial sequence relationships, we chose GPT-2 as the base model because it has a strong ability to generate autoregressive sequences.

[0058] Model configuration: Assume we choose a relatively large GPT-2 variant, such as a model with 12 Transformer blocks, each containing 16 attention heads, and a hidden layer size of 768. These parameters can be adjusted appropriately based on the complexity of the actual trajectory data and available computing resources.

[0059] Loading pre-trained weights: First, download and load the pre-trained GPT-2 model parameters. These parameters have been pre-trained on large unlabeled text data, giving the model a certain general language understanding ability.

[0060] Then we study the laws of time and space.

[0061] Data preparation: The raw flight trajectory data (latitude and longitude coordinates) is converted into Geohash encoding and arranged into a sequence according to time order. For example, the historical trajectory sequence of a certain aircraft is: "u3ffcp", "u3ffcr", "u3ffcs"..., where each step represents the geographic coordinates at a timestamp.

[0062] Sequence padding: Since GPT-2 requires a fixed sequence length, for shorter trajectory sequences, we pad the end with "[PAD]" markers to the required length; while for excessively long sequences, we truncate the nearest fixed-length sequence and add "[BOS]" and "[EOS]" markers at the beginning and end, respectively, to indicate the start and end of the sequence.

[0063] Training objective setting: The training objective is to predict several consecutive points following a given trajectory segment. In the training set, there is a one-to-one correspondence between the input sequence and the target sequence. The model is trained by minimizing the cross-entropy loss between the two, that is, the difference between the model's output sequence and the Geohash encoding of the actual flight trajectory in the next time segment.

[0064] Finally, there is the learning of attention mechanisms.

[0065] Self-attention mechanism: In the GPT-2 model, the self-attention mechanism allows the model to learn the correlation between the Geohash code at each location and all other codes before and after it. In flight trajectory prediction scenarios, this mechanism helps the model understand the mutual influence of trajectory points between different time steps and the potential spatiotemporal trends.

[0066] Attention score calculation: In each attention layer, the input sequence is mapped to three sets of vectors: Query, Key, and Value. The dot product of Query and Key is calculated and normalized to obtain the attention weights. These weights are then applied to the Value vector and weighted summed to generate the context vector for each position.

[0067] As training progresses, the model will automatically identify and focus on turning points or velocity changes crucial for trajectory prediction, such as Geohash codes indicating sudden changes in direction or rapid increases in speed. In practical applications, given historical trajectory data of flight target A, the model will gradually adapt and learn effective spatiotemporal patterns. Through continuous iterative optimization, it will ultimately improve the accuracy of predicting future trajectory points. For example, after training, when the model receives the first n Geohash code sequences, it can effectively predict the Geohash codes for the (n+1)th and even multiple subsequent flight positions based on its learned knowledge.

[0068] After completing the training of the model, this application embodiment also constructs a typical target trajectory prediction sample set for fine-tuning the large model, thereby improving the accuracy of the model in the flight target trajectory prediction task.

[0069] Specifically, the step of fine-tuning the large model based on a typical target trajectory prediction sample set to obtain the target model includes: acquiring flight trajectory data related to the flight target from historical data based on the flight mode and flight context corresponding to the flight target, and constructing a typical target trajectory prediction sample set based on the flight trajectory data; taking each sample data in the typical target trajectory prediction sample set as input, and outputting the predicted trajectory corresponding to the sample data as the label to generate a fine-tuning dataset; and fine-tuning the large model based on the fine-tuning dataset to obtain the target model.

[0070] In practical applications, firstly, typical flight target trajectories covering various flight modes and scenarios are selected from historical data to construct a prediction sample set. Each sample is input by a past trajectory sequence and its corresponding future real trajectory is used as a label. Then, the large model that has been initially trained is fine-tuned using the above prediction sample set to optimize the model's ability to predict the trajectory of specific flight targets.

[0071] In practical applications, the finely tuned model is used to predict the future trajectory of flight targets.

[0072] In this embodiment of the application, the step of generating character-type trajectory data corresponding to the flight target based on the target model and converting the character-type trajectory data into target geographic coordinates through Geohash reverse decoding includes: inputting historical trajectory data of the flight target to be tested into the target model, receiving predicted character-type trajectory data corresponding to the flight target output by the target model; converting the predicted character-type trajectory data into at least one corresponding geographic coordinate based on the Geohash algorithm, and generating a trajectory prediction map of the flight target by concatenating at least one of the geographic coordinates.

[0073] In practical applications, firstly, some historical trajectory data of the target to be predicted (in Geohash encoding form) is input into the model, and the model will output a series of string sequences representing future trajectory points.

[0074] Then, the predicted string sequence is parsed one by one and restored to latitude and longitude coordinates. The Geohash inverse decoding algorithm is used to restore the accurate geographic coordinates. Finally, these predicted latitude and longitude coordinates are concatenated to generate a continuous trajectory prediction map of the future flight target, thus providing strong support for route planning, traffic control or safety early warning.

[0075] This application provides a novel method for predicting flight target trajectories by combining the Geohash algorithm and a large-scale language model. It has the advantages of strong data processing capabilities, good real-time performance, high accuracy, and good generalization ability.

[0076] The flight target trajectory prediction method based on the Geohash algorithm provided in this application achieves efficient prediction of flight target trajectories through the following key steps: performing fine-grained outlier detection and cleaning on the original trajectory data to ensure data quality and integrity; converting latitude and longitude coordinates into character-based data suitable for large-scale model training using the Geohash algorithm; constructing a large-scale model training process based on the character-based trajectory dataset, and using a pre-trained model with a Transformer architecture (such as GPT-2) to mine the spatiotemporal patterns of the trajectory; fine-tuning the model to improve accuracy in specific flight target trajectory prediction tasks; applying the trained model to predict future trajectories, and then using Geohash for reverse decoding to restore the actual geographical coordinates.

[0077] On the one hand, it improves the accuracy of flight target trajectory prediction, making flight path prediction more accurate and beneficial for route planning and traffic management. On the other hand, it effectively handles large-scale, complex, and variable flight trajectory data, solving the bottleneck problem of traditional methods in data processing. Furthermore, by combining Geohash encoding technology, it transforms latitude and longitude data that was originally difficult for large models to accept into a suitable form, greatly enhancing the application potential of large models in trajectory prediction. Moreover, through pre-trained models and fine-tuning techniques, it uncovers deep-seated patterns of flight targets in time and space, achieving a more comprehensive understanding and prediction of flight behavior. Finally, it realizes the automation and intelligence of flight target trajectory prediction, providing strong technical support for related fields.

[0078] Corresponding to the above method embodiments, this specification also provides an embodiment of a flight target trajectory prediction device based on the Geohash algorithm, see [link to documentation]. Figure 2 , Figure 2 This is a block diagram of a flight target trajectory prediction device based on the Geohash algorithm provided in an embodiment of this application, such as... Figure 2 As shown, it specifically includes the following modules.

[0079] The cleaning module 202 is configured to perform outlier detection and cleaning on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data. The first conversion module 204 is configured to use the Geohash algorithm to convert the original geographic coordinates into character data for large model training, generating a character trajectory dataset; The construction module 206 is configured to construct a large model training process based on the character-type trajectory dataset, and to use the pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target. The fine-tuning module 208 is configured to fine-tune the large model based on a typical target trajectory prediction sample set to obtain the target model. The typical target trajectory prediction sample set is a sample set generated based on flight trajectories related to the flight target in historical data. The second conversion module 210 is configured to generate character-type trajectory data corresponding to the flight target based on the target model, and convert the character-type trajectory data into target geographic coordinates through Geohash reverse decoding; The generation module 212 is configured to generate a trajectory prediction map of the flight target based on the target's geographic coordinates.

[0080] Furthermore, the cleaning module 202 is further configured as follows: Using statistical methods, the original trajectory data is analyzed to determine abnormal geographic coordinate points. The abnormal geographic coordinate points include a first abnormal geographic coordinate point and a second abnormal geographic coordinate point. The first abnormal geographic coordinate point is the geographic coordinate point corresponding to the abnormal data in the original trajectory data of the flight target. The second abnormal geographic coordinate point is the geographic coordinate point where the location area corresponding to the original trajectory data of the flight target is located in an illegal flight area. The original geographic coordinates are obtained by processing the abnormal geographic coordinate points.

[0081] Furthermore, the cleaning module 202 is further configured as follows: Based on the original geographic coordinates corresponding to the original trajectory data, calculate the standard deviation of each original geographic coordinate point relative to the sample mean, and calculate the standardized score corresponding to the standard deviation based on the target formula. If the standardized score is greater than a preset standardized score threshold, the original geographic coordinates are determined to be abnormal geographic coordinates.

[0082] Furthermore, the cleaning module 202 is further configured as follows: The first abnormal geographic coordinate point is corrected. If the correction result of the first abnormal geographic coordinate point does not meet the reasonable interpolation, the first geographic coordinate point is removed. Remove the second abnormal geographic coordinate point.

[0083] Furthermore, the cleaning module 202 is further configured as follows: Determine the data characteristics corresponding to the data sequence where the first abnormal geographic coordinate point is located; Based on the data characteristics, a corresponding data correction method is determined, and based on the data correction method, the first abnormal geographic coordinate point is corrected.

[0084] Furthermore, the cleaning module 202 is further configured as follows: By normalizing the original geographic coordinates, the original geographic coordinates are converted into corresponding latitude and longitude integers; Based on a preset precision level, the latitude and longitude integers are split bit by bit to obtain at least one set of latitude and longitude integers, and the set of latitude and longitude integers is assigned to the corresponding encoding interval; Each of the aforementioned encoding intervals is character-encoded, and by concatenating the character codes corresponding to each of the aforementioned encoding intervals, character-type data for training large models is obtained; Based on the character-type data, a character-type trajectory dataset is obtained.

[0085] Furthermore, the building module 206 is further configured as follows: Input the character-type trajectory dataset into the target language model, and output the spatiotemporal pattern corresponding to the flight trajectory of the flight target by iterative learning of the character-type trajectory dataset; By optimizing the key points in the spatiotemporal pattern through an attention mechanism, the target spatiotemporal pattern is obtained.

[0086] Furthermore, the fine-tuning module 208 is further configured as follows: Based on the flight mode and flight context corresponding to the flight target, flight trajectory data related to the flight target is obtained from historical data, and a typical target trajectory prediction sample set is constructed based on the flight trajectory data; Each sample data in the typical target trajectory prediction sample set is input with the corresponding historical trajectory sequence and output with the predicted trajectory corresponding to the sample data as the label to generate a fine-tuning dataset. Based on the fine-tuning dataset, the large model is fine-tuned to obtain the target model.

[0087] Furthermore, the second conversion module 210 is further configured to: Input the historical trajectory data of the flight target to be tested into the target model, and receive the predicted character-type trajectory data of the flight target output by the target model; Based on the Geohash algorithm, the predicted character trajectory data is converted into at least one corresponding geographic coordinate. By concatenating at least one of the geographic coordinates, a trajectory prediction map of the flight target is generated.

[0088] This application provides a novel flight target trajectory prediction device by combining the Geohash algorithm and a large-scale language model. It has the advantages of strong data processing capabilities, good real-time performance, high accuracy, and good generalization ability.

[0089] Figure 3 This is a structural block diagram of a computing device provided in an embodiment of this application. The components of the computing device 300 include, but are not limited to, a memory 310 and a processor 320. The processor 320 is connected to the memory 310 via a bus 330, and a database 350 is used to store data. The computing device 300 also includes an access device 340.

[0090] In one embodiment of this application, the aforementioned components of the computing device 300 and Figure 3 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 3 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this application. Those skilled in the art can add or replace other components as needed.

[0091] The computing device 300 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 300 can also be a mobile or stationary server.

[0092] The processor 320 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above method.

[0093] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the above method belong to the same concept, and all details not described in detail in the technical solution of the computing device can be referred to the description of the technical solution of the above method.

[0094] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described method.

[0095] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the method described above belong to the same concept, and all details not described in detail in the technical solution of the storage medium can be referred to the description of the technical solution of the method described above.

[0096] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0097] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0098] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0099] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for predicting the trajectory of a flight target based on the Geohash algorithm, characterized in that, include: Anomaly detection and cleaning are performed on the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data. The original trajectory data includes first trajectory data corresponding to the flight target collected in real time, and / or second trajectory data corresponding to the flight target stored in advance. The original geographic coordinates are converted into character-type data for large model training using the Geohash algorithm, generating a character-type trajectory dataset; A large model training process based on the aforementioned character-type trajectory dataset is constructed, and the spatiotemporal patterns of the flight trajectory corresponding to the flight target are mined using the pre-trained model of the target neural network architecture. Based on the typical target trajectory prediction sample set, the large model is fine-tuned to obtain the target model. The typical target trajectory prediction sample set is a sample set generated based on flight trajectories related to the flight target in historical data. Based on the target model, character-type trajectory data corresponding to the flight target is generated, and the character-type trajectory data is converted into target geographic coordinates through Geohash reverse decoding. Based on the target's geographic coordinates, a trajectory prediction map of the flight target is generated; The step of using the Geohash algorithm to convert the original geographic coordinates into character-type data for large model training, generating a character-type trajectory dataset, also includes: By normalizing the original geographic coordinates, the original geographic coordinates are converted into corresponding latitude and longitude integers; Based on a preset precision level, the latitude and longitude integers are split bit by bit to obtain at least one set of latitude and longitude integers, and the set of latitude and longitude integers is assigned to the corresponding encoding interval; Each of the aforementioned encoding intervals is character-encoded, and by concatenating the character codes corresponding to each of the aforementioned encoding intervals, character-type data for training large models is obtained; Based on the character-type data, a character-type trajectory dataset is obtained; The process of fine-tuning the large model based on a typical target trajectory prediction sample set to obtain the target model includes: Based on the flight mode and flight context corresponding to the flight target, flight trajectory data related to the flight target is obtained from historical data, and a typical target trajectory prediction sample set is constructed based on the flight trajectory data; Each sample data in the typical target trajectory prediction sample set is input with the corresponding historical trajectory sequence and output with the predicted trajectory corresponding to the sample data as the label to generate a fine-tuning dataset. Based on the fine-tuning dataset, the large model is fine-tuned to obtain the target model.

2. The method according to claim 1, characterized in that, The process of outlier detection and cleaning of the original trajectory data to obtain the original geographic coordinates corresponding to the original trajectory data includes: Using statistical methods, the original trajectory data is analyzed to determine abnormal geographic coordinate points. The abnormal geographic coordinate points include a first abnormal geographic coordinate point and a second abnormal geographic coordinate point. The first abnormal geographic coordinate point is the geographic coordinate point corresponding to the abnormal data in the original trajectory data of the flight target. The second abnormal geographic coordinate point is the geographic coordinate point where the location area corresponding to the original trajectory data of the flight target is located in an illegal flight area. The original geographic coordinates are obtained by processing the abnormal geographic coordinate points.

3. The method according to claim 2, characterized in that, The step of using statistical methods to analyze the original trajectory data and determine abnormal geographic coordinate points includes: Based on the original geographic coordinates corresponding to the original trajectory data, calculate the standard deviation of each original geographic coordinate point relative to the sample mean, and calculate the standardized score corresponding to the standard deviation based on the target formula. If the standardized score is greater than a preset standardized score threshold, the original geographic coordinate point is determined to be an abnormal geographic coordinate point.

4. The method according to claim 2, characterized in that, The process of processing the abnormal geographic coordinate points to obtain the original geographic coordinates includes: The first abnormal geographic coordinate point is corrected. If the correction result of the first abnormal geographic coordinate point does not meet the reasonable interpolation, the first abnormal geographic coordinate point is removed. Remove the second abnormal geographic coordinate point.

5. The method according to claim 4, characterized in that, The correction of the first abnormal geographic coordinate point includes: Determine the data characteristics corresponding to the data sequence where the first abnormal geographic coordinate point is located; Based on the data characteristics, a corresponding data correction method is determined, and based on the data correction method, the first abnormal geographic coordinate point is corrected.

6. The method according to claim 1, characterized in that, The process of constructing a large model training procedure based on the character-type trajectory dataset, which utilizes a pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target, includes: Input the character-type trajectory dataset into the target language model, and output the spatiotemporal pattern corresponding to the flight trajectory of the flight target by iterative learning of the character-type trajectory dataset; By optimizing the key points in the spatiotemporal pattern through an attention mechanism, the target spatiotemporal pattern is obtained.

7. The method according to claim 1, characterized in that, The process of generating character-based trajectory data corresponding to the flight target based on the target model, and converting the character-based trajectory data into target geographic coordinates through Geohash reverse decoding, includes: Input the historical trajectory data of the flight target to be tested into the target model, and receive the predicted character-type trajectory data of the flight target output by the target model; Based on the Geohash algorithm, the predicted character trajectory data is converted into at least one corresponding geographic coordinate. By concatenating at least one of the geographic coordinates, a trajectory prediction map of the flight target is generated.

8. A flight target trajectory prediction device based on the Geohash algorithm, characterized in that, include: The cleaning module is configured to perform outlier detection and cleaning on the raw trajectory data to obtain the original geographic coordinates corresponding to the raw trajectory data. The first conversion module is configured to use the Geohash algorithm to convert the original geographic coordinates into character data for large model training, generating a character trajectory dataset. The step of using the Geohash algorithm to convert the original geographic coordinates into character-type data for large model training, generating a character-type trajectory dataset, also includes: By normalizing the original geographic coordinates, the original geographic coordinates are converted into corresponding latitude and longitude integers; Based on a preset precision level, the latitude and longitude integers are split bit by bit to obtain at least one set of latitude and longitude integers, and the set of latitude and longitude integers is assigned to the corresponding encoding interval; Each of the aforementioned encoding intervals is character-encoded, and by concatenating the character codes corresponding to each of the aforementioned encoding intervals, character-type data for training large models is obtained; Based on the character-type data, a character-type trajectory dataset is obtained; The construction module is configured to build a large model training process based on the character-type trajectory dataset, and to use the pre-trained model of the target neural network architecture to mine the spatiotemporal patterns of the flight trajectory corresponding to the flight target. The fine-tuning module is configured to fine-tune the large model based on a typical target trajectory prediction sample set to obtain a target model. The typical target trajectory prediction sample set is a sample set generated from historical data related to flight trajectories of the flight target. The fine-tuning of the large model based on the typical target trajectory prediction sample set to obtain the target model includes: Based on the flight mode and flight context corresponding to the flight target, flight trajectory data related to the flight target is obtained from historical data, and a typical target trajectory prediction sample set is constructed based on the flight trajectory data; Each sample data in the typical target trajectory prediction sample set is input with the corresponding historical trajectory sequence and output with the predicted trajectory corresponding to the sample data as the label to generate a fine-tuning dataset. Based on the fine-tuning dataset, the large model is fine-tuned to obtain the target model; The second conversion module is configured to generate character-type trajectory data corresponding to the flight target based on the target model, and convert the character-type trajectory data into target geographic coordinates through Geohash reverse decoding; The generation module is configured to generate a trajectory prediction map of the flight target based on the target's geographic coordinates.

Citation Information

Patent Citations

  • Flight trajectory prediction method and device based on Koopman neural network

    CN116841307A

  • Target trajectory prediction method and system with domain generalization ability

    CN116912661A