An AI-based unmanned aerial vehicle flight data management method and system

Abstract

This invention discloses an AI-based method and system for managing drone flight data, relating to the field of data management technology. The method includes: Step 1: Establishing a flight data source repository and a mission echo repository; writing flight data into the flight data source repository and constructing mission echo records in the mission echo repository; Step 2: Establishing a data echo channel and generating mission echo segments; Step 3: Generating echo identity tags based on an improved xLSTM model; Step 4: Writing the echo identity tags into the corresponding flight data and mission echo records; Step 5: Performing consistency comparison; Step 6: When the echo identity tags are consistent, marking them as reliable flight data; when the echo identity tags are inconsistent, marking them as abnormal echo data and transferring them to an isolated echo zone; Step 7: Isolating and managing abnormal echo data. This invention utilizes a dual-repository echo channel and an improved xLSTM model to achieve reliable traceability and abnormal isolation management of drone flight data.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and in particular to an AI-based method and system for managing drone flight data. Background Technology

[0002] With the increasing application of drones in scenarios such as inspection, surveying, logistics, and emergency operations, the demand for complete preservation, reliable traceability, and anomaly isolation management of flight data is constantly increasing. Current drone flight data management methods mainly rely on centralized storage of raw logs, retrieval by task number, or simple operation records for data traceability. However, these methods commonly suffer from the following problems in practical applications: Each UAV flight generates a large volume of flight process data over a long period. Current methods typically store all data as a continuous log, lacking mission-side records corresponding to the takeoff, cruise, operation, return, and landing phases. This results in low location efficiency when searching for critical flight moments, requiring traversal through a large amount of raw data. Flight data is prone to misalignment, missing data, or abnormal rewriting when operational events occur. Existing methods struggle to synchronously read back the echo identity tags between flight data and mission echo records and perform consistency comparisons when operational events occur. Furthermore, traditional data management methods lack isolation management mechanisms for abnormal echo data, which may continue to be used, migrated, or archived, leading to the spread of anomalies, difficulties in tracing responsibility, and a decline in the reliability of flight data.

[0003] Therefore, how to provide an AI-based method and system for managing drone flight data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose an AI-based method and system for managing UAV flight data. This invention utilizes a flight data source repository, a mission echo repository, and a data echo channel, and combines an improved xLSTM model to generate echo identity tags. This enables mission phase mapping of UAV flight data, echo identity tag generation, and consistency comparison, timely verification of the correspondence between flight data and mission echo records, distinguishing between reliable flight data and abnormal echo data, and isolating and managing abnormal echo data to prevent abnormal data from continuing to participate in subsequent calls or transfers. This improves the reliable traceability, anomaly identification, and data security management level of UAV flight data.

[0005] According to an embodiment of the present invention, an AI-based method for managing drone flight data includes the following steps: Step 1: Establish the flight data source repository and the mission echo repository. Write the flight data generated by each flight of the UAV into the flight data source repository, and build the mission echo record in the mission echo repository. Step 2: Establish a data echo channel between the flight data source and the mission echo source, perform echo localization on the flight data based on the mission echo record, and generate mission echo segments; Step 3: Input the mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. Step 4: Write the echo identity tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo repository through the data echo channel; Step 5: When an operation event occurs in the flight data in the original flight data repository, the corresponding flight data and the echo identity tag in the corresponding mission echo record are read back through the data echo channel, and a consistency comparison is performed; Step 6: When the echo identification tags match, mark the corresponding flight data as reliable flight data; when the echo identification tags do not match, mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone. Step 7: Based on the isolated echo records in the isolated echo zone, manage the isolated echo data.

[0006] Optionally, step one specifically includes: Collect flight data generated by the UAV for each flight. The flight data includes flight mission number, UAV number, flight time, positioning trajectory data, attitude status data, link status data, payload status data, and control command data. The flight data is completely written into the original flight data repository according to the flight mission number, UAV number, and flight time. Obtain the mission phase configuration corresponding to the flight mission, which includes the takeoff phase, cruise phase, operation phase, return phase, and landing phase; Flight data is categorized into the corresponding takeoff phase, cruise phase, operation phase, return phase, or landing phase according to flight time, thus obtaining the flight phase sequence. Based on the flight mission number, UAV number, flight phase sequence, and the start and end times of each flight phase, a mission echo record is constructed. Configure a task echo index for the task echo record and write the task echo record to the task echo repository.

[0007] Optionally, step two specifically involves: Obtain the mission echo index, flight mission number, UAV number, flight phase sequence, and start and end times of each flight phase corresponding to the mission echo record; Based on the flight mission number and UAV number, the corresponding flight data is searched in the original flight data repository, and the flight data within the corresponding flight stage is extracted according to the start and end times of each flight stage to obtain the stage flight data segment. The mission echo index, flight phase sequence, and phase flight data segments are mapped to establish a data echo channel between the flight data source repository and the mission echo repository. In the data echo channel, the flight data segments of each stage are read frame by frame according to the flight time and compared. When any one of the following changes in adjacent flight times: the trajectory direction corresponding to the positioning trajectory data, the attitude angle corresponding to the attitude state data, the link connection state corresponding to the link state data, the load working state corresponding to the load state data, or the control command category corresponding to the control command data, the next flight time is determined as the echo positioning time, and the flight data corresponding to the echo positioning time is determined as the flight echo node. The type of echo node corresponding to the flight echo node is determined based on the data type that has changed, and the flight phase identifier and the location of the original data in the original data repository are obtained. The flight phase identifier, echo location time, echo node type, and original data location are combined into an echo node record, and each echo node record is arranged according to the echo location time to generate a mission echo segment.

[0008] Optionally, the improved xLSTM model specifically includes a flight echo embedding layer, a scalar memory structure, a flight echo receiving layer, an exponential gating structure, a matrix memory structure, and an echo identity output layer; The flight echo embedding layer reads the records of each echo node in the mission echo segment, encodes and splices the flight phase identifier, echo location time, echo node type and original data location, and obtains the flight echo token through linear mapping and GELU activation processing. The scalar memory structure reads the flight echo token according to the echo localization time, and performs temporal recursion on adjacent flight echo tokens through the scalar memory state to obtain single-line echo memory features. The flight echo receiving layer is set between the scalar memory structure and the matrix memory structure, and performs stage receiving of single-line echo memory features according to the flight stage identifier to obtain the received echo features. The exponential gating structure optimizes the flight echo continuity of the received echo features to obtain stable echo features. The matrix memory structure receives stable echo features and writes the stable echo features into the matrix memory state. The matrix memory state is used to remember and express the correspondence between flight phase identifiers, echo location times, and echo node types, thus obtaining matrix echo memory features. The echo identity output layer sequentially performs linear mapping, LayerNorm normalization, GELU activation, and Softmax mapping on the matrix echo memory features to obtain the echo identity probability distribution corresponding to each echo node record, and generates the corresponding echo identity tag based on the echo identity probability distribution corresponding to each echo node record.

[0009] Optionally, the step of performing stage-based succession of single-line echo memory features according to flight stage identifiers to obtain succession echo features specifically involves: Obtain the flight phase identifier, echo location time, and echo node type corresponding to the single-line echo memory feature; According to the flight phase identifier, the single-line echo memory features are written into the phase receiving slots corresponding to the takeoff phase, cruise phase, operation phase, return phase and landing phase to obtain the phase echo feature group. Within each stage receiving slot, single-line echo memory features in the stage echo feature group are arranged according to the echo positioning time to generate an echo receiving chain within the stage. Read the single-line echo memory features at the end of the previous stage and the single-line echo memory features at the beginning of the next stage in the order of flight stages, and compare the echo node type corresponding to the single-line echo memory features at the end of the previous stage with the echo node type corresponding to the single-line echo memory features at the beginning of the next stage. When the echo nodes are of the same type, the single-line echo memory feature at the end of the previous stage, the same type of continuation identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the same type of stage continuation feature. When the echo node types are different, the single-line echo memory feature at the end of the previous stage, the heterogeneous transition identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the heterogeneous stage transition feature. Arrange the echo succession chain within a stage, the continuity features of similar stages, and the transition features of dissimilar stages according to the echo location time to obtain the succession echo features.

[0010] Optionally, the optimization of the flight echo continuity of the received echo features to obtain stable echo features specifically involves: The intra-stage echo succession chain, similar stage continuity feature, and dissimilar stage transition feature in the echo succession feature are respectively input into the exponential gating structure. The exponential gating structure performs differential mapping on the adjacent single-line echo memory features in the intra-stage echo relay chain to obtain the intra-stage differential response features. By splicing and mapping the same-type succession identifier, the single-line echo memory feature at the end of the previous stage and the single-line echo memory feature at the beginning of the next stage in the same-type stage succession feature, the same-type succession response feature is obtained. The heterogeneous transition identifier, the single-line echo memory feature at the end of the previous stage, and the single-line echo memory feature at the beginning of the next stage in the heterogeneous stage transition features are spliced ​​and mapped to obtain the heterogeneous transition response features. The intra-stage differential response characteristics, similar succession response characteristics, and dissimilar transition response characteristics are input into the exponential function to obtain the intra-stage gating value, similar succession gating value, and dissimilar transition gating value. The in-stage gating value is multiplied element-wise with the in-stage echo relay chain to obtain the in-stage gating relay chain. The similar type of continuation gate value is multiplied element-wise with the similar type of stage continuation feature, and the residuals are superimposed to obtain the similar type of gated continuation feature. The heterogeneous transition gating value is multiplied element-wise with the heterogeneous stage transition feature, and then concatenated and mapped with the heterogeneous stage transition feature to obtain the heterogeneous gating transition feature. The gated connection chain, similar gated connection features, and dissimilar gated transition features within the stage are spliced ​​together according to the echo localization time, and then subjected to residual connection and LayerNorm normalization to obtain stable echo features.

[0011] Optionally, step five specifically includes: Events generated when flight data in the original flight data repository is viewed, exported, copied, modified, deleted, migrated, compressed and archived, or restored and retrieved are defined as operation events; When an operation event occurs in the flight data in the original flight data repository, the location of the original data in the repository corresponding to the operation event is obtained; Based on the location of the original warehouse data, find the corresponding echo node record in the data echo channel, and determine the corresponding task echo record based on the echo node record; The echo identity markers of the corresponding flight data in the original flight data repository are read back through the data echo channel, and the echo identity markers of the corresponding mission echo records in the mission echo repository are read back. The echo identification tags of the corresponding flight data are compared with the echo identification tags of the corresponding mission echo records for consistency. When two echo identity tags are the same, the echo identity tags are determined to be consistent; When two echo identity tags are different, it is determined that the echo identity tags are inconsistent.

[0012] Optionally, step six specifically includes: When the echo identity tags match, the corresponding flight data is marked as trustworthy flight data, and the original data position of the corresponding flight data in the original data repository is maintained. When the echo identification tags are inconsistent, the corresponding flight data will be marked as abnormal echo data, and an isolated echo zone will be established in the original flight data repository. The isolated echo zone is a management area in the original flight data repository that is independent of the trusted flight data, and is used to store abnormal echo data and isolated echo records. Transfer abnormal echo data to the isolated echo zone, and retain the original warehouse data position corresponding to the abnormal echo data before it was transferred to the isolated echo zone; Based on the operation event type, the location of the original warehouse data, the echo identity tag corresponding to the flight data, the echo identity tag corresponding to the mission echo record, and the isolation write time, an isolation echo record is generated. The isolated echo records are bound to the abnormal echo data, and the echo relationship between the isolated echo records and the corresponding echo node records and the corresponding task echo records is maintained through the data echo channel.

[0013] Optionally, step seven specifically includes: Obtain abnormal echo data and isolation echo records from the isolated echo zone; The location of the original warehouse data, the type of operation event, and the time of isolation write are determined based on the isolated echo record; Mark the management status of abnormal echo data as isolated management status; Write the original data location, operation event type, isolation write time, and isolation management status back to the original flight data repository; Based on the isolation management status, abnormal echo data is restricted from being viewed, exported, copied, modified, deleted, migrated, compressed, archived, or restored.

[0014] An AI-based drone flight data management system according to an embodiment of the present invention includes the following modules: The dual-warehouse construction module is used to establish a flight data origin warehouse and a mission echo warehouse. It writes the flight data generated by each flight of the UAV into the flight data origin warehouse and builds mission echo records in the mission echo warehouse. The data echo channel construction module is used to establish a data echo channel between the flight data source and the mission echo source, perform echo localization of flight data based on mission echo records, and generate mission echo segments. The echo identity tag generation module is used to input the mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. The echo identification tag writing module is used to write the echo identification tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo store through the data echo channel; The operation event comparison module is used to read back the corresponding flight data and the echo identity tag in the corresponding mission echo record through the data echo channel when an operation event occurs in the flight data in the original flight data repository, and perform a consistency comparison. The anomaly isolation processing module is used to mark the corresponding flight data as reliable flight data when the echo identity tags are consistent, and to mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone when the echo identity tags are inconsistent. The isolation management write-back module is used to isolate and manage abnormal echo data according to the isolation echo records in the isolation echo zone, and write the management status of the abnormal echo data back to the flight data source.

[0015] The beneficial effects of this invention are: This invention addresses the reliable management needs of UAV flight data in long-term storage, mission traceability, and anomaly control. It establishes a data management system that integrates a flight data origin repository, a mission echo repository, and a data echo channel. This system separates complete flight data from mission phase references and enables accurate rereading through mission echo records. Consequently, continuous flight logs are no longer simply stored in chronological order but form a stable correspondence with the takeoff, cruise, operation, return, and landing phases. A clear link is established between key flight moments, echo node records, and the location of the origin repository data, reducing redundant retrieval from massive amounts of raw data and improving flight data location and traceability efficiency.

[0016] This invention further inputs mission echo segments into an improved xLSTM model, adds a flight echo relay layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure. This allows for continuous relay of echo information between different flight phases, reducing echo breakage, misalignment, and drift caused by phase switching, and improving the stability and accuracy of echo identification tagging. When an operational event occurs in the flight data, the system can synchronously read back the echo identification tags on both sides through the data echo channel and complete a consistency comparison, promptly distinguishing between reliable flight data and abnormal echo data. After abnormal echo data is transferred to the isolation echo zone, it continues to be isolated and managed according to the isolation echo record, preventing abnormal data from participating in calls, migrations, or archiving again, thereby improving the reliability, traceability, anomaly isolation capability, and overall management security of UAV flight data. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of an AI-based drone flight data management method proposed in this invention; Figure 2 This is a schematic diagram of the structure of an AI-based drone flight data management system proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figure 1 An AI-based method for managing drone flight data includes the following steps: Step 1: Establish the flight data source repository and the mission echo repository. Write the flight data generated by each flight of the UAV into the flight data source repository, and build the mission echo record in the mission echo repository. Step 2: Establish a data echo channel between the flight data source and the mission echo source, perform echo localization on the flight data based on the mission echo record, and generate mission echo segments; Step 3: Input the mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. Step 4: Write the echo identity tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo repository through the data echo channel; Step 5: When an operation event occurs in the flight data in the original flight data repository, the corresponding flight data and the echo identity tag in the corresponding mission echo record are read back through the data echo channel, and a consistency comparison is performed; Step 6: When the echo identification tags match, mark the corresponding flight data as reliable flight data; when the echo identification tags do not match, mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone. Step 7: Based on the isolated echo records in the isolated echo zone, manage the isolated echo data.

[0020] In this embodiment, step one specifically includes: Collect flight data generated by each UAV flight. The flight data includes flight mission number, UAV number, flight time, positioning trajectory data, attitude status data, link status data, payload status data, and control command data. Write the flight data completely into the original flight data repository according to the flight mission number, UAV number, and flight time. Obtain the mission phase configuration corresponding to the flight mission. The mission phase configuration includes takeoff phase, cruise phase, operation phase, return phase, and landing phase. Flight data is categorized into the corresponding takeoff phase, cruise phase, operation phase, return phase, or landing phase according to flight time, thus obtaining the flight phase sequence. Based on the flight mission number, UAV number, flight phase sequence, and the start and end times of each flight phase, a mission echo record is constructed. Configure a task echo index for the task echo record and write the task echo record to the task echo repository; This invention separates the complete preservation of UAV flight data from the phased mapping of the mission process by constructing a flight data origin repository and a mission echo repository. The flight data origin repository is used to completely preserve positioning trajectory data, attitude status data, link status data, payload status data, and control command data according to the flight mission number, UAV number, and flight time, ensuring that the original flight process is not compressed or rewritten, which facilitates review, tracing, and management. The mission echo repository does not repeatedly preserve complete flight data, but generates mission echo records according to the mission phase configuration. It forms the flight phase sequence according to the flight time of the takeoff phase, cruise phase, operation phase, return phase, and landing phase, and records the start and end times of each flight phase, so that the flight mission is transformed from a continuous data stream into a mission-side record that can be located, compared, and read back. Through the aforementioned dual-warehouse structure, the flight data source warehouse provides the data ontology, the mission echo warehouse provides mission phase references, and the data echo channel can perform echo localization of flight data based on mission echo records. This avoids disordered searching in a large amount of raw flight data, improves the accuracy and efficiency of flight echo node extraction, echo identity tag writing, and operational event consistency comparison, thereby enhancing the traceability, reliability, and anomaly isolation capabilities of UAV flight data management.

[0021] In this embodiment, step two specifically involves: Obtain the mission echo index, flight mission number, UAV number, flight phase sequence, and start and end times of each flight phase corresponding to the mission echo record; Based on the flight mission number and UAV number, the corresponding flight data is searched in the original flight data repository, and the flight data within the corresponding flight stage is extracted according to the start and end times of each flight stage to obtain the stage flight data segment. The mission echo index, flight phase sequence, and phase flight data segments are mapped to establish a data echo channel between the flight data source repository and the mission echo repository. In the data echo channel, the flight data segments of each stage are read frame by frame according to the flight time and compared. When any one of the following changes in adjacent flight times: the trajectory direction corresponding to the positioning trajectory data, the attitude angle corresponding to the attitude state data, the link connection state corresponding to the link state data, the load working state corresponding to the load state data, or the control command category corresponding to the control command data, the next flight time is determined as the echo positioning time, and the flight data corresponding to the echo positioning time is determined as the flight echo node. Determine the echo node type corresponding to the flight echo node based on the changed data type, and obtain the flight phase identifier corresponding to the flight echo node and the original data location in the flight data original repository. The flight phase identifier, echo location time, echo node type, and original data location are combined into an echo node record, and each echo node record is arranged according to the echo location time to generate a mission echo segment. In the specific implementation process, the data echo channel is implemented by mapping the lightweight index on the mission side to the physical address on the original data warehouse side. The original flight data warehouse saves complete flight data according to 1 second as a flight moment, while the mission echo warehouse saves the mission echo index and mission echo record, without repeatedly saving positioning trajectory data, attitude status data, link status data, payload status data and control command data.

[0022] Preferably, during the frame-by-frame comparison stage, the included angle of the trajectory direction corresponding to the positioning trajectory data at adjacent flight times is calculated. When the included angle of the trajectory direction is greater than or equal to 5 degrees, it is determined that the trajectory direction corresponding to the positioning trajectory data has changed. The difference in attitude angles corresponding to the attitude state data at adjacent flight times is calculated. When the difference in attitude angles is greater than or equal to 3 degrees, it is determined that the attitude angle corresponding to the attitude state data has changed. When the link connection status changes from connected to disconnected or from disconnected to connected, it is determined that the link connection status corresponding to the link status data has changed. When the payload operating status changes from off to on or from on to off, it is determined that the payload operating status corresponding to the payload status data has changed. When the control command type changes from the control command type of the previous flight time to another control command type of the next flight time, it is determined that the control command type corresponding to the control command data has changed. Therefore, the mission echo fragments only retain key echo nodes that can point back to the original flight data repository, without carrying the full flight data content, thus forming a low-coupling, high-retrieval, and efficient storage architecture. By decoupling the business of the original flight data repository and the mission echo repository, the lightweight time-series stage on the mission side is used to directly map the massive amount of raw data on the physical side, avoiding redundant storage and blind traversal of the full data. When generating echo identity tags and performing consistency comparisons, key flight moments can be located directly based on the mission echo fragments, improving the speed of anomaly location and the efficiency of flight data management.

[0023] In this embodiment, the improved xLSTM model specifically includes a flight echo embedding layer, a scalar memory structure, a flight echo receiving layer, an exponential gating structure, a matrix memory structure, and an echo identity output layer. The flight echo embedding layer reads the records of each echo node in the mission echo segment, encodes and splices the flight phase identifier, echo location time, echo node type and original data location, and obtains the flight echo token through linear mapping and GELU activation processing. Specifically, flight phase identifiers are numbered 1 to 5 according to the takeoff phase, cruise phase, operation phase, return phase, and landing phase, respectively. Echo node types are numbered 1 to 5 according to changes in positioning trajectory, attitude state, link state, payload state, and control command, respectively. The echo positioning time is represented by a second-level offset relative to the start time of the corresponding flight mission, and the second-level offset is divided by the mission duration in seconds to form a time normalization value. The original data location is encoded in three segments according to the partition number, file block number, and frame sequence number, and each segment is mapped to a fixed-length value. During encoding splicing, the flight phase number, time normalization value, echo node type number, and three-segment original location code are arranged in a fixed order to form a six-dimensional basic encoding vector corresponding to a single echo node record. The six-dimensional basic encoding vector is then input into a linear mapping matrix to be mapped to a 128-dimensional flight echo vector, and the flight echo token is obtained through GELU activation processing.

[0024] In the above way, the stage attributes, time attributes, type attributes and original warehouse location attributes in the mission echo fragments are unified into the same vector space. This preserves both the stage position of the flight echo node in the flight mission and the physical position of the flight echo node pointing back to the original warehouse of the flight data. This allows the scalar memory structure to recursively perform stable recursion according to the echo location time, reduces feature offset caused by inconsistent scales of different fields, and improves the accuracy of echo identity tag generation. The scalar memory structure reads the flight echo token according to the echo localization time, and performs temporal recursion on adjacent flight echo tokens through the scalar memory state to obtain the single-line echo memory feature. Specifically, the scalar memory structure reads the flight echo tokens from first to last according to the echo localization time, initializes the scalar memory state corresponding to the first flight echo token to 0, performs a linear mapping on the current flight echo token to obtain the current input scalar, retains and reads the scalar memory state corresponding to the adjacent previous flight echo token to obtain the preceding memory scalar, adds the current input scalar to the preceding memory scalar, and generates the current recursive coefficient through the Sigmoid function, multiplies the current input scalar to obtain the current written scalar, multiplies the preceding memory scalar by the value of 1 minus the current recursive coefficient to obtain the preceding retained scalar, and adds the current written scalar to the preceding retained scalar to obtain the scalar memory state corresponding to the current flight echo token.

[0025] During the continuous recursion process, the scalar memory state is updated item by item according to the echo location time. Each update is formed by the currently written scalar and the previously retained scalar, ensuring that the newly added echo information in the current flight echo token and the historical echo information in the previous flight echo token are continuously transmitted in the same scalar memory state. To avoid early nodes being overwritten too quickly during long-duration flights, the current recursion coefficient is set with a lower limit of 0.2 and an upper limit of 0.8. When the Sigmoid function output value is less than 0.2, the current recursion coefficient is corrected to 0.2; when the Sigmoid function output value is greater than 0.8, the current recursion coefficient is corrected to 0.8. Through the above recursion method, the temporal relationship between adjacent flight echo tokens is compressed into a single-line echo memory feature, which reduces the repeated reading of complete flight data and provides a continuous and stable temporal input for the flight echo relay layer. The flight echo receiving layer is set between the scalar memory structure and the matrix memory structure, and the single-line echo memory features are received in stages according to the flight stage identifier to obtain the received echo features. The exponential gating structure optimizes the flight echo continuity of the received echo characteristics to obtain stable echo characteristics; The matrix memory structure receives stable echo features and writes them into the matrix memory state. The matrix memory state is used to remember and express the correspondence between flight phase identifiers, echo location times and echo node types, thus obtaining matrix echo memory features. Specifically, the stable echo features are global echo features covering the takeoff, cruise, operation, return, and landing phases. After receiving the stable echo features, the matrix memory structure allocates matrix positions according to the flight phase identifier, echo location time, and echo node type. The flight phase identifier is mapped to 5 phase channels, the echo node type is mapped to 5 type channels, and the echo location time is formed according to the node arrangement order in the mission echo segment. The matrix memory state is constructed according to the flight phase channel, time position, and echo node type channel, so that each echo node is recorded in an independent matrix position in the matrix memory state.

[0026] During the matrix memory state construction process, the feature fragments corresponding to each echo node record in the stable echo features are assigned to the corresponding matrix positions according to the flight stage identifier, time position, and echo node type, and are unified into 128-dimensional matrix memory units through linear mapping. At the same time, the original data position corresponding to the echo node record is indexed and bound to the corresponding matrix position, and the original data position does not participate in the feature operation of the matrix memory state. After the matrix memory state construction is completed, one-dimensional convolution processing is performed along the time position direction, linear mapping processing is performed along the flight stage channel direction, and channel compression processing is performed along the echo node type channel direction to obtain the matrix echo memory features.

[0027] Through the above processing, the stable echo features are transformed into a matrix representation jointly defined by the flight phase, echo location time, and echo node type. This enables the matrix echo memory features to form a correspondence with each echo node record in the mission echo segment, facilitating the echo identity output layer to generate corresponding echo identity tags for each echo node record. At the same time, the original warehouse data location is only retained as an index, avoiding the mixing of physical address information and continuous vector features in calculations, thus improving the data consistency of the matrix echo memory features. The echo identity output layer sequentially performs linear mapping, LayerNorm normalization, GELU activation, and Softmax mapping on the matrix echo memory features to obtain the echo identity probability distribution corresponding to each echo node record, and generates the corresponding echo identity tag based on the echo identity probability distribution corresponding to each echo node record. Specifically, the echo identity probability distribution includes credible echo probability, missing echo probability, misaligned echo probability, and drifting echo probability. For each echo node record, the category with the highest probability value in the corresponding echo identity probability distribution is determined as the echo identity category of that echo node record, and a corresponding echo identity tag is generated. When the credible echo probability is the highest, a credible echo identity tag is generated. When the missing echo probability, misaligned echo probability, or drifting echo probability is the highest, a missing echo identity tag, a misaligned echo identity tag, or a drifting echo identity tag is generated, respectively. Each echo identity tag is bound to the corresponding echo node record, and the original data location in the corresponding echo node record corresponds to the flight data in the original flight data repository. The task echo index to which the task echo segment belongs corresponds to the task echo record in the task echo repository, which is used for subsequent consistency comparison.

[0028] In this embodiment, the single-line echo memory features are stage-by-stage transitioned according to the flight stage identifiers to obtain the transitioned echo features, specifically as follows: Obtain the flight phase identifier, echo location time, and echo node type corresponding to the single-line echo memory feature; According to the flight phase identifier, the single-line echo memory features are written into the phase receiving slots corresponding to the takeoff phase, cruise phase, operation phase, return phase and landing phase to obtain the phase echo feature group. Within each stage receiving slot, single-line echo memory features in the stage echo feature group are arranged according to the echo positioning time to generate an echo receiving chain within the stage. Read the single-line echo memory features at the end of the previous stage and the single-line echo memory features at the beginning of the next stage in the order of flight stages, and compare the echo node type corresponding to the single-line echo memory features at the end of the previous stage with the echo node type corresponding to the single-line echo memory features at the beginning of the next stage. When the echo nodes are of the same type, the single-line echo memory feature at the end of the previous stage, the same type of continuation identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the same type of stage continuation feature. When the echo node types are different, the single-line echo memory feature at the end of the previous stage, the heterogeneous transition identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the heterogeneous stage transition feature. Arrange the echo succession chain within a stage, the continuity features of similar stages, and the transition features of dissimilar stages according to the echo location time to obtain the succession echo features. In the specific implementation process, the stage receiving slot is set with 5 fixed slots, and the slot numbers correspond sequentially to the takeoff stage, cruise stage, operation stage, return stage, and landing stage. The slot capacity is set to retain a maximum of 128 single-line echo memory features per stage. When the number of single-line echo memory features in the same stage exceeds 128, 128 features are extracted at equal intervals according to the echo positioning time and written into the corresponding stage receiving slot. Each stage receiving slot is set with a beginning position, an end position, and a middle position. The beginning position stores the earliest single-line echo memory feature of the corresponding stage, the end position stores the latest single-line echo memory feature of the corresponding stage, and the middle position stores the single-line echo memory features within the stage according to the echo positioning time order, so that the echo receiving chain within the stage can retain the echo evolution order within that stage.

[0029] When adjacent phases are connected, the same-type continuity identifier is set to a value of 1, and the different-type transition identifier is set to a value of 0. These, together with the single-line echo memory features at the end of the previous phase and the single-line echo memory features at the beginning of the next phase, form a phase handover record. In this way, the connection echo features are not just single time-series features, but a global echo representation that simultaneously includes the continuity relationship within a phase and the handover relationship between phases. This processing can avoid the loss of boundary information between takeoff, cruise, operation, return, and landing phases due to compression by the scalar memory structure. It enables the exponential gating structure to identify the continuation of the same-type echo and the switching of different-type echoes, improving the ability of echo identity tags to express the consistency of cross-phase flight data.

[0030] In this embodiment, the flight echo continuity of the received echo characteristics is optimized to obtain stable echo characteristics, specifically as follows: The intra-stage echo succession chain, similar stage continuity feature, and dissimilar stage transition feature in the echo succession feature are respectively input into the exponential gating structure. The exponential gating structure performs differential mapping on the adjacent single-line echo memory features in the intra-stage echo relay chain to obtain the intra-stage differential response features. By splicing and mapping the same-type succession identifier, the single-line echo memory feature at the end of the previous stage and the single-line echo memory feature at the beginning of the next stage in the same-type stage succession feature, the same-type succession response feature is obtained. The heterogeneous transition identifier, the single-line echo memory feature at the end of the previous stage, and the single-line echo memory feature at the beginning of the next stage in the heterogeneous stage transition features are spliced ​​and mapped to obtain the heterogeneous transition response features. The intra-stage differential response characteristics, similar succession response characteristics, and dissimilar transition response characteristics are input into the exponential function to obtain the intra-stage gating value, similar succession gating value, and dissimilar transition gating value. Multiply the in-stage gating value element by element with the in-stage echo relay chain to obtain the in-stage gating relay chain. The gating values ​​of the same type of continuation are multiplied element by element with the continuation features of the same type of stage, and the residuals are superimposed to obtain the gating features of the same type of continuation. The heterogeneous transition gating value is multiplied element-wise with the heterogeneous stage transition feature, and then concatenated with the heterogeneous stage transition feature to obtain the heterogeneous gating transition feature. The gated connection chain, similar gated connection features, and dissimilar gated transition features within the stage are spliced ​​together according to the echo localization time, and then subjected to residual connection and LayerNorm normalization to obtain stable echo features. In the specific implementation process, the single-line echo memory feature in the echo succession chain within each stage is set as a 128-dimensional vector. For the first single-line echo memory feature within each stage, since there is no preceding single-line echo memory feature, the corresponding difference vector is set as a 128-dimensional zero vector, serving as the stage difference boundary. Starting from the second single-line echo memory feature, the current single-line echo memory feature is differentially processed with the adjacent preceding single-line echo memory feature to obtain a 128-dimensional difference vector. Then, the stage difference response feature is obtained through linear mapping. (The text then mentions similar stage succession features and dissimilar stage features.) All transition features adopt a three-segment input format. The single-line echo memory feature at the end of the previous stage and the single-line echo memory feature at the beginning of the next stage are both set as 128-dimensional vectors. The same-type continuation identifier is set to 1, and the different-type transition identifier is set to 0. The identifier is expanded into a 128-dimensional identifier vector. During the splicing mapping, the single-line echo memory feature at the end of the previous stage, the identifier vector, and the single-line echo memory feature at the beginning of the next stage are spliced ​​according to the channel dimension. Then, they are compressed into 128-dimensional response features through linear mapping, so that changes within the stage, continuation of the same-type stage, and transition of different-type stages enter a unified dimensional space.

[0031] The exponential function is used to convert intra-stage differential response features, similar successive response features, and dissimilar transitional response features into gate values ​​ranging from 0 to 1. The lower limit of the gate value is set to 0.15, and the upper limit is set to 0.95. When the output is less than 0.15, it is corrected to 0.15; when the output is greater than 0.95, it is corrected to 0.95 to prevent key echo information from being completely blocked or noise information from being amplified indiscriminately. When superimposing residuals, the similar successive gate value is multiplied element-wise with the similar stage successive features, and then added to the 128-dimensional response features compressed by the linear mapping, so that continuous information with the same echo node type in adjacent flight stages can be preserved. During splicing mapping... The method involves element-wise multiplying the heterogeneous transition gating value with the heterogeneous stage transition feature, then concatenating and linearly mapping the gating value along the channel dimension. This reorganizes the switching information of different echo node types in adjacent flight stages into a transition expression. In this invention, the same-type stage continuation feature corresponds to the continuation of the same echo node type between adjacent flight stages. Residual superposition preserves the original continuation backbone and the effective response after gating. The heterogeneous stage transition feature corresponds to the switching of different echo node types between adjacent flight stages. Concatenation mapping rearranges the feature channels before and after the transition, reducing semantic confusion caused by directly adding different echo types. Through this processing, stable echo features can simultaneously preserve continuous changes within stages, continuation of the same type between stages, and heterogeneous switching between stages, reducing echo breaks at flight stage boundaries and improving the stability of the matrix memory structure in representing the echo identity throughout the flight mission.

[0032] In this invention, the improved xLSTM model is a structured improvement model formed on the long sequence memory framework of the existing xLSTM model for UAV flight data management scenarios. The existing xLSTM model relies on scalar memory structure, matrix memory structure and exponential gating structure to maintain the state of continuous sequences, which is suitable for handling data dependencies with long time spans. This invention retains this basic memory path, so that the task echo segments can still form a single-line temporal recursion along the echo localization time, and further forms a multi-dimensional memory representation through the matrix memory structure, thereby ensuring the continuity of the improved model with the existing xLSTM model in the basic computational framework.

[0033] Compared to existing xLSTM models, this invention does not directly feed single-line echo memory features into the matrix memory structure. Instead, it adds a flight echo relay layer between the scalar memory structure and the matrix memory structure. This allows the flight echo information in the single-line time series to be relayed in stages according to takeoff, cruise, operation, return, and landing phases, forming echo relay chains within stages, similar stage continuity features, and dissimilar stage transition features. Simultaneously, the exponential gating structure is no longer updated solely based on ordinary time series states. Instead, it forms gating responses for intra-stage continuity, inter-stage similar continuity, and inter-stage dissimilar transitions in the relayed echo features, ensuring that echo information at flight stage transitions can be stably preserved or reorganized. These improvements enable the model to simultaneously express task stage relationships, echo node type relationships, and echo location timing relationships, reducing echo breaks at flight stage boundaries and semantic confusion between different echo types, and improving the ability of echo identity markers to distinguish between reliable echoes, missing echoes, misaligned echoes, and drifting echoes.

[0034] In this embodiment, step five specifically includes: Events generated when flight data in the original flight data repository is viewed, exported, copied, modified, deleted, migrated, compressed and archived, or restored and retrieved are defined as operation events; When an operation event occurs in the flight data in the original flight data repository, the location of the original data in the repository corresponding to the operation event is obtained; Based on the location of the original warehouse data, find the corresponding echo node record in the data echo channel, and determine the corresponding task echo record based on the echo node record; The echo identity markers of the corresponding flight data in the original flight data repository are read back through the data echo channel, and the echo identity markers of the corresponding mission echo records in the mission echo repository are read back. The echo identification tags of the corresponding flight data are compared with the echo identification tags of the corresponding mission echo records for consistency. When two echo identity tags are the same, the echo identity tags are determined to be consistent; When two echo identity tags are different, it is determined that the echo identity tags are inconsistent; This invention unifies viewing, exporting, copying, modifying, deleting, migrating, compressing and archiving, and restoring / retrieving data into a single operation event. This ensures that flight data in the original flight data repository triggers echo identity verification before use or alteration, avoiding the omission of high-risk operations such as exporting, copying, and restoring / retrieving data due to checks only at the modification stage. After an operation event occurs, the location of the original data repository serves as the positioning entry point. The corresponding echo node record is searched in the data echo channel, and the corresponding task echo record is further identified, establishing an accurate correspondence between the data in the original flight data repository and the task-side record in the task echo repository. Subsequently, the echo identity markers in the corresponding flight data and the corresponding task echo record are read back and directly compared for consistency. This method enables credibility verification before flight data enters external use, internal migration, or status changes, improving the anti-tampering capability, traceability efficiency, and anomaly isolation accuracy in the UAV flight data management process.

[0035] In this embodiment, step six specifically includes: When the echo identity tags match, the corresponding flight data is marked as trustworthy flight data, and the original data position of the corresponding flight data in the original data repository is maintained. When the echo identification tags are inconsistent, the corresponding flight data will be marked as abnormal echo data, and an isolated echo zone will be established in the original flight data repository. The isolated echo zone is a management area in the original flight data repository that is independent of the trusted flight data. It is used to store abnormal echo data and isolated echo records. Transfer abnormal echo data to the isolated echo zone, and retain the original warehouse data position corresponding to the abnormal echo data before it was transferred to the isolated echo zone; Based on the operation event type, the location of the original warehouse data, the echo identity tag corresponding to the flight data, the echo identity tag corresponding to the mission echo record, and the isolation write time, an isolation echo record is generated. The isolated echo record is bound to the abnormal echo data, and the echo relationship between the isolated echo record and the corresponding echo node record and the corresponding task echo record is maintained through the data echo channel; In this invention, the isolated echo record simultaneously saves the operation event type, the original data location, the echo identification markers on both sides, and the isolation write time. It can clearly record the source of the anomaly and the triggering process. The data echo channel maintains the retracement relationship between the isolated echo record and the corresponding echo node record and the corresponding task echo record. It does not destroy the original storage structure of the flight data original repository and the task echo repository, and facilitates traceability, repair or freezing, thereby improving the security and traceability of UAV flight data management.

[0036] In this embodiment, step seven specifically includes: Obtain abnormal echo data and isolation echo records from the isolated echo zone; Determine the original data location, operation event type, and isolation write time corresponding to the abnormal echo data based on the isolation echo record; Mark the management status of abnormal echo data as isolated management status; Write the original data location, operation event type, isolation write time, and isolation management status back to the original flight data repository; Based on the isolation management status, abnormal echo data can be restricted from being viewed, exported, copied, modified, deleted, migrated, compressed and archived, or restored and retrieved. This invention writes the original data location, operation event type, isolation write time, and isolation management status of abnormal echo data back to the flight data original repository. This gives the abnormal echo data a clear status identifier within the existing data management system, preventing abnormal data from being viewed, exported, copied, modified, deleted, migrated, compressed, archived, or restored as ordinary trusted data. Once the isolation management status corresponds to the original data location, the flight data original repository can directly identify abnormal echo data before subsequent data retrieval, eliminating the need for repeated full echo identity comparisons and improving data management efficiency. Simultaneously, the operation event type and isolation write time record the source of the anomaly and the time of isolation, facilitating subsequent accountability and anomaly review. Through these processes, abnormal echo data is incorporated into an independent management process, establishing a clear boundary between trusted and abnormal flight data, thus improving the security, traceability, and data status maintenance capabilities of UAV flight data management.

[0037] refer to Figure 2 An AI-based drone flight data management system includes the following modules: The dual-warehouse construction module is used to establish a flight data origin warehouse and a mission echo warehouse. It writes the flight data generated by each flight of the UAV into the flight data origin warehouse and builds mission echo records in the mission echo warehouse. The data echo channel construction module is used to establish a data echo channel between the flight data source and the mission echo source, perform echo localization of flight data based on mission echo records, and generate mission echo segments. The echo identity tag generation module is used to input mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. The echo identification tag writing module is used to write the echo identification tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo store through the data echo channel; The operation event comparison module is used to read back the corresponding flight data and the echo identity tag in the corresponding mission echo record through the data echo channel when an operation event occurs in the flight data in the original flight data repository, and perform a consistency comparison. The anomaly isolation processing module is used to mark the corresponding flight data as reliable flight data when the echo identity tags are consistent, and to mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone when the echo identity tags are inconsistent. The isolation management write-back module is used to isolate and manage abnormal echo data based on the isolation echo records in the isolation echo zone, and write the management status of the abnormal echo data back to the flight data source.

[0038] Example 1: To verify the feasibility of this invention in practice, it was applied to a UAV inspection data management platform. This platform receives flight data daily from multiple UAVs performing route inspections, area mapping, and emergency verification tasks. Each task typically lasts 35 to 70 minutes. The original management method primarily stored flight logs centrally according to task numbers. When data was subsequently exported, copied, archived, or restored, the reliability of the data could only be determined manually by reviewing the log files and operation records. This was particularly problematic when flight tasks spanned multiple stages and data files were migrated multiple times, leading to issues such as slow location at critical moments, continued flow of abnormal data, and unclear correspondence between task records and original flight data.

[0039] In this embodiment, the platform first establishes a flight data origin repository and a mission echo repository for each flight. The flight data origin repository stores complete flight data according to the flight mission number, UAV number, and flight time. The mission echo repository constructs mission echo records according to the takeoff phase, cruise phase, operation phase, return phase, and landing phase. Taking a certain inspection mission as an example, the total mission duration is 52 minutes. The system writes 3120 second-level flight times into the flight data origin repository and records 5 flight phases and their corresponding start and end times in the mission echo repository. Subsequently, the data echo channel extracts stage flight data segments based on the mission echo records, compares the changes in trajectory direction, attitude angle, link connection status, payload working status, and control command type frame by frame, extracts a total of 146 flight echo nodes, and generates mission echo segments. After the mission echo segments are input into the improved xLSTM model, the flight echo receiving layer performs stage receiving of single-line echo memory features according to the flight phase identifier. The exponential gating structure stably expresses the continuity within the stage, the similar continuation between stages, and the heterogeneous transition, and finally generates the echo identity tag corresponding to each echo node record. When flight data is viewed, exported, copied, modified, deleted, migrated, compressed and archived, or restored and retrieved, the platform compares the consistency of the echo identity tags on both sides through the data echo channel. If the tags are inconsistent, the corresponding flight data is marked as abnormal echo data and transferred to the isolation echo zone. Subsequently, the isolation management status is written back based on the isolation echo record.

[0040] To verify the practical effectiveness of this invention, three comparison schemes were set up. Comparison Scheme 1 is the traditional centralized log management scheme, which saves all logs only according to flight mission number and relies on manual querying; Comparison Scheme 2 is the mission index management scheme, which establishes mission indexes according to flight phases but does not generate echo identity markers; Comparison Scheme 3 is the ordinary xLSTM recognition scheme, which inputs mission segments into a standard xLSTM model but does not set up a flight echo continuation layer or optimize flight echo continuity. Test data includes 30 consecutive days of operational data, totaling 480 UAV flight missions, accumulating approximately 186.4 GB of flight data. 92 abnormal echo data points were manually verified. The comparison results are shown in Table 1.

[0041] Table 1 Comparison of Flight Data Echolocation and Identity Verification Results

[0042] According to the data in Table 1 above, under the same conditions of 480 test tasks and 92 manually confirmed abnormal data, different solutions show significant differences in their ability to identify and isolate abnormal echo data. Solution 1, using traditional centralized log management, achieved an anomaly identification accuracy of only 78.6%, with 17 abnormal data missed for isolation. This indicates that relying solely on task numbers and original log retrieval is insufficient to promptly detect abnormal changes in flight data during viewing, exporting, copying, migrating, or recovery. Solution 2, after introducing a task index, improved the anomaly identification accuracy to 85.4% and reduced the number of missed isolations to 11, indicating that task-stage records can improve data location efficiency. However, the lack of consistency comparison of echo identity markers still makes it easy to miss some abnormal data. Solution 3... The ordinary xLSTM model further improved the anomaly identification accuracy to 91.8% and reduced the number of missing isolations to 6, indicating that the time series model can enhance the anomaly identification capability. However, without the flight echo continuation layer and flight echo continuity optimization, echo breaks or misalignments still exist at stage transitions. The anomaly identification accuracy of the method of this invention reached 97.5%, with only 1 missing isolation of anomalous data. This verifies that the flight data source, mission echo, data echo channel, and improved xLSTM model working together can more accurately identify anomalous echo data and complete isolation management in a timely manner.

[0043] This embodiment verifies the feasibility and effectiveness of the present invention in UAV flight data management scenarios. By establishing a flight data source repository and a mission echo repository, and using the data echo channel to map flight data to mission echo records, the system can quickly locate key flight moments, reducing the workload of manually searching for original logs. After improving the xLSTM model to generate echo identity tags, consistency comparison can be performed when flight data is viewed, exported, copied, modified, migrated, or restored, promptly distinguishing between reliable flight data and abnormal echo data, thereby reducing the risk of abnormal data continuing to circulate and improving the reliable traceability and security management level of flight data.

[0044] 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 equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An AI-based method for managing drone flight data, characterized in that, Includes the following steps: Step 1: Establish the flight data source repository and the mission echo repository. Write the flight data generated by each flight of the UAV into the flight data source repository, and build the mission echo record in the mission echo repository. Step 2: Establish a data echo channel between the flight data source and the mission echo source, perform echo localization on the flight data based on the mission echo record, and generate mission echo segments. Step 3: Input the mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. Step 4: Write the echo identity tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo repository through the data echo channel; Step 5: When an operation event occurs in the flight data in the original flight data repository, the corresponding flight data and the echo identity tag in the corresponding mission echo record are read back through the data echo channel, and a consistency comparison is performed; Step 6: When the echo identification tags match, mark the corresponding flight data as reliable flight data; when the echo identification tags do not match, mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone. Step 7: Based on the isolated echo records in the isolated echo zone, manage the isolated echo data.

2. The AI-based drone flight data management method according to claim 1, characterized in that, Step one specifically involves: Collect flight data generated by the UAV for each flight. The flight data includes flight mission number, UAV number, flight time, positioning trajectory data, attitude status data, link status data, payload status data, and control command data. The flight data is completely written into the original flight data repository according to the flight mission number, UAV number, and flight time. Obtain the mission phase configuration corresponding to the flight mission, which includes the takeoff phase, cruise phase, operation phase, return phase, and landing phase; Flight data is categorized into the corresponding takeoff phase, cruise phase, operation phase, return phase, or landing phase according to flight time, thus obtaining the flight phase sequence. Based on the flight mission number, UAV number, flight phase sequence, and the start and end times of each flight phase, a mission echo record is constructed. Configure a task echo index for the task echo record and write the task echo record to the task echo repository.

3. The AI-based drone flight data management method according to claim 1, characterized in that, Step two specifically involves: Obtain the mission echo index, flight mission number, UAV number, flight phase sequence, and start and end times of each flight phase corresponding to the mission echo record; Based on the flight mission number and UAV number, the corresponding flight data is searched in the original flight data repository, and the flight data within the corresponding flight stage is extracted according to the start and end times of each flight stage to obtain the stage flight data segment. The mission echo index, flight phase sequence, and phase flight data segments are mapped to establish a data echo channel between the flight data source repository and the mission echo repository. In the data echo channel, the flight data segments of each stage are read frame by frame according to the flight time and compared. When any one of the following changes in adjacent flight times: the trajectory direction corresponding to the positioning trajectory data, the attitude angle corresponding to the attitude state data, the link connection state corresponding to the link state data, the load working state corresponding to the load state data, or the control command category corresponding to the control command data, the next flight time is determined as the echo positioning time, and the flight data corresponding to the echo positioning time is determined as the flight echo node. The type of echo node corresponding to the flight echo node is determined based on the data type that has changed, and the flight phase identifier and the location of the original data in the original data repository are obtained. The flight phase identifier, echo location time, echo node type, and original data location are combined into an echo node record, and each echo node record is arranged according to the echo location time to generate a mission echo segment.

4. The AI-based drone flight data management method according to claim 1, characterized in that, The improved xLSTM model specifically includes a flight echo embedding layer, a scalar memory structure, a flight echo receiving layer, an exponential gating structure, a matrix memory structure, and an echo identity output layer; The flight echo embedding layer reads the records of each echo node in the mission echo segment, encodes and splices the flight phase identifier, echo location time, echo node type and original data location, and obtains the flight echo token through linear mapping and GELU activation processing. The scalar memory structure reads the flight echo token according to the echo localization time, and performs temporal recursion on adjacent flight echo tokens through the scalar memory state to obtain single-line echo memory features. The flight echo receiving layer is set between the scalar memory structure and the matrix memory structure, and performs stage receiving of single-line echo memory features according to the flight stage identifier to obtain the received echo features. The exponential gating structure optimizes the flight echo continuity of the received echo features to obtain stable echo features. The matrix memory structure receives stable echo features and writes the stable echo features into the matrix memory state. The matrix memory state is used to remember and express the correspondence between flight phase identifiers, echo location times, and echo node types, thus obtaining matrix echo memory features. The echo identity output layer sequentially performs linear mapping, LayerNorm normalization, GELU activation, and Softmax mapping on the matrix echo memory features to obtain the echo identity probability distribution corresponding to each echo node record, and generates the corresponding echo identity tag based on the echo identity probability distribution corresponding to each echo node record.

5. The AI-based drone flight data management method according to claim 4, characterized in that, The process of performing stage-based transfer of single-line echo memory features according to flight stage identifiers to obtain transferred echo features is as follows: Obtain the flight phase identifier, echo location time, and echo node type corresponding to the single-line echo memory feature; According to the flight phase identifier, the single-line echo memory features are written into the phase receiving slots corresponding to the takeoff phase, cruise phase, operation phase, return phase and landing phase to obtain the phase echo feature group. Within each stage receiving slot, single-line echo memory features in the stage echo feature group are arranged according to the echo positioning time to generate an echo receiving chain within the stage. Read the single-line echo memory features at the end of the previous stage and the single-line echo memory features at the beginning of the next stage in the order of flight stages, and compare the echo node type corresponding to the single-line echo memory features at the end of the previous stage with the echo node type corresponding to the single-line echo memory features at the beginning of the next stage. When the echo node types are the same, the single-line echo memory feature at the end of the previous stage, the same type of continuation identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the same type of stage continuation feature. When the echo node types are different, the single-line echo memory feature at the end of the previous stage, the heterogeneous transition identifier, and the single-line echo memory feature at the beginning of the next stage are connected in sequence to generate the heterogeneous stage transition feature. Arrange the echo succession chain within a stage, the continuity features of similar stages, and the transition features of dissimilar stages according to the echo location time to obtain the succession echo features.

6. The AI-based drone flight data management method according to claim 4, characterized in that, The optimization of flight echo continuity of the received echo features to obtain stable echo features specifically involves: The intra-stage echo succession chain, similar stage continuity feature, and dissimilar stage transition feature in the echo succession feature are respectively input into the exponential gating structure. The exponential gating structure performs differential mapping on the adjacent single-line echo memory features in the intra-stage echo relay chain to obtain the intra-stage differential response features. By splicing and mapping the same-type succession identifier, the single-line echo memory feature at the end of the previous stage and the single-line echo memory feature at the beginning of the next stage in the same-type stage succession feature, the same-type succession response feature is obtained. The heterogeneous transition identifier, the single-line echo memory feature at the end of the previous stage, and the single-line echo memory feature at the beginning of the next stage in the heterogeneous stage transition features are spliced ​​and mapped to obtain the heterogeneous transition response features. By inputting the intra-stage differential response characteristics, similar succession response characteristics, and dissimilar transition response characteristics into the exponential function, the intra-stage gating value, similar succession gating value, and dissimilar transition gating value are obtained. The in-stage gating value is multiplied element-wise with the in-stage echo relay chain to obtain the in-stage gating relay chain. The similar type of continuation gate value is multiplied element by element with the similar type of stage continuation feature, and the residuals are superimposed to obtain the similar type of gated continuation feature. The heterogeneous transition gating value is multiplied element-wise with the heterogeneous stage transition feature, and then concatenated and mapped with the heterogeneous stage transition feature to obtain the heterogeneous gating transition feature. The gated connection chain, similar gated connection features, and dissimilar gated transition features within the stage are spliced ​​together according to the echo localization time, and then subjected to residual connection and LayerNorm normalization to obtain stable echo features.

7. The AI-based drone flight data management method according to claim 1, characterized in that, Step five specifically involves: Events generated when flight data in the original flight data repository is viewed, exported, copied, modified, deleted, migrated, compressed and archived, or restored and retrieved are defined as operation events; When an operation event occurs in the flight data in the original flight data repository, the location of the original data in the repository corresponding to the operation event is obtained; Based on the location of the original warehouse data, find the corresponding echo node record in the data echo channel, and determine the corresponding task echo record based on the echo node record; The echo identity markers of the corresponding flight data in the original flight data repository are read back through the data echo channel, and the echo identity markers of the corresponding mission echo records in the mission echo repository are read back. The echo identification tags of the corresponding flight data are compared with the echo identification tags of the corresponding mission echo records for consistency. When two echo identity tags are the same, the echo identity tags are determined to be consistent; When two echo identity tags are different, it is determined that the echo identity tags are inconsistent.

8. The AI-based drone flight data management method according to claim 1, characterized in that, Step six specifically involves: When the echo identity tags match, the corresponding flight data is marked as trustworthy flight data, and the original data position of the corresponding flight data in the original data repository is maintained. When the echo identification tags are inconsistent, the corresponding flight data will be marked as abnormal echo data, and an isolated echo zone will be established in the original flight data repository. The isolated echo zone is a management area in the original flight data repository that is independent of the trusted flight data, and is used to store abnormal echo data and isolated echo records. Transfer abnormal echo data to the isolated echo zone, and retain the original warehouse data position corresponding to the abnormal echo data before it was transferred to the isolated echo zone; Based on the operation event type, the location of the original warehouse data, the echo identity tag corresponding to the flight data, the echo identity tag corresponding to the mission echo record, and the isolation write time, an isolation echo record is generated. The isolated echo records are bound to the abnormal echo data, and the echo relationship between the isolated echo records and the corresponding echo node records and the corresponding task echo records is maintained through the data echo channel.

9. The AI-based drone flight data management method according to claim 1, characterized in that, Step seven specifically involves: Obtain abnormal echo data and isolation echo records from the isolated echo zone; The location of the original warehouse data, the type of operation event, and the time of isolation write are determined based on the isolated echo record; Mark the management status of abnormal echo data as isolated management status; Write the original data location, operation event type, isolation write time, and isolation management status back to the original flight data repository; Based on the isolation management status, abnormal echo data is restricted from being viewed, exported, copied, modified, deleted, migrated, compressed, archived, or restored.

10. An AI-based drone flight data management system, executing the AI-based drone flight data management method according to any one of claims 1 to 9, characterized in that, Includes the following modules: The dual-warehouse construction module is used to establish a flight data origin warehouse and a mission echo warehouse. It writes the flight data generated by each flight of the UAV into the flight data origin warehouse and builds mission echo records in the mission echo warehouse. The data echo channel construction module is used to establish a data echo channel between the flight data source and the mission echo source, perform echo localization of flight data based on mission echo records, and generate mission echo segments. The echo identity tag generation module is used to input the mission echo fragments into the improved xLSTM model. The improved xLSTM model adds a flight echo receiving layer between the scalar memory structure and the matrix memory structure, and optimizes the flight echo continuity of the exponential gating structure to generate echo identity tags. The echo identification tag writing module is used to write the echo identification tag into the corresponding flight data in the flight data source and the corresponding mission echo record in the mission echo store through the data echo channel; The operation event comparison module is used to read back the corresponding flight data and the echo identity tag in the corresponding mission echo record through the data echo channel when an operation event occurs in the flight data in the original flight data repository, and perform a consistency comparison. The anomaly isolation processing module is used to mark the corresponding flight data as reliable flight data when the echo identity tags are consistent, and to mark the corresponding flight data as abnormal echo data and transfer it to the isolated echo zone when the echo identity tags are inconsistent. The isolation management write-back module is used to isolate and manage abnormal echo data according to the isolation echo records in the isolation echo zone, and write the management status of the abnormal echo data back to the flight data source.

Images