Multi-source unmanned aerial vehicle flight record analysis method and system based on UTM cloud
By using the multi-source UAV flight record analysis method in the UTM cloud, the problem of multi-source data fusion and accurate analysis has been solved. It has achieved dynamic adaptation and data standardization of cross-manufacturer flight control systems, improved the accuracy of abnormal flight event identification and operational error judgment, and met the safety monitoring and accountability requirements in large-scale UAV operation scenarios.
Patent Information
- Application Number
- CN202511243907.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-02
- Publication Date
- 2025-11-07
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing UAV flight record analysis technologies have significant limitations in multi-source data fusion, intelligent analysis, and accurate judgment, making it difficult to achieve unified collection, secure storage, and accurate analysis of multi-source flight records. This makes it impossible to meet the needs of flight safety monitoring, operation optimization, and accountability in large-scale UAV operation scenarios.
By employing a multi-source UAV flight record analysis method based on UTM cloud, a multi-source flight control docking module is used to establish data docking with UAV flight control systems from different manufacturers. Data is uploaded to the cloud storage module in real time and analyzed in multiple dimensions. An intelligent analysis module is used to identify abnormal flight events, and Bayesian optimization algorithm and random forest classifier are combined to dynamically adjust the threshold, thereby achieving data standardization, secure storage and accurate analysis.
It achieves dynamic adaptation across different manufacturers' flight control systems, improves data recognition accuracy and compatibility, reduces storage costs, and enhances the accuracy of identifying abnormal flight events and determining operational errors, thus meeting the safety monitoring and accountability requirements in large-scale UAV operation scenarios.
Smart Images

Figure CN120913292A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of flight record analysis, in particular to a multi-source unmanned aerial vehicle flight record analysis method and system based on a UTM cloud. BACKGROUND
[0002] With the rapid development of unmanned aerial vehicle technology, its application scenarios have expanded to aerial photography, high-altitude cleaning, logistics transportation, power inspection, surveying and mapping, and other fields. Deep analysis of flight record data (such as flight trajectory, operation instructions, and device state parameters) is critical to ensuring flight safety and optimizing operation processes. However, existing unmanned aerial vehicle flight record analysis technology has significant limitations in multi-source data fusion, intelligent analysis, and accurate determination, including the following specific problems: I. Multi-source flight control data interfacing and standardization difficulties Different manufacturers' unmanned aerial vehicle flight control systems have different technical standards, communication protocols, and data formats, resulting in "data island" phenomenon, making it difficult to effectively integrate multi-source flight record data.
[0003] II. Limitations of flight record data storage and transmission Unmanned aerial vehicle flight record data has characteristics such as "multiple types (time sequence trajectory data, discrete operation instructions), strong real-time (abnormal events require second-level response), and safety sensitivity (including accurate geographic location, operator ID)". Existing storage and transmission solutions have the following defects: low data compression and transmission efficiency, weak security mechanism, and imbalance between storage cost and access efficiency. III. Insufficient accuracy of abnormal flight event identification Existing abnormal identification technology relies on static thresholds and single indicators, which cannot adapt to complex flight environments and dynamic device states.
[0004] IV. Subjectivity and one-sidedness of human operation error determination Existing technology relies on manual review or simple rule matching to determine the cause of abnormal events, lacking quantitative analysis and multi-dimensional verification.
[0005] The above problems make it difficult for existing technology to achieve unified collection, safe storage, and accurate analysis of multi-source unmanned aerial vehicle flight records, and cannot meet the needs of flight safety monitoring, operation optimization, and responsibility tracing in large-scale unmanned aerial vehicle operation scenarios. Therefore, there is an urgent need for a multi-source unmanned aerial vehicle flight record analysis technology that can break through "data islands", dynamically adapt to environmental and device states, and quantitatively determine operation errors. SUMMARY
[0006] In order to overcome the prior art, the purpose of the present application is to provide a multi-source unmanned aerial vehicle flight record analysis method and system based on UTM cloud, which is used to solve the technical problems of multi-source flight control data docking and standardization, flight record data storage and transmission limitations, insufficient precision of abnormal flight event identification, and subjective and one-sidedness of manual operation error determination, so as to realize unified collection, safe storage and accurate analysis of multi-source flight record data, and meet the flight safety monitoring, operation optimization and responsibility tracing requirements under large-scale unmanned aerial vehicle operation scenarios.
[0007] To solve the above problems, the technical scheme adopted by the present application is as follows: A multi-source unmanned aerial vehicle flight record analysis method based on UTM cloud, comprising the following steps: Through the multi-source flight control docking module, data docking is established with the flight control systems of at least two different manufacturers, and flight record data including flight trajectory, operation instruction and equipment state parameter are analyzed and collected; The flight record data is uploaded to the cloud storage module in real time for centralized storage; The intelligent analysis module is used to analyze the flight record data in multiple dimensions, including: real-time monitoring of flight status to identify abnormal flight events; when an abnormality is detected, historical and real-time flight record data are called, and the cause is determined whether it is a manual operation error through data correlation analysis; Among them, the external environment parameters are introduced as correction factors to normalize the basic indicators and composite indicators in the identification process, and the Bayesian optimization algorithm is used to dynamically adjust the abnormality judgment threshold; The determination process is based on a three-dimensional operation baseline model and an operation error identification model trained by a random forest classifier to output an operation error probability score.
[0008] Preferably, when analyzing and collecting flight record data, it includes: Through the adaptive protocol conversion unit built in the multi-source flight control docking module, the manufacturer type of the flight control system is identified based on the pre-stored communication protocol feature library, and the protocol analysis rule is automatically loaded for data standardization conversion; before data transmission, the abnormal values are filtered through the preset threshold range to ensure that the data meet the analysis standard.
[0009] Preferably, when identifying the manufacturer type of the flight control system, it includes: The dynamic feature library update unit built in the multi-source flight control docking module periodically acquires the latest feature data of each manufacturer's communication protocol from the cloud to update the local communication protocol feature library; The initial handshake data packet of the accessed flight control is extracted, the protocol type identifier, device model code and firmware version number are analyzed, and the multi-dimensional matching calculation confidence is matched with the feature library; When the confidence is higher than the threshold, the manufacturer type is directly determined; when the confidence is lower than the threshold, a fuzzy matching and self-learning mechanism is started, the communication characteristic parameters are recorded, and a to-be-identified queue is generated, and the characteristic library is updated after being confirmed by an administrator.
[0010] Preferably, when the data standardization conversion is performed, the following is included: According to the firmware version number of the flight control system, the protocol release time and the historical analysis success rate, a dynamic priority is allocated to the multi-version protocol rules of the same manufacturer; When the old version protocol is detected, the basic analysis rule and the historical patch package are automatically loaded, and the abandoned field is marked as a compatible reserved field; A pre-defined standardization field set is stored, and the correspondence between the private fields of each manufacturer and the standard fields and the unit conversion formula are stored; The converted data needs to pass through field integrity verification and logical consistency verification, and the field repair mechanism is triggered when the verification fails.
[0011] Preferably, when the data standardization conversion is performed, the following is included: The field analysis failure rate and the semantic conflict frequency are counted in real time by deploying an analysis error monitoring unit, and when the error rate of a single batch of data exceeds the threshold, the current rule loading is automatically suspended and the standby rule is started; The error data segment and the original protocol frame are uploaded to the cloud intelligent analysis module, the rule correction suggestion is generated combined with the historical successful cases, and the local protocol analysis rule and the multi-level semantic mapping library are updated after being confirmed by an administrator; For encrypted flight record data, temporary decryption parameters are requested from the cloud key management module based on the real-time updated encryption algorithm identifier and version number mapping relationship, and the decrypted data needs to pass through secondary verification.
[0012] Preferably, when the real-time uploading to the cloud storage module for centralized storage is performed, the following is included: The data is dynamically blocked according to the data type characteristics, the Douglas-Peucker algorithm is used for simplification and compression of the trajectory data block, the LZ77 dictionary compression is used for the device state data block with a compression ratio of 1:3~1:5, and the operation instruction data block is not compressed; a multi-dimensional verification identifier is attached to the data block header; The transmission layer uses a TLS 1.3 encrypted channel, the sensitive fields in the data layer are independently encrypted using the AES-256-GCM algorithm, and the encryption key is dynamically generated by the cloud key management module and periodically rotated; According to the real-time and importance, the data is divided into hot, warm and cold data layers, the hot data is stored in the cloud SSD cluster, the warm data is stored in the hybrid storage array, and the cold data is automatically migrated to the object storage; Three copies and erasure code hybrid backup are used, the uploading progress is recorded in real time, the breakpoint resume is triggered when the network is interrupted or the node fails, and only the missing or damaged data block is retransmitted by comparing the local cache checksum and the cloud index.
[0013] Preferably, when uploading to the cloud storage module in real time for centralized storage, it also includes: Deploy a network state awareness unit to monitor the upload link bandwidth fluctuation, packet loss rate and delay in real time at a preset sampling frequency, automatically enable the data priority queue when the link quality deteriorates, and dynamically adjust the MTU value and TCP congestion control algorithm; The cloud storage module performs hierarchical verification after receiving the data, automatically triggers the data repair process when the verification fails, and feeds back the abnormal data characteristics to optimize the subsequent preprocessing rules; Deploy edge computing nodes in areas with intensive UAV operations, temporarily cache flight record data locally, and only upload preprocessed incremental data; By analyzing historical data access frequency to establish a dynamic migration threshold, automatically migrate data at different levels.
[0014] Preferably, when identifying abnormal flight events, it includes: Establish a basic index feature library based on real-time collected flight trajectories, device state parameters and operation instruction sequences; Generate composite indicators through feature engineering and establish a composite indicator feature library; Introduce external environmental parameters as correction factors to normalize the basic indicators and composite indicators; Train a model-task baseline model based on historical non-abnormal flight data to generate abnormal judgment thresholds for each indicator in different operation scenarios; Combine the current flight phase, remaining power and device health, and dynamically adjust the abnormal judgment thresholds of each indicator through the Bayesian optimization algorithm; When the normalized basic indicators and / or composite indicators exceed the corresponding real-time abnormal judgment thresholds, it is determined that an abnormal flight event has occurred.
[0015] Preferably, when determining whether the cause of the abnormality is human operation error, it includes: Based on historical non-abnormal flight data of the same type of UAV in the same operation scenario, extract operator operation instruction timing, physiological response and task adaptation features, build a three-dimensional operator-model-task operation baseline model, and generate an operation deviation threshold matrix for different task types; Intercept the operation instruction sequence of the abnormal event, and align the timing with the three-dimensional operation baseline model through dynamic time warping algorithm to calculate the operation deviation index; Retrieve the fault precursor parameters in the device health profile before the fault, and calculate the correlation between the operation deviation index and the device fault precursor parameters through the Pearson correlation coefficient; Input the environmental parameters at the time of the abnormality into the environmental influence recognition model trained by the decision tree classifier, output the environmental interference score to judge the environmental influence weight; The operation deviation index, the equipment state verification result and the environmental interference score are input into an operation failure identification model trained by a random forest classifier, and an operation failure probability score is output.
[0016] A multi-source unmanned aerial vehicle flight record analysis system based on a UTM cloud end comprises: A multi-source flight control interfacing module is configured to establish data interfacing with flight control systems of at least two different manufacturers of unmanned aerial vehicles, analyze and collect flight record data including flight trajectories, operation instructions and equipment state parameters, and perform real-time uploading; A cloud end storage module is configured to receive and centrally store the flight record data uploaded in real time by the multi-source flight control interfacing module; An intelligent analysis module is configured to perform multi-dimensional analysis on the flight record data in the cloud end storage module, including: real-time monitoring of flight states to identify abnormal flight events; when an abnormality is detected, historical and real-time flight record data are retrieved, and whether the cause is human operation failure is determined through data correlation analysis; In the identification process, external environmental parameters are introduced as correction factors to normalize the basic indicators and composite indicators, and the abnormality determination threshold is dynamically adjusted through a Bayesian optimization algorithm. In the determination process, an operation failure probability score is output based on a three-dimensional operation baseline model and an operation failure identification model trained by a random forest classifier.
[0017] Compared with the prior art, the present application has the following advantages: (1) Dynamically adapt to multi-manufacturer protocol updates Through the cooperation of the dynamic feature library updating unit and the cloud end, the latest features of each manufacturer's communication protocol are synchronized in real time, and the initial handshake data packet is matched in multiple dimensions and the confidence is calculated, so that the accuracy of flight control manufacturer identification is greatly improved, solving the problem of access failure caused by the lag of traditional static rule library. For low-confidence devices, fuzzy matching and self-learning mechanism are started, and after the administrator confirms, the feature library is updated, gradually covering niche brands and customized devices, and the device compatibility is improved.
[0018] (2) Standardized conversion compatible with multiple versions of protocols Based on the firmware version, the protocol release time and the success rate of analysis, the priority is dynamically allocated, the patch package is automatically loaded for the old version of the protocol, and the abandoned field is marked, the private field conversion accuracy is high through the pre-defined standardized field set and the unit conversion formula. The converted data is checked for integrity and logic, and the field repair mechanism is used, so that the data anomaly repair success rate is significantly improved, ensuring the uniformity of data formats across manufacturers.
[0019] (3) Hierarchical storage and intelligent compression to reduce costs Data is divided into hot, warm and cold layers according to real-time and importance. The storage amount of trajectory data is greatly reduced after being compressed by the Douglas-Peucker algorithm. The device state data is compressed by the LZ77 algorithm to achieve a compression ratio of 1:3~1:5, and the overall storage cost is significantly reduced. Cold data is automatically migrated to object storage, and hot data is retained in the SSD cluster, with an access delay controlled within 10ms.
[0020] (4) Multiple encryption and fault tolerance mechanism to improve data security The transmission layer uses a TLS 1.3 encrypted channel, and sensitive fields are independently encrypted by AES-256-GCM. The key is dynamically rotated daily, and the risk of data leakage is significantly reduced. 3 copies + erasure code hybrid backup combined with breakpoint resume, the data recovery time of single point failure is shortened to within 5 minutes, and the retransmission efficiency is greatly improved after network interruption.
[0021] (5) Environment-adaptive anomaly judgment model External parameters such as wind speed and electromagnetic interference are introduced as correction factors, and the threshold is dynamically adjusted by the Bayesian optimization algorithm to shorten the response time of abnormal flight event identification. Combined with the model of aircraft type-task baseline, the accuracy of abnormal judgment in complex scenarios is high, and the misjudgment rate is significantly reduced compared with the traditional static threshold method.
[0022] (6) Three-dimensional operation baseline to realize quantitative evaluation of operation errors Based on the operator-aircraft-task three-dimensional model and the random forest classifier, the operation error probability score is output, combined with the device failure precursor parameters and environmental interference weight, the accuracy of human operation error attribution is significantly improved, providing data support for flight accident responsibility division and operation training.
[0023] The application will be further described in detail below in combination with the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS
[0024] Figure 1 is a multi-source unmanned aerial vehicle flight record analysis method step diagram of an embodiment of the application; Figure 2 is a real-time identification flowchart of the type of the manufacturer to which the embodiment of the application belongs; Figure 3 is a multi-dimensional data preprocessing and transmission optimization flowchart of the embodiment of the application. DETAILED DESCRIPTION
[0025] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.
[0026] It should be understood, of course, that the drawings are designed for the purpose of illustration only and not as a definition of the limits of the application.
[0027] The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the application or its application or uses.
[0028] Techniques, methods, and apparatus known to those of ordinary skill in the relevant art can not be discussed in detail herein. However, where appropriate, such techniques, methods, and apparatus should be considered as being part of the specification.
[0029] Embodiment 1, see Figure 1 The application provides a multi-source unmanned aerial vehicle flight record analysis method as shown in Figure 1 The application provides a multi-source unmanned aerial vehicle flight record analysis method based on a UTM cloud, comprising the following steps: S101. Data docking is established between a multi-source flight control docking module and flight control systems of at least two unmanned aerial vehicles of different manufacturers, flight record data generated by the flight control systems of the unmanned aerial vehicles are analyzed and collected, and the flight record data at least includes flight trajectories, operation instructions and device state parameters; S102. The flight record data is uploaded to a cloud storage module in real time for centralized storage; S103. The stored flight record data is analyzed by an intelligent analysis module in multiple dimensions, including: Real-time monitoring of unmanned aerial vehicle flight states and identification of abnormal flight events; When an abnormal flight event is detected, historical flight record data and real-time flight record data of the corresponding unmanned aerial vehicle are called, and whether the cause of the abnormal flight event is human operation error is determined through data correlation analysis.
[0030] Background description: With the rapid development of unmanned aerial vehicle technology, its application scenarios have been expanded to aerial photography, cleaning, logistics, surveying and mapping, power inspection and other fields. Due to differences in technical standards, communication protocols and data formats, unmanned aerial vehicle flight control systems of different manufacturers form a "data island" problem. Specifically, it is shown that: Protocol incompatibility: private communication protocols are used in flight control systems of various manufacturers, which leads to difficulties in cross-manufacturer data docking, and traditional single protocol analysis solutions cannot meet the multi-source data collection requirements; Data format confusion: there are significant differences in field definition, unit conversion (such as height units "meters" and "feet") and data frame structure of flight record data (such as trajectory, device parameters and operation instructions), which directly affects the centralized storage and subsequent analysis of data; Data quality varies: Raw flight data often contains sensor noise, transmission errors or outliers (such as voltage jumps, positioning drift), if not pre-processed directly uploaded, will lead to subsequent intelligent analysis module (such as abnormal event identification) misjudgment; Protocol dynamic update lag: Manufacturers will update the flight control system firmware from time to time, add protocol fields or encryption algorithms, and traditional static protocol parsing rules are difficult to adapt, resulting in data collection interruption or parsing failure. Based on this: In the above step S101, when analyzing and collecting the flight record data generated by each unmanned aerial vehicle flight control system, including: Through the adaptive protocol conversion unit built in the multi-source flight control docking module, the communication protocol characteristics of each manufacturer's flight control system are pre-stored to form a communication protocol characteristic library; Real-time identification of the type of unmanned aerial vehicle flight control system connected, and automatically load the corresponding protocol analysis rule to perform standardized conversion of data format; Among them, the original flight record data is filtered before data transmission, and the abnormal value filtering is realized by comparing the preset equipment state parameter threshold range, so as to ensure that the data uploaded to the cloud storage module meets the input standard of the subsequent intelligent analysis module.
[0031] In the embodiments of the present application, it needs to be further explained that the communication protocol characteristics pre-stored in the communication protocol characteristic library of the present embodiment specifically include: Basic protocol identification: protocol type (such as TCP / UDP), device model code, firmware version number, version number mapping relationship and other metadata for identity recognition; Data frame structure characteristics: start / end flag of data frame, field length, check bit algorithm (such as CRC32) and field order; Private field mapping relationship: the corresponding relationship between the fields defined by each manufacturer (such as "lat / lng" "GPS coordinates") and the standardized fields (such as "real-time longitude and latitude"); Encryption and security characteristics: encryption algorithm identification (such as AES-256), key exchange protocol version and data signature rule; Abnormal value distribution law: the normal fluctuation range and typical noise characteristics of each equipment state parameter (such as battery voltage, motor temperature).
[0032] These characteristics are stored in a structured database as a "dictionary" for subsequent identification of manufacturer type and data format analysis.
[0033] Before data transmission, this embodiment filters out abnormal values (such as voltage fluctuations and positioning drift) in the raw data by comparing it with preset device status parameter threshold ranges (such as normal battery voltage range and GPS signal strength threshold). This prevents noisy or erroneous data from entering the cloud and reduces the risk of misjudgment in subsequent analysis. For example, if the battery voltage of a drone suddenly exceeds the threshold range, the system will filter out this abnormal value to ensure that the uploaded data reflects the true device status.
[0034] Background Description: Due to differences in manufacturer technical standards and dynamic protocol updates, existing UAV flight control systems suffer from problems such as low recognition accuracy, poor compatibility, and high maintenance costs when connecting devices from different manufacturers. Specifically: (1) Static protocol feature library cannot adapt to dynamic update requirements Traditional flight control interface solutions rely on predefined static protocol rules, while drone manufacturers periodically update flight control firmware (such as adding protocol fields, upgrading encryption algorithms, or adjusting data frame structures). For example, after a manufacturer releases firmware version V2.0, it adds a "battery cycle count" field and modifies the encryption method for positioning data. Since the static rules do not include these new features, data parsing may fail or fields may be lost, requiring manual intervention to update the rules and causing data acquisition interruptions.
[0035] (2) The accuracy of manufacturer type identification is insufficient, and the mismatch rate is high. Traditional identification methods rely on matching based on a single feature (such as the device model code), making them susceptible to firmware version differences or customization modifications. For example, different series of drones from the same manufacturer may use similar model codes but have different protocol details, resulting in a mismatch rate exceeding 15%. If the flight control system has been modified by a third party (such as by changing the firmware), it may be completely unrecognizable, requiring manual configuration of protocol rules for each drone, which is extremely inefficient.
[0036] (3) Lack of effective processing mechanisms in low confidence scenarios When connecting to flight control systems of unknown model or with ambiguous protocol characteristics (such as equipment from niche manufacturers or older models), traditional solutions, lacking self-learning capabilities, can only refuse access or adopt generic parsing rules, resulting in a data loss rate of over 30%. For example, a surveying company using a customized flight control system had a protocol field length that deviated from the standard value by 20%, making it impossible for static rules to parse, requiring on-site debugging by technicians, which severely impacted operational efficiency.
[0037] (4) The maintenance of the feature library depends on manual labor and has a long iteration cycle. The updating of protocol features by various manufacturers requires manual collection, verification, and manual entry into the system, resulting in a significant lag in feature database updates compared to actual protocol changes. Based on this: See Figure 2The real-time identification flowchart of the type of the manufacturer to which the unmanned aerial vehicle flight control system belongs, in one possible embodiment, comprises the following steps: Through the dynamic feature library updating unit built in the multi-source flight control docking module, the latest feature data of the communication protocol of the flight control system of each manufacturer is obtained from the cloud regularly, and the locally pre-stored communication protocol feature library is updated in real time. In the identification process, first, the initial handshake data packet of the access flight control system is extracted, the protocol type identifier, device model code and firmware version number are analyzed, and multi-dimensional matching is performed with the updated communication protocol feature library, and the matching confidence of each manufacturer type is calculated. When the highest matching confidence is higher than the preset threshold, the type of the manufacturer to which it belongs is directly determined and the corresponding protocol analysis rule is loaded. When the highest matching confidence is lower than the preset threshold, the fuzzy matching and self-learning mechanism is started: Record the communication feature parameters of the flight control system, and generate a to-be-identified queue. Through the intelligent analysis module, an artificial confirmation request is sent to the system administrator, and the real-time communication log segment of the flight control system is synchronously pushed. After receiving the correct manufacturer type feedback from the administrator, the communication feature parameters of the flight control system are automatically added to the protocol feature library, and are marked as high-priority identification items, realizing self-iterative optimization of the feature library.
[0038] In the embodiments of the present application, it is further explained that the latest feature data of the communication protocol of the flight control system of each manufacturer is obtained from the cloud regularly, including: new protocol fields (such as manufacturer-defined state parameters), encryption algorithm identifier (such as AES-256 version), version number mapping relationship (firmware version and protocol version corresponding rule).
[0039] These data are synchronized to the local communication protocol feature library in real time, ensuring that the library contains the latest manufacturer protocol features, solving the problem that traditional static rules cannot adapt to protocol dynamic updates.
[0040] When the unmanned aerial vehicle flight control system is accessed, the system first extracts its initial handshake data packet, analyzes the key information: protocol type identifier (such as TCP port number), device model code (manufacturer-defined code), firmware version number (such as V2.3.1). The analysis information is matched with the updated communication protocol feature library in multiple dimensions, and the matching confidence of each manufacturer type (the matching degree of the protocol type, device model, firmware version, etc. features) is calculated. High confidence scenario: if the highest matching confidence is greater than or equal to the preset threshold (such as 90%), the manufacturer type is directly determined, and the corresponding protocol analysis rule (such as field mapping relationship, data format conversion logic) is automatically loaded.
[0041] When the highest matching confidence is less than a preset threshold (e.g., 90%), a fuzzy matching and self-learning mechanism is started (e.g., when an unknown model, protocol features are ambiguous, or a third-party modified device is encountered): The communication characteristic parameters of the flight control system are recorded, including: data frame structure (field order, length), abnormal value distribution rule (e.g., sensor noise characteristics), private field definition (manufacturer-defined field), and a to-be-identified queue is generated.
[0042] The intelligent analysis module sends an artificial confirmation request to the administrator and synchronously pushes real-time communication log segments (e.g., data frame examples, abnormal fields).
[0043] After receiving the correct manufacturer type feedback from the administrator, the communication characteristic parameters of the flight control system are automatically added to the protocol feature library and marked as a high-priority identification item (preferentially matched next time the same type of device is accessed), realizing self-iterative optimization of the feature library.
[0044] This embodiment realizes a closed-loop mechanism of "dynamic updating - intelligent matching - self-learning optimization" through cloud collaborative updating, real-time response to manufacturer protocol upgrades (e.g., addition of new fields, update of encryption algorithms), and avoidance of data collection interruption. Multi-dimensional feature matching and confidence calculation reduce the risk of single feature mis-matching (e.g., traditional misjudgment caused by reliance on device model alone). Artificial feedback in low-confidence scenarios and feature library iteration gradually cover niche manufacturers or old models, improving system compatibility. This embodiment realizes a closed-loop mechanism of "dynamic updating - intelligent matching - self-learning optimization", solving the problems of low recognition accuracy and poor compatibility when accessing cross-manufacturer unmanned aerial vehicle flight control systems, and laying a foundation for subsequent data standardization analysis and flight record analysis.
[0045] BACKGROUND DESCRIPTION: With the diversification of unmanned aerial vehicle application scenarios (e.g., aerial photography, logistics, surveying and mapping, etc.), different manufacturers' flight control systems form "protocol barriers" and "data islands" due to differences in technical standards, leading to multiple challenges in the standardization of multi-source flight record data analysis, specifically: (1) Insufficient multi-version protocol compatibility, static analysis rules lagging behind Unmanned aerial vehicle manufacturers will update the flight control firmware from time to time (e.g., add new protocol fields, upgrade encryption algorithms, or adjust data frame structure), and traditional analysis schemes rely on static predefined rules, which cannot dynamically adapt to protocol changes. For example: A manufacturer's V2.0 version firmware adds a "battery cycle count" field and modifies the positioning data encryption method, and static rules fail to include this feature, resulting in failed data analysis; Old devices (e.g., V1.0 version protocol that has stopped being maintained) lack a downward compatibility mechanism, with a data field loss rate as high as 30%, which cannot meet the needs of centralized storage and analysis.
[0046] (2) Field semantics are confusing and data formats are not unified Private field definitions, unit conversions, and data structures vary significantly between manufacturers, making data standardization difficult. Confusing field names: Private fields such as "lat / lng", "GPS coordinates", and "location information" all represent geographic locations, but the mapping relationships are not clear. Units are not unified: Height units are mixed with "meters" and "feet", and voltage units are different between "mV" and "V", directly affecting cross-device data comparison. Lack of verification mechanism: Raw data often contains sensor noise (such as voltage jumps) or logical contradictions (such as "flight speed" and "horizontal acceleration" physical relationship anomalies), and unverified error data can lead to misjudgment in subsequent analysis modules.
[0047] (3) Low efficiency of error handling, relying on manual intervention Traditional systems lack automated repair capabilities for parsing exceptions (such as format errors, data overflow): When the error rate of a single batch of data exceeds 5%, manual intervention is required to adjust the rules, resulting in interruptions in data collection; Error characteristics cannot be fed back to the rule library, similar problems repeatedly occur, and the success rate of parsing is consistently below 90%, affecting data quality stability.
[0048] (4) Conflict between encrypted data security and decryption mechanism Sensitive data (such as precise geographic location, operator ID) needs to be encrypted for transmission, but different manufacturers use different encryption algorithms (such as AES-256, RSA) and key exchange protocols: The traditional decryption method relies on fixed keys and cannot adapt to manufacturers' dynamic updates of encryption algorithm identifiers, resulting in a decryption failure rate of over 20%; There is a risk of tampering during data transmission, and there is a lack of security mechanisms such as hash verification, making it difficult to ensure data integrity.
[0049] (5) Protocol dynamic update and rule iteration disconnection After manufacturers update their protocols (such as adding new fields, adjusting encryption algorithms), traditional static rule libraries require manual collection, verification, and entry of new features, with a long iteration cycle.
[0050] In summary, existing technologies have poor compatibility, low standardization, delayed error handling, insufficient security, and iteration disconnection in multi-source flight control protocol parsing. Based on this: In one possible embodiment, the corresponding protocol parsing rules are automatically loaded for data format standardization conversion, including: Dynamic rule priority and multi-version compatibility mechanism: Establish a protocol rule priority queue: Based on the firmware version number of the flight control system, the protocol release time, and the historical parsing success rate, assign dynamic priorities to multiple versions of protocol rules from the same manufacturer. Supports backward compatibility parsing: When the flight control system is detected to be using an old version protocol (such as the V1.0 version that the manufacturer has stopped maintaining), the basic parsing rules and historical patch packages are automatically loaded, and obsolete fields are marked as compatibility reserved fields to ensure the integrity of data conversion for old equipment. Semantic field mapping and standardized validation: Construct a multi-level semantic mapping library: predefine a standardized field set and store the correspondence between each manufacturer's private fields and standard fields, as well as unit conversion formulas; Perform standardized validation: The transformed data must pass field integrity validation and logical consistency validation. If the validation fails, a field repair mechanism will be triggered. Real-time error feedback analysis and self-optimization: Deploy a parsing error monitoring unit: Real-time statistics on field parsing failure rate (such as format errors, data overflow) and semantic conflict count (such as contradictory data in the same field). When the error rate of a single batch of data exceeds 5%, the current rule loading will be automatically paused and a backup rule will be started. Closed-loop optimization process: Upload erroneous data fragments and original protocol frames to the cloud-based intelligent analysis module; generate rule correction suggestions based on historical successful cases (such as adding field mapping relationships and adjusting data type verification range); after administrator confirmation, update the local protocol parsing rules and multi-level semantic mapping library, and mark the rule as a high-priority verification item, so that the optimization effect is verified first when the same type of device is connected next time, realizing rule self-iteration; Encrypted data decryption and secure conversion: For encrypted flight record data (such as data packets containing sensitive geographical locations), based on the real-time updated encryption algorithm identifier and version number mapping relationship, a temporary decryption parameter is requested from the cloud key management module (using asymmetric encryption transmission). The decrypted data needs to undergo secondary verification: comparing the hash values of the data packets before and after decryption to ensure that the data has not been tampered with. At the same time, abnormal data packets that fail to decrypt are isolated and stored, triggering a manual review process to prevent erroneous data from entering subsequent analysis stages.
[0051] In this embodiment of the invention, it is necessary to further explain that the "dynamic rule priority and multi-version compatibility mechanism" enables adaptive parsing of different manufacturers' and different version protocols, ensuring data compatibility between new and old devices. Specifically, rule versions with a parsing success rate >95% and a release date within the past year are prioritized for loading, ensuring high reliability and timeliness. For older protocols, for example, although the "GPS coordinates" field in an older device protocol may be obsolete in the new version, it can still be parsed into the standard "real-time latitude and longitude" field through a patch package.
[0052] "Semantic field mapping and standardized validation" unifies the proprietary data formats of different manufacturers, ensuring data consistency and availability across devices. Predefined common fields (such as "latitude and longitude," "flight altitude," and "remaining battery power") serve as a unified data standard, storing the correspondence between each manufacturer's proprietary fields and the standard fields (e.g., "lat / lng," "GPS coordinates," and "location information" are all mapped to the standard field "real-time latitude and longitude"). Built-in unit conversion formulas are included (e.g., converting "mV" to "V," retaining two decimal places). It also ensures that required standard fields (such as latitude and longitude and battery voltage) are complete, avoiding the omission of critical data, and verifies the rationality of physical relationships between data (e.g., the matching of "flight speed" and "horizontal acceleration"). If validation fails, missing fields are supplemented through historical data interpolation (e.g., if altitude data is missing at a certain moment, it is supplemented by fitting the trajectory 10 seconds before and after).
[0053] "Real-time parsing error feedback and self-optimization" dynamically monitors parsing anomalies and improves rule adaptation capabilities through closed-loop optimization.
[0054] "Encrypted data decryption and secure conversion" ensures the secure transmission and integrity of sensitive data (such as geographic location and operator ID).
[0055] This embodiment solves the compatibility problem of multiple version protocols through dynamic rule adaptation, achieves cross-manufacturer data unification through semantic mapping, improves parsing robustness through error self-optimization, and ensures data privacy and integrity through secure decryption. Ultimately, it provides a standardized and high-quality data foundation for the centralized storage and intelligent analysis of UAV flight records (such as abnormal event identification and operational error judgment).
[0056] In step S102 above, when the flight record data is uploaded to the cloud storage module for centralized storage in real time, it includes: multi-dimensional data preprocessing and transmission optimization, distributed storage and fault tolerance mechanism, real-time quality monitoring and dynamic adjustment, and edge-cloud collaborative storage optimization.
[0057] Background Description: Drone flight log data is characterized by its diverse types (continuous time-series data such as flight trajectories and device status parameters, and discrete event data such as operation commands), high real-time requirements (abnormal events require second-level response), and security sensitivity (containing precise geographical location, operator ID, and other privacy information). Traditional data preprocessing and transmission schemes have the following shortcomings: Data compression lacks specificity: using general compression algorithms (such as ZIP) to process trajectory data results in insufficient compression ratio (usually only 1:2) and loss of accuracy at key inflection points (error > 1 meter); excessive compression of discrete operation commands can easily lead to the loss of key information such as mode switching events, affecting the accuracy of subsequent operation error analysis.
[0058] The transmission security mechanism is weak: it relies only on transport layer encryption (such as TLS 1.2), and sensitive fields in the data layer are not independently protected. Moreover, the encryption key is fixed for a long time (such as rotating it every six months), which poses a risk of mass decryption after theft. There is a lack of data integrity verification, and malicious tampering of data can enter the storage system through transmission verification.
[0059] Inefficient data block management: The lack of a dynamic block partitioning and verification mechanism means that entire blocks must be retransmitted when transmission errors occur (e.g., a 10MB data block needs to be retransmitted due to a 1-byte error), resulting in wasted bandwidth and increased latency. Therefore: See Figure 3 The flowchart of multi-dimensional data preprocessing and transmission optimization, in one possible embodiment, includes: Adaptive Segmentation and Compression: Dynamic segmentation is performed based on the type characteristics of flight record data. Trajectory data blocks are compressed using trajectory points based on the Douglas-Peucker algorithm, while device status data blocks are compressed using the LZ77 dictionary compression algorithm. The compression ratio is dynamically adjusted to the range of 1:3 to 1:5. No compression is performed on operation command data blocks. A multi-dimensional checksum is added to the header of each data block. Transmission encryption and dynamic key management: A two-layer encryption mechanism is adopted. The transport layer uses the TLS 1.3 protocol to encrypt the transmission channel, and the data layer uses the AES-256-GCM algorithm to independently encrypt sensitive fields. The encryption key is dynamically generated and rotated periodically through the cloud key management module. The key transmission uses ECC-384 asymmetric encryption to ensure security.
[0060] In this embodiment of the invention, it is necessary to further explain that the flight record data is segmented differently based on its type characteristics: Continuous time-series data (flight trajectory, equipment status parameters) is dynamically divided according to time windows (e.g., 1 second / block) or data volume thresholds (e.g., 1 MB / block) to ensure strong temporal correlation within data blocks, facilitating subsequent compression and analysis. Discrete event data (operation commands) is independently segmented according to the event triggering order (e.g., one block for a single command) to avoid mixing with continuous data and causing parsing confusion. The core objective is to adapt the segmentation granularity to the data characteristics, providing an optimized basis for the differentiated compression algorithm while reducing the retransmission cost in case of transmission errors (only retransmitting corrupted blocks).
[0061] This embodiment selects an adaptation algorithm based on the redundancy characteristics of different data types: Flight trajectory data: The Douglas-Peucker algorithm was used to simplify and compress the trajectory points.
[0062] Principle: By recursively retaining key inflection points in the trajectory that deviate from the fitted straight line by more than a threshold (≤ 0.5 meters), and eliminating redundant intermediate points, the data volume is reduced (compression ratio up to 1:5~1:10) while ensuring positioning accuracy.
[0063] Example: A straight flight trajectory originally contains 1000 positioning points, after compression by the algorithm, only the start point, end point and 2 inflection points are retained, reducing the data volume by 90%.
[0064] Device state parameters: Using LZ77 dictionary compression algorithm, taking advantage of the time series repetition characteristics of device parameters (such as battery voltage, motor temperature), recording the data sequence (dictionary) that has appeared through sliding window, subsequent repeated sequence is replaced by "distance + length" pointer, compression ratio is dynamically adjusted to 1:3~1:5. Its advantages are: for periodic fluctuation data (such as motor temperature gradient), the compression efficiency is higher than that of general algorithm (such as ZIP), and the decompression speed is fast, suitable for real-time transmission scene.
[0065] Operation instruction data: not compressed, because: the instruction is a discrete event (such as mode switching, rod quantity input), the data volume is small (single instruction < 100 bytes), the compression benefit is low and the key operation details (such as instruction interval timing) may be lost.
[0066] This embodiment adds block sequence number, timestamp, data type code, compression algorithm version and other multi-dimensional verification identifiers to the header of each data block to ensure that the cloud storage module accurately identifies the data block type, decompression algorithm and integrity. If the verification fails during transmission, the damaged block can be quickly located and retransmission is triggered.
[0067] This embodiment uses a "transmission layer encryption + data layer encryption" in-depth defense mechanism: Transmission layer encryption: use TLS 1.3 protocol to encrypt the transmission channel.
[0068] Principle: Through the handshake protocol to negotiate the session key, all data in the transmission link is encrypted (symmetric encryption algorithm such as AES-GCM), to prevent man-in-the-middle eavesdropping or tampering with the transmission content.
[0069] Advantages: Compared with TLS 1.2, the handshake delay is reduced by 50%, supports 0-RTT fast reconnection, and adapts to the high dynamic network environment of unmanned aerial vehicles.
[0070] Data layer encryption: sensitive fields (such as precise geographic location, operator ID) are independently encrypted using AES-256-GCM algorithm.
[0071] Principle: An independent encryption key is generated for sensitive fields in a single data, and after encryption, it is transmitted together with non-sensitive fields. The cloud needs to decrypt the key to parse sensitive information.
[0072] Necessity: Avoid the full leakage of sensitive data after the transmission layer encryption is cracked, and realize "field-level" fine protection.
[0073] The key dynamic management mechanism of the embodiment includes key generation and rotation and key transmission security; Among them, the key generation and rotation: the cloud key management module dynamically generates an encryption key based on the device unique identifier, timestamp and random number, and the rotation period is ≤24 hours, ensuring that the key life cycle is controllable. Example: Each UAV generates a new AES-256 key every day, which is automatically invalidated the next day, so even if the key is leaked, it will only affect the data of a single day.
[0074] Key transmission security: ECC-384 asymmetric encryption algorithm is used to transmit temporary decryption parameters. Principle: The UAV side encrypts the key with the cloud public key, and the cloud side decrypts it with the private key, avoiding key theft during transmission (ECC-384 is more secure than RSA-2048, and the key length is shorter, suitable for resource-constrained devices).
[0075] In summary, the embodiment greatly reduces the data volume through the "block-compression-check" closed loop, while ensuring trajectory accuracy and command integrity, combined with TLS 1.3 and AES-256-GCM, and high-frequency key rotation, significantly reducing the risk of data leakage.
[0076] Background description: With the expansion of UAV application scale, GB-level flight data is generated daily per flight, and the traditional centralized storage architecture faces prominent contradictions in cost and performance, specifically: Storage level missing: all data is stored in high-performance disk arrays, resulting in an access delay of >50ms for hot data (such as real-time state parameters), which cannot meet the real-time response requirements of abnormal events; Cold data (such as data archived three months ago) accounts for more than 70%, but still occupies high-cost storage resources, and storage cost accounts for >60%.
[0077] Fault tolerance and disaster recovery capability is insufficient: using local dual backup, regional failure (such as power failure) can cause data loss, and the recovery time is >24 hours; After network interruption, historical data needs to be retransmitted in full, and the time-consuming of single flight data retransmission is >30 minutes, which seriously affects the continuity of the operation. Based on this: In one possible embodiment, the distributed storage and fault tolerance mechanism includes: Classify flight record data according to real-time and importance: Hot data layer is stored in cloud SSD cluster, access delay ≤10ms; Warm data layer is stored in hybrid storage array; Through the data life cycle management strategy, the cold data layer is automatically migrated to low-cost object storage, reducing storage costs; Cross-region redundant backup and breakpoint resume: A hybrid backup mechanism of 3 replicas + erasure coding is adopted. The primary replica is stored on a local cloud node, while the other 2 replicas and erasure coding shards are stored on a disaster recovery node at a distance of ≥500km to ensure data recoverability in the event of a single node failure. The data block upload progress is recorded in real time during transmission. When the network is interrupted or a node fails, the breakpoint resume mechanism is triggered. By comparing the checksum of the uploaded blocks in the local cache with the cloud index, only the missing or damaged data blocks are retransmitted, reducing the consumption of invalid traffic.
[0078] In this embodiment of the invention, it is necessary to further explain that, based on the real-time access requirements and differences in the importance of the data, the flight record data is divided into three levels, matched with storage media of different performance, to balance storage costs and access efficiency: (1) Hot data layer Data range: Real-time status parameters (such as battery voltage and GPS signal strength) for the most recent hour, and key data triggered by abnormal events (such as abnormal flight trajectory segments).
[0079] Storage medium: Cloud SSD cluster (solid-state drives).
[0080] Principle: Utilizing the low latency characteristics of SSDs (access latency ≤10ms), it ensures rapid response in high-frequency access scenarios such as abnormal event identification and real-time monitoring, meeting the needs of second-level analysis.
[0081] (2) Warm data layer Data range: historical trajectories and operation logs (such as lever input sequences and mode switching records) within 7 days.
[0082] Storage medium: Hybrid storage array (combination of SSD and HDD).
[0083] Principle: Balancing performance and cost, supporting medium-frequency historical data queries (such as manual operation review and routine flight analysis), and avoiding the hot data layer storage resources being occupied by non-real-time data.
[0084] (3) Cold data layer Data range: Archived data exceeding 30 days (such as long-term job statistics and non-critical historical records).
[0085] Storage medium: Low-cost object storage (such as S3-compatible storage).
[0086] Principle: By implementing data lifecycle management strategies (such as automatic compression and archiving of data that has not been accessed for 3 months), the storage cost of cold data is reduced, while supporting on-demand retrieval (such as regulatory compliance audits).
[0087] This embodiment realizes the dual improvement of data risk resistance and transmission stability through multi-copy backup, erasure code technology and breakpoint resume mechanism: (1) 3 copies + erasure code hybrid backup Backup strategy: The main copy is stored in the local cloud node to ensure low-latency access; 2 copies and erasure code shards are stored in off-site disaster recovery nodes with a distance of ≥500km.
[0088] Principle: Copy mechanism: Multi-copy storage (3 copies) can resist single node failure (such as hard disk damage, node downtime), ensuring data integrity; Erasure code technology: The data is divided into multiple shards and a check block is generated, even if some shards are lost, the complete data can still be reconstructed through the check block, saving more than 50% storage space compared to full copy backup.
[0089] Objective: To achieve "local fast access + off-site disaster recovery" and ensure data recoverability in extreme scenarios (such as regional power failure, natural disasters).
[0090] (2) Breakpoint resume mechanism Transmission optimization: Real-time record data block upload progress, save the checksum (such as hash value) of the uploaded block through local cache; After network interruption or node failure recovery, compare the local checksum with the cloud index and only retransmit the missing or damaged data block.
[0091] Principle: Avoid full data retransmission, reduce invalid traffic consumption (such as network fluctuations only need to supplement the transmission of faulty blocks), improve transmission efficiency in weak network environments (such as mountainous areas, urban canyon scenarios).
[0092] In summary, this embodiment realizes "efficient access to hot data - low-cost archiving of cold data" resource optimization through hierarchical storage strategy, and guarantees data reliability and continuity in extreme failure and network fluctuations through cross-regional hybrid backup and breakpoint resume, providing underlying technical support for centralized management, real-time analysis and long-term archiving of unmanned aerial vehicle flight records.
[0093] Background description: Unmanned aerial vehicle operating environment network fluctuation is significant (such as bandwidth from 100Mbps to 10Mbps in mountainous areas, urban canyon scenarios), traditional fixed transmission parameters (such as MTU, congestion algorithm) cause the following problems: Key data transmission delay: Non-critical data (such as historical trajectory) occupies resources when bandwidth drops, causing device state parameters (such as battery voltage, GPS signal strength) upload delay > 5 seconds, affecting real-time identification of abnormal events.
[0094] Poor link adaptability: packet loss rate > 5% without triggering congestion algorithm switching (e.g. from CUBIC to BBR), data transmission success rate < 80%; lack of real-time link quality monitoring (e.g. bandwidth fluctuation sampling frequency < 1 Hz), unable to dynamically adjust transmission strategy. Based on this: In one possible embodiment, real-time quality monitoring and dynamic adjustment, including: Transmission link quality closed-loop optimization: deploy network state awareness unit to monitor bandwidth fluctuation, packet loss rate and delay of the transmission link in real time at a preset sampling frequency, and when link quality degradation is detected: Automatic enablement of data priority queue, priority transmission of device state parameters and abnormal event identification data, and temporary suspension of transmission of non-critical historical trajectory data; Dynamic adjustment of MTU (maximum transmission unit) value and TCP congestion control algorithm to ensure critical data transmission success rate; Storage integrity verification and self-recovery: the cloud storage module performs hierarchical verification after receiving data: Data block level verification (compare transmission checksum); Field level logical verification (e.g. latitude and longitude range reasonableness, speed and acceleration physical relationship verification); When verification fails, automatically trigger data repair process and feed back abnormal data features to intelligent analysis module to optimize subsequent data preprocessing rules.
[0095] In the embodiments of the present application, it is further explained that the embodiments monitor network status in real time, dynamically adjust transmission strategy, guarantee priority transmission and transmission success rate of critical data (such as device state parameters and abnormal event identification), and adapt to complex and variable network environment (such as bandwidth fluctuation in mountainous areas and urban valleys) in unmanned aerial vehicle operation scenarios.
[0096] (1) Real-time link state monitoring Monitoring indicators: deploy network state awareness unit to collect bandwidth fluctuation, packet loss rate and delay of the link in real time at a sampling frequency of 10 Hz (once every 0.1 second).
[0097] Triggering conditions: when link quality degradation (such as packet loss rate > 3% or delay > 200 ms) is detected, automatically start optimization process.
[0098] (2) Dynamic transmission strategy optimization Data priority scheduling: enable data priority queue, prioritize transmission of device state parameters (such as battery voltage, GPS signal strength) and abnormal event identification data (such as abnormal flight trajectory segment), and temporarily suspend transmission of non-critical historical trajectory data to ensure timeliness of real-time monitoring and abnormal response.
[0099] Network parameter adaptive adjustment: MTU dynamic adjustment: Adjust the maximum transmission unit (MTU) in the range of 1200-1500 bytes according to the link bandwidth and packet loss rate (reduce MTU to reduce fragmentation packet loss when bandwidth is low, increase MTU to improve transmission efficiency when bandwidth is high).
[0100] TCP congestion control algorithm switching: Adaptive switching between BBR algorithm (suitable for high-bandwidth low-latency links) and CUBIC algorithm (suitable for links with high packet loss rate), balancing throughput and stability.
[0101] Principle: Through the closed-loop mechanism of "real-time monitoring-threshold triggering-dynamic optimization", realize the on-demand allocation of network resources, and ensure the reliable transmission of key data in poor network environment.
[0102] This embodiment ensures the integrity of stored data through a multi-level verification mechanism, and automatically repairs when abnormalities are found, reducing manual intervention, while feeding back data anomaly features to optimize subsequent preprocessing rules.
[0103] (1) Three-layer verification mechanism Data block level verification: After receiving the data block, compare the checksum generated during transmission (such as CRC32, MD5) to ensure that the data block has not been tampered with or damaged during transmission.
[0104] Field-level logical verification: Reasonable verification of standardized fields, such as: Physical parameter range verification: Latitude and longitude are within a reasonable geographic range (such as latitude -90°~90°), battery voltage meets device model threshold; Physical relationship verification: Matching of flight speed and horizontal acceleration (such as sudden change in speed accompanied by abnormal acceleration), to avoid logical contradiction data entering the analysis process.
[0105] (2) Self-healing and feedback optimization Automatic repair process: If the verification fails, the system triggers the data repair mechanism: For missing fields (such as missing altitude data at a certain time), use historical data interpolation algorithms (such as linear interpolation, moving average) to supplement; For damaged data blocks (such as checksum mismatch), request the terminal to retransmit the damaged block to avoid full data retransmission.
[0106] Abnormal feature feedback: Feedback the data features of failed verification (such as abnormal value distribution, field missing rules) to the intelligent analysis module, to optimize subsequent data preprocessing rules (such as adjusting abnormal value filtering threshold, enhancing the verification logic of specific fields), forming a closed loop of "verification-repair-optimization".
[0107] Principle: Through the multi-layer verification of "data block level-field level", the physical integrity and logical rationality of the stored data are ensured, combined with automatic repair and rule optimization, the cost of manual intervention is reduced, and the adaptability of the system to complex data scenarios is improved.
[0108] In summary, the embodiment ensures real-time transmission of critical data in weak network environment through 10Hz high-frequency monitoring and dynamic priority scheduling; transmission link optimization focuses on "process reliability", storage verification focuses on "result integrity", forming end-to-end data quality protection; abnormal feature feedback and preprocessing rule optimization make the system continuously evolve with data characteristics, reducing long-term maintenance costs. The above mechanism effectively solves the problems of poor network fluctuation adaptability, high data error rate, and frequent manual intervention in the "transmission-storage" link of unmanned aerial vehicle flight record data, providing high-quality data input for subsequent intelligent analysis (such as abnormal event identification, operation error judgment).
[0109] Background Description: In the unmanned aerial vehicle operation intensive area (such as logistics park, surveying and mapping construction site), a large number of equipment (>100 / square kilometer) upload data at the same time, the traditional "full upload" mode causes the following problems: Core network bandwidth bottleneck: full upload of raw data causes core network bandwidth occupancy rate > 90%, data queuing delay > 20 seconds, which cannot meet the real-time monitoring demand.
[0110] High cost of cold data storage: cold data (such as archived data not accessed within 30 days) accounts for more than 60%, but is still stored in high-performance media, resulting in waste of storage resources; lack of dynamic migration mechanism, cold data needs to be loaded from low-speed storage when accessed, query delay > 30 seconds, affecting the efficiency of post-fault analysis. Based on this: In one possible embodiment, edge-cloud collaborative storage optimization includes: Edge node pre-caching and incremental synchronization: deploying edge computing nodes in unmanned aerial vehicle operation intensive areas for local temporary caching of flight record data, uploading only pre-processed incremental data to the cloud, reducing core network bandwidth occupancy; Intelligent cold and hot data migration trigger mechanism: by analyzing historical data access frequency, a dynamic migration threshold is established, when a certain type of data in the cold data layer is accessed continuously for multiple times, it is automatically promoted to the warm data layer, shortening the subsequent access delay.
[0111] In the embodiment of the present application, it needs to be further explained that the edge node pre-caching and incremental synchronization mechanism provided by the embodiment aims to reduce core network bandwidth occupancy and improve data transmission efficiency in unmanned aerial vehicle operation intensive areas (such as logistics park, surveying and mapping construction site) through local edge nodes. The working principle is as follows: (1) Edge node deployment and local caching In areas with high drone operation density (e.g., 1 edge computing node per square kilometer), locally cache flight record data temporarily (cache duration ≤ 2 hours). The cached content includes original flight trajectory, device status parameters, and operation instructions.
[0112] Cache function: Avoid core network congestion caused by simultaneous full data upload from a large number of drones. Implement a hierarchical processing of "near-end temporary storage - preprocessing - on-demand upload" of data.
[0113] (2) Incremental data extraction and upload Edge nodes preprocess cached data and only upload incremental data to the cloud, including: Deviation from historical baseline: e.g., abnormal segments of flight trajectory deviating from historical average path, values of device status parameters (e.g., battery voltage, motor temperature) exceeding normal fluctuation range; Abnormal event segments: e.g., sensor failures, operation instruction mutations, and other critical event data.
[0114] Non-incremental data filtering: For regular data consistent with historical baseline (e.g., trajectory during stable cruising phase, normal device status parameters), only short-term cache at edge nodes and then automatically clean up, without the need to upload to the cloud.
[0115] (3) Bandwidth optimization effect Through the "local caching + incremental upload" mode, the core network bandwidth occupancy is greatly reduced, avoiding full data queuing delay and ensuring the real-time nature of abnormal event data.
[0116] The intelligent cold and hot data migration trigger mechanism provided in this embodiment dynamically adjusts the storage hierarchy based on data access frequency, balancing storage cost and access efficiency.
[0117] By analyzing historical data access frequency, establish migration trigger rules: For example: In the past 30 days, the access frequency of abnormal event data is 8.3 times that of ordinary data. Mark such high-frequency access data as "potential hot data"; Set dynamic migration threshold: When a certain type of data in the cold data layer is accessed continuously for 3 times, it is determined that the "access demand has increased", and migration is automatically triggered: Migrate cold data that meets the trigger conditions (e.g., recent frequently queried historical abnormal event records) from low-cost object storage to warm data layer (hybrid storage array); Effect: When accessing this type of data subsequently, there is no need to load from low-speed storage, and the query delay is shortened from > 30 seconds to < 5 seconds, improving the efficiency of post-fault analysis, compliance audit, and other scenarios.
[0118] Warm data layer: data that has not been accessed for 30 consecutive days is automatically downgraded to the cold data layer, freeing up high-performance storage resources and reducing overall storage costs (cold data storage costs are only 1 / 5 of the hot data layer).
[0119] The embodiment reduces the bandwidth occupation of the core network by "localization staging + incremental filtering", solves the data upload congestion problem in dense areas, dynamically adjusts the storage level based on the access frequency, realizes the resource optimization of "efficient access of hot data-low cost archiving of cold data", and balances performance and cost; The closed loop of "edge preprocessing-cloud hierarchical storage-dynamic on-demand migration" is formed by the cooperation of the two, supporting efficient management and analysis of unmanned aerial vehicle flight record data.
[0120] BACKGROUND DESCRIPTION: As the application scenarios of unmanned aerial vehicles become diversified (such as aerial photography, logistics, surveying and mapping, power inspection, etc.), the complexity of flight environment and the risk of device operation have significantly increased. The traditional abnormal flight event identification scheme is difficult to meet the high-precision and high-adaptability monitoring requirements due to its dependence on single-dimensional indicators, fixed thresholds and static models. Therefore: In the above step S103, when identifying an abnormal flight event, the following steps are included: Multi-dimensional abnormal feature index system construction and dynamic weight distribution: Core index layer: based on the real-time collected flight trajectory, device state parameters and operation instruction sequence, a basic index feature library containing multiple basic indexes is established; Derivative index layer: composite indexes such as "operation-response delay deviation rate" and "energy consumption abnormality coefficient" are generated through feature engineering, and a composite index feature library containing multiple composite indexes is established; Environment correlation layer: external environment parameters are introduced as correction factors to normalize the basic indexes and composite indexes; Adaptive threshold dynamic adjustment mechanism: Baseline model construction: based on historical abnormal-free flight data, a model-task baseline model is trained to generate abnormal judgment thresholds for each index under different operation scenarios; Real-time threshold correction: combined with the current flight stage, remaining power and device health, the abnormal judgment threshold of each index is dynamically adjusted through the Bayesian optimization algorithm, and the threshold adjustment range is limited within ±30% of the baseline value; When the normalized basic index and / or composite index exceeds the corresponding real-time abnormal judgment threshold, an abnormal flight event is determined.
[0121] In the embodiment of the present application, it needs to be further explained that the embodiment captures flight abnormal features through three layers of indexes, and combines environmental factors for correction, ensuring the comprehensiveness and accuracy of the indexes.
[0122] (1) Core index layer: basic features are directly collected Flight trajectory indicators include latitude and longitude drift (deviation between actual position and planned path), altitude fluctuation range (standard deviation of altitude change per unit time), and horizontal positioning accuracy (range of GPS signal deviation), which directly reflect the spatial motion status of the UAV.
[0123] Equipment status parameters include battery voltage drop rate (voltage change per unit time), motor temperature gradient (temperature change rate over time), and GPS signal-to-noise ratio (signal strength to noise ratio), which characterize the health of the equipment hardware.
[0124] Operation command sequence: such as the frequency of sudden changes in joystick input (the number of instantaneous changes in the joystick's amplitude) and mode switching interval (the time interval between automatic / manual mode switching), reflects the stability of the operator's control behavior. Purpose: To establish a basic indicator feature library as the "raw data source" for anomaly identification.
[0125] (2) Derived index layer: composite feature engineering enhancement By combining core metrics into composite metrics through feature engineering, the discriminative power of abnormal patterns can be improved. Operation-response delay deviation rate: The calculation formula is (difference between actual executed command and theoretical response time) / standard threshold, which measures the timeliness of the UAV's response to operation commands.
[0126] Energy consumption anomaly coefficient: The ratio of energy consumption during the current flight phase (such as energy consumption per unit distance during cruise phase) to the historical average for the same phase, identifying abnormal energy consumption patterns (such as a sudden increase in energy consumption caused by battery failure).
[0127] Purpose: To establish a composite indicator feature library, make up for the limitations of single indicators, and capture more complex anomaly patterns.
[0128] (3) Environmental correlation layer: normalization correction of external factors External environmental parameters (wind speed, atmospheric pressure, electromagnetic interference intensity) are introduced as correction factors to normalize the core and composite indicators: For example, the altitude fluctuation of drones will naturally increase in strong winds, and the judgment criteria for altitude fluctuation need to be corrected by wind speed parameters to avoid misjudgments caused by environmental interference.
[0129] Function: To eliminate the interference of environmental factors on indicators and ensure the comparability of indicators in different scenarios.
[0130] This embodiment combines a baseline model with real-time correction to dynamically adjust the anomaly detection threshold, avoiding the limitations of a fixed threshold.
[0131] (1) Baseline model construction: Scenario-based threshold benchmark Training data: Train model based on historical normal flight data (>1000 flights) for each model-task pair.
[0132] Scenario-based threshold: Generate abnormality judgment threshold for different operation scenarios (aerial photography, logistics, surveying and mapping, etc.). For example: Surveying and mapping tasks require high positioning accuracy, and the horizontal positioning accuracy baseline is set to ±0.5 meters; Aerial photography tasks allow larger errors, with a baseline of ±1.2 meters.
[0133] Effect: Provide initial judgment criteria for each index to adapt to the characteristics of different tasks.
[0134] (2) Real-time threshold correction: dynamically adapt to the current state Adjust the threshold according to the real-time flight state, and realize dynamic optimization through Bayesian optimization algorithm: Correction factors: Flight phase: The take-off / landing phase allows greater height fluctuations (threshold relaxation), while the cruising phase is tightened; Remaining power: When the power is less than 30%, the battery voltage drop rate threshold is tightened (more sensitive to abnormal power consumption); Device health: Adjust the key component parameter threshold according to historical fault records (e.g. if the motor has experienced a thermal fault, the temperature gradient threshold is lowered).
[0135] Adjustment range: The threshold adjustment is limited within ±30% of the baseline value to avoid excessive correction leading to misjudgment.
[0136] Effect: Make the threshold adapt to the real-time working condition, improve the flexibility and accuracy of abnormality identification.
[0137] In summary, the abnormal flight event identification of the embodiment realizes precise monitoring of the flight state of the unmanned aerial vehicle through the three-layer mechanism of "multi-dimensional index system capturing features - environmental correction eliminating interference - dynamic threshold adapting to scenarios".
[0138] Background description: In the cause determination of abnormal flight events of unmanned aerial vehicles, the existing technology lacks scenario-based operation baseline, has insufficient time sequence dynamic alignment capability, lacks non-human factor filtering mechanism, lacks quantitative evaluation, and has weak model generalization, making it difficult to accurately determine whether the cause of the abnormal flight event is human operation error. Based on this: In the above step S103, when determining whether the cause of the abnormal flight event is human operation error through data correlation analysis, it includes: Operator-model-task three-dimensional operation baseline modeling and deviation quantification: Baseline model construction: Based on historical data of at least 100 flights of the same type of unmanned aerial vehicle in the same operating scenario without abnormal flight, the timing characteristics of the operator's operation instructions, physiological response characteristics and task adaptation characteristics are extracted, a three-dimensional operation baseline model including the operator-aircraft-task is constructed, and an operation deviation threshold matrix under different task types is generated; Real-time deviation calculation: intercept the operation instruction sequence of the abnormal flight event, perform timing alignment with the three-dimensional operation baseline model through the dynamic time warping (DTW) algorithm, calculate the operation deviation index, and the operation deviation index includes: instruction amplitude deviation rate, operation rhythm disorder degree and mode switching error rate; Multi-dimensional correlation filtering and non-human factor exclusion: Device state correlation verification: retrieve the device failure precursor parameters in the device health record data of the abnormal flight event during the same period, calculate the correlation between the operation deviation index and the device failure precursor parameters through the Pearson correlation coefficient; Environmental interference correlation verification: taking the non-human abnormal flight event cases caused by historical environmental interference as training samples, training an environmental influence recognition model through a decision tree classifier; The environmental parameters (wind speed mutation, electromagnetic interference intensity) at the time of the abnormal flight event are input into the trained environmental influence recognition model, and the environmental interference score is output, which is used to judge the influence weight of the environmental parameters on the flight state; Operation failure mode recognition and probability grading: Feature engineering and model training: extracting operation deviation index, device state verification result and environmental interference score from historical human operation failure cases, training an operation failure recognition model using a random forest classifier; Real-time classification and probability output: input the operation deviation index, device state verification result and environmental interference score of the real-time abnormal flight event into the trained operation failure recognition model, output the operation failure probability score (0-100 points), and when the operation failure probability score is greater than or equal to 60 points, it is determined that the cause contains human operation failure factors.
[0139] What needs to be further explained in the embodiment of the application is that the embodiment recognizes potential operation abnormalities by constructing a "normal operation benchmark" and comparing the deviation degree of actual operation and the benchmark.
[0140] (1) Baseline model construction Data basis: based on historical data of at least 100 flights of the same type of unmanned aerial vehicle in the same operating scenario (such as aerial photography, logistics), ensure that the sample size is statistically representative.
[0141] Feature extraction: three types of key features are extracted from the data to form a three-dimensional baseline: Operation instruction timing characteristics: stick input amplitude, instruction interval length, mode switching logic, etc. Physiological response characteristics: emergency operation response time, continuous operation stability (such as joystick operation smoothness); Task adaptation characteristics: waypoint planning efficiency (such as the deviation rate of actual waypoints from planned waypoints), obstacle avoidance operation accuracy.
[0142] Threshold matrix generation: generate operation deviation threshold matrix for different task types (such as mapping, logistics), for example: Mapping tasks require high precision, and stick input fluctuation threshold is ±8%; Logistics tasks allow greater operational flexibility, with a threshold of ±15%.
[0143] (2) Real-time deviation calculation Data interception: intercept the operation instruction sequence from 30 seconds before the abnormal event to the end of the event, focusing on the operation behavior in the key time period.
[0144] Timing alignment: align the real-time instruction sequence with the three-dimensional baseline model through the dynamic time warping (DTW) algorithm, solving the timing difference problem under different operation rhythms (such as slower operation of beginners and faster operation of experienced operators).
[0145] Deviation index calculation: Instruction amplitude deviation rate: (actual input / baseline mean-1), measures the degree of deviation of operation intensity from the normal range; Operation rhythm disorder degree: (instruction interval standard deviation / baseline standard deviation), reflects the stability of operation interval; Mode switching error rate: (number of invalid switching times / total switching times), identifies operation errors such as accidental mode switching button pressing.
[0146] This embodiment excludes non-human factors such as equipment failure and environmental interference through statistical analysis and machine learning model, focusing on operator responsibility.
[0147] Device state verification: retrieve the pre-fault parameters in the device health profile (such as motor temperature gradient, IMU drift), calculate the correlation between operation deviation indicators and device parameters through Pearson correlation coefficient, when the correlation coefficient is <0.3, exclude equipment failure as the dominant factor.
[0148] Environmental interference verification: train decision tree classifier with historical environmental interference cases (such as strong wind, electromagnetic interference) to generate environmental impact recognition model, when environmental interference score <40 points, determine as potential human operation dominant event.
[0149] This embodiment trains machine learning model based on historical human operation error cases, integrates multi-dimensional feature output quantitative probability, and realizes objective judgment.
[0150] (1) Model training Feature input: Operation deviation indicators in historical human operation error cases (such as instruction amplitude deviation rate), device state verification results (correlation coefficient), environmental interference score.
[0151] Algorithm selection: Random forest classifier is used to improve classification accuracy in complex scenarios (such as distinguishing between "rod input overloading" and "mode switching error" and other different error types) through multi-decision tree ensemble learning.
[0152] (2) Real-time judgment Input data: Operation deviation indicators of real-time abnormal events, device state verification results, and environmental interference scores.
[0153] Probability output: The model outputs an operation error probability score (0-100 points). When the score is greater than or equal to 60 points, it is determined that the abnormal event is caused by human operation error factors.
[0154] This embodiment realizes accurate identification of human operation errors through a closed-loop process of "baseline modeling - deviation quantification - interference exclusion - probability determination", combining statistical analysis and machine learning. Its core innovation lies in: Three-dimensional baseline: Considering the interactive influence of operators, machine types, and tasks, avoiding the limitations of single baseline; Multi-factor filtering: Through data correlation analysis, the device and environmental interference are stripped, improving the specificity of the judgment; Probability grading: The machine learning model outputs quantitative results based on historical cases, taking into account accuracy and flexibility.
[0155] Embodiment 2: The present application provides a multi-source unmanned aerial vehicle flight record analysis system based on UTM cloud, comprising: a multi-source flight control docking module, a cloud storage module and an intelligent analysis module.
[0156] The multi-source flight control docking module is configured to establish a communication connection with at least two different manufacturers' unmanned aerial vehicle flight control systems, analyze and collect flight record data generated by each unmanned aerial vehicle flight control system, and upload in real time. The flight record data at least includes flight trajectory, operation instruction and device state parameter.
[0157] The cloud storage module is configured to receive and centrally store the flight record data uploaded in real time by the multi-source flight control docking module.
[0158] The intelligent analysis module is configured to perform multi-dimensional analysis on the flight record data in the cloud storage module, including: Real-time monitoring of unmanned aerial vehicle flight state, identifying abnormal flight events; When an abnormal flight event is detected, historical flight record data and real-time flight record data of the corresponding unmanned aerial vehicle are called, and whether the cause of the abnormal flight event is a human operation error is determined through data correlation analysis.
[0159] In a possible embodiment, the multi-source flight control docking module is built-in with an adaptive protocol conversion unit for pre-storing communication protocol characteristics of flight control systems of various manufacturers to form a communication protocol characteristic library.
[0160] In a possible embodiment, the multi-source flight control docking module is built-in with a dynamic characteristic library updating unit for periodically acquiring the latest characteristic data of the communication protocols of the flight control systems of various manufacturers from the cloud to update the locally pre-stored communication protocol characteristic library in real time.
[0161] In a possible embodiment, the multi-source flight control docking module is deployed with an error analysis monitoring unit for real-time statistics of field parsing failure rate and semantic conflict times.
[0162] Finally: the above only for the preferred embodiments of the present application, and not for limiting the present application, any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.
Claims
1. A method for analyzing multi-source unmanned aerial vehicle flight records based on a UTM cloud, characterized in that, The method comprises the following steps: Through the multi-source flight control docking module, data docking is established with the flight control systems of at least two different manufacturers, flight record data including flight trajectory, operation instructions and equipment state parameters are analyzed and collected; The flight record data is uploaded to the cloud storage module in real time for centralized storage; Through the intelligent analysis module, multi-dimensional analysis is performed on the flight record data, including: real-time monitoring of flight status to identify abnormal flight events; when an abnormality is detected, historical and real-time flight record data are called, and whether the cause is human operation error is determined through data correlation analysis; In the identification process, external environmental parameters are introduced as correction factors to normalize the basic indicators and composite indicators, and the abnormality determination threshold is dynamically adjusted through the Bayesian optimization algorithm; The determination process is based on a three-dimensional operation baseline model and an operation error identification model trained by a random forest classifier to output an operation error probability score.
2. The method of claim 1, wherein, When analyzing and collecting flight record data, the following steps are included: Through the adaptive protocol conversion unit built in the multi-source flight control docking module, the manufacturer type of the flight control system is identified based on the pre-stored communication protocol feature library, the protocol analysis rules are automatically loaded for data standardization conversion; before data transmission, abnormal values are filtered through a pre-set threshold range to ensure that the data meet the analysis standards.
3. The method of claim 2, wherein, When identifying the manufacturer type of the flight control system, the following steps are included: The dynamic feature library update unit built in the multi-source flight control docking module periodically obtains the latest feature data of each manufacturer's communication protocol from the cloud to update the local communication protocol feature library; The initial handshake data packet of the accessed flight control is extracted, the protocol type identifier, device model code and firmware version number are analyzed, and the feature library is multi-dimensionally matched to calculate the confidence; When the confidence is higher than the threshold, the manufacturer type is directly determined; when the confidence is lower than the threshold, a fuzzy matching and self-learning mechanism is started, the communication feature parameters are recorded and a to-be-identified queue is generated, and the feature library is updated after being confirmed by the administrator.
4. The method of claim 2, wherein, When performing data standardization conversion, the following steps are included: According to the firmware version number of the flight control system, the protocol release time and the historical analysis success rate, dynamic priority is assigned to the multi-version protocol rules of the same manufacturer; When an old version protocol is detected, the basic analysis rules and historical patches are automatically loaded, and the abandoned fields are marked as compatible reserved fields; A set of predefined standardized fields is stored, and the correspondence between the private fields and the standard fields of each manufacturer and the unit conversion formula are stored; The converted data needs to pass through field integrity verification and logical consistency verification, and the field repair mechanism is triggered when the verification fails.
5. The method of claim 4, wherein, When performing data standardization conversion, the following steps are included: The error monitoring unit is deployed to real-time statistics the field analysis failure rate and the semantic conflict frequency, and when the error rate of a single batch of data exceeds the threshold, the current rule loading is automatically suspended and the standby rule is started; The error data segment and the original protocol frame are uploaded to the cloud intelligent analysis module, rule correction suggestions are generated based on historical successful cases, and the local protocol analysis rules and multi-level semantic mapping library are updated after being confirmed by the administrator; For encrypted flight record data, temporary decryption parameters are requested from the cloud key management module based on the real-time updated encryption algorithm identifier and version number mapping relationship, and the decrypted data needs to pass through secondary verification.
6. The method of claim 1, wherein, When uploading to the cloud storage module for centralized storage in real time, the following steps are included: The data is dynamically divided into blocks according to the data type characteristics, the trajectory type data block is compressed by the Douglas-Peucker algorithm, the device state type data block is compressed by the LZ77 dictionary compression with a compression ratio of 1:3~1:5, and the operation instruction type data block is not compressed; a multi-dimensional verification identifier is added to the data block header; The transmission layer uses a TLS 1.3 encrypted channel, and the sensitive fields in the data layer are independently encrypted using the AES-256-GCM algorithm. The encryption key is dynamically generated by the cloud key management module and is periodically rotated; The data is divided into hot, warm, and cold data layers according to real-time and importance. Hot data is stored in a cloud SSD cluster, warm data is stored in a hybrid storage array, and cold data is automatically migrated to object storage. A 3-replica + erasure code hybrid backup is used, and the upload progress is recorded in real time. When the network is interrupted or the node fails, the breakpoint resume is triggered. By comparing the local cache checksum and the cloud index, only the missing or damaged data block is retransmitted.
7. The method of claim 6, wherein, When the real-time upload is stored in the cloud storage module, it also includes: A network state sensing unit is deployed to monitor the upload link bandwidth fluctuation, packet loss rate, and delay in real time at a preset sampling frequency. When the link quality deteriorates, the data priority queue is automatically enabled, and the MTU value and TCP congestion control algorithm are dynamically adjusted; The cloud storage module performs layered verification after receiving the data. If the verification fails, the data repair process is automatically triggered, and the abnormal data characteristics are fed back to optimize the subsequent preprocessing rules; In the area where the unmanned aerial vehicle is operating intensively, an edge computing node is deployed to temporarily cache the flight record data locally and only upload the preprocessed incremental data. By analyzing the historical data access frequency, a dynamic migration threshold is established to automatically migrate the data at different levels.
8. The method of claim 1, wherein, When an abnormal flight event is identified, it includes: A basic index feature library is established based on the real-time collected flight trajectory, device state parameters, and operation instruction sequence; A composite index is generated through feature engineering, and a composite index feature library is established; External environmental parameters are introduced as correction factors to normalize the basic index and composite index; A model-task baseline model is trained based on historical abnormal flight data to generate abnormal judgment thresholds for each index in different operation scenarios. Combined with the current flight stage, remaining power, and device health, the abnormal judgment threshold of each index is dynamically adjusted through the Bayesian optimization algorithm. When the normalized basic index and / or composite index exceeds the corresponding real-time abnormal judgment threshold, an abnormal flight event is determined.
9. The method according to claim 1 or 8, characterized in that, When determining whether the abnormal cause is human operation error, it includes: Based on the historical abnormal flight data of the same type of unmanned aerial vehicle in the same operation scenario, the operator operation instruction sequence, physiological response, and task adaptation characteristics are extracted to build a three-dimensional operator-model-task operation baseline model and generate an operation deviation threshold matrix for different task types. The abnormal event operation instruction sequence is intercepted, and the operation deviation index is calculated by aligning the time sequence with the three-dimensional operation baseline model through dynamic time warping algorithm. The correlation between the operation deviation index and the device failure precursor parameters is calculated through the Pearson correlation coefficient by calling the failure precursor parameters in the same period device health profile. An environmental impact recognition model trained by inputting environmental parameters at the time of abnormal occurrence into a decision tree classifier outputs an environmental disturbance score to determine the environmental impact weight; An operation failure recognition model trained by inputting the operation deviation index, the equipment state verification result and the environmental disturbance score into a random forest classifier outputs an operation failure probability score. 10.A multi-source unmanned aerial vehicle flight record analysis system based on a UTM cloud, characterized in that, Comprise: Multi-source flight control docking module: configured to establish data docking with flight control systems of at least two different manufacturers, analyze and collect flight record data including flight trajectory, operation instruction and equipment state parameter, and perform real-time uploading; Cloud storage module: configured to receive and centrally store the flight record data uploaded in real time by the multi-source flight control docking module; Intelligent analysis module: configured to perform multi-dimensional analysis on the flight record data in the cloud storage module, including: real-time monitoring of flight state to identify abnormal flight events; when an abnormality is detected, historical and real-time flight record data are called to determine whether the cause is human operation failure through data correlation analysis; In the recognition process, external environmental parameters are introduced as correction factors to normalize the basic index and composite index, and the abnormality determination threshold is dynamically adjusted through the Bayesian optimization algorithm; The determination process is based on a three-dimensional operation baseline model and an operation failure recognition model trained by a random forest classifier to output an operation failure probability score.
Citation Information
Cited By
Heterogeneous link cloud native security communication system for low-altitude unmanned aerial vehicle
CN121619558A
