A Method and System for Accident Tracing by Power Line Inspection Drones Based on Multi-Source Data Fusion

Abstract

This invention discloses a method and system for tracing the source of accidents involving power line inspection drones based on multi-source data fusion. The method first synchronizes the drone and remote controller time, collects five types of multi-source data including flight status and user voice data, adds a unified millisecond-level timestamp, and writes it to a two-layer cache module. Then, based on the flight status data, it determines anomalies according to a dual threshold rule of warning and triggering, generates trigger signals, extracts multi-source data for key time periods using a differentiated dynamic window strategy, packages it into an encrypted evidence file, and uploads its hash value to a third-party evidence storage platform or blockchain for evidence storage. Finally, it decrypts the file to achieve synchronized playback of the multi-source data over a unified timeline, quantifies causal inference, calculates a comprehensive operational response score, and outputs a structured accident analysis report. Using this invention, accident scenarios can be reconstructed, improving the accuracy and objectivity of source tracing, providing quantitative and verifiable evidence for determining responsibility in power line inspection drone accidents, and solving industry pain points.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a method and system for tracing the source of accidents involving power line inspection drones based on multi-source data fusion. Background Technology

[0002] With the large-scale application of power inspection drones in scenarios such as power transmission lines, substations, and distribution substations, the frequency of flight accidents has increased significantly, and the determination of liability for accidents has become a prominent pain point in the power inspection industry and the field of drone application.

[0003] Existing technologies mainly rely on flight data recorded by the UAV flight control system, such as motor speed, attitude angle, altitude, voltage, flight speed, and flight path, for accident analysis. However, flight data can only reflect changes in the physical state of the UAV and cannot reconstruct the following key dimensions of information specific to the power inspection scenario before and after the accident: (1) the subjective intentions and verbal instructions of the inspection operator: such as the operator's verbal reaction after observing abnormal UAV prompts and line hazards, and the content of communication with ground command personnel and inspection colleagues; (2) The timing and rationality of operation instructions: such as the precise timing of joystick operation, flight mode switching, and gimbal adjustment; (3) Remote control screen prompts: such as key warning information such as motor stall, signal loss, and power saturation. Such information is directly related to the accident causes in the power inspection scenario; (4) Flight screen and inspection environment information: such as the visual presentation of obstacles and flight attitude.

[0004] In actual accident cases involving power line inspection drones, the lack or fragmentation of the aforementioned information has led to serious difficulties in determining liability. There is an urgent need for a new method for tracing the source of power line inspection drone accidents, capable of achieving full-dimensional data fusion in power line inspection scenarios. This would provide strong support for determining accident liability and ensure the safe and orderly conduct of power line inspection work. Summary of the Invention

[0005] In view of the technical problems mentioned in the background, the purpose of this invention is to provide a method and system for tracing the source of power inspection drone accidents based on multi-source data fusion.

[0006] To achieve the objectives of this invention, the technical solution provided by this invention is as follows: First aspect This invention provides a method for tracing the source of accidents using power line inspection drones based on multi-source data fusion, comprising the following steps: Step S1: Synchronize the time between the drone and the remote controller; Step S2: Collect multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data; the multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. Step S3: Continuously write the spatiotemporally aligned multi-source data into the cache module; Step S4: Based on the UAV flight status data, determine whether a trigger signal is generated according to the preset judgment rules; if a trigger signal is generated, proceed to step S5; Step S5: After receiving the trigger signal, extract multi-source data for key time periods from the cache module according to the dynamic window strategy; Step S6: Combine and package the extracted multi-source data for key time periods in chronological order to generate encrypted evidence files; Step S7: Upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate; Step S8: Decrypt the encrypted evidence file to obtain decrypted multi-source data; based on the decrypted multi-source data, perform synchronous playback and causal inference, and output an accident analysis report.

[0007] Second aspect The present invention also provides a power inspection drone accident tracing system based on multi-source data fusion, used to execute the power inspection drone accident tracing method based on multi-source data fusion, including the following units: time synchronization unit, data acquisition unit, writing unit, trigger signal generation unit, data extraction unit, evidence file generation unit, uploading unit, and analysis report output unit; The time synchronization unit is used to synchronize the time between the drone and the remote controller; The data acquisition unit is used to acquire multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data. The multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. The writing unit is used to continuously write spatiotemporally aligned multi-source data into the cache module; The trigger signal generation unit is used to determine whether to generate a trigger signal based on the UAV flight status data and a preset judgment rule; if a trigger signal is generated, the data extraction unit is executed. The data extraction unit is used to extract multi-source data for key time periods from the cache module according to a dynamic window strategy after receiving a trigger signal. The evidence file generation unit is used to associate and package the extracted multi-source data from key time periods in chronological order to generate encrypted evidence files; The uploading unit is used to upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate. The analysis report output unit is used to decrypt the encrypted evidence file to obtain decrypted multi-source data; based on the decrypted multi-source data, it performs synchronous playback and causal inference, and outputs an accident analysis report.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention breaks through the limitations of existing technologies that rely solely on UAV flight status data. It simultaneously collects five types of multi-source data: UAV flight status data, remote controller user voice data, operation command data, screen recording data, and flight image data. Furthermore, it adds a unified millisecond-level timestamp to all data to achieve precise spatiotemporal alignment. This invention fully reconstructs the entire scene before and after the accident, including the UAV's physical state, operator's operational intentions, environmental communication, screen warning prompts, and visual information of the flight environment. It overcomes the technical shortcomings of traditional traceability methods, such as single data dimensions and incomplete scene reconstruction.

[0009] In addition, this invention establishes dual quantitative judgment rules for warning thresholds and trigger thresholds for multiple types of UAV abnormal events. It standardizes and accurately detects six common accident anomalies such as motor stall and attitude angle exceeding limits, and generates trigger signals containing event type and trigger timestamp. For compound abnormal events, data is extracted based on the earliest trigger timestamp, which defines a precise time range and analysis object for subsequent accident tracing, thereby improving the accuracy of anomaly detection and the targeting of tracing.

[0010] Furthermore, the system enables synchronized playback of flight data curves, operation commands, user voice, flight footage, and remote control screen content on a unified timeline within the same user interface. Simultaneously, OCR recognition of screen prompts and differentiated color-coding of abnormal events, screen prompt events, operation command events, and user voice events on the timeline create a visually appealing unified timeline annotation chart. This makes the temporal relationships of various accident-related events readily apparent, reducing the difficulty of manual analysis. By establishing quantitative causal inference rules based on the temporal relationships between screen prompt events and abnormal events, and quantifying the strength of their causal association through confidence level values, the system clarifies causal determination conclusions at different time differences. This replaces traditional subjective manual accident cause analysis methods, providing standardized and quantifiable evidence for determining accident causality, and enhancing the objectivity and credibility of the tracing conclusions. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the process for tracing the source of power inspection drone accidents based on multi-source data fusion, provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that the acquisition of data and collection of information in this application are legal, compliant, or obtained with the consent of the subject of the data collection.

[0014] like Figure 1 As shown, this embodiment of the invention provides a method for tracing the source of accidents using power line inspection drones based on multi-source data fusion, including the following steps: Step S1: Synchronize the time between the drone and the remote controller; Specific implementation method: (1) GPS / BeiDou dual-mode satellite time synchronization is used as the main time source, and the time synchronization accuracy is better than 1ms; (2) When satellite signals are unavailable, automatically switch to Network Time Protocol (NTP) or Precision Time Protocol (PTP) for synchronization; (3) The time offset between the UAV and the remote controller is periodically exchanged through a heartbeat packet mechanism and calibrated every 100ms to ensure that the clock drift during long-term flight does not exceed 2ms; (4) All data frames are added with a 64-bit millisecond-level timestamp based on the synchronous clock during acquisition, with an accuracy of 1ms. The timestamp includes the device ID field, which facilitates the fusion of data from multiple devices.

[0015] Step S2: Collect multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data; It should be noted that the multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. The UAV flight status data is collected through the UAV flight control system at a sampling frequency of 100Hz, including motor speed, three-axis attitude angle (accuracy 0.01°), altitude (barometer + ultrasonic fusion, accuracy 0.1m), battery voltage (accuracy 0.01V), communication signal strength (RSSI, accuracy 1dBm), positioning data, etc. Among them, the remote control user voice data is collected through a microphone array with a sampling rate of 16kHz and a sampling accuracy of 16bit. It supports dual-microphone beamforming, effectively reduces noise, and automatically marks valid voice segments with voice activity detection (VAD). The remote control operation command data is collected through the joystick and button sensors, with a sampling frequency of 100Hz, a joystick position accuracy of 0.1%, a button status sampling frequency of 1000Hz, and support for channel-level timestamp marking. Among them, the remote control screen recording data is collected through the screen recording module, with a resolution of no less than 1280×720, a frame rate of 30fps, H.264 hard encoding compression, and supports real-time OCR text recognition and index creation. The drone flight footage is captured by an onboard FPV camera with a resolution of no less than 1920×1080, a frame rate of 30fps, H.265 encoding, and supports video frame-level timestamp embedding.

[0016] It should be noted that the technical solution of this invention integrates five types of multi-source data: flight status, user voice, operation commands, screen recording, and flight footage. Combined with a unified millisecond-level timestamp, it achieves precise spatiotemporal alignment, fully reconstructs the full-dimensional scenario of the accident, and solves the problems of single-dimensional data and incomplete scenario reconstruction in traditional source tracing.

[0017] Step S3: Continuously write the spatiotemporally aligned multi-source data into the cache module; It should be noted that the caching module includes a first-level cache area and a second-level cache area; the first-level cache area is a first-in-first-out (FIFO) circular buffer with a preset capacity to store the full multimodal data of the most recent 10 minutes, and adopts memory-mapped file technology with a write latency of less than 1ms; the second-level cache area is local flash memory, used to automatically persist multi-source data of key periods in the first-level cache area to the second-level cache area to form a backup when the storage of the first-level cache area is full during the current storage period (when the data in the first-level cache area during the previous storage period is about to be overwritten).

[0018] In addition, the storage duration of the first-level cache area can be dynamically adjusted and customized, and the storage duration can be automatically extended to the required duration when state fluctuations are detected.

[0019] Step S4: Based on the UAV flight status data, determine whether a trigger signal is generated according to the preset judgment rules; if a trigger signal is generated, proceed to step S5; The preset determination rule is to perform the following determination within each determination period: (1) If the deviation between the motor speed command value and the actual speed is >60% and the state lasts for >50ms, the motor stall is determined to have reached the warning threshold, and the storage time of the first-level cache area is extended; if the deviation between the motor speed command value and the actual speed is >80% and the state lasts for >100ms, the motor stall is determined to have reached the trigger threshold, and a trigger signal for the motor stall abnormal event is generated. (2) If the absolute value of the roll angle or pitch angle is >30°, it is determined that the attitude angle exceeds the warning threshold and the storage time of the first-level cache area is extended; if the absolute value of the roll angle or pitch angle is >45°, it is determined that the attitude angle exceeds the trigger threshold and a trigger signal for the attitude angle exceeding the abnormal event is generated. (3) If the altitude drop rate is >6m / s, it is determined that the altitude drop has reached the warning threshold, and the storage time of the first-level cache area is extended; if the altitude drop rate is >10m / s, it is determined that the altitude drop has reached the trigger threshold, and a trigger signal for the altitude drop abnormal event is generated. (4) If the voltage drops by more than 0.8V within 1 second, it is determined that the voltage drop has reached the warning threshold and the storage time of the first-level cache area is extended; if the voltage drops by more than 1.5V within 1 second, it is determined that the voltage drop trigger threshold is reached and a voltage drop abnormal event trigger signal is generated. (5) If the received signal strength indicator RSSI < -75dBm and the state lasts for >1s, it is determined that the communication signal is weak and reaches the warning threshold, and the storage time of the first-level buffer area is extended; if the received signal strength indicator RSSI < -85dBm and the state lasts for >1s, it is determined that the communication signal is weak and reaches the trigger threshold, and a trigger signal for the communication signal weak abnormal event is generated. (6) When the number of positioning satellites is less than 8 and the status lasts for more than 3 seconds, the positioning loss is determined to have reached the warning threshold, and the storage time of the first-level cache area is increased; when the number of positioning satellites is less than 6 and the status lasts for more than 3 seconds, the positioning loss is determined to have reached the trigger threshold, and a trigger signal for the positioning loss abnormal event is generated.

[0020] It should be noted that the prediction and judgment rules adopt a two-level mechanism of warning threshold and trigger threshold. The warning threshold is used to expand the cache retention range in advance, and the trigger threshold is used to generate a formal trigger signal. When multiple abnormal events occur in succession within 1 second, the system merges them into a compound abnormal event to avoid fragmentation of evidence files due to multiple triggers. The trigger signal includes the abnormal event type and trigger timestamp, which facilitates subsequent analysis.

[0021] Step S5: After receiving the trigger signal, extract multi-source data for key time periods from the cache module according to the dynamic window strategy; The extraction of multi-source data for key time periods according to the dynamic window strategy includes the following: (1) For voltage drop anomaly events, weak communication signal anomaly events, and location loss anomaly events, extract multi-source data from 30 seconds before to 60 seconds after the trigger timestamp; (2) For motor stalling abnormal events, extract multi-source data from 60 seconds before to 90 seconds after the trigger timestamp; (3) For attitude angle over-limit abnormal events and altitude drop abnormal events, extract multi-source data from 90 seconds before to 120 seconds after the trigger timestamp; (4) When there are more than two abnormal events that occur within the same judgment period, the multiple abnormal events are merged into a composite abnormal event, and the multi-source data from 120 seconds before to 120 seconds after the trigger timestamp of the first abnormal event is extracted.

[0022] In addition, at the boundaries of the extraction window, buffer data of 1 second before and after is automatically retained to prevent key evidence from being truncated; after extraction, the continuity of timestamps of the five types of data is verified. If there is a timestamp jump of more than 100ms, it is automatically marked and an attempt is made to fill in the missing data segment from the secondary cache area.

[0023] This invention achieves complete retention and accurate extraction of critical incident data by designing a dual-layer caching module and a differentiated dynamic window data extraction strategy, combined with dynamic adjustment of the caching duration for anomaly warnings, ensuring the integrity of the evidence chain. Furthermore, by establishing standardized detection rules with dual thresholds for six types of abnormal events, and using the earliest trigger timestamp as a benchmark for composite anomalies, it achieves precise incident triggering and source tracing, improving the accuracy of anomaly detection.

[0024] Step S6: Combine and package the extracted multi-source data for key time periods in chronological order to generate encrypted evidence files; Step S7: Upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate; Step S8: Decrypt the encrypted evidence file to obtain decrypted multi-source data; based on the decrypted multi-source data, perform synchronous playback and causal inference, and output an accident analysis report.

[0025] It should be noted that the synchronized playback refers to the simultaneous playback of the following content on the same user interface with a unified timeline: (1) Flight data curve panel, including a multi-source data curve overlay composed of altitude curve, attitude angle curve, motor speed curve, voltage curve and communication signal strength curve; (2) Operation instruction panel, including a two-dimensional vector diagram of the joystick position and a sequence diagram of button events; (3) User voice panel, including waveform display and automatic speech-to-text transcription, with voice synchronized with the timeline; (4) Flight video window, used for playing flight video; (5) Remote control screen video window, used to synchronously play screen recording content; This invention enables synchronized playback of multi-dimensional data over a unified timeline, combined with differentiated visual annotations of events, making accident timeline information readily available and reducing the difficulty of manual analysis.

[0026] During synchronized playback, the following operations are automatically performed to generate a unified timeline annotation diagram: (1) Recognize all system prompt text from the screen recording using OCR technology, extract the text content, text appearance time, and text disappearance time, and generate a list of text events; (2) Align the text events in the text event list with the flight video frames at the millisecond level according to the timestamp; (3) On the flight video window, all system prompt texts corresponding to the current moment are superimposed in real time in the form of semi-transparent subtitles, and the time of appearance of the texts is marked; (4) Mark the appearance time of all system prompt texts with red vertical lines on a unified timeline; (5) Automatically label the following key events on a unified timeline using different colors: Abnormal events include attitude angle exceeding limits, sudden altitude drop, sudden voltage drop, and motor stall, which are marked in red. Screen notification events: Includes the time and content of all system notification texts, highlighted in orange; Operation command events: including joystick zeroing, mode switching, and button operations, marked in blue; User voice events: Valid voice segments detected by voice activity detection are highlighted in green.

[0027] The rules for causal inference are as follows: (1) During synchronous playback, if the screen prompt event precedes the abnormal event by more than 500 ms, the confidence level that the screen prompt event is the cause of the abnormal event is greater than 85%; (2) During synchronous playback, if the screen prompt event precedes the abnormal event by less than 500ms, then the screen prompt event is the cause of the abnormal event with a confidence level of 85% > 50%. The two are highly correlated and need to be analyzed in conjunction with the comprehensive score of operation response. The calculation method for the comprehensive score of the operation response is as follows: S = St × Kp; Where S is the comprehensive score for operational response, St is the basic score for response timeliness, and Kp is the operational rationality coefficient. The response time base score St is calculated according to the following rules: from the time the system prompt text appears or the trigger timestamp to the time of the first joystick movement or button press extracted from the remote control operation command data, if the response time is <1 second, then St=100; if 1 second < response time < 3 seconds, then St=80; if the response time is >3 seconds or there is no response, then St=50. The operation rationality coefficient Kp is calculated according to the following rules: if the operation conforms to the recommended strategy, then Kp=1.0; if the operation does not conform to the recommended strategy or there is no effective operation, then Kp=0.5.

[0028] By constructing a dual evaluation system for response timeliness and operational rationality, and quantifying the comprehensive score of operational response through formulas, the gap in operator operation evaluation is filled, providing a quantitative basis for accident liability division.

[0029] (3) During synchronous playback, if an abnormal event precedes the screen notification event by more than 500 ms, the confidence level of the screen notification event as an abnormal event result is greater than 85%; (4) During synchronous playback, if only screen prompt events exist and no abnormal events occur, the confidence level that the screen prompt events are false alarms or have been handled in a timely manner is <30%; (5) During synchronous playback, if there are only abnormal events and no screen prompt events, the confidence level of the existence of undetected abnormal event causes is 50%.

[0030] This invention formulates quantitative causal inference rules based on temporal relationships, and quantifies the causal correlation strength of events by measuring confidence, replacing subjective analysis and improving the objectivity and credibility of accident cause determination.

[0031] The accident analysis report includes the following: (1) Basic information of abnormal events: power inspection drone ID, flight duration, trigger timestamp, and abnormal event type; (2) Explanation of the integrity of evidence: scope of multi-source data collection, consistency of timestamps, and encryption status of encrypted evidence files; (3) Key event sequence table: List all abnormal events, screen prompt events, and operation command events in chronological order; (4) Causal inference conclusions: the order of occurrence of screen prompt events and abnormal events, confidence level, and causal inference logic; (5) Operational response evaluation results: the basic score for response timeliness, the score for operational rationality coefficient, and the comprehensive score for operational response; (6) Visualized evidence charts: multi-source data curve overlay charts and unified time axis annotation charts for key periods; (7) Original evidence index: hash value of encrypted evidence file, evidence certificate of third-party evidence storage platform / blockchain network.

[0032] By automatically generating structured accident analysis reports and standardizing and integrating the results of the entire source tracing process, the system provides a clear and comprehensive reference for accident handling and liability determination, thereby improving the practicality of the source tracing results. At the same time, the overall solution is customized to suit the power inspection industry scenario, accurately addressing the industry pain point of difficulty in determining liability for drone accidents in this field, and promoting the standardization of industry applications.

[0033] In addition, the present invention also provides a power inspection drone accident tracing system based on multi-source data fusion, which is used to execute the power inspection drone accident tracing method based on multi-source data fusion, and includes the following units: time synchronization unit, data acquisition unit, writing unit, trigger signal generation unit, data extraction unit, evidence file generation unit, uploading unit, and analysis report output unit. The time synchronization unit is used to synchronize the time between the drone and the remote controller; The data acquisition unit is used to acquire multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data. The multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. The writing unit is used to continuously write spatiotemporally aligned multi-source data into the cache module; The trigger signal generation unit is used to determine whether to generate a trigger signal based on the UAV flight status data and a preset judgment rule; if a trigger signal is generated, the data extraction unit is executed. The data extraction unit is used to extract multi-source data for key time periods from the cache module according to a dynamic window strategy after receiving a trigger signal. The evidence file generation unit is used to associate and package the extracted multi-source data from key time periods in chronological order to generate encrypted evidence files; The uploading unit is used to upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate. The analysis report output unit is used to decrypt the encrypted evidence file to obtain decrypted multi-source data; based on the decrypted multi-source data, it performs synchronous playback and causal inference, and outputs an accident analysis report.

[0034] Finally, it should be noted that the above embodiments are merely illustrative and explanatory of the present invention, and are not intended to limit the present invention to the scope of the described embodiments. Furthermore, those skilled in the art will understand that the present invention is not limited to the above embodiments, and many more variations and modifications can be made based on the teachings of the present invention, all of which fall within the scope of protection claimed by the present invention.

Claims

1. A method for tracing the source of accidents involving power line inspection drones based on multi-source data fusion, characterized in that, Includes the following steps: Step S1: Synchronize the time between the drone and the remote controller; Step S2: Collect multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data; the multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. Step S3: Continuously write the spatiotemporally aligned multi-source data into the cache module; Step S4: Based on the UAV flight status data, determine whether to generate a trigger signal according to the preset judgment rules; If a trigger signal is generated, proceed to step S5; Step S5: After receiving the trigger signal, extract multi-source data for key time periods from the cache module according to the dynamic window strategy; Step S6: Combine the extracted multi-source data for key time periods into a chronological sequence and package them to generate encrypted evidence files; Step S7: Upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate; Step S8: Decrypt the encrypted evidence file to obtain decrypted multi-source data; Based on the decrypted multi-source data, synchronous playback and causal inference are performed to output an accident analysis report.

2. The method for tracing the source of power line inspection drone accidents based on multi-source data fusion as described in claim 1, characterized in that, In step S3, the cache module includes a first-level cache area and a second-level cache area; the first-level cache area is a first-in-first-out (FIFO) circular buffer; the second-level cache area is local flash memory, which is used to automatically persist multi-source data of key periods in the first-level cache area to the second-level cache area to form a backup when the storage of the first-level cache area is full within the current storage period.

3. The method for tracing the source of power line inspection drone accidents based on multi-source data fusion according to claim 2, characterized in that, In step S4, the preset determination rule is to perform the following determination within each determination period: (1) If the deviation between the motor speed command value and the actual speed is >60% and the state lasts for >50ms, the motor stall is determined to have reached the warning threshold, and the storage time of the first-level cache area is extended; if the deviation between the motor speed command value and the actual speed is >80% and the state lasts for >100ms, the motor stall is determined to have reached the trigger threshold, and a trigger signal for the motor stall abnormal event is generated. (2) If the absolute value of the roll angle or pitch angle is >30°, it is determined that the attitude angle exceeds the warning threshold and the storage time of the first-level cache area is extended; if the absolute value of the roll angle or pitch angle is >45°, it is determined that the attitude angle exceeds the trigger threshold and a trigger signal for the attitude angle exceeding the abnormal event is generated. (3) If the altitude drop rate is >6m / s, it is determined that the altitude drop has reached the warning threshold, and the storage time of the first-level cache area is extended; if the altitude drop rate is >10m / s, it is determined that the altitude drop has reached the trigger threshold, and a trigger signal for the altitude drop abnormal event is generated. (4) If the voltage drops by more than 0.8V within 1 second, it is determined that the voltage drop has reached the warning threshold and the storage time of the first-level cache area is extended; if the voltage drops by more than 1.5V within 1 second, it is determined that the voltage drop trigger threshold is reached and a voltage drop abnormal event trigger signal is generated. (5) If the received signal strength indicator RSSI < -75dBm and the state lasts for >1s, it is determined that the communication signal is weak and reaches the warning threshold, and the storage time of the first-level buffer area is extended; if the received signal strength indicator RSSI < -85dBm and the state lasts for >1s, it is determined that the communication signal is weak and reaches the trigger threshold, and a trigger signal for the communication signal weak abnormal event is generated. (6) When the number of positioning satellites is less than 8 and the status lasts for more than 3 seconds, the positioning loss is determined to have reached the warning threshold, and the storage time of the first-level cache area is increased; when the number of positioning satellites is less than 6 and the status lasts for more than 3 seconds, the positioning loss is determined to have reached the trigger threshold, and a trigger signal for the positioning loss abnormal event is generated.

4. The method for tracing the source of power line inspection drone accidents based on multi-source data fusion according to claim 3, characterized in that, The trigger signal includes the abnormal event type and the trigger timestamp.

5. The method for tracing the source of power line inspection drone accidents based on multi-source data fusion according to claim 4, characterized in that, In step S5, the extraction of multi-source data for key time periods according to the dynamic window strategy includes the following: (1) For voltage drop anomaly events, weak communication signal anomaly events, and location loss anomaly events, extract multi-source data from 30 seconds before to 60 seconds after the trigger timestamp; (2) For motor stalling abnormal events, extract multi-source data from 60 seconds before to 90 seconds after the trigger timestamp; (3) For attitude angle over-limit abnormal events and altitude drop abnormal events, extract multi-source data from 90 seconds before to 120 seconds after the trigger timestamp; (4) When there are more than two abnormal events that occur within the same judgment period, the multiple abnormal events are merged into a composite abnormal event, and the multi-source data from 120 seconds before to 120 seconds after the trigger timestamp of the first abnormal event is extracted.

6. The method for tracing the source of power line inspection drone accidents based on multi-source data fusion according to claim 5, characterized in that, In step S8, the synchronized playback refers to the synchronized playback of the following content on the same user interface with a unified timeline: (1) Flight data curve panel, including a multi-source data curve overlay composed of altitude curve, attitude angle curve, motor speed curve, voltage curve and communication signal strength curve; (2) Operation instruction panel, including a two-dimensional vector diagram of the joystick position and a sequence diagram of button events; (3) User voice panel, including waveform display and automatic speech-to-text transcription, with voice synchronized with the timeline; (4) Flight video window, used for playing flight video; (5) Remote control screen video window, used to synchronously play screen recording content; During synchronized playback, the following operations are automatically performed to generate a unified timeline annotation diagram: (1) Recognize all system prompt text from the screen recording using OCR technology, extract the text content, text appearance time, and text disappearance time, and generate a list of text events; (2) Align the text events in the text event list with the flight video frames at the millisecond level according to the timestamp; (3) On the flight video window, all system prompt texts corresponding to the current moment are superimposed in real time in the form of semi-transparent subtitles, and the time of appearance of the texts is marked; (4) Mark the appearance time of all system prompt texts with red vertical lines on a unified timeline; (5) Automatically label the following key events on a unified timeline using different colors: Abnormal events include attitude angle exceeding limits, sudden altitude drop, sudden voltage drop, and motor stall, which are marked in red. Screen notification events: Includes the time and content of all system notification texts, highlighted in orange; Operation command events: including joystick zeroing, mode switching, and button operations, marked in blue; User voice events: Valid voice segments detected by voice activity detection are highlighted in green.

7. A method for tracing the source of power line inspection drone accidents based on multi-source data fusion as described in claim 6, characterized in that, In step S8, the rules for causal inference are as follows: (1) During synchronous playback, if the screen prompt event precedes the abnormal event by more than 500 ms, the confidence level that the screen prompt event is the cause of the abnormal event is greater than 85%; (2) During synchronous playback, if the screen prompt event precedes the abnormal event by less than 500ms, then the screen prompt event is the cause of the abnormal event with a confidence level of 85% > 50%. The two are highly correlated and need to be analyzed in conjunction with the comprehensive score of operation response. (3) During synchronous playback, if an abnormal event precedes the screen notification event by more than 500 ms, the confidence level of the screen notification event as an abnormal event result is greater than 85%; (4) During synchronous playback, if only screen prompt events exist and no abnormal events occur, the confidence level that the screen prompt events are false alarms or have been handled in a timely manner is <30%; (5) During synchronous playback, if there are only abnormal events and no screen prompt events, the confidence level of the existence of undetected abnormal event causes is 50%.

8. A method for tracing the source of power line inspection drone accidents based on multi-source data fusion as described in claim 7, characterized in that, The calculation method for the comprehensive score of the operation response is as follows: S = St × Kp; Where S is the comprehensive score for operational response, St is the basic score for response timeliness, and Kp is the operational rationality coefficient. The response time base score St is calculated according to the following rules: from the time the system prompt text appears or the trigger timestamp to the time of the first joystick movement or button press extracted from the remote control operation command data, if the response time is <1 second, then St=100; if 1 second < response time < 3 seconds, then St=80; if the response time is >3 seconds or there is no response, then St=50. The operation rationality coefficient Kp is calculated according to the following rules: if the operation conforms to the recommended strategy, then Kp=1.0; if the operation does not conform to the recommended strategy or there is no effective operation, then Kp=0.

5.

9. A method for tracing the source of power line inspection drone accidents based on multi-source data fusion as described in claim 8, characterized in that, The accident analysis report includes the following: (1) Basic information of abnormal events: power inspection drone ID, flight duration, trigger timestamp, and abnormal event type; (2) Explanation of the integrity of evidence: scope of multi-source data collection, consistency of timestamps, and encryption status of encrypted evidence files; (3) Key event sequence table: List all abnormal events, screen prompt events, and operation command events in chronological order; (4) Causal inference conclusions: the order of occurrence of screen prompt events and abnormal events, confidence level, and causal inference logic; (5) Operational response evaluation results: the basic score for response timeliness, the score for operational rationality coefficient, and the comprehensive score for operational response; (6) Visualized evidence charts: multi-source data curve overlay charts and unified time axis annotation charts for key periods; (7) Original evidence index: hash value of encrypted evidence file, evidence certificate of third-party evidence storage platform / blockchain network.

10. A power line inspection drone accident tracing system based on multi-source data fusion, used to execute the power line inspection drone accident tracing method based on multi-source data fusion as described in any one of claims 1-9, characterized in that, It includes the following units: time synchronization unit, data acquisition unit, writing unit, trigger signal generation unit, data extraction unit, evidence file generation unit, uploading unit, and analysis report output unit; The time synchronization unit is used to synchronize the time between the drone and the remote controller; The data acquisition unit is used to acquire multi-source data and add a uniform millisecond-level timestamp to each frame of data for spatiotemporal alignment of multi-source data. The multi-source data includes UAV flight status data to reflect changes in the physical state of the UAV, remote controller user voice data to record the operator's subjective intentions and communication with the environment, remote controller operation command data to record the operator's operation sequence, remote controller screen recording data to capture key text information, and UAV flight image data to provide visual evidence of the flight environment. The writing unit is used to continuously write spatiotemporally aligned multi-source data into the cache module; The trigger signal generation unit is used to determine whether to generate a trigger signal based on the UAV flight status data and according to a preset judgment rule. If a trigger signal is generated, the data extraction unit will be executed; The data extraction unit is used to extract multi-source data for key time periods from the cache module according to a dynamic window strategy after receiving a trigger signal. The evidence file generation unit is used to associate and package the extracted multi-source data from key time periods in chronological order to generate encrypted evidence files; The uploading unit is used to upload the hash value of the encrypted evidence file to a third-party evidence storage platform or blockchain network to generate an immutable evidence storage certificate. The analysis report output unit is used to decrypt the encrypted evidence file to obtain decrypted multi-source data; Based on the decrypted multi-source data, synchronous playback and causal inference are performed to output an accident analysis report.

Images