Unmanned aerial vehicle flight parameter real-time monitoring method and system supporting remote rollback

Abstract

The application relates to the technical field of unmanned aerial vehicle flight parameter monitoring, and discloses an unmanned aerial vehicle flight parameter real-time monitoring method and system supporting remote backtracking, which comprises the following steps: according to collected real-time flight parameters of an unmanned aerial vehicle, predicting a flight state change trend of the unmanned aerial vehicle, expanding parameters, and screening a first expanded parameter set; calculating the correlation degree between parameters, and constructing a first correlation degree baseline; screening a core parameter set, screening expanded parameters with a correlation entropy greater than a preset entropy value, obtaining a second expanded parameter set; calculating the priority of corresponding expanded parameters in the second expanded parameter set, layering the parameters, mapping the parameters to the same space according to corresponding layers, and fusing the parameters, to obtain fused parameters; analyzing the state of the unmanned aerial vehicle in real time, identifying fault parameters, and backtracking in time, to perform real-time monitoring and analysis on unmanned aerial vehicle flight parameters; and the application can quickly locate fault sources and improve parameter monitoring backtracking effects through correlation entropy and layered fusion.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle flight parameter monitoring, and more particularly to an unmanned aerial vehicle flight parameter real-time monitoring method and system supporting remote backtracking. BACKGROUND

[0002] The flight safety of unmanned aerial vehicles is of great importance, and therefore real-time monitoring and fault diagnosis of flight parameters become critical. In the prior art, unmanned aerial vehicle flight parameter monitoring relies on simple threshold alarms or basic data analysis methods, such as detecting whether a parameter is out of limits by pre-setting a fixed threshold or identifying outliers using a statistical model, which can only trigger an alarm after a fault occurs and cannot predict fault trends in advance, resulting in insufficient response time and difficulty in tracing the origin and propagation path of the fault, making it difficult for maintenance personnel to quickly locate the root cause.

[0003] The prior art has the following problems: based on a single threshold alarm, it cannot handle gradual faults or associated faults, and the false alarm rate and the missed alarm rate are high; it ignores the dynamic correlation between parameters, and the monitoring effect is not good under complex flight conditions; it lacks effective causal analysis methods, and cannot quickly backtrack after a fault occurs, resulting in long fault response time and poor fault handling effect; to solve at least one of the above problems, the present application proposes an unmanned aerial vehicle flight parameter real-time monitoring method and system supporting remote backtracking. SUMMARY

[0004] In view of the deficiencies in the prior art, the purpose of the present application is to provide an unmanned aerial vehicle flight parameter real-time monitoring method and system supporting remote backtracking, which can effectively solve the problems in the background art. The specific technical solutions of the present application are as follows:

[0005] The unmanned aerial vehicle flight parameter real-time monitoring method supporting remote backtracking comprises:

[0006] According to the collected real-time flight parameters of the unmanned aerial vehicle, the flight state change trend of the unmanned aerial vehicle is predicted, the parameters are expanded, and a first expanded parameter set is selected;

[0007] Based on the flight parameters obtained under the normal flight state of the unmanned aerial vehicle, the correlation degree between the parameters is calculated, and a first correlation degree baseline is constructed;

[0008] The real-time flight parameters of the unmanned aerial vehicle are clustered, a core parameter set is selected, the correlation degree between each expanded parameter in the first expanded parameter set and the core parameter set is calculated, compared with the first correlation degree baseline, the correlation entropy of each expanded parameter is calculated, the expanded parameters with correlation entropy greater than a preset entropy value are selected, and a second expanded parameter set is obtained;

[0009] The configuration parameter hierarchical fusion mechanism is configured to calculate the priority of the corresponding extended parameter in the second extended parameter set, hierarchically arrange the parameters, map the parameters to the same space according to the corresponding level and fuse the parameters to obtain the fused parameters.

[0010] The real-time flight parameters of the unmanned aerial vehicle are clustered to filter out a core parameter set, the matching degree between each extended parameter in the first extended parameter set and the core parameter set is analyzed, a time lag value is calculated, the extended parameters are compensated according to the time lag value, the correlation degree between the compensated extended parameters and the core parameter set is calculated, and the correlation entropy is obtained by comparing the correlation degree with the first correlation degree baseline.

[0011] Specifically, the flight state change trend of the unmanned aerial vehicle is predicted according to the collected real-time flight parameters of the unmanned aerial vehicle, the parameters are expanded, and a first extended parameter set is filtered out, including:

[0012] According to the collected real-time flight parameters of the unmanned aerial vehicle and the preset flight task instruction, a task transition point is identified, and a flight state change trend corresponding to the task transition point is predicted through a preset state analysis model.

[0013] Based on the flight state change trend, the parameters are expanded, and a first extended parameter set is filtered out.

[0014] Specifically, the correlation degree between the parameters is calculated based on the flight parameters obtained under the normal flight state of the unmanned aerial vehicle, and a first correlation degree baseline is constructed, including:

[0015] Based on the flight parameters obtained under the normal flight state of the unmanned aerial vehicle, different flight working conditions are obtained by analyzing the flight height, speed and task mode, the flight parameters are divided, and a plurality of parameter slices are obtained.

[0016] For each parameter slice, the parameter correlation degree is analyzed, the mutual information value between the parameters is calculated, and a corresponding parameter correlation matrix is constructed.

[0017] According to the parameter correlation matrix, a first correlation degree baseline is constructed.

[0018] Specifically, the real-time flight parameters of the unmanned aerial vehicle are clustered to filter out a core parameter set, the matching degree between each extended parameter in the first extended parameter set and the core parameter set is analyzed, a time lag value is calculated, the extended parameters are compensated according to the time lag value, the correlation degree between the compensated extended parameters and the core parameter set is calculated, and the correlation entropy is obtained by comparing the correlation degree with the first correlation degree baseline.

[0019] The real-time flight parameters of the unmanned aerial vehicle are clustered to filter out a core parameter set;

[0020] The matching degree between each extended parameter in the first extended parameter set and the core parameter set is analyzed, a time lag value is calculated, the extended parameters are compensated according to the time lag value, the correlation degree between the compensated extended parameters and the core parameter set is calculated, and the correlation entropy is obtained by comparing the correlation degree with the first correlation degree baseline.

[0021] Screen out the expansion parameter with the correlation entropy greater than the preset entropy value to obtain a second expansion parameter set.

[0022] Specifically, the matching degree between each expansion parameter in the first expansion parameter set and the core parameter set is analyzed, a time lag value is calculated, the expansion parameter is compensated according to the time lag value, the correlation degree between the compensated expansion parameter and the core parameter set is calculated, and the correlation entropy is obtained by comparing with the first correlation degree baseline, including:

[0023] The matching degree between each expansion parameter in the first expansion parameter set and the core parameter set is analyzed, and the corresponding matching path is screened out, and the time lag value is calculated according to the matching path;

[0024] The expansion parameter is time lag compensated according to the time lag value, the mutual information value between the compensated expansion parameter and the core parameter set is calculated, the mutual information change rate is obtained by comparing the mutual information value with the mutual information average value in the first correlation degree baseline;

[0025] The joint probability distribution between each expansion parameter and the core parameter is analyzed, the mismatch degree of the joint behavior mode between the parameters is calculated, and the mismatch degree is obtained.

[0026] The mutual information change rate and the mismatch degree are weighted and fused to obtain the correlation entropy.

[0027] Specifically, the parameter hierarchical fusion mechanism includes:

[0028] The comprehensive value of the corresponding expansion parameter in the second expansion parameter set is analyzed by a preset parameter value analysis model, the corresponding priority is calculated, and the parameters are layered according to the priority;

[0029] According to the corresponding level, the parameters are respectively mapped to the same space and fused to obtain the fused parameters.

[0030] Specifically, the comprehensive value of the corresponding expansion parameter in the second expansion parameter set is analyzed by a preset parameter value analysis model, the corresponding priority is calculated, and the parameters are layered according to the priority, including:

[0031] The comprehensive value of the corresponding expansion parameter in the second expansion parameter set is analyzed and calculated by a preset parameter value analysis model from three dimensions of static representation, dynamic response and backtracking positioning, and a parameter value matrix is constructed;

[0032] Based on the parameter value matrix, the influence degree of each expansion parameter on the unmanned aerial vehicle state is analyzed, projected to the corresponding flight task space, and a parameter task mapping matrix is constructed;

[0033] The parameter value matrix and the parameter task mapping matrix are fused to calculate the parameter priority;

[0034] The parameters corresponding to each extended parameter are layered according to parameter priority.

[0035] Specifically, the parameters are respectively mapped into the same space and fused according to the corresponding hierarchy to obtain the fused parameters, including:

[0036] The time sequence characteristics and spectral energy distribution of the extended parameters in each hierarchy are analyzed according to the corresponding hierarchy, the energy proportion of the Fourier transform of the first priority hierarchy parameter is calculated to construct a first descriptor, the deviation degree of the second priority hierarchy parameter from the reference parameter is calculated to construct a second descriptor, and the mutual information value between the third priority hierarchy parameter and the adjacent layer parameter is calculated to construct a third descriptor;

[0037] According to the first descriptor, the corresponding real-time parameters are scale transformed to be mapped into the first parameters; according to the second descriptor, the corresponding real-time parameters are mapped into the second parameters in combination with the mutual information value between the real-time parameters; and according to the third descriptor, the corresponding real-time parameters are isometrically mapped to obtain the third parameters;

[0038] The first parameters, the second parameters and the third parameters are weighted and fused to obtain the fused parameters.

[0039] Specifically, the real-time analysis of the unmanned aerial vehicle state according to the fused parameters, the identification of the fault parameters and the forward tracing in time, the filtering of the fault propagation chain between the parameters through the preset causal analysis model, including:

[0040] The real-time analysis of the unmanned aerial vehicle state according to the fused parameters, the filtering of the fault parameters;

[0041] According to the preset causal analysis model, the forward tracing in time of the fault parameters, the calculation of the transfer entropy between the parameters, and the filtering of the fault propagation chain between the parameters.

[0042] The unmanned aerial vehicle flight parameter real-time monitoring system supporting remote tracing is used to realize the unmanned aerial vehicle flight parameter real-time monitoring method supporting remote tracing, and includes:

[0043] The parameter extension module, according to the collected real-time flight parameters of the unmanned aerial vehicle, predicts the flight state change trend of the unmanned aerial vehicle to extend the parameters, and filters out a first set of extended parameters;

[0044] The correlation baseline construction module, based on the flight parameters obtained under the normal flight state of the unmanned aerial vehicle, calculates the correlation degree between the parameters to construct a first correlation degree baseline;

[0045] The parameter filtering module clusters the real-time flight parameters of the UAV, filters out the core parameter set, calculates the correlation degree between each extended parameter in the first extended parameter set and the core parameter set, compares it with the first correlation degree baseline, calculates the correlation entropy of each extended parameter, filters out the extended parameters whose correlation entropy is greater than the preset entropy value, and obtains the second extended parameter set.

[0046] The parameter fusion module configures a hierarchical parameter fusion mechanism, calculates the priority of the corresponding extended parameters in the second extended parameter set, hierarchically divides the parameters, maps the parameters to the same space according to the corresponding level and fuses them to obtain fused parameters;

[0047] The parameter backtracking module analyzes the UAV status in real time based on the fused parameters, identifies fault parameters, and backtracks them in time. It uses a preset causal analysis model to filter out the fault propagation chain between parameters, so as to monitor and analyze the UAV flight parameters in real time.

[0048] The beneficial effects of this application are as follows: Based on flight mission instructions and state change trends, parameter expansion is predicted to generate a first expanded parameter set, enhancing the comprehensiveness and foresight of the parameter set. Under normal flight conditions, a first correlation baseline is constructed based on parameter slices and mutual information values, providing a dynamic benchmark for anomaly detection. A core parameter set is obtained through clustering, and the correlation entropy between the expanded parameters and the core parameter set is calculated to select a second expanded parameter set, improving fault sensitivity. Priority is calculated through a parameter value analysis model, parameters are hierarchically layered and mapped to a high-dimensional space for fusion, resulting in fused parameters. Real-time analysis of the UAV's state is performed, and a causal analysis model is used to trace back and identify fault propagation chains, enabling parameter backtracking and fault analysis. Through correlation entropy and hierarchical fusion, false alarms and false negatives are reduced, improving fault detection accuracy. Screening fault propagation chains allows for rapid location of fault root causes, enhancing the effectiveness of real-time parameter monitoring and backtracking. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the real-time monitoring method for UAV flight parameters that supports remote backtracking, as described in Embodiment 1 of this application.

[0050] Figure 2 This is a flowchart illustrating the correlation entropy calculation process in Embodiment 1 of this application;

[0051] Figure 3 This is a schematic diagram of the hierarchical structure of the extended parameters in Embodiment 1 of this application;

[0052] Figure 4 This is a schematic diagram of the structure of the real-time monitoring system for UAV flight parameters that supports remote backtracking, as described in Embodiment 1 of this application. Detailed Implementation

[0053] The present application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0054] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0055] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0056] Example 1:

[0057] refer to Figure 1 The image shows a specific implementation of the real-time monitoring method for UAV flight parameters that supports remote backtracking, as described in this application, including:

[0058] S101. Based on the collected real-time flight parameters of the UAV, predict the flight status change trend of the UAV and expand the parameters to select the first extended parameter set.

[0059] S102. Based on the flight parameters obtained under the normal flight state of the UAV, calculate the correlation between the parameters and construct the first correlation baseline.

[0060] S103. Cluster the real-time flight parameters of the UAV, select the core parameter set, calculate the correlation degree between each extended parameter in the first extended parameter set and the core parameter set, compare it with the first correlation degree baseline, calculate the correlation entropy of each extended parameter, select the extended parameters whose correlation entropy is greater than the preset entropy value, and obtain the second extended parameter set.

[0061] S104. Configure a hierarchical fusion mechanism for parameters, calculate the priority of the corresponding extended parameters in the second extended parameter set, hierarchically classify the parameters, map the parameters to the same space according to the corresponding level and fuse them to obtain fused parameters;

[0062] S105. Analyze the UAV status in real time based on the fused parameters, identify fault parameters and trace back in time, and filter out the fault propagation chain between parameters through a preset causal analysis model in order to monitor and analyze the UAV flight parameters in real time.

[0063] Currently, drone flight parameter monitoring relies heavily on simple threshold alarms or static models, which can only detect parameters that deviate significantly from the normal range. They cannot cope with complex and gradual faults, monitor individual parameters in isolation, and ignore the inherent correlation between dynamic changes in flight parameters. This results in insufficient predictive ability of the system for potential faults, and after a fault occurs, it is also impossible to effectively trace its root cause and propagation path. As a result, it is difficult to meet the requirements of high reliability and flight safety for drones in complex missions.

[0064] In this embodiment, based on the collected real-time flight parameters of the UAV, the preset flight mission command is parsed and compared with the position and speed in the real-time flight parameters to identify mission transition points, including but not limited to the transition from cruising to hovering. A preset state analysis model is used to predict the flight state change trend at the corresponding mission transition point. The state analysis model includes, but is not limited to, a time-series prediction model built based on a Long Short-Term Memory (LSTM) network. The input of this model is the sequence of flight parameters within the current and historical window and the code of the mission command to be executed. The flight parameter sequence includes, but is not limited to, attitude angle, speed, and altitude. The output is the trajectory of the change of key state parameters within several future sampling periods. The key state parameters include, but are not limited to, the expected pitch angle, roll angle, and thrust.

[0065] Based on the predicted flight status change trend of the UAV at the mission transition point, the parameters are expanded, and a first set of expanded parameters is selected. For example, if an imminent climb is predicted, the ratio of vertical acceleration to engine command thrust is calculated in real time as a new derived parameter. At the same time, parameters whose importance increases under this trend are activated or enhanced. By analyzing and predicting the UAV status and expanding the parameters, subtle deviations between parameter behavior and flight intentions can be predicted before the actual parameter exceedance caused by a fault. This increases the lead time for fault warnings and can effectively prevent the fault from developing into a functional failure.

[0066] Specifically, based on flight parameters acquired during normal UAV flight, the parameter data is sliced ​​according to key operational dimensions such as flight altitude, speed, and mission mode. For each slice, the correlation between parameters is calculated to construct a first correlation baseline. For example, low altitude-low speed-hovering is a typical operational condition slice. For the data within each slice, the set of parameters to be monitored is selected, and the mutual information value between every two parameters is calculated. By constructing the first correlation baseline, the problem of poor adaptability of fixed thresholds or single static models in complex and variable flight environments can be overcome. By analyzing mutual information and operational condition slices, it is possible to accurately determine whether the correlation between parameters is normal under the current specific flight conditions, enhance anomaly detection capabilities and adaptability to different flight modes, and improve the accuracy of anomaly identification.

[0067] Specifically, the real-time flight parameters of the UAV are clustered to select a core parameter set. The correlation degree between each extended parameter in the first extended parameter set and the core parameter set is calculated and compared with the first correlation degree baseline. The time lag value is determined by calculating the matching path and time lag compensation is performed. The correlation entropy of each extended parameter is calculated, and extended parameters with correlation entropy greater than the preset entropy value are selected to obtain the second extended parameter set. By calculating the correlation entropy and combining it with statistical dependence and joint distribution pattern changes, abnormal correlation relationships of parameters can be monitored. This can effectively filter out parameters with fluctuating values ​​but not closely related to the core state, improve the accuracy of the selected key parameters, analyze accurate fault parameters, and improve the processing efficiency of system parameter monitoring.

[0068] Furthermore, a hierarchical parameter fusion mechanism is configured to calculate the priority of corresponding extended parameters in the second extended parameter set, hierarchically classify real-time parameters, and adopt different strategies according to the corresponding level to map real-time parameters to the same high-dimensional space and fuse them to obtain fused parameters. Through priority hierarchical classification, the importance of key parameter information in the fusion result can be ensured, and interference from secondary information can be avoided. Mapping parameters with different characteristics to a unified high-dimensional space and fusing them to obtain fused parameters can retain the core state information in the original parameter set. Through feature extraction and dimensionality reduction, the computational load of subsequent processing can be reduced, and the efficiency of fault monitoring and analysis can be improved.

[0069] Specifically, the system analyzes the UAV status in real time based on fused parameters, identifies fault parameters, and traces them back in time. A pre-defined causal analysis model analyzes the causal flow between fault parameters and other parameters over time. This model uses transit entropy as a measure of causal strength, calculating the transit entropy from candidate parameters to the fault parameter. The calculation process is based on conditional probability, quantifying the additional information provided by the past values ​​of candidate parameters for predicting the future values ​​of fault parameters when their own past values ​​are known. The system also calculates the transit entropy from all potentially fault-related parameters to the fault parameter and performs statistical significance tests. This process identifies the fault propagation chain between parameters for real-time monitoring and analysis of UAV flight parameters. Through fault propagation chain analysis, the system can analyze the propagation and development of faults, pinpoint the initial abnormal parameters, provide accurate parameter references for fault maintenance, enable precise maintenance decisions, shorten troubleshooting time, and improve operational efficiency and system reliability.

[0070] This application predicts parameter expansion based on flight mission instructions and state change trends, generating a first expanded parameter set. This enhances the comprehensiveness and foresight of the parameter set. Under normal flight conditions, a first correlation baseline is constructed based on parameter slices and mutual information values, providing a dynamic benchmark for anomaly detection. A core parameter set is obtained through clustering, and the correlation entropy between the expanded parameters and the core parameter set is calculated. A second expanded parameter set is then selected, improving fault sensitivity. Priority is calculated using a parameter value analysis model, parameters are hierarchically layered, and mapped to a high-dimensional space for fusion, resulting in fused parameters. Real-time analysis of the UAV's state is performed, and a causal analysis model is used to trace back and identify fault propagation chains, enabling parameter backtracking and fault analysis. Through correlation entropy and hierarchical fusion, false alarms and false negatives are reduced, improving fault detection accuracy. Screening fault propagation chains allows for rapid location of fault root causes, enhancing the effectiveness of real-time parameter monitoring and backtracking.

[0071] Furthermore, based on the collected real-time flight parameters of the UAV, the parameters are expanded according to the predicted flight status change trend of the UAV, and a first set of expanded parameters is selected, including:

[0072] S201. Based on the collected real-time flight parameters of the UAV and the preset flight mission instructions, identify the mission transition points and predict the flight status change trend of the corresponding mission transition points through the preset state analysis model.

[0073] S202. Based on the flight state change trend, the parameters are expanded, and a first set of expanded parameters is selected.

[0074] In this embodiment, based on the collected real-time flight parameters of the UAV and the preset flight mission instructions, the real-time flight parameters of the UAV include, but are not limited to, the current geographical location, altitude, and speed, while the flight mission instructions include, but are not limited to, waypoints and action command sets. The real-time flight parameters are compared with the next target state required by the mission instructions to identify the corresponding mission transition point. For example, when the UAV approaches the predetermined waypoint and is about to execute the hovering command, the difference between the current state and the target state identifies the mission transition point from level flight to hovering. The flight state change trend at the corresponding mission transition point is predicted by a preset state analysis model. The state analysis model includes, but is not limited to, an LSTM neural network model pre-trained based on the UAV dynamics model and a large amount of historical flight data, capable of simulating the transition process between different tasks. The model predicts the trajectory of the flight state change at a specific transition point, thus obtaining the flight state change trend.

[0075] Specifically, during the training phase, the state analysis model uses historical normal flight data to learn dynamic patterns during task transitions. The model's input is a multi-dimensional time series within a fixed time window (e.g., the past 5 seconds), including core state parameters such as attitude angles, three-axis acceleration, altitude, and airspeed, as well as a one-hot encoded vector representing the task type for the next stage. The model's output is the predicted sequence of changes in the core state parameters within a future prediction window (e.g., the next 3 seconds). Through training, the model establishes a mapping relationship from the current state and task intent to the short-term future state trajectory. Model analysis yields quantitative predictions of upcoming flight states, including but not limited to attitude adjustment rate and deceleration curves.

[0076] It should be noted that by analyzing and predicting the flight status change trend at the task transition point, the future status change can be predicted based on the current status. This allows for the early prediction of the normal flight trend and trajectory of the UAV in the coming period, providing a trajectory benchmark for detecting abnormal states. It can also detect early signs of failure when the absolute value of the parameters has not yet exceeded the standard, but the dynamic trend has deviated from the expected, thus improving the system's accuracy in identifying potential risks.

[0077] Specifically, based on the trend of flight status changes, parameters are expanded, and new derived parameters are generated according to the predicted flight status. For example, if the UAV is about to climb, the expected ratio of vertical acceleration to engine thrust is used as a new monitoring variable. The monitoring weight of subsequent scenario parameters is increased based on mission transition points. For example, when a landing mission transition point is identified, the monitoring weight of parameters such as landing gear status and altitude, which are not heavily monitored during level flight, is increased. From the original parameter set and the expanded new parameters, parameters reflecting flight status characteristics are selected to obtain the first expanded parameter set. By dynamically expanding the monitored parameter set, it can be ensured that at any given time, monitoring resources are concentrated on the parameter dimensions most relevant to the fault, improving the targeting and efficiency of the parameter monitoring system and providing an accurate data foundation for operational condition correlation analysis and anomaly detection.

[0078] Furthermore, based on the flight parameters acquired during normal UAV flight, the correlation between the parameters is calculated to construct a first correlation baseline, including:

[0079] S301. Based on the flight parameters obtained under normal flight conditions of the UAV, analyze the flight altitude, speed and mission mode to obtain different flight conditions, divide the flight parameters, and obtain multiple parameter slices.

[0080] S302. For each parameter slice, analyze the parameter correlation, calculate the mutual information value between parameters, and construct the corresponding parameter correlation matrix.

[0081] S303. Construct the first correlation baseline according to the parameter correlation matrix.

[0082] In this embodiment, based on historical flight parameters acquired during normal UAV flight, each sampled data point is labeled according to three key dimensions: flight altitude, speed, and mission mode, resulting in multiple sets of operational condition labels, including but not limited to high-altitude-high-speed-cruise mode and low-altitude-low-speed-hovering mode. Based on these operational condition labels, a data density-based clustering method is used to divide all flight parameter data into different parameter slices. The data within each parameter slice originates from the same or similar flight operational conditions. By analyzing flight operational conditions to divide the parameters, the global flight data is divided into parameter slices corresponding to different operational conditions. This allows for analysis of the flight conditions corresponding to each parameter slice, avoiding misjudgments caused by interference between correlation patterns under different operational conditions and improving the accuracy of the correlation baseline.

[0083] For each parameter slice, the mutual information value between parameters is calculated by analyzing the parameter correlation. This process iterates through all parameter pairs within the slice. For each parameter pair, the KL divergence between the product of the joint probability distribution and the product of their respective marginal probability distributions is calculated to obtain the mutual information value of the parameter pair across the entire slice data sequence. The mutual information value reflects how much uncertainty can be reduced for one parameter once the value of another parameter is known. The calculated mutual information values ​​for all parameter pairs are then integrated sequentially to construct the corresponding parameter correlation matrix. The matrix elements reflect the degree of intrinsic correlation between each pair of parameters under the current flight conditions. By calculating mutual information values ​​and analyzing the intrinsic relationships between parameters, it is possible to effectively analyze parameter correlations that exhibit asynchronous numerical changes but deep dependencies, thereby improving the ability to identify faults caused by nonlinear correlations.

[0084] Specifically, based on the parameter correlation matrix, each calculated parameter correlation matrix is ​​bound to the corresponding operating condition. The bound data units are then integrated in chronological order to construct a first correlation baseline covering all known normal flight conditions, including combinations of three dimensions: flight altitude, flight speed, and mission mode. By constructing this first correlation baseline, a dynamic behavioral benchmark is provided for the real-time parameter monitoring process. During real-time monitoring, the corresponding normal correlation pattern can be quickly retrieved from this baseline based on the current real-time flight condition of the UAV. Based on dynamically changing local standards that are highly relevant to the current flight situation, the accuracy and adaptability of the fault diagnosis process can be improved, effectively reducing false alarms caused by normal changes in flight status.

[0085] Furthermore, the real-time flight parameters of the UAV are clustered to select a core parameter set. The correlation degree between each extended parameter in the first extended parameter set and the core parameter set is calculated and compared with the first correlation degree baseline. The correlation entropy of each extended parameter is calculated, and extended parameters with a correlation entropy greater than a preset entropy value are selected to obtain a second extended parameter set, including:

[0086] S401. Cluster the real-time flight parameters of the UAV and filter out the core parameter set;

[0087] S402. Analyze the matching degree between each extended parameter in the first extended parameter set and the core parameter set, calculate the time lag value, compensate the extended parameters according to the time lag value, calculate the correlation degree between the compensated extended parameters and the core parameter set, compare it with the first correlation degree baseline, and obtain the correlation entropy.

[0088] S403. Filter out the extended parameters whose correlation entropy is greater than the preset entropy value to obtain the second extended parameter set.

[0089] In this embodiment, density-based clustering is used to cluster the real-time flight parameters of the UAV, analyze the change patterns and magnitudes of parameter values ​​over time, automatically group parameters with similar behaviors into the same cluster, extract the central parameter of each cluster as the corresponding core parameter, and filter out the core parameter set. Clustering can quickly extract the core parameter subset reflecting the core state of the system from the high-dimensional parameter space, providing an accurate data reference benchmark for subsequent analysis. It can also perform deep correlation analysis based on the extended parameters most related to the core state, improving the efficiency and relevance of extended parameter calculation.

[0090] Specifically, the matching degree between each extended parameter in the first extended parameter set and the core parameter set is analyzed. The cross-correlation coefficient is calculated to filter the time offset that maximizes the synergy between the two, obtaining the time lag value. The extended parameters are compensated according to the time lag value, and the correlation degree between the compensated extended parameters and the core parameter set is calculated. This is compared with the first correlation degree baseline to obtain the correlation entropy. By analyzing the time lag compensation and correlation entropy, the dynamic correlation anomalies between parameters can be analyzed. This can effectively identify the hidden correlation disruptions that are ignored due to fault propagation delays. Through multi-dimensional evaluation and analysis, the comprehensiveness and accuracy of the abnormal correlation judgment results can be improved, and misjudgments caused by noise or normal fluctuations can be reduced.

[0091] Specifically, by analyzing and statistically processing historical normal and fault-free abnormal data, and setting entropy values ​​based on fault identification accuracy, extended parameters with associated entropy values ​​greater than preset entropy values ​​are filtered out, resulting in a second set of extended parameters. These parameters reflect whether they are under abnormal influence or are themselves the source of the abnormality. By filtering out this second set of extended parameters, fault information can be extracted. This process eliminates a large number of extended parameters that are normally associated with the core state or fluctuate within a reasonable range. Limited computing resources can be concentrated on key parameters reflecting the fault situation, improving the signal-to-noise ratio and analysis efficiency of the system parameter monitoring process, and enhancing the accuracy of fault location and analysis.

[0092] like Figure 2 As shown, the matching degree between each extended parameter in the first extended parameter set and the core parameter set is analyzed, the time lag value is calculated, the extended parameters are compensated according to the time lag value, the correlation degree between the compensated extended parameters and the core parameter set is calculated, and compared with the first correlation degree baseline to obtain the correlation entropy, including:

[0093] S501. Analyze the degree of matching between each extended parameter in the first extended parameter set and the core parameter set, filter out the corresponding matching paths, and calculate the time lag value according to the matching paths.

[0094] S502. Perform time lag compensation on the extended parameters according to the time lag value, calculate the mutual information value between the compensated extended parameters and the core parameter set, and compare the mutual information value with the average mutual information value in the first correlation baseline to obtain the mutual information change rate.

[0095] S503. Analyze the joint probability distribution between each extended parameter and the core parameter, calculate the degree of mismatch between the joint behavior patterns of the parameters, and obtain the degree of mismatch.

[0096] S504. The mutual information change rate and mismatch degree are weighted and fused to obtain the correlation entropy.

[0097] In this embodiment, by analyzing historical normal data and prior knowledge of the system's physical topology, matching paths between each extended parameter in the first extended parameter set and the core parameter set are selected. Transitional parameters between the extended parameters and the core parameters are then matched. For example, the correlation between the extended parameter motor current and the core parameter flight airspeed is analyzed to obtain the transitional parameter propeller thrust. For each matching path, time-series mutual information analysis is used to find the time offset that maximizes the synergy between the parameters at both ends of the path in the normal data, and the time lag value corresponding to each pair of parameters is calculated. By calculating the corresponding time lag values ​​for different parameter pairs, the internal dynamic processes of the UAV system can be reflected, providing data support for time lag compensation. This avoids errors in correlation calculation caused by using a uniform lag, and provides data support for accurately analyzing anomaly propagation.

[0098] Specifically, based on the calculated time lag value, the time series of each extended parameter in the first extended parameter set is shifted accordingly to align it with the corresponding parameter in the core parameter set in causal time. The mutual information value between the compensated extended parameter and the core parameter set is calculated. From the first correlation baseline, the average mutual information value of the compensated parameter under normal conditions is retrieved under the current flight conditions. The ratio of the real-time calculated mutual information value to the average mutual information value under normal conditions is calculated to obtain the mutual information change rate. Time lag compensation can restore the true instantaneous correlation between parameters, and can sensitively detect the attenuation or enhancement of information flow between parameters due to faults. By calculating the mutual information change rate, abnormalities in the correlation strength can be reflected, and early faults can be detected where the parameter value itself is not obviously abnormal, but the dynamic information interaction between it and the core state has already encountered problems, thus improving the sensitivity of fault detection.

[0099] Specifically, the joint probability distribution between each extended parameter and the core parameter is analyzed and compared with the normal joint probability distribution under the same working conditions obtained from the first correlation baseline. The difference between the two distributions is calculated using the KL divergence statistical distance metric to reflect the degree of mismatch in the joint behavior patterns between parameters, thus obtaining the mismatch degree. By analyzing the abnormal cooperative behavior patterns between parameters, fault types in which the correlation strength does not change much but the cooperative pattern has been distorted can be captured. This can make up for the shortcomings of relying solely on the mutual information change rate for anomaly analysis and improve the comprehensiveness and accuracy of the fault identification process.

[0100] The calculated mutual information change rate and mismatch degree are normalized, and weights are assigned based on their contribution to different types of anomalies in historical fault data. The normalized mutual information change rate and mismatch degree are then weighted and fused according to these weights to obtain the correlation entropy, which reflects the degree of disruption of the normal correlation between the extended parameter and the core parameter set. A higher correlation entropy value indicates a more severe disruption of the normal correlation between the extended parameter and the core parameter set, and a higher probability of anomaly. By calculating the correlation entropy and performing anomaly correlation analysis, intensity and pattern information can be integrated, avoiding the limitations of a single indicator, effectively capturing and amplifying parameter correlation disruption, reducing the probability of false positives and false negatives, and improving the accuracy of parameter monitoring.

[0101] Furthermore, the parameter hierarchical fusion mechanism includes:

[0102] S601. Using a preset parameter value analysis model, analyze the comprehensive value of the corresponding extended parameters in the second extended parameter set, calculate the corresponding priority, and stratify the parameters according to the priority.

[0103] S602. According to the corresponding level, map the parameters to the same space and merge them to obtain the merged parameters.

[0104] In this embodiment, a preset parameter value analysis model is used to analyze the comprehensive value of the corresponding extended parameters in the second extended parameter set from the perspectives of static representation value, dynamic response value, and backtracking positioning value. The corresponding priorities are calculated, and the real-time parameters are layered according to the priorities. By calculating the comprehensive value of the extended parameters and layering the parameters, it is possible to ensure that the high-value fault indication parameters are highlighted during the parameter fusion process, thereby improving the signal-to-noise ratio and fault information density of the final state representation vector.

[0105] Specifically, the input to the parameter value analysis model is the metadata of each parameter in the second extended parameter set and its performance in historical and real-time data. The evaluation process includes: obtaining the static characterization value by setting basic weight scores based on the parameter's physical position and functional criticality in the UAV system dynamics model; calculating the dynamic response value by analyzing the parameter's performance in historical known failure events; determining the backtracking value by statistically analyzing the frequency of the parameter as a root cause node or key propagation node based on the historical failure case library, with higher frequencies corresponding to greater backtracking values; the model quantifies and scores each dimension, assigns preset weights for weighted summation, and outputs the comprehensive value of each parameter.

[0106] Specifically, according to the corresponding levels, real-time parameters are mapped to the same high-dimensional space and fused using different mapping methods to obtain fused parameters. For parameters at different value levels, corresponding features are extracted and weighted fusion is performed in the high-dimensional space. This retains the core dynamic information from key parameters, integrates steady-state deviations and related contextual information, provides accurate data references for state identification and fault diagnosis, and improves the analytical performance and decision reliability of the parameter monitoring system.

[0107] Furthermore, using a pre-defined parameter value analysis model, the comprehensive value of the corresponding extended parameters in the second extended parameter set is analyzed, the corresponding priorities are calculated, and the real-time parameters are stratified according to priority, including:

[0108] S701. Using a pre-defined parameter value analysis model, the comprehensive value of the corresponding extended parameters in the second extended parameter set is analyzed and calculated from three dimensions: static characterization, dynamic response, and backtracking location, and a parameter value matrix is ​​constructed.

[0109] S702. Based on the parameter value matrix, analyze the degree of influence of each extended parameter on the UAV state, project it onto the corresponding flight mission space, and construct a parameter mission mapping matrix.

[0110] S703. Merge the parameter value matrix and the parameter task mapping matrix to calculate the parameter priority;

[0111] S704. Stratify the parameters corresponding to each extended parameter according to parameter priority.

[0112] In this embodiment, a pre-defined parameter value analysis model is used to evaluate each extended parameter in the second extended parameter set from three orthogonal dimensions. The static characterization dimension evaluates the fundamental and irreplaceable nature of the parameter in the UAV system dynamics model. For example, a parameter reflecting the core power output has a higher static value than a parameter reflecting the auxiliary environmental temperature. The dynamic response dimension evaluates the parameter's sensitivity and response speed to various fault stimuli, quantified by analyzing the parameter's response amplitude change rate and response delay to known faults in historical data. The backtracking and positioning dimension evaluates the frequency and importance of the parameter as a root node or key intermediate node in the fault causal chain based on a historical fault case library. The model calculates a corresponding score for each parameter in each of the three dimensions to obtain a comprehensive value. The parameters are then arranged according to their comprehensive value and corresponding dimensions to construct a parameter value matrix. By constructing the parameter value matrix, accurate data support is provided for analyzing the comprehensive value of parameters, avoiding the one-sidedness of ranking based on a single indicator and improving the accuracy and comprehensiveness of parameter value analysis and evaluation.

[0113] Specifically, the input to the parameter value analysis model is the identifier and related data of each parameter in the second extended parameter set. Model evaluation is conducted on three independent dimensions: static characterization value based on the physical nature of the parameter and the system functional architecture analysis; dynamic response value obtained by analyzing historical fault datasets to calculate the mean absolute value of the first-order difference of each parameter in the early stages of various confirmed fault cases and the average lead time of parameter changes relative to standard fault alarm tags; and retrospective location value obtained by calculating the frequency with which the parameter was identified as a fault root cause node or key propagation node in history based on historical fault root cause analysis reports. For each parameter, its score in the three dimensions is calculated to form a three-dimensional value vector. Arranging the value vectors of all parameters row-wise constitutes a parameter value matrix.

[0114] Specifically, based on the parameter value matrix, the impact of each extended parameter on the UAV's state is analyzed. These parameters are projected onto the current flight mission space, which is defined according to the current mission instructions and their key performance indicators. Through predefined mapping relationships—obtained by learning from extensive historical data—the criticality of each parameter in ensuring mission success is determined for the specific flight mission. For example, in a precision hovering mission, the GPS positioning accuracy parameter has a high mission mapping value. The criticality score of a parameter in the current mission is obtained through mapping relationship analysis, constructing a corresponding parameter mission mapping matrix that reflects the weight adjustment of parameter value within the specific mission context. By projecting parameters onto the flight mission space, the parameter importance analysis process can be dynamically adjusted according to the flight mission, improving the reliability of monitoring strategies in complex environments.

[0115] The parameter value matrix and parameter task mapping matrix are fused. The vector formed by each row of the parameter value matrix is ​​multiplied by the task criticality score of the parameter in the parameter task mapping matrix with weights to calculate the priority score of each parameter. All parameters are then sorted in descending order according to their priority scores to obtain the parameter priority. By integrating intrinsic value and contextual value, the importance of the parameter itself and the urgency of the current task can be combined. In complex task environments, this ensures that the most critical parameters receive the highest level of attention, thereby improving the effectiveness of the parameter monitoring process.

[0116] like Figure 3 As shown, the real-time parameters corresponding to each extended parameter are layered according to parameter priority, divided into high, medium, and low priority levels. The real-time data stream corresponding to each extended parameter is then assigned to different levels. For example, the top 20% of parameters by priority score are assigned to the highest priority level, the middle 60% to the middle priority level, and the bottom 20% to the lowest priority level. Layering parameters by priority provides a structural reference for layered data fusion, enabling the setting of corresponding computational granularity, fusion weights, and transmission strategies for parameters at different levels. This allows for on-demand allocation of monitoring resources, improving the system's computational efficiency and monitoring performance.

[0117] Furthermore, according to the corresponding levels, the real-time parameters are mapped to the same high-dimensional space and fused to obtain fused parameters, including:

[0118] S801. Analyze the temporal characteristics and spectral energy distribution of the extended parameters in each level according to the corresponding level, calculate the energy proportion of the first priority level parameter after Fourier transform to construct the first descriptor, calculate the deviation of the second priority level parameter from the reference parameter to construct the second descriptor, and calculate the mutual information value between the third priority level parameter and the adjacent level parameter to construct the third descriptor.

[0119] S802. According to the first descriptor, the corresponding real-time parameters are scaled and mapped to the first parameters; according to the second descriptor and the mutual information value between the real-time parameters, the corresponding real-time parameters are mapped to the second parameters; according to the third descriptor, the corresponding real-time parameters are equidistantly mapped to obtain the third parameters.

[0120] S803. The first parameter, the second parameter, and the third parameter are weighted and fused to obtain the fused parameter.

[0121] In this embodiment, the temporal characteristics and spectral energy distribution of extended parameters within each level are analyzed according to the corresponding hierarchy. The first priority level corresponds to the highest priority level, the second priority level corresponds to the intermediate priority level, and the third priority level corresponds to the lowest priority level. For parameters of the highest priority level, the parameters reflect the changes in the core dynamic characteristics of the system. The temporal characteristics and spectral energy distribution of the dynamic process are analyzed. By performing Fourier transform on them, their spectral components are analyzed, and the energy proportion of the main frequency band in the total energy is calculated to construct a first descriptor that reflects the pattern and stability of the core dynamic behavior of the system. For parameters of the intermediate priority level, the parameters can indicate the steady-state deviation or asymptotic anomaly of the system state. By calculating the deviation of their real-time values ​​from the baseline parameters obtained by statistically analyzing historical normal data under the same operating conditions, the baseline parameters include but are not limited to the mean, and the deviation includes but is not limited to calculating the Euclidean distance, a second descriptor is constructed to quantify the degree of deviation from the normal operating conditions. For parameters of the lowest priority level, the auxiliary correlation between them and higher-level parameters is reflected. By calculating the mutual information value between them and the parameters of adjacent layers, a third descriptor is constructed to reflect the degree of support of the lower-level parameters for the upper-level state.

[0122] It should be noted that by constructing corresponding feature descriptors for parameters at different levels, the essential features of heterogeneous and multi-scale parameter information can be accurately extracted, and the dynamic patterns of the highest priority level parameters, the steady-state shifts of the intermediate priority level parameters, and the associated context of the lowest priority level parameters can be quantified. This provides a feature basis for parameter mapping fusion and can preserve the value of parameters with different properties during the fusion process.

[0123] According to the first descriptor, the first parameter is generated by multiplying the parameter value by its main frequency band energy proportion and scaling the real-time parameters of the highest priority level, thus encoding the dynamic features into the parameter representation. According to the second descriptor, the second parameter is generated by combining the mutual information values ​​between the parameters of this level and mapping the real-time parameters of the intermediate priority level to the second parameter by combining the deviation of each parameter with the relevant mutual information values. According to the third descriptor, the third parameter is obtained by performing equidistant mapping on the real-time parameters of the lowest priority level through principal component analysis. By mapping and transforming the parameters, parameters with different physical meanings and dimensions can be transformed into feature representations with the same mathematical properties. The data is standardized to generate feature inputs that are more suitable for state recognition and fault diagnosis.

[0124] Specifically, the first, second, and third parameters are assigned fusion weights based on hierarchical priority. These weights are then weighted and fused to obtain the fused parameters. Weighted fusion overcomes the redundancy and high dimensionality of the original parameter set, retains the most relevant sensitive information about the fault, ensures the dominance of important features through weighting, reduces the computational complexity of subsequent state analysis and fault identification, improves parameter processing speed and judgment accuracy, and enhances the performance and reliability of the parameter monitoring system.

[0125] Furthermore, based on the fused parameters, the drone's status is analyzed in real time, fault parameters are identified, and the fault propagation chain between parameters is filtered out through a pre-set causal analysis model, including:

[0126] S901: Analyze the UAV status in real time based on the fused parameters and filter out fault parameters;

[0127] S902. By using a preset causal analysis model, the fault parameters are traced back in time to calculate the propagation entropy between the parameters and filter out the fault propagation chain between the parameters.

[0128] In this embodiment, the UAV status is analyzed in real time based on fused parameters. A random forest model trained on a large amount of normal flight data is used to analyze the degree of anomaly of the current state corresponding to the real-time fused parameters. When the degree of anomaly exceeds a preset threshold, a system fault is determined, and one or more specific fault parameters with the highest degree of responsibility are selected from the second extended parameter set. Based on the fused parameters, system faults can be located to specific fault parameters, shortening fault identification time and improving fault analysis efficiency.

[0129] Specifically, using a pre-defined causal analysis model, starting with the fault parameter as the endpoint, the model traces backward along the timeline. Within a time window prior to the fault occurrence, the model calculates the propagation entropy values ​​of all other parameters causally related to the fault parameter in the second extended parameter set. By comparing the magnitude and temporal order of the propagation entropy of all parameters to the fault parameter, the most significant causal links are selected. These links are then connected according to chronological order and causal strength to construct the fault propagation chain. By tracing back to the root cause of the fault, the initial location of the anomaly in the system can be pinpointed, providing accurate decision-making basis for precise maintenance processes and improving operational efficiency and overall system reliability.

[0130] like Figure 4 As shown, a real-time monitoring system for UAV flight parameters supporting remote backtracking is used to implement a method for real-time monitoring of UAV flight parameters supporting remote backtracking, including:

[0131] The parameter expansion module expands the parameters based on the collected real-time flight parameters of the UAV, predicts the flight status change trend of the UAV, and selects the first expanded parameter set.

[0132] The correlation baseline construction module calculates the correlation between parameters based on the flight parameters obtained under normal flight conditions of the UAV and constructs the first correlation baseline.

[0133] The parameter filtering module clusters the real-time flight parameters of the UAV, filters out the core parameter set, calculates the correlation degree between each extended parameter in the first extended parameter set and the core parameter set, compares it with the first correlation degree baseline, calculates the correlation entropy of each extended parameter, filters out the extended parameters whose correlation entropy is greater than the preset entropy value, and obtains the second extended parameter set.

[0134] The parameter fusion module configures a hierarchical parameter fusion mechanism, calculates the priority of the corresponding extended parameters in the second extended parameter set, hierarchically divides the parameters, maps the parameters to the same space according to the corresponding level and fuses them to obtain fused parameters;

[0135] The parameter backtracking module analyzes the UAV status in real time based on the fused parameters, identifies fault parameters, and backtracks them in time. It uses a preset causal analysis model to filter out the fault propagation chain between parameters, so as to monitor and analyze the UAV flight parameters in real time.

[0136] In this embodiment, the parameter expansion module receives real-time flight parameters from the UAV, parses flight mission instructions, and uses a state analysis model to predict the changing trend of flight state. It dynamically expands and filters out the first expanded parameter set, predicts the expanded flight parameters in advance, and can capture early abnormal signs that contradict the flight intention in advance, improving the system's early warning capability and creating a time window for proactive intervention. The correlation baseline construction module processes historical parameters acquired by the UAV under normal flight conditions, divides the working condition slices according to flight altitude, speed, and mission mode, calculates the mutual information values ​​between parameters, and constructs a dynamic first correlation baseline. This provides a normal behavior benchmark for the system, enabling anomaly detection to adapt to complex multi-working-condition environments and improving the precision and reliability of the fault discrimination benchmark.

[0137] Specifically, the parameter screening module clusters real-time parameters to determine the core parameter set. It then calculates the correlation entropy between each parameter in the first extended parameter set and the core parameter set after time-delay compensation, filtering out parameters with excessive correlation entropy to form the second extended parameter set. This composite index of correlation entropy enables sensitive and robust detection of abnormal parameter correlations, improving the signal-to-noise ratio and efficiency of subsequent processing. The parameter fusion module evaluates the static, dynamic, and retrospective value of each parameter in the second extended parameter set using a parameter value analysis model. It calculates their priority and hierarchically categorizes them, mapping parameters at different levels to a high-dimensional space according to a specific strategy and fusing them into a single fused parameter. This value-oriented hierarchical fusion mechanism generates information-rich system state characteristic representations, providing accurate data support for anomaly diagnosis while optimizing the allocation of computing resources. The parameter retrospective module identifies fault parameters in real-time based on the fused parameters and, through a causal analysis model centered on propagation entropy, traces back along the timeline to filter out the fault propagation chain between parameters. This enables parameter tracing from fault phenomena to their root causes, quickly locating fault parameters, analyzing the causal origin and propagation path of the fault, and improving the accuracy and efficiency of fault diagnosis.

[0138] Example 2:

[0139] This embodiment describes the overall process of the technical solution of the present invention in the context of a specific application scenario of a logistics delivery drone. In this embodiment, the drone performs a cargo delivery mission from a regional distribution station to a residential community. The predetermined flight mission command sequence is as follows: autonomously take off from the warehouse platform, climb to a cruising altitude of 100 meters, cruise towards the target point at a speed of 15 meters per second, perform precise hovering upon arrival at the target point, then deliver the cargo, and finally return to base.

[0140] In this embodiment, the UAV flight control system collects raw flight parameters in real time, including airspeed, altitude, three-axis attitude angles, three-axis acceleration, GPS position, battery voltage, total current, motor speeds, and ESC temperature, and sends them to a remote monitoring server. The server initiates the monitoring method of this invention. During the initial takeoff and climb phase of the mission, the system identifies the current mission transition point from takeoff to cruise based on a predetermined flight mission command sequence. Using an LSTM state analysis model pre-trained with a large amount of normal takeoff data, the system predicts the expected state change trend of the UAV within the next 5 seconds as follows: altitude continuously increases, airspeed accelerates to 15 m / s, and pitch angle first becomes positive and then tends towards level flight. Based on this predicted trend, the system expands the parameter set. For example, it generates the parameter of the ratio of altitude change rate to average motor speed and increases the monitoring weight of parameters such as climb rate and motor load rate, forming a first extended parameter set containing raw parameters and derived parameters.

[0141] The server has pre-stored a first correlation baseline constructed from historical normal flight data. This baseline categorizes and stores the parameter correlation matrix according to different flight conditions. For example, in the cruise-high speed slice, the baseline indicates a high mutual information value between airspeed and total motor current, and a lower mutual information value with lateral acceleration. Simultaneously, the system performs cluster analysis on the currently received flight parameters. The DBSCAN algorithm clusters parameters such as airspeed, altitude, pitch angle, and main motor speed into a core parameter set, which represents the core flight state.

[0142] Once the UAV enters the stable cruise phase, the system begins in-depth analysis of each parameter in the first extended parameter set. Taking the newly introduced derived parameter, the ratio of altitude change rate to average motor speed, as an example, denoted as parameter P, the system analyzes its matching degree with the core parameter set. It finds that the change in parameter P lags behind the airspeed change by approximately 300 milliseconds, and performs time lag compensation accordingly. After compensation, the average mutual information value between parameter P and the core parameter set under the current cruise state is calculated and compared with the normal average mutual information value under the cruise-high-speed condition in the first correlation baseline. It is found that the mutual information change rate has decreased by approximately 40%. Simultaneously, the joint probability distribution of the mutual information change rate and the core parameters under the current data window is calculated and compared with the normal joint distribution in the baseline, yielding a high KL divergence, which is used as the mismatch degree. The mutual information change rate and mismatch degree are weighted and fused to obtain the correlation entropy of parameter P. Similarly, after calculating the correlation entropy of all extended parameters, parameters with correlation entropy exceeding a preset threshold are selected. If the correlation entropy of parameter P and the temperature of motor number three is significantly high, these parameters are selected to form the second extended parameter set, indicating that the normal correlation between these parameters and the core flight state has been disrupted.

[0143] Furthermore, a parameter hierarchical fusion mechanism is initiated, and the parameter value analysis model evaluates the parameters in the second extended parameter set from three dimensions: static, dynamic, and retrospective. For example, the temperature of motor 3 has high value in both static characterization and dynamic response, while parameter P has high value in dynamic response. Based on the current cruise mission mode, the model calculates that the temperature of motor 3 has the highest priority, followed by parameter P. Accordingly, the raw temperature data stream corresponding to the temperature of motor 3 is divided into a high-priority layer, and parameter P is divided into a medium-priority layer.

[0144] Specifically, a hierarchical mapping and fusion process is performed. For high-priority temperature data, its time-series spectral features are extracted, revealing a significant increase in the proportion of low-frequency fluctuation energy. Based on this, a first descriptor is constructed, and the original temperature values ​​are scaled and mapped to a first parameter vector. For the medium-priority parameter P, its deviation from the historical health benchmark is calculated, a second descriptor is constructed, and its mutual information value with the current parameter is combined to map it to a second parameter vector. Finally, the first and second parameter vectors are weighted and fused to generate a fused parameter vector representing the current abnormal state of the system.

[0145] Based on this fused parameter vector, the system's state classifier, specifically a pre-trained random forest model, identifies the current state as belonging to the motor system efficiency degradation fault category. Through feature importance backtracking, the fault is located to the specific temperature parameter of motor number three. At this point, the system does not merely issue an overheat alarm but automatically triggers causal backtracking analysis. Using the temperature of motor number three as the fault endpoint, the system traces back 30 seconds of data, using a transitive entropy causal analysis model to calculate the causal information flow from each candidate parameter to that temperature parameter. Analysis reveals that the link with significant transitive entropy and preceding time sequence is: abnormal fluctuation of the duty cycle of the motor number three drive signal (time t1) → increase in the harmonic components of the motor number three current (time t2) → accelerated temperature rise of motor number three (time t3). Thus, the system automatically filters and presents this clear fault propagation chain.

[0146] In this embodiment, remote maintenance personnel can not only receive early warnings of declining motor system performance before the drone's cockpit instruments display abnormalities, but also intuitively see a complete fault evolution report: the root cause of the fault is an abnormal ESC drive signal, leading to deterioration of motor current quality, ultimately manifesting as motor overheating. This allows ground personnel to immediately notify the drone to return safely ahead of schedule and accurately guide on-site maintenance personnel to replace the faulty ESC module, thus achieving a closed loop from intelligent early warning to precise diagnosis.

[0147] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A method for real-time monitoring of UAV flight parameters supporting remote backtracking, characterized in that, include: Based on the collected real-time flight parameters of the UAV and the preset flight mission commands, the mission transition points are identified, and the flight status change trend of the corresponding mission transition points is predicted through the preset state analysis model. Based on the flight state change trend, the parameters are expanded, and a first expanded parameter set is selected. Based on the flight parameters obtained under normal flight conditions of the UAV, the correlation between the parameters is calculated, and the first correlation baseline is constructed. Cluster the real-time flight parameters of the UAV and filter out the core parameter set; Analyze the degree of matching between each extended parameter in the first extended parameter set and the core parameter set, filter out the corresponding matching paths, and calculate the time lag value according to the matching paths; Time lag compensation is performed on the extended parameters according to the time lag value. The mutual information value between the compensated extended parameters and the core parameter set is calculated. The mutual information value is compared with the average mutual information value in the first correlation baseline to obtain the mutual information change rate. Analyze the joint probability distribution between each extended parameter and the core parameter, calculate the degree of mismatch in the joint behavior patterns between parameters, and obtain the degree of mismatch. The mutual information change rate and mismatch degree are weighted and fused to obtain the association entropy; The extended parameters with correlation entropy greater than the preset entropy value are selected to obtain the second extended parameter set; Configure a hierarchical fusion mechanism for parameters, calculate the priority of the corresponding extended parameters in the second extended parameter set, hierarchically classify the parameters, map the parameters to the same space according to the corresponding level and fuse them to obtain fused parameters; The system analyzes the UAV status in real time based on fused parameters, identifies fault parameters and traces them back in time. It also filters out the fault propagation chain between parameters through a preset causal analysis model, so as to monitor and analyze the UAV flight parameters in real time.

2. The method for real-time monitoring of UAV flight parameters supporting remote backtracking according to claim 1, characterized in that, The process of calculating the correlation between flight parameters obtained under normal UAV flight conditions and constructing a first correlation baseline includes: Based on the flight parameters obtained under normal flight conditions of the UAV, the flight altitude, speed and mission mode are analyzed to obtain different flight conditions, and the flight parameters are divided to obtain multiple parameter slices. For each parameter slice, analyze the parameter correlation, calculate the mutual information value between parameters, and construct the corresponding parameter correlation matrix; Based on the aforementioned parameter correlation matrix, construct the first correlation baseline.

3. The method for real-time monitoring of UAV flight parameters supporting remote backtracking according to claim 1, characterized in that, The parameter hierarchical fusion mechanism includes: Using a pre-defined parameter value analysis model, the comprehensive value of the corresponding extended parameters in the second extended parameter set is analyzed, the corresponding priority is calculated, and the parameters are stratified according to the priority. According to the corresponding level, the parameters are mapped to the same space and fused to obtain the fused parameters.

4. The method for real-time monitoring of UAV flight parameters supporting remote backtracking according to claim 3, characterized in that, The process involves analyzing the comprehensive value of corresponding extended parameters in the second extended parameter set using a preset parameter value analysis model, calculating their corresponding priorities, and stratifying the parameters according to their priorities, including: Using a pre-defined parameter value analysis model, the comprehensive value of the corresponding extended parameters in the second extended parameter set is analyzed and calculated from three dimensions: static representation, dynamic response, and backtracking location, and a parameter value matrix is ​​constructed. Based on the parameter value matrix, the influence of each extended parameter on the UAV state is analyzed, projected onto the corresponding flight mission space, and a parameter mission mapping matrix is ​​constructed. The parameter value matrix and the parameter task mapping matrix are fused together to calculate the parameter priority; The parameters corresponding to each extended parameter are stratified according to parameter priority.

5. The method for real-time monitoring of UAV flight parameters supporting remote backtracking according to claim 4, characterized in that, The process of mapping parameters to the same space according to corresponding levels and then fusing them to obtain fused parameters includes: According to the analysis of the temporal characteristics and spectral energy distribution of the extended parameters in each level, the energy proportion of the first priority level parameter after Fourier transform is calculated to construct the first descriptor, the deviation of the second priority level parameter from the reference parameter is calculated to construct the second descriptor, and the mutual information value between the third priority level parameter and the adjacent level parameter is calculated to construct the third descriptor. According to the first descriptor, the corresponding real-time parameters are scaled and mapped to the first parameter; according to the second descriptor and the mutual information value between the real-time parameters, the corresponding real-time parameters are mapped to the second parameter; according to the third descriptor, the corresponding real-time parameters are equidistantly mapped to obtain the third parameter. The first, second, and third parameters are weighted and fused to obtain the fused parameters.

6. The method for real-time monitoring of UAV flight parameters supporting remote backtracking according to claim 1, characterized in that, The process of analyzing the UAV status in real time based on fused parameters, identifying fault parameters and tracing back in time, and filtering out the fault propagation chain between parameters through a preset causal analysis model includes: The drone status is analyzed in real time based on the fused parameters, and fault parameters are filtered out. By using a pre-defined causal analysis model, the transmission entropy between parameters is calculated by tracing the fault parameters backward in time, and the fault propagation chain between parameters is then identified.

7. A real-time monitoring system for UAV flight parameters supporting remote backtracking, characterized in that, A method for real-time monitoring of UAV flight parameters supporting remote backtracking as described in any one of claims 1 to 6, comprising: The parameter expansion module expands the parameters based on the collected real-time flight parameters of the UAV, predicts the flight status change trend of the UAV, and selects the first expanded parameter set. The correlation baseline construction module calculates the correlation between parameters based on the flight parameters obtained under normal flight conditions of the UAV and constructs the first correlation baseline. The parameter filtering module clusters the real-time flight parameters of the UAV, filters out the core parameter set, calculates the correlation degree between each extended parameter in the first extended parameter set and the core parameter set, compares it with the first correlation degree baseline, calculates the correlation entropy of each extended parameter, filters out the extended parameters whose correlation entropy is greater than the preset entropy value, and obtains the second extended parameter set. The parameter fusion module configures a hierarchical parameter fusion mechanism, calculates the priority of the corresponding extended parameters in the second extended parameter set, hierarchically divides the parameters, maps the parameters to the same space according to the corresponding level and fuses them to obtain fused parameters; The parameter backtracking module analyzes the UAV status in real time based on the fused parameters, identifies fault parameters, and backtracks them in time. It uses a preset causal analysis model to filter out the fault propagation chain between parameters, so as to monitor and analyze the UAV flight parameters in real time.

Images