Unmanned aerial vehicle multi-dimensional risk accident tracing method based on dynamic network model

Abstract

The invention provides an unmanned aerial vehicle multi-dimensional risk accident tracing method based on a dynamic network model, and belongs to the technical field of unmanned aerial vehicle operation risk accident tracing analysis. Constructing a risk evolution network model; calculating model parameters; and risk accident traceability analysis. According to the invention, relevance among unmanned aerial vehicle operation risk factors is determined, and traceability analysis is carried out on various unmanned aerial vehicle risk accidents; the problem that a traditional tracing method lacks timeliness, accuracy and comprehensiveness is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of source tracing analysis of UAV operation risk accidents, and particularly to a multi-dimensional risk accident source tracing method for UAVs based on a dynamic network model. Background Art

[0002] With the rapid development and wide application of UAV technology, the operation safety of UAVs has gradually become the focus of attention. The risk factors in UAV operation are complex and diverse, involving multiple aspects such as operators, maintenance personnel, the performance of UAVs themselves, and the operation environment. The existing technologies mainly conduct risk analysis based on the statistical results after historical flight accidents, lacking in-depth analysis of the dynamic correlations over time between various risk factors under different environmental impacts, making it difficult to trace the root causes of accidents in real time and accurately, and unable to effectively prevent and control the occurrence of UAV risk accidents.

[0003] In view of the limitations of the existing technologies, the present invention proposes a multi-dimensional risk accident source tracing method for UAVs based on a dynamic network model. Summary of the Invention

[0004] The present invention provides a multi-dimensional risk accident source tracing method for UAVs based on a dynamic network model. This method provides an innovative solution for ensuring the safe and efficient operation of UAVs by comprehensively evaluating multi-dimensional parameters, dynamically analyzing operation risks in real time, and effectively tracing the risk propagation paths; aims to clarify the correlations between UAV operation risk factors and conduct source tracing analysis on various UAV risk accidents; and is used to solve the problems of lack of timeliness, accuracy, and comprehensiveness in traditional source tracing methods.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A multi-dimensional risk accident source tracing method for UAVs based on a dynamic network model, comprising: Risk factor identification: Analyze historical accident data, flight logs, and real-time operation environment data, and identify potential risk factors in the UAV operation process based on the three dimensions of human-machine-environment. The potential risk factors include human factors, UAV performance risk factors, and operation environment risk factors; Construct a risk evolution network model: Regard the potential risk factors as network nodes and the transmission relationships between risk factors as network edges to construct a UAV operation risk evolution network model; Model parameter calculation: Based on the UAV operation risk evolution network model, introduce a time node degree to calculate the node degree, introduce a risk propagation intensity to calculate the closeness centrality, calculate the betweenness centrality, introduce a dynamic attenuation factor to calculate the page rank value, and introduce an environmental risk factor to calculate the edge betweenness; Risk accident traceability analysis: Based on the dynamic changes of risk propagation over time, a risk loss coefficient is introduced, and various network characteristic indicators are comprehensively analyzed to evaluate the importance of nodes, determine the key paths of risk propagation, and extract the key risk propagation chains to obtain the analysis results.

[0006] In this specification, the formula for calculating the node degree by introducing the time node degree is as follows: ; In the formula represents the time node degree of node ; represents the total number of network nodes; represents whether there is a risk propagation edge between node and node at time . If it exists, the value is 1; if it does not exist, the value is 0.

[0007] In this specification, the formula for calculating the closeness centrality of risk propagation intensity is as follows: ; ; In the formula, is the number of edges contained in the shortest path starting from node and ending at node ; represents the total number of nodes; represents the risk propagation intensity from node to node .

[0008] In this specification, the formula for calculating the betweenness centrality is as follows: ; In the formula, represents the total number of shortest paths between node and node ; represents the number of shortest paths from node to node passing through node .

[0009] In this specification, the formula for calculating the page rank value by introducing the dynamic decay factor is as follows: ; ; ; In the formula: is the page rank value of the node, is the total number of nodes; is the initial attenuation factor, is the adjustment amplitude, is the time function, is the preset constant; is connected to the node; is the edge weight of; is the degree of the node.

[0010] In this specification, the formula for introducing the environmental risk factor to calculate the betweenness centrality is as follows: ; In the formula is the betweenness centrality of; is and the environmental risk factors of; is the number of paths passing through the edge between the node and the node .

[0011] In this specification, the risk loss coefficient is introduced, and the formula for comprehensively analyzing various network characteristic indicators is as follows: ; In the formula, is the risk loss function, , are respectively the time node degree, closeness centrality, betweenness centrality, PageRank value, and betweenness centrality of the node ; are respectively the weight coefficients of the time node degree, closeness centrality, betweenness centrality, PageRank value, and betweenness centrality, and the range is [0,1], and ; is the risk loss coefficient.

[0012] In this specification, the LSTM network is used to predict the time function ; ; Its hidden state is updated by the formula: ; where is the Sigmoid function, is the weight matrix from the LSTM hidden state to the output, is the bias term, is the input feature within the time window.

[0013] In this specification, constructing a risk evolution network model includes: Establishing a hierarchical network structure, including a human factor layer, an airborne layer, and an environmental layer; Using a hypergraph to represent the cross-layer risk propagation path, and the hyperedge weight is calculated by the following formula: ; is a dynamic learning parameter, and is an environmental risk factor.

[0014] In this specification, Kalman filtering is used to fuse multi-sensor data to calculate the real-time environmental risk factor ; ; wherein, is the normalized reading of the th sensor, is the temperature parameter, is the normalization coefficient.

[0015] In summary, the present invention has at least the following beneficial effects: The present invention can trace the root cause of UAV accidents more accurately and in real time, and can be used to prevent and control the occurrence of UAV risk accidents. Brief Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 is a schematic diagram of the UAV multi-dimensional risk accident traceability method based on a dynamic network model involved in the present invention.

[0018] Figure 2 is a schematic diagram of the UAV operation risk evolution network model involved in the present invention.

[0019] Figure 3 is a schematic diagram of the hierarchical structure division of UAV operation risk factors involved in the present invention. Detailed Embodiments

[0020] In the following text, only some exemplary embodiments are briefly described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the embodiments of the present invention. Therefore, the drawings and the description are considered to be exemplary in nature rather than restrictive.

[0021] The following disclosure provides many different embodiments or examples for implementing different structures of the embodiments of the present invention. To simplify the disclosure of the embodiments of the present invention, the components and settings of specific examples are described below. Of course, they are only examples and are not intended to limit the embodiments of the present invention. In addition, the embodiments of the present invention may repeat reference numerals and / or reference letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed.

[0022] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] As Figure 1 shown, this embodiment provides a multi-dimensional risk accident tracing method for unmanned aerial vehicles based on a dynamic network model, including: Risk factor identification: Analyze historical accident data, flight logs, and real-time operation environment data, and identify potential risk factors during the operation of unmanned aerial vehicles based on the three dimensions of human-machine-environment. The potential risk factors include human factors, unmanned aerial vehicle's own performance risk factors, and operation environment risk factors; Construct a risk evolution network model: Regard the potential risk factors as network nodes and the transmission relationship between risk factors as network edges to construct a risk evolution network model for the operation of unmanned aerial vehicles; Model parameter calculation: Based on the risk evolution network model of unmanned aerial vehicle operation, introduce the time node degree theory to calculate the node degree, introduce the risk propagation intensity to calculate the closeness centrality, calculate the betweenness centrality, introduce the dynamic decay factor to calculate the page rank value, and introduce the environmental risk factor to calculate the edge betweenness; Risk accident tracing analysis: Based on the dynamic change of risk propagation over time, introduce the risk loss coefficient, comprehensively analyze various network characteristic indicators, evaluate the importance of nodes, determine the key path of risk propagation, and extract the key risk propagation chain to obtain the analysis result.

[0024] In some embodiments, the formula for calculating the node degree by introducing the time node degree is as follows: ; In the formula represents the time node degree of node ; represents the total number of network nodes; represents at time node Whether there is a risk propagation edge between and

[0025] is determined as follows: if it exists, the value is 1; if it does not exist, the value is 0. By using 0 and 1, the node degree calculation process is split to reduce the overall calculation workload (usually, all nodes are calculated, and by splitting, the calculation of some unconnected nodes can be reduced). In some embodiments, the formula for calculating the risk propagation intensity close to centrality is as follows: In the formula, starting from node and ending at node is the number of edges contained in the shortest path; represents the total number of nodes; represents the risk propagation intensity from node to node ; is the number of times the path from node to appears in the historical accident data. The higher the frequency of occurrence, the greater the risk propagation intensity.

[0026] The risk propagation intensity represents the reliability degree of the risk propagation path from node to node in the historical data (such as UAV historical accident reports, sensor monitoring logs, simulation test data, etc.), reflecting the stability, repeatability, and risk conduction efficiency of this path in risk propagation. It means that, for example, the fewer the number of edges in the shortest path and the greater the risk propagation intensity , the faster the risk propagation speed from node to . The greater the value of the closeness centrality , the higher the importance of the node.

[0027] In some embodiments, the formula for calculating the betweenness centrality is as follows: ; In the formula, represents the total number of shortest paths between node and node ; represents the number of shortest paths from node to node passing through node .

[0028] In some embodiments, a dynamic attenuation factor is introduced The formula for calculating the page rank value is as follows: ; ; ; In the formula: is the page rank value of the node, is the total number of nodes; is the initial attenuation factor, is the adjustment amplitude, is a time function, is a preset constant, usually taking the value of 0.1; is the node connected to ; is 's edge weight; is the degree of the node; The size depends on the risk propagation probability of the edge , that is, after the node occurs, the conditional probability that the node will occur accordingly; ; Among them, is the number of times that the nodes and occur simultaneously, is the number of times that the node occurs alone.

[0029] In some embodiments, the formula for introducing an environmental risk factor to calculate the edge betweenness is as follows: ; In the formula is the edge betweenness of ; is the environmental risk factor of and ; is the number of paths passing through the edge between the node and the node .

[0030] In some embodiments, a risk loss coefficient is introduced, and the formula for comprehensively analyzing various network characteristic indicators is as follows: ; In the formula, is the risk loss function, , are the nodes respectively The time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness; are the weight coefficients of the time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness, respectively, and the range is [0, 1] for all, and ; is the risk loss coefficient.

[0031] Among them, represents the comprehensive loss degree that may be caused once the node (risk factor or risk event) occurs or is triggered, including personal injury, property loss, social impact, and economic cost, etc. According to the relevant regulations and standards of unmanned aerial vehicles, the accident level is divided into intact, slightly damaged, severely damaged, and completely lost. Therefore, in the present invention, the accident level is directly mapped to the value of the risk loss coefficient . In the future, when the amount of historical accident data is sufficient, the accident level can be further classified and graded in detail to map different risk loss coefficient values.

[0032]

[0033] In some embodiments, an LSTM network is used to predict the time function ; ; Its hidden state The update formula is: ; Among them is the Sigmoid function, is the weight matrix from the LSTM hidden state to the output, is the bias term, is the input feature within the time window.

[0034] In some embodiments, constructing a risk evolution network model includes: Establishing a hierarchical network structure, including a human factor layer, an airborne layer, and an environmental layer; Using a hypergraph to represent the cross-layer risk propagation path, and the hyperedge weight is calculated by the following formula: ; is a dynamic learning parameter, is the environmental risk factor.

[0035] In some embodiments, Kalman filtering is used to fuse multi-sensor data to calculate the real-time environmental risk factor ; ; Among them, is the normalized reading of the th sensor, is the temperature parameter, is the normalization coefficient.

[0036] The technical concept of the present invention is as follows: Specifically, analyze historical accident data, flight logs and real-time operation environment data, identify potential risk factors in UAV operation based on the three dimensions of "human - machine - environment", and construct a risk evolution network model for UAV operation. Considering the dynamic change of risk propagation over time, comprehensively evaluate the importance of risk nodes by improving parameters such as calculating node degree, closeness centrality, betweenness centrality and PageRank (PR) algorithm, etc.; meanwhile, evaluate risk edges based on edge betweenness under different environmental risks to determine the key risk propagation paths in different environments. The present invention introduces a comprehensive evaluation method based on risk loss coefficient to quantitatively analyze the influence of node degree, closeness centrality, betweenness centrality, PageRank (PR) algorithm and edge betweenness on UAV operation, and determine key risk nodes and key propagation paths.

[0037] The specific steps are as follows: (1) Risk factor identification Analyze historical accident data, flight logs and real-time operation environment data, and identify potential risk factors in the process of UAV operation based on the three dimensions of "human - machine - environment", including human factors, UAV own performance risk factors and operation environment risk factors.

[0038] (2) Construct a risk evolution network model Regard the identified potential risk factors as network nodes and the transfer relationship between risk factors as network edges to construct a risk evolution network model for UAV operation.

[0039] (3) Model parameter calculation Considering the dynamic change of risk propagation over time, comprehensively evaluate the importance and randomness of risk nodes by improving parameters such as calculating node degree, closeness centrality, betweenness centrality and PageRank ( PR ) algorithm, etc.; meanwhile, evaluate risk edges based on edge betweenness under different environmental risks to determine the key risk propagation paths in different environments.

[0040] (4) Risk accident traceability analysis Based on the dynamic change of risk propagation over time, the present invention introduces a time decay factor, an environmental risk change factor and a risk loss coefficient, comprehensively analyzes various network characteristic indexes, evaluates the importance of nodes, determines the key risk propagation paths and extracts key risk propagation chains.

[0041] Example 1: Risk factor identification Identify potential risk factors during the operation of UAVs from three dimensions of "human - machine - environment" as follows: Human factors include inspection and maintenance errors, command center dispatching errors, flight dynamic monitoring deviations, dangerous goods loading, illegal interference, aircraft operation errors, insufficient operator experience, failure to comply with standard operating procedures, flight path planning defects, etc.

[0042] The performance factors of UAVs themselves include electronic instrument failures, flight program failures, airframe damage, power system failures, communication system failures, navigation and positioning system failures, flight control system failures, power supply system failures, insufficient battery endurance, battery overheating, mechanical component aging, design and manufacturing defects, excessive total flight hours of the aircraft, immature aircraft product technology, overloaded aircraft, aircraft emergency system failures, etc.

[0043] Operating environment factors include bird strikes, strong winds / gusts / wind shear, thunderstorms, showers, icing, magnetic field interference, aerial obstacles, high environmental temperature, low visibility, terrain obstacles, flight conflict risks, etc.

[0044] Example 2: Construct a risk evolution network model Regard the above - mentioned risk factors as network nodes and the transfer relationships between risk factors as network edges to construct a risk evolution network model for UAV operation, as Figure 2 shown. Figure 2 In it, each time the risk is transferred, the edge will become thickened. The thickness of the edge represents the number of times of risk transfer, that is, the weight of the edge. The thicker the edge, the more types of risks are transferred during the risk transfer process.

[0045] The risk evolution network model for UAV operation contains 44 nodes and 335 edges. Among them, the nodes are divided into human factor nodes (A1 - A9), UAV performance nodes (B1 - B16), operating environment nodes (C1 - C8), and 12 risk event nodes (D1 - D12), as shown in Table 1.

[0046] Table 1. Risk factors for UAV operation node risk factor node risk factor A1 inspection and maintenance error B15 overloaded aircraft A2 dispatching error of command center B16 failure of aircraft emergency system A3 deviation of flight dynamic monitoring C1 bird strike A4 hazardous goods loading C2 strong wind / gust / wind shear, thunderstorm, shower, icing A5 illegal interference C3 magnetic field interference A6 aircraft operation error C4 aerial obstacle A7 insufficient operator experience C5 too high ambient temperature A8 failure to comply with standard operating procedures C6 low visibility A9 defect in flight path planning C7 terrain obstacle B1 failure of electronic instrument C8 risk of flight conflict B2 failure of flight procedure D1 aircraft stall B3 airframe damage D2 aircraft yaw B4 failure of power system D3 aircraft out of control B5 failure of communication system D4 fire B6 failure of navigation and positioning system D5 explosion B7 failure of flight control system D6 mid-air collision B8 failure of power supply system D7 forced landing B9 insufficient battery endurance D8 mid-air disintegration B10 battery overheat D9 ground collision B11 aging of mechanical components D10 plane crash B12 design and manufacturing defect D11 injury to third-party personnel B13 too high total flight hours of aircraft D12 property loss of third party B14 immaturity of aircraft product technology Use the ISM (Interpretive Structural Modeling) to divide the risk factors into a hierarchical structure. According to the degree of influence of different risk factors on risk accidents, it is divided into five levels: root factors (the fifth level), structural factors (the fourth level), indirect factors (the third level), direct factors (the second level), and risk events (the first level), which can more directly understand the relationship between each risk factor and risk events. For example, from the perspective of risk prevention, the connection between direct factors and risk events can be disconnected targeted to achieve the effect of risk prevention. As Figure 3 shown.

[0047] Example 3: Model Parameter Calculation Node Degree Calculation: The degree of a node is an important metric. Nodes with higher degrees are considered more important in the network. Due to the characteristic that risk changes over time, the time node degree is introduced . Therefore, for node , its degree is expressed as follows, and the calculation is shown in the following formula: ; In the formula represents the time node degree of node ; represents the total number of network nodes. represents whether there is a risk propagation edge between node and node at time . For example, during the flight of a drone, as time goes by, the battery power gradually decreases, the risk of battery failure increases, and the risk propagation path related to the battery may change dynamically, which can reflect the dynamic characteristics of the risk propagation network in real time.

[0048] Closeness Centrality Calculation: Closeness centrality represents the average distance between a node in the network and other nodes. The larger the value of a node's closeness centrality, the higher the importance of the node. Nodes with high closeness centrality can establish connections with other nodes in the network faster, and the risk propagation speed on these nodes is relatively fast. The risk propagation intensity is introduced. Paths with high risk propagation intensity have a greater impact on the closeness centrality of nodes, and can more accurately evaluate the importance of nodes in risk propagation. The calculation is shown in the following formula: ; In the formula, is the number of edges contained in the shortest path starting from node and ending at node ; represents the total number of nodes; represents the risk propagation intensity from node to node .

[0049] Betweenness Centrality Calculation: Betweenness centrality measures the proportion of all shortest paths in the network passing through a certain node. When risk propagates, it often passes through these key nodes between different subgroups. The calculation is shown in the following formula: ; In the formula, represents the total number of shortest paths between node and node ; represents the number of shortest paths from node to node The number of shortest paths passing through the node .

[0050] PageRank (PR) algorithm calculation: Using the PageRank (PR) algorithm to simulate random walks on the network graph, at each step, it randomly jumps from the current node to the next node along the directed edge. By calculating the stationary probability of this random walk visiting each node over a long period of time, the PR value of each node in the network model can be obtained. To improve the simulation accuracy, a dynamic decay factor is introduced, which represents the probability of a certain node continuing to traverse backward at any given moment. The calculation is shown in the following formula: ; ; ; In the formula: is the PR value of the node, is the total number of nodes; is the initial decay factor, is the adjustment amplitude, usually taking the value of 0.25, is a time function, is a preset constant, usually taking the value of 0.1; for example, during the intensive flight mission stage when the risk spreads quickly, can be appropriately reduced to highlight the impact of recent risk propagation; is the node connected to; is edge weight; is degree of the node.

[0051] Edge betweenness calculation: An environmental risk factor is introduced in the edge betweenness calculation to consider the impact of environmental factors on the risk propagation path. It is determined according to the specific environmental conditions (such as weather, magnetic field, etc.) on the degree of influence of risk propagation passing through the edge from node to . In bad weather, some risk propagation paths are more affected. Through this calculation formula, key risk edges can be more accurately identified. The calculation is shown in the following formula: ; In the formula is the edge betweenness of; is the environmental risk factor; is the number of paths passing through the edge between node and node .

[0052] Example 4: Hierarchical Division of Risk Factors Construct an adjacency matrix: An adjacency matrix is a matrix used to describe the relationships between various influencing factors. Determine the relationships between various influencing factors and construct an adjacency matrix A= ( bij ) n×n , and represent the logical relationships between various factors in matrix form, where n = 44. According to the rules of the adjacency matrix, the relationship between two factors is represented by the matrix element bij : ; Reachability matrix calculation: The reachability matrix represents the connection channels between a certain factor in the system and other factors, and is a data interpretation of the hierarchical structure model. The calculation of the reachability matrix needs to satisfy Boolean operations, where is the identity matrix: ; Hierarchical calculation: Hierarchical division requires factor extraction and hierarchical construction based on the decomposition results. All factors are divided into several levels according to the reachability matrix to construct the hierarchical structure of the influencing factors.

[0053] First step, the set of elements corresponding to the columns where all matrix elements in the th row of the reachability matrix are 1 is defined as the reachable set ; Second step, the set of elements corresponding to the rows where all matrix elements in the th row of the reachability matrix are 1 is defined as the antecedent set ; Third step, define the common set of the reachable set and the antecedent set . When the condition of is met, the corresponding factor level extraction is performed. Finally, the UAV operation risks are divided into five levels: root factors, structural factors, indirect factors, direct factors, and risk events.

[0054] Example 5: Traceability Analysis of Risk Accidents By comprehensively analyzing various network characteristic indicators, deeply explore the key transmission process between system risk factors. According to five important evaluation indicators: node degree, closeness centrality, betweenness centrality, PageRank ( PageRank, PR ) algorithm, and edge betweenness, judge the key risk nodes and key risk edges, extract the key risk propagation chain, and trace the root cause of the accident.

[0055] By tracing the root cause of accidents in real time and accurately, the occurrence of similar UAV risk accidents can be effectively avoided, achieving the goals of risk prevention and risk control. For example, if a certain type of UAV accident occurs frequently, through root cause analysis, it is found that the fundamental reason is that the battery product process of this type of UAV does not meet the requirements, resulting in frequent crashes of the UAV and causing losses to personnel and property. For the risk nodes related to battery quality, measures such as choosing to replace other mature brand UAVs to complete the flight mission and requiring the manufacturer to upgrade the battery product can be taken to prevent such accidents from happening again. That is, tracing the source is the basis of prevention, and the two are in a progressive relationship rather than a contradiction.

[0056] Comprehensively analyze each propagation path in the key risk propagation chain, and clarify the transmission mechanism and influencing factors of risks between different nodes. To make the comprehensive assessment pay more attention to serious risks, a risk loss coefficient is introduced , which is determined by the losses caused by different risk events. The risk loss function is shown in the following formula: ; In the formula, is the risk loss function, , are the time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness of node respectively; are the weight coefficients of the time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness respectively, and their ranges are all [0, 1], and ; is the risk loss coefficient; can be determined by the analytic hierarchy process combined with expert experience. The core role of the risk loss function is to identify the comprehensive risk contribution degree of each node in the UAV operation risk network through quantitative analysis, and achieve the identification of key risk nodes, propagation paths and the accurate tracing of risk accidents. That is, the risk loss function is the core hub of the technical chain of "multi-dimensional risk dynamic modeling → real-time quantitative assessment → accurate tracing decision-making" of the present invention. Its essence is to transform the complex UAV operation risk system into computable and interpretable quantitative indicators through mathematical modeling, solving the problems of strong subjectivity and poor timeliness in traditional risk assessment methods, and providing a data-driven scientific basis for the safe operation of UAVs.

[0057] Example 6: Adaptive adjustment of dynamic parameters By introducing a dynamic adjustment mechanism, the model can automatically adjust the weights of nodes and edges in the network according to real-time flight data and environmental changes, so as to better adapt to the dynamic changes during the operation of the UAV and improve the accuracy and timeliness of tracing risk accidents. At the same time, the model automatically optimizes the parameter calculation method during continuous operation, adjusts the algorithm parameters according to the actual risk assessment results, improves the accuracy and adaptability of parameter calculation, and provides a more reliable basis for tracing risk accidents.

[0058] (I) Prediction of Dynamic Risk Decay Coefficient Based on LSTM Use the LSTM network to perform time series modeling on the degradation data of key components such as UAV batteries and motors, and predict the dynamic risk decay coefficient at different time points. The LSTM network can capture long-term dependencies in time series data and is suitable for dealing with dynamic changes during UAV operation.

[0059] 1. LSTM Network Structure The LSTM network consists of an input gate, a forget gate, and an output gate, and can effectively handle long-term dependency problems in time series data. Its basic unit structure is as follows: ; Among them, and are the activation values of the input gate, forget gate, and output gate respectively; is the value of the candidate memory cell; is the memory cell at the current time; is the hidden state at the current time; is the weight matrix; is the bias term; is the Sigmoid activation function; is the hyperbolic tangent activation function; is the input feature at the current time; is the hidden state at the previous time; is the memory cell at the previous time.

[0060] 2. Prediction Formula for Dynamic Risk Decay Coefficient The prediction formula for the dynamic risk decay coefficient based on the LSTM network is as follows: ; ; Among them, is the cumulative time of risk propagation (unit: second); is the dynamic risk decay coefficient; is the weight matrix from the LSTM hidden state to the output; is the bias term; weight matrix and the bias term The initial values are combined with the physical meaning of UAV risk propagation, and the initial weights are set by the method of expert scoring to ensure that the model conforms to the prior logic; is the Sigmoid function; is the hidden state of LSTM at time.

[0061] 3. Input Features and Output Labels The input feature is the timestamp of historical risk data, and the auxiliary input features are yaw distance, wind speed, battery voltage, temperature, charge-discharge times, ambient temperature, etc. The output is the risk attenuation coefficient at the next moment .

[0062] The attenuation coefficient is defined to describe the characteristic that the risk gradually weakens over time, and the influence of risk propagation will decrease over time with the increase. The Sigmoid function adopted in the present invention , has a value range of (0, 1). By adjusting the value of the bias term , the risk attenuation coefficient is made to satisfy when approaching 0, (the risk is not attenuated); when increases, gradually approaches 0 (the risk is attenuated to near disappearance).

[0063] For example, a minor fault that occurred early caused the battery to overheat slightly but did not trigger subsequent accidents, and its contribution to the current risk will decrease over time. Let , then: seconds, ; when seconds; , the risk is basically attenuated.

[0064] 1) Data Preprocessing Step 1: Collect time series data Collect the timestamps and auxiliary features (wind speed, yaw distance, battery power) of historical risk events, etc., such as predicting the attenuation coefficient of the yaw risk caused by strong wind.

[0065]

[0066] Step 2: Construct input-label pairs Sliding window: Set the window size to n, and use the continuous n time-step data as the input and the data at the next time step as the label.

[0067] Input features: ; Output label: ; For example, input features: [duration of strong wind, yaw distance]; label: at the next moment . Suppose (the weight of the duration of strong wind is positive, and the weight of the yaw distance is negative), , the weight matrix and the bias term are continuously optimized through model training, so that fits the true decay curve, then there is .

[0068] When the duration of strong wind is longer, is larger, the risk decay is slow, the yaw distance is smaller, is smaller, and the risk has decayed.

[0069] Step 3: Data normalization Normalize the timestamps of historical risk times and auxiliary features (yaw distance, wind speed, battery power, etc.) and scale them to the range [0, 1] to avoid slow model convergence caused by differences in feature value ranges and improve training efficiency.

[0070] (II) Real-time prediction of environmental risk factors based on reinforcement learning Build a reinforcement learning model (RL) to dynamically adjust the edge betweenness weight according to real-time environmental data (wind speed, magnetic field strength, precipitation level). The reinforcement learning model can dynamically adjust the strategy according to environmental feedback to adapt to real-time changing environmental conditions.

[0071] 1. Reinforcement learning model architecture Adopt an Actor-Critic architecture based on PPO (Proximal Policy Optimization), where the Actor network outputs actions and the Critic network evaluates state values.

[0072] State space: current environmental parameters (wind speed, magnetic field strength, precipitation level) + UAV flight state (altitude, speed).

[0073] Action space: adjustment increment for adjusting the edge betweenness weight Δw , with the range limited to [-0.1, 0.1] to avoid mutations.

[0074] Reward function: accuracy of risk propagation path prediction (verified by historical accident data).

[0075] 2. Reinforcement learning update formula The update formula of the reinforcement learning model is as follows: ; ; Among them, are the parameters of the Actor network; is the learning rate, adjusted according to the convergence speed, with a value of 0.01. is the gradient of the parameter ; is the probability of taking action in state ; is the estimated value of the advantage function; is the value estimation of the state by the Critic network; is the reward obtained after taking action in state ; is the discount factor; is the new state transferred to after executing action ;

[0076] 3. Real-time prediction formula for environmental risk factors The real-time prediction formula for environmental risk factors based on the reinforcement learning model is as follows: ; where is the dimension of the state space, is the dimension of the action space, is the reward function, is the wind speed; is the magnetic field strength; is the precipitation level; is the UAV altitude; is the UAV speed; is the remaining battery power; is the adjustment increment of the betweenness weight; is the accuracy reward; is the safety reward; is the stability penalty.

[0077] (III) Parameter adaptive adjustment based on feedback optimization The model automatically optimizes the parameter calculation method during continuous operation, and adjusts the algorithm parameters according to the feedback of the actual risk assessment results to improve the accuracy and adaptability of parameter calculation.

[0078] 1. Feedback optimization mechanism The core of the feedback optimization mechanism is to compare the risk assessment results with the actual accident data, calculate the assessment error, and dynamically adjust the model parameters according to the error.

[0079] 2. Parameter adaptive adjustment The parameter adaptive adjustment formula is as follows: ; wherein, is the evaluation error, is the risk value predicted by the model; is the actual risk value; are the old model parameters; are the updated model parameters; .

[0080] 3. Dynamically adjust the weights of nodes and edges According to real-time flight data and environmental changes, dynamically adjust the weights of nodes and edges in the network so that the model can better adapt to the dynamic changes during the operation of the UAV.

[0081] ; wherein, is the old weight of node ; is the new weight of node ; is the dynamic risk attenuation coefficient; is the node weight adjustment coefficient; is the old weight of the edge; is the new weight of the edge; is the adjustment increment of the edge betweenness weight; is the edge weight adjustment coefficient.

[0082] (IV) Technical effects Through the above dynamic adjustment mechanism, the model can automatically adjust the weights of nodes and edges in the network according to real-time flight data and environmental changes, improving the accuracy and timeliness of risk accident traceability. At the same time, the model automatically optimizes the parameter calculation method during continuous operation, and adjusts the algorithm parameters according to the actual risk assessment results, improving the accuracy and adaptability of parameter calculation. The specific technical effects are as follows: 1. From experience-driven to data-driven: The adjustment of dynamic parameters changes from experience-driven to data-driven, improving the self-adaptability of the model to different scenarios.

[0083] 2. Reduce manual intervention: Reduce manual intervention and respond to sudden environmental changes (such as strong wind attacks) in real time.

[0084] 3. Reduce errors: LSTM+RL dynamic prediction effectively reduces errors.

[0085] 4. Improve the risk path coverage rate: Hypergraph modeling breaks through the limitations of traditional single-layer networks, supports multi-hop risk propagation analysis, and improves the risk path coverage rate.

[0086] 5. Reduce subjective bias: Eliminate the subjective bias of manually defined nodes / edges and improve the objectivity of the model.

[0087] 6. Improve the accuracy of risk perception: The quantification of environmental risk factors has objective criteria, reducing the dependence on subjective experience. The fusion of multi-sensor data improves the accuracy of risk perception in complex environments.

[0088] Example 7: Multilayer network modeling and cross-dimensional risk propagation Hierarchical network structure design: Human factor layer: The nodes are operator behaviors (false touch, instruction delay), ground station status; the edges are the propagation effects of operation errors on the instruction link.

[0089] Airborne layer: The nodes are UAV hardware (battery, motor, sensor), software modules; the edges are the fault transfer relationships (such as insufficient battery voltage causing the motor to stop rotating).

[0090] Environmental layer: The nodes are weather (wind speed, precipitation), electromagnetic interference sources, geographical obstacles; the edges are the influence paths of environmental factors on the UAV.

[0091] Cross-layer interaction modeling: Use a hypergraph to represent cross-layer risk propagation, for example: Hyperedge 1: Operator false touch (human factor layer) → Instruction delay (airborne layer) → Obstacle avoidance failure (environmental layer).

[0092] Define cross-layer weights ; , and optimize it through the gradient descent method, with the initial value (based on historical accident data statistics).

[0093] Enhance dynamic learning parameters Belongs to the cross-layer interaction weight coefficient and is updated through Q-learning: ; is the learning rate, adjusted according to the convergence speed, with a value of 0.01. is the immediate reward calculated based on real environment feedback or historical data, reflecting the effect of the current action in the real scenario, and the calculation method: ; Example: When the UAV triggers an obstacle avoidance action and successfully avoids obstacles under the magnetic field intensity , then (no event but risk exists).

[0094] If the obstacle avoidance is not triggered and a collision occurs, then .

[0095] is the estimated value of the expected reward for the current action of the model, which is output by the value function or the Critic network in Q-learning. The calculation method is: ; ; where is the expected value of taking action in state . is the value function (output of the Critic network) of state .

[0096] Example: If the model predicts that the value for triggering obstacle avoidance under the magnetic field intensity is , then .

[0097] Technical effect: Accurately identify cross-dimensional compound risks (such as the crash path caused by the superposition of "operation error + battery aging + strong wind"). The hypergraph modeling breaks through the limitations of traditional single-layer networks and supports multi-hop risk propagation analysis. The hypergraph cross-layer modeling improves the coverage rate of risk paths.

[0098] Example 8: Data-driven network structure optimization Causal network construction: Based on historical accident data, the PC algorithm (Peter-Clark) is used to automatically deduce the causal relationships between nodes from the data, replacing 335 manually defined edges.

[0099] Node relationship correction rule: If , then it is considered that there is a causal edge . 2.0 is an empirical value, which can be adjusted through statistical significance tests (such as p <0.05).

[0100] Dynamic topology update mechanism: Introduce a graph neural network (GNN) to update the network structure in real time: Node embedding: ; is the embedding vector of node in the th layer, is the embedding vector of node in the th layer, is the neighbor set of node , is a trainable weight matrix, is node in the Embedding vector of the layer; is the vector concatenation operation; Edge weight update: is the embedding vector of the node, is the embedding vector of the node, is the multi-layer perceptron; Recalculate the network topology according to the sensor data every 30 seconds; Graph Neural Network (GNN) parameters: Node embedding dimension: l = 128, representing the hidden space dimension of node features, the node embedding dimension l The larger it is, the richer the node feature information that the model can capture; Weight matrix W∈Rl×l : Learned through end-to-end training; Update period: 30 seconds, set according to the UAV sensor sampling frequency (≥2Hz).

[0101] Technical effect: Eliminate the subjective deviation of manually defined nodes / edges, and improve the objectivity of the model. The dynamic topology adapts to the changes in the UAV state (such as cutting off the connections of non-critical nodes when the battery is exhausted). PC algorithm + GNN dynamic topology, reducing subjective deviation.

[0102] Example 9: Quantification improvement of environmental risk factors Environmental risk database construction: Establish an environmental parameter-risk weight mapping table:

[0103] Definition: The influence intensity of environmental parameters on the risk propagation path, mapped by the look-up table method:

[0104] Sensor fusion correction: Use Kalman filter to fuse multi-sensor data (barometer, magnetometer, GPS) and calculate the environmental risk factor in real time: .

[0105] Temperature parameter T : Controls the smoothness of the weight distribution, default T = 1.0, high temperature ( T > 1) enhances the influence of small weights, and low temperature ( T < 1) highlights the dominant sensor.

[0106] Technical effect: The quantification of environmental risk factors has an objective standard, reducing the dependence on subjective experience. The fusion of multi-sensor data improves the risk perception accuracy in complex environments.

[0107] Example 10: Scenario: When the UAV is performing an inspection mission in a strong magnetic field area, the battery voltage suddenly drops. Dynamic parameter adjustment: LSTM detects an abnormal voltage drop rate and triggers the RL model to adjust the environmental weight (The magnetic field risk coefficient rises to 0.8).

[0108] Multi-layer network warning: Hypergraph identifies cross-layer paths: Battery failure (onboard layer) → Navigation signal loss (environmental layer) → Tower collision risk (human factor layer).

[0109] Data-driven decision-making: GNN updates the network topology, cuts off the power supply of unnecessary modules, and gives priority to ensuring the operation of the obstacle avoidance system.

[0110] Example 11: Adaptive adjustment of dynamic parameters Data preparation: Collect historical data of UAV battery voltage, temperature, and environmental wind speed, time window n = 10; LSTM training: Construct a two-layer LSTM network (hidden layer dimension l = 64), input sequence length , output ; Real-time prediction: Deploy the model to the UAV edge computing module and update it every 5 seconds ; Effect verification: In the scenario of battery aging, the risk warning time of the optimized model is 120 seconds earlier than that of the traditional method.

[0111] Example 12: Cross-layer hypergraph modeling Hierarchical definition: Human factor layer nodes: Operator instruction delay, interface accidental touch; Onboard layer nodes: Battery voltage, motor speed, GPS module; Environmental layer nodes: Wind speed, magnetic field intensity, visibility; Hyperedge construction: Define composite risk paths, for example: Hyperedge 1 (operation delay → motor speed decrease → strong wind interference → crash risk), initial weight ; Dynamic update: When the magnetic field intensity exceeds 50 μT, automatically increase the weight of the relevant hyperedge to .

[0112] Example 13: Quantification of environmental factors Sensor calibration: Set the magnetic field meter range to , normalize ; Weight calculation: In a strong magnetic field environment In, ; Risk integration: If there is wind speed , the total environmental risk weight is 0.3 + 0.9 = 1.2 (an emergency obstacle avoidance is triggered when it exceeds the threshold of 1.0).

[0113] Example 14: is the LSTM predicted value Input data: Time window The battery voltage sequence within: ; The current environmental risk factor .

[0114] LSTM prediction : The LSTM output hidden state ; The Sigmoid function is defined as: ; Calculate ; Among them, ; ; Therefore, 。

[0115] Calculate the cross-layer weight: ; If , a high-risk warning is triggered.

[0116] Example 15: The specific process of the reinforcement learning (RL) modeling of the environmental risk factor 1. Selection and architecture of the reinforcement learning model: Model type: Actor-Critic architecture based on PPO (Proximal Policy Optimization); Basis for selection: PPO performs stably in continuous action spaces and high-dimensional state spaces and is suitable for dynamic adjustment . Eenv .

[0117] Network structure:

[0118] State space: .

[0119] Environmental parameters : Wind speed , magnetic field intensity [0, 100] μT, precipitation level (no rain), 1 (light rain), 2 (heavy rain)}.

[0120] UAV state : UAV altitude , UAV speed , remaining battery power .

[0121] Action space: .

[0122] Definition: The increment of each step adjustment is limited to to avoid mutations.

[0123] Update rule: .

[0124] Reward function: ; Accuracy reward: ; : Model prediction value, : True value deduced from historical accidents.

[0125] Safety reward : ; Stability penalty: ; (Suppress frequent adjustments) 2. Training process Data preparation: Historical dataset: Time series data containing environmental parameters, UAV state, and accident records (sampling frequency 1 Hz ).

[0126] Synthetic data augmentation: Expand extreme scenarios (such as sudden magnetic field increase + strong wind) through Gaussian noise injection and adversarial sample generation.

[0127] Simulation environment setup: Use Gazebo or AirSim simulation platform to simulate the flight scenario of the UAV in a dynamic environment.

[0128] Risk event trigger: When threshold, randomly inject faults (such as GPS signal loss).

[0129] Training steps: Initialization: The weights of the Actor and Critic networks are randomly initialized, and the capacity of the experience replay pool N = 106.

[0130] Interactive sampling: Execute the current policy in the simulation environment to collect trajectories .

[0131] Policy Optimization: Critic Update: Minimize the Temporal Difference Error (TD Error): ; is the value estimate of the state by the Critic network . is the discount factor, is the new state transferred after executing the action , is the expected operation; Actor Update: Maximize the PPO objective function: ; where the advantage function , is the probability density of selecting the action when the Actor network parameters are in the state , are the old policy network parameters (the Actor network before update), is the clipping threshold, is the clipping function.

[0132] Iterative Convergence: Train until the Critic loss and the reward curve is stable.

[0133] 3. Implementation (Deployment and Verification) Model Lightweighting: Convert the trained PPO model to a TensorRT engine and deploy it to the drone edge computing unit (such as NVIDIA Jetson).

[0134] Inference Latency Requirement: Single-step prediction < 10ms.

[0135] Real-time Prediction Process: Data Acquisition: Read from sensors (wind speed, magnetic field, battery power, etc.).

[0136] Action Generation: The Actor network outputs , update .

[0137] Risk Decision-making: If , trigger the obstacle avoidance or return-to-base command.

[0138] Effect Verification: Test Scenario: Strong magnetic field superimposed with battery aging .

[0139] Result Comparison Table: method accuracy rate of risk warning false alarm rate before RL dynamic adjustment 65% 25% after RL dynamic adjustment 92% 8% The above-described embodiments are used to illustrate the present invention, not to limit the present invention. Therefore, changes in the exemplified numerical values or replacement of equivalent elements should still fall within the scope of the present invention.

[0140] From the above detailed description, those of ordinary skill in the art can clearly understand that the present invention can indeed achieve the aforementioned objectives and has actually met the requirements of the patent law.

[0141] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the appended claims are intended to be construed as including the preferred embodiments and all changes and modifications falling within the scope of the present invention. The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0142] It should be noted that the above description of the process is only for illustration and explanation and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the process under the guidance of this specification. However, these modifications and changes are still within the scope of this specification.

[0143] The basic concepts have been described above. Obviously, for those of ordinary skill in the art after reading this application, the above invention disclosure is only an example and does not constitute a limitation to this application. Although not explicitly stated here, those of ordinary skill in the art may make various modifications, improvements, and corrections to this application. Such modifications, improvements, and corrections are proposed in this application, so such modifications, improvements, and corrections still belong to the spirit and scope of the exemplary embodiments of this application.

[0144] At the same time, this application uses specific terms to describe the embodiments of this application. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of this application. Therefore, it should be emphasized and noted that the "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned two or more times in different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of this application can be appropriately combined.

[0145] In addition, those of ordinary skill in the art can understand that various aspects of the present application can be illustrated and described by several patentable types or situations, including any new and useful process, machine, product, or composition of matter, or any new and useful improvement thereof. Therefore, various aspects of the present application can be implemented entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above-mentioned hardware or software can all be referred to as "units", "modules", or "systems". In addition, various aspects of the present application can take the form of a computer program product embodied in one or more computer-readable media, in which computer-readable program code is included.

[0146] The computer program code required for the operation of each part of the present application can be written in any one or more of the above programming languages, including object-oriented programming languages such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, etc., conventional procedural programming languages such as C programming language, VisualBasic, Fortran2103, Perl, COBOL2102, PHP, ABAP, dynamic programming languages such as Python, Ruby, and Groovy, or other programming languages. The program code can run entirely on the user's computer, or run on the user's computer as an independent software package, or run partially on the user's computer and partially on a remote computer, or run entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any network form, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (for example, through the Internet), or in a cloud computing environment, or used as a service such as software as a service (SaaS).

[0147] In addition, unless clearly stated in the claims, the order of the processing elements and sequences, the use of numbers and letters, or the use of other names in the present application is not used to limit the order of the processes and methods of the present application. Although some currently considered useful embodiments of the invention are discussed through various examples in the above disclosure, it should be understood that such details only serve the purpose of illustration, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that conform to the essence and scope of the embodiments of the present application. For example, although the implementation of the above various components can be embodied in a hardware device, it can also be implemented as a pure software solution. For example, it can be installed on an existing server or mobile device.

[0148] Similarly, it should be noted that, in order to simplify the description disclosed in this application and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of this application, sometimes multiple features are incorporated into one embodiment, drawing or description thereof. However, this method of this application should not be construed as reflecting an intention that the claimed subject matter requires more features than are expressly recited in each claim. On the contrary, the subject matter of the invention should have fewer features than the above single embodiment.

Claims

1. A method for tracing the source of multi-dimensional risk accidents of unmanned aerial vehicles based on a dynamic network model, characterized in that, Including: Risk factor identification: Analyze historical accident data, flight logs, and real-time operation environment data, and identify potential risk factors during the operation of unmanned aerial vehicles (UAVs) based on three dimensions of human-machine-environment. The potential risk factors include human factors, UAV's own performance risk factors, and operation environment risk factors; Construct a risk evolution network model: Regard potential risk factors as network nodes and the transmission relationships between risk factors as network edges to construct a risk evolution network model for UAV operation; Model parameter calculation: Based on the risk evolution network model of UAV operation, introduce the time node degree theory to calculate the node degree, introduce the risk propagation intensity to calculate the closeness centrality, calculate the betweenness centrality, introduce the dynamic decay factor to calculate the page rank value, and introduce the environmental risk factor to calculate the edge betweenness; Risk accident traceability analysis: Based on the dynamic change of risk propagation over time, introduce the risk loss coefficient, comprehensively analyze various network characteristic indicators, evaluate the importance of nodes, determine the key path of risk propagation, and extract the key risk propagation chain to obtain the analysis results.

2. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein The formula for calculating the node degree by introducing the time node degree theory is as follows: ； where represents the time - node degree of node ; represents the total number of network nodes; represents whether there is a risk - propagation edge between node at time and . If it exists, the value is 1; if not, the value is 0.

3. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein Introduce the formula for calculating the proximity centrality of risk propagation intensity as follows: ； In the formula, The number of edges contained in the shortest path starting from node and ending at node ; Represents the total number of nodes; Represents the risk propagation intensity from node to node ; Is the number of times the path from node i to j appears in the historical accident data.

4. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein, Calculating betweenness centrality The formula is as follows: ； In the formula, represents the total number of shortest paths between node and node ; represents the number of shortest paths from node to node through node 5. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein Introduce a dynamic decay factor The formula for calculating the page ranking value is as follows: ； ； ； Wherein: is the web page sorting value of the node; is the total number of nodes; is the initial attenuation factor; is the adjustment amplitude; is a time function; is a preset constant; is connected to the node; is the edge weight of; is the degree of the node; The size depends on the risk propagation probability of the edge i.e., after the node occurs, the conditional probability that the node will occur accordingly; ； Wherein, is the number of occurrences of and occurring simultaneously, is the number of occurrences of occurring alone.

6. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein The formula for calculating the edge betweenness by introducing the environmental risk factor is as follows: ； where is the edge betweenness; is and the environmental risk factor; is the number of paths between node and node through edge 7. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein The formula for introducing the risk loss coefficient and comprehensively analyzing various network characteristic indicators is as follows: ； Wherein, is the risk loss function, , are respectively the time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness of node ; are respectively the weight coefficients of the time node degree, closeness centrality, betweenness centrality, PageRank value, and edge betweenness, and the ranges are all [0, 1], and ; is the risk loss coefficient.

8. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 5, wherein Predicting the time function using an LSTM network ; ； Its hidden state The update formula is as follows: ； wherein is the Sigmoid function, is the weight matrix from the LSTM hidden state to the output, is the bias term, is the input feature within the time window.

9. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, characterized in that Constructing the risk evolution network model includes: Establish a hierarchical network structure, including the human factor layer, the airborne layer, and the environmental layer; Use a hypergraph to represent the cross-layer risk propagation path, and the hyperedge weight is calculated by the following formula: ； is a dynamic learning parameter, is an environmental risk factor.

10. The method for tracing the source of multi-dimensional risk accidents of an unmanned aerial vehicle based on a dynamic network model according to claim 1, wherein Use Kalman filtering to fuse multi-sensor data and calculate the real-time environmental risk factor ； ； wherein, is the normalized reading of the th sensor, is the temperature parameter, is the normalization coefficient.

Images