A vehicle-mounted unmanned aerial vehicle scheduling method based on real-time rendezvous risk assessment

Abstract

The present application belongs to the technical field of vehicle-mounted unmanned aerial vehicle cooperation, and in particular to a vehicle-mounted unmanned aerial vehicle scheduling method based on real-time rendezvous risk assessment. The method comprises the following steps: S1: quantifying traffic risk probability; S2: quantifying energy consumption risk probability; S3: quantifying environmental risk probability; S4: performing dynamic scheduling decisions according to the comprehensive risk value and the predicted variance of each candidate point. The present application effectively captures the nonlinear characteristics of dynamic traffic flow fluctuations by introducing gradient boosting decision trees for residual iteration, significantly improving the accuracy of vehicle-mounted platform arrival time prediction. Traffic delays, energy consumption depletion, and environmental disturbances are decoupled as independent failure events, and dimensionality reduction fusion is performed using reliability theory to accurately define the absolute safety red line of the air-ground cooperative system. The rendezvous decision is driven by the comprehensive risk value, effectively overcoming the inertia of traditional static scheduling when facing sudden disturbances, and building a dynamic closed loop of "prediction-evaluation-decision".

Description

Technical Field

[0001] This invention relates to the field of vehicle-mounted drone collaboration technology, specifically a vehicle-mounted drone scheduling method based on real-time rendezvous risk assessment. Background Technology

[0002] With the development of the low-altitude economy and smart logistics, vehicle-mounted drones are increasingly being used in emergency rescue, last-mile delivery, and other fields. In air-ground collaborative missions, the real-time rendezvous between the vehicle-mounted platform and the drone is a core element in ensuring mission continuity and system safety. However, actual traffic and the low-altitude environment are highly uncertain. Sudden road network congestion, nonlinear energy decay of drones, and rapidly changing weather disturbances can all significantly limit the success rate of air-ground collaborative rendezvous.

[0003] Most existing drone dispatching solutions rely on offline pre-planning based on static road conditions, lacking in-depth online assessment of dynamic disturbances. This results in sluggish system response when faced with sudden traffic congestion or severe weather. Furthermore, existing risk assessment models are often one-dimensional, failing to deeply characterize the heterogeneous features between traffic conditions, drone energy consumption, and meteorological environment. They also lack a coupling and fusion mechanism for multi-dimensional independent risks, making it difficult to accurately quantify and dynamically defend against the system's overall safety threshold. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] To address the aforementioned technical problems, according to one aspect of the present invention, the present invention provides the following technical solution:

[0006] A method for scheduling vehicle-mounted drones based on real-time rendezvous risk assessment includes the following steps:

[0007] S1: Based on historical speed sequences, real-time speed and time characteristics of road traffic state data, construct a spatiotemporal prediction model for road conditions, predict the travel time distribution of the vehicle platform to each candidate meeting point, and quantify the probability of traffic risks.

[0008] S2: Based on the drone's flight path and remaining battery power, calculate the total energy consumption demand for the drone rendezvous and quantify the probability of energy consumption risk.

[0009] S3: Based on environmental perception data, assess external environmental disturbances and quantify the probability of environmental risks;

[0010] S4: Integrate traffic risk probability, energy consumption risk probability, and environmental risk probability to calculate the comprehensive rendezvous risk value of each candidate rendezvous point, and execute dynamic scheduling decisions based on the comprehensive risk value and prediction variance of each candidate point.

[0011] As a preferred embodiment of the vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment described in this invention, the specific method for constructing the spatiotemporal prediction model of road conditions in step S1 is as follows: the road network is represented as a directed graph, wherein each road segment has a segment from node i to node j. Road section In the past moment Historical traffic speed ,in For the current moment, Historical time interval parameters, road segments At the present moment Real-time traffic speed The time-related features are used as inputs; a gradient boosting decision tree model is used to calculate the road segment. Predicted vehicle speed within the future prediction time window t+τ (t+τ), where τ is the prediction time span parameter;

[0012] Based on the predicted vehicle speed, calculate the predicted travel time for the road segment: (t+τ)= .

[0013] As a preferred embodiment of the vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment described in this invention, the specific steps of the gradient boosting decision tree model are as follows:

[0014] Constructing a feature input vector: This involves inputting the historical traffic speeds of each road segment. Real-time traffic speed And categorical variable parameters of congestion characteristics in the current time period. Input vector ;

[0015] Initialization: Initialize the prediction baseline value The historical global average speed;

[0016] Pseudo-residual: In the m-th iteration, the negative gradient is calculated as a pseudo-residual, and the m-th regression tree is trained. Fit the pseudo residual;

[0017] Model Update: Introducing Learning Rate Update the model:

[0018] After the iteration converges, the final prediction model is output, and the prediction speed is obtained.

[0019] As a preferred embodiment of the vehicle-mounted drone scheduling method based on real-time rendezvous risk assessment described in this invention, the specific method for quantifying the traffic risk probability in S1 is as follows:

[0020] Based on the predicted travel time of each road segment (t+τ) represents the distance from the current location of the vehicle platform to the candidate rendezvous point. driving route The predicted travel times for each segment of the driving path are summed to obtain the candidate rendezvous points reached by the onboard platform. Predicted total travel time:

[0021] = ;

[0022] in, Indicates that by node Pointing to node The section of road; Indicates candidate meeting points; This indicates the distance from the vehicle platform's current location to the candidate rendezvous point. The driving route; (t+τ) represents the road segment The predicted travel time at the future predicted time t+τ; This indicates that the vehicle platform has arrived at the candidate rendezvous point from its current location. The predicted total travel time;

[0023] The predicted driving time Model it as a random variable and calculate its probability distribution;

[0024] Based on the drone's maximum waiting time Calculate the probability of the onboard platform's ability to reach the candidate rendezvous point within the time constraint. = ( );

[0025] in, ( () indicates the probability of the event within the parentheses occurring;

[0026] Calculate the probability of traffic risk =1- .

[0027] As a preferred embodiment of the vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment described in this invention, the specific method for quantifying the energy consumption risk probability in S2 is as follows:

[0028] Based on the UAV flight path length d and flight altitude change Calculate the energy consumption of drone flight: = d+ ,in, This represents the energy consumption coefficient per unit flight of a drone. This represents the energy consumption coefficient per unit climb / descent for a drone;

[0029] Based on the time difference between the predicted travel time from the vehicle platform and the time the drone arrives at the rendezvous point, the energy consumption of the drone while hovering is calculated: = ( - ),in, This represents the power consumption coefficient per unit time when a drone is hovering in the air. Indicates the time when the drone arrives at the rendezvous point;

[0030] Adding the flight energy consumption and the hovering energy consumption together, we obtain the total energy consumption requirement for the UAV to perform the rendezvous mission: = + ;

[0031] Calculate the remaining battery power of the drone With the total energy consumption requirement Energy margin between: E= - ;

[0032] When the energy margin is less than or equal to zero, the energy consumption risk of the drone is defined as the maximum risk value;

[0033] When the energy margin is greater than zero, the probability of drone energy consumption risk is calculated using an exponential decay function:

[0034] =exp(-k E), where k represents the exponential decay sensitivity coefficient parameter of the rate at which energy consumption risk amplifies with the reduction of electricity consumption.

[0035] As a preferred embodiment of the vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment described in this invention, the specific method for quantifying the environmental risk probability in S3 is as follows:

[0036] Dimensionless processing: Obtaining candidate meeting points Environmental sensing data, selecting actual wind speed Current visibility and current environmental interference values Perform normalization processing;

[0037] Candidate meeting points Wind speed failure normalization index value ( )= ,in Maximum safe wind speed allowed for drones; candidate rendezvous points Visibility failure normalized index value (q)= ,in Maximum visible distance; candidate rendezvous point Failure environmental interference failure normalized index value ( )= ,in The maximum environmental disturbance value among all candidate points;

[0038] Weight determination: The weight vectors for wind speed, visibility, and environmental disturbances were calculated using the analytic hierarchy process (AHP). , and After normalization, the weight coefficients are determined as follows: ;

[0039] Calculating the comprehensive environmental risk index: candidate convergence points Comprehensive environmental risk indicators: ,in, This refers to the numbering of environmental impact factors;

[0040] Output the final environmental risk probability: ,in, This represents the maximum possible value of the comprehensive environmental risk index.

[0041] As a preferred embodiment of the vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment described in this invention, the specific method of S4 is as follows:

[0042] The combined rendezvous risk value of each candidate rendezvous point is obtained by fusing the probabilities of various risks based on the reliability model of the parallel system. =1- And sorted in ascending order, among which, Indicates the probability of traffic risk. Indicates the probability of energy consumption risk. Indicates the probability of environmental risk. This represents the overall convergence risk value;

[0043] Candidate points with significantly deteriorating comprehensive risk values ​​in the ranking results are directly eliminated; and the variance of the prediction uncertainty index for nodes that pass the initial screening is extracted. If a candidate point has a low expected risk difference, but If the value is significantly larger than other candidate points, the system determines that the spatiotemporal cooperative structure of that point is fragile and easily subject to disturbances and instantaneous failure, and it is also eliminated.

[0044] Finally, the candidate point with the highest risk and data confidence is selected as the actual rendezvous point, and the spatiotemporal trajectory self-organizing alignment instruction is issued.

[0045] Compared with the prior art, the beneficial effects of this invention are: high prediction confidence: by introducing a gradient boosting decision tree for residual iteration, the nonlinear characteristics of dynamic traffic flow fluctuations are effectively captured, significantly improving the accuracy of onboard platform arrival time prediction.

[0046] Comprehensive risk characterization: Traffic delays, energy depletion, and environmental disturbances are decoupled into independent failure events, and reliability theory is used to reduce the dimension and integrate them, thus accurately defining the absolute safety red line of the air-ground cooperative system.

[0047] Strong robustness against disturbances: The decision-making process driven by the comprehensive risk value effectively overcomes the lag of traditional static scheduling in the face of sudden disturbances, and constructs a dynamic closed loop of "prediction-evaluation-decision". Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and detailed embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0049] Figure 1 This is a flowchart of a vehicle-mounted drone scheduling method based on real-time rendezvous risk assessment according to the present invention. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0052] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0053] Please see Figure 1A method for scheduling vehicle-mounted drones based on real-time rendezvous risk assessment includes the following steps:

[0054] S1: Based on historical speed sequences, real-time speed and time characteristics of road traffic state data, construct a spatiotemporal prediction model for road conditions, predict the travel time distribution of the vehicle platform to each candidate meeting point, and quantify the probability of traffic risks.

[0055] The specific method for constructing a spatiotemporal prediction model for road conditions is to represent the road network as a directed graph, where each road segment has segments from node i to node j. Road section In the past moment Historical traffic speed ,in For the current moment, Historical time interval parameters, road segments At the present moment Real-time traffic speed The time-related features are used as inputs; a gradient boosting decision tree model is used to calculate the road segment. Predicted vehicle speed within the future prediction time window t+τ (t+τ), where τ is the prediction time span parameter;

[0056] Based on the predicted vehicle speed, calculate the predicted travel time for the road segment: (t+τ)= ;

[0057] The specific steps of the gradient boosting decision tree model are as follows:

[0058] Constructing a feature input vector: This involves inputting the historical traffic speeds of each road segment. Real-time traffic speed And categorical variable parameters of congestion characteristics in the current time period. Input vector ;

[0059] Initialization: Initialize the prediction baseline value The historical global average speed;

[0060] Pseudo-residual: In the m-th iteration, the negative gradient is calculated as a pseudo-residual, and the m-th regression tree is trained. Fit the pseudo residual;

[0061] Model Update: Introducing Learning Rate Update the model:

[0062] After iterative convergence, the final prediction model is output, and the prediction speed is obtained.

[0063] The specific methods for quantifying the probability of traffic risks are as follows:

[0064] Based on the predicted travel time of each road segment (t+τ) represents the distance from the current location of the vehicle platform to the candidate rendezvous point. driving route The predicted travel times for each segment of the driving path are summed to obtain the candidate rendezvous points reached by the onboard platform. Predicted total travel time:

[0065] = ;

[0066] in, Indicates that by node Pointing to node The section of road; Indicates candidate meeting points; This indicates the distance from the vehicle platform's current location to the candidate rendezvous point. The driving route; (t+τ) represents the road segment The predicted travel time at the future predicted time t+τ; This indicates that the vehicle platform has arrived at the candidate rendezvous point from its current location. The predicted total travel time.

[0067] The predicted driving time Model it as a random variable and calculate its probability distribution;

[0068] Based on the drone's maximum waiting time Calculate the probability of the onboard platform's ability to reach the candidate rendezvous point within the time constraint. = ( ), ( () indicates the probability of the event within the parentheses occurring;

[0069] Calculate the probability of traffic risk =1- .

[0070] S2: Based on the drone's flight path and remaining battery power, calculate the total energy consumption demand for the drone rendezvous and quantify the probability of energy consumption risk.

[0071] The specific method for quantifying the probability of energy consumption risk is as follows:

[0072] Based on the UAV flight path length d and flight altitude change Calculate the energy consumption of drone flight: = d+ ,in, This represents the energy consumption coefficient per unit flight of a drone. This represents the energy consumption coefficient per unit climb / descent for a drone;

[0073] Based on the time difference between the predicted travel time from the vehicle platform and the time the drone arrives at the rendezvous point, the energy consumption of the drone while hovering is calculated: = ( - ),in, This represents the power consumption coefficient per unit time when a drone is hovering in the air. Indicates the time when the drone arrives at the rendezvous point;

[0074] Adding the flight energy consumption and the hovering energy consumption together, we obtain the total energy consumption requirement for the UAV to perform the rendezvous mission: = + ;

[0075] Calculate the remaining battery power of the drone With the total energy consumption requirement Energy margin between: E= - ;

[0076] When the energy margin is less than or equal to zero, the energy consumption risk of the drone is defined as the maximum risk value;

[0077] When the energy margin is greater than zero, the probability of drone energy consumption risk is calculated using an exponential decay function:

[0078] =exp(-k E), where k represents the exponential decay sensitivity coefficient parameter of the rate at which energy consumption risk amplifies with the reduction of electricity consumption.

[0079] S3: Based on environmental perception data, assess external environmental disturbances and quantify the probability of environmental risks;

[0080] The specific methods for quantifying the probability of environmental risks are as follows:

[0081] Dimensionless processing: Obtaining candidate meeting points Environmental sensing data, selecting actual wind speed Current visibility and current environmental interference values Perform normalization processing;

[0082] Candidate meeting points Wind speed failure normalization index value ( )= ,in Maximum safe wind speed allowed for drones; candidate rendezvous points Visibility failure normalized index value ( )= ,in Maximum visible distance; candidate rendezvous point Failure environmental interference failure normalized index value (q)= ,in The maximum environmental disturbance value among all candidate points;

[0083] Weight determination: The weight vectors for wind speed, visibility, and environmental disturbances were calculated using the analytic hierarchy process (AHP). , and After normalization, the weight coefficients are determined as follows: ;

[0084] Calculating the comprehensive environmental risk index: candidate convergence points Comprehensive environmental risk indicators: ,in, This refers to the numbering of environmental impact factors;

[0085] Output the final environmental risk probability: ,in, This represents the maximum possible value of the comprehensive environmental risk index.

[0086] S4: Integrate traffic risk probability, energy consumption risk probability and environmental risk probability to calculate the comprehensive rendezvous risk value of each candidate rendezvous point, and execute dynamic scheduling decisions based on the comprehensive risk value and prediction variance of each candidate point.

[0087] The specific method is as follows:

[0088] The combined rendezvous risk value of each candidate rendezvous point is obtained by fusing the probabilities of various risks based on the reliability model of the parallel system. =1- And sorted in ascending order; among them, Indicates the probability of traffic risk. Indicates the probability of energy consumption risk. Indicates the probability of environmental risk. This represents the overall convergence risk value;

[0089] Candidate points with significantly deteriorating comprehensive risk values ​​in the ranking results are directly eliminated; and the variance of the prediction uncertainty index for nodes that pass the initial screening is extracted. If a candidate point has a low expected risk difference, but a high variance in its distribution... If the value is significantly larger than other candidate points, the system determines that the spatiotemporal cooperative structure of that point is fragile and easily subject to disturbances and instantaneous failure, and it is also eliminated.

[0090] Finally, the candidate point with the highest risk and data confidence is selected as the actual rendezvous point, and the spatiotemporal trajectory self-organizing alignment instruction is issued.

[0091] Example:

[0092] Taking a single-vehicle-single-drone collaborative delivery scenario as an example, let the vehicle's current position be V, the drone's current position be U, and the system have three candidate rendezvous points A, B, and C. The drone's flight speed is known. Energy consumption per unit flight speed of a drone at 60km / h =15Wh / km, hovering power =80W, current remaining battery power =120Wh. Maximum allowable latency of the system. t=5min.

[0093] The parameters for each candidate rendezvous point are as follows:

[0094] Point A: Vehicle travel distance 4km, drone flight distance The maximum wind speed is 15 m / s, with a range of 3 km and a maximum visibility of 6 km. The maximum visibility is 10 km, and the interference I is 0.22.

[0095] Point B: Vehicle travel distance is 6km, drone flight distance is... 2km, wind speed is 6m / s, maximum wind speed is 15m / s, visibility is 8km, maximum visibility is 10km, interference I is 0.35;

[0096] Point C: Vehicle travel distance is 4.5km, drone flight distance is... 2.5km, wind speed 4m / s, maximum wind speed 15m / s, visibility 9km, maximum visibility 10km, interference I is 0.15;

[0097] Step S1: Based on historical speed sequences, real-time speed, and time-related road traffic state data, construct a spatiotemporal prediction model for road conditions to predict the travel time distribution of the vehicle platform to each candidate meeting point and quantify the probability of traffic risks, as detailed below:

[0098] Step S11: Represent the road network as a directed graph, where each segment has a length. Based on historical traffic speed Real-time traffic speed Using time-related features as input, a gradient boosting decision tree model is used to calculate the predicted speed of road segments. (t+τ), specifically as follows:

[0099] The system represents the road network as a directed graph, and sets the model learning rate. =0.1.

[0100] For candidate point A, extract the mean of the historical vehicle speed sequence for road segment A. =45km / h, real-time traffic speed =38km / h (minor fluctuation), and current time categorical variable parameters =0 (non-congestion-prone periods). Input vector Initialize the prediction baseline value =45km / h, the pseudo residual is predicted in the first round of iterations. =-30km / h, updated prediction value is =42km / h, the second iteration outputs a slightly smoothed residual. =-15km / h, updated prediction value is =42+0.1 (-15) = 40.5 km / h, the residual output in the 3rd iteration. =-5km / h, iterative convergence, final output vehicle speed prediction value =40km / h.

[0101] For candidate point B, extract the historical average vehicle speed of road segment B. =48km / h, real-time traffic speed =25km / h, and time characteristic variables =1, initialize the prediction baseline value =48km / h After three-tree residual iteration (outputting -50km / h, -20km / h, and -10km / h respectively), the iteration converges, and the final vehicle speed prediction value is output. =40km / h

[0102] For candidate point C, extract the historical average vehicle speed of road segment C. =55km / h, real-time traffic speed =52km / h. And time-related variables. =0, initialize the prediction baseline value The predicted vehicle speed is 55 km / h. After iterative iteration using three tree residuals (outputting -20 km / h, -15 km / h, and -15 km / h respectively), the iteration converges, and the final predicted vehicle speed is output. =50km / h

[0103] Step S12: Calculate the predicted travel time for the road segment and model its probability distribution: based on the formula (t+τ)= ;

[0104] Predicted travel time at point A =4 / 40 = 0.1h = 6min, modeling it as a normal distribution, the expected value is... = 6. Variance, an indicator of forecast uncertainty =2;

[0105] Predicted travel time at point B =6 / 40=9min, its expected distribution =9, due to the sudden congestion characteristics causing drastic changes in the residuals, the variance of the prediction uncertainty index is set to... =4;

[0106] Predicted travel time at point C =4.5 / 50=5.4min. Due to the extremely high random fluctuation of speed on this road section, its expected distribution is... =5.4, the variance of the prediction uncertainty index is set to... =8;

[0107] Step S13: Calculate the probability of the onboard platform's arrival capability and quantify the probability of traffic risk. This is based on the allowable delay time constraint. Set the maximum permitted arrival time for vehicles. , For the arrival time of the drone, According to the formula Perform the calculation:

[0108] Time constraints at point A =2+5=7min, after system correction and calculation, the corresponding standard normal distribution variable Z=-0.5, and the positional capability probability is obtained by looking up the table. =Φ(-0.5)=0.3085, then the probability of traffic risk is... =1-0.8413=0.1587;

[0109] Time constraints at point B =3+5=8min, the standard normal distribution variable Z=(8-6) / 2=1, and the probability of positional ability is obtained by looking up the table. =Φ(1)=0.8413, then the probability of traffic risk is... =1-0.3085=0.6915;

[0110] Time constraints at point C =2.5 + 5 = 7.5 min. After adjusting for specific historical fluctuations, the corresponding standard normal distribution variable Z = 0.2625. The probability of displacement is obtained from the table. =Φ(0.2625)=0.6035, then the probability of traffic risk is... =1-0.6035=0.3965;

[0111] Step S2: Based on the drone's flight path and remaining battery power, calculate the total energy consumption requirement for the drone rendezvous and quantify the probability of energy consumption risk. Details are as follows:

[0112] (1) Calculate the flight energy consumption of the UAV Energy consumption during hovering Total energy consumption:

[0113] The flight energy consumption at point A is 3. 15 = 45Wh. Based on the data in the example, the drone needs to hover for 6 - 3 = 3 minutes = 0.05 hours, and the hovering energy consumption is 80W. 0.05h = 4Wh, total energy consumption ;

[0114] The flight energy consumption at point B is 2. 15 = 30Wh. Based on the data in the example, the drone needs to hover for 9 - 2 = 7 minutes = 0.117 hours, and the hovering energy consumption is 80W. 0.117h = 9.36Wh, total energy consumption

[0115] The energy consumption for flight at point C is 2. 15 = 37.5Wh. Based on the data in the example, the drone needs to hover for 5.4 – 2.5 = 2.9 min = 0.0483 h, and the hovering energy consumption is 80W. 0.0483h = 3.864Wh, total energy consumption

[0116] (2) Calculate the remaining battery power of the drone Total energy demand Energy margin between : ;

[0117] =120-49=71Wh;

[0118] =120-39.36=80.64Wh;

[0119] =120-41.364=78.636Wh;

[0120] (3) Calculate the probability of UAV energy consumption risk using the exponential decay function. =exp(-k E);

[0121] =exp(-0.02 71)=exp(-1.42) 0.2417;

[0122] =exp(-0.02 80.64)=exp(-1.6128) 0.1993;

[0123] =exp(-0.02 78.636)=exp(-1.5727)=0.2075

[0124] Step S3: Based on environmental perception data, assess external environmental disturbances and quantify the probability of environmental risks, as detailed below:

[0125] (1) Acquire environmental perception data and perform dimensionless normalization processing:

[0126] According to the formula: Wind speed risk ( )= ,in, This refers to the actual wind speed. For maximum safe wind speed;

[0127] Visibility risk ( )= ,in, For current visibility, Maximum visible distance;

[0128] Environmental disturbance risk ( )= ,in, This represents the current environmental interference value (from sensor data or historical data statistics). The maximum interference value (take the maximum interference value among all candidate points);

[0129] Normalized value at point A =8 / 15=0.5333, =1-6 / 10=0.4, =0.22;

[0130] Normalized value at point B =6 / 15=0.4, =1-8 / 10=0.2, =0.35;

[0131] Normalized value at point C =4 / 15=0.2667, =1-9 / 10=0.1, =0.15;

[0132] (2) The weight coefficients of each environmental factor were determined using the analytic hierarchy process (AHP):

[0133] Obtaining large feature vectors =3.071984, =3.032613, =3.009049, and the final weights of wind speed, visibility, and environmental disturbances were calculated to be 3.009049. =0.6333、 =0.2605、 =0.1062;

[0134] (3) Calculate the comprehensive environmental risk index and output the final environmental risk probability: based on the formula Maximum possible environmental indicators Treated as 1, that is ;

[0135] Environmental risk probability at point A (A)= (A)+ (A)+ (A) = 0.4653;

[0136] Environmental risk probability at point B (B)= (B)+ (B)+ (B) = 0.3426;

[0137] Environmental risk probability at point C (C)= (C)+ (C)+ (C) = 0.2108;

[0138] Step S4: Integrate traffic risk probability, energy consumption risk probability, and environmental risk probability to calculate the comprehensive rendezvous risk value for each candidate rendezvous point. Then, execute dynamic scheduling decisions based on the comprehensive risk value and prediction variance of each candidate point, as detailed below:

[0139] (1) Parallel fusion calculation of multidimensional risks based on the reliability model of parallel systems: each risk probability is regarded as an independent failure event, and the fusion formula is used. =1- Calculate the overall risk value of each candidate point.

[0140] Comprehensive risks at point A: =1-(1-0.1587)×(1-0.2417)×(1-0.4653)=0.6589;

[0141] Comprehensive risks at point B: =0.8376;

[0142] Comprehensive risks at point C: =0.6225;

[0143] (2) Based on the comprehensive risk ranking results, identify and eliminate high-risk convergence points: The system sorts the comprehensive risk values ​​of each point in ascending order. Candidate point B has the highest comprehensive convergence risk value and extremely high traffic risk. It is a node that is severely deteriorated and does not meet the coordination requirements. The system first directly eliminates and avoids it.

[0144] (3) Combine spatiotemporal prediction uncertainty indicators to identify and eliminate high-volatility vulnerability points: extract the variance of the predicted travel time of nodes that pass the initial screening. Although candidate point C has the lowest expected value for multidimensional coupling risk, its travel time distribution variance is significantly high. The system determines that point C is a highly volatile and vulnerable node, extremely susceptible to transient coordination failures due to random traffic fluctuations during operation; therefore, it is also eliminated.

[0145] (4) Robustly select actual meeting points and issue spatiotemporal trajectory self-organizing alignment instructions: In comparison, although the comprehensive risk value of candidate point A is slightly higher than that of point C, its traffic flow evolution is extremely stable, and the variance of the prediction uncertainty index is only a small percentage. =1.5. This indicates that candidate point A has the highest data confidence and the system is the most robust against dynamic disturbances. The system eventually converges and selects candidate point A, which has controlled risk and the highest data confidence, as the actual rendezvous point. It then issues spatiotemporal trajectory self-organizing alignment instructions to the vehicle platform and the drone, triggering the adaptive dynamic scheduling of the rendezvous strategy.

[0146] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for scheduling vehicle-mounted unmanned aerial vehicles (UAVs) based on real-time rendezvous risk assessment, characterized in that, Includes the following steps: S1: Based on historical speed sequences, real-time speed and time characteristics of road traffic state data, construct a spatiotemporal prediction model for road conditions, predict the travel time distribution of the vehicle platform to each candidate meeting point, and quantify the probability of traffic risks. S2: Based on the drone's flight path and remaining battery power, calculate the total energy consumption demand for the drone rendezvous and quantify the probability of energy consumption risk. S3: Based on environmental perception data, assess external environmental disturbances and quantify the probability of environmental risks; S4: Integrate traffic risk probability, energy consumption risk probability, and environmental risk probability to calculate the comprehensive rendezvous risk value of each candidate rendezvous point, and execute dynamic scheduling decisions based on the comprehensive risk value and prediction variance of each candidate point.

2. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 1, characterized in that, The specific method for constructing the spatiotemporal prediction model of road conditions in S1 is as follows: the road network is represented as a directed graph, where each road segment has a segment from node i to node j. Road section In the past moment Historical traffic speed ,in For the current moment, Historical time interval parameters, road segments At the present moment Real-time traffic speed and time-related features are used as inputs; Calculate road segments using gradient boosting decision tree model Predicted vehicle speed within the future prediction time window t+τ (t+τ), where τ is the prediction time span parameter; Based on the predicted vehicle speed, calculate the predicted travel time for the road segment: (t+τ)= .

3. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 2, characterized in that, The specific steps of the gradient boosting decision tree model are as follows: Constructing a feature input vector: This involves inputting the historical traffic speeds of each road segment. Real-time traffic speed And categorical variable parameters of congestion characteristics in the current time period. Input vector ; Initialization: Initialize the prediction baseline value The historical global average speed; Pseudo-residual: In the m-th iteration, the negative gradient is calculated as a pseudo-residual, and the m-th regression tree is trained. Fit the pseudo residual; Model Update: Introducing Learning Rate Perform model updates: After the iteration converges, the final prediction model is output, and the prediction speed is obtained.

4. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 3, characterized in that, The specific method for quantifying the probability of traffic risk in S1 is as follows: Based on the predicted travel time of each road segment (t+τ) represents the distance from the current location of the vehicle platform to the candidate rendezvous point. driving route The predicted travel times for each segment of the driving path are summed to obtain the candidate rendezvous points reached by the onboard platform. Predicted total travel time: = ； in, Indicates that by node Pointing to node The section of road; Indicates candidate meeting points; This indicates the distance from the vehicle platform's current location to the candidate rendezvous point. The driving route; (t+τ) represents the road segment The predicted travel time at the future predicted time t+τ; This indicates that the vehicle platform has arrived at the candidate rendezvous point from its current location. The predicted total travel time; The predicted driving time Model it as a random variable and calculate its probability distribution; Based on the drone's maximum waiting time Calculate the probability of the onboard platform's ability to reach the candidate rendezvous point within the time constraint. = ( ); in, ( () indicates the probability of the event within the parentheses occurring; Calculate the probability of traffic risk =1- .

5. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 1, characterized in that, The specific method for quantifying the probability of energy consumption risk in S2 is as follows: Based on the UAV flight path length d and flight altitude change Calculate the energy consumption of drone flight: = d+ ,in, This represents the energy consumption coefficient per unit flight of a drone. This represents the energy consumption coefficient per unit climb / descent for a drone; Based on the time difference between the predicted travel time from the vehicle platform and the time the drone arrives at the rendezvous point, the energy consumption of the drone while hovering is calculated: = ( - ),in, This represents the power consumption coefficient per unit time when a drone is hovering in the air. Indicates the time when the drone arrives at the rendezvous point; Adding the flight energy consumption and the hovering energy consumption together, we obtain the total energy consumption requirement for the UAV to perform the rendezvous mission: = + ; Calculate the remaining battery power of the drone With the total energy consumption requirement Energy margin between: E= - ; When the energy margin is less than or equal to zero, the energy consumption risk of the drone is defined as the maximum risk value; When the energy margin is greater than zero, the probability of drone energy consumption risk is calculated using an exponential decay function: =exp(-k E), where k represents the exponential decay sensitivity coefficient parameter of the rate at which energy consumption risk amplifies with the reduction of electricity consumption.

6. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 1, characterized in that, The specific method for quantifying the probability of environmental risk in S3 is as follows: Dimensionless processing: Obtaining candidate meeting points Environmental sensing data, selecting actual wind speed Current visibility and current environmental interference values Perform normalization processing; Candidate meeting points Wind speed failure normalization index value ( )= ,in The maximum safe wind speed allowed for drones; Candidate meeting points Visibility failure normalized index value (q)= ,in Maximum visible distance; candidate rendezvous point Failure environmental interference failure normalized index value ( )= ,in The maximum environmental disturbance value among all candidate points; Weight determination: The weight vectors for wind speed, visibility, and environmental disturbances were calculated using the analytic hierarchy process (AHP). , and After normalization, the weight coefficients are determined as follows: ; Calculating the comprehensive environmental risk index: candidate convergence points Comprehensive environmental risk indicators: ,in, This refers to the numbering of environmental impact factors; Output the final environmental risk probability: ,in, This represents the maximum possible value of the comprehensive environmental risk index.

7. The vehicle-mounted UAV scheduling method based on real-time rendezvous risk assessment according to claim 1, characterized in that, The specific method of S4 is as follows: The combined rendezvous risk value of each candidate rendezvous point is obtained by fusing the probabilities of various risks based on the reliability model of the parallel system. =1- And sorted in ascending order, among which, Indicates the probability of traffic risk. Indicates the probability of energy consumption risk. Indicates the probability of environmental risk. This represents the overall convergence risk value; Candidate points with significantly deteriorating comprehensive risk values ​​in the ranking results are directly eliminated; and the variance of the prediction uncertainty index for nodes that pass the initial screening is extracted. If a candidate point has a low expected risk difference, but If the value is significantly larger than other candidate points, the system determines that the spatiotemporal cooperative structure of that point is fragile and easily subject to disturbances and instantaneous failure, and it is also eliminated. Finally, the candidate point with the highest risk and data confidence is selected as the actual rendezvous point, and the spatiotemporal trajectory self-organizing alignment instruction is issued.

Images