CN-122001805-A - Vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP

CN122001805ACN 122001805 ACN122001805 ACN 122001805ACN-122001805-A

Abstract

The application provides a vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP, which is based on the priority of service data to be transmitted and the real-time LQS score of a link, and realizes dynamic creation, distribution, switching and destruction of sub-streams for the service data to be transmitted through a dynamic sub-stream scheduling module, so as to complete data transmission between the vehicle and the unmanned aerial vehicle; according to the application, the LQS score is obtained based on the calculation of the core index of the available link, the core index covers the influence on the transmission link caused by the change conditions of the distance, the gesture and the like caused by the high-dynamic relative motion of the unmanned aerial vehicle and the vehicle, the calculated LQS score can reflect the high-dynamic relative motion of the unmanned aerial vehicle and the vehicle, the dynamic sub-stream scheduling module is ensured to be capable of adjusting the sub-stream allocation strategy in real time according to the dynamic position change of the adaptive vehicle and the unmanned aerial vehicle, and the link quality fluctuation is dealt with.

Inventors

LIANG TENG
MENG LINGSHUANG
LI HOUGUO
LI XIANGZHI
WANG YONGPING

Assignees

无锡车联天下智能科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (9)

1. The vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP is characterized by comprising the following steps of: s1, constructing a link state sensing module, and collecting core indexes of all available links between a vehicle-mounted unmanned aerial vehicle and a vehicle machine in real time; the core indexes comprise end-to-end time delay RTT, packet loss rate PLR, available bandwidth BW and signal strength RSSI; s2, constructing a data priority mapping module; the input of the data priority mapping module is business data to be transmitted; Aiming at scenes in the communication of the intelligent vehicle and the unmanned aerial vehicle, respectively defining priorities of various types of data in the communication of the intelligent vehicle and the unmanned aerial vehicle, and pre-storing the data in a data priority mapping module; After receiving the service data to be transmitted, the data priority mapping module generates MPTCP transmission priority corresponding to the service data according to the data type mapping of the service data and outputs the MPTCP transmission priority; s3, constructing a link quality score model; In the link quality score model, calculating a link quality score LQS for each available link in the system in real time based on the core index; S4, constructing a dynamic sub-stream scheduling module; Aiming at each service data to be transmitted, the dynamic sub-stream scheduling module realizes dynamic creation, distribution, switching and destruction of sub-streams according to the transmission priority corresponding to the service data to be transmitted and the real-time link quality scores LQS of all current links, and realizes communication data interaction between intelligent vehicles and unmanned aerial vehicles; The sub-stream creation operation comprises the steps of respectively creating MPTCP sub-streams for each type of available links according to the types of the current available links, and setting the initial state as activation; The sub-stream distribution operation comprises the steps of setting corresponding transmission modes for each type of transmission priority based on the real-time link quality score LQS, wherein the transmission modes define the types and the transmission mechanisms of available links adopted in multi-sub-stream transmission; The sub-stream switching operation comprises setting link switching standard or data processing method in the data transmission process for each type of transmission priority based on real-time link quality score LQS; the sub-stream destroying operation comprises the steps of evaluating each link according to the real-time link quality fraction LQS, and destroying the sub-stream of the link when the LQS of the evaluated link in the continuous N sampling periods is smaller than a preset smaller threshold value; S5, after the vehicle-to-machine communication is established, the link state sensing module calculates the core index of each available link and sends the core index into the link quality score model, and calculates the link quality score LQS for each available link; The dynamic sub-stream scheduling module completes sub-stream dynamic creation aiming at the current available link type; Once the service data to be transmitted is generated, the service data is sent to the data priority mapping module to obtain the corresponding transmission priority; the dynamic sub-stream scheduling module realizes data transmission for each piece of service data to be transmitted based on the link quality score LQS of each available link and the transmission priority corresponding to each piece of service data to be transmitted.
2. The method for communication data interaction between a vehicle and an unmanned aerial vehicle based on MPTCP of claim 1, wherein the method for calculating the link quality score is as follows: LQS=w 1 ·N RTT +w 2 ·N PLR +w 3 ·N BW +w 4 ·N RSSI ; Wherein, w 1 is the dynamic weight of the end-to-end time delay, w 2 is the dynamic weight of the packet loss rate, w 3 is the dynamic weight of the available bandwidth, and w 4 is the dynamic weight of the signal strength; n RTT is the normalized value of the end-to-end delay; N RTT =(RTT raw -RTT min )/(RTT max -RTT min ); RTT raw is actually monitored link delay data, RTT min is the minimum value of the actual measured delay of the corresponding link, RTT max is the maximum value of the actual measured delay of the corresponding link; N PLR is the normalized value of the packet loss rate; N PLR =1-PLR raw ; PLR raw is the actual monitored link packet loss rate; N BW is a normalized value of the available bandwidth; N BW =(BW raw -BW min )/(BW max -BW min ); BW raw is the actual monitored available bandwidth of the link, BW min is the actual measured bandwidth minimum value of the corresponding link, BW max is the actual measured bandwidth maximum value of the corresponding link; N RSSI is the normalized value of the signal intensity; N RSSI =(RSSI raw -RSSI min )/(RSSI max -RSSI min ); RSSI raw is the actual monitored link signal strength, RSSI min is the minimum measured signal strength of the corresponding link, and RSSI max is the maximum measured signal strength of the corresponding link.
3. The method for communication data interaction between a vehicle and an unmanned aerial vehicle based on MPTCP of claim 1, wherein the data types of the service data comprise control instructions, positioning data, pose data, perceived image data, point cloud data, state monitoring data and decision interaction data; The transmission priorities comprise a highest transmission priority P0, a high transmission priority P1, a medium-high transmission priority P2 and a medium transmission priority P3, wherein a control instruction corresponds to the highest transmission priority P0, positioning data and pose data correspond to the high transmission priority P1, perception image data and point cloud data correspond to the medium-high transmission priority P2, and state monitoring data and decision interaction data correspond to the medium transmission priority P3.
4. The method for data interaction between a vehicle and an unmanned aerial vehicle based on MPTCP according to claim 1, further comprising a method for dynamically adjusting weights in the link quality score model, specifically comprising the following steps: a1, constructing a data acquisition layer; The data acquisition layer acquires the data priority distribution data and the core indexes of all available links in all the current service data to be transmitted, calculates the normalized value of the core indexes and sends the normalized value to the state coding layer, wherein the data priority distribution data comprises a high priority duty ratio and a low priority duty ratio; a2, constructing a state coding layer; the state coding layer comprises a 1-layer full-connection layer and an activation function, wherein input data is compressed into a state vector S and sent to a lightweight DQN decision layer; S=[P01 ratio ,P23 ratio ,N RTT ,N PLR ,N BW ,N RSSI ,link load ,reward last ]; Wherein P01 ratio represents the duty ratio of high priority data in the service data to be transmitted, P23 ratio represents the duty ratio of low priority data in the service data to be transmitted, N RTT is the normalized value of end-to-end time delay, N PLR is the normalized value of packet loss rate, N BW is the normalized value of available bandwidth, N RSSI is the normalized value of signal intensity, and reward last represents the last round of rewarding value; link load represents the link total load rate: link load =(Σload x )/N; N is the total number of active links; Load x is the single link Load rate of the xth effective link, load x =TR x /BW x ; TR x is the actual transmission rate of the single link of the xth effective link, and BW x is the real-time available bandwidth of the single link; a3, constructing a lightweight DQN decision layer based on the DQN model; a4, constructing a transmission effect feedback layer; Constructing a lightweight experience pool in the transmission effect feedback layer, wherein the lightweight experience pool stores latest QN pieces of experience data, and each piece of experience data comprises a state vector S, an action a and a reward R; a5, constructing a lightweight reinforcement learning model with dynamic weights; The lightweight reinforcement learning model comprises a data acquisition layer, a state coding layer, a lightweight DQN decision layer and a transmission effect feedback layer which are sequentially connected; a6, constructing a feedback and learning mechanism, which specifically comprises the following steps: b1, setting an increment learning triggering condition; the increment learning triggering condition is that priority duty ratio data of service data to be transmitted and normalization values of four core indexes included in the state vector S input each time are respectively compared with corresponding temporary reference values, and if the priority duty ratio data and the normalization values of the four core indexes accord with mutation conditions, online increment learning is executed; The abrupt change condition includes that the data change amount of P01 ratio or P23 ratio is more than 20% compared with the temporary reference value, or the change amount of the normalized value of any one core index is more than 10% compared with the temporary reference value; b2, constructing an offline pre-training method; In offline pre-training, training the lightweight DQN decision layer based on a typical scene according to a preset offline training round to obtain a basic training model corresponding to the lightweight DQN decision layer; b3, constructing an online incremental learning method; in online incremental learning, using the data in the lightweight experience pool to retrain the lightweight DQN decision layer according to a preset incremental training round, and optimizing an action selection strategy of a model; b4, arranging the lightweight reinforcement learning model subjected to offline pre-training in an unmanned plane and an intelligent vehicle; If the state coding layer generates the state vector S for the first time, the state vector S is marked as a state vector to be processed, and the step b6 is executed; if the state coding layer does not generate the state vector S for the first time, the state vector S is marked as a state vector to be processed; b5, comparing the state vector to be processed with the temporary reference value; If the increment learning triggering condition is met, executing a step b7; if the increment learning triggering condition is not met, executing a step b8; b6, directly sending the state vector to be processed into a basic training model of the lightweight DQN decision layer, selecting a group of weights with highest Q values in the output of the basic training model as an optimal weight combination, and carrying out subsequent calculation; Simultaneously, all data included in the state vector S generated by the state coding layer are recorded as temporary reference values; b7, directly sending the state vector to be processed into a basic training model of the lightweight DQN decision layer, selecting a group of weights with highest Q values in the output of the basic training model as optimal weight combinations, and carrying out subsequent calculation; triggering the online increment learning and optimizing an action selection strategy; And b8, acquiring the optimal weight combination corresponding to the temporary reference value, and taking the optimal weight combination as the optimal weight combination corresponding to the current state vector to be processed to perform subsequent calculation.
5. The method for communication data interaction between a vehicle and an unmanned aerial vehicle based on MPTCP of claim 4, wherein in step a3, the method specifically comprises the following steps of; a31, constructing an action space; Discretizing the weight adjustment into N action core actions, wherein each action corresponds to a group of weight during the calculation of the link quality score; each set of weights meets the following weight constraint: constraint 1 the sum of all weights is 1; Constraint 2, setting basic weight range constraint for each weight according to actual communication scene requirements between the unmanned aerial vehicle and the vehicle; a32, adjusting the Q network structure in the DQN model to be: the input layer is that 8 neurons correspond to 8-dimensional input vectors S; the hidden layer is composed of 16 neurons, and RelU functions are used as the activation functions; The output layer is used for outputting Q value estimation of each action aiming at N action neurons and N action actions; a33, designing a reward function R for the lightweight DQN decision layer; R=0.7*R priority +0.2*R link +0.1*R stable ; Wherein R priority is priority standard-reaching rewards used for rewarding high priority data transmission standard-reaching rate; r link is a link utilization reward for punishing frequent and large-scale adjustment of weight; R stable is a weight stabilization reward for punishing frequent and large-scale adjustment of weight.
6. The method for data interaction between vehicle and unmanned aerial vehicle communication based on MPTCP of claim 4, wherein the lightweight reinforcement learning model further comprises a weight constraint layer and a link quality score LQS calculation layer; The link quality score LQS calculation layer is arranged between the weight constraint layer and the transmission effect feedback layer; The weight constraint layer is used for setting constraint conditions for weight values, and all weight value combinations output by the lightweight DQN decision layer are constrained and only the weight value combinations meeting the constraint conditions are output to a subsequent flow; the weight constraint layer comprises a sum constraint, a range constraint and a step length constraint; the comprehensive constraint is that the sum of the limiting ownership weights is 1; the range constraint is that a weight range conforming to the physical meaning is set for each weight; The step length constraint is that in the actual deployment operation process, when the optimal weight value combination is calculated as a weight updating value, the weight adjustment step length of each weight value is limited to be less than or equal to 0.05; The link quality score LQS calculation layer calculates the LQS score according to the optimal weight output by the lightweight DQN decision layer and the corresponding state vector S, and sends the LQS score, the weight value and the rewarding function R to the transmission effect feedback layer.
7. The method for exchanging data between a vehicle and an unmanned aerial vehicle based on MPTCP of claim 1, wherein the transmission mode comprises dual sub-stream redundancy transmission, dual sub-stream aggregation transmission, dual sub-stream bandwidth aggregation transmission and multi-sub-stream redundancy transmission.
8. The method for exchanging data between a vehicle and an unmanned aerial vehicle based on MPTCP of claim 1, further comprising a congestion control adaptation module and an anomaly recovery module; The congestion control adaptation module adopts a differential congestion control algorithm to independently adjust a congestion window for each sub-flow according to different congestion characteristics of heterogeneous links; The abnormal recovery module is used for realizing recovery operation aiming at an abnormal scene in a link, wherein the abnormal scene comprises link interruption and data packet loss; the link terminal recovery operation comprises switching to a standby sub-stream within 50ms when the link of the high priority data is interrupted, and triggering a retransmission mechanism; The data packet loss recovery operation comprises the steps of starting a fast retransmission and selective acknowledgement SACK mechanism for P0 and P1 data, setting retransmission timeout time RTO, retransmitting only key frames for P2 data, enabling non-key frames to be lost and not retransmitted, and supplementing lost data by means of periodic transmission for P3 data.
9. The method for communicating data between a vehicle and a drone based on MPTCP of claim 1, wherein the available links include C-V2X, wiFi and 5G.

Description

Vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP Technical Field The invention relates to the technical field of intelligent traffic control, in particular to a vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP. Background With the rapid fusion of intelligent networking automobile and unmanned aerial vehicle technology, communication data interaction between a vehicle and an Unmanned Aerial Vehicle (UAV) has become a core support in the fields of intelligent transportation, air-ground collaborative awareness and the like, and the application scene of the intelligent networking automobile comprises a plurality of aspects such as vehicle-mounted task unloading, road condition collaborative monitoring, emergency communication relay and the like. However, the high-speed movement of the vehicle and the flexible flight of the unmanned aerial vehicle can cause frequent switching of communication links, interference of wireless channels, dynamic topology change, bandwidth fluctuation and other problems, and the problems all provide serious challenges for the traditional single-path communication protocol. A technician introduces a multipath transmission control protocol (MPTCP) into the vehicle to drone communication to address the deficiencies of the single communication protocol. The vehicle and unmanned aerial vehicle communication belongs to the field of air-ground heterogeneous communication, and is characterized in that both communication parties have high mobility, and the communication environment is complex and changeable. The intelligent network-connected automobile can generate mass sensing data such as road condition and vehicle state data in the running process, and the unmanned aerial vehicle can make up for the defect of uneven coverage of a ground network by virtue of the flexibility of the intelligent network-connected automobile, and can be used as an air edge node to realize data acquisition, task unloading and communication relay, construct a three-dimensional mobile edge computing network and realize cross-domain sharing and collaborative scheduling of computing resources. The core requirements of the communication are concentrated on three points, namely real-time performance, namely, the transmission delay of a vehicle control instruction and an unmanned aerial vehicle flight scheduling instruction is controlled to be in a millisecond level, collaborative failure caused by delay is avoided, reliability, the problems of link interruption, data packet loss and the like are solved, the type of key data such as safety early warning information is not lost, and flexibility, namely, the communication link seamless switching and load balancing are realized by adapting to the dynamic position change of the vehicle and the unmanned aerial vehicle. However, in practical application, it is found that when the existing MPTCP technology is applied to communication between a vehicle and an unmanned aerial vehicle, there are still problems of insufficient route selection adaptability, mismatching between data scheduling and scene requirements, insufficient pertinence of congestion control mechanism, insufficient optimization and lack of safety and energy consumption, insufficient multi-protocol coordination, and the like, and these pain points severely restrict the performance improvement of communication data interaction between the vehicle based on MPTCP and the unmanned aerial vehicle, and cannot fully satisfy the practical application requirements of scenes such as intelligent traffic, air-space coordination perception, and the like. Disclosure of Invention In order to solve the problem that the existing MPTCP technology still cannot meet the communication requirements of intelligent vehicles and unmanned aerial vehicles, the application provides a vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP, which can adapt to the high mobility and dynamic link characteristics of the vehicles and the unmanned aerial vehicles and simultaneously gives consideration to real-time performance, reliability and safety. The technical scheme of the invention is that the vehicle and unmanned aerial vehicle communication data interaction method based on MPTCP is characterized by comprising the following steps: s1, constructing a link state sensing module, and collecting core indexes of all available links between a vehicle-mounted unmanned aerial vehicle and a vehicle machine in real time; the core indexes comprise end-to-end time delay RTT, packet loss rate PLR, available bandwidth BW and signal strength RSSI; s2, constructing a data priority mapping module; the input of the data priority mapping module is business data to be transmitted; Aiming at scenes in the communication of the intelligent vehicle and the unmanned aerial vehicle, respectively defining priorities of various types of data in the communic