An Adaptive Link Monitoring System for Software-Defined Vehicle Networks

Abstract

The present invention discloses a link monitoring system for an adaptive software-defined vehicle network, comprising: a controller, a routing calculation module, a link evaluation module, an adaptive threshold module, and a link detection module; the routing calculation module is used to calculate a route according to a built-in routing algorithm thereof; the link evaluation module is used to evaluate all links on the route path and obtain an evaluation value of the link quality by using a fuzzy logic method; the adaptive threshold module uses a Q-learning method to determine a detection threshold according to the state of the link; the link detection module is used to detect packet loss on the link and send a detection report to the controller; the link monitoring system provided by the present invention combines link availability, link stability, and link load factors to obtain an evaluation value of the link quality. The detection threshold can be adaptively adjusted according to the detection result of the link. Selective detection is performed by using the global information of the SDVN, and the detection overhead is reduced by adjusting the detection range.

Description

Technical Field

[0001] The present disclosure relates to the technical field of adaptive link state awareness, and more particularly to a link monitoring system for an adaptive software-defined vehicle network. Background Art

[0002] Vehicular Ad Hoc Network (VANET) is an important part of the construction of intelligent transportation systems. VANET is a promising wireless network technology that can provide information services for passengers and drivers and promote road safety. However, the high mobility of vehicles and the rapidly changing environment pose some challenges to achieving this goal. Different from other ad hoc networks, the network topology in VANET changes frequently. The rapid change of link state makes data transmission difficult, and data packets may be lost due to link failures, which is particularly serious in multi-hop routing. In a distributed vehicle network, vehicles usually obtain the local topology by broadcasting beacon messages. Vehicles are only allowed to make local routing decisions based on the network information within one hop, which undoubtedly increases the possibility of packet loss. In recent years, software-defined vehicular networks (SDVNs) have attracted wide attention. The design of separating the control plane from the data plane allows the controller to supervise all logical functions and make decisions on the data plane. The data plane consists of vehicles and base stations. Compared with traditional distributed routing methods with local optimality, the routing protocol based on SDVN can obtain better routing through global network information. In SDVN, the controller obtains network topology information through beacon messages periodically sent by vehicles. However, the link state may change during the beacon time interval. A node may select a neighbor outside the transmission range as the next hop for data transmission, which is called link information expiration. Worse still, if the controller does not obtain the link state information in time, the failed link can be reused. In order to enable the controller to quickly obtain the link state, most SDVN-based routing protocols assume that the controller can obtain the link state information in real time, so as to set the beacon update frequency large enough. However, too large a beacon update frequency will result in huge overhead and reduced throughput. In addition, beacon information is likely to be lost.

[0003] To solve the problem of expired link information, the following practices currently exist in the prior art: Some researchers consider using adaptive beacons to detect failed links in the network; some authors proposed a Competitive Adaptive Position Update (CAPU) algorithm for routing. In CAPU, if the predicted position and the true position of the next hop exceed a predetermined threshold, the vehicle will send a beacon message (including position and speed) for update to ensure the correct local topology. Some people also proposed a dynamic beacon scheme to improve the accuracy of node neighbor information by changing the beacon interval. The beacon interval mainly depends on the moving speed of the node and the number of adjacent nodes. Some authors proposed an Adaptive Beacon-based Opportunistic Routing (ABOR) scheme, which updates beacons based on two rules. First, according to the minimum link lifetime between nodes and neighbors, the link lifetime between predicted nodes and the time to send the next beacon message are set. Second, the beacon interval also changes with the change of the forwarding set. Although the adaptive beacon scheme can solve the link failure problem to a certain extent, too high a beacon frequency will result in a large overhead, and data packets will still be lost within the beacon update interval. More importantly, these adaptive beacon schemes are designed based on a distributed network and are not suitable for SDVN.

[0004] Different from the adaptive beacon scheme that actively updates the link state, many studies focus on link failures caused by malicious nodes. These schemes reduce the data packet loss rate by isolating malicious nodes in the network. Some authors introduced a homomorphic linear verifier to determine the real cause of packet loss. Since this scheme is collusion-proof, the source node needs to have high computing power. Although this scheme can identify the real cause of packet loss, it is only applicable to static networks. Some authors introduced a statistical technique to mitigate malicious nodes in mobile ad hoc networks. This scheme uses non-parametric statistical methods to determine malicious nodes in mobile ad hoc networks. Some authors proposed a malicious node detection scheme based on the packet delivery rate information of the base station. First, a packet transmission scheme was proposed to ensure that other nodes except the BS cannot obtain the content of the packet. Second, some authors proposed a malicious link detection scheme, which can detect the packet loss situation of each link during the routing process. Although the above link detection methods can detect failed links, their main purpose is to detect malicious nodes in the network, so it is difficult to achieve timeliness and low overhead. In the vehicle network, link failures caused by vehicle mobility deserve more attention, so a scheme that can obtain the link state in a timely manner is needed. Summary of the Invention

[0005] In view of this, the present invention discloses an adaptive link monitoring system for a software-defined vehicle network, enabling the controller to quickly obtain the link state.

[0006] The technical solution provided by the present invention is specifically an adaptive link monitoring system for a software-defined vehicle network, including: a controller, a routing calculation module, a link evaluation module, an adaptive threshold module, and a link detection module;

[0007] After receiving a routing request, the routing calculation module is used to calculate a route according to its built-in routing algorithm; the link evaluation module is used to evaluate all links on the routing path and obtain an evaluation value of the link quality by using a fuzzy logic method; the adaptive threshold module uses the Q-learning method to determine a detection threshold according to the link state, and determines the link detection range by comparing the detection threshold with the link evaluation value; the link detection module is used to detect packet loss on the link and send a detection report to the controller; the controller is used to obtain the link state in a timely manner by selectively detecting the links on the route.

[0008] Further, the fuzzy logic method adopted in the link evaluation module is: obtaining the quality evaluation of the link by considering the position, direction, speed, and link load of the vehicle.

[0009] Further, the link evaluation module obtaining the evaluation value of the link quality by using the fuzzy logic method specifically includes the following:

[0010] 1) Calculating link failure factors, including link availability time, link stability factor, and link load;

[0011] Link availability time: Assuming that the direction and speed of the vehicle are constant, the link lifetime is calculated as:

[0012]

[0013] a = v i cosθ i -v j cosθ j

[0014] b = x i -x j

[0015] c = v i sinθ i -v j sinθ j

[0016] d = y i -y j

[0017] In Equation (1), v i , v j are the speeds of vehicles i and j, θ i , θj is the moving direction of vehicles i and j; (x i , y i ) and (x j , y j ) are the coordinates of vehicles i and j respectively; Since there will be a delay in the process of transmitting data packets, it is also necessary to consider the order of the links in the routing table. The link available time (LAT) represents the remaining life cycle of the link when the data packet arrives; LAT can be expressed as

[0018] LAT i,j = LET i,j - delay i,j (2)

[0019] where the delay includes data packet processing delay, processing delay, contention delay, and queuing delay;

[0020] The link stability factor: Link stability is determined according to the change in link distance; If the distance of the link changes less within a given time period, the link is more stable, and vice versa. Therefore, the link stability factor can be expressed as:

[0021]

[0022] where is the distance between vehicles i and j at time t0. is the distance between vehicles i and j at time t1. r is the transmission range of the vehicle.

[0023] The link load factor: It is calculated according to the MAC layer interface queue length of the node; Assume that each vehicle regularly sends the interface queue length to the controller through beacon messages. Taking vehicle A as an example, q i is the i-th sample value representing the queue length at the current moment, and N is the total number of queue length samples collected during the whole process. Therefore, the average traffic load of vehicle A can be expressed as

[0024]

[0025] 2) Fuzzification, mapping, and combination of IF-THEN rules: The fuzzification is the process of converting values into fuzzy values through membership functions. The three factors of link available time, link stability factor, and link load are defined as three levels: {low, middle, high}; After calculating the fuzzy values of link available time, link stability factor, and link load, the controller uses the predefined IF-THEN rule combination to map the fuzzy values into the fuzzy output, representing the weight of the link; The linguistic variables of the weight are defined as {Perfect, Good, Acceptable, Unprefect, Bad, Verybad}.

[0026] 3) Defuzzification: The output membership function is used to convert the fuzzy value into a numerical value, and the conversion process is called defuzzification; the centroid method (COG) is used for defuzzification processing.

[0027] Further, a detection threshold is set in the threshold selection module to determine which links need to be detected; if the fuzzy evaluation value of a link is less than the threshold, the link is detected; Q-learning is used to dynamically adjust the detection threshold, and the RL model is defined as <S, A, R>, consisting of a set of state spaces S, an action space A, and a reward function R.

[0028] Further, the state space: The environment around the link is used as the state of reinforcement learning; different average vehicle speeds and vehicle densities within the link transmission range are divided into different states. On the link monitoring system, vehicle speed and density are continuous values, so the states are separated in the RL model;

[0029] The action space: Due to fuzzy logic evaluation, the link quality is mapped between 0 and 1, and the detection threshold can be a value between 0 and 1. To accelerate the learning process, the action space is defined as a discrete space.

[0030] The reward function: The link detection result is used to define the reward function; the feedback information obtained from link detection can be divided into the following four types: where C1 represents that the fuzzy evaluation value of the link is less than the detection threshold and packet loss occurs; C2 represents that the fuzzy evaluation value of the link is greater than the detection threshold, that is, the link is not detected; C3 represents that the fuzzy evaluation value of the link is less than the detection threshold and the packet is not lost; C4 represents that the fuzzy evaluation value is greater than the detection threshold and the packet is not lost; different reward functions are set for different feedback situations.

[0031] Further, the specific implementation method of the link detection module is as follows: Set χ as the detection value, and selectively detect the link. When the detection value χ is 1, it means detecting the link, and when χ is 0, it means not detecting the link;

[0032] If the link needs to be detected, the link source node will start a timer when sending a data packet to the next hop, and a timeout value is set for the timer The evidence of packet loss is whether the link source node receives an ACK message from the next node within the timeout value Set the value of as follows:

[0033]

[0034] where t trans is the transmission delay, t prop is the processing delay, tq is the queuing delay, Δt is the time to ensure the upper limit of the delay, and Δt = 0.01; by setting a threshold c mis to determine the link state. If the number of packet losses on the link exceeds c within a given time mis , it is determined that the link has failed.

[0035] A link monitoring system for an adaptive software-defined vehicle network provided by the present invention first evaluates the possibility of link failure based on a fuzzy logic-based link quality assessment method. By combining link availability, link stability, and link load factors, an evaluation value of the link quality is obtained.

[0036] An adaptive threshold adjustment method is also adopted to determine the detection range of the link that achieves low detection overhead and high detection ratio. Compared with the fixed detection threshold method, a reinforcement learning method is adopted, which can adaptively adjust the detection threshold according to the detection results of the link.

[0037] Finally, an adaptive link state awareness scheme is adopted to overcome packet loss caused by expired link information in SDVN. This scheme determines the link state by detecting packet losses on the link. Compared with traditional packet loss detection methods, this scheme uses the global information of SDVN for selective detection and reduces the detection overhead by adjusting the detection range.

[0038] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the disclosure of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0041] Figure 1 Schematic diagram of a link monitoring system framework for an adaptive software-defined vehicle network provided by an embodiment of the present invention;

[0042] Figure 2 Membership function graph of link availability provided by an embodiment of the present invention;

[0043] Figure 3 Membership function graph of link stability provided by an embodiment of the present invention;

[0044] Figure 4 It is the membership function graph of the link load provided by the disclosed embodiment of the present invention;

[0045] Figure 5 It is the fuzzy logic output function graph provided by the disclosed embodiment of the present invention;

[0046] Figure 6 It is the schematic diagram of the link detection process provided by the disclosed embodiment of the present invention;

[0047] Figure 7 It is the effect diagram of the data packet delivery rate under different vehicle numbers provided by the disclosed embodiment of the present invention;

[0048] Figure 8 It is the effect diagram of the data packet delivery rate under different S-D distances provided by the disclosed embodiment of the present invention;

[0049] Figure 9 It is the effect diagram of the end-to-end delay under different vehicle numbers provided by the disclosed embodiment of the present invention;

[0050] Figure 10 It is the effect diagram of the end-to-end delay under different S-D distances provided by the disclosed embodiment of the present invention;

[0051] Figure 11 It is the effect diagram of the communication overhead under different vehicle numbers provided by the disclosed embodiment of the present invention;

[0052] Figure 12 It is the effect diagram of the link detection coverage rate under different vehicle numbers provided by the disclosed embodiment of the present invention;

[0053] Figure 13 It is the effect diagram of the link detection coverage rate under different S-D distances provided by the disclosed embodiment of the present invention;

[0054] Figure 14 It is the effect diagram of the link detection false positive rate under different vehicle numbers provided by the disclosed embodiment of the present invention;

[0055] Figure 15 It is the effect diagram of the link detection false positive rate under different S-D distances provided by the disclosed embodiment of the present invention. Detailed implementation manners

[0056] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present invention. On the contrary, they are only examples of the system consistent with some aspects of the present invention as detailed in the appended claims.

[0057] Considering the existing technologies, in order to provide a solution that can obtain the link status in a timely manner, this solution should meet the following three conditions. First, the link failure must be detected in a timely manner. If the link failure cannot be detected in a timely manner, reusing the failed link will result in more packet losses. Second, this solution does not generate a huge overhead. For example, if we detect every link in the network, it will undoubtedly affect the network performance because a large amount of bandwidth will be occupied. Finally, due to the highly dynamic characteristics of VANET, this solution needs to adaptively detect the links in the network. The most obvious manifestation of link failure is packet loss at the network layer. Therefore, we can determine the link status by detecting packet loss.

[0058] Specifically, this implementation provides an Adaptive Link Monitoring System for Software-Defined Vehicular Networks (ALPS), which enables the controller to obtain the link status in a timely manner by selectively detecting the links on the route. ALPS determines the link status by detecting packet loss on the link. The schematic diagram of ALPS is as Figure 1 shown. The framework includes four modules: a routing calculation module, a link evaluation module, an adaptive threshold module, and a link detection module. First, after receiving a routing request, the routing calculation module calculates the route according to the built-in routing algorithm. Then, the link evaluation module evaluates all the links on the routing path. It combines link load, link stability, and link availability, and uses the fuzzy logic method to obtain the evaluation value of the link quality. Then, the adaptive threshold module determines the detection threshold according to the link status. The Q-learning method is used to adaptively select the detection threshold. By comparing the detection threshold with the link evaluation value, the range of link detection is determined. Finally, the link detection module detects the packet loss on the link and sends the detection report to the controller.

[0059] As Figure 1As shown, vehicle N1 sends a routing request to the controller, and the controller wishes to send data to vehicle N4. After obtaining the routing request, the routing calculation module calculates the route (N1 - N2 - N3 - N4) through the routing algorithm and sends it to the link evaluation module. The link evaluation module evaluates all the links on the routing path using an evaluation method based on fuzzy logic and sends the evaluation results to the adaptive threshold module. The adaptive threshold module determines the detection threshold by querying the Q table, determines the detection range (N3 - N4) by combining the evaluation results, and sends it to the routing calculation module. The routing calculation module adds the detection range to the path reply and sends the routing reply to vehicle N1 through the controller. After receiving the routing reply, vehicle N1 sends the data packet (N1 - N2 - N3 - N4) according to the route. Due to link failure, the data packet is lost on the link (N3 - N4). Since the link (N3 - N4) is within the detection range, the link detection module can detect the data packet loss and send a detection report to the controller. The controller updates the link status according to the link detection result.

[0060] The above link evaluation module: In the problems of this implementation scheme, in addition to the vehicle exceeding the transmission range due to high mobility, queue overflow caused by high data rate is also a reason for link information expiration. Therefore, it is a challenging task to consider all the above problems and their conflicts and uncertainties in the link evaluation. This implementation scheme uses fuzzy logic to solve this problem. The goal is to obtain the link quality evaluation by considering the vehicle's position, direction, speed, and link load.

[0061] (1) Calculation of link failure factors

[0062] Link availability time: First, assuming that the vehicle's direction and speed are constant, we calculate the link lifetime as

[0063]

[0064] a = v i cosθ i -v j cosθ j

[0065] b = x i -x j

[0066] c = v i sinθ i -v j sinθ j

[0067] d = y i -y j

[0068] where, v i and v j are the speeds of vehicles i and j, and θ i and θ j are the moving directions of vehicles i and j. (x i , y i ) and (x j , y j ) are the coordinates of vehicles i and j respectively. Since there will be some delays during the transmission of data packets, the order of the links in the routing table needs to be considered. The Link Available Time (LAT) represents the remaining lifetime of the link when the data packet arrives. We can simply estimate the arrival time of the data packet at each node in the routing table according to the delay. Therefore, LAT can be expressed as

[0069] LAT i,j = LET i,j - delay i,j (2)

[0070] where the delay includes data packet processing delay, processing delay, contention delay, and queuing delay.

[0071] Link stability factor: Link stability is determined according to the change in link distance. If the distance of the link changes less within a given time period, the link is more stable, and vice versa. Therefore, the link stability factor can be expressed as

[0072]

[0073] where is the distance between vehicles i and j at time t0. is the distance between vehicles i and j at time t1. r is the transmission range of the vehicle.

[0074] Link load factor: Calculated according to the queue length of the node's MAC layer interface. In our model, each vehicle regularly sends the interface queue length to the controller through beacon messages. Taking vehicle A as an example, q i is the i-th sample value representing the queue length at the current moment, and N is the total number of queue length samples collected during the whole process. Therefore, the average traffic load of vehicle A can be expressed as

[0075]

[0076] 2) Fuzzification, mapping, and combination of IF-THEN rules: The process of converting values into fuzzy values through the membership function is called fuzzification. The three factors of link available time, link stability factor, and link load are defined as three levels: {low, middle, high} Figure 2 , Figure 3 , Figure 4Fuzzy membership functions for link available time, link stability factor, and link load factor respectively. After calculating the fuzzy values of link available time, link stability factor, and link load, the controller maps the fuzzy values to fuzzy outputs using predefined IF-THEN rule combinations to represent the weights of the links. The linguistic variables of the weights are defined as {Perfect, Good, Acceptable, Unprefect, Bad, Verybad}.

[0077] Table 1: Fuzzy Rules

[0078]

[0079] It should be noted that multiple rules can be applied to the same fuzzy value. Here, the Min-Max method is used to combine the results of multiple rules.

[0080] (3) Defuzzification: Figure 5 The output membership function is given to convert the fuzzy value into a numerical value. The conversion process is called defuzzification. This implementation uses the center of gravity method (COG) for defuzzification.

[0081] Threshold selection module: Although the links on the route are evaluated and the link quality of the links is mapped to [0, 1], how to determine the links that need to be detected is still a problem, and a detection threshold needs to be set to determine which links need to be detected. If the fuzzy evaluation value of the link is less than this threshold, the link is detected. Due to the highly dynamic nature and complex communication environment of the vehicle network, fixed detection threshold selection cannot adapt to this change, thus having a negative impact on the accuracy of link detection. Therefore, Q-learning is used to dynamically adjust the detection threshold. The RL model is defined as (S, A, R), consisting of a set of states S, an action set A, and a reward function R;

[0082] State space: Since the vehicle network has a highly dynamic and complex communication environment, the cause of link failure is not only related to the state of the link itself but also to the surrounding environment, and it is difficult to determine the impact of the surrounding environment on link quality. Therefore, we use the environment around the link as the state of reinforcement learning and divide the different average vehicle speeds and vehicle densities within the link transmission range into different states. In VANET, vehicle speed and density are continuous values, so we separate the states in the RL model.

[0083] Action space: Due to fuzzy logic evaluation, the link quality is mapped between 0 and 1, and the detection threshold can be a value between 0 and 1. To accelerate the learning process, the action space is defined as a discrete space.

[0084] Reward function: We use the link detection results to define the reward function. The feedback information obtained from the link detection can be divided into the following four types. C1 indicates that the fuzzy evaluation value of the link is less than the detection threshold and the data packet is lost. C2 indicates that the fuzzy evaluation value of the link is greater than the detection threshold, that is, the link is not detected. C3 indicates that the fuzzy evaluation value of the link is less than the detection threshold and the message is not lost. C4 indicates that the fuzzy evaluation value is greater than the detection threshold and the message is not lost. Different reward functions are set for different feedback situations.

[0085] Link detection module: After determining the scope of link detection, in order to selectively detect links, χ is set as the detection value. When the detection value χ is 1, it indicates detecting the link, and when χ is 0, it indicates not detecting the link. If a link needs to be detected, the link source node will start a timer when sending a data packet to the next hop. A timeout value is set for the timer. The evidence of packet loss is whether the link source node receives an ACK message from the next node within the timeout value Therefore, setting the value is crucial. If the timer expires before receiving the ACK message, it will wrongly consider that the data packet has been lost. On the contrary, if is too large, packet loss cannot be detected in time. In this implementation, the value is set as follows:

[0086]

[0087] where t trans is the transmission delay, t prop is the processing delay, t q is the queuing delay, Δt is the time to ensure the delay upper limit, and Δt = 0.01. It should be noted that packet loss does not mean the link is unavailable. Packets may also be lost for other reasons (for example, due to the busy data channel of the forwarding node exceeding the maximum backoff limit and due to interference from hidden nodes resulting in transmission errors). Therefore, the link state is determined by setting a threshold c mis If the number of lost packets on the link exceeds c mis within a given time, we determine that the link fails.

[0088] As Figure 6 shown, the link (A - B), as one of the links in the routing path, respectively describes the link detection processes from the sender and receiver of the data packet. When vehicle A sends a data packet to vehicle B, vehicle A determines whether to detect the link (A - B) according to the value of χ1. If χ1 is 1, vehicle A starts a timer when sending the data packet. If vehicle A is in If it does not receive an ACK message from vehicle B within the timeout value, it will determine that the data packet has been lost and record the data packet loss value on the link (A - B) as 1.

[0089] When vehicle B receives a data packet from A, it decides whether to send an ACK message to vehicle A based on the value of χ1. If the value of χ1 is 0, it means that the link (A - B) does not need to be detected, and vehicle B will not send an ACK message to A.

[0090] The present invention will be further explained below in conjunction with specific embodiments, but it is not used to limit the protection scope of the present invention.

[0091] Python is used as the programming language to complete the experiment simulation. The experimental topology map is Tiexi District, Shenyang City, China. The size of the map is set to 2500 meters × 1500 meters. The map data is sourced from OpenStreetMap. We use SUMO to obtain the trajectory data of vehicles. The transmission range of vehicles is set to 300 meters. Every second, 10 pairs of vehicles are randomly selected as the source node and the destination node in the path, and the source node sends 4 data packets to the destination node every second. For comparison, we uniformly use the Dijkstra algorithm as the routing algorithm to calculate the routing.

[0092] Evaluation Metrics

[0093] (1) Packet Delivery Ratio (PDR). The ratio of data packets successfully delivered to the destination node.

[0094] (2) Average End - to - End Delay (AD). The average end - to - end delay is the average amount of time it takes for a data packet to be successfully sent from the source to the destination.

[0095] (3) Communication Overhead (CO). Communication overhead refers to the number of data packets sent to maintain the network topology. In this article, this data packet includes beacon messages, ACK messages, and detection reports.

[0096] (4) Link Detection Coverage Rate (CR). The link detection coverage rate is the ratio of the detected unavailable links to the actual unavailable links.

[0097] (5) False Positive Rate (FPR). The ratio of the links that successfully transmit data packets within the detection range to all the links that successfully transmit data packets.

[0098] To analyze the performance of ALPS in different environments, we established two groups of experiments.

[0099] (1) Different numbers of vehicles. To analyze the influence of different network densities, the number of road vehicles is set to 100, 200, 300, 400, and 500 vehicles respectively.

[0100] (2) Different S-D distances. To analyze the impact of different S-D distances, in the simulation, the S-D distances were set to 500m, 1000m, 1500m, 2000m, and 2500m respectively.

[0101] Comparison method: We attached ALPS to the Fixed Beaconing scheme (FB) and the Adaptive Beaconing scheme (ABOR). Most SDVN routing protocols adopt the FB scheme, where vehicles periodically send beacon information to the controller. The beacon frequency of the ABOR scheme varies according to three factors: vehicle movement, changes in the vehicle topology, and the number of vehicles in the forwarding concentration. To make the ABOR scheme suitable for the SDVN network, vehicles will send beacon information to the controller at an adaptive frequency instead of broadcasting beacon information to the surrounding environment. In the experiment, we set the fixed beacon interval to 1s and the minimum interval of the adaptive beacon to 0.2s. In addition, to evaluate the adaptive threshold method, we compared it with the fixed threshold method.

[0102] Evaluation of PDR: The packet delivery ratio under different vehicle densities is as Figure 7 shown. When the network density is low, the PDR is usually also low. When the number of vehicles is 100, the vehicle distribution is sparse, and nodes may be far from each other, making it impossible for vehicles to find an available route to the destination. As the vehicle density increases, this situation will be improved. With the addition of ALPS, the PDR has been significantly improved for both the FB scheme and the ABOR scheme. This is because ALPS can detect failed links in a timely manner, thus avoiding the use of unavailable links to transmit data.

[0103] Figure 8 shows the PDR of different S-D distances among 500 vehicles. As the S-D distance increases, the PDR of all schemes decreases. The reason is that the greater the transmission distance, the more hops the packet needs to reach the target node, thus increasing the risk of packet loss. It can be seen from the experimental results that in different situations, after adding ALPS, the value of PDR has increased by 5% - 15%. Evaluation of AD: The average end-to-end delay under different vehicle densities is as Figure 9 shown. As shown, as the number of vehicles increases, the AD of all schemes increases. The reason is that the calculation time of the Dijkstra algorithm is positively correlated with the network density; that is, the more nodes in the network, the longer the calculation time. After adding ALPS, the AD of neither the FB scheme nor the ABOR scheme has increased significantly. This is caused by the following reasons. First, ALPS needs to evaluate the links in the route, which undoubtedly increases the calculation time of the route. However, since ALPS can obtain the link status in a timely manner, the number of lost packets is reduced, thus reducing the number of packet retransmissions.Figure 10 Shows the AD for different S-D distances among 500 vehicles. As the S-D distance increases, the AD of all scenarios also increases. This is because the greater the transmission distance, the more hops a data packet needs to reach the destination. Similar to the case of different vehicle densities, the increase in ALPS has no serious impact on the AD.

[0104] CO evaluation: Figure 11 Shows the communication overhead under different vehicle densities. As the number of vehicles increases, the CO of all scenarios shows an upward trend. The reason is that the more vehicles there are, the more beacons are needed to maintain the global network information in the controller. As shown in the figure, the CO of FB is positively correlated with the number. In contrast, the CO of the ABOR scenario does not change significantly. This is because the ABOR scenario mainly updates the beacon message according to the speed of the vehicle, and not all vehicles need to send beacons regularly. Compared with the FB and ABOR scenarios, as ALPS increases, the communication overhead increases slightly. This is caused by the following reasons. First, the ALPS scenario is added as an additional scenario to FB and ABOR. Therefore, it has all the beacon messages of the original scenarios. In addition, the ALPS scenario uses vehicles to send ACK messages to detect the links in the road and sends the detection report to the controller after determining the link failure. As shown in the figure, we can see that compared with the CO caused by beacon information, the additional overhead caused by ALPS is acceptable.

[0105] CR evaluation: Figure 12 Shows the link detection coverage under different vehicle densities. As the detection threshold increases, the CR also increases. This is because the larger the detection threshold, the more links are detected. Compared with the fixed threshold, the CR of the adaptive threshold is only slightly lower than the maximum threshold of 0.7 in the experiment, which indicates that the adaptive threshold can obtain better detection results through reinforcement learning. The CR for different S-D distances is as Figure 13 shown. As the transmission distance increases, the CR shows a downward trend. The main reason is that the greater the transmission distance, the more hops the data needs to reach the destination, which increases the risk of link unavailability. Similar to the case of different vehicle densities, the CR of the adaptive threshold is only slightly lower than the maximum threshold of 0.7 in the experiment.

[0106] FPR evaluation: Figure 14The evaluation of [[ID=]] shows the false positive rate for different numbers of vehicles. As the detection threshold increases, the FPR also increases. The main reason is that the larger the detection threshold, the more links without lost data packets in the detection range, and the higher the FPR. Compared with the fixed threshold, the FPR of the adaptive threshold is between 0.6 and 0.7. When the threshold is 0.6, the FPR is closer to the FPR. This is because the adaptive threshold not only considers one factor of the FPR, but also ensures a relatively high CR, so it does not reach the minimum FPR. Figure 15 shows the FPR at different S-D distances. Similar to the case of different vehicle densities, the FPR of the adaptive threshold is not optimal. By comparing Figure 13 the CR of different threshold methods in [[ID=]], we can find that the adaptive threshold method can obtain a relatively high CR at a lower FPR, which cannot be achieved by the fixed threshold method.

[0107] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include known common knowledge or conventional technical means in the technical field not disclosed by the present invention. The specification and examples are only to be considered exemplary, and the true scope and spirit of the present invention are pointed out by the claims.

Claims

1. An adaptive link monitoring system for software-defined vehicle networks, characterized in that, Including: A controller, a routing calculation module, a link evaluation module, an adaptive threshold module, and a link detection module; After receiving a routing request, the routing calculation module is used to calculate a route according to its built-in routing algorithm; the link evaluation module is used to evaluate all links on the routing path and obtain an evaluation value of the link quality by using a fuzzy logic method; the adaptive threshold module uses the Q-learning method to determine a detection threshold according to the state of the link, and determines the link detection range by comparing the detection threshold with the link evaluation value; the link detection module is used to detect packet loss on the link and send a detection report to the controller; the controller is used to timely obtain the link state by selectively detecting the links on the route; The fuzzy logic method adopted in the link evaluation module is: obtaining the link quality evaluation by considering the position, direction, speed and link load of the vehicle; The link evaluation module obtaining the evaluation value of the link quality by using the fuzzy logic method specifically includes the following: 1) Calculating link failure factors, including link availability time, link stability factor, and link load; Link availability time: Assuming that the direction and speed of the vehicle are constant, the link lifetime is calculated as: In formula (1), v i , v j are the speeds of vehicles i and j, θ i , θ j are the moving directions of vehicles i and j; (x i , y i ) and (x j , y j ) are the coordinates of vehicles i and j respectively; r represents the transmission range of the vehicle; since there will be a delay in the process of transmitting data packets, it is also necessary to consider the order of the links in the routing table, and the link available time (LAT) represents the remaining life cycle of the link when the data packet arrives; LAT can be expressed as follows: LAT i,j = LET i,j - delay i,j (2) where the delay includes packet processing delay, processing delay, contention delay, and queuing delay; The link stability factor: The link stability is determined according to the change of the link distance; if the distance of the link changes less within a given time period, the link is more stable. Therefore, the link stability factor can be expressed as: Among them, is the distance between vehicles i and j at time t0, is the distance between vehicles i and j at time t1, and r is the transmission range of the vehicle; The link load factor: is calculated based on the MAC layer interface queue length of the node; assume that each vehicle regularly sends the interface queue length to the controller via beacon messages. Taking vehicle A as an example, q i is the i-th sample value representing the queue length at the current moment, and N is the total number of queue length samples collected during the entire process. Therefore, the average traffic load of vehicle A can be expressed as: 2) Fuzzification, mapping and combining IF-THEN rules: The fuzzification is a process of converting values into fuzzy values through a membership function. The three factors of link available time, link stability factor, and link load are defined as three levels of {low, middle, high}; after calculating the fuzzy values of link available time, link stability factor, and link load, the controller uses a predefined combination of IF-THEN rules to map the fuzzy values into a fuzzy output, representing the weight of the link; the linguistic variables of the weight are defined as {Perfect, Good, Acceptable, Unprefect, Bad, Verybad}; 3) Defuzzification: An output membership function is adopted to convert the fuzzy value into a numerical value, and the conversion process is called defuzzification; the center of gravity method (COG) is used for defuzzification processing; A detection threshold is set in the adaptive threshold module to determine which links need to be detected; if the fuzzy evaluation value of the link is less than the threshold, the link is detected; Q-learning is used to dynamically adjust the detection threshold, and the RL model is defined as <S, A, R>, consisting of a set of state spaces S, an action space A, and a reward function R; The state space: Using the environment around the link as the state of reinforcement learning; different average vehicle speeds v and vehicle densities p within the link transmission range are divided into different states. On the link monitoring system, the vehicle speed and density are continuous values, so the states are separated in the RL model; The action space: Due to fuzzy logic evaluation, the link quality mapping is between 0 and 1, and the detection threshold can be a value between 0 and 1. To accelerate the learning process, the action space is defined as a discrete space; The reward function: The link detection result is used to define the reward function; The feedback information obtained from link detection can be divided into the following four types: where C1 represents that the fuzzy evaluation value of the link is less than the detection threshold and the data packet is lost; C2 represents that the fuzzy evaluation value of the link is greater than the detection threshold, that is, the link is not detected; C3 represents that the fuzzy evaluation value of the link is less than the detection threshold and the message is not lost; C4 represents that the fuzzy evaluation value is greater than the detection threshold and the message is not lost; Different reward functions are set for different feedback situations; The specific implementation method of the link detection module is: Set χ as the detection value to selectively detect the link. When the detection value χ is 1, it means detecting the link, and when χ is 0, it means not detecting the link; If a link needs to be detected, the link source node will start a timer when sending a data packet to the next hop and set a timeout value for the timer Evidence of packet loss is whether the link source node receives an ACK message from the next node within the timeout value, and will set the value as follows: where t trans is the transmission delay, t prop is the processing delay, t q is the queuing delay, Δt is the time to ensure the upper limit of the delay, Δt = 0.01; by setting a threshold c mis to determine the link state, if the number of packet losses on the link exceeds c mis within a given time, determine that the link has failed.

Images