An integrated prediction method for adaptive routing in airborne ad hoc networks

Abstract

This invention discloses an adaptive routing integrated prediction method for airborne ad hoc networks, which belongs to the field of communications technology. The integrated prediction method of the present invention uses a long short-term memory (LSTM) model to predict and obtain the maneuverability, available buffer size, and link timeout (LET) of each neighboring drone (NU), avoiding drones with high maneuverability, high flow rate, and weak links to establish an appropriate path. The routing decision problem is expressed as an optimization problem, and the proposed entropy weighted multi-metric (EWMM) method is used to quickly make integrated routing decisions. The proposed integrated prediction and decision process takes into account multi-metric factors that may cause current and future packet loss or delay. Simulation results prove the effectiveness of the LSTM integrated prediction (JP) model, demonstrating that the JPE protocol outperforms the PAP and SPA protocols.

Description

[Technical Field]

[0001] The present invention relates to an adaptive routing and combined prediction method for airborne ad hoc networks, and belongs to the field of communication technology. [Background technology]

[0002] Drones (unmanned aerial vehicles) have become increasingly popular in recent years for military and civilian applications such as aerial photography, communications, agriculture, and search and rescue. FANETs enable coordination and cooperation between drones, enabling them to perform missions that would be difficult or impossible with a single drone.

[0003] Although several studies have proposed routing solutions for FANETs, ​​these conventional methods have limitations in meeting the specific requirements of FANETs in terms of routing, maintenance, and decision-making. These limitations include the following: First, traditional topology-based routing protocols have difficulty maintaining and updating routing paths and routing tables due to the large number of nodes, complex connections, and dynamic topology. Second, recent reinforcement learning-based routing protocols have issues such as long search and promotion times, slow convergence speed, inaccurate state modeling, and poor scalability. Third, swarm intelligence-based routing protocols tend to focus on finding quadratic optimal solutions, making it difficult to model actual fitness functions and set termination conditions. Finally, many existing prediction-based routing methods have difficulty adaptively predicting multiple metrics and making routing decisions that jointly consider multiple metrics. Therefore, these limitations of existing routing methods further increase the demand for more advanced and innovative routing algorithms to effectively handle the problems of FANETs. Summary of the Invention [Problem to be solved by the invention]

[0004] In view of the above problems, the present invention provides an adaptive routing integrated prediction method for flying ad hoc networks, which improves the reliability of transmission links by predicting mobility and data traffic using multi-metric prediction and decision-making, and finds suitable routing paths using an entropy weight-based method. [Means for solving the problem]

[0005] In order to solve the above-mentioned problems, the present invention proposes the following solutions.

[0006] The present invention provides an adaptive routing integrated prediction method for airborne ad hoc networks, which includes the following steps:

[0007] Step S1: We construct a 3D flying ad-hoc network, i.e., FANETS, consisting of normal drones and disaster area drones. The drones follow the same Gauss-Markov migration model to determine the drone's position at time t. JPEG2025536585000002.jpg621, Formula (1): JPEG2025536585000003.jpg771; Formula (2): JPEG2025536585000004.jpg771; Formula (3): JPEG2025536585000005.jpg750; In the formula: JPEG2025536585000006.jpg43 is the drone speed, S t is the drone's speed at time t, JPEG2025536585000007.jpg10128 is horizontal, JPEG2025536585000008.jpg981 is the pitch angle, the pitch angle at time t JPEG2025536585000009.jpg1077, JPEG2025536585000010.jpg43, For JPEG2025536585000011.jpg43 and for JPEG2025536585000012.jpg43, t is the iteration, step size JPEG2025536585000013.jpg33 is within the range of [0, 1], JPEG2025536585000014.jpg43, JPEG2025536585000015.jpg43 and JPEG2025536585000016.jpg43 is the average value, JPEG2025536585000017.jpg69, JPEG2025536585000018.jpg69 and JPEG2025536585000019.jpg59 are random variables of the Gaussian distribution. Here, it is assumed that each drone has a buffer of the same length and packet loss occurs due to buffer overflow.

[0008] Step S2: Drone set, routing request, history sequence Input JPEG2025536585000020.jpg44, and the execution time T. When the system time t < T, the adaptive prediction of JPEG2025536585000021.jpg49 of the drone is performed by the long short-term memory integrated prediction module to find an appropriate transmission path. JPEG2025536585000021.jpg49 of the drone is adaptively predicted to find an appropriate transmission path.

[0009] Step S3: When one drone receives the routing request execution, it starts routing using the shortest path method in Equation (6).

[0010] Step S4: Use the timeout timer and the acknowledgment response ACKs to control the routing time and retransmission. When the drone receives the ACK message, the routing is successful. When there is a timeout, routing is performed again. Otherwise, the routing fails and the loop from Step S3 to Step S4 ends.

[0011] Step S5: Update t, return to step S2 and execute the steps above until the loop is completed.

[0012] Preferably, an integrated long short-term memory prediction module is utilized to predict available buffer size, transmission rate, and link expiration time.

[0013] Preferably, in step S2, the drone to be predicted using the long short-term memory integrated prediction module The index contents of JPEG2025536585000022.jpg49 are: JPEG2025536585000023.jpg771 The contents of the discrete-time queueing system of neighboring drones (NU) are integer time slots. JPEG2025536585000024.jpg433 JPEG2025536585000025.jpg58, the initial state of the system JPEG2025536585000026.jpg610 is a non-negative real random number, and the future state is the arrival and transmission process of a random packet. JPEG2025536585000027.jpg57 and JPEG2025536585000028.jpg57, and packet arrival is represented by a random process, JPEG2025536585000029.jpg718 is a random time slot JPEG2025536585000030.jpg433 is a real-valued random variable sequence defined in . JPEG2025536585000031.jpg57 is the NU queue slot This is the number of packets that can be processed by JPEG2025536585000032.jpg42. JPEG2025536585000033.jpg57 is the slot JPEG2025536585000034.jpg42 is the number of new packets arriving, JPEG2025536585000035.jpg57 is assumed to be non-negative. JPEG2025536585000036.jpg54 is the amount of data sent, and LET is the time difference. JPEG2025536585000037.jpg619 is calculated. JPEG2025536585000038.jpg44 is the arriving packet, calculated as the product of the predicted PAR and the time interval JPEG2025536585000039.jpg55 is defined in LET and the channel transmission rate JPEG2025536585000040.jpg57 It is represented as JPEG2025536585000041.jpg1764, In the formula: JPEG2025536585000042.jpg45 is the current (current) bandwidth of the drone (CU), JPEG2025536585000043.jpg44 is the transmission power of the CU (transmission power), JPEG2025536585000044.jpg43 is a log-normal shadow component, with an average value of 0 dB and a standard deviation of JPEG2025536585000045.jpg44, JPEG2025536585000046.jpg513 is the fading width between CU c and NU i. Rayleigh distribution modeling is adopted, and the expected value is JPEG2025536585000047.jpg728, JPEG2025536585000048.jpg56 is the power of additive Gaussian white noise, λ is the path fading exponent, λ is generally chosen in the range [2, 4], JPEG2025536585000049.jpg68 represents the distance between CU c and NU i, Next, LET It is displayed as JPEG2025536585000050.jpg27128, In the formula, R is the communication range of the drone, JPEG2025536585000051.jpg35128JPEG2025536585000052.jpg45 and JPEG2025536585000053.jpg45 is the speed value, JPEG2025536585000054.jpg521 is the horizontal value, JPEG2025536585000055.jpg45 and JPEG2025536585000056.jpg45 is the pitch angle value, m and n represent CU and NU, and the routing decision is more favorable for higher LK.

[0014] Preferably, in step S2, finding a suitable route is to use a multi-metric method based on entropy weighting to evaluate the suitability of the routing route, which includes the following content: first, normalize each decision-making factor as follows: In the JPEG2025536585000057.jpg1772 format, JPEG2025536585000058.jpg55 is the decision metric of the jth metric of the i-th NU, where the metric includes transmission rate, link duration, NU location, NU speed, etc. In the formula, N is the number of units, M is the number of decision indicators, JPEG2025536585000059.jpg44 is the efficiency coefficient that controls the numerical acquisition range of the jth decision index, and JPEG2025536585000060.jpg1419, The CU obtains the entropy of the numerical acquisition range of the jth decision index by the following formula: JPEG2025536585000061.jpg14128; Here, JPEG2025536585000062.jpg55 represents the jth network metric ratio of NUi, JPEG2025536585000063.jpg1161; The entropy weighting of the jth decision metric is defined as follows: JPEG2025536585000064.jpg1275; Based on the normalized value and metric entropy obtained above, the availability of the i-th NU acquired by the CU is as follows: JPEG2025536585000065.jpg1472; The transmission probability of the i-th NU that the CU acquires is as follows: JPEG2025536585000066.jpg1178; The above five equations convert the NU-metric matrix into a decision probability vector: Formula (4): JPEG2025536585000067.jpg1755; If some NUs are inconsistent in the decision vector of the above formula (4), the CU calculates the required time by dividing the current packet size by the transmission rate. If the required time is smaller than the link expiration time LET, the CU deletes the NU and routes it. At the same time, the CU deletes the NU with an available buffer size smaller than the PAR timer link expiration time LET, or makes it out of the communication range within LET. After the deletion operation, the remaining NUs are named candidate NUs, i.e., the CNU set. Based on the SDN control system, the centralized controller CC gives the most probable routing decision on the right-hand side of Equation (4), JPEG2025536585000068.jpg4385; In the formula, N BS is the number of base stations, JPEG2025536585000069.jpg613 is a CNU collection Candidate path in JPEG2025536585000070.jpg43 The adaptive routing integrated prediction method for flying ad hoc networks as described in claim 3, characterized in that the probability is JPEG2025536585000071.jpg43, and reliable routing is obtained by solving the shortest path algorithm using the above formula.

[0015] The computational complexity of the JPE routing protocol is JPEG2025536585000072.jpg745, JPEG2025536585000073.jpg45 is the length of the input sequence of the LSTM-based JP module. JPEG2025536585000074.jpg44 is the maximum number of NUs, E is the number of drones in FANETS, V is the number of connections in FANETS, JPEG2025536585000075.jpg34 is the time interval between routing requests. [Effects of the Invention]

[0016] Compared with the prior art, the technical solution of the present invention has the following advantages:

[0017] This invention is based on a joint prediction entropy weighting (JPE) routing protocol, which predicts drone link mobility and data traffic through a joint prediction module of a long short-term memory model, forming the basis for routing decisions. The routing decision problem is expressed as an optimization problem, and the proposed entropy weighting multi-metric (EWMM) method can be used to quickly make joint routing decisions. The JPE protocol, a solution for FANETs provided by this invention, is adaptive to traffic flow, mobility prediction, routing decisions, and actual assignments. Furthermore, the joint prediction and decision process takes into account multi-metric factors that cause current and future packet loss or delay, thereby improving the reliability of transmission links. [Brief explanation of the drawings]

[0018] [Figure 1] FIG. 1 is an architecture diagram of the LSTM-based JP module of the present invention. [Figure 2] 10 is a flowchart illustrating the operation of the JPE routing protocol of the present invention. [Figure 3] FIG. 10 is a diagram comparing delay performance with different numbers of drones in an embodiment of the present invention. [Figure 4]FIG. 10 is a comparison diagram of delay performance for different CRs (unit: m) in an embodiment of the present invention. [Figure 5] FIG. 10 is a comparison diagram of delay performance at different AVs (unit: m / s) in an embodiment of the present invention. [Figure 6] FIG. 10 is a comparison diagram of delay performance in the case of different APARs (unit: Kb / s) according to an embodiment of the present invention. [Figure 7] This is a delay box diagram when APAR, AV, and CR are set to 10 Kb / s, 5 m / s, and 28 m, and the number of drones is changed in an embodiment of the present invention. [Figure 8] FIG. 10 is a delay box diagram for different CRs (unit: m) according to an embodiment of the present invention. [Figure 9] FIG. 10 is a delay box diagram for different AVs (unit: m / s) according to an embodiment of the present invention. [Figure 10] FIG. 10 is a delay box diagram for different APARs (unit: Kb / s) in an embodiment of the present invention. [Figure 11] FIG. 10 is a diagram comparing PDR performance with different numbers of drones in an embodiment of the present invention. [Figure 12] FIG. 10 is a diagram comparing PDR performance in the case of different CRs (unit: m) according to an embodiment of the present invention. [Figure 13] FIG. 10 is a diagram comparing PDR performance in the case of different AVs (unit: m / s) according to an embodiment of the present invention. [Figure 14] FIG. 10 is a diagram comparing PDR performance at different APARs (unit: Kb / s) in an embodiment of the present invention. [Figure 15] 10A-10C are PDR box diagrams with different drone numbers in an embodiment of the present invention. [Figure 16] FIG. 1 is a PDR box diagram for different CRs (unit: m) according to an embodiment of the present invention. [Figure 17] FIG. 1 is a PDR box diagram for different AVs (unit: m / s) according to an embodiment of the present invention. [Figure 18] FIG. 1 is a PDR box diagram for different APARs (unit: Kb / s) according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0019] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described in conjunction with the following specific embodiments and corresponding drawings, which are merely some of the embodiments rather than all of the embodiments of the present invention.

[0020] The present invention provides an integrated prediction method for adaptive routing in flying ad-hoc networks, including an adaptive and integrated prediction-driven routing decision scheme. As shown in Figure 2, the method includes the following steps:

[0021] Step S1: Construct a 3D flying ad hoc network (i.e., FANETS) consisting of normal drones (unmanned aerial vehicles) and drones (unmanned aerial vehicles) in disaster areas. The drones behave according to a similar Gauss-Markov transition model, and the drone's position at time t is This is expressed as JPEG2025536585000076.jpg621, and the following equation is obtained. Formula (1): JPEG2025536585000077.jpg771; Formula (2): JPEG2025536585000078.jpg771; Formula (3): JPEG2025536585000079.jpg750; In the formula: JPEG2025536585000080.jpg43 is the speed of the drone, JPEG2025536585000081.jpg10128 is horizontal, JPEG2025536585000082.jpg981 is the pitch angle, JPEG2025536585000083.jpg1077. JPEG2025536585000084.jpg43、 JPEG2025536585000085.jpg43 and For JPEG2025536585000086.jpg43, t is the iteration, step size For JPEG2025536585000087.jpg33 within the range of [0, 1], JPEG2025536585000088.jpg43, JPEG2025536585000089.jpg43 and JPEG2025536585000090.jpg43 is the average value, JPEG2025536585000091.jpg69, JPEG2025536585000092.jpg69 and JPEG2025536585000093.jpg59 are random variables of the Gaussian distribution. Here, it is assumed that each drone has a buffer of the same length and packet loss occurs due to buffer overflow.

[0022] Step S2: Drone set (collection), routing request, history sequence JPEG2025536585000094.jpg44, input the execution time T. If the system time t < T, the JP module of LSTM adaptively predicts the JPEG2025536585000095.jpg49 of the drone and finds an appropriate transmission path.

[0023] Here, the index content of JPEG2025536585000096.jpg49 of the drone predicted by the JP module of LSTM includes the following content. JPEG2025536585000097.jpg771 The content of the discrete-time waiting queue system of the neighboring drones (NU) is the integer time slot JPEG2025536585000098.jpg433 at Represented by JPEG2025536585000099.jpg58. System initial state JPEG2025536585000100.jpg610 is a non-negative real random number. The future state is the arrival and transmission process of a random packet. JPEG2025536585000101.jpg57 and The arrival of a packet is determined by a random process: JPEG2025536585000103.jpg718, this is a random time slot JPEG2025536585000104.jpg433 is a real-valued random variable sequence defined in JPEG2025536585000105.jpg57 is the NU queue slot This is the number of packets that can be processed by JPEG2025536585000106.jpg42. JPEG2025536585000107.jpg57 is the slot JPEG2025536585000108.jpg42 is the number of new packets arriving, JPEG2025536585000109.jpg57, assuming it is non-negative: JPEG2025536585000110.jpg54 is the amount of data sent, and LET is the time difference. It is calculated as JPEG2025536585000111.jpg619. JPEG2025536585000112.jpg44 is the arriving packet, calculated as the product of the predicted PAR and the time interval JPEG2025536585000113.jpg55 is defined in LET and the channel transmission rate JPEG2025536585000114.jpg57 will be displayed as follows. In the JPEG2025536585000115.jpg1764 format, JPEG2025536585000116.jpg45 is the current (current) bandwidth of the drone (CU), JPEG2025536585000117.jpg44 is the transmission power of the CU (transmission power), JPEG2025536585000118.jpg43 is a log-normal shadow component, with an average value of 0 dB and a standard deviation of JPEG2025536585000119.jpg44, JPEG2025536585000120.jpg513 is the fading width between CU c and NU i. Rayleigh distribution modeling is adopted, and the expected value is JPEG2025536585000121.jpg728, JPEG2025536585000122.jpg56 is the power of additive Gaussian white noise, λ is the path fading exponent, λ is generally chosen in the range [2, 4], JPEG2025536585000123.jpg79 represents the distance between CU c and NU i.

[0024] LET is displayed as follows: JPEG2025536585000124.jpg27128; In the formula, R is the communication range of the drone, JPEG2025536585000125.jpg35128JPEG2025536585000126.jpg45 and JPEG2025536585000127.jpg45 is the speed value, JPEG2025536585000128.jpg521 is the horizontal value, JPEG2025536585000129.jpg45 and JPEG2025536585000130.jpg45 is the pitch angle value, m and n represent CU and NU, and the routing decision is more favorable for higher LK.

[0025] The present invention considers the following metrics: 1) available buffer size, 2) transmission rate, and 3) link expiration time (LET). By considering these metrics in the present and future and highlighting irrelevant NUs, the network can automatically and adaptively obtain an appropriate routing path. To make an efficient routing decision, all the aforementioned metrics need to be considered simultaneously.

[0026] To make integrated and fast decision-making, this embodiment proposes to use an entropy weighting-based multi-metric (EWMM) method to measure the suitability of routing paths, which includes the following steps: First, normalize each decision-making factor as follows: In the JPEG2025536585000131.jpg1772 format, JPEG2025536585000132.jpg55 is the decision metric of the jth metric of the ith NU, and the metric includes transmission rate, link duration, NU position, NU rate, etc. In the formula, N is the number of units, M is the number of decision indicators, JPEG2025536585000133.jpg44 is the efficiency coefficient that controls the numerical acquisition range of the jth decision index, and JPEG2025536585000134.jpg1419.

[0027] The CU obtains the entropy of the numerical value acquisition range of the j-th decision index by the following formula. JPEG2025536585000135.jpg14128; Here, JPEG2025536585000136.jpg55 represents the jth network metric ratio of NUi, JPEG2025536585000137.jpg1161; The entropy weighting of the jth decision metric is defined as follows: JPEG2025536585000138.jpg1275; Based on the normalized value and metric entropy obtained above, the availability of the i-th NU acquired by the CU is as follows: JPEG2025536585000139.jpg1472; The transmission probability of the i-th NU that the CU acquires is JPEG2025536585000140.jpg1178, The above five equations convert the NU-metric matrix into a decision probability vector, Formula (4): JPEG2025536585000141.jpg1755; If some NUs are inconsistent in the decision vector of the above formula (4), the CU calculates the required time by dividing the current packet size by the transmission rate, and if the required time is smaller than the link expiration time LET, deletes the NU and performs routing. At the same time, the CU deletes NUs with available buffer sizes smaller than the PAR timer link expiration time LET, or places them outside the communication range within the LET (deletes NUs outside the communication range within the LET). After the deletion operation, the remaining NUs are called candidate NUs, i.e., a set of CNUs. In this embodiment, the following procedure is performed using three NUs to be deleted as an example. JPEG2025536585000142.jpg27128The underlined parts indicate the parts that are deleted to reduce the decision space. After the deletion operation, these remaining NUs are named candidate NUs (CNUs), which can ensure the comprehensive and rational routing decisions of the EWMM method.

[0028] Considering the SDN-based control architecture, the CC can make the routing decision with the highest probability on the right side of equation (4) as follows: Formula (6): In the JPEG2025536585000143.jpg4385 formula, N BS is the number of base stations, JPEG2025536585000144.jpg613 is a CNU collection Candidate path in JPEG2025536585000145.jpg43 JPEG2025536585000146.jpg43 probability,A reliable routing solution can be obtained by solving the shortest path algorithm using the above formula.

[0029] Step S3: When a drone receives a routing request execution, it starts routing using the shortest path method according to Equation (6).

[0030] Step S4: Use the timeout timer and ACKs (acknowledgements / acknowledgements) to control the routing time and retransmissions. If the drone receives an ACK message, the routing is successful; if it times out, it tries routing again; otherwise, the routing fails and the loop from step S3 to step S4 is terminated.

[0031] Step S5: Update t and return to step S2 described above until the loop is completed.

[0032] The JPE routing protocol of the present invention is adaptive in terms of traffic and mobility prediction, routing decision and actual deployment. By choosing to use a responsive approach, the key parameters of the computation path are calculated using a LSTM-based JP module. This effectively demonstrates how the predictive properties support routing by adaptively calculating JPEG2025536585000147.jpg44, eliminating conflicting NUs, and obtaining CNUs. When a drone requests packet transmission, the CUs select the most probable path and use timeout timers and ACKs to control routing time and retransmissions. The computational complexity of the JPE routing protocol can be expressed as follows: JPEG2025536585000148.jpg745Here, JPEG2025536585000149.jpg45 is the length of the input sequence of the LSTM-based JP module. JPEG2025536585000150.jpg44 is the maximum number of NUs, E is the number of drones in FANETS, V is the number of connections in FANETS, JPEG2025536585000151.jpg34 is the time interval between routing requests. This complexity is determined by the size of the NUs, the frequency of routing requests, the number of drones, and their density.

[0033] To verify the effectiveness of the present invention, in this embodiment, the JPE routing protocol (hereinafter referred to as the JPE routing protocol) is compared with the following protocols: The optimal solution is achieved using a brute force search algorithm based on the optimal path solution formula (6). In this case, the performance is the best. The Packet Arrival Prediction Routing Protocol or PAP routes packets with constrained ordering based on predicted traffic and mobility, and eliminates the decision module based on entropy weighting. The Random Packet Transmission Algorithm or SPA routes packets based on the metric entropy decision method of the current topology, and removes future metrics from the decision module. Figures 3 to 18 show the comparison results of experimental simulations.

[0034] The experimental simulation content of the JPE routing protocol of the present invention is as follows.

[0035] The simulation equipment includes a disaster assistance (DA) center, a base station (BS), and multiple drones that must transmit to the BS via hop-by-hop routing or direct transmission. All drones in the simulation experiment fly within the network area using the same Gauss-Markov model and receive packets based on the distance to the data center using the same Poisson packet arrival model. That is, drones move according to the following equations (1) through (3) and receive packets from ground nodes based on the Poisson packet arrival model and APAR. Packets arrive at the drones according to their distance to the DA center, but are constrained by the overall PAR of the FANET. Packets can be lost in the transmission channel due to link interruptions, buffer overflows, or transmission timing errors. Degraded channel quality, expansion of CU queues, or additional hops can also increase latency. The main simulation parameters are listed in Table 1.

[0036] Table 1 Simulation parameters JPEG2025536585000152.jpg68147

[0037] In the experiments, we set the average packet arrival rate (APAR), average drone velocity (AV), and drone communication radius (CR) to 10 kb / s, 5 m / s, and 28 m, respectively, and observed the latency performance by varying the number of drones. The simulation results are shown in Figures 3 and 7. We observe that the latency of the four protocols increases sharply when the number of drones reaches 100, then drops sharply above 100 and increases again above 120. As shown in the box diagram in Figure 7, compared to SPA and PAP, JPE achieves the best latency performance, reaching near-optimal performance. This is because the LSTM-based JP and EWMM routing decision modules are practical and can derive suboptimal solutions that are closest to the optimal solution. In other words, the proposed LSTM-based JP module can adaptively evaluate each link by comprehensively considering mobility and data traffic, and the EWMM routing decision method can output routing decisions that take into account current and future metrics. Furthermore, future metrics are very important, and SPA produces the worst delay. Compared with the comparative SPA solution, the JPE method of this embodiment significantly reduces the delay, reducing it by 66.94% on average.

[0038] For different CRs, in this embodiment, the APAR is set to 10 Kb / s, the AV is set to 5 m / s, and the number of drones is set to 35. The CR ranges from 20 to 80 m. The latency trends were observed. The simulation results are shown in Figures 4 and 8. As the CR increases, the number of connections between drones increases, and the number of drones that lose connection decreases, resulting in an increase in latency before the CR reaches 70 m. However, when a single drone's CR reaches 70 m, it can connect with all drones within a 100 x 100 x 300 area. This is a very high CR, and the NU of the CU becomes very large, resulting in stable connectivity. This significantly increases the number of hops and reduces latency performance. Furthermore, considering LET, JPE achieves optimal latency performance. Compared to SPA, JPE is clearly closer to the optimal value, achieving an average latency improvement of 15.69%. In dynamic scenarios, SPA, which only uses total entropy weighting and does not have adaptive prediction, is unable to make better decisions, resulting in the worst latency performance.

[0039] Figures 5 and 9 show the latency performance of different AVs under the constant conditions of APAR of 10 Kb / s, drone count of 35, and CR of 28 m. The results show that as AVs increase, drone maneuverability and the associated topology drift increase, resulting in a significant increase in latency for all four protocols. Furthermore, a comparison of the box plots in Figure 9 reveals that JPE achieves optimal performance, due to its mobility prediction and EWMM routing decision methods working well and quickly finding a suitable path. As shown in Figures 5 and 9, compared to SPA, JPE can improve latency performance by an average of 68.88%.

[0040] Finally, we plotted the relationship between delay performance and different APARs. Here, we set AV to 5 m / s, the number of drones to 35, and CR to 28 m. The relationship diagrams are shown in Figures 6 and 10. Similarly, as APAR increases, a sudden increase in delay can cause bursts of data flow (a sudden increase in data flow), potentially leading to transmission buffer overflow. Data flow is another important parameter that can cause packet loss, and buffer overflow definitely means packet loss. Therefore, these results demonstrate that the LSTM-based JP module can adequately predict data flow. Furthermore, because the three key indicators reasonably consider the current and future situations and are incorporated into the EWMM method, the method performed well in handling different APARs. Considering the box diagram, Figure 10 shows the overall delay performance. Except for the optimal case, JPE delay is the smallest, with an average 24.52% reduction in delay compared to SPA.

[0041] Regarding PDR performance, the simulation in this embodiment was performed with APAR set to 10 Kb / s, AV set to 5 m / s, and CR set to 28 m, and the number of drones was varied. The results are shown in Figures 11 and 15. The PDR of the four protocols showed three stages of change, with JPE outperforming PAP and SPA. The reason for this is that increasing the number of drones increases the number of connections between drones, which in turn increases the number of hops, ultimately forming stable connections. Among them, JPE showed the optimal PDR. This is because JPE uses prediction and entropy decision-making to consider both current and future indicators, allowing it to avoid low-persistence links with high traffic volume (data flow), high mobility, and many relay NUs, while also considering both current and future indicators. This also confirms that JPE outperforms PAP and SPA for all drones. The box diagram of the four schemes in Figure 15 shows that the PDR of the JPE scheme increased by an average of 30.98% compared to the SPA scheme.

[0042] Figures 13 and 17 show how JPE adapts to different AVs. As shown, the PDR performance of all four protocols decreases as AVs increase. This is because increased drone mobility makes it more difficult to maintain stable connections and avoid packet loss. However, JPE can mitigate this impact by accurately predicting drone maneuverability and making integrated routing decisions based on that information. JPE's ability to improve PDR performance in high-mobility environments is due to its LSTM-based JP module. The JP module takes into account various factors, such as the drone's speed and acceleration, and the environmental wind speed and direction, and can accurately predict drone maneuverability parameters, which are important for maintaining stable connections. The JP module also uses the EWMM routing decision method, which evaluates each drone's available buffer size, transmission rate, and LET to make routing decisions. Furthermore, JPE's mobility prediction capability allows it to derive new topologies. CUs can make routing decisions without losing connectivity. By eliminating the need for a self-consistency unit, JPE improves decision efficiency and accuracy, and combined with JPE's ability to avoid packet loss due to mobility and buffer overflow, JPE becomes a superior routing protocol to PAP and SPA. This is further highlighted in Figure 17, which shows JPE's high value and stable PDR performance, with an average improvement of 27.88% compared to SPA.

[0043] Figure 14 shows the PDR curves for different APARs with the number of drones set to 35, AV set to 5 m / s, and CR set to 28 m. As can be seen, the PDRs of all four protocols decrease as the APAR increases. This is because an increase in APAR makes it difficult for the buffer to efficiently relay more packets. Because APAR is a crucial factor, changing it can cause significant buffer overflow and packet loss. To mitigate this situation, JPE's LSTM-based JP module takes PAR into account, while the EWMM method considers the available buffer size as the basic metric for decision-making. The impact of different APARs on JPE's integration process is minimized. Furthermore, JPE's high-performance and fast prediction capabilities effectively reduce packet loss due to buffer overflow. From the PDRs in Figure 18, JPE outperforms PAP and SPA in terms of rate amplitude and stability. This is because a larger APAR increases data flow, making relay failures (especially PAP and SPA) more likely to occur. Overall, compared to SPA, JPE improved the PDR performance by an average of 43.33%, as shown in Figs. 14 and 18 .

[0044] Finally, this embodiment summarizes the improvements of PAP, JPE, and Optimal, using the SPA protocol as a reference. For easier understanding, this embodiment assumes that delay improvement is the same evaluation metric as PDR, i.e., physical delay reduction to maintain the improvement rate. Furthermore, this embodiment sets the y-axis to multiple scenes. Simulation results show that Optimal outperforms the others, while JPE is suboptimal, and PAP outperforms SPA. Averaging the improvement rates, we conclude that JPE can improve PDR and delay by 30.13% and 31.24%, respectively, compared to SPA.

[0045] The above content is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Without departing from the scope of the technical solution of the present invention, those skilled in the relevant technical field may make any form of substitution or modification to the technical solution and technical content disclosed in the present invention, which will not depart from the content of the technical solution of the present invention and will still fall within the protection scope of the present invention.

Claims

1. 1. An adaptive routing integrated prediction method for an airborne ad hoc network, comprising: Step S1: We construct a three-dimensional flying ad-hoc network, i.e., FANETS, consisting of normal drones and disaster area drones. The drones follow the same Gauss-Markov transition model to determine the position of the drone at time t. year, 、 、 and In the formula: is the speed of the drone, S ｔ is the drone's speed at time t, is horizontal, is the pitch angle, and the pitch angle at time t 、 、 and where t is the iteration and step size is in the range [0, 1], 、 and is the mean value, 、 and is a Gaussian distributed random variable, where it is assumed that each drone has a buffer of the same length and packet loss occurs due to buffer overflow. Step S2: Drone set, routing request, history sequence , and the execution time T are input. If the system time t<T, the drone's and performing adaptive prediction to find a suitable transmission path. Step S3: When one drone receives the routing request execution, it starts routing using the shortest path method according to Equation (6); Step S4: Using the timeout timer and the ACK acknowledgment ACKs to control the routing time and retransmission, if the drone receives the ACK message, the routing is successful, if the timeout occurs, the drone tries to route again, otherwise the routing fails, and the loop from step S3 to step S4 is terminated. Step S5: Update t, return to step S2, and execute the above steps until the loop is completed.

2. The adaptive routing integrated prediction method for an airborne ad hoc network according to claim 1, characterized in that in step S2, an integrated prediction module of long short-term memory is used to predict available buffer size, transmission rate, and link expiration time.

3. In step S2, the drone's prediction is made using the long-short-term memory integrated prediction module. The indicator contents are as follows: The contents of the discrete-time queueing system of neighboring drones (NUs) are integer time slots. in and the initial state of the system is is a non-negative real random number, and the future state is the arrival and transmission process of a random packet. and The arrival of packets is determined by a random process, is a random time slot is a real-valued random variable sequence defined by The queue of NU is slot is the number of packets that can be processed by is a slot is the number of new packets arriving at Assuming that is non-negative, is the amount of transmission, and LET is the time difference. It is calculated as follows: is the arriving packet, calculated as the product of the predicted PAR and the time interval is defined in LET and the transmission rate of the channel teeth is expressed as In the formula: is the current (current) bandwidth of drones (CU), is the transmission power of the CU (transmission power), is the log-normal shadow component, the mean value is 0 dB, and the standard deviation is 、 is the fading width between CU c and NU i, and adopts Rayleigh distribution modeling, and the expected value is and is a power of additive Gaussian white noise, λ is the path fading exponent, λ is typically chosen in the range [2, 4], represents the distance between CU c and NU i; Next, LET It is displayed as In the formula, R is the communication range of the drone, and is the velocity value, is the horizontal value, and The adaptive routing integrated prediction method for flying ad hoc networks according to claim 1, characterized in that m and n represent the pitch angle value, m and n represent CU and NU, and the routing decision is more favorable when LK is higher.

4. In step S2, finding a suitable route is to use a multi-metric method based on entropy weighting to evaluate the suitability of the routing route, which includes the following content: first, normalize each decision-making factor as follows: In the formula: is the decision metric of the jth metric of the i-th NU, where the metric includes transmission rate, link duration, NU location, NU speed, etc. In the formula, N is the number of units, M is the number of decision indicators, is an efficiency coefficient that controls the numerical acquisition range of the jth decision index, and and The CU obtains the entropy of the numerical value acquisition range of the jth decision index by the following formula: ； Here, represents the j-th network metric ratio of NUi, ； The entropy weighting of the jth decision index is defined as follows: ； Based on the normalized value and metric entropy obtained above, the availability of the i-th NU that the CU obtains is: ； The transmission probability of the i-th NU that the CU acquires is: ； The above five equations convert the NU-metric matrix into a decision probability vector: Formula (4): ； If some NUs are inconsistent in the decision vector of the above formula (4), the CU calculates the required time by dividing the current packet size by the transmission rate. If the required time is smaller than the link expiration time LET, the CU deletes the NU and routes it. At the same time, the CU deletes the NU with an available buffer size smaller than the PAR timer link expiration time LET, or makes it out of the communication range within LET. After the deletion operation, the remaining NUs are named candidate NUs, i.e., CNU set. Based on the SDN control system, the centralized controller CC provides the most probable routing decision on the right-hand side of equation (4), ； In the formula, N ＢＳ is the number of base stations, is a CNU set Candidate paths within The adaptive routing integrated prediction method for flying ad hoc networks according to claim 3, characterized in that the probability is [times] = [times] , and the reliable routing is obtained by solving the shortest path algorithm according to the above formula.

5. The computational complexity of the JPE routing protocol is and is the length of the input sequence of the LSTM-based JP module, is the maximum number of NUs, E is the number of drones in FANETS, V is the number of connections in FANETS, 2. The adaptive routing integrated prediction method for airborne ad hoc networks according to claim 1, wherein τ is the time interval between routing requests.

Images