A method and system for predicting the formation routing performance of unmanned aerial vehicles based on a pigeon flock neural network

Abstract

The present invention discloses a method and system for predicting the routing performance of an unmanned aerial vehicle (UAV) formation based on a pigeon flock neural network, including the following steps: S1: According to the acquired historical data, use the pigeon flock neural network to perform offline training on the routing performance prediction model, and solve to obtain an optimized routing performance prediction model; S2: Periodically and real-time sense and obtain the current network situation and UAV formation situation of the UAV formation; S3: Input the obtained current network situation and UAV formation situation into the optimized routing performance prediction model to predict the routing performance of neighbor nodes, and use the predicted routing performance prediction matrix as the basis for routing selection. The present invention does not require the whole network topology as prior knowledge, greatly reducing the control overhead. At the same time, it does not require online training by UAVs, greatly reducing the computational amount.

Description

Technical Field

[0001] The present invention relates to the field of network switching technology, and more specifically, to a method and system for predicting the routing performance of an unmanned aerial vehicle (UAV) formation based on a pigeon flock neural network. Background Art

[0002] With the development of UAV control technology, UAV formations have been widely used in various fields. The cooperative control of UAV formations is based on the ability of UAVs to communicate with each other. Therefore, the routing method of UAV formations has become a research hotspot.

[0003] Due to the frequent relative movement of communication nodes within the UAV formation, and the limitations of airborne communication equipment and CPUs due to UAV payloads, the design of the routing method for UAV formations needs to consider the characteristics of frequent topological changes, insufficient bandwidth, and insufficient computing resources. Since traditional routing protocols are not suitable for the network environment of UAV formations, in recent years, many scholars have designed routing methods based on machine learning, such as the adaptive routing method and system for unmanned systems networks based on deep reinforcement learning proposed by the Institute of Scientific Computing Technology, China, and the distributed intelligent routing method for UAV network slicing proposed by the University of Electronic Science and Technology.

[0004] The current machine learning-based routing methods for UAV formations have two problems. On the one hand, some routing methods do not adopt the offline training method, and their online operation requires a large amount of power consumption and occupies a large amount of computing resources, such as the QoS routing method for flying ad-hoc networks based on Q-learning proposed by the Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, the routing optimization method and system based on graph neural networks and deep reinforcement learning proposed by Huazhong University of Science and Technology, and the network routing planning method and system based on the BP neural network ant colony algorithm proposed by Zhejiang Gongshang University. On the other hand, due to the inherent instability of machine learning algorithms, the routing methods based on machine learning have difficult-to-solve instability problems, and their reliability is relatively low. Directly performing routing selection based on machine learning algorithms may lead to network failures.

[0005] The routing performance is predicted using machine learning algorithms to provide a reference for route selection instead of directly performing route selection. The above strategy can effectively improve the stability of the network. Nanjing University of Posts and Telecommunications proposed a QoS routing algorithm for software-defined networks based on delay prediction and double ant colony, which provides a reference for route selection by predicting the delay, but this algorithm is only applicable to the software-defined network architecture. Guangdong Polytechnic Normal University proposed a routing method, device, computer device, and storage medium based on resource prediction, which uses neural networks and reinforcement learning to predict the changing trend of network resources and provides a reference for route selection, but this routing method does not consider the frequent movement of nodes. Chongqing University of Posts and Telecommunications proposed a multipath routing protocol method for predicting the mobility of UAV ad hoc networks. By predicting the node mobility, the reliability of the route is evaluated. This method focuses on considering the movement law of nodes and the link expiration time, and is mainly used to predict the probability that a certain route path may fail, without providing predictions of other performances of the route. Summary of the Invention

[0006] To solve the problems of the deficiencies and defects in the above prior art, the present invention provides a method and system for predicting the routing performance of UAV formations based on pigeon flock neural networks, which does not require the whole network topology as prior knowledge, greatly reducing the control overhead, and at the same time does not require online training by UAVs, greatly reducing the computational amount.

[0007] To achieve the above object of the present invention, the following technical solutions are adopted:

[0008] A method for predicting the routing performance of UAV formations based on pigeon flock neural networks, the method comprising the following steps:

[0009] S1: According to the obtained historical data, use the pigeon flock neural network to perform offline training on the routing performance prediction model, and solve to obtain an optimized routing performance prediction model;

[0010] S2: Periodically and real-time sense and obtain the current network situation and UAV formation situation of the UAV formation;

[0011] S3: Input the obtained current network situation and UAV formation situation into the optimized routing performance prediction model, predict the routing performance of neighbor nodes, and use the predicted routing performance prediction matrix as the basis for route selection.

[0012] Preferably, the historical data includes the data collected by the UAV formation during previous missions, expressed as a matrix The rows of the matrix represent a complete training pair, that is, the matrix consists of N data training pairs constitute, i = 1, 2,..., N data , that is:

[0013] Ddata = [T data (1) T data (2) … T data (N data )] T

[0014] Each training pair consists of the forward speed v of the drone pro , forward distance r pro , delay T with neighbor nodes de , maximum bandwidth B, neighbor node degree Λ, hop count h of neighbor nodes, and total delay T to , that is:

[0015] T data (i) = [v pro (i) r pro (i) T de (i) B(i) Λ(i) T to (i) h(i)]

[0016] where v pro (i) represents the forward speed of the i-th drone, r pro (i) represents the forward distance of the i-th drone, T de (i) represents the delay between the i-th drone and its neighbor nodes, B(i) represents the maximum bandwidth of the i-th drone, Λ(i) represents the neighbor node degree of the i-th drone, T to (i) represents the total delay of the i-th drone, and h(i) represents the hop count of the neighbor nodes of the i-th drone.

[0017] Furthermore, when offline training the routing performance prediction model using the pigeon flock neural network, there are three pigeon flocks, namely pigeon flock pigeon flock and pigeon flock Each pigeon flock contains N data pigeon individuals. The i-th pigeon individual in pigeon flock pigeon flock and pigeon flock is the mapping of the i-th drone in the unmanned formation in space space and space respectively. Among them, N N1 and N N2 represent the number of neurons in the first hidden layer and the second hidden layer of the pigeon flock neural network respectively.

[0018] Furthermore, the expression of the routing performance prediction model is as follows:

[0019]

[0020] Among them, W uh , W hh , W hy all represent the weighted matrix, and u represents the input of the routing performance prediction model.

[0021] Furthermore, the pigeon flock neural network is used to solve the routing performance prediction model offline, specifically as follows:

[0022] S101: Import historical data to obtain N data training pairs;

[0023] S102: Process historical data to obtain the input u(i) of the pigeon flock neural network and the theoretical value y T (i) of the output layer from each training pair:

[0024]

[0025] S103: Initialize, set the iteration number j = 1, and select the maximum iteration number j max ; Set the number of neurons N N1 in the first hidden layer of the pigeon flock neural network, the number of neurons N N2 in the second hidden layer, the iteration step sizes γ uh , γ hh and γ hy , and the initial values of the weighted matrices W uh (i,j), W hh (i,j) and W hy (i,j);

[0026] S104: For the input and output of all training pairs, calculate the output of the first hidden layer in the pigeon flock neural network, the output of the second hidden layer, and the output y(i,j) of the output layer:

[0027]

[0028] S105: Regard the weighted matrices W uh (i,j), W hh (i,j) and W hy (i,j) as the positions of the i-th pigeon individual in the 3 pigeon flocks at the j-th iteration respectively, and calculate the landmark operator of the i-th pigeon in each pigeon flock in this iteration;

[0029] S106: Calculate the positions of the leading pigeons uh corresponding to the pigeon flocks W hh (i,j), W hy (i,j) and W and

[0030]

[0031] S107: If the number of iterations is less than j max , the number of iterations j = j + 1, and go to S108; otherwise, go to S109;

[0032] S108: The pigeon flock W uh (i, j), W hh (i, j) and W hy (i, j) of all pigeons approach the leading pigeon, that is, according to and update W uh (i, j), W hh (i, j) and W hy (i, j), and go to S104;

[0033] S109: Training is completed, and is the optimal solution of the weighted matrix and obtain the solved routing performance prediction model, and its expression is as follows:

[0034]

[0035] Furthermore, calculate the landmark operator of the i-th pigeon in each pigeon flock for this iteration. The specific calculation formula is as follows:

[0036]

[0037] where fitness y (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W hy (i, j), fitness N2 (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W hh (i, j), fitness N1 (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W uh (i, j), e y (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W hy (i, j), e N2 (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W hh (i, j), e N1 (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W uh (i, j).

[0038] Furthermore, according to and Update W uh (i, j), W hh (i, j) and W hy (i, j) is as follows:

[0039]

[0040] In the formula, κ hy represents the update coefficient of the weighted matrix W hy (i, j), κ hh represents the update coefficient of the weighted matrix W hh (i, j), κ uh represents the update coefficient of the weighted matrix W uh represents the update coefficient of the weighted matrix W

[0041] Preferably, the formation situation of the UAVs includes the current speed of the UAVs themselves and the current positions of all UAVs in the formation; the current network situation includes the time delay between the UAVs and their neighbors and the maximum bandwidth between the UAVs and their neighbors.

[0042] Furthermore, the prediction of the routing performance of neighbor nodes is as follows:

[0043] S301: Determine whether the current node has neighbors. If there are no neighbors and the routing performance prediction matrix cannot be obtained, directly enter S307; otherwise, enter S302;

[0044] S302: Obtain the destination node from the packet to be forwarded. If the destination node is a neighbor, there is no need to provide the routing performance prediction matrix, and directly enter step 307; otherwise, enter S303;

[0045] S303: Use the forward rate v a of the current node x n relative to the neighbor node x pro (a, n), the forward distance r pro (a, n), the time delay T de (a, n) between the current node and the neighbor node, the maximum bandwidth B(a, n) between the current node and the neighbor node, and the neighbor node degree Λ(n) as the input of the routing performance prediction model:

[0046] u(a, n) = [v pro (a, n) r pro (a, n) T de (a, n) B(a, n) Λ(n)]

[0047] where n max is the number of neighbors of the current node, n = 1, 2,..., n max ;

[0048] The hop count prediction value of neighbor node x is calculated according to the routing performance prediction model n and the total delay prediction value

[0049]

[0050] S304: If all neighbor nodes have been traversed, proceed to the next step; otherwise, go to S303;

[0051] S305: Organize the routing performance prediction results of all neighbor nodes into a routing performance prediction matrix:

[0052]

[0053] Use the routing performance prediction matrix as the basis for routing selection and proceed to S306;

[0054] S306: If the task is completed, end the task; otherwise, go to S301.

[0055] A UAV formation routing performance prediction system based on a pigeon flock neural network, including an offline training module based on a pigeon flock neural network, a formation and network situation awareness module, and a routing performance prediction model;

[0056] Among them, the offline training module based on the pigeon flock neural network uses the pigeon flock neural network to perform offline training on the routing performance prediction model according to the acquired historical data, and obtains an optimized routing performance prediction model;

[0057] The formation and network situation awareness module periodically and real-time perceives and obtains the current network situation and UAV formation situation of the UAV formation;

[0058] The routing performance prediction module inputs the acquired current network situation and UAV formation situation into the optimized routing performance prediction model, predicts the routing performance of neighbor nodes, and uses the predicted routing performance prediction matrix as the basis for routing selection.

[0059] The present invention designs for the first time the routing performance prediction of neighbor nodes based on a pigeon flock neural network, providing a basis for routing selection, and having the following beneficial effects:

[0060] (1) It does not require the whole network topology as prior knowledge and is suitable for UAV formations with frequent relative movements

[0061] The present invention does not need to obtain the whole network topology of the UAV formation, and only predicts the routing performance according to neighbor node information and the position of the UAV formation. On the one hand, it is not affected by the frequent relative movement of the UAV formation.

[0062] ​(2) Offline training is adopted, which has relatively low requirements for the computing performance of the UAV platform.

[0063] Using historical data, the process of training and solving the routing performance prediction model based on the pigeon flock neural network is carried out offline, without occupying the scarce computing resources of the UAV platform.

[0064] (3) Based on the small computational amount of the pigeon flock neural network, the requirements for the computing performance of the offline training platform are relatively low.

[0065] Using the pigeon flock neural network to solve the routing performance prediction model, there is no need for derivative and differential processes, which simplifies the training process and greatly reduces the computational amount.

[0066] (4) Compatible with other routing protocols or routing methods.

[0067] The final output of the present invention is a routing performance prediction matrix, which provides a reference for route selection and does not directly perform route selection and forwarding. Therefore, it is compatible with other routing protocols or routing methods. Description of the Drawings

[0068] Figure 1 is a flowchart of the method for predicting the routing performance of UAV formations based on the pigeon flock neural network of the present invention.

[0069] Figure 2 is an example of the UAV formation of the present invention.

[0070] Figure 3 is a flowchart of the offline training of the routing performance prediction model using the pigeon flock neural network of the present invention.

[0071] Figure 4 is a schematic diagram of the solution process of the pigeon flock neural network of the present invention.

[0072] Figure 5 is the workflow of the routing performance prediction module of the present invention for predicting the routing performance of neighbor nodes. Detailed Embodiments

[0073] The present invention will be described in detail below with reference to the drawings and specific embodiments.

[0074] Embodiment 1

[0075] As Figure 1 shown, a method for predicting the routing performance of UAV formations based on the pigeon flock neural network, the method includes the following steps:

[0076] S1: According to the obtained historical data, use the pigeon flock neural network to perform offline training on the routing performance prediction model, and solve to obtain an optimized routing performance prediction model;

[0077] S2: Periodically and real - time sense and obtain the current network situation and the UAV formation situation of the UAV formation;

[0078] S3: Input the obtained current network situation and the UAV formation situation into the optimized routing performance prediction model, predict the routing performance of neighbor nodes, and use the predicted routing performance prediction matrix as the basis for routing selection.

[0079] This embodiment takes the UAV formation as shown in Figure 2 as an example. Let x a be the current node, x n be the current node, x t be the current node. Some elements of the input for its routing performance prediction

[0080] u(a,n)=[v pro (a,n) r pro (a,n) T de (a,n) B(a,n) Λ(n)] are calculated according to the following formula:

[0081] v pro (a,n)=||v pro (a,n)||=||v(a)cos[θ v (a,n)]||

[0082] r pro (a,n)=||r pro (a,n)||=||r(a,t)cos[θ r (a,n)]||

[0083] Among them, v pro (a,n) is the forward speed, v(a) is the speed of the current node, θ v (a,n) is the included angle between v(a) and r(a,n), r pro (a,n) is the forward position, r(a,n) is the position of x n relative to x a The position of r(a,t) is the position of x t relative to x a .

[0084] The delay T de (a,n) between the current node and neighbor nodes, the maximum bandwidth B(a,n) between the current node and neighbor nodes are reported by the channel devices of UAV nodes. The neighbor node degree Λ(n) is obtained according to neighbor node information. According to the example of Figure 2 , since neighbor node x n has 3 neighbor nodes in addition to the current node x a , therefore, Λ(n)=4.

[0085] On the one hand, the present invention does not require the whole network topology as prior knowledge, greatly reducing the control overhead. On the other hand, it does not require online training by the unmanned aerial vehicle (UAV), greatly reducing the computational complexity. Therefore, the present invention is suitable for the UAV formation network environment with high-dynamic topology and narrow bandwidth characteristics and the UAV platform with insufficient computing resources.

[0086] In a specific embodiment, the historical data includes the data collected by the UAV formation during previous missions, represented as a matrix The rows of the matrix represent a complete training pair, that is, the matrix is composed of N data training pairs where i = 1, 2, …, N data , that is:

[0087] D data = [T data (1) T data (2) … T data (N data )] T

[0088] Each training pair consists of the forward speed v pro of the UAV, the forward distance r pro , the time delay T de with the neighbor node, the maximum bandwidth B, the neighbor node degree Λ, the hop count h of the neighbor node, and the total time delay T to of the UAV, that is:

[0089] T data (i) = [v pro (i) r pro (i) T de (i) B(i) Λ(i) T to (i) h(i)]

[0090] where v pro (i) represents the forward speed of the i-th UAV, r pro (i) represents the forward distance of the i-th UAV, T de (i) represents the time delay between the i-th UAV and the neighbor node, B(i) represents the maximum bandwidth of the i-th UAV, Λ(i) represents the neighbor node degree of the i-th UAV, T to (i) represents the total time delay of the i-th UAV, and h(i) represents the hop count of the neighbor node of the i-th UAV.

[0091] In a specific embodiment, when using the pigeon flock neural network to perform offline training on the routing performance prediction model, there are three pigeon flocks, namely pigeon flock pigeon flock and pigeon flock Each pigeon flock contains N data pigeon individuals. The pigeon flocks The pigeon flocks and the pigeon flocks The i-th pigeon flock individuals of the pigeon flocks are respectively the mapping of the i-th unmanned aerial vehicle without formation in space Space and space where N N1 and N N2 respectively represent the number of neurons in the first hidden layer and the second hidden layer of the pigeon flock neural network.

[0092] In a specific embodiment, the expression of the routing performance prediction model is as follows:

[0093]

[0094] where W uh 、W hh 、W hy all represent weight matrices, and u represents the input of the routing performance prediction model.

[0095] In a specific embodiment, the pigeon flock neural network is used to solve the routing performance prediction model offline, as Figure 3 shown, specifically as follows:

[0096] S101: Import historical data to obtain N data training pairs;

[0097] S102: Process historical data to obtain the input u(i) of the pigeon flock neural network and the theoretical value y T (i) of the output layer from each training pair:

[0098]

[0099] S103: Initialize, set the iteration number j = 1, and select the maximum iteration number j max ; Set the number of neurons N N1 in the first hidden layer of the pigeon flock neural network, the number of neurons N N2 in the second hidden layer, the iteration step sizes γ uh 、γ hh and γ hy , and the initial values of the weight matrices W uh (i,j), W hh (i,j) and W hy (i,j);

[0100] S104: For the input and output of all training pairs, calculate the output of the first hidden layer in the pigeon flock neural network The output of the second hidden layer and the output y(i, j) of the output layer:

[0101] This embodiment calculates the output of the first hidden layer in the pigeon flock neural network the output of the second hidden layer and the output y(i, j) of the output layer, specifically as follows:

[0102]

[0103] S105: Consider the weighted matrices W uh (i, j), W hh (i, j), and W hy (i, j) as the positions of the i-th pigeon individual in the j-th iteration of 3 pigeon flocks respectively, and calculate the landmark operator of the i-th pigeon in each pigeon flock in this iteration;

[0104] This embodiment calculates the landmark operator of the i-th pigeon in each pigeon flock in this iteration, and the specific calculation formula is as follows:

[0105]

[0106] where fitness y (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W hy (i, j), fitness N2 (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W hh (i, j), fitness N1 (i, j) is the landmark operator of the i-th pigeon in the pigeon flock W uh (i, j), e y (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W hy (i, j), e N2 (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W hh (i, j), e N1 (i, j) is the landmark error metric function of the i-th pigeon in the pigeon flock W uh (i, j).

[0107] S106: Calculate the positions of the leading pigeons corresponding to the pigeon flocks W uh (i, j), W hh (i, j), and W hy (i, j) respectively in this iteration and

[0108] This embodiment calculates the positions of the leading pigeons corresponding to the pigeon flocks W uh (i, j), W hh (i, j), and W hyThe position of the leading pigeon corresponding to (i,j) and

[0109]

[0110] S107: If the number of iterations is less than j max , the number of iterations j = j + 1, and go to S108; otherwise, go to S109;

[0111] S108: The pigeon flock W uh (i,j), W hh (i,j) and W hy (i,j) all pigeons approach the leading pigeon, that is, according to and update W uh (i,j), W hh (i,j) and W hy (i,j), and go to S104;

[0112] In this embodiment, according to and update W uh (i,j), W hh (i,j) and W hy (i,j), specifically as follows:

[0113]

[0114] In the formula, κ hy represents the update coefficient of the weighted matrix W hy (i,j), κ hh represents the update coefficient of the weighted matrix W hh (i,j), κ uh represents the update coefficient of the weighted matrix W uh (i,j).

[0115] S109: Training is completed, and are the optimal solutions of the weighted matrix and obtain the optimized routing performance prediction model, and its expression is as follows:

[0116]

[0117] As Figure 4 shown, the routing performance prediction model of the UAV formation

[0118] When solving, adopt the pigeon flock neural network solving method to obtain the weighted matrix W uh , W hhand W hy Optimal solution and The solution process includes 3 pigeon flocks, namely the W N1 pigeon flock located in the N uh ×5 dimensional Euclidean space, the W N2 pigeon flock located in the N N1 ×N hh dimensional Euclidean space, and the W N2 pigeon flock located in the N2×N hy dimensional Euclidean space. Using the W uh (i,j), W hh (i,j) and W hy (i,j) corresponding to each training pair in the historical data as the members of the W uh pigeon flock, W hh pigeon flock and W hy pigeon flock respectively, and continuously optimize the positions of the leading pigeons of the 3 pigeon flocks through iteration based on the landmark operator and The optimal solution can be obtained without operations such as derivation or differentiation.

[0119] In a specific embodiment, the formation situation of the UAVs includes the current speed of the UAV itself and the current positions of all UAVs in the formation; the current network situation includes the delay between the UAV and its neighbors and the maximum bandwidth between the UAV and its neighbors.

[0120] In a specific embodiment, the prediction of the routing performance of neighbor nodes is as follows:

[0121] S301: Determine whether the current node has neighbors. If there are no neighbors, the routing performance prediction matrix cannot be obtained, and directly enter S307; otherwise, enter S302;

[0122] S302: Obtain the destination node from the packet to be forwarded. If the destination node is a neighbor, there is no need to provide the routing performance prediction matrix, and directly enter step 307; otherwise, enter S303;

[0123] S303: Take the forward rate v a of the current node x n relative to the neighbor node x pro (a,n), the forward distance r pro (a,n), the delay T de (a,n) between the current node and the neighbor node, the maximum bandwidth B(a,n) between the current node and the neighbor node, and the neighbor node degree Λ(n) as the input of the routing performance prediction model:

[0124] u(a,n) = [v pro (a,n) rpro (a,n) T de (a,n) B(a,n) Λ(n)]

[0125] where n max is the number of neighbors of the current node, n = 1, 2, …, n max ;

[0126] The hop count prediction value of neighbor node x n is calculated according to the routing performance prediction model and the total delay prediction value

[0127]

[0128] S304: If all neighbor nodes have been traversed, proceed to the next step; otherwise, go to S303;

[0129] S305: Organize the routing performance prediction results of all neighbor nodes into a routing performance prediction matrix:

[0130]

[0131] Use the routing performance prediction matrix as the basis for routing selection and proceed to S306;

[0132] S306: If the task is completed, end the task; otherwise, go to S301.

[0133] Embodiment 2

[0134] Based on the method for predicting the routing performance of an unmanned aerial vehicle formation based on a pigeon flock neural network described in Embodiment 1, this embodiment also provides a system for predicting the routing performance of an unmanned aerial vehicle formation based on a pigeon flock neural network, including an offline training module based on a pigeon flock neural network and a formation and network situation awareness module;

[0135] Among them, the offline training module based on a pigeon flock neural network performs offline training on the pigeon flock neural network according to the acquired historical data to obtain a routing performance prediction model, a routing performance prediction model;

[0136] Among them, the offline training module based on a pigeon flock neural network performs offline training on the routing performance prediction model using the pigeon flock neural network according to the acquired historical data, and obtains an optimized routing performance prediction model;

[0137] The formation and network situation awareness module periodically and real - time perceives and obtains the current network situation and the formation situation of the unmanned aerial vehicle formation;

[0138] The described routing performance prediction module inputs the obtained current network situation and UAV formation situation into the optimized routing performance prediction model to predict the routing performance of neighbor nodes, and uses the predicted routing performance prediction matrix as the basis for routing selection.

[0139] Among them, the offline training module based on pigeon flock neural network implements steps S101 - S109 in Embodiment 1.

[0140] The described routing performance prediction module implements steps S301 - S306 in the embodiment.

[0141] Embodiment 3

[0142] This embodiment also provides a computer system, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for predicting the routing performance of UAV formations based on pigeon flock neural network described in Embodiment 1.

[0143] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. A method for predicting the routing performance of an unmanned aerial vehicle formation based on a pigeon flock neural network, characterized in that: The method described above includes the following steps: S1: According to the obtained historical data, use the pigeon flock neural network to perform offline training on the routing performance prediction model, and solve to obtain the optimized routing performance prediction model; S2: Periodically and real-time sense and obtain the current network situation and the unmanned aerial vehicle (UAV) formation situation of the UAV formation; S3: Input the obtained current network situation and the UAV formation situation into the optimized routing performance prediction model, predict the routing performance of neighbor nodes, and use the predicted routing performance prediction matrix as the basis for routing selection; Among them, the historical data includes the data collected by the UAV formation in previous missions, which is represented as a matrix The rows of the matrix represent a complete training pair, that is, the matrix consists of N data training pairs constitute, i = 1, 2,..., N data , that is: D data = [T data (1)T data (2)…T data (N data )] T Each training pair consists of the forward rate v of the drone pro , the forward distance r pro , the time delay T with the neighbor node de , the maximum bandwidth B, the neighbor node degree Λ, the hop count h of the neighbor node, and the total time delay T to That is: T data (i) = [v pro (i)r pro (i)T de (i)B(i)Λ(i)T to (i)h(i)] where, v pro (i) represents the forward rate of the i-th drone, r pro (i) represents the forward distance of the i-th drone, T de (i) represents the time delay between the i-th drone and its neighbor nodes, B(i) represents the maximum bandwidth of the i-th drone, Λ(i) represents the neighbor node degree of the i-th drone, T to (i) represents the total time delay of the i-th drone, h(i) represents the hop count of the neighbor nodes of the i-th drone; and, When the routing performance prediction model is offline trained using the pigeon flock neural network, there are three pigeon flocks, namely pigeon flock Pigeon flock and pigeon flock Each pigeon flock contains N data pigeon individuals. The i-th pigeon flock individuals of pigeon flock Pigeon flock and pigeon flock are the mappings of the i-th unmanned aerial vehicle in the non-formation in space Space and space respectively. Among them, N N1 and N N2 represent the number of neurons in the first hidden layer and the second hidden layer of the pigeon flock neural network respectively; and, The expression of the routing performance prediction model is as follows: Among them, W uh , W hh , W hy all represent weighted matrices, and u represents the input of the routing performance prediction model.

2. The method for predicting the formation routing performance of drones based on a pigeon flock neural network according to claim 1, wherein: Use the pigeon flock neural network to perform offline solution on the routing performance prediction model, specifically as follows: S101: Import historical data to obtain N data training pairs; S102: Process historical data to obtain the input u(i) of the pigeon flock neural network and the theoretical value y(i) of the output layer from each training pair. T (i): S103: Initialize, set the iteration number j = 1, and select the maximum iteration number j max ; Set the number of neurons N in the first hidden layer of the pigeon flock neural network N1 , the number of neurons N in the second hidden layer N2 , the iteration step size γ uh 、γ hh and γ hy , the weight matrix W uh (i, j), W hh (i, j) and W hy (i, j) initial values; S104: Calculate the outputs of the first hidden layer, the second hidden layer and the output y(i, j) of the output layer of the pigeon flock neural network for the inputs and outputs of all training pairs respectively: ​ S105: Treat the weighted matrices W uh (i, j), W hh (i, j) and W hy (i, j) as the positions of the i-th pigeon individual in the j-th iteration of three pigeon flocks respectively, and calculate the landmark operators of the i-th pigeon in each pigeon flock for this iteration; S106: Calculate the pigeon groups W uh (i, j), W hh (i, j) and W hy (i, j) corresponding leading pigeon positions and S107: If the number of iterations is less than j max , the number of iterations j = j + 1, and proceed to S108; otherwise, proceed to S109; S108: Pigeon flock W uh (i, j), W hh (i, j) and W hy All pigeons in (i, j) move closer to the leading pigeon, that is, update W according to and Update W uh (i, j), W hh (i, j) and W hy (i, j), and enter S104; S109: Training completed, and is the optimal solution of the weighted matrix and obtain the optimized routing performance prediction model, and its expression is as follows:

3. The method for predicting the routing performance of an unmanned aerial vehicle formation based on a pigeon flock neural network according to claim 2, wherein: Calculate the landmark operator of the i-th pigeon in each pigeon flock in this iteration. The specific calculation formula is as follows: Among them, fitness y (i, j) is the landmark operator of the i-th pigeon in pigeon flock W hy (i, j) is the landmark operator of the i-th pigeon, fitness N2 (i, j) is the landmark operator of the i-th pigeon in pigeon flock W hh (i, j) is the landmark operator of the i-th pigeon, fitness N1 (i, j) is the landmark operator of the i-th pigeon in pigeon flock W uh (i, j) is the landmark operator of the i-th pigeon, e y (i, j) is the landmark operator of the i-th pigeon in pigeon flock W hy (i, j) is the landmark error metric function of the i-th pigeon, e N2 (i, j) is the landmark operator of the i-th pigeon in pigeon flock W hh (i, j) is the landmark error metric function of the i-th pigeon, e N1 (i, j) is the landmark operator of the i-th pigeon in pigeon flock W uh (i, j) is the landmark error metric function of the i-th pigeon.

4. The method for predicting the formation routing performance of drones based on a pigeon flock neural network according to claim 3, wherein: According to and Update W uh (i, j), W hh (i, j) and W hy (i, j) as follows: where κ hy represents the weighting matrix W hy the update coefficient of (i, j), κ hh represents the weighting matrix W hh the update coefficient of (i, j), κ uh represents the weighting matrix W uh the update coefficient of (i, j).

5. The method for predicting the formation routing performance of unmanned aerial vehicles based on a pigeon flock neural network according to claim 1, wherein: The UAV formation situation includes the current speed of the UAV itself and the current positions of all UAVs in the formation; the current network situation includes the delay between the UAV and its neighbors and the maximum bandwidth between the UAV and its neighbors.

6. The method for predicting the formation routing performance of drones based on a pigeon flock neural network according to claim 1, characterized in that: The prediction of the routing performance of neighbor nodes is specifically as follows: S301: Determine whether the current node has neighbors. If there are no neighbors and the routing performance prediction matrix cannot be obtained, directly enter S307; otherwise, enter S302; S302: Obtain the destination node from the packets to be forwarded. If the destination node is a neighbor, there is no need to provide the routing performance prediction matrix, and directly enter step 307; otherwise, enter S303; S303: Use the forward rate v n of this node xa relative to the neighbor node x pro (a,n), the forward distance r pro (a,n), the delay T de (a,n) between this node and the neighbor node, the maximum bandwidth B(a,n) between this node and the neighbor node, and the neighbor node degree Λ(n) as the input of the routing performance prediction model: u(a,n) = [v pro (a,n) r pro (a,n) T de (a,n) B(a,n) Λ(n)] where n max is the number of neighbors of the current node, and n = 1, 2, …, n max ; The hop count prediction value of neighbor node x calculated according to the routing performance prediction model n and the total delay prediction value are obtained S304: If all neighbor nodes have been traversed, enter the next step; otherwise, enter S303; S305: Organize the routing performance prediction results of all neighbor nodes into a routing performance prediction matrix: Use the routing performance prediction matrix as the basis for routing selection, and enter S306; S306: If the task is completed, end the task; otherwise, enter S301.

7. A system for predicting the routing performance of an unmanned aerial vehicle formation based on the pigeon flock neural network according to any one of claims 1 to 6, characterized in that: It includes an offline training module based on the pigeon flock neural network, a formation and network situation perception module, and a routing performance prediction model; Among them, the offline training module based on the pigeon flock neural network performs offline training on the routing performance prediction model using the pigeon flock neural network according to the obtained historical data, and solves to obtain the optimized routing performance prediction model; The formation and network situation perception module periodically and real-time senses and obtains the current network situation and the UAV formation situation of the UAV formation; The routing performance prediction module inputs the obtained current network situation and the UAV formation situation into the optimized routing performance prediction model, predicts the routing performance of neighbor nodes, and uses the predicted routing performance prediction matrix as the basis for routing selection.

Images