Human-Machine / Unmanned-Aerial-Vehicle Co-Fusion Region Search Control Method Based on Biological Positive and Negative Feedback

Abstract

The present invention discloses a method for searching for a co - integration area of manned aircraft and unmanned aircraft based on biological positive and negative feedback, including: Step 1, establishing a spatial model of a heterogeneous cluster of manned aircraft and unmanned aircraft; Step 2, controlling the cluster of manned aircraft and unmanned aircraft based on model predictive control; Step 3, establishing a biological - inspired neural network environment model based on stimulus response; Step 4, establishing a decision - making model for the unmanned aircraft and setting corresponding efficiency functions; Step 5, establishing a rolling pigeon flock optimization model strategy based on the model predictive control model; Step 6, establishing a multi - unmanned - aircraft information fusion strategy based on a timestamp synchronization mechanism; Step 7, outputting a search result map of the co - integration area of manned aircraft and unmanned aircraft with biological positive and negative feedback model predictive control. This method aims to, under the condition of unknown target and environmental information, the manned aircraft guides the unmanned aircraft through the cluster space state instruction to conduct area search in an unknown environment through a biological excitation - inhibition model, improving the search efficiency and environmental self - adaptability.

Description

Technical Field

[0001] The present invention discloses a co - operative area search control method for manned - aircraft / unmanned - aircraft based on biological positive and negative feedback, belonging to the technical field of autonomous search control for human - machine hybrid clusters. Background Technique

[0002] In recent years, unmanned aircraft have developed rapidly. As a new type of combat force, they have continuously penetrated into all walks of life from traditional information support and information guarantee fields such as reconnaissance and surveillance, hydrology and meteorology, and can undertake more important tasks in aspects such as tracking, positioning, remote control, telemetry, and digital transmission.

[0003] Generally, the problem of unmanned - aircraft cooperative area search is the cooperation between weak agents, while the cooperation of manned - aircraft / unmanned - aircraft in area search is the cooperation between strong agents and weak agents. The traditional center - less distributed control mode requires unmanned aircraft to have strong independent computing, analysis, decision - making, and autonomous control capabilities. This strategy is difficult to implement under the current situation where the autonomous capabilities of unmanned aircraft are not high. Considering that the manned aircraft is located at the core of the cluster space state to command and control its loyal wingman unmanned aircraft, an operator can control a heterogeneous cluster of multiple unmanned aircraft. Although unmanned - aircraft technology is constantly improving, unmanned aircraft still have "inherent drawbacks". When operating alone, they cannot meet the autonomous - capability conditions for high - dynamic confrontation and face great risks. Therefore, having a simple and efficient human - in - the - loop intervention control through a manned - aircraft / unmanned - aircraft combat platform with command control is the basis for realizing cooperative combat.

[0004] Cooperative area search of unmanned aerial vehicles (UAVs) has become an important research topic because it can timely detect targets to be rescued in unknown areas and provide the target locations for subsequent rescue personnel. For the area coverage search algorithm of UAVs, heuristic algorithms are mostly used for solving. Common ones include heuristic-based coverage algorithms, pheromone-based multi-UAV coverage algorithms, and multi-UAV coverage algorithms based on swarm intelligence algorithms such as ant colony and particle swarm. However, although these algorithms can improve the coverage rate to a certain extent, the computational complexity of the algorithms is relatively large, and affected by the large number of the group, the distributed characteristics of mutual division of labor and cooperation among the groups are not fully realized. Therefore, it is necessary to determine the interaction rules and interaction methods between UAVs so that the UAV swarm can quickly and effectively create a map with complete information. For this purpose, a new bio-inspired neural network search model is introduced to represent the environmental state information in the coverage process. This bio-positive and negative feedback model of excitation and inhibition can enable UAVs to quickly interact with the environment and the UAV swarm in unknown areas during area coverage, and relatively completely utilize and create map information. The stimulus-response model is a biologically inspired computational model used to describe the collective behavior of individuals in complex systems. Taking the ant colony as an example, this model abstracts the process of information acquisition, transmission, and utilization of ants into a simple model with four elements: information source, pheromone, feedback mechanism, and critical threshold. This model reveals the generation mechanism of group behavior, that is, through information transmission and regulation between individuals, the overall coordination and adaptability are achieved.

[0005] For the control process of heterogeneous UAV clusters in discrete space, in practical engineering, classical control methods often have difficulty obtaining satisfactory results because it is difficult to handle nonlinear, multi-constrained, uncertain, and time-varying control systems. The Model Predictive Control (MPC) method is based on model prediction, rolling optimization, and feedforward-feedback control structure. At each discrete time, according to the obtained current measurement information, the optimal solution within a certain time domain is solved, and the first element of the obtained control sequence is used to act on the controlled object; at the next sampling moment, the above process is repeated, and the new measurement value is used as the initial condition for predicting the future dynamics of the system at this time, and the problem is refreshed for solution.

[0006] In summary, in order to enable manned aircraft to guide UAVs through the bio-excitation inhibition model to conduct area search in an unknown environment under the conditions of unknown target and environmental information, improving the search efficiency and environmental adaptability has become an urgent problem to be solved in the existing technology. Summary of the Invention

[0007] In view of the problems existing in the prior art, the present invention provides a method for searching for a co - existence area of manned aircraft and unmanned aircraft based on biological positive and negative feedback. The purpose is to enable the manned aircraft to guide the unmanned aircraft through the biological excitation - inhibition model to search for areas in an unknown environment under the condition of unknown target and environmental information, so as to improve the search efficiency and environmental adaptability.

[0008] A method for controlling the co - existence area search of manned aircraft and unmanned aircraft based on biological positive and negative feedback of the present invention includes:

[0009] Step 1: Establish a heterogeneous cluster space model of manned aircraft and unmanned aircraft

[0010] Based on the three translational degrees of freedom and two rotational degrees of freedom included in the airspeed, heading, and flight path three - channel autopilot configured for each unmanned aircraft, an equivalent second - order system dynamics model of the unmanned aircraft is established. Using the individual state variables of the unmanned aircraft in the earth coordinate system and the spatial configuration between the unmanned aircraft and the manned aircraft, a cluster space state model in the cluster coordinate system is established.

[0011] Step 2: Control of the manned aircraft / unmanned aircraft cluster based on model predictive control

[0012] Based on the position error and velocity error of the cluster space state, a state - space equation is established and model predictive control is used for control.

[0013] Step 3: Establish a bio - inspired neural network environment model based on stimulus response

[0014] An environment model is established through the field - of - view angle range of the unmanned aircraft, the number of surrounding associated neurons, the threshold effect model generated by environmental factors, and the inhibition and excitation of grid - map environmental information.

[0015] Step 4: Build a decision - making model for the unmanned aircraft and set the corresponding effectiveness function

[0016] The decision - making model of the unmanned aircraft is mainly established with the control input of the moving angle of the unmanned aircraft, and the coverage search effectiveness function is set by the coverage neuron activity value gain, turning angle, and threat collision.

[0017] Step 5: Establish a rolling pigeon - flock optimization model strategy based on the model predictive control model

[0018] According to the neural activity value of the grid map, the value of the search cost function is calculated. During the prediction within the prediction interval by predicting L steps, during multiple predictions of L steps, the prediction with the maximum search benefit needs to be selected to establish a rolling pigeon - flock optimization model.

[0019] Step 6: Establish a multi - unmanned - aircraft information fusion strategy based on the timestamp synchronization mechanism

[0020] Use the bio-grid neuron activity value map and timestamp map of each UAV to synchronize the information fusion among multiple UAVs.

[0021] Step 7: Output the search control result map of the manned / UAV co-fusion area with positive and negative biological feedback

[0022] By judging whether the iteration time reaches the maximum iteration time, if it reaches, jump out of the loop and output the manned / UAV model predictive control result and the biological positive and negative feedback search neuron activity value map; otherwise, continue the iterative optimization process to search the entire area.

[0023] The above-mentioned scheme further constructs the manned / UAV heterogeneous cluster space model in Step 1 as follows:

[0024]

[0025] In the formula: [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 respectively represent the position coordinates of the state variables of 3 UAV individuals in the geodetic coordinate system, [χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 represent the translation and rotation postures of 3 UAV individuals in the geodetic coordinate system, [x c , y c , z c , χ c , γ c represent the parameter values of the cluster space {C} relative to the geodetic coordinate system, that is, the position of the manned aircraft in space and its translation and rotation postures. Here, the subscript c represents the meaning of the cluster center. (χ c1 , χ c2 , χ c3 ) and (γ c1 , γ c2 , γ c3 ) represent the rotation posture variables of the UAV individuals in the cluster coordinate system. The values of {p, q, α, β} determine the configuration of the UAV cluster. p and q are variables for controlling the forward following distance of the cluster configuration. α represents the pitch command angle of the cluster, and β represents the variable for controlling the lateral following distance of the cluster configuration. g 1 , g 2 ..., g15 Represents the functional relationship of the internal variables of the individual state variable r of the UAV and the cluster space state variable c. (h, l, d) represents the intermediate vector in the calculation process, and (p c , p 1 , p 2 , P c , P 1 , P 2 ) represents the coordinate point in the calculation process.

[0026] The manned / UAV cluster control based on model predictive control in the second step includes

[0027] U k = [u(k|k), u(k + 1|k),..., u(k + i - 1|k), u(k + N - 1|k)] T (12)

[0028] where k represents the sampling time, and u(k) is the input vector.

[0029] The establishment of the bio-inspired neural network environment model based on stimulus response in the third step includes:

[0030] The basic stimulus response model

[0031]

[0032] where i and j represent the number of UAVs and neurons respectively, x j is the stimulation degree of the biological neuron j, is the response threshold of the UAV i, x k is the stimulation degree of other neurons associated with the neuron j, m is the number of associated other neurons, T(x j ) is the probability that the UAV i performs the task x j at this time.

[0033] And the bio-inspired neural network model

[0034]

[0035] where h is the number of associated neurons connected to the i-th neuron. And represent the excitatory and inhibitory inputs respectively. is the activity value of the neuron j adjacent to the neuron i. are all positive constants, is the decay rate of and are respectively The upper and lower limits, i.e.,

[0036] The fourth step of building a decision-making model for the UAV and setting the corresponding effectiveness function includes:

[0037] UAV r i The next state can be expressed as

[0038]

[0039] Where, And respectively represent the position, speed, and the angle between the current moving direction and the positive direction of the horizontal axis. The subscript ri represents the serial number of the UAV, which is a positive integer.

[0040] The combined collaborative area search cost function is

[0041] f(k) = α 1 f E (k) + α 2 f S (k) + α 3 f C (k) (29)

[0042] Where, α 1 , α 2 , α 3 are the weight coefficients corresponding to different cost functions, f E neuron grid map excitation cost function, f S turning angle limit cost function, and f C collision risk avoidance cost function.

[0043] The fifth step of establishing a rolling pigeon flock optimization model strategy based on the model predictive control model includes:

[0044] Within the prediction interval, by predicting L steps and accumulating the cost function f for each step, the sum of the cost function values for all UAVs in L steps is expressed as

[0045]

[0046] After calculating the cost function value by predicting L steps each time, determine the control input at this time

[0047] {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (31)

[0048] Apply the first input value to the drone swarm.

[0049] During the prediction process of multiple L steps, it is necessary to select the prediction with the greatest search benefit. Considering the characteristics of the pigeon flock optimization PIO in terms of simple parameter design, fast iteration speed, and strong global optimization ability, PIO is selected for optimization and solution.

[0050] Step 6: Establish a multi-drone information fusion strategy based on the timestamp synchronization mechanism

[0051] When information is exchanged between drones, since the activity value judgment of the current biological grid neuron activity map is related not only to the current drone but also to the activity values detected by other drones within the range, each drone needs to maintain not only a biological grid neuron activity value map but also a timestamp map to synchronize the information exchange process during the search process. Let the timestamp be t i,c , representing the update of the search map for the nearest update of grid c by drone i.

[0052] Step 7: Output the search control result map of the manned / unmanned aircraft co-fusion area with positive and negative biological feedback

[0053] Judge whether the iteration time reaches the maximum iteration time k max , if it reaches, then jump out of the loop and output the prediction control result of the manned / unmanned aircraft model and the biological positive and negative feedback search neuron activity value map; otherwise, continue the iterative optimization process to search the entire area.

[0054] In the above solution, further

[0055] In step 1, it is assumed that each drone is equipped with an autopilot with three channels of airspeed, heading, and flight path, including three translational degrees of freedom and two rotational degrees of freedom. The equivalent second-order system dynamics model of the drone is established as follows:

[0056]

[0057] At the same time, consider the practicality constraints of the drone:

[0058]

[0059] In the formula: all i mentioned in step 1 are positive integers, referring to the i-th drone, are the velocities of the three axes of drone i, V i , χ i and γ i are the horizontal velocity, heading angle, and altitude change rate respectively, V i c , and are the control inputs on the three loops of the autopilot, where the superscript c represents the meaning of control, and τ v , τ χ and τ λ are three time constants respectively, V max , V min , n max and γ max are all greater than 0, representing the maximum speed, minimum horizontal speed, maximum lateral overload and maximum altitude change rate respectively. γ min is the minimum altitude change rate and is less than 0. g is the acceleration due to gravity taken as 9.8 m / s 2 .

[0060] The individual state variables of the UAV are expressed in the earth coordinate system {G} as follows:

[0061] r = [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 , χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 T (3)

[0062] In the formula: [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 represent the position coordinates of the individual state variables of the three UAVs in the earth coordinate system respectively. [χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 represent the translation and rotation postures of the three UAVs in the earth coordinate system respectively.

[0063] The cluster space state variables in the cluster coordinate system {C} are established as follows

[0064] c = [x c , y c , z c ​, χ c , γ c , χ c1 , χ c2 , χ c3 , γ c1 , γ c2 , γ c3 , p, q, α, β] T (4)

[0065] In the formula: [x c , y c , z c , χ c , γ c represents the parameter values of the cluster space {C} relative to the geodetic coordinate system, that is, the position of the manned aircraft in space and its translational and rotational postures. Here, the subscript c represents the meaning of the cluster center. (χ c1 , χ c2 , χ c3 ) and (γ c1 , γ c2 , γ c3 ) represent the rotational attitude variables of the UAV individual in the cluster coordinate system. The values of {p, q, α, β} determine the configuration of the UAV cluster. p and q represent the variables controlling the forward following distance of the cluster configuration. α represents the pitch command angle of the cluster, and β represents the variable controlling the lateral following distance of the cluster configuration.

[0066] Based on the selection of the UAV individual state variable r and the cluster space state variable c above, establish their kinematic conversion relationship. g 1 , g 2 ..., g 15 represents the functional relationship of the internal variables of the UAV individual state variable r and the cluster space state variable c. KIN(·) represents the function of the conversion of these two variables.

[0067]

[0068] Among them, (h, l, d) represents the intermediate vector in the calculation process. (p c , p 1 , p 2 , P c , P 1 , P 2 ) represents the coordinate points in the calculation process.

[0069]

[0070] Considering the relationship between the UAV individual speed variable and the cluster space speed variable , J rDenote the derivative of the individual state variable and the derivative of the swarm space state variable The Jacobian matrix for conversion between them.

[0071]

[0072] Where

[0073]

[0074] Taking the second-order differential of the UAV position in the model of Equation (1) gives the relevant second-order system control input as follows

[0075]

[0076] In the second step, the position error and velocity error of the swarm space state are expressed as e i = c i - c d and The discrete state-space equation with respect to the error is obtained as follows:

[0077]

[0078] That is

[0079]

[0080] where Δt is the sampling time, k represents the sampling instant, u(k) is the input vector, x(k) ∈ R n is the system variable, and A and B are the coefficient matrices of the system. The model predictive control model MPC is established through this state-space equation of the error. For the system to generate e(0) in one iteration process, the control input sequence can make the position error and velocity error converge to 0 within a finite time.

[0081] Assume that starting from the k-th moment, the input of the system is

[0082] U k = [u(k|k), u(k + 1|k),..., u(k + i - 1|k), u(k + N - 1|k)] T (12)

[0083] where N represents the prediction horizon.

[0084] The prediction of the system state quantity by MPC is expressed as

[0085] X k = [x(k|k), x(k + 1|k), x(k + 2|k),..., x(k + N|k)] T (13)

[0086] In this state equation, since the error is used as the state variable, the reference value r is 0, and the output y of the system is y = x. Considering that the MPC problem is solved by the quadratic programming method, the general form is

[0087]

[0088] Therefore, the cost function at this time considering the weighted sum of errors, the weighting of system inputs, and the terminal error at the end of the prediction is expressed as

[0089]

[0090] According to the discrete state-space expressions in equations (12) and (13), the expressions of each state in x(k) are listed as follows

[0091] After simplification, we can get

[0092]

[0093] That is

[0094] X k = Mx k + CU k (18)

[0095] According to equations (15) and (18), matrix simplification gives

[0096]

[0097] Among them, let the error weight parameter matrix The input weight parameter matrix The prediction end terminal parameter matrix

[0098] In the third step, each UAV individual is mapped to an ant individual, the stimulus inhibition intensity of the area is mapped to the state of the area at this time, and each UAV has a fixed response threshold, through the distance between UAVs, the number of surrounding individuals, and the inhibition and excitation of the grid map environment information. The basic stimulus-response model is as follows

[0099]

[0100] where i and j represent the number of UAVs and neurons respectively, x j is the stimulation degree of the biological neuron j, is the response threshold of the UAV i, x k is the stimulation degree of other neurons associated with the neuron j, m is the number of associated other neurons, T(x j) is the probability that UAV i executes mission x at this time j The probability.

[0101] In the decision-making stage, each UAV is assumed to be an agent with an omnidirectional 360° viewing angle ability but with a maximum turning limit angle θ max and a minimum turning limit angle θ min and can communicate within a certain range L max First, the regional search environment is rasterized into a two-dimensional grid, and then a bio-inspired neural network model on a two-dimensional plane is established on the grid map. Each grid intersection represents a biological neuron and has a corresponding neuron activity value. Within the range L max it will have excitatory and inhibitory effects on other biological neurons, where N c represents the current biological neuron, N a represents the associated biological neuron associated with the current biological neuron, and N e represents the biological neuron that has no connection with the current biological neuron. At this time, the bio-inspired neural network model is

[0102]

[0103] where h is the number of associated neurons connected to the i-th neuron. and represent excitatory and inhibitory inputs respectively. is the activity value of neuron j adjacent to neuron i. are all positive constants, is the decay rate of and are respectively the upper and lower limits of That is, the connection weight between neuron i and neuron j in the same layer is called the lateral connection weight, denoted as

[0104]

[0105] where, |p i -p j | refers to the distance between two neurons, and a and R n are both positive constants.

[0106] Each grid point in the grid map has three states, and these three states will directly affect the neuron activity value of this grid point, which are: not searched, searched, and obstacle. These three states are represented by state flag bits

[0107]

[0108] Among them, S(G i ) represents the state of the i-th grid vertex in the grid map. When the grid vertex has not been searched, S(G i ) = 1. When the grid vertex has been searched, S(G i ) = 0. When there is an obstacle at the grid vertex, S(G i ) = -1.

[0109] In the fourth step described above, during the top-level decision-making and planning process, the UAV is simplified to immediately change its moving direction and move forward at a constant speed without considering acceleration or other complex factors. The next state of UAV r i is expressed as

[0110]

[0111] where and represent the position, speed, and the angle between the current moving direction and the positive x-axis respectively. The subscript ri represents the serial number of the UAV, which is a positive integer. In order to search the unknown area more effectively, each UAV will accumulate multiples based on the given minimum angle θ min , and the formula is expressed as

[0112] θ ri (t + 1) = θ ri (t) ± εθ min (25)

[0113] where ε is a positive integer in the interval.

[0114] The designed coverage search efficiency function is mainly aimed at guiding the UAV to search the uncovered area. During the search process, there are certain turning angle limit conditions and possible collision risks. Therefore, the neuron grid map excitation cost function f E , the turning angle limit cost function f S , and the collision risk avoidance cost function f C will be considered.

[0115] The UAV moves towards the area with a higher neuron activity value. Therefore, the neuron activity value incremental gain is the sum of the neuron activity values within the range covered by the UAV for each step, which is expressed as

[0116]

[0117] where is the l-th grid point within the current coverage range of the UAV. Ω is the number of grid points that the UAV can cover at the current position.

[0118] When the UAV is in motion, a too large turning angle will cause excessive consumption of the UAV's energy. Therefore, the turning cost function is designed as

[0119]

[0120] where θ max is the maximum turning limit angle of each UAV.

[0121] During the movement of the UAV, it is necessary to avoid obstacles according to the judgment of the threat distance L max with other UAVs. When there are H robots within the communication range L max , the collision cost function is established as

[0122]

[0123] where L cr represents the distance between the current UAV and other UAVs within the coverage range.

[0124] According to the design of different weights of the cost, three factors are mainly considered for the impact on the cooperative area search of the UAV, so as to improve the adaptability of the entire cooperative area search algorithm. Therefore, the combined cooperative area search cost function is

[0125] f(k) = α 1 f E (k) + α 2 f S (k) + α 3 f C (k) (29)

[0126] where α 1 , α 2 , α 3 are the weight coefficients corresponding to different cost functions.

[0127] In the fifth step, according to equations (20) and (28), the neural activity value of the grid map changes with the movement of the UAV, and thus the value of the search cost function is calculated. Due to the characteristics of rolling prediction, within the prediction interval, by predicting L steps and accumulating the cost function f of each step, the sum of the cost function values of all UAVs for L steps is expressed as

[0128]

[0129] After calculating the cost function value by predicting L steps each time, the control input at this time is determined

[0130] {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (31)

[0131] Apply the first input value among the UAV swarm.

[0132] During the prediction process of multiple L steps, it is necessary to select the prediction with the maximum search benefit. Considering the characteristics of the pigeon flock optimization PIO in simple parameter design, fast iteration speed, and strong global optimization ability, PIO is selected for optimization and solution.

[0133] Step 6: Establish a multi-UAV information fusion strategy based on the timestamp synchronization mechanism

[0134] When information is exchanged between UAVs, since the activity value judgment of the current biological grid neuron activity map is not only related to the current UAV but also related to the activity values detected by other UAVs within the range, each UAV needs to maintain not only a biological grid neuron activity value map but also a timestamp map to synchronize the information exchange process during the search process. Let the timestamp be t i,c , representing the most recent update of the search map of UAV i for grid c.

[0135] Step 7: Output the search control result map of the manned / UAV co-fusion area with positive and negative biological feedback

[0136] Judge whether the iteration time reaches the maximum iteration time k max , if it reaches, break out of the loop and output the prediction control results of the manned / UAV model and the biological positive and negative feedback search neuron activity value map; otherwise, continue the iterative optimization process to search the entire area.

[0137] In addition, in the above scheme, the PIO algorithm in step 5 has the following four steps:

[0138] 1) Initialization

[0139] Initialize the population size P of the pigeon flock, the dimension D of the solution space, the map and compass factor R, and the number of iterations N c1 and N c2 .

[0140] 2) Map and compass operator

[0141] According to the initialized pigeon flock algorithm parameters, substitute them into the map and compass operator. The iteration formula of this operator is as follows:

[0142]

[0143] In the formula, the position of the i-th pigeon is represented as X i , and the velocity is represented as V i , R is the map and compass parameter, t represents the number of iterations, rand is a random number between [0, 1], and X g represents the globally optimal position. When t reaches the number of iterations N c1 , the map and compass operator is exited. Equation (30) is the fitness function f(X i t ) during the iteration process.

[0144] 3) Landmark operator

[0145] According to the initialized pigeon flock algorithm parameters, the parameters of the landmark operator X that exits the map and compass operator i t and the fitness function value f(X i t ) are substituted into the landmark operator. The iteration formula of this operator is as follows:

[0146]

[0147] In the formula, N p represents the number of individuals in the population, t represents the number of iterations, X c represents the position of the central pigeon after each iteration, X i represents the position of the i-th pigeon, and f(X i ) represents the fitness function at this time of X i , that is, Equation (29). rand is a random number on [0, 1]. When t reaches the number of iterations N c2 , the landmark operator is exited.

[0148] 4) Loop. All particles iterate within the boundary range through initialization, the map and compass operator, and the landmark operator until the maximum number of generations. After predicting the fitness function value for L steps each time, the optimal control input U m , that is, Equation (33), is obtained, and its first component u m (k) is applied to the UAV, and finally the search efficiency coverage is maximized.

[0149] {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (33).

[0150] Three rules are set in Step 6 to update the timestamp map.

[0151] 1) When the UAV searches the current grid, the update of the timestamp at this time comes from its own behavior, and the timestamp t of the searched grid c is updated at this time. i,c It is the number of update moments at the current time.

[0152] 2) When the updated bio-grid neuron activity value map of the current grid searched by the UAV comes from the information fusion of multiple UAVs within the communication range, the timestamp of the current searched grid c is updated to the timestamp closest to the current moment.

[0153] 3) When there are other UAVs within the range of the current UAV, information interaction is required. At this time, not only the bio-grid neuron activity value map is interacted, but also the map is updated to the current latest timestamp map.

[0154] In the entire UAV system, the activity value map results represented by the latest timestamp within the confidence communication range (including the local machine) in other areas are selected to avoid the UAVs from repeatedly accessing some grid areas that have been determined.

[0155] The present invention is a method for searching the co-fusion area of manned aircraft / UAVs based on the bio-positive and negative feedback model prediction control. Its advantages and effects are as follows: First, it provides a stable control for the manned aircraft to control the UAV through MPC relying on simple instructions, reduces the load of the manned aircraft operator, and provides a stable and loyal wingman framework structure; Second, it proposes a method for establishing a grid map based on the bio-positive and negative feedback mechanism, improves the environmental perception ability of the UAV, reduces the probability of collision avoidance between UAVs, and has a certain environmental adaptability; Third, it has the characteristics of distributed optimization solution, reduces the situation of repeated search within the search range, and the optimization algorithm can also converge quickly, improving the coverage efficiency. Brief Description of the Drawings

[0156] Figure 1 Schematic diagram of two-dimensional bio-inspired neural network grid

[0157] Figure 2 Algorithm flowchart based on positive and negative feedback model prediction control

[0158] Figure 3 Simulation result diagram (3D view) of manned aircraft / UAV collaborative scanning line area search

[0159] Figure 4 Simulation result diagram (top view) of manned aircraft / UAV collaborative scanning line area search

[0160] Figure 5 Simulation result diagram 1 (3D view) of UAV collaborative area search based on bio-positive and negative feedback

[0161] Figure 6Simulation 1 Results Diagram of UAV Cooperative Area Search Based on Biological Positive and Negative Feedback (Top View)

[0162] Figure 7 Simulation 2 Results Diagram of UAV Cooperative Area Search Based on Biological Positive and Negative Feedback (Stereogram)

[0163] Figure 8 Simulation 2 Results Diagram of UAV Cooperative Area Search Based on Biological Positive and Negative Feedback (Top View)

[0164] Explanation of Symbols in the Figure is as Follows:

[0165] L max —— Maximum communication range radius of UAV

[0166] θ min —— Minimum turning limit angle of UAV

[0167] θ max —— Maximum turning limit angle of UAV

[0168] N c —— Represents the current bio-inspired neuron

[0169] N a —— Represents the neighbor neuron associated with the current bio-inspired neuron

[0170] N e —— Represents the neuron not directly related to the current bio-inspired neuron Specific Embodiment

[0171] The following combines the accompanying drawings and specific embodiments to verify the effectiveness of the method proposed by the present invention.

[0172] Embodiment 1

[0173] Combined with Figure 1 and 2 , a human-machine / UAV co-integration area search control method based on biological positive and negative feedback includes:

[0174] Step 1: Build a heterogeneous cluster space model of human-machine / UAV

[0175] Assume that each UAV is equipped with an autopilot with three channels of airspeed, heading and track, including three translational degrees of freedom and two rotational degrees of freedom. The equivalent second-order system dynamics model of the UAV is established as follows:

[0176]

[0177] At the same time, consider the practicality constraints of UAV:

[0178]

[0179] In the formula: all i mentioned in Step 1 are positive integers, referring to the i-th unmanned aerial vehicle (UAV). are the velocities of the UAV i in three axial directions, V i , χ i and γ i are the horizontal velocity, heading angle and altitude change rate respectively, V i c , and are the control inputs on three loops of the autopilot respectively, where the superscript c represents the meaning of control, τ v , τ χ and τ λ are three time constants respectively, V max , V min , n max and γ max are all greater than 0, representing the maximum velocity, minimum horizontal velocity, maximum lateral overload and maximum altitude change rate respectively, γ min is the minimum altitude change rate and is less than 0, and g is the gravitational acceleration taken as 9.8 m / s 2 .

[0180] The individual state variables of the UAV in the geodetic coordinate system {G} are expressed as:

[0181] r = [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 , χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 T (3)

[0182] In the formula: [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 represent the position coordinates of the individual state variables of three UAVs in the geodetic coordinate system respectively, [χ 1 , χ 2 , χ 3 , γ 1 , γ​2 , γ 3 represent the translation and rotation postures of three UAV individuals in the geodetic coordinate system respectively.

[0183] The cluster space state variables in the cluster coordinate system {C} are established as follows

[0184] c = [x c , y c , z c , χ c , γ c , χ c1 , χ c2 , χ c3 , γ c1 , γ c2 , γ c3 , p, q, α, β] T (4)

[0185] In the formula: [x c , y c , z c , χ c , γ c represents the parameter values of the cluster space {C} relative to the geodetic coordinate system, that is, the position of the manned aircraft in space and its translation and rotation postures. The subscript c here represents the meaning of the cluster center. (χ c1 , χ c2 , χ c3 ) and (γ c1 , γ c2 , γ c3 ) represent the rotation attitude variables of the UAV individuals in the cluster coordinate system. The values of {p, q, α, β} determine the configuration of the UAV cluster. The variables p and q represent the variables that can control the forward following distance of the cluster configuration. α represents the pitch command angle of the cluster, and β represents the variable that can control the lateral following distance of the cluster configuration.

[0186] Based on the selection of the above UAV individual state variable r and cluster space state variable c, the kinematic conversion relationship can be established. g 1 , g 2 ..., g 15 represents the functional relationship between the internal variables of the UAV individual state variable r and the cluster space state variable c. KIN(·) represents the function of the conversion of these two variables.

[0187]

[0188] Among them, (h, l, d) represents the intermediate vector in the calculation process. (p c , p 1 , p 2 , P c,P 1 ,P 2 ) represents the coordinate points in the calculation process.

[0189]

[0190] Consider the individual speed variable of the UAV and the swarm space speed variable The relationship between them, J r represents the derivative of the individual state variable and the Jacobian matrix for the conversion between the derivative of the swarm space state variable and the derivative of the swarm space state variable.

[0191]

[0192] Among them,

[0193]

[0194] Taking the second-order differential of the UAV position in the model of Equation (1), the relevant second-order system control input can be obtained as follows

[0195]

[0196] Step 2: Manned / UAV Swarm Control Based on Model Predictive Control

[0197] Express the position error and speed error of the swarm space state as e i = c i - c d and Obtain the discrete state-space equation about the error:

[0198]

[0199] That is

[0200]

[0201] Among them, Δt is the sampling time, k represents the sampling moment, u(k) is the input vector, and x(k) ∈ R n is the system variable, and A and B are the coefficient matrices of the system. The model predictive control (Model Predictive Control, abbreviated as MPC) model can be established through the state-space equation of this error. The purpose is to make e(0) generated in one iteration process of the system, and the control input sequence can make the position error and speed error converge to 0 within a limited time.

[0202] Assume that starting from the k moment, the input of the system is

[0203] U k=[u(k|k), u(k + 1|k),..., u(k + i - 1|k), u(k + N - 1|k)] T (12)

[0204] where N represents the prediction horizon.

[0205] The prediction of the system state variables by MPC is expressed as

[0206] X k =[x(k|k), x(k + 1|k), x(k + 2|k),..., x(k + N|k)] T (13)

[0207] In this state equation, since the error is used as the state variable, the reference value r is 0, and the system output y = x. Considering that the MPC problem is solved by the method of quadratic programming, the general form is

[0208]

[0209] Therefore, the cost function at this time considering the weighted sum of errors, the weighting of system inputs, and the terminal error at the end of the prediction is expressed as

[0210]

[0211] According to the discrete state - space expressions in Eqs. (12) and (13), the expressions of each state in x(k) can be listed as follows

[0212] After simplification, it can be obtained that

[0213]

[0214] That is

[0215] X k = Mx k + CU k (18)

[0216] According to the matrix simplification of Eqs. (15) and (18), it can be obtained that

[0217]

[0218] Among them, let the error weight parameter matrix Input weight parameter matrix Terminal parameter matrix at the end of the prediction

[0219] Step 3: Establish a bio - inspired neural network environment model based on stimulus - response

[0220] The stimulus-response model is a computational model of group behavior, which originates from the self-organization phenomenon of collective behavior in organisms such as insects. The core idea of this model is that based on the stimuli generated by environmental factors, individuals take corresponding actions and leave information, thus influencing and promoting the behaviors of other individuals. Since each grid vertex in the grid map uses a probabilistic representation form, which is more in line with the actual situation, based on the stimulus-response principle and model, each UAV individual is mapped to an ant individual, the stimulus inhibition intensity of the area is mapped to the state of the area at this time, and each UAV itself has a fixed response threshold, through the distance between UAVs, the number of surrounding individuals, and the inhibition and excitation of the grid map environmental information. The basic stimulus-response model is as follows

[0221]

[0222] where i and j represent the numbers of UAVs and neurons respectively, and x j is the degree of stimulation of biological neuron j, is the response threshold of UAV i, and x k is the degree of stimulation of other neurons associated with neuron j, m is the number of associated other neurons, and T(x j ) is the probability that UAV i performs task x j at this time. When a certain threshold is reached, specific actions can be executed. Similar to ants, when the pheromone concentration reaches a certain level, ants will start to act.

[0223] In the decision-making stage, each UAV is assumed to be an agent with a 360° omnidirectional viewing angle ability but with a maximum turning limit angle θ max and a minimum turning limit angle θ min , and can communicate within a certain range L max . First, the regional search environment is rasterized into two dimensions, and then a bio-inspired neural network model in a two-dimensional plane is established on the grid map. Each grid intersection represents a biological neuron and has a corresponding neuron activity value, which will cause stimulatory and inhibitory effects on other biological neurons within the range L max . As shown in Figure 1 , where N c represents the current biological neuron, N a represents the associated biological neuron associated with the current biological neuron, and N e represents the biological neuron that has no connection with the current biological neuron.

[0224] At this time, the bio-inspired neural network model is

[0225]

[0226] Among them, h is the number of associated neurons that the i-th neuron is connected to; and respectively represent excitatory and inhibitory inputs; is the activity value of neuron j adjacent to neuron i; are all positive constants, is the decay rate of and are respectively the upper and lower limits of The connection weight between neuron i and neuron j in the same layer is called the lateral connection weight, denoted as

[0227]

[0228] where, |p i - p j | refers to the distance between two neurons, and a and R n are both positive constants.

[0229] Each grid point in the grid map has three states, and these three states will directly affect the neuron activity values of this grid point, which are: not searched, searched, and obstacle. These three states are represented by status flags

[0230]

[0231] where, S(G i ) represents the state of the i-th grid vertex in the grid map; when the grid vertex has not been searched, S(G i ) = 1; when the grid vertex has been searched, S(G i ) = 0; when there is an obstacle at the grid vertex, S(G i ) = -1.

[0232] Step 4: Establish a decision-making model for the UAV and set the corresponding effectiveness function

[0233] During the top-level decision-making and planning process, the UAV is simplified to be able to immediately change its moving direction and move forward at a constant speed, without considering acceleration or other complex factors. The next state of the UAV r i can be expressed as

[0234]

[0235] where, and respectively represent the position, speed, and the angle between the current moving direction and the positive x-axis. The subscript ri represents the serial number of the UAV, which is a positive integer. In order to search the unknown area more effectively, each UAV will accumulate multiples based on the given minimum angle θ min The formula is expressed as

[0236]

[0237] where ε is a positive integer in the

[0238] The designed coverage search efficiency function is mainly used to guide the UAV to search the uncovered area. During the search process, there are certain turning angle limit conditions and possible collision risks. Therefore, the neuron grid map excitation cost function f E , the turning angle limit cost function f S and the collision risk avoidance cost function f C will be considered

[0239] The UAV moves towards the area with a higher neuron activity value. Therefore, the neuron activity value increment benefit is the sum of the neuron activity values within the range covered by the UAV in each step, which is expressed as

[0240]

[0241] where is the l-th grid point within the current coverage range of the UAV; Ω is the number of grid points that the UAV can cover at the current position

[0242] When the UAV is moving, a too large turning angle will cause excessive consumption of the UAV's energy. Therefore, the turning cost function is designed as

[0243]

[0244] where θ max is the maximum turning limit angle of each UAV

[0245] During the movement of the UAV, it is necessary to avoid obstacles according to the judgment of the threat distance L max from other UAVs. If the distance is too close, it will not only cause the danger of collision, but also easily cause the overlapping of the coverage ranges of the UAVs, which will greatly reduce the coverage efficiency. When there are H robots within the communication range L max , the collision cost function is established as

[0246]

[0247] where Lcr Indicates the distance between the current UAV and other UAVs within the coverage range.

[0248] According to the design of different weights for the cost, three factors affecting the collaborative area search of UAVs can be considered with emphasis, so as to improve the adaptability of the entire collaborative area search algorithm. Therefore, the combined collaborative area search cost function is

[0249] f(k) = α 1 f E (k) + α 2 f S (k) + α 3 f C (k) (29)

[0250] Among them, α 1 , α 2 , α 3 Are the weight coefficients corresponding to different cost functions.

[0251] Step 5: Establish a rolling pigeon flock optimization model strategy based on the model predictive control model

[0252] According to equations (20) and (28), the neural activity value of the grid map changes with the movement of the UAV, so as to calculate the value of the search cost function. Due to the characteristics of rolling prediction, within the prediction interval, by predicting L steps and accumulating the cost function f of each step, the sum of the cost function values of all UAVs for L steps is obtained and expressed as

[0253]

[0254] In the process of multiple L-step predictions, it is necessary to select the prediction with the maximum search benefit. Considering the characteristics of the pigeon-inspired optimization (PIO) in simple parameter design, fast iteration speed, and strong global optimization ability. Select PIO for optimization and solution. The PIO algorithm has the following four steps:

[0255] 1) Initialization

[0256] Initialize the population size P of the pigeon flock, the dimension D of the solution space, the map and compass factor R, and the number of iterations N c1 and N c2 .

[0257] 2) Map and compass operator

[0258] According to the initialized pigeon flock algorithm parameters, substitute them into the map and compass operator. The iteration formula of this operator is as follows:

[0259]

[0260] In the formula, the position of the i-th pigeon is represented as X i , and the velocity is represented as V i , R is the map and compass parameter, t represents the number of iterations, rand is a random number between [0, 1], and X g represents the globally optimal position. When t reaches the number of iterations N c1 , the map and compass operator is exited. Equation (30) is the fitness function f(X i t ) during the iterative process.

[0261] 3) Landmark operator

[0262] According to the initialized pigeon flock algorithm parameters, the parameters of the landmark operator X i t and the fitness function value f(X i t ) that exit the map and compass operator are substituted into the landmark operator. The iterative formula of this operator is as follows:

[0263]

[0264] In the formula, N p represents the number of individuals in the population, t represents the number of iterations, X c represents the position of the central pigeon after each iteration, X i represents the position of the i-th pigeon, and f(X i ) represents the fitness function at this time of X i , that is, Equation (29). rand is a random number on [0, 1]. When t reaches the number of iterations N c2 , the landmark operator is exited.

[0265] 4) Loop. All particles are iterated within the boundary range through initialization, the map and compass operator, and the landmark operator until the maximum number of generations. After predicting the fitness function value for L steps each time, the optimal control input U m , that is, Equation (33), is obtained, and its first component u m (k) is applied to the UAV, finally achieving the maximization of the search efficiency coverage.

[0266] {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (33)

[0267] Step Six: Establish a multi-UAV information fusion strategy based on the timestamp synchronization mechanism

[0268] When information is exchanged between UAVs, since the determination of the activity value of the current biological grid neuron activity map is related not only to the current UAV but also to the activity values detected by other UAVs within the range, each UAV needs to maintain not only a biological grid neuron activity value map but also a timestamp map to synchronize the information exchange process during the search. Let the timestamp be t i,c , which represents the update of the search map for the grid c by UAV i at the most recent time. And three rules are set here to update the timestamp map.

[0269] 1) When a UAV searches the current grid, the update of the timestamp at this time comes from its own behavior, and the timestamp t of the currently searched grid c is updated i,c is the number of the current update moment.

[0270] 2) When the updated biological grid neuron activity value map of the current grid search by the UAV comes from the information fusion of multiple UAVs within the communication range, the timestamp of the currently searched grid c is updated to the timestamp closest to the current moment.

[0271] 3) When there are other UAVs within the range of the current UAV, information needs to be exchanged. At this time, not only the biological grid neuron activity value map is exchanged, but also the timestamp map is updated to the current latest one.

[0272] 4) In the entire UAV system, the activity value map result represented by the latest timestamp within the confidence communication range (including the local machine) in other regions is selected to avoid UAVs from repeatedly accessing some determined grid areas.

[0273] Step Seven: Output the search control result map of the manned / unmanned aircraft co - integration area with biological positive and negative feedback (generate the best collaborative area search simulation trajectory map in the corresponding figure)

[0274] Judge whether the iteration time has reached the maximum iteration time k max , if it has reached, then jump out of the loop and output the manned / unmanned aircraft model predictive control result and the biological positive and negative feedback search neuron activity value map; otherwise, continue the iterative optimization process to search the entire area.

[0275] Specific Application Example 2

[0276] The experimental computer is configured with an Intel Core i7 - 10700 processor, a main frequency of 2.90Ghz, and 16G of memory. The software is the MATLAB R2022b version.

[0277] The specific steps of a manned / unmanned aircraft co - integration area search control method based on biological positive and negative feedback are as follows:

[0278] Step 1: Build a manned / unmanned aerial vehicle heterogeneous cluster space model

[0279] Initialize each parameter of the unmanned aerial vehicle, where the airspeed time constant τ v = 0.01, the heading angle time constant τ χ = 0.01, the track control time constant τ γ = 0.01, the minimum speed limit V min = 5 m / s, the maximum speed limit V max = 20 m / s, the maximum lateral overload n max = 1, the minimum lateral overload n min = -1, the maximum track inclination angle The minimum track inclination angle The error proportionality coefficient k p = 0.425, the error derivative proportionality coefficient k d = 0.50, set the initial positions of three unmanned aerial vehicles as x = [85, 90, 95], y = [85, 90, 95], z = [95, 100, 90], the initial position of the manned aerial vehicle as [x, y, z] = [100, 100, 100], and set the unmanned aerial vehicle cluster space configuration parameters as [30, 30, 0, π / 3]. Substitute these parameters into the manned / unmanned aerial vehicle heterogeneous cluster space model, that is, Equation (6), to obtain the input control interface Equation (9), the cluster space state c Equation (4), and the derivative of the cluster space state Equation (7).

[0280] Step 2: Manned / unmanned aerial vehicle cluster control based on model predictive control

[0281] Considering the manned / unmanned aerial vehicle model in the cluster space in Step 1, according to the cluster space state c and its derivative c, where the time step is Δt = 0.1, the discrete state space equation regarding the error is

[0282]

[0283] Initialize the error weighted sum parameter matrix The input weighted sum parameter R = 1, the terminal error parameter matrix The prediction horizon N = 3, and initially set a trajectory. Input the first value of the obtained input u(k) to the second-order control input of the unmanned aerial vehicle, that is, Equation (9). Examine the control effect of the MPC control on the manned / unmanned aerial vehicle cluster space model to realize the guidance process of the existing track and complete the verification of the initial scan line search control effect.

[0284] Step 3: Establish a biological-inspired neural network excitation-inhibition environment model

[0285] Initialization parameter: Decay rate of biological neuron activity value Upper and lower limits

[0286] Initial value of neuron activity Determine the distance parameter of neighboring neurons Correlation neighbor weight parameter a = 0.1, maximum turning limit angle θ max = π, minimum turning limit angle θ min = π / 12, maximum communication range Position of the obstacle Response threshold of UAV i According to the obstacle area, unsearched area, and searched area in the grid map, which is expressed by Equation (23), and then substituted into Equation (21) to obtain the neuron activity value of each grid area, and then substituted into the stimulus-response model Equation (20) to obtain the probability that the UAV will execute a certain grid area at this time, Equation (20).

[0287] Step 4: Establish the decision-making model of the UAV and set the corresponding efficiency function

[0288] According to the grid area obtained in Step 3, then set the initial position of the UAV and Two different initial positions correspond to different initial positions of two instances, initial directions [π / 4, 7π / 4, 5π / 4, 3π / 4], neuron activity value increment benefit ratio coefficient α 1 = 1, UAV turning cost function ratio coefficient α 2 = 0.01, UAV anti-collision cost function ratio coefficient α 3 = 0.1, calculate all possible movement directions θ of the UAV obtained in Equations (24) and (25) ri , and then according to the set cost function ratio coefficient, calculate the search cost function value of the UAV moving in the possible direction at this time, Equation (29).

[0289] Step 5: Based on the model predictive control model, establish a rolling pigeon flock optimization model strategy

[0290] At this time, set the prediction interval L = 3, population size P = 30, dimension of the solution space D = 8, map and compass factor R = 0.65, number of iterations N c1 = 25, number of iterations N c2 = 30, by calculating the cost function values of P individuals in the population under the prediction interval, find the optimal value through pigeon flock optimization, and allow mutation and migration within a certain range, and then find the optimal U mThat is, Equation (33). Input the first vector in Equation (33) into Equation (24) to complete one iteration of the regional search.

[0291] Step Six: Establish a multi-UAV information fusion strategy based on the timestamp synchronization mechanism

[0292] According to Steps Three and Four, when the UAVs perform information interaction, each UAV synchronously records the next timestamp t for the searched area i,c , that is, record the iteration time each time in an array. When there are other UAVs within the communication range, update it to the neuron activity value of the latest timestamp among all UAVs including itself.

[0293] Step Seven: Output the control result graph of the manned / UAV co-fusion regional search with biological positive and negative feedback

[0294] First, in testing the control effect of MPC on the scanning line search of the manned / UAV heterogeneous model, the total iteration time is 160, and the sampling time step is 0.1. When the iteration time reaches 160, output the search simulation results of the manned and unmanned UAVs under MPC control to lay the foundation for the guidance control. Then, test the role of the biological positive and negative feedback mechanism on the UAVs. The total iteration time is 50, and the sampling time step is 0.1. When the iteration time reaches 50, input the simulation graphs of the UAV collaborative regional search under the biological positive and negative feedback mechanism under the conditions of different initial positions, grid sizes of 50×50 and 80×80 (this condition only has different initial positions given in Step Four and different grid sizes, and the rest of the simulation parameters are the same).

[0295] The present invention adopts three test simulations. First, use 1 manned aircraft and 3 UAVs to analyze the control effect of the manned / UAV collaborative scanning line search under MPC control. After laying the foundation for the guidance, then construct an environmental map of the UAV collaborative regional search through the biological positive and negative feedback mechanism, and conduct trajectory point analysis on the two search processes under different initial conditions of 4 UAVs (different initial conditions refer to different initial input positions of the UAVs and different grid sizes, and the rest of the parameters are the same in the two test simulations) in order to complete the collaborative regional search task.

Claims

1. A method for controlling the co - fusion area search of manned / unmanned aerial vehicles based on biological positive and negative feedback, characterized in that, it includes: Step 1: Build a spatial model of a heterogeneous cluster of manned / unmanned aerial vehicles Based on the three translational degrees of freedom and two rotational degrees of freedom included in the airspeed, heading, and flight path three - channel autopilot configured for each unmanned aerial vehicle, establish an equivalent second - order system dynamics model of the unmanned aerial vehicle. Using the individual state variables of the unmanned aerial vehicle in the earth coordinate system and the spatial configuration between the unmanned aerial vehicle and the manned aerial vehicle, establish a cluster - space state model in the cluster coordinate system; Step 2: Cluster control of manned / unmanned aerial vehicles based on model predictive control Establish a state - space equation based on the position error and velocity error of the cluster - space state, and use model predictive control for control; Step 3: Establish a bio - inspired neural network environment model based on stimulus - response Establish an environment model through the field - of - view angle range of the unmanned aerial vehicle, the number of surrounding associated neurons, the threshold effect model generated by environmental factors, and the inhibition and excitation of grid - map environmental information; Step 4: Build a decision - making model for the unmanned aerial vehicle and set the corresponding effectiveness function The decision - making model of the unmanned aerial vehicle is mainly established with the control input of the moving angle of the unmanned aerial vehicle, and the coverage search effectiveness function is set by considering the benefits of the active values of the covering neurons, the turning angle, and the threat of collision; Step 5: Establish a rolling pigeon - flock optimization model strategy based on the model predictive control model Calculate the value of the search cost function according to the neural activity value of the grid map. During the prediction within the prediction interval for L steps, during multiple L - step prediction processes, it is necessary to select the prediction with the maximum search benefit to establish a rolling pigeon - flock optimization model; Step 6: Establish a multi - unmanned - aerial - vehicle information fusion strategy based on the timestamp synchronization mechanism Use the biological grid neuron activity value map and timestamp map of each unmanned aerial vehicle to synchronize the information fusion between multiple unmanned aerial vehicles; Step 7: Output the control result map of the co - fusion area search of manned / unmanned aerial vehicles with biological positive and negative feedback By judging whether the iteration time reaches the maximum iteration time, if it reaches, then jump out of the loop and output the model predictive control result of the manned / unmanned aerial vehicle model and the biological positive and negative feedback search neuron activity value map. Otherwise, continue the iterative optimization process to search the entire area.

2. The method for controlling the co - fusion area search of manned / unmanned aerial vehicles based on biological positive and negative feedback according to claim 1, characterized in that: The building of the spatial model of the heterogeneous cluster of manned / unmanned aerial vehicles in Step 1 is as follows: where: [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 respectively represent the position coordinates of the state variables of 3 UAV individuals in the geodetic coordinate system, [χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 represents the translation and rotation postures of 3 UAV individuals in the geodetic coordinate system, [x c , y c , z c , χ c , γ c represents the parameter values of the cluster space {C} relative to the geodetic coordinate system, that is, the position of the manned aircraft in space and its translation and rotation postures. Here, the subscript c represents the meaning of the cluster center. (χ c1 , χ c2 , χ c3 ) and (γ c1 , γ c2 , γ c3 ) represent the rotation attitude variables of the UAV individuals in the cluster coordinate system. The values of {p, q, α, β} determine the configuration of the UAV cluster. p and q represent the variables that control the forward following distance of the cluster configuration. α represents the pitch command angle of the cluster. β represents the variable that controls the lateral following distance of the cluster configuration. g 1 , g 2 ..., g 15 represent the functional relationship between the state variable r of the UAV individual and the internal variables of the cluster space state variable c. (h, l, d) represents the intermediate vector in the calculation process. (p c , p 1 , p 2 , P c , P 1 , P 2 ) represent the coordinate points in the calculation process; The cluster control of manned / unmanned aerial vehicles based on model predictive control in Step 2 includes U k = [u(k|k), u(k + 1|k),..., u(k + i - 1|k), u(k + N - 1|k)] T (12) where k represents the sampling time, and u(k) is the input vector; The establishment of the bio - inspired neural network environment model based on stimulus - response in Step 3 includes: The basic stimulus - response model where \(i\) and \(j\) represent the number of drones and neurons respectively, \(x\) j j is the degree of stimulation of biological neuron \(j\), is the response threshold of drone \(i\), \(x\) k k is the degree of stimulation of other neurons associated with neuron \(j\), \(m\) is the number of associated other neurons, \(T(x\) j j ) is the probability that drone \(i\) performs task \(x\) j j at this time; and the bio - inspired neural network model where h is the number of associated neurons that the i-th neuron is connected to; and represent excitatory and inhibitory inputs respectively; is the activity value of neuron j adjacent to neuron i; are all positive constants, is the decay rate of, and are respectively the upper and lower limits of, that is The building of the decision - making model for the unmanned aerial vehicle and the setting of the corresponding effectiveness function in Step 4 includes: Drone r i The next state can be represented as Among them, and respectively represent the position, velocity, and the angle between the current moving direction and the positive x-axis. The subscript ri represents the serial number of the UAV, which is a positive integer; The combined collaborative area search cost function is f(k) = α 1 f E (k) + α 2 f S (k) + α 3 f C (k) (29) Among them, α 1 , α 2 , α 3 are weight coefficients corresponding to different cost functions, f E neuron grid map excitation cost function, f S turning angle limit cost function, and f C collision risk avoidance cost function; The establishment of the rolling pigeon - flock optimization model strategy based on the model predictive control model in Step 5 includes: Within the prediction interval, by predicting for L steps and accumulating the cost function f for each step, the sum of the cost function values for all drones over L steps is obtained and denoted as After calculating the cost function value by predicting L steps each time, determine the control input at this time {u m (k),u m (k + 1),u m (k + 2),...,u m (k + L - 1)} (31) Apply the first input value among them to the drone swarm During multiple L-step prediction processes, it is necessary to select the prediction with the greatest search benefit. Considering the characteristics of the pigeon-inspired optimization (PIO) in terms of simple parameter design, fast iteration speed, and strong global optimization ability, select PIO for optimization and solution Step 6: Establish a multi-drone information fusion strategy based on a timestamp synchronization mechanism When information is exchanged between drones, since the determination of the activity value of the current biological grid neuron activity map is related not only to the current drone but also to the activity values detected by other drones within the range, each drone needs to maintain not only a biological grid neuron activity value map but also a timestamp map to synchronize the information exchange process during the search process; let the timestamp be t i,c , indicating the update of the search map for the nearest update of grid c by drone i; Step 7: Output the search control result graph of the manned / unmanned aircraft co-fusion area with biological positive and negative feedback Determine whether the iteration time has reached the maximum iteration time k max , if it has reached, then jump out of the loop, output the model predictive control results of the manned / unmanned aerial vehicle and the map of the activity values of the biological positive and negative feedback search neurons, otherwise, continue the iterative optimization process to search the entire area.

3. The manned / unmanned aircraft co-fusion area search control method based on biological positive and negative feedback according to claim 1, characterized in that In the said Step 1, assume that each drone is equipped with an autopilot with three channels of airspeed, heading, and trajectory, including three translational degrees of freedom and two rotational degrees of freedom. The equivalent second-order system dynamics model of the drone is established as follows: At the same time, consider the practicality constraints of the drone where: all the \(i\) mentioned in step one are positive integers, referring to the \(i\)-th drone, are the velocities of the three axes of drone \(i\), \(V\) i , \(\chi\) i and \(\gamma\) i are the horizontal velocity, heading angle and altitude change rate respectively, \(V\) i c , and are the control inputs on the three loops of the autopilot respectively, where the superscript \(c\) represents the meaning of control, \(\tau\) v , \(\tau\) χ and \(\tau\) λ are the three time constants respectively, \(V\) max , \(V\) min , \(n\) max and \(\gamma\) max are all greater than 0, representing the maximum speed, minimum horizontal speed, maximum lateral overload and maximum altitude change rate respectively, \(\gamma\) min is the minimum altitude change rate, and is less than 0, \(g\) is the gravitational acceleration taken as \(9.8m / s\) 2 ; Express the individual state variables of the drone in the earth coordinate system {G} as: r = [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 , χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 T (3)​ where: [x 1 , x 2 , x 3 , y 1 , y 2 , y 3 , z 1 , z 2 , z 3 respectively represent the position coordinates of the state variables of three UAV individuals in the geodetic coordinate system, and [χ 1 , χ 2 , χ 3 , γ 1 , γ 2 , γ 3 respectively represent the translation and rotation postures of three UAV individuals in the geodetic coordinate system; Establish the cluster space state variables in the cluster coordinate system {C} as follows c = [x c , y c , z c , χ c , γ c , χ c1 , χ c2 , χ c3 , γ c1 , γ c2 , γ c3 , p, q, α, β] T (4) Where: [x c , y c , z c , χ c , γ c represents the parameter values of the cluster space {C} relative to the geodetic coordinate system, that is, the position of the manned aircraft in space and its translational and rotational postures. Here, the subscript c represents the meaning of the cluster center. (χ c1 , χ c2 , χ c3 ) and (γ c1 , γ c2 , γ c3 ) represent the rotational posture variables of the UAV individuals in the cluster coordinate system. The values of {p, q, α, β} determine the configuration of the UAV cluster. p and q represent the variables controlling the forward following distance of the cluster configuration. α represents the pitch command angle of the cluster. β represents the variable controlling the lateral following distance of the cluster configuration; Based on the selection of the above-mentioned individual drone state variable r and swarm space state variable c, establish their kinematic conversion relationship, g 1 ,g 2 ...,g 15 represents the functional relationship of the internal variables of the individual drone state variable r and the swarm space state variable c, and KIN(·) represents the function for the conversion of these two variables; Among them, (h, l, d) represents the intermediate vector of the calculation process, and (p c , p 1 , p 2 , P c , P 1 , P 2 ) represents the coordinate points in the calculation process; Consider the individual speed variable of the UAV and the swarm space speed variable The relationship between them, J r represents the Jacobian matrix for the conversion between the derivative of the individual state variable and the derivative of the swarm space state variable ; where Taking the second-order differential of the drone position in the model of Equation (1), the relevant second-order system control input can be obtained as follows In the second step, the position error and velocity error of the cluster space state are expressed as e i = c i - c d and The discrete state space equation about the error is obtained: That is where Δt is the sampling time, k represents the sampling instant, u(k) is the input vector, and x(k) ∈ R n is the system variable, and A and B are the coefficient matrices of the system; a model predictive control model MPC is established through the state-space equation of this error; e(0) is generated during one iteration of the system, and the control input sequence can make the position error and speed error converge to 0 within a finite time; Assume that starting from time k, the input of the system is U k = [u(k|k), u(k + 1|k),..., u(k + i - 1|k), u(k + N - 1|k)] T (12) where N represents the prediction interval The prediction of the system state quantity by MPC is expressed as X k = [x(k|k), x(k + 1|k), x(k + 2|k),..., x(k + N|k)] T (13) In this state equation, since the error is used as the state variable, the reference value r is 0, and the system output y = x. Considering that the MPC problem is solved by the quadratic programming method, the general form is Therefore, the cost function at this time considering the weighted sum of errors, the weighting of system inputs, and the terminal error at the end of the prediction is expressed as According to the discrete state space expressions of Equations (12) and (13), list the expressions of each state in x(k) as follows After simplification, we get That is X k = Mx k + CU k (18) According to Equations (15) and (18), perform matrix simplification to obtain where the error weight parameter matrix is the input weight parameter matrix the predicted end terminal parameter matrix In the said Step 3, each drone individual is mapped to an ant individual, the stimulus inhibition intensity of the area is mapped to the state of the area at this time, and each drone itself has a fixed response threshold, through the distance between drones, the number of surrounding individuals, and the inhibition and excitation of the grid map environment information. The basic stimulus response model is as follows where \(i\) and \(j\) represent the number of drones and neurons respectively, \(x\) j is the degree of stimulation of biological neuron \(j\), is the response threshold of drone \(i\), \(x\) k is the degree of stimulation of other neurons associated with neuron \(j\), \(m\) is the number of associated other neurons, \(T(x\) j ) is the probability that drone \(i\) performs task \(x\) j at this time; In the decision-making stage, each UAV is assumed to be an agent with a full 360° perspective ability but with a maximum turning limit angle θ max and a minimum turning limit angle θ min and can communicate within a certain range L max First, the two-dimensional grid of the regional search environment is rasterized, and then a bio-inspired neural network model on a two-dimensional plane is established on the grid map. Each grid intersection represents a biological neuron and has a corresponding neuron activity value. Within the range L max it will have a stimulating and inhibitory effect on other biological neurons, where N c represents the current biological neuron, N a represents the associated biological neuron associated with the current biological neuron, and N e represents the biological neuron that has no connection with the current biological neuron; at this time, the bio-inspired neural network model is where h is the number of associated neurons associated with the i-th neuron; and represent excitatory and inhibitory inputs respectively; is the activity value of neuron j adjacent to neuron i; are all positive constants, is the decay rate of, and are respectively the upper and lower limits of, that is The connection weight between neuron i and neuron j in the same layer is called the lateral connection weight, denoted as where |p i -p j | refers to the distance between two neurons, and a and R n are both positive constants; Each grid point in the grid map will have three states, and these three states will directly affect the neuron activity value of this grid point, which are: not searched, already searched, and obstacle. These three states are represented by state flags Among them, S(G i ) represents the state of the i-th grid vertex in the grid map; when the grid vertex has not been searched, S(G i ) = 1; when the grid vertex has been searched, S(G i ) = 0; when there is an obstacle at the grid vertex, S(G i ) = -1; In the fourth step, during the top-level decision-making and planning process, the drone is simplified to immediately change its moving direction and move forward at a constant speed without considering acceleration or other complex factors; the next state of the drone r i is represented as Among them, and respectively represent the position, speed, and the angle between the current moving direction and the positive direction of the horizontal axis. The subscript ri represents the serial number of the UAV, which is a positive integer. In order to more effectively search the unknown area, each UAV will accumulate multiples based on the given minimum angle θ min The formula is expressed as where ε is a positive integer in the interval; The designed coverage search efficiency function is mainly aimed at guiding the UAV to search for uncovered areas. During the search process, there are certain turning angle limit conditions and possible collision risks. Therefore, the excitation cost function f of the neuron grid map will be considered E , the turning angle limit cost function f S and the collision risk avoidance cost function f C ; The drone moves towards the area with a higher neuron activity value. Therefore, the neuron activity value incremental benefit is expressed as the sum of the neuron activity values within the range covered by the drone for each step Among them, is the l-th grid point within the current coverage range of the UAV; Ω is the number of grid points that the UAV can cover at the current position; When the drone is moving, a large turning angle will cause excessive consumption of the drone's energy. Therefore, a turning cost function is designed here and expressed as where θ max is the maximum turning limit angle of each drone; During the movement of the drone, it is necessary to avoid obstacles according to the judged threat distance L from other drones max When avoiding obstacles, when there are H robots within the communication range L max the collision cost function is established as Among them, L cr represents the distance between the current UAV and other UAVs within the coverage range; According to the design of different weights for costs, three factors affecting the cooperative area search of UAVs are considered with emphasis, so as to improve the adaptability of the entire cooperative area search algorithm. Therefore, the combined cooperative area search cost function is f(k) = α 1 f E (k) + α 2 f S (k) + α 3 f C (k) (29) where α 1 , α 2 , α 3 are weight coefficients corresponding to different cost functions; In the fifth step, according to equations (20) and (28), the neural activity value of the grid map changes with the movement of the UAVs, and thus the value of the search cost function is calculated. Due to the characteristics of rolling prediction, within the prediction interval, by predicting L steps and accumulating the cost function f for each step, the sum of the cost function values for all UAVs in L steps is obtained and expressed as After calculating the cost function value by predicting L steps each time, determine the control input at this time {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (31) Apply the first input value to the UAV swarm During the prediction process of multiple L steps, it is necessary to select the prediction with the maximum search benefit. Considering the characteristics of the pigeon-inspired optimization (PIO) in simple parameter design, fast iteration speed, and strong global optimization ability, PIO is selected for optimization and solution Step Six: Establish a multi-UAV information fusion strategy based on the timestamp synchronization mechanism When information is exchanged between drones, since the determination of the activity value of the current biological grid neuron activity map is related not only to the current drone but also to the activity values detected by other drones within the range, each drone needs to maintain not only a biological grid neuron activity value map but also a timestamp map to synchronize the information exchange process during the search process; let the timestamp be t i,c , indicating the update of the search map for the nearest update of grid c by drone i; Step Seven: Output the control result map of the manned / unmanned aerial vehicle co-integration area search with biological positive and negative feedback Determine whether the iteration time has reached the maximum iteration time k max , if it has reached, then jump out of the loop, output the model predictive control results of the manned / unmanned aerial vehicle and the map of the activity values of the biological positive and negative feedback search neurons, otherwise, continue with the iterative optimization process to search the entire area.

4. The manned / unmanned aerial vehicle co-integration area search control method based on biological positive and negative feedback according to claim 2 or 3, characterized in that The PIO algorithm in the fifth step has the following four steps 1) Initialization Initialize the population size of the pigeon flock as P, the dimension of the solution space as D, the map and compass factor as R, and the number of iterations as N c1 and N c2 ; 2) Map and compass operator According to the initialized pigeon-inspired algorithm parameters, substitute them into the map and compass operator. The iteration formula of this operator is as follows Wherein, the position of the i-th pigeon is denoted as X i , and the velocity is denoted as V i , R is the map and compass parameter, t represents the number of iterations, rand is a random number between [0, 1], and X g represents the globally optimal position. When t reaches the number of iterations N c1 , the map and compass operator is jumped out. Equation (30) is the fitness function f(X i t ) for the iterative process; 3) Landmark operator According to the initialized pigeon flock algorithm parameters, the parameter landmark operator X that jumps out of the map and compass operator i t and the fitness function value f(X i t ) are substituted into the landmark operator, and the iterative formula of this operator is as follows: Where, N p represents the number of individuals in the population, t represents the number of iterations, X c represents the position of the central pigeon after each iteration, X i represents the position of the i-th pigeon, f(X i ) represents X i At this time, the fitness function is Equation (29), rand is a random number on [0, 1], and when t reaches the number of iterations N c2 it jumps out of the landmark operator; 4) Loop; all particles iterate within the boundary range through initialization, map and compass operators, and landmark operators until the maximum number of generations. After predicting the fitness function value for L steps each time, the optimal control input U is obtained. m That is, Equation (33), and its first component u m (k) acts on the UAV, and finally realizes the maximization of search efficiency coverage. {u m (k), u m (k + 1), u m (k + 2),..., u m (k + L - 1)} (33).

5. The manned / unmanned aerial vehicle co-integration area search control method based on biological positive and negative feedback according to claim 2 or 3, characterized in that The following three rules are set in the sixth step to update the timestamp map 1) When the drone searches the current grid, the update of the timestamp at this time comes from its own behavior, and the timestamp t of the searched grid c is updated at this time. i,c is the number of updated moments at this time; 2) When the updated biological grid neuron activity value map of the current grid searched by the UAV comes from the information fusion of multiple UAVs within the communication range, the timestamp of the current search grid c is updated to the timestamp closest to the current moment 3) When there are other UAVs within the range of the current UAV, information interaction is required. At this time, not only the biological grid neuron activity value map is interacted, but also it will be updated to the current latest timestamp map In the entire UAV system, for other areas, select the activity value map result represented by the latest timestamp within the confidence communication range to avoid the UAVs from repeatedly accessing some grid areas that have been determined 6. The manned / unmanned aerial vehicle co-integration area search control method based on biological positive and negative feedback according to claim 4, characterized in that The following three rules are set in the sixth step to update the timestamp map 1) When the drone searches the current grid, the update of the timestamp at this time comes from its own behavior, and the timestamp t of the searched grid c is updated at this time i,c is the number of updated moments at this time; 2) When the updated biological grid neuron activity value map of the current grid searched by the UAV comes from the information fusion of multiple UAVs within the communication range, the timestamp of the current search grid c is updated to the timestamp closest to the current moment 3) When there are other UAVs within the range of the current UAV, information interaction is required. At this time, not only the biological grid neuron activity value map is interacted, but also it will be updated to the current latest timestamp map 4) In the entire UAV system, the activity value map results represented by the latest timestamp within the confidence communication range are selected in other areas, so as to avoid the UAVs from repeatedly visiting some grid areas that have been determined.

Images