CN108830373A - The modeling method that the extensive intelligent group of imitative starling cluster flight independently cooperates with - Google Patents

Abstract

The present invention proposes a large-scale intelligent group autonomous collaborative modeling method imitating starling flock flight, including initializing the population and its parameters, setting the initial state, all Agents take the initial position as the individual optimal position, and All Agents regard the optimal value of their nearest 6 or 7 neighbors as the local optimum; for fitness function calculation, set the anisotropy of the distribution of neighbors around each Agent individual, but its nearest 6 or 7 neighbors is isotropic, the interaction between Agent individuals depends on the topological distance; select the update partner, including selecting one of the nearest 6 or 7 neighbors for an Agent as its update partner; define the interaction relationship between Agents; Update the agent's speed and position. The present invention will have very broad application prospects in military, emergency, industrial, medical and other fields such as dense formation of UAV clusters, emergency evacuation of large-scale crowds, large-scale robot group collaborative operations, and disease spread control.

Description

A Modeling Method for Autonomous Collaboration of Large-Scale Intelligent Groups Imitating Starling Flocking Flying

technical field

The invention belongs to the technical field of artificial intelligence swarm intelligence applications, and in particular relates to a modeling method for large-scale intelligent swarm autonomous collaboration imitating starling flocking flight.

Background technique

Many application fields, such as dense formation of UAV clusters, emergency evacuation of large public places, collaborative operation of industrial robot groups, etc., require large-scale intelligent agents to work together to complete. In intelligent swarm application systems, agents (such as sensors, robots, aircraft, etc.) have limited individual capabilities, but their swarms can demonstrate efficient collaborative capabilities and advanced levels of intelligent coordination. With the continuous development of technologies such as computer networks, communications, and distributed computing, many practical application systems often become very large and complex, making it impossible for a single agent to effectively implement them due to the limitations of individual knowledge and computing resources. Processing and management, so large-scale intelligent group collaboration plays a very important role in many applications. How to maximize the effect of the teamwork of agents, the research on the theory of intelligent group collaboration has always been an important topic and key to swarm intelligence. , to maximize the joint completion of complex tasks that are difficult for a single agent, the research on large-scale intelligent group systems has very important practical significance.

In recent years, many scholars at home and abroad have carried out in-depth and extensive research on the coordination problem, studying the swarm control problem from the local control strategy, and making the swarm cooperation emerge the desired behavior by designing local rules and control strategies. The research on flocking behavior originated from the simulation of flight behavior of flocks of birds by Reynolds et al. (1987). They proposed that individuals in a flock follow three simple heuristic rules, namely, Cohesion, Separation, and Alignment. A Boids model of cluster behavior was constructed. From the perspective of system control, this work is essentially an idea and strategy that does not rely on the central control mechanism and uses local rule control strategies to achieve cluster coordination. Although it is simple, it is very effective. On the basis of the three principles of Reynolds, some scholars have proposed a large number of swarm motion models, among which the more classic ones are the Swarm model, the Cucker-Smale model, and the Cavagna model. Kwong et al. (2003) described Swarm as a collection of interacting adjacent individuals, and performed swarm behavior control simulations on the Swarm model, including gathering, moving in circles, moving around "8" shapes, and moving in a straight line. Wang et al. (2005) designed a cluster local rule model by adding local feedback information to control the cluster behavior of the mobile autonomous agent group, and verified the feasibility of the method. Cucker and Smale (2007) used the adjacency matrix to describe the interaction strength between individuals, and proposed a model that focused on describing the interaction of speed among individuals, and only considered the alignment principle among the three principles. However, these methods have obvious shortcomings. The relationship between group behavior and group model parameters is unknown, and a large number of simulation experiments are needed to determine the appropriate parameter values. The control requirements of the system are far-reaching. Cavagna et al. (2015) tried to establish a general theory to unify the swarm motion model, and gave the inertial spin model of swarm motion, and the information propagated outward during the flight of birds is similar to spin waves in magnetic materials. However, the cluster system is usually highly complex, and the cluster behavior is extremely diverse. The control of a large number of cluster behaviors through the system parameter adjustment method has certain limitations, so only relying on local control strategies cannot satisfy the effective control of large-scale cluster systems.

In recent years, biologists have conducted in-depth observation and research on starling flocks. At dusk, over some areas, tens of thousands or even hundreds of thousands of starlings gather and fly together. The peculiarity is that the entire flock of birds is completely synchronized with each other during the flight, and the flight mechanism is similar to that of an avalanche and an avalanche. Equilibrium-critical systems with instantaneous transitions in crystal formation, but almost instantaneous signal processing speeds, have attracted widespread interest from researchers around the world. Ballerini et al. (2008) used computer vision technology to record the three-dimensional position of a specific individual in a flock of starlings, and found that there was anisotropy in the distribution of individuals in the flock. In a large flock of starlings, individuals use topological distance (topological distance) to interact with their nearest 6-7 individuals, which is not determined by the metric distance (metric distance) of individuals in the flock. Bode et al. (2010) developed an individual-based animal group movement model based on the observations of Ballerini et al., but this model lacks a group guidance mechanism, and the movement direction of the entire group is aimless. Young et al. (2013) gave the reason for this observation through the method of system theory. When there are uncertain factors in perception, the interaction between the agent and its 6 or 7 neighbors helps to optimize the team cohesion and balance between individual efforts. These latest important biological discoveries provide a new idea for the collaborative application of large-scale intelligent groups. Cavagna (2010) proposed to use polarization to measure the overall degree of order of starling flocks. Under the displacement-velocity framework, this topological interaction mechanism was introduced into the particle swarm algorithm to make it fly in a flight characteristic of starlings. At the same time, it has better adaptability, but a complete implementation scheme has not yet been proposed. Hereford and Blum (2011) proposed the FlockOpt algorithm, which improved the Boids model and combined the starling group motion model with the swarm intelligence algorithm to solve the problem of finding the optimal value in a single-peak search space, but failed to solve the multi-peak search question. Netjinda et al. (2015) introduced the collective response behavior of European starlings into the particle swarm algorithm to increase the diversity of the population, achieve a wider range of search space, and avoid suboptimal solutions. Qiu Huaxin and Duan Haibin (2017) tried to introduce this bird swarm flight mechanism into the practical application of UAV autonomous swarm formation control. The preliminary research results show that it is feasible to combine this mechanism with swarm intelligence collaborative research.

A summary of the latest research results at home and abroad has failed to break through the autonomous collaboration of intelligent groups in the order of magnitude.

Contents of the invention

The purpose of the present invention is to provide a large-scale intelligent group autonomous collaboration modeling method inspired by starling flocking behavior. Different from the existing particle swarm algorithm, genetic algorithm, etc., combined with the latest research results of starling behavior animal habits and computational biology, the mechanism of starling biological cluster behavior and the internal function and coordination mechanism of non-centered self-organization are analyzed. Establish a mapping mechanism for the collaborative application of starling flocks to large-scale intelligent groups.

In order to achieve the above-mentioned purpose, the large-scale intelligent group autonomous collaboration modeling method of imitating starling flocking behavior mechanism provided by the present invention comprises the following steps,

Step S1, initialize the population and parameters, including setting that in the initial state, all agents use the initial position as the individual optimal position, and all agents use the initial position as the individual local optimal position;

Step S2, the calculation of the fitness function, including following the following rules (1) and (2) to construct the framework of the topological action mechanism,

Rule ⑴, the distribution of neighbors around each Agent individual is anisotropic, but its nearest 6 or 7 neighbors are isotropic;

Rule ⑵, the interaction between Agent individuals depends on the topological distance;

Step S3, selecting an update partner, including selecting one of the nearest 6 or 7 neighbors for Agent i as an update partner Agent j;

Step S4, defining the interaction relationship between Agents;

Step S5, update the speed and position of the Agent, and return to Step S2 until the Agent group reaches the destination or the number of cycles reaches the maximum evolutionary number.

And, in step S3, update companion Agent j to follow the following rule (3) to select, then calculate the fitness value according to rule (4),

Rule ⑶, according to the principle of p _j ~ 1/d _ij , Agent i selects and updates its companion Agent j from the nearest 6 or 7 neighbors within a certain field of vision radius rV around it,

Among them, p _j is the probability, d _ij is the distance between Agent i and Agent j;

Rule ⑷, use the fitness function to evaluate the selected update companion Agent j, including the preset fitness function threshold f _threshold ; if the fitness value of Agent j is greater than f _threshold , the fitness value is poor, and Agent j is rejected Eliminated, Agent i maintains its original flight mode; otherwise, Agent i chooses Agent j as its update companion.

Moreover, in step S4, the interaction relationship between Agents is defined, and the implementation method is as follows,

Define the three radius parameters of repulsion radius rE, retention radius rM and attraction radius rA, and define the topological relationship between Agents to satisfy the following rules,

Rule ⑸, repulsion rule, Agent i repels other Agent j within its close range, that is, if d _ij < rE, then modify the flight direction of Agent i to fly away from Agent j;

Rule ⑹, keep the rule, Agent i follows other Agent j within the range of medium distance;

Rule ⑺, attracting rules, Agent i attracts other Agent j within a relatively long distance range, that is, if d _ij >rA, then modify the flight direction of Agent i to fly in the direction of Agent j;

Rule ⑻, if d _ij >rA, then there is no interaction between Agent i and Agent j.

Moreover, in step S5, update the speed and position of the Agent, including introducing the polarization factor according to the following rules (9) to control the Agent population,

Rule ⑼, by defining the polarization Φ to measure the overall order of the group, reflecting the consistency of the overall flight direction of the group,

Among them, v _i is the speed of Agent i, and ||v _i || is the norm of calculating v _i in its metric space; when Φ=0, it indicates that the overall flight direction of the cluster is chaotic; when Φ→1, it indicates that the cluster The whole basically faces the same direction.

Moreover, in step S5, update the speed and position of the Agent, including limiting the updated speed to satisfy the rule ⑽,

Rule ⑽, each agent's speed range is limited to [v _min , v _max ], where v _min and v _max are the minimum and maximum speeds of the individual Agent, respectively.

Moreover, it is used for dense formation of UAV clusters.

The technical characteristics that the present invention is different from prior art are as follows:

1. Cavagna (2010) proposed to use polarization to measure the overall order of starling flocks. Under the displacement-velocity framework, this topological interaction mechanism was introduced into the particle swarm algorithm to make it fly with starlings. At the same time, it has better adaptability, but due to the complexity of flock flight of starlings, no practical and complete realization scheme has been proposed yet. The present invention is the first to propose a large-scale intelligent group autonomous collaboration modeling method imitating starling flock flight, as well as a complete implementation plan and specific steps.

2. The essence of intelligent group collaboration is that under the stimulation of the environment, a large number of Agent individuals jointly load some of the same behavior rules, and through the feedback of these rules, a certain macroscopic orderly phenomenon is generated. The present invention proposes 10 basic simple rules that intelligent groups should follow in collaboration, that is, rules (1) to rules (10).

3. The present invention adopts the idea of "non-centered self-organization synergy theory". The individual Agent does not have a global perspective, it only cares about its nearest 6 or 7 neighbors, and searches for the optimal value in the local area. This local idea is different from the current The global view of classical particle swarm optimization is quite different.

4. Hereford and Blum (2011) proposed the FlockOpt algorithm, improved the Boids model, combined the starling group motion model with the swarm intelligence algorithm, and solved the problem of finding the optimal value in the unimodal search space, but failed to solve the problem of multiple Peak search problem. Under the displacement-velocity framework, the present invention introduces the topological interaction mechanism into the algorithm, and makes new settings for the parameters, so that it has stronger adaptability while having the flight characteristics of the flock of starlings. search space, and can also find optimal values in multimodal search spaces.

5. The idea proposed by the present invention is based on the displacement-velocity framework, and the number of intelligent groups is not limited, so it can be applied to the collaboration of large-scale intelligent groups.

Therefore, the present invention has beneficial effect:

The research on the autonomous coordination mechanism of large-scale intelligent groups has very important practical significance and application value. Inspired by the intensive flight and mutual coordination of starling flocks, the flocking behavior is introduced into the cooperation of large-scale intelligent groups, providing a new research idea for solving the two problems of intelligent individual self-adaptive regulation and intelligent group self-organization and coordination. , research methods and calculation models will have very broad application prospects in military, emergency, industrial, medical and other fields such as dense formation of UAV clusters, emergency evacuation of large-scale crowds, large-scale robot group collaborative operations, and disease spread control. .

Description of drawings

Fig. 1 is the implementation steps of large-scale intelligent group autonomous collaborative modeling imitating starling group flight of the present invention.

Detailed ways

The invention introduces the biological intelligence of starling group flight into large-scale intelligent group collaborative application, and proposes a large-scale intelligent group autonomous collaborative modeling method imitating the starling group behavior mechanism. The basic idea is: in a set of individuals (called Agents) {agent ₁ , agent ₂ ,...,agent _N } with a certain group size N in the D-dimensional space, each Agent has a certain flying speed and can determine its flying direction with distance. The initial values of Agent's position and velocity are determined by random numbers. All agents search the nearest 6 or 7 neighbors around them in the solution space. Each Agent is endowed with a certain memory function, which can remember the optimal position it has found. Its fitness value is determined by the fitness function, and the quality of its current position is evaluated accordingly. After flying per unit time, each agent dynamically adjusts its flight direction, speed and position according to the speed update equation and position update equation. Weight the center, and keep the cluster flight direction consistent through the polarization factor.

The key is:

1) Intelligent groups follow basic simple rules in collaboration, and summarize 10 basic rules, namely, rules ⑴ to rules ⑽.

2) The present invention adopts the idea of "no center theory". Individual Agent does not have a global view, it only cares about its nearest 6 or 7 neighbors, which has locality, which is completely different from the global view of particle swarm algorithm in the prior art .

Referring to Fig. 1, the embodiment of the present invention takes paying attention to 7 neighbors as an example, and the specific steps for implementation are as follows:

Step 1: Initialize the population and its parameters

Initialize the population and its parameters to prepare a replacement population for subsequent steps, including:

① Set the population size N, the maximum evolutionary algebra iter _max , and define the initial value of the evolutionary algebra k=0, where k<

iter _max .

②Set the maximum speed v _max and the minimum speed v _min , and define the initial speed for each Agent as Here rand() is defined as a random number in the interval (0,1).

③ The space range is defined as [x _min , x _max ], and the initial position for each Agent is defined as Here rand() is defined as a random number in the interval (0,1).

④ Define the initial value of individual historical optimal fitness f _i ⁽⁰⁾ = ∞, the initial value of historical local fitness

In the initial state, all agents take their initial positions as their individual optimal positions, and all agents use their initial positions as their individual local optimal positions.

Step 2: Fitness function calculation

Step 2.1: Mechanism framework for topological action

The latest zoological research shows that in a large flock of starlings, a starling individual interacts with its nearest 6 or 7 individuals; the distribution of neighbors around each individual is anisotropic, but its nearest 7 However, each neighbor is isotropic; the interaction between individuals depends on the topological structure, rather than measuring distance, therefore, follow the following basic rules (1) and rules (2), construct the topological action mechanism framework, as the basis of the method steps of the present invention, Reflect the theme of collaboration.

Rule ⑴: The distribution of neighbors around each Agent individual is anisotropic, but its 7 nearest neighbors are isotropic.

Anisotropy refers to the characteristic that each individual in the group moves in different directions; while isotropy refers to the characteristic that each individual moves in roughly the same direction.

At the beginning when there is no coordination, the individuals in the entire Agent group fly in their own direction. From the overall point of view, the direction of movement is chaotic, showing the characteristics of anisotropy. After a period of time, the individual Agents make self-adaptive adjustments according to their seven nearest neighbors. Finally, overall, the movement direction of the group is roughly the same, showing the characteristics of isotropy.

Rule ⑵: The interaction between Agent individuals depends on the topological distance, which reflects the topology-distance relationship rather than the metric-distance framework.

The "topology-distance relationship" here can actually be understood as a non-metric distance relationship, which is a relative distance. The actual distance length is not important. The key is to determine the interaction relationship between two Agents through this topology-distance .

In the follow-up step 4.2, the realization of this interaction is illustrated, that is, the use of the classic three basic rules of "repulsion-retention-attraction", namely rule ⑸ ~ rule ⑺, while satisfying rule ⑻.

Based on the framework of the topological action mechanism above, the polarization value Φ is defined in the subsequent step 5, which is used to reflect the consistency of the overall flight direction of the Agent group. The present invention proposes an autonomous collaboration method and steps, and how to judge the final effect of the collaboration realized according to these steps can be evaluated by using the defined Φ. If Φ is close to 0, it indicates that the overall flying direction of the cluster is chaotic; otherwise, it indicates that the overall cluster is basically flying in the same direction.

Step 2.2: Update the individual historical optimal fitness value

During specific implementation, the user can predetermine the fitness function f according to specific application problems, and set its threshold as f _threshold . The fitness function f( _xi ) of _xi is defined as the evaluation of the current best position of the i-th individual Agent i tending to the target point. In the k-th iterative evolution, calculate the current fitness value f _i ^(k) of Agent i. If f _i ^(k) >f _i ^(k-1) ,k>1, then update the individual historical optimal position in That is, the initial optimal position of Agent i,

Step 2.3: Update local optimal fitness value

Use the fitness function f( _xi ) to calculate the local optimal fitness value of Agent i's 7 nearest neighbors

if but

Update the local optimal fitness value

And update the local optimal position

Step 3: Select Update Companion

Step 3.1: Obtain the 7 nearest neighbors of the Agent individual

Obtain seven nearest neighbors {j ₁ ,j ₂ ,…,j ₇ } for Agent i within a certain field of view radius rV.

Step 3.2: Calculation of the roulette wheel selection mechanism, select and update companions for the Agent individual

Choose its update partner (update partner) j for Agent i, following rule (3).

Rule ⑶: According to the principle of p _j ~ 1/d _ij , select update partner j from the seven nearest neighbors {j ₁ ,j ₂ ,…,j ₇ } for Agent i within a certain field of view radius rV around it.

Among them, p _j is the probability, d _ij is the distance between Agent i and Agent j. The smaller the distance, the greater the probability of selection. Use the roulette selection mechanism to calculate p _j , that is

d _iq refers to the distance between Agent i and Agent q.

where q is the count used to sum the distances of Agent i to its 7 neighbors, thus taking values from 1 to 7.

Rule ⑷: Use f(x _j ) to evaluate the selected update partner j to ensure that the partner with better fitness value is selected. According to the preset fitness function threshold f _threshold , if f(x _j )>f _threshold , the fitness value is poor, Agent j is eliminated, and Agent i keeps its original flight mode; otherwise, Agent i chooses Agent j as Your own update companion.

In the present invention, the Agent individual only cares about the nearest 7 neighbors within a certain range of vision, randomly selects one of them after calculating the probability according to the rule (3), and then judges whether the neighbor is good or bad after calculating the fitness value according to the rule (4). Will be the "role model" for this Agent.

Step 4: Define the interaction relationship between Agents

According to the selected update partner j, determine the interaction relationship between Agent i and Agent j. Specific steps are as follows:

Step 4.1: Define the radius parameter

Define three radius parameters: repulsion radius rE, retention radius rM, and attraction radius rA. The repulsion radius is the minimum distance kept between the Agent and its updating companions to avoid collisions between the two. Sets the hold radius, i.e. within the medium distance between the agent and its update peers, the agents maintain the same motion pattern between each other. At the same time, in order to maintain the cohesion of the whole group, the agent is attracted by the distant peers, and the attraction radius is set to the maximum range of the agent's search space, that is, the field of view radius rV.

Step 4.2: Define the topological relationship between Agents

Each agent embodies the topology-distance relationship in the process of movement, and follows the classic three basic rules of "repulsion-retention-attraction", namely rule ⑸ to rule ⑺, while satisfying rule ⑻.

Rule ⑸: Repulsion rule, Agent i repels other Agent j within its close range, that is, if d _ij < rE, then modify the flight direction of Agent i to fly away from Agent j.

Rule ⑹: keep the rule, Agent i follows other Agent j within the range of medium distance.

Rule ⑺: Attraction rule, Agent i attracts other Agent j within its long-distance range, that is, if d _ij >rA, then modify the flight direction of Agent i to fly in the direction of Agent j.

Rule ⑻: If d _ij >rA, then there is no interaction between Agent i and Agent j.

Step 5: Update the Agent's speed and position

In specific implementation, after the update is executed, it can be designed to return to step S2 until the agent group reaches the destination or the number of cycles reaches the maximum evolutionary generation, so as to realize the update process of continuous iteration.

Step 5.1: Introduce polarization factors to control the Agent population

In view of the fact that the traditional model lacks a group guidance mechanism, and the movement direction of the entire group is non-target, the traditional model is improved, and the trend of each Agent in the group is controlled to guide it to find the optimal value in the unimodal search space.

Rule ⑼: By defining the polarization Φ to measure the overall order of the group, reflecting the consistency of the overall flight direction of the group, that is

Among them, v _i is the velocity of Agent i, and ||v _i || is the norm of computing v _i in its metric space. When Φ=0, it indicates that the overall flying direction of the cluster is chaotic; when Φ→1, it indicates that the overall cluster is basically facing the same direction.

Step 5.2: Calculate the kinetic energy of the agent

The kinetic energy of Agent i is defined as formula (3), namely

Among them, m is the quality of Agent i, assuming that the quality of Agent individual is unit 1. v _i is the speed of Agent i. The total kinetic energy of the population is Population kinetic energy reflects evolutionary rhythm. If the kinetic energy is large, it indicates that the evolutionary rhythm is fast, and the inertia weight ω needs to be adjusted to keep decreasing; on the contrary, it indicates that the evolutionary rhythm is slow, and ω needs to be increased to ensure that the individual does not over-concentrate.

Step 5.3: Define the Velocity Update Equation

Introduce the pbest variable (the optimal position found by the particle) in the particle swarm optimization algorithm (PSO), that is, the optimal position found by the particle, and use the formula (4) to calculate, that is

pbest=x _best -x (4)

Among them, x is the current position of the Agent, and x _best is the historical best position of the Agent.

Under the displacement-velocity framework, the topological interaction mechanism is introduced into the algorithm, and new parameters are set to make it more adaptable while having the flight characteristics of starling flocks. , and can also find the optimal value in a multimodal search space. The speed update equations of Agent i in k+1 iterations are formula (5) and formula (6).

Among them, N _i is the set of 7 neighbors of Agent i, and n is used to identify one of the set of N _i for calculation. is the average of the speeds of the 7 neighbors. ω is the inertia weight, which is adaptively obtained by formula (7), namely

ω＝ω _max -(ω _max -ω _min )×k/iter _max +α×rand( )×e ^-E (7)

k is the evolution algebra, the present invention uses the superscript (k) to identify the parameters of the kth iterative evolution, α is the control factor, and ω _min and ω _max are the weight dynamic ranges. E is the kinetic energy of the Agent. Through the function of E, it plays a certain buffer role in the control of the inertia weight ω, so that it can be adjusted adaptively, and the convergence of the algorithm can be guaranteed to the greatest extent.

pbest _i is the optimal position experienced by Agent i, and lbest _i is the optimal position experienced by all 7 neighbors of Agent i, which is a local extremum, that is, lbest _i =max{pbest _j ,j=1,2,… ,7}.

In the traditional particle swarm optimization algorithm, the position update formula uses gbest _i to represent the global optimal position. Different from the traditional particle swarm optimization algorithm, the present invention is based on "no-center theory", that is, the principle of local 7 neighbors, and uses lbest _i variable to indicate the optimal position experienced by all 7 neighbors of Agent i.

c ₁ and c ₂ are acceleration constants, usually c ₁ and c ₂ =2.

c ₃ is the topology learning factor, as shown in the formula (8), which plays a role in maintaining the diversity of the population, so that the realization algorithm can search in a wider range.

When ω decreases, the population gradually loses its diversity. At this time, the topological effect is strengthened to enhance the information interaction between the Agent individual and other Agent individuals, so as to prevent the population from converging to a local point. In order to ensure the accuracy, by default, after finding the local optimal solution after two-thirds of the generations, the population diversity will no longer be maintained. At this time, set c ₃ =0; in other cases, c ₃ =1-ω.

The above updated speed satisfies rule ⑽, namely

Rule ⑽: The speed range of each Agent is limited to [v _min ,v _max ]. At the same time, in the process of iterative evolution, if the agent's position exceeds the boundary position when flying at a certain speed, its position is limited to the boundary value [x _min , x _max ].

Step 5.4: Define the Position Update Equation

The position is updated according to the update speed, and the position update equation of Agent i in k+1 iterations is shown in formula (9).

Among them, τ is unit time, and τ is usually set to 1.

Repeat steps 2 to 5 above, evolutionary algebra k=k+1, Agent i selects its seven nearest neighbors as interaction objects. After being weighted with the speed of adjacent individuals, the speed and position of the Agent individual are updated through continuous iteration, so that the behavior of the entire cluster tends to be consistent until the group reaches the destination. Alternatively, if k ≥ itermax, that is, the maximum number of evolutionary generations is reached, the loop is terminated.

During specific implementation, computer software technology can be used to realize the automatic operation of the above process. For implementation reference, the following table is provided:

Table 1: Variables and their descriptions

The large-scale intelligent group autonomous cooperation modeling method proposed by the present invention imitates starling cluster flight can be applied in many fields, especially in the practical application of large-scale unmanned aerial vehicle cluster dense formation.

For example, in a group of drones of a certain size and size in a certain space, each drone has a certain flight speed and can determine the direction and distance of its flight. Its fitness value is determined by the fitness function (defined as the reciprocal of the distance between the current position and the destination point position), and the quality of its current position is evaluated accordingly. All UAVs search the nearest 6 or 7 neighbors around them within their field of view, and choose their update companions from them. A roulette selection mechanism is used to retain companions with better fitness values or eliminate companions with poor fitness values. According to the repulsion radius, retention radius and attraction radius, the interaction force between it and the update partner is calculated, and the repulsion-retention-attraction topological relationship between the two is determined. After flying per unit time, each UAV dynamically adjusts its flight direction, speed and position according to the speed update equation and position update equation, and flies to its own optimal position and the weighted local optimal position of 6 or 7 neighbors. center. During the flight, the consistency of the flight direction of the UAV cluster is maintained through the polarization factor, and the evolutionary rhythm is adjusted through the kinetic energy of the population. According to this method, the UAV cluster formation can be formed, which can fly autonomously like a large-scale starling flock.

In the future environment of informatization, networking, and system confrontation, decentralized and autonomous large-scale UAVs will cooperate closely with each other to realize autonomous formation flight of large-scale UAV groups. Compared with a single UAV system, UAV clusters can use collective behavior to cooperate with each other to efficiently and to the greatest extent jointly complete complex tasks that are difficult for a single UAV, and form a scale advantage.

Claims (6)

1. A large-scale intelligent group autonomous collaborative modeling method imitating starling group flight, it is characterized in that: comprise the following steps, Step S1, initialize the population and parameters, including setting that in the initial state, all agents use the initial position as the individual optimal position, and all agents use the initial position as the individual local optimal position; Step S2, fitness function calculation, including following the following rules ⑴ and ⑵, constructing the topological action mechanism framework, rule ⑴, the distribution of neighbors around each Agent individual is anisotropic, but its nearest 6 or 7 neighbors are anisotropic the same sex; rule (2), the interaction between Agent individuals depends on the topological distance; Step S3, selecting an update partner, including selecting one of the nearest 6 or 7 neighbors for Agent i as an update partner Agent j; Step S4, defining the interaction relationship between Agents; Step S5, update the speed and position of the Agent, and return to Step S2 until the Agent group reaches the destination or the number of cycles reaches the maximum evolutionary number. 2. The modeling method of the large-scale intelligent group autonomous collaboration of imitation European starling cluster flight as claimed in claim 1, is characterized in that: in step S3, update companion Agent j to follow following rule (3) to select, then calculate according to rule (4) fitness value, Rule ⑶, according to the principle of p _j ~ 1/d _ij , Agent i selects and updates its companion Agent j from the nearest 6 or 7 neighbors within a certain field of vision radius rV around it, Among them, p _j is the probability, d _ij is the distance between Agent i and Agent j; Rule ⑷, use the fitness function to evaluate the selected update companion Agent j, including the preset fitness function threshold f _threshold ; if the fitness value of Agent j is greater than f _threshold , the fitness value is poor, and Agent j is rejected Eliminated, Agent i maintains its original flight mode; otherwise, Agent i chooses Agent j as its update companion. 3. The modeling method of the large-scale intelligent group autonomous collaboration of imitation European starling group flight as claimed in claim 1, it is characterized in that: in step S4, define the interaction relation between Agent, the realization mode is as follows, Define the three radius parameters of repulsion radius rE, retention radius rM and attraction radius rA, and define the topological relationship between Agents to satisfy the following rules, Rule ⑸, repulsion rule, Agent i repels other Agent j within its close range, that is, if d _ij < rE, then modify the flight direction of Agent i to fly away from Agent j; Rule ⑹, keep the rule, Agent i follows other Agent j within the range of medium distance; Rule ⑺, attracting rules, Agent i attracts other Agent j within a relatively long distance range, that is, if d _ij >rA, then modify the flight direction of Agent i to fly in the direction of Agent j; Rule ⑻, if d _ij >rA, then there is no interaction between Agent i and Agent j. 4. The modeling method of large-scale intelligent group autonomous cooperation imitating starling flock flight as claimed in claim 1, characterized in that: in step S5, update the agent's speed and position, including introducing poles according to the following rules (9) Chemical factors to control the Agent population, Rule ⑼, by defining the polarization Φ to measure the overall order of the group, reflecting the consistency of the overall flight direction of the group, Among them, v _i is the speed of Agent i, and ||v _i || is the norm of calculating v _i in its metric space; when Φ=0, it indicates that the overall flight direction of the cluster is chaotic; when Φ→1, it indicates that the cluster The whole basically faces the same direction. 5. The modeling method of large-scale intelligent group autonomous cooperation imitating European starling cluster flight as claimed in claim 1, characterized in that: in step S5, update the agent's speed and position, including limiting the updated speed to meet Rule ⑽, rule ⑽, each agent's speed range is limited to [v _min , v _max ], where v _min and v _max are the minimum and maximum speeds of the individual Agent, respectively. 6. as claimed in claim 1 or 2 or 3 or 4 or 5, the modeling method of the large-scale intelligent group autonomous cooperation of imitation European starling cluster flight is characterized in that: it is used for the dense formation of unmanned aerial vehicle cluster.