A particle swarm search method based on communication-free robots

Abstract

This invention proposes a particle swarm search method based on robots that do not require communication. In a two-dimensional or three-dimensional optimization search task, multiple robots that do not have the ability to communicate with each other are dispersed within the search space. The group of robots is considered a particle swarm, and each robot is considered a particle. The S&W particles in the corresponding particle swarm algorithm are replaced with S&W robots, and the S&W particle set is replaced with the S&W robot set. Environmental modeling is performed, and the field source signal value detected by the robot is used as the fitness value indicator of the particle swarm algorithm. This invention improves the traditional particle swarm algorithm by utilizing the one-way active detection capability of particles. The particle swarm does not need to communicate with each other to share fitness and position information. This makes the particle swarm algorithm, for the first time, capable of iterative search within the optimization problem space without communication. The method is applied to swarm robot search tasks and exhibits excellent search performance.

Description

Technical Field

[0001] The present invention relates to the fields of swarm intelligence and swarm robot target search, and in particular to a particle swarm algorithm variant without the need for communication between particles and a target search method for swarm robots using the variant. Background Art

[0002] The particle swarm optimization (PSO) algorithm, also known as particle swarm optimization or particle swarm optimization (PSA), is a heuristic evolutionary algorithm proposed by Kennedy and Eberhart in 1995. It simulates the foraging motion of flocks of birds or schools of fish. It effectively searches for and finds candidate solutions in problems with high-dimensional solution spaces, effectively solving global optimization problems. Due to its simple parameters and rapid convergence, the PSO has rapidly developed into a mainstream global optimization technique and has been successfully applied to solve practical problems in various fields. However, due to the randomness of the particle search within the algorithm, this method does not guarantee that the optimal solution it finds is the true optimal solution. Consequently, numerous researchers have proposed variations of the PSO algorithm, using various methods to balance the swarm's global and local search capabilities to effectively improve its optimization performance.

[0003] Since its inception, most current particle swarm optimization (PSO) variants have required communication between particles to share their fitness values ​​and find the optimal position they have searched, thereby achieving effective collaboration. Because current PSO algorithms rely on communication between particles, they are no longer applicable when faced with poor or no communication.

[0004] For scenarios where communication is not required, some researchers have proposed source localization algorithms for 2D or 3D space. These algorithms can solve single-mode, single-source problems in 2D or 3D space without communication. However, these algorithms do not consider high-dimensional or multi-mode, multi-source problems. Therefore, they cannot be directly applied to high-dimensional or multi-mode problems without further extension.

[0005] Swarm robotics is a self-coordinating system of multiple robots that merge into a single organism to perform a common task. Swarm robotics technology has emerged in the field of artificial swarm intelligence and in biological research on insects, ants, and other natural phenomena where swarm behavior occurs. Swarm robotics typically consists of robots of the same type that interact with each other and the environment using simple sensors. Swarm robotics aims to solve problems that are intractable for individual robots or to achieve goals more efficiently through the use of a group.

[0006] If robots are considered particles and swarms of robots are considered particle swarms, the particle swarm algorithm (PSO) can be applied to swarm robotic search tasks. However, relying on inter-robot communication to share fitness values ​​is challenging in some application scenarios. For example, acoustic communication in underwater environments is affected by high transmission loss, high ambient noise, and high propagation delay; strong electronic / electromagnetic interference in the search space can make communication channels unreliable; swarm robots may encounter serious software or hardware failures that prevent them from establishing communication channels; and robots may be in a competitive relationship and unwilling to share their searched fitness data. Therefore, it is necessary to develop a PSO variant that exhibits excellent search performance while eliminating the need for inter-particle communication and apply it to swarm robotic search tasks. Summary of the Invention

[0007] To address the aforementioned shortcomings of existing particle swarm algorithm variants, which are unable to be applied to scenarios where inter-robot communication is not required, the present invention proposes a particle swarm search method based on inter-robot communication. This particle swarm algorithm variant utilizes particles' own active detection capabilities to detect the positions and velocities of other particles, rather than relying directly on inter-particle communication. Each particle simultaneously searches the problem space following its selected leader and the detected swarm center position. In the particle swarm algorithm, pbest represents the particle's individual best historical position (i.e., personal best), which corresponds to the particle's optimal fitness. In the present invention, S&W stands for stop-and-wait; an S&W particle is one that remains at its pbest position with a zero velocity. When a single particle fails to continuously update its pbest within a preset number of generations, K, it reverts to its pbest position, becomes an S&W particle, and joins the S&W particle set. When the total number of particles in the S&W particle set reaches or exceeds a preset threshold, M, each S&W particle randomly selects another S&W particle as its leader. Based on this, the particle swarm can iteratively search in the search space without communication until the search ends and the optimal solution is obtained. By applying this particle swarm algorithm variant that does not require communication between particles to a swarm robot search task, the present invention realizes a particle swarm search method that does not require communication between robots.

[0008] The present invention is achieved through the following technical solutions:

[0009] A particle swarm search method based on robots without communication is proposed. In an actual two-dimensional or three-dimensional optimization search task, multiple robots without the ability to communicate with each other are dispersed in the search space. The group of robots is regarded as a particle swarm, and each robot is regarded as a particle. The S&W particles (set) in the corresponding particle swarm algorithm are replaced with S&W robots (set). The environment is modeled and the field source signal value detected by the robots is used as the fitness value indicator of the particle swarm algorithm. The specific steps are as follows:

[0010] Step 1.1: Initialize the particle swarm optimization algorithm parameter settings: number of particles N, dimension of the optimization problem D, inertia weight ω, constant acceleration factors c1 and c2, threshold K for failure of continuous update of pbest, particle number threshold M of S&W particle set, and maximum number of iterations T.

[0011] Step 1.2: Randomly initialize the initial positions and velocities of N robots, that is, the position of robot i and speed where i∈{1, 2, ..., N}.

[0012] Step 1.3: Randomly initialize the leaders of N robots. For robot i, its leader is marked as l i .

[0013] Step 1.4: The robots do not have the ability to communicate with each other, but obtain the ability to locate and sense speed through one-way active detection, including but not limited to visual observation, radar detection, and sonar search. The robot detects the position of other robots and calculates the center position of the group robot through the positions of all robots.

[0014]

[0015] Wherein, d∈{1, 2, ..., D} is the d-th dimension search space of the optimization problem.

[0016] Step 1.5: Traverse all robots in the swarm. If the traversed robot i has selected a leader, execute steps 1.6 and 1.7, where i∈{1, 2, ..., N}.

[0017] Step 1.6: Update the robot's velocity and position. The update formulas are as follows:

[0018]

[0019]

[0020] Among them, r1 and r2 are D-dimensional random number vectors; Indicates the leader position selected by robot i.

[0021] Step 1.7: Evaluate the robot's fitness based on its current position, update pbest, and determine whether the robot can become an S&W robot (S&W is the abbreviation of stop-and-wait); if it can become an S&W robot, fall back to pbest and wait.

[0022] Step 1.8: Traverse all S&W robots in the swarm, and execute step 1.9 for each S&W robot.

[0023] Step 1.9: Check whether the number of robots in the S&W robot set reaches or exceeds the threshold M. If it reaches or exceeds the threshold M, randomly select a leader for the S&W robot.

[0024] Step 1.10: If the current number of iterations is greater than the maximum number of iterations T or the optimal target source is successfully searched, proceed to step 1.11; otherwise, return to step 1.4 for the next iteration.

[0025] Step 1.11: Return the best pbest and its fitness value among all robots as the final search result.

[0026] Furthermore, in the above step 1.3, for each robot in the swarm of robots {1, 2, ..., N}, another robot in the swarm is randomly selected as its leader:

[0027] l i =randi({1, 2,...,N}-{i});

[0028] Among them, randi(.) is a random selection function used to randomly select a robot from the group of robots, l i Is the leader of robot i.

[0029] In the above step 1.4, in the actual application of the optimization search task, the robot's one-way active detection capability can be achieved by configuring cameras, millimeter-wave radars, or sonar devices in the optimization search task in the air, on the ground, on the sea surface, or on the seabed, to one-way actively detect the positions of other robots in the group and perceive the speed of other robots.

[0030] In step 1.6 above, robot i simultaneously follows the leader position it chooses and the center position of the swarm robot to update its speed.

[0031] In the above step 1.7, it is divided into the following sub-steps:

[0032] Step 1.7.1: Each robot updates its pbest the number of times it fails to update continuously. For the minimum optimization problem, the update formula is:

[0033]

[0034] For the maximum optimization problem, the update formula is:

[0035]

[0036] Among them, f(.) is the fitness function of the optimization problem to be searched, k i is the number of consecutive failed updates of pbest for robot i.

[0037] Step 1.7.2: Each robot determines the number of times k that its pbest consecutive updates have failed i Whether the threshold K is reached or exceeded. If it is reached or exceeded, the robot will fall back to its pbest position, clear its previously recorded leader, reset its speed to zero, and become an S&W robot:

[0038] if k i ≥K

[0039] X i =pbest;l i =0; V i =(0 1 , 0 2 ,...,0 D );

[0040] end

[0041] In the above step 1.9, it is divided into the following sub-steps:

[0042] Step 1.9.1: Each S&W robot actively detects other S&W robots and adds all detected S&W robots to the S&W robot set:

[0043] i∈P if V i =(0 1 , 0 2 ,...,0 D )

[0044] The zero velocity vector V i =(0 1 , 0 2 ,...,0 D ) indicates that particle i is detected as an S&W robot. P represents the set of detected S&W robots, and P includes all S&W robots.

[0045] Step 1.9.2: When the number of robots in the S&W robot set P reaches or exceeds the threshold M, each S&W robot randomly selects another S&W robot as its leader:

[0046] l i =randi(P-{i})if|P|≥M and i∈P

[0047] Where |P| represents the number of robots in the S&W robot set, and randi(.) is a random selection function used to randomly select another S&W robot.

[0048] Beneficial effects:

[0049] Traditional particle swarm algorithms require particles to communicate with each other to share their fitness values ​​and position information. This invention improves on the traditional particle swarm algorithm by utilizing the particles' unidirectional active detection capability. This eliminates the need for particle swarms to communicate and share fitness values ​​and position information. This enables the particle swarm algorithm to iteratively search within the optimization problem space without communication for the first time, and has been applied to swarm robot search tasks. Comparative experiments using this method with two other particle swarm algorithm variants, SPSO and CSO, demonstrated that the proposed method exhibits exceptional search performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Schematic diagram of the two-dimensional multimodal search task scenario and its fitness value.

[0051] Figure 2 Schematic diagram of the contour lines of the fitness value in the search space in a two-dimensional multi-modal search task scenario.

[0052] Figure 3 The figure is a flow chart of applying the method of the present invention when the swarm robots are regarded as a particle swarm.

[0053] Figure 4 Schematic diagram of the search process (first iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to a state at a different number of iterations in the search process.

[0054] Figure 5 Schematic diagram of the search process (fourth iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to a state at a different number of iterations in the search process.

[0055] Figure 6 Schematic diagram of the search process (sixth iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to a state at a different number of iterations in the search process.

[0056] Figure 7Schematic diagram of the search process (seventh iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to a state at a different number of iterations in the search process.

[0057] Figure 8 Schematic diagram of the search process (tenth iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to a state at a different number of iterations in the search process.

[0058] Figure 9 This is a schematic diagram of the search process (fifty-fifth iteration) in a simulated dual-source environment using the method of the present invention, where each diagram corresponds to the state at a different number of iterations in the search process.

[0059] Figure 10 Graph showing the convergence of a swarm robot applying the method of the present invention and a comparative particle swarm algorithm variant under the example function f1 of CEC2017.

[0060] Figure 11 Convergence curves of swarm robots applying the method of the present invention and a comparative particle swarm algorithm variant under the example function f9 of CEC2017.

[0061] Figure 12 In the example function f at CEC2017 11 Below, the convergence curves of the swarm robot applying the method of the present invention and the comparative particle swarm algorithm variant are shown.

[0062] Figure 13 In the example function f at CEC2017 16 Below, the convergence curves of the swarm robot applying the method of the present invention and the comparative particle swarm algorithm variant are shown.

[0063] Figure 14 In the example function f at CEC2017 22 Below, the convergence curves of the swarm robot applying the method of the present invention and the comparative particle swarm algorithm variant are shown.

[0064] Figure 15 In the example function f at CEC2017 26 Below, the convergence curves of the swarm robot applying the method of the present invention and the comparative particle swarm algorithm variant are shown. DETAILED DESCRIPTION

[0065] By fully leveraging the active detection capabilities of robots, the present invention, for the first time, extends a variant of the particle swarm algorithm that does not require communication between particles to a search task scenario where communication between robots is not required. During the implementation of the present invention, the swarm robots are regarded as a particle swarm, and each robot is regarded as a particle. The S&W particles (set) in the particle swarm algorithm variant are replaced with S&W particles (set). The particle swarm algorithm variant described in the present invention cleverly designs a mechanism for particles to follow their leader and continuously update pbest (individual historical optimal position) during the particle swarm center search, and if the particle fails to fall back to its pbest position, it can fully utilize the historical optimal solution of the particle search; at the same time, the present invention designs a mechanism for S&W particles to randomly select a leader from the S&W particle set, thereby increasing the leader diversity of the particle swarm. The two mechanisms introduced in the present invention can effectively balance the global exploration and local development capabilities of the particle swarm algorithm. Compared with the particle swarm algorithm variants SPSO and CSO that require communication between particles, experiments show that the method of the present invention has a more outstanding optimization search capability.

[0066] The present invention is based on a particle swarm search method that does not require communication between robots. In actual two-dimensional or three-dimensional optimization search tasks (refer to Figure 1 and Figure 2 Schematic diagram of a scenario for an optimization search problem, in which there are multiple local extreme values ​​and a global optimal value, and the field source signal value and its position corresponding to the unique global optimal value need to be found in the search space. Multiple robots that do not have the ability to communicate with each other are dispersed in the search space. The group robots are regarded as a particle swarm, and each robot is regarded as a particle. The S&W particles (set) in the corresponding particle swarm algorithm are replaced with S&W robots (sets), and environmental modeling is performed. The field source signal value detected by the robot is used as the fitness value indicator of the particle swarm algorithm. The method of the present invention can be applied to group robot search tasks and can demonstrate excellent search performance. The so-called particle swarm search task based on the absence of communication between robots in the embodiment, that is, the task of the technical solution of the present invention, is an example and not a limitation, and can be applied to the following scenarios:

[0067] Localizing the source of hazardous gas leaks: In an area where a hazardous gas leak exists but the source is unknown, multiple physical robots with active detection capabilities but no intercommunication capabilities can locate the leak source based on the gas concentration values ​​detected at different locations and the actively detected positions and speeds of other robots.

[0068] Post-disaster rescue: To locate and rescue survivors under the rubble after a disaster, robots equipped with sensing equipment such as radar life detectors are used to detect the strength of vital signs at various locations on the scene when communication conditions are not available, and the search method of the present invention is used to locate the location with the strongest vital signs.

[0069] In the embodiments, the so-called robots are robots that have active detection capabilities but lack intercommunication capabilities. Examples, but not limitations, include drones, unmanned vehicles, and unmanned underwater vehicles. The active detection capabilities described in the present invention include, but are not limited to, the ability to obtain position and velocity identification through visual observation, radar detection, sonar search, and other means.

[0070] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0071] Example

[0072] like Figure 3 As shown, this embodiment utilizes a particle swarm search method that does not require communication between robots. In the actual application of optimization search tasks, the group of robots needs to search for the field strength value and location of the optimal source. Multiple robots that do not have the ability to communicate with each other are dispersed in the search space. The group of robots is regarded as a particle swarm, and each robot is regarded as a particle. The S&W particles (set) in the corresponding particle swarm algorithm are replaced with S&W robots (sets). Environmental modeling is performed, and the field source signal value detected by the robot is used as the fitness value indicator of the particle swarm algorithm. Specifically, the following steps are included:

[0073] Step 1.1: Initialize the particle swarm optimization algorithm parameter settings: number of particles N = 50, dimension of the optimization problem D = 50, inertia weight ω = 0.72984, constant acceleration factors c1 = 1.496172 and c2 = 1.496172, threshold for pbest continuous update failure K = 2, particle number threshold of the S&W particle set M = 25, and maximum number of iterations T = 10000.

[0074] Step 1.2: Randomly initialize the initial positions and velocities of N robots, that is, the position of robot i and speed where i∈{1, 2, ..., N}.

[0075] Step 1.3: Randomly initialize the leaders of N robots. For robot i, its leader is marked as l i .

[0076] Step 1.4: The robots do not have the ability to communicate with each other, but obtain the ability to locate and sense speed through one-way active detection, including but not limited to visual observation, radar detection or sonar search. The robots use the active detection capability to detect the positions of other robots and calculate the center position of the group robot through the positions of all robots.

[0077]

[0078] Wherein, d∈{1, 2, ..., D} is the d-th dimension search space of the optimization problem.

[0079] Step 1.5: Traverse all robots in the swarm. If the traversed robot i has selected a leader, execute steps 1.6 and 1.7, where i∈{1, 2, ..., N}.

[0080] Step 1.6: Update the robot's velocity and position. The update formulas are as follows:

[0081]

[0082]

[0083] Among them, r1 and r2 are D-dimensional random number vectors; Indicates the leader position selected by robot i.

[0084] Step 1.7: Evaluate the robot's fitness based on its current position, update pbest, and determine whether the robot can become an S&W robot (S&W is the abbreviation of stop-and-wait); if it can become an S&W robot, fall back to pbest and wait.

[0085] Step 1.8: Traverse all S&W robots in the swarm, and execute step 1.9 for each S&W robot.

[0086] Step 1.9: Check whether the number of robots in the S&W robot set reaches or exceeds the threshold M. If it reaches or exceeds the threshold M, randomly select a leader for the S&W robot.

[0087] Step 1.10: If the current number of iterations is greater than the maximum number of iterations T or the optimal target source is successfully searched, proceed to step 1.11; otherwise, return to step 1.4 for the next iteration.

[0088] Step 1.11: Return the best pbest and its fitness value among all robots as the final search result.

[0089] Furthermore, in the above step 1.3, for each robot in the swarm of robots {1, 2, ..., N}, another robot in the swarm is randomly selected as its leader:

[0090] l i =randi({1, 2,...,N}-{i});

[0091] Among them, randi(.) is a random selection function used to randomly select a robot from the group of robots, l iIs the leader of robot i.

[0092] In the above step 1.4, in the actual application of the optimization search task, the robot's one-way active detection capability can be achieved by configuring cameras, millimeter-wave radars, or sonar devices in the optimization search task in the air, on the ground, on the sea surface, or on the seabed, to one-way actively detect the positions of other robots in the group and perceive the speed of other robots.

[0093] In step 1.6 above, robot i simultaneously follows the leader position it chooses and the center position of the swarm robot to update its speed.

[0094] In the above step 1.7, it is divided into the following sub-steps:

[0095] Step 1.7.1: Each robot updates its pbest the number of times it fails to update continuously. For the minimum optimization problem, the update formula is:

[0096]

[0097] For the maximum optimization problem, the update formula is:

[0098]

[0099] Among them, f(.) is the fitness function of the optimization problem to be searched, k i is the number of consecutive failed updates of pbest for robot i.

[0100] Step 1.7.2: Each robot determines the number of times k that its pbest consecutive updates have failed i Whether the threshold K is reached or exceeded. If it is reached or exceeded, the robot will fall back to its pbest position, clear its previously recorded leader, reset its speed to zero, and become an S&W robot:

[0101] if k i ≥K

[0102] X i =pbest;l i =0; V i =(0 1 , 0 2 ,...,0 D );

[0103] end

[0104] In the above step 1.9, it is divided into the following sub-steps:

[0105] Step 1.9.1: Each S&W robot actively detects other S&W robots and adds all detected S&W robots to the S&W robot set:

[0106] i∈P if V i =(0 1 , 0 2 ,...,0 D )

[0107] The zero velocity vector V i =(0 1 , 0 2 ,...,0 D ) indicates that robot i is detected as an S&W robot. P represents the set of detected S&W robots, which includes all S&W robots.

[0108] Step 1.9.2: When the number of robots in the S&W robot set P reaches or exceeds the threshold M, each S&W robot randomly selects another S&W robot as its leader:

[0109] l i =randi(P-{i})if|P|≥M and i∈P

[0110] Where |P| represents the number of robots in the S&W robot set, and randi(.) is a random selection function used to randomly select another S&W robot.

[0111] Figures 4 to 9 This is the process of the swarm robot using the above embodiment method to perform an optimal search in a two-dimensional search space with two local field source signal extreme values. There are multiple robots that do not have the ability to communicate with each other scattered in the search space. The swarm robot is regarded as a particle group and each robot is regarded as a particle. The parameters are set as N = 20, D = 2, ω = 0.72984, c1 = 1.496172, c2 = 1.496172, K = 3, M = 10. The direction of the arrow indicates that the robot follows the leader it selects. Figure 4 As shown in , at the initialization of the first iteration, all robots randomly select another robot in the group as the leader. Figure 5 As shown in , at the 4th iteration, the three robots failed to update pbest continuously and the number of failures reached or exceeded the threshold K = 3, so they became S&W robots. Figure 6 As shown in , at the 6th iteration, the total number of S&W robots exceeds the threshold M = 10, so each S&W robot randomly selects another S&W robot as the leader. Figure 7As shown in , at the 7th iteration, the S&W robot that selected the leader in the previous generation begins to follow its leader and the center of the swarm robot to search, and at the same time, the other three robots in the swarm robot become new S&W robots. Figure 8 As shown in , at the 10th iteration, the total number of newly generated S&W robots exceeds the threshold M = 10 again, and these S&W robots randomly select a leader again. Figure 9 As shown in Figure 3, at iteration 55, as the search proceeds, all robots converge to the global optimal source position.

[0112] Verification of this embodiment on the CEC2017 benchmark function test set:

[0113] In order to more intuitively verify the performance of the present invention in the optimization search problem, 29 functions in the CEC2017 (N. Awad, M. Ali, J. Liang, B. Qu, and P. Suganthan, "Problem definitions and evaluation criteria for the CEC 2017 special session and competition on single objective real-parameter numerical optimization," Nanyang Technological University, Singapore and Computational Intelligence Laboratory, Zhengzhou University, Zhengzhou, China, Technical Report, vol. 10, 2017.) benchmark function test set (excluding the unstable function Function2) are used to simulate the minimization problem in a complex signal environment, where the target source of each function has only one minimum value, and the remaining extreme values ​​are all false targets.

[0114] The proposed particle swarm optimization variant (CfPSO) that does not require communication between robots is applied to a swarm robot search task. The results are compared using the CEC2017 benchmark function test set. Each algorithm is run 51 times on each function, and the average of the search results is taken.

[0115] Table 1 - Comparison of PSO algorithm variant parameter settings

[0116]

[0117] The detailed data of the convergence accuracy comparison experiment is shown in Table 2, where Mean represents the average search result of 51 runs, Std. represents the variance of the search results, Rank represents the ranking of the compared algorithms, the best search result of each function is marked in bold, and AR is the average ranking of each algorithm after summing up the Rank values ​​on the 29 functions and dividing by 29.

[0118] Convergence accuracy comparison experiments show that CfPSO outperforms SPSO on 24 of the 29 functions and outperforms CSO on all functions. CfPSO's overall average ranking is 1.17, significantly ahead of SPSO and CSO.

[0119] Table 2-CEC2017 Convergence Accuracy Comparison Experiment

[0120]

[0121] Figures 10 to 15 Shows 6 example functions (f1,f9,f 11 ,f 16 ,f 22 ,f 26). CfPSO represents the method of the present invention, while SPSO and CSO are particle swarm optimization variants used for comparison. The convergence curves show that CfPSO converges faster than SPSO. While CSO converges at a similar speed to CfPSO, its convergence accuracy is significantly worse than CfPSO.

[0122] Comparative experiments on the CEC2017 test function set show that CfPSO achieves better convergence accuracy and speed than SPSO and CSO. This demonstrates that the CfPSO method can significantly improve the optimization search capabilities of swarm robots without requiring inter-robot communication, allowing them to quickly and efficiently locate the optimal target source.

[0123] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A particle swarm search method based on robots without communication, characterized in that: In a two-dimensional or three-dimensional optimization search task, multiple robots without the ability to communicate with each other are dispersed in the search space. The group of robots is regarded as a particle swarm, and each robot is regarded as a particle. The S&W particles in the corresponding particle swarm algorithm are replaced with S&W robots, and the S&W particle set is replaced with the S&W robot set. The environment modeling is performed, and the field source signal value detected by the robot is used as the fitness value indicator of the particle swarm algorithm. The specific steps are as follows: Step 1.1: Initialize the particle swarm optimization algorithm parameter settings: number of particles N, dimension of the optimization problem D, inertia weight ω, constant acceleration factors c1 and c2, threshold K for continuous update failure of individual historical optimal position pbest, particle number threshold M of the S&W particle set, and maximum number of iterations T; Step 1.2: Randomly initialize the initial positions and velocities of N robots, that is, the position of robot i and speed where i∈{1,2,…,N}; Step 1.3: Randomly initialize the leaders of N robots. For robot i, its leader is marked as l i ; Step 1.4: The robot detects the position of other robots and calculates the center position of the group robot through the positions of all robots Where d∈{1,2,…,D} is the d-th dimension search space of the optimization problem; Step 1.5: Traverse all robots in the swarm. If the traversed robot i has selected a leader, execute steps 1.6 and 1.7, where i∈{1,2,…,N}; Step 1.6: Update the robot's velocity and position. The update formulas are as follows: Among them, r1 and r2 are D-dimensional random number vectors; represents the leader position selected by robot i; Step 1.7: Evaluate the robot's fitness based on its current position, update pbest, and determine whether the robot can become an S&W robot; if it can become an S&W robot, fall back to pbest and wait; Step 1.8: Traverse all S&W robots in the swarm, and execute step 1.9 for each S&W robot; Step 1.9: Check whether the number of robots in the S&W robot set reaches or exceeds the threshold M. If it reaches or exceeds the threshold M, randomly select a leader for the S&W robot; Step 1.10: If the current number of iterations is greater than the maximum number of iterations T or the optimal target source is successfully found, proceed to step 1.11; otherwise, return to step 1.4 for the next iteration. Step 1.11: Return the best pbest and its fitness value among all robots as the final search result.

2. The particle swarm search method based on inter-robot communication without communication as claimed in claim 1, characterized in that: In step 1.3, each robot in the swarm of robots {1, 2, ..., N} randomly selects another robot in the swarm as its leader: l i =randi({1,2,…,W}-{i}); Among them, randi(.) is a random selection function used to randomly select a robot from the group of robots, l i Is the leader of robot i.

3. The particle swarm search method based on inter-robot communication without communication as claimed in claim 1, characterized in that: In step 1.6, robot i simultaneously follows the leader position it chooses and the center position of the swarm robot to update its speed.

4. The particle swarm search method based on inter-robot communication without communication as claimed in claim 1, characterized in that: In the step 1.7, it is divided into the following sub-steps: Step 1.7.1: Each robot updates its pbest the number of times it fails to update continuously. For the minimum optimization problem, the update formula is: For the maximum optimization problem, the update formula is: Among them, f(.) is the fitness function of the optimization problem to be searched, k i is the number of consecutive failed updates of pbest of robot i; Step 1.7.2: Each robot determines the number of times k that its pbest consecutive updates have failed i Whether the threshold K is reached or exceeded. If it is reached or exceeded, the robot will fall back to its pbest position, clear its previously recorded leader, reset its speed to zero, and become an S&W robot: if k i ≥K X i =pbest;l i =0;V i =(0 1 ,0 2 ,…,0 D ); End.

5. The particle swarm search method based on inter-robot communication without communication as claimed in claim 1, characterized in that: In the step 1.9, it is divided into the following sub-steps: Step 1.9.1: Each S&W robot actively detects other S&W robots and adds all detected S&W robots to the S&W robot set: i∈P if V i =(0 1 ,0 2 ,…,0 D ) The zero velocity vector V i =(0 1 ,0 2 ,…,0 D ) represents the detected particle i as an S&W robot; P represents the set of detected S&W robots, and P includes all S&W robots; Step 1.9.2: When the number of robots in the S&W robot set P reaches or exceeds the threshold M, each S&W robot randomly selects another S&W robot as its leader: l i =randi(P-{i}) if|P|≥M and i∈P Where |P| represents the number of robots in the S&W robot set, and randi(.) is a random selection function used to randomly select another S&W robot.

Images