Multi-agent collaborative region segmentation method and system in dynamic environment

Abstract

The present invention relates to the field of multi-intelligent control, and provides a multi-agent collaborative area segmentation method and system in a dynamic environment, including: S1: obtaining a task area, dividing the task area horizontally, and obtaining multiple strips; S2: setting a segmentation strategy, obtaining the task amount and controller of each strip according to the segmentation strategy, and controlling multiple agents to vertically segment each strip in turn through the task amount and controller of each strip. The present invention has a small amount of calculation, and is more practical and efficient when implementing tasks such as cleaning that require traversing the task area; the distributed algorithm makes each agent an independent individual, the system is more robust, and the hardware requirements are smaller; the algorithm in a dynamic environment is theoretically analyzed and experimentally verified, and can be effectively applied to a dynamic environment where the workload density does not increase.

Description

Technical Field

[0001] The present invention relates to the field of multi-intelligent control, and in particular to a multi-agent collaborative region segmentation method and system in a dynamic environment. Background Art

[0002] With the rapid development of information technology, intelligent agents are present in all aspects of human life, helping humans to complete various tasks efficiently. Compared with a single intelligent agent, a multi-agent system supports distributed control, has good modularity and is easy to expand. When faced with specific complex tasks, a multi-agent system based on distributed control can often be gradually solved through coordination and cooperation. In order to improve the working efficiency of the multi-agent system, researchers proposed the Voronoi region partitioning algorithm, which aims to improve the coverage of the multi-agent system through reasonable partitioning, and at the same time improve the overall efficiency of the multi-agent system in completing tasks. The algorithm has been widely used in many fields such as geometry, crystallography, architecture, geography, meteorology and information systems. However, a major defect of the algorithm is that each node must be able to communicate with other nodes in order to move. If a node at the edge of the coverage area has only one neighbor node and the node fails, the network cannot move other nodes to compensate and rebuild the communication network. Therefore, the algorithm often requires the deployment of a large number of nodes, which greatly increases the cost.

[0003] In order to improve the applicability of the partitioning algorithm, the researchers proposed an equal-task region partitioning algorithm based on the widely used resource allocation strategy, that is, balancing the allocation of resources to workloads. In essence, this algorithm uses the idea of ​​distributed control. Compared with the Voronoi region partitioning algorithm, the equal-task region partitioning algorithm usually only needs to know the information of neighbors to complete the equal-task partitioning of the entire rectangular area, which makes it less demanding on the performance of a single intelligent agent, more in line with the requirements of practical application, reduces communication costs, and improves work efficiency.

[0004] The above new partitioning algorithm still has some shortcomings in existing practice:

[0005] Most existing equal-task segmentation algorithms assume that the division of the workspace is calculated centrally, which is in sharp contrast to the fully distributed architecture for multi-agent systems. At the same time, due to the lack of a fully distributed architecture, the applicability of equal-task region segmentation algorithms in limited-scale multi-agent systems running in known environments is limited; in the current research process, researchers often only consider the performance of algorithms in static environments, ignoring more complex dynamic environments, while the actual application environment is mainly dynamic environments. In addition, existing technologies still have problems such as high cost, poor applicability, and low efficiency.

[0006] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a multi-agent collaborative region segmentation method in a dynamic environment, comprising:

[0008] S1: Acquire a task area, and divide the task area horizontally to obtain multiple strips;

[0009] S2: Setting a segmentation strategy, obtaining the task amount and controller of each strip according to the segmentation strategy, and controlling multiple agents to vertically segment each strip in turn through the task amount and controller of each strip.

[0010] Preferably, step S1 specifically comprises:

[0011] S11: Select a length of I a , width is I b The rectangular area is taken as the task area D, and the lower left corner of the task area D is taken as the origin to construct a rectangular coordinate system;

[0012] S12: Divide the task area D into two parts with a width of d and a length of i a The number of stripes is q, I b =q×d.

[0013] Preferably, S2 is specifically:

[0014] S21: Record the strip as A k , k is the number of the strip; the initial value of k is 1 and the maximum value is q; the strip at the bottom of the task area D is marked as A 1 ;

[0015] S22: n agents will strip A 1 The strips are divided vertically into n sub-strips of random size. The sub-strip numbers correspond to the agent numbers one by one and are all represented by i. 2 Scan and get A 2 The workload and controller;

[0016] S23: Starting from k=2, n agents follow strip A k The task volume and controller will be stripe A k Split vertically into n sub-strips of equal size, and at the same time k+1 Scan and get A k+1 The amount of tasks and controller; let k = k + 1

[0017] S24: Repeat step S23 for q-1 times in total until all q strips are divided vertically.

[0018] Preferably, strip A k The steps to obtain the task volume are:

[0019] The task load distribution function in the static environment on the task area D is set to ρ(x, y). The task load distribution function is continuous and positive on D. Then, in strip A K The distribution of tasks along the x-axis is:

[0020]

[0021] Among them, τ represents the integral variable; y k = k × d, indicating the vertical boundary position of the kth strip; y k-1 =(k-1)×d;

[0022] In a static environment, in strip A K The task volume of the i-th sub-strip after vertical splitting is:

[0023]

[0024] in, Indicates that the i-th agent is in strip A k The workload on i-1 and x i represents the horizontal boundary position of agent i, and x i-1 is the common boundary position of agent i-1 and agent i, x i is the common boundary position of agent i and agent i+1;

[0025] In a dynamic environment, in stripe A K The task volume of the i-th sub-strip after vertical splitting is:

[0026]

[0027] Among them, ρ(x, y, t) represents the distribution function of the dynamic workload in area D, that is, the dynamic density function; t represents the running time of agent i.

[0028] Preferably, strip A k The controller acquisition steps are:

[0029] Strip A k The boundary of each sub-strip is denoted as i, and the controller of the agent split on the boundary i is denoted as u i ; The dynamic partition of the agent on the right boundary of sub-strip i is:

[0030]

[0031] in, represents the moving speed of the common boundary position of agent i and agent i+1;

[0032] Controller i The expression is:

[0033] u i =κ(m i+1 -m i )

[0034] Among them, κ represents the control gain, which is a positive number that can be set manually.

[0035] A multi-agent collaborative region segmentation system in a dynamic environment, comprising:

[0036] A stripe horizontal segmentation module is used to obtain a task area and horizontally segment the task area to obtain a plurality of stripes;

[0037] The strip vertical segmentation module is used to set the segmentation strategy, obtain the task amount and controller of each strip according to the segmentation strategy, and control multiple agents to vertically segment each strip in turn through the task amount and controller of each strip.

[0038] The present invention has the following beneficial effects:

[0039] 1. Compared with the traditional Voronoi segmentation algorithm, the present invention has a small amount of calculation, is more practical and more efficient when implementing tasks such as cleaning that require traversing the task area;

[0040] 2. The distributed algorithm makes each agent an independent individual, making the system more robust and requiring less hardware;

[0041] 3. The algorithm in dynamic environment is theoretically analyzed and experimentally verified, and can be effectively applied to dynamic environment where workload density does not increase. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0043] Figure 2 It is a schematic diagram of the segmentation of the task area;

[0044] Figure 3 is the segmentation of the first task volume distribution function;

[0045] Figure 4 is the segmentation of the second task volume distribution function;

[0046] Figure 5 is the task size of the first task distribution function;

[0047] Figure 6 is the task amount size of the second task amount distribution function;

[0048] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0050] Reference Figure 1 The present invention aims to improve the problems of high cost, poor applicability and low efficiency in traditional region segmentation methods. By improving the control architecture of the multi-agent system, a multi-agent collaborative region segmentation method in a dynamic environment is provided, including:

[0051] S1: Acquire a task area, and divide the task area horizontally to obtain multiple strips;

[0052] S2: Setting a segmentation strategy, obtaining the task amount and controller of each strip according to the segmentation strategy, and controlling multiple agents to vertically segment each strip in turn through the task amount and controller of each strip.

[0053] refer to Figure 2 In this embodiment, step S1 is specifically as follows:

[0054] S11: Select a length of I a , width is I b The rectangular area is taken as the task area D, and the lower left corner of the task area D is taken as the origin to construct a rectangular coordinate system;

[0055] S12: Divide the task area D into two parts with a width of d and a length of i a The number of stripes is q, I b =q×d.

[0056] In this embodiment, S2 is specifically:

[0057] S21: Record the strip as A k , k is the number of the strip; the initial value of k is 1 and the maximum value is q; the strip at the bottom of the task area D is marked as A 1 ;

[0058] S22: n agents will strip A 1 The strips are divided vertically into n sub-strips of random size. The sub-strip numbers correspond to the agent numbers one by one and are all represented by i.2 Scan and get A 2 The workload and controller;

[0059] S23: Starting from k=2, n agents follow strip A k The task volume and controller will be stripe A k Split vertically into n sub-strips of equal size, and at the same time k+1 Scan and get A k+1 The amount of tasks and controller; let k = k + 1

[0060] S24: Repeat step S23 for q-1 times in total until all q strips are divided vertically.

[0061] Specifically, n agents are arranged in the lowest strip of the horizontally divided rectangular area. The working area on the first strip (k=1) will be randomly divided vertically. At the same time, the working area on the second strip (k=2) will be scanned with the same amount of tasks until the first strip A 1 After the working area on the A is randomly divided vertically, all agents move to the second strip A. 2 Inside, to A 2 The strip is divided vertically in equal amounts, and A 3 The strip scans the same amount of tasks; and so on, when the agent completes A K When the area within the strip is divided vertically in equal amounts, the agent is also dividing A k+1 The working area on the strip is scanned with equal task amount until the vertical segmentation of the entire rectangular area is completed; only the area on the first strip is randomly segmented vertically, and the remaining strips are all segmented vertically with equal amount.

[0062] In this embodiment, reference Figure 2 , mainly analyzing the division of tasks on the Kth strip into equal amounts; first, a rectangular coordinate system is made with the lower left corner of the rectangular area as the origin, and the agents arranged on the first strip are numbered from left to right as 1, 2, 3, ···, i, ···, n-1, n, and x i represents the final position of the i-th agent on the x-axis (and the right boundary position of sub-strip i); for each agent, the starting position on the y-axis is (K-1)×d, the ending position is K×d, and the starting position on the x-axis is x i-1 , the end position is x i ; Obviously, the starting position of the first agent on the x-axis is x 1 , the final position x of the nth agent on the x-axis n For I b; Assuming that the density function ρ(x, y, t) on region D is known, we can infer the i-th agent in strip A through calculus knowledge K The task size m i and in strip A K The total amount of tasks on M K ;

[0063] In order to achieve “equal task division”, all m i should be close to M K / n, the algorithm changes the task amount of two adjacent agents by splitting the boundary; for the right boundary x of the i-th sub-strip i ,Due to the uncertainty, it is impossible to divide the work area in advance, and online partitioning operations must be performed;

[0064] Band A k The steps to obtain the task volume are:

[0065] The task load distribution function in the static environment on the task area D is set to ρ(x, y). The task load distribution function is continuous and positive on D. Then, in strip A K The distribution of tasks along the x-axis is:

[0066]

[0067] Among them, τ represents the integral variable; y k = k × d, indicating the vertical boundary position of the kth strip; y k-1 =(k-1)×d;

[0068] In a static environment, in strip A K The task volume of the i-th sub-strip after vertical splitting is:

[0069]

[0070] Among them, m i Indicates that the i-th agent is in strip A k The workload on i-1 and x i represents the horizontal boundary position of agent i, and x i-1 is the common boundary position of agent i-1 and agent i, x i is the common boundary position of agent i and agent i+1;

[0071] Specifically, the relationship between the amount of tasks that the agent is responsible for and the working range of the agent in a static environment is considered. Based on the task distribution function, the task amount is obtained by integrating the x and y positions respectively; based on this, the task amount in a dynamic environment can be obtained in general as follows:

[0072] In a dynamic environment, in stripe A K The task volume of the i-th sub-strip after vertical splitting is:

[0073]

[0074] Among them, ρ(x, y, t) represents the distribution function of the dynamic workload in area D, that is, the dynamic density function; t represents the running time of agent i.

[0075] In this embodiment, the equal task area segmentation algorithm considers how to update the boundary to balance the workload of each agent on the sub-strip, and then complete the segmentation of each stripe to improve the working efficiency of the entire multi-agent system;

[0076] Band A k The controller acquisition steps are:

[0077] Strip A k The boundary of each sub-strip is denoted as i, and the controller of the agent split on the boundary i is denoted as u i ;

[0078] The dynamic partition of the agent on the right boundary of sub-strip i is:

[0079]

[0080] in, represents the moving speed of the common boundary position of agent i and agent i+1;

[0081] Controller i The expression is:

[0082] u i =κ(m i+1 -m i )

[0083] Among them, κ represents the control gain, which is a positive number that can be set manually.

[0084] Convergence proof of the controller's control algorithm:

[0085] The focus is on analyzing the convergence of the equal-task region segmentation algorithm after adding the controller; at the same time, in order to meet the requirements of practical applications, the number of iterations of the control algorithm (how long it takes) should be limited; first, consider the partitions on each strip. Secondly, design a suitable Lyapunov function. According to the design idea of ​​the equal-task region segmentation algorithm, it is relatively easy to obtain a suitable Lyapunov function, and then complete the theoretical proof through Lyapunov's second method, that is, for a scalar function with continuous first-order partial derivatives, if H k Positive definite, we can derive H k The derivative of is negative definite, then the control algorithm is convergent, then this H k It is a Lyapunov function of the control system; the following will complete the convergence analysis of the control algorithm step by step:

[0086]

[0087] In Stripe A K Find the average workload:

[0088]

[0089] In Stripe A K Select the Lyapunov function above:

[0090]

[0091] The task load distribution function in a dynamic environment is constantly changing with time. In order to meet the stability requirements of the algorithm, it is usually necessary to k (x, t) with certain restrictions; considering the task volume distribution function ω k (x, t) gradually decays with time;

[0092] Therefore, let's assume:

[0093]

[0094] Where α is a constant and α<0, α represents the rate of change of the task load distribution function with respect to time;

[0095] According to the square mean ≥ arithmetic mean, we can get:

[0096]

[0097] In summary, H k The derivative along the system trajectory satisfies:

[0098]

[0099] Obviously, since κ>0, ω k (xi ,t)>0 and α<0, so in this strip A K In addition to the balance point, This shows that the equilibrium point is asymptotically stable, thus concluding that the algorithm is stable.

[0100] Numerical simulation:

[0101] According to the design idea of ​​the equal-task region segmentation algorithm in a dynamic environment, a program is written on MATLAB to perform simulation verification; assuming that the number of initial agents is 6 and the initial boundary position x 0_0 =0, x 1_0 =1, x 2_0 =3, x 3_0 =5, x 4_0 =7, x 5_0 =12, x 6_0 =18, where x 0_0 、x 6_0 is the left and right boundaries of the strip, which is a fixed value and does not change; at the same time, according to the known task volume distribution function and the boundary position of the task area, by establishing a good mathematical model, the initial task volume m of each agent in charge of the task area can be obtained 1_0 、m 2_0 、m 3_0 、m 4_0 、m 5_0 、m 6_0 ;

[0102] To facilitate computer program processing:

[0103] x i (k+1)=x i (k)+Δtκ(m i+1 -m i )

[0104] where x i (k) represents the time x at time k i The size of x i (k+1) means that at time k+1 x i The size of , Δt represents the difference between two moments, and it is obvious that Δt is always greater than zero;

[0105] Take Δt = 0.0001, κ = 10; get the following simulation results for different task volume distributions. At the same time, analyze the algorithm performance from the workload partitioning situation;

[0106] Take the first task volume distribution function ρ(x,y,t)=(x+0.5)(1+exp(-0.05t-2))

[0107] The workload is divided into Figure 3 As shown in the figure, under the premise of Δt = 0.0001, the fitting is basically completed after about 0.2s. It can be seen that the task volume is close to 30.33, which meets the expected goal of the algorithm, that is, the distribution of the task volume on each participating agent is consistent;

[0108] Take the second task volume distribution function ρ(x,y,t)=(1.5+0.5sin(x)+x)(1+exp(-0.05t-2))

[0109] The workload is divided into Figure 4 As shown in the figure, under the premise of Δt=0.0001, the fitting is basically completed after about 0.28s. It can be seen that the task volume is close to 31.45, which meets the expected goal of the algorithm, that is, the distribution of the task volume on each participating intelligent agent is consistent.

[0110] Control algorithm application:

[0111] The design and numerical simulation of the equal-task region segmentation algorithm have been completed. On the basis of simulation, in order to further approach the actual use of the equal-task region segmentation algorithm in multi-agent systems, it is necessary to further apply the control algorithm to the wireless communication network based on ZigBee technology. Under this experimental platform, unlike pure MATLAB numerical simulation, the actual algorithm application often needs to take into account various interferences. For multi-agent systems, the real-time and anti-interference performance of communication will also affect the performance of the algorithm. On the wireless communication network based on ZigBee technology, the real-time and anti-interference performance of communication can be greatly guaranteed to support the application of the designed equal-task region segmentation algorithm.

[0112] For the entire experimental link, the ultimate goal is to verify whether the algorithm can successfully complete the task of equal task division on the experimental platform based on ZigBee technology, that is, in a multi-agent system, whether the amount of tasks each agent is responsible for can eventually converge within a limited time range; in order to achieve this goal, this experiment uses a variety of task volume distribution functions in different environments to test the effectiveness and reliability of the algorithm. In order to better conduct comparative analysis, the selected task volume distribution function refers to the situation in the simulation process;

[0113] In the experiment, the communication data between ZigBee nodes is collected on the computer through the serial port, and a program for serial port collection and real-time plotting display is written on MATLAB to complete the data processing; the communication frequency is set to 0.05s / time, and the LED light is set to alternately light up and off to mark the data sending and receiving;

[0114] Take the first task volume distribution function ρ(x,y,t)=(x+0.5)(1+exp(-0.05t-2))

[0115] like Figure 5 As shown in the figure, compared with the simulation results, the change pattern of the task volume is basically the same. However, in the process of the task volume of each agent tending to be consistent, it can be clearly seen that compared with the simulation results, the update of the task volume of each agent has experienced many twists and turns;

[0116] Take the second task volume distribution function ρ(x,y,t)=(x+0.5)(1+exp(-0.05t-2))

[0117] like Figure 6 As shown in the figure, unlike the previous situation, the changes in the agent's task volume have experienced larger fluctuations. In the period of 0-800 communications, the agent's task volume m1, m2, m3, m4, m5, and m6 have experienced intermittent large-scale fluctuations; in the range of 50-200 and 200-350 times, the agent's task volume m6 even remained unchanged for a time.

[0118] In summary, the application of this control algorithm in the wireless communication network based on ZigBee technology has been completed, which further verifies the effectiveness and reliability of the algorithm.

[0119] The present invention provides a multi-agent collaborative region segmentation system in a dynamic environment, comprising:

[0120] A stripe horizontal segmentation module is used to obtain a task area and horizontally segment the task area to obtain a plurality of stripes;

[0121] The strip vertical segmentation module is used to set the segmentation strategy, obtain the task amount and controller of each strip according to the segmentation strategy, and control multiple agents to vertically segment each strip in turn through the task amount and controller of each strip.

[0122] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0123] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0124] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A multi-agent collaborative region segmentation method in a dynamic environment, characterized in that: include: S1: Acquire a task area, and divide the task area horizontally to obtain multiple strips; S2: setting a segmentation strategy, obtaining the task amount and controller of each strip according to the segmentation strategy, and controlling multiple agents to vertically segment each strip in turn through the task amount and controller of each strip; S2 is specifically: S21: Record the strip as A k , k is the number of the strip; the initial value of k is 1, and the maximum value is q; the strip at the bottom of the task area D is recorded as A1; S22: n agents vertically divide strip A1 into n sub-strips of random size. The sub-strip numbers correspond to the agent numbers one by one and are both represented by i. At the same time, strip A2 is scanned to obtain the task volume and controller of A2. S23: Starting from k=2, n agents follow strip A k The task volume and controller will be stripe A k Split vertically into n sub-strips of equal size, and at the same time k+1 Scan and get A k+1 The amount of tasks and controller; let k = k + 1 S24: repeat step S23 for q-1 times in total until all q strips are vertically divided; Band A k The steps to obtain the task volume are: The task load distribution function in the static environment on the task area D is set to ρ(x, y). The task load distribution function is continuous and positive on D. Then, in strip A K The distribution of tasks along the upper x-axis is: Among them, τ represents the integral variable; y k = k × d, indicating the vertical boundary position of the kth strip; y k-1 =(k-1)×d; In a static environment, in strip A K The task volume of the i-th sub-strip after vertical splitting is: in, Indicates that the i-th agent is in strip A k The workload on i-1 and x i represents the horizontal boundary position of agent i, and x i-1 is the common boundary position of agent i-1 and agent i, x i is the common boundary position of agent i and agent i+1; In a dynamic environment, in stripe A K The task volume of the i-th sub-strip after vertical splitting is: Where ρ(x, y, t) represents the distribution function of the dynamic workload in region D, i.e., the dynamic density function; t represents the running time of agent i; Band A k The controller acquisition steps are: Strip A k The boundary of each sub-strip is denoted as i, and the controller of the agent split on the boundary i is denoted as u i ; The dynamic partition of the agent on the right boundary of sub-strip i is: in, represents the moving speed of the common boundary position of agent i and agent i+1; Controller i The expression is: u i =κ(m i+1 -m i ) Among them, κ represents the control gain, which is a positive number set artificially, and m i represents the workload of agent i, m i+1 Represents the workload of agent i+1.

2. The multi-agent collaborative region segmentation method in a dynamic environment according to claim 1, characterized in that: Step S1 is specifically as follows: S11: Select a length of I a , width is I b The rectangular area is taken as the task area D, and the lower left corner of the task area D is taken as the origin to construct a rectangular coordinate system; S12: Divide the task area D into two parts with a width of d and a length of i a The number of stripes is q, I b =q×d.

3. A multi-agent collaborative region segmentation system in a dynamic environment, used to implement the multi-agent collaborative region segmentation method in a dynamic environment as described in any one of claims 1-2, characterized in that: include: A stripe horizontal segmentation module is used to obtain a task area and horizontally segment the task area to obtain a plurality of stripes; The strip vertical segmentation module is used to set the segmentation strategy, obtain the task amount and controller of each strip according to the segmentation strategy, and control multiple agents to vertically segment each strip in turn through the task amount and controller of each strip.

Images