Multi-agent system consistency algorithm based on dynamic autonomous reinforcement weak connection

Abstract

The invention provides a multi-agent system consistency algorithm based on dynamic autonomous reinforcement weak connection. The multi-agent system consistency algorithm comprises the steps of neighbor perception, spokesman selection strategy, connection strength index calculation, weight calculation, consistency algorithm updating and the like. Firstly, redundant neighbors are effectively removed through a spokesman selection strategy, and the calculation process is simplified. Secondly, a novel index is introduced to evaluate the connection strength in the distributed switching topology, a reasonable weight matrix is designed, intelligent agents at the two ends of the weak connection are promoted to gather together, and therefore connection breakage is prevented. On the whole, the system can still be prevented from being split into a plurality of clusters under the condition of no connectivity maintenance algorithm, and finally consistency is achieved. The algorithm effectively solves the problem that in the prior art, connectivity needs to be maintained with high computing power or the system is split due to lack of connectivity maintenance.

Description

Technical Field

[0001] The present invention belongs to the field of multi-agent systems, and specifically relates to a multi-agent system consistency algorithm based on dynamic autonomous reinforcement of weak connections under discrete time switching topology. Technical Background

[0002] With the rapid development of artificial intelligence technology, multi-agent systems, as an important research direction of distributed artificial intelligence, have gradually attracted widespread attention. The design inspiration of these systems comes from the self-organizing characteristics of group behavior in nature, such as the foraging behavior of ants, the migration of geese, and the collective swimming of fish. These natural phenomena demonstrate the coordination, stability, and adaptability of groups in dynamic environments, providing rich references for the research of multi-agent systems.

[0003] In multi-agent systems, the consistency problem is the basis for achieving collaborative control, collective decision-making and task allocation. The so-called consistency refers to the consensus of the agents in the system on the status of position, speed, etc. The network topology has an important impact on the consistency problem, especially in the consistency research under the switching topology, its complexity and challenge are more prominent, which has attracted widespread attention in the academic community.

[0004] Although some algorithms have been proposed to enhance the consistency of multi-agent systems with multiple switching topologies, they can be divided into two categories: one is based on connectivity maintenance algorithms. Although they can ensure that the entire system converges to one cluster, connectivity maintenance algorithms consume a lot of computing power, which greatly limits the application scope of multi-agent systems; the other is traditional consistency algorithms without connectivity maintenance, which will cause the weak connections of some agents to break, and finally split the system into multiple different clusters. Therefore, it is very necessary to develop a new consistency algorithm for multi-agent systems so that the system can reach consistency without the need for connectivity maintenance algorithms. Summary of the invention

[0005] The present invention provides a multi-agent system consistency algorithm that dynamically and autonomously strengthens weak connections. It can achieve system consistency without the need for a connectivity maintenance algorithm, reduces agent performance requirements, and increases the application scope of the multi-agent system. The specific implementation steps are as follows: a) Neighbor Awareness: In a distributed multi-agent system with discrete-time switching topology, the agent v i At time t, it perceives its neighboring agent set N i (t). b) Spokesperson selection: Agent v i According to the spokesperson selection strategy, a set of spokespersons with different states are selected from the neighbors. c) Connection strength index calculation: Agent v i Calculate its and set Any neighbor v j ALC ij The value of (t). d) Weight calculation: Calculate the agent v i and Middle Neighbor v j The connection weight W ij (t). e) Consensus algorithm update: Agent v i according to The state and weight of the neighbors in the network are used to update the state of the network itself through the multi-agent system consensus algorithm. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 This is a flow chart of the consensus algorithm for a multi-agent system based on dynamic autonomous reinforcement of weak connections. Figure 2 is the agent selection strategy process, the dotted line is the abandoned neighbor, the solid line is the selected neighbor, and the white line is the agent v i , black is the agent v i Neighbors. Figure 3 The intelligent agent refers to the state of its neighbors through new weights, identifies connections that are easily broken and assigns them larger weights to prevent the system from splitting. Figure 4 This is the experimental effect diagram. Figure 4 (a) is the initial topology, Figure 4 (b) The traditional multi-agent consensus algorithm without connectivity maintenance algorithm causes the system to split into many clusters; Figure 4 (c) It is the present invention that enables the system to achieve consistency without connectivity maintenance. DETAILED DESCRIPTION

[0006] To facilitate understanding of the technical solution of the present invention, specific implementations of each step are described in detail below with reference to the accompanying drawings.

[0007] In multi-agent systems, graph theory and matrix theory are usually used to analyze consistency problems. In the present invention, an undirected graph G = (V, E, A) is used to represent the topological structure, where: Node set V = {v1,v2,…,v n} represents the set of agents; ·Edge Set Represents the connection relationship between agents at time t; · Adjacency matrix A(t) = (a ij (t)) n×nIt is used to represent the connection status between the agents of the entire system at time t, and

[0008] The communication radius of each agent in a multi-agent system is limited. Let r be the communication radius of each agent. i Agents within the communication range are called their neighbors. The details of each step are as follows:

[0009] Step 1: Neighbor perception: The agent exchanges information with neighbors within the communication range to obtain the status information of the neighbors.

[0010] Step 2: Spokesperson selection: The agent traverses the neighbors and selects the neighbors with different states to obtain the spokesperson set. This process eliminates homogeneous neighbors and improves the accuracy and efficiency of subsequent calculations.

[0011] Step 3: Calculate the connection strength index. The ALC(t) index is used to measure the The connection strength index of all neighbors in is calculated according to the following steps:

[0012] Step 3.1: Calculate the agent v i and The maximum number of edges that can exist in a community of all neighbors in : Among them, |…| means calculating the number of elements in the set, and the following expressions are similar.

[0013] Step 3.2: Calculate a value that can roughly determine the agent v i and The strength of any neighbor connection in C τ (t) coefficient: in Represents v i and The subscript il indicates that the calculated coefficient is about e il (t) This connection, Indicates that both agents are The set of all connections on a set, Indicates the number of elements in a set. Represents the agent v at time t i and neighbors l The connection between C τ The coefficients are expressed similarly below.

[0014] Step 3.3: Calculate an evaluation The coefficient of the closeness of the community composed of all agents in in Represents the agent v at time t i C λ coefficient.

[0015] Step 3.4: Calculate the connection strength index ALC(t): ALC ij (t)Indicator and Multiply the coefficients to get The tighter the community formed by all agents in C λ The larger the community is, the less likely it is for the community to split. ij (t) is larger, indicating that it is closer to the neighbor v j The stronger the connection.

[0016] Step 4: Weight calculation. The weight calculation steps are as follows:

[0017] Step 4.1: Calculate ALC ij The negation of (t): In the consensus algorithm, we prevent this kind of neighbor disconnection and the system from splitting into more clusters, so we use ALC ij (t) The indicator is negated.

[0018] Step 4.2: Calculate weights: θ is a trade-off factor that adjusts The larger the θ, The stronger the indicator's ability to regulate weak connections, when θ = 1, the system will only converge to one cluster, and the convergence effect is the best.

[0018] Step 5: Update the consensus algorithm to incorporate the new weight W ij (t), the update consistency algorithm is: Where α is the adjustment factor, which is a number greater than 0 and less than 1.

[0019] Compared with existing consistency algorithms, the present invention has the following advantages: compared with existing consistency algorithms that require connectivity maintenance algorithms, the present invention does not require connectivity maintenance algorithms and can prevent system splitting, and consumes less computing power; compared with traditional consistency algorithms that do not require connectivity maintenance algorithms, traditional algorithms will cause the system to split into multiple clusters and cannot achieve consistency, while the present invention can enable the system to achieve consistency.

Claims

1. A consensus algorithm for multi-agent systems based on dynamic autonomous reinforcement of weak connections, characterized by The steps are as follows: Step S1: In a distributed multi-agent system with discrete-time switching topology, any agent v in the system i Scan its neighboring agent set N at time t i (t); Step S2: Agent v j Filter out a set of agents with different states from their neighbors Step S3: Agent v i Calculate its and set Any neighbor v j ALC ij (t) the value of the indicator; Step S4: Calculate the agent v i and Middle Neighbor v j The connection weight W ij (t); Step S5: Agent v i according to The state and weight of the neighbors in the network are used to update the state of the network itself through the multi-agent system consensus algorithm.

2. According to claim 1, the consensus algorithm of multi-agent system based on dynamic autonomous reinforcement of weak connections is characterized in that: The initial topology of the multi-agent system is connected, and the basis for judging whether the system has reached consistency is that all the agents in the system have reached the same state.

3. The multi-agent system consensus algorithm based on dynamic autonomous reinforcement of weak connections according to claim 1 is characterized in that: ALC ij The calculation formula of the indicator (t) is: In the above formula, e jk (t) represents the agent v at time t j and v k The connection between Indicates that both agents are The set of all connections on a set, Indicates the number of elements in a set; Represents the agent v at time t i C λ coefficient; Represents the agent v at time t i and neighbors v l The connection between C τ coefficient.

4. The multi-agent system consensus algorithm based on dynamic autonomous reinforcement of weak connections according to claim 1 is characterized in that: Weight W ij The calculation formula of (t) is: Among them A ij (t) represents the i-th row and j-th column in the adjacency matrix. i and v j When there is a connection between ij (t) is 1, otherwise it is 0; θ is a constant, ranging from 0 to 1.

5. The multi-agent system consensus algorithm based on dynamic autonomous reinforcement of weak connections according to claim 1 is characterized in that: The consensus algorithm is as follows: Among them, α is the adjustment factor, which is a number greater than 0 and less than 1.

6. The multi-agent system consensus algorithm based on dynamic autonomous reinforcement of weak connections according to claim 1 is characterized in that: Agent v i The update process only refers to some neighbors Information from all neighbors N is not referenced i (t).

Images