Multi-agent system modeling and model detection method based on cognitive process calculation

Abstract

The invention discloses a multi-agent system modeling and verification framework based on cognitive process calculation, and aims to solve the problems of inaccuracy and incompleteness of multi-agent system modeling and verification in the prior art. According to the method, firstly, a multi-agent system is modeled by adopting value transmission process calculation, and a cognitive concept is introduced, so that a novel cognitive process calculation (EPC) is constructed for formally describing agent behavior interaction and cognitive states; on the basis, alternating time sequential logic (ATL) is expanded, and novel cognitive alternating sequential logic (ATLE) is formed and used for expressing complex attributes related to the cognitive state of the intelligent agent. The invention further provides a model checking algorithm aiming at the ATLE specification, the state space of the multi-agent system can be systematically explored, and whether the specified attribute is met or not is evaluated. According to the invention, the feasibility and practicability of the proposed framework are verified through case research of cooperative attack of the unmanned aerial vehicle. According to the method, a more accurate and comprehensive tool can be provided for modeling and verification of the multi-agent system, a reference is provided for design and optimization of the multi-agent system, and a quantifiable basis is provided for task scheduling, resource allocation and strategy selection.

Description

Technical Field

[0001] The present invention discloses a multi-agent system modeling and model detection method based on cognitive process calculation, which is mainly used to capture the complex cognitive and communication interactions between agents, and to verify the security and reliability of the system by reasoning the beliefs, knowledge and uncertainty of the agents through cognitive logic. The present invention is a novel modeling and model detection method for multi-agent systems, which provides precise mathematical language to describe the interactions between agents, and uses model detection technology to verify various properties of the system. Background Art

[0002] In recent years, with the explosive development of artificial intelligence, more and more applications have emerged, and multi-agent systems are part of these applications. More and more industrial products are beginning to consider adopting multi-agent architectures, such as multi-UAV systems and multi-robot systems.

[0003] Logic-based technology is a traditional and ancient technology for understanding human thinking. It is close to the human cognitive process. Using logic-based technology to model, reason, and verify multi-agents is undoubtedly the best way to ensure the safety and reliability of cognitive multi-agent systems. In addition, logic-based methods have always been the core of many successful agent-based models and technologies. Generally speaking, cognitive agents in cognitive multi-agent systems directly use logic-based models and technologies to perform rational processes, knowledge representation, expressive communication, and effective coordination.

[0004] Process calculus is a mathematical tool based on algebraic theory that is used to describe concurrent systems. It brings the following advantages to the modeling of cognitive multi-agent systems: it provides a high degree of abstraction, formally represents individual interactions, and ignores non-critical details; it allows precise definition of system behavior to ensure compliance with design specifications; it is naturally suitable for describing concurrent activities and is a powerful multi-agent system analysis tool; it supports formal verification and improves system reliability; its modular design makes it easy to understand, modify, and expand; it helps analyze the dynamic behavior of the system and optimize system performance; it is particularly suitable for capturing the interactions between nodes in a distributed system and understanding system complexity.

[0005] Model checking technology has significant advantages in verifying the properties of multi-agent systems, mainly in automated verification, handling complex scenarios, verifying multiple logical formulas, and optimizing state explosion problems. This technology can automatically check whether the system meets specific properties and is suitable for non-deterministic complex systems. It can not only verify temporal formulas, but also cognitive formulas, making the verification more comprehensive. These advantages make model checking an important tool for multi-agent system verification.

[0006] In general, modeling multi-agent systems through process calculus and then verifying the system's properties through model checking techniques can help improve the reliability, maintainability, and scalability of system design and support in-depth analysis of system behavior. Summary of the invention

[0007] Formal methods have been increasingly widely used in safety-critical software engineering. A system should be specified using formal models, such as automata, Petri nets, and process algebras, in order to be formally verified. This paper proposes a calculus for describing the cognitive and interactive behaviors of multi-agent systems: cognitive calculus (EPC) and a new temporal logic for EPC: cognitive alternating temporal logic ATLE. Finally, a new model checking algorithm is proposed for the semantic description of ATLE.

[0008] The present invention models and checks the cognition and interaction of a multi-agent system based on EPC and ATLE, and mainly includes the following steps:

[0009] Step 1: Model the multi-agent system using the newly proposed cognitive process calculus (EPC).

[0010] Define the types of agents and their basic actions; First, it is necessary to clarify what types of agents exist in the system, such as drones, robots, sensors, etc. Each type of agent may have different functions and capabilities.

[0011] Using the grammatical rules of process algebra, we construct process expressions that describe the behavior of agents based on their behaviors and interactions. Process expressions can describe the behavior of a single agent or the interactions between multiple agents. Based on the structure of process expressions, processes can be divided into different types, such as basic processes, compound processes, parallel processes, etc.

[0012] Combine the process expressions of all agents to form a model of a multi-agent system. The model can contain multiple agents, as well as the interactions and communication relationships between them. The system model provides a holistic view of the multi-agent system and can clearly show the behavior and interaction of all agents in the system.

[0013] Step 2: Based on EPC, a new alternating temporal logic of cognition (ATLE) is proposed, and ATLE is used to define the system-related properties of the multi-agent system.

[0014] Define ATLE logic formulas that describe system properties, such as safety, liveness, fairness, etc. This step mainly uses temporal logic operators and policy agent collection operators to build complex ATLE property formulas. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Flowchart of the modeling and model checking method for multi-agent systems based on cognitive process calculus.

[0016] Figure 2 Schematic diagram of multiple drones' coordinated combat.

[0017] Figure 3 This is a state transition diagram of multiple UAV collaborative combat. DETAILED DESCRIPTION

[0018] As attached Figure 1 The specific implementation method of the invention is as follows:

[0019] Step 1: Model the multi-agent system using the newly proposed cognitive process calculus (EPC).

[0020] Given a multi-agent system, there are N multi-agents. A process represents one or more executions of a single agent, and the smallest unit of process execution is an action, which is an atomic operation that represents the basic capabilities of a process. A name is a basic concept used to identify a communication channel or a subject of communication. In the grammar, names have special meanings; they can represent a channel, a process, or another communication entity. Define the set is a data set, is a cognitive set, which is the basic unit of process sending and receiving. Action represents the role and behavior of the agent in the communication event, and its formal definition is as follows:

[0021]

[0022] Where x is data or knowledge. and x is the ability of the process to send and receive data or cognition. τ is a process internal action, referred to as internal action.

[0023] Using the grammatical rules of process algebra, according to the behavior and interaction mode of the agent, a process expression describing the behavior of the agent is constructed. The specific process P is defined by the BNF paradigm as follows:

[0024]

[0025] In the above content, 0 indicates a process without an action prefix, that is, a "dead" process that does not perform any operation; aP indicates that process P is guarded by an action prefix, and P can only be executed after the action is completed; (vz)P indicates a restriction on the name z, that is, z can only be used inside process P; (PIP) indicates that two independent processes are executed concurrently; P+Q indicates the uncertain selection execution of two processes; Where P is a procedure expression, which can be defined recursively.

[0026] There are two constraint constructions in cognitive calculus: restricted names (vx)P and input x(y).P denotes the constraint of x and y in P, respectively. The set of constrained names and the set of free names of a process P are denoted by and respectively. Processes that differ only in the names of their constraints are called and are denoted by . This paper does not distinguish between processes of. Substitution is a partial mapping from the set of agents N to the set of agents N, denoted by σ. Substitution is a postfix operator with higher precedence than any other operator in the language.

[0027] Define a marking process and combine the process expressions of all agents to form a marking process for a multi-agent system.

[0028] The representation of a multi-agent system consists of a global environment and several agents. The behavior of each agent is controlled by a marking process P. The marking process M is formally defined as follows:

[0029] M:=({P} Ag )|(M|M)|(vz)M

[0030] where z∈Name. {P} Ag represents the marked process composed of the agent Ag, and the behavior of Ag is controlled by the process P. Since the process can be compound, P here can also be multiple processes; M|M represents the compound on the marked process, indicating that the system is composed of multiple sub-marked processes; (vz)M means that z is a constraint name, and its scope is M;

[0031] Structural congruence ≡ is the minimum equivalence relation and satisfies the following formula:

[0032] P|Q≡Q|PP|0≡P

[0033] P1|(P2|P3)≡(P1|P2)|P3

[0034] P1+(P2+P3)≡(P1+P2)+P3P+0≡P

[0035] (va)(vb)P≡(vb)(va)P

[0036]

[0037] A process specifies the actions to be performed, while a state encapsulates the current cognitive condition and all other relevant data. The concept of state is broader than that of process. In cognitive process calculus modeling, cognitive state and process are independent. This means that the state may change even if the process is consistent, or the state may remain the same even if the process is different, and they are related to each other through mapping functions.

[0038] The operational semantics of EPC is based on the calculus of value operations. The SOS rules of EPC are listed in the table, which describe the behavior of the process. They are as follows:

[0039] Table 1. SOS rules for processes

[0040]

[0041] Table 2. SOS rules for marking processes

[0042]

[0043] This patent uses a set of symbols and functions to represent specific professional semantics. Ag is the set of multi-agents in a multi-agent system; S = S1×S2....×S n is a non-empty set of global states; S i represents a non-empty set of states of agent i; MS is a non-empty set of marked processes. SMS is a non-empty set of tuples of states and marked processes. Acts is a non-empty set of actions. Ag Represents the set of actions that the Ag set can perform. AP is a non-empty set of atomic propositions. T = {S × AP → Bool} is an abstract function that maps propositions to the global state S. It is a transition relationship, which indicates the state evolution of the system.

[0044] Multi-agent systems are built on cognitive states, and their evolution depends on changes in these cognitive states. The conclusions generated by structured operational semantics (SOS) rules have the following form:

[0045]

[0046] This means that under the marking process M, state s evolves into state s′ by executing action ACT, and the marking process changes from M to M′. Table 1 involves the structured operational semantics rules of the process, while Table 2 involves the structured operational semantics rules of the marking process. In addition, the cognitive state of each agent can be represented by an abstract function, which is defined as follows:

[0047] h i (s)=es i

[0048] Function H is responsible for abstracting the original proposition from the global state, or comprehensively abstracting the cognitive state related to the agent from the entire state context. T = {S × AP-→BOOL} represents an abstract function that acts as a mapping mechanism from the Cartesian product of the global state S and the set of atomic propositions AP to Boolean values. This function is designed to evaluate the truth value of a proposition under a given global state.

[0049] In multi-agent systems, strategy is a key concept. Consider a marking process M. First, we define Ag = {1, ..., k} as a set of agents. For a subset A∈2 ∑ , A partial strategy It maps the next action that the agents in set A can perform asynchronously. When an agent not in set A performs an action, the action is arbitrary; therefore, in some states of executing part of the strategy, it is the agents not in set A that perform certain actions. Maps the execution strategy in all states Acts A ={α i , τ ij |i,j∈A,α∈{x <y>, and The formal definition of is as follows:

[0050]

[0051]

[0052] If there is a marked process M, then OUT(FA, s) represents all state paths formed by iteratively executing the strategies in FA starting from (s, M). A path starts from (s0, M0): ∏ = (s0, M0)(S1, M1)....(s K， M κ ), ∏∈OUT(F A , S0), this sequence complies with the following conditions: F A (s1) = ACT1 and Represents the first element of the sequence, ∏[I]=(s1,M1).

[0053] The present invention adopts a suitable data structure to represent the characteristics and operation rules of each part of the EPC, and realizes efficient modeling of the EPC. The specific structure is as follows:

[0054]

[0055]

[0056] Step 2: Based on EPC, a new alternating temporal logic of cognition (ATLE) is proposed, and ATLE is used to define the system-related properties of the multi-agent system.

[0057] In the atomic proposition set AP, let the number of agent sets be represented by Ag = {1, ..., n}, and let the strategic agent set A belong to Ag. The ATLE formula φ indicates that a certain property holds under A, and its specific definition is as follows:

[0058]

[0059] Where ap∈AP, φ, φ1, φ2 are ATLE formulas. In addition, the path formula Fφ=true Uφ, Where K i φ、E G φ、C G φ represents the knowledge state of agent a about formula φ, all agents in the system G know the knowledge state of φ, and all agents in G know and all agents in G know that other agents in G also know the knowledge state of φ.

[0060] Given a labeled process M, a global state s, and an ATLE formula φ, the satisfiability of φ is denoted as (s, M) |= φ, and the satisfiability property satisfies the following definitions:

[0061] · if and only if T(s, ap) = true;

[0062] · if and only if

[0063] · if and only if

[0064] · if and only if there exists a and strategy for any for any path π ∈ out(f A , s), there is

[0065] · if and only if there exists a and strategy for any for any path π ∈ out(f A , s), for all i ≥ 0, there is

[0066] · if and only if there exists a and strategy for any for any path π ∈ (f A , s), there exists 0 ≤ i < ∞, there is

[0067] · if and only if there exists a and strategy for any for any path π ∈ out(f A , s), there exists i ≥ 0, there is and for any 0 ≤ j < i, there is

[0068] · if and only if h i (s) = h i (s′), there is

[0069] · if and only if for all agents i ∈ G, there is

[0070] · If and only if all agents i∈G, R i =((s,M),(s′,M′))|h i (S) = h i (s′), YesR i 's transitive closure, we have For any

[0071] · If and only if there exists an agent i∈G, there is

[0072] Step 3: Provide a specific model detection algorithm that can recursively calculate the performance formula layer by layer, starting from the innermost atomic formula to the outer formula, to obtain the value of the attribute.

[0073] In order to parse and calculate the performance attribute formula of the multi-agent system, the present invention proposes a data structure based on a syntax tree. The syntax tree is recursively constructed by decomposing the formula into basic operators and atomic formula nodes to represent the logical structure of the formula layer by layer. Each node of the syntax tree corresponds to an element in the formula, the root node represents the highest level operator of the formula, and the leaf node represents the most basic atomic formula.

[0074] For atomic formulas, they are leaf nodes in the syntax tree; for logical operators, The basic formula composed of V, the root node is v, the left child node and the right child node represent the sub-formulas φ1 and φ2 on both sides of the operator respectively, The root node is The left child node is φ. For K containing cognitive operators i φ、E G φ、C G φ、D G φ, K, E, C, and D are the root nodes of the syntax tree constructed by φ; for < >Xφ, whose root node is < >X, the child node is the syntax tree of φ; for < >Gφ, whose root node is< >G, the child node is the syntax tree of φ; for < >Fφ, whose root node is< >F, the child node is the syntax tree of φ; for < >φ1Uφ2, its left child node is the syntax tree of φ1, and its right child node is the syntax tree of φ2. The specific structure is as follows:

[0075]

[0076]

[0077] Next, we propose a model checking algorithm based on recursive calculation for modeling multi-agent systems using EPC. This algorithm combines ATLE and EPC structures, and recursively calculates nested formulas to determine whether the formula is true. Specifically, the algorithm starts with the calculation of atomic formulas, and gradually recursively processes operators, knowledge logic formulas, and temporal logic formulas to finally obtain the formula result. The flowchart of the algorithm is shown below: Figure 2 shown.

[0078] 3.1 Input model EPC and ATLE formula φ, recursively calculate CHECK(EPC, φ)

[0079] 3.2 When φ is an atomic formula f, the result returns the value of the atomic formula in the current state s;

[0080] 3.3 When When , the result returns the opposite value of CHECK(EPC, φ);

[0081] 3.4 When φ=φ1Vφ2, the result is returned (CHECK(EPC, φ1)∪CHECK(EPC, φ2));

[0082] 3.5 When φ=K i (φ), call CHECKK(EPC, φ, i) to obtain the set of states res that are cognitively equivalent to s, where all states in s and res satisfy h i (s′)≠h i (s), then return true, otherwise return false;

[0083] 3.6 When φ=E G (φ), call CHECKE(EPC, φ, G); if all agents i in G have CHECKK(s, M, φ, i) is true, then return true, otherwise return false;

[0084] 3.7 When φ=D G (φ), call CHECKD(EPC,φ,G); if there is agent i in G and CHECKK(s,M,φ,i) is true, then return true, otherwise return false;

[0085] 3.8 When φ=C G (φ), call CHECKC(EPC,φ,G); first initialize the transitive closure For any binary relation (s, s′) that belongs to the transitive closure, all states satisfy h i (s′)≠h i (s), then return true, otherwise return false;

[0086] 3.9 When φ=< >X(φ), call CHECKX(EPC, φ, A); traverse all strategies about A, if there is a strategy that generates all paths, all paths satisfy the next state φ, then return true, otherwise return false;

[0087] 3.11 When φ=< >G(φ), call CHECKG(EPC,φ,A); traverse all strategies about A, if there is a strategy that generates all paths, all states of all paths satisfy the state φ, then return true, otherwise return false;

[0088] 3.12 When φ=< >F(φ), call CHECKF(EPC,φ,A); traverse all the strategies about A, if there is a state that satisfies the state φ for all the paths generated by a strategy, then return true, otherwise return false;

[0089] 3.13 When φ=< >φ1Uφ2, call CHECKU(EPC,φ1,φ2,A); traverse all strategies about A, if there is a strategy that generates all paths, all paths have a state that satisfies state φ2 and all previous states in this state satisfy φ2, then return true, otherwise return false;

[0090] The above calculation process is as follows:

[0091]

[0092]

[0093]

[0094] Implementation Cases

[0095] With the rapid development of drone technology, drones are widely used in military and civilian fields, such as reconnaissance, strike, transportation, etc. In the military field, drone swarm combat has gradually become an important combat mode, where multiple drones work together to complete combat missions, such as attacking and reconnaissance on enemy targets.

[0096] The multi-UAV cooperative combat system faces many challenges, such as: Cooperative control: How to coordinate the actions of multiple UAVs to complete the task in a coordinated manner and avoid conflicts. Information sharing: How to achieve information sharing between UAVs, such as target information, attack results, etc., to improve combat efficiency and effectiveness. Security: How to ensure the security of UAV swarm combat and prevent UAVs from being attacked or interfered by the enemy. Reliability: How to ensure the reliability of UAV swarm combat and ensure that UAVs can complete the task as expected.

[0097] In order to solve the above challenges, advanced modeling and verification methods are needed to analyze and evaluate the multi-UAV cooperative combat system. This case study takes the multi-UAV cooperative attack system as an example to explore how to use EPC and ATLE for system modeling and verification to ensure system safety and reliability.

[0098] The background of this case study is the actual application requirements of multi-UAV cooperative combat system, which aims to provide new ideas and methods for the safety and reliability verification of multi-agent system. We built a system model for the cooperative combat of multiple unmanned aerial vehicles, in which three UAVs attack a location under the command of the ground command center (see Figure 3 ).

[0099] Assume a series of atomic propositions: p i : Drone i successfully hits the target, q i : The drone receives an attack command. The actions in the system are: "com" is an action that issues a command and transmits cognition, while "attack" is the command for the drone to attack. The four agents are defined as follows:

[0100]

[0101]

[0102]

[0103] (1)Ag={UAV0,UAV1,GCS}.

[0104] (2)S={s0,s1,s2,s3}.

[0105] (3)ES={es0,es1,es2,es3}.

[0106] (4)

[0107] (5)

[0108]

[0109] (6)AP={p0,p1,q0,q1}.

[0110] (7)

[0111] (8)

[0112]

[0113] (9)

[0114] (10)h i ={(s0,UAV0)=is 00 ,(s0,UAV1)=is 01 ,(s0,GCS)=is 03 ,(s1,UAV0)=is 00 ,(s1,UAV1)=is 11 (s1, GCS) is 12 ,(s2,UAV0)=is 20 (s2, UAV1) is 21 (s2, GCS) is 22 ,(s3,UAV0)=is 30 ,(s3,UAV1)=is 31 ,(s3,GCS)=is 23 .....}.

[0115] (11)T={(s0, p0)=false, (s0, p1)=false, (s0, q0)=false, (s0, q1)=false, (s1, p0)=false, (s1, p1)=false, (s1, q0)=true, (s1, q1)=true, (s2, p0)=false, (s2, p1) =true, (s2, q0) = true, (s2, q1) = true, (s3, p0) = true, (s3, p1) = false, (s3, q0) = true, (s3, q1) = true, (s4, p0) = true, (s4, p1) = true, (s4, q0) = true, (s4, q1) = true}.

[0116] The above is some key information based on the definition of the marking process. Among them, Ag is the set of agents in the system, S is the set of states, ES is the set of cognitive states, PS is the set of processes of each agent, and MS is the set of all marking processes that can be generated after the entire system changes as a marking process. The rest of the definitions can be seen in the formal definition in the previous section. After the formal definition, these can be used as inputs, and the ATLE formula description of the definition of related attributes can also be used as inputs, and the model detection algorithm can be used to verify the correctness of the attributes.

[0117] See Figure 3 , (s0, M0) is the initial preparation state of the system. GCS issues a command action to order the drone cluster to attack the designated target and enters the state (s1, M1). (s1, M1) is the attack preparation state of the two drones. The two drones execute attack actions respectively and enter the (s2, M2) and (s3, M3) states respectively. After that, they transmit commands to GCS to enter the (s4, M0) state. Then GCS can continue to issue attacks and enter the (s1, M1) state to enter a cyclic link.

[0118]

[0119] In particular, the policies for the first four attributes are unrestricted; so the agent can perform any action that can be performed. However, there is a partial policy function for the fifth attribute. Come. Among them so And the 5th property is false.< / y>

Claims

1. A multi-agent system modeling and model checking method based on cognitive process calculus, characterized by: The main steps include: (1) Define the system model based on EPC; (2) Define system-related attributes based on ATLE; (3) Verify the attributes according to the ATLE model detection algorithm.

2. The multi-agent system modeling and model checking method for cognitive process calculation according to claim 1, characterized in that: The described step 1(1) is specifically: (1) First, define the actions that the agent can perform. These actions include internal actions, sending actions, and receiving actions. The content sent or received can be knowledge or specific values. (2) Then, based on the behavior and interaction mode of the agent, a specific process expression describing the agent's behavior is constructed; (3) Finally, the process expressions of all agents are combined to form a model of the multi-agent system. The model can contain multiple agents, as well as the interactions and communication relationships between them. The system model provides a holistic view of the multi-agent system and can clearly show the behavior and interaction of all agents in the system.

3. The multi-agent system modeling and model checking method for cognitive process calculation according to claim 1, characterized in that: The described step 1(2) is specifically: (1) First, define the basic formula, which consists of atomic propositions and logical operators. composition. (2) Constructing cognitive logic formula: Introducing knowledge operators K, K i φ is used to describe the knowledge state of agent i about formula φ. Introduce knowledge operators E, E G (φ) indicates that all agents in the system G know the knowledge state of φ. Introducing knowledge operator C, C G (φ) represents the public knowledge in the system, which means that all agents in G know and all agents in G know that other agents in G also know the knowledge state of φ. Introducing knowledge operators C, D G (φ) represents the distributed knowledge in the system, which means that φ is a distributed knowledge in the agent set G. (3) Constructing temporal logic formulas: Use temporal logic operators and strategic agent aggregates to construct complex ATLE attribute formulas. (4) Construct a complete ATLE attribute formula: By combining the atomic formula, the timing operator and the knowledge operator, a complete ATLE attribute formula is constructed. For example: >X(K i (φ1Vφ2)), which indicates that there is a strategy in the agent set A, so that in the next step, agent a knows that φ1Vφ2 is true.

4. The multi-agent system modeling and model checking method for cognitive process calculation according to claim 1, characterized in that: The described step 1(3) is specifically: (1) Input the system model EPC of the multi-agent system and the ATL formula φ, and recursively calculate CHECK(EPC, φ); (2) When φ is an atomic formula f, the result returns the value of the atomic formula in the current state s; (3) When When , the result returns the opposite value of CHECK(EPC, φ); (4) When φ=φ1Vφ2, the result is returned as (CHECK(EPC, φ1)∪CHECK(EPC, φ2)); (5) When φ=K i (φ), call CHECKK(EPC, φ, i); (6) When φ=E G (φ), call CHECKE(EPC, φ, G); (7) When φ=D G (φ), call CHECKD(EPC, φ, G); (8) When φ=C G (φ), call CHECKC(EPC, φ, G); (9) When φ=< >X(φ), call CHECKX(EPC, φ, A); (10) When φ=< >G(φ), call CHECKG(EPC, φ, A); (11) When φ=< >F(φ), call CHECKF(EPC, φ, A); (12) When φ=< >φ1Uφ2, call CHECKU(EPC,φ1,φ2,A).

Images