Multi-agent negotiation method, system, equipment and medium

Abstract

The invention provides a multi-agent negotiation method, system and device and a medium, and belongs to the technical field of distributed optimization. The method comprises the following steps: for each agent, sending a local sub-population optimal solution as a candidate solution to an adjacent agent, receiving the candidate solution of the adjacent agent, calculating local fitness when a local candidate solution value and an adjacent agent candidate solution value are adopted for each shared variable, and generating a fitness vector; the adjacent agents exchange fitness vectors, the global fitness of each shared variable is calculated, a better candidate solution is selected as a consensus solution by comparing the global fitness, and a competition result vector is generated; updating a penalty switch according to the competition result vector, updating candidate solutions of competition failure in the local sub-population by using a consensus solution, and updating a penalty base objective function; gradient estimation is carried out on the shared variable, the adjacent agents exchange gradient estimation results, and if the gradient directions of the two agents on the shared variable are consistent, penalty term constraints are removed.

Description

Technical Field

[0001] This invention belongs to the field of distributed optimization technology, and more specifically relates to a multi-agent negotiation method, system, device and medium. Background Technology

[0002] In modern network systems, distributed optimization has become a core technological requirement for scenarios such as smart grid dispatching, collaborative sensing in wireless sensor networks, and collaborative control of industrial IoT devices. These scenarios require multiple autonomous agents to optimize global objectives by interacting with neighboring nodes, even in the absence of global information. As network scale expands and application scenarios become more complex, the coupling between agents becomes increasingly tight. The efficiency of reaching consensus on shared variables directly determines the overall performance of the distributed system. Therefore, an efficient multi-agent negotiation mechanism becomes crucial for improving system collaboration capabilities. Among existing distributed optimization techniques, gradient-based methods are one of the earliest developed solutions, achieving convergence of the optimization objective through iterative updates of gradient information. However, these methods heavily rely on the assumptions of mathematical differentiability and convexity of the optimization problem, making them difficult to apply to black-box problems (such as industrial control processes without explicit mathematical models) and non-convex problems (such as multimodal function optimization scenarios) that are prevalent in practical engineering, thus greatly limiting their application scope. Furthermore, gradient methods are sensitive to initial values, prone to getting trapped in local optima, and exhibit poor convergence stability in scenarios where agent objectives conflict. While traditional distributed evolutionary algorithms can overcome the mathematical limitations of gradient methods and achieve global optimization through population evolution, they suffer from significant shortcomings in the multi-agent negotiation phase. On one hand, existing algorithms generally lack structured consensus guidance mechanisms. Agents adjust their local decisions merely through simple information exchange, leading to persistent conflicts in shared variables during iteration, resulting in long consensus-building cycles and severely impacting the convergence speed of distributed optimization. On the other hand, their penalty mechanisms often employ fixed weights or empirical parameters, failing to dynamically adapt to real-time changes in goal conflicts among agents. Excessive penalty for conflict-free variables reduces the degrees of freedom in local optimization, while insufficient penalty for highly conflicting variables hinders consensus formation, resulting in low collaborative efficiency. Furthermore, existing negotiation methods fall short in balancing communication costs and negotiation efficiency. In pursuit of consensus accuracy, some methods require frequent exchange of large amounts of global information, leading to an exponential increase in communication burden with network size. This can easily result in communication congestion and latency accumulation in large-scale agent networks. Conversely, methods that oversimplify communication content can lead to decreased negotiation accuracy due to insufficient information, making it difficult to achieve efficient and reliable collaborative optimization in real-world large-scale distributed systems. These issues make it difficult for existing technologies to meet the practical needs of efficient multi-agent collaboration in complex network environments. Summary of the Invention To address the above problems, the present invention aims to provide a multi-agent negotiation method, system, device, and medium. Through a structured negotiation process and dynamic conflict detection, it achieves efficient consensus on shared variables among agents, thereby improving the collaborative efficiency and convergence performance of distributed optimization.

[0003] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, embodiments of this application provide a multi-agent negotiation method, including: For each agent, the local subpopulation optimal solution is sent as a candidate solution to neighboring agents, and the candidate solutions of neighboring agents are received. For each shared variable, the local fitness is calculated when using the local candidate solution value and the candidate solution value of neighboring agents, and a fitness vector is generated. Neighboring agents exchange fitness vectors, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector. Update the penalty switch based on the competition result vector, update the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and update the penalty basis objective function; Gradient estimation is performed on the shared variables, and adjacent agents exchange gradient estimation results. If the gradient directions of the two agents on the shared variables are consistent, the penalty term constraint is removed.

[0004] In an optional implementation, for each agent, the local subpopulation optimal solution is sent as a candidate solution to neighboring agents, and candidate solutions from neighboring agents are received. For each shared variable, the local fitness is calculated using both the local candidate solution value and the neighboring agent's candidate solution value, generating a fitness vector, including: Any agent i sends the optimal solution of its local subpopulation as a candidate solution to its neighboring agent j, and synchronously receives the candidate solutions sent by agent j. For agent i, for each shared variable d, calculate the local fitness when using local candidate solutions. Local fitness when using candidate solution values ​​of agent j ; For agent j, for each shared variable d, calculate the local fitness when using local candidate solutions. Local fitness when using candidate solution values ​​of agent i .

[0005] In an optional implementation, the neighboring agents exchange fitness vectors and calculate the global fitness for each shared variable, including: Agents i and j exchange the local fitness calculated using the candidate solution values ​​of neighboring agents; Through formula Calculate the global fitness A of the candidate solutions for agent i; Through formula Calculate the global adequacy B of the candidate solutions for agent j.

[0006] In an optional implementation, the step of selecting a better candidate solution as the consensus solution by comparing global fitness and generating a competition result vector includes: Compare A and B using the consensus solution selection rules; If the current problem is a maximization problem, determine whether A > B; if so, then take the candidate solution of agent i as the consensus solution. If not, then the candidate solution of agent j will be used as the consensus solution. ; If the current problem is a minimization problem, determine whether A < B; if so, then take the candidate solution of agent i as the consensus solution. If not, then the candidate solution of agent j will be used as the consensus solution. ; When the candidate solution of agent i is used as the consensus solution At that time, generate a competition result vector. ; When the candidate solution of agent j is used as the consensus solution At that time, generate a competition result vector. .

[0007] In an optional implementation, the step of updating the penalty switch based on the competition result vector, updating the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and updating the penalty basis objective function includes: Based on competition outcome vector Update penalty switch ;like ,but ; like ,but ; if Then use consensus solution Update the candidate solution values ​​for agent i; The penalty basis objective function is further updated as follows:

[0008] in, Let N be the original objective function of agent i. i Let i be the set of neighboring agents of agent i. I i,j Let w be the subset of shared variables between agents i and j, and w be the penalty weight.

[0009] In an optional implementation, the penalty weight ; Where λ is the preset scaling factor, and n is the total number of agents. Let be the consensus solution for agent k.

[0010] In an optional implementation, the gradient estimation of the shared variable, the exchange of gradient estimation results between adjacent agents, and the removal of the penalty term constraint if the gradient directions of the two agents on the shared variable are consistent, include: For the shared variable d, through the formula Calculate the positive perturbation gradient of agent i using the formula Calculate the negative perturbation gradient of agent i; where, It is a d-dimensional unit vector. The perturbation step size; For the shared variable d, through the formula Calculate the positive perturbation gradient of agent j using the formula Calculate the negative perturbation gradient of agent j; if × >0, and × If >0, then set .

[0011] Secondly, embodiments of this application also provide a multi-agent negotiation system, including: The evaluation module is used to send the local subpopulation optimal solution as a candidate solution to neighboring agents for each agent, and receive the candidate solutions of neighboring agents. It calculates the local fitness for each shared variable using the local candidate solution value and the candidate solution value of neighboring agents, and generates a fitness vector. The competition module is used for neighboring agents to exchange fitness vectors, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector. The shared module is used to update the penalty switch based on the competition result vector, update the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and update the penalty basis objective function. The conflict detection module is used to perform gradient estimation on shared variables. Neighboring agents exchange gradient estimation results. If the gradient directions of the two agents on the shared variables are consistent, the penalty term constraint is removed.

[0012] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the multi-agent negotiation method as described in any of the above.

[0013] Fourthly, embodiments of this application also provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the multi-agent negotiation method as described in any of the above claims.

[0014] As can be seen from the above technical solutions, the present invention has the following advantages: The multi-agent negotiation method provided in this application achieves efficient consensus on shared variables among multiple agents in a network environment through a structured negotiation process and dynamic conflict detection. This method exchanges optimal solutions and cross-evaluates fitness during the evaluation phase; determines a consensus solution based on global fitness during the competition phase; guides population evolution by adjusting a penalty switch during the sharing phase; and dynamically optimizes the penalty mechanism through gradient estimation during the conflict detection phase. This method improves the collaborative efficiency of agents in distributed optimization, is applicable to black-box, non-convex network distributed optimization problems, and has advantages such as low communication cost, high consensus efficiency, and adaptive conflict handling.

[0015] This application extends the decision-making of a single agent to a global consideration of multiple objectives by integrating the local evaluation results of an individual and neighboring agents. This ensures that the final consensus solution not only meets individual needs but also takes into account the overall optimality, thus avoiding decision-making bias caused by information isolation.

[0016] This application relies on clear consensus selection rules and result feedback mechanisms to provide agents with a unified decision-making standard. By directly updating the failed competing solutions, it promotes the alignment of individuals with the group consensus, significantly reducing decision-making conflicts among multiple agents and improving the efficiency and consistency of consensus-building.

[0017] This application flexibly adjusts the constraint intensity based on the consensus competition results, applies targeted penalties to behaviors that deviate from the consensus, and dynamically changes the penalty intensity with the global goal. This both promotes the agent to move closer to the consensus and avoids excessive constraints from affecting the individual's optimization autonomy, thus achieving a balance between cooperation and autonomy.

[0018] This application determines whether the intelligent agents have the same optimization direction on the shared goal, and automatically removes unnecessary constraints in areas with the same direction, avoiding ineffective negotiation costs and allowing negotiation resources to be concentrated on the points of disagreement that really need coordination, thereby improving the overall optimization efficiency.

[0019] This application standardizes each stage, from information exchange and evaluation calculations to consensus updates and constraint adjustments, through clear rules. This reduces the risk of confusion caused by operational ambiguity, ensures that the multi-agent negotiation process is repeatable and controllable, and maintains long-term stable operation. Attached Figure Description

[0020] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the multi-agent negotiation method provided in this application.

[0022] Figure 2 This is a schematic diagram of the structure of the multi-agent negotiation system provided in this application.

[0023] Figure 3 A schematic diagram of the structure of the electronic device provided in this application. Detailed Implementation

[0024] The various embodiments of this disclosure will be described more fully in the detailed steps of the multi-agent negotiation method described below. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0025] In the following, the terms “comprising” or “may include”, which may be used in various embodiments of this disclosure, indicate the presence of the disclosed functions, operations, or elements, and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in various embodiments of this disclosure, the terms “comprising,” “having,” and their cognates are intended only to indicate a particular feature, number, step, operation, element, component, or combination of the foregoing, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or the possibility of adding one or more combinations of the foregoing.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 The diagram shown is a flowchart of a multi-agent negotiation method in a specific embodiment. The method includes: S1: For each agent, send the local subpopulation optimal solution as a candidate solution to neighboring agents, and receive the candidate solutions from neighboring agents. For each shared variable, calculate the local fitness when using the local candidate solution value and the candidate solution value from neighboring agents, and generate a fitness vector.

[0028] In a specific implementation, this step is the evaluation phase, used for local candidate solution interaction and fitness vector generation.

[0029] First, any agent i selects the optimal solution from its local subpopulation as a candidate solution and sends it to its neighboring agent j; at the same time, it synchronously receives the candidate solution sent by agent j, thus completing the basic information exchange.

[0030] Then, for each shared variable d, the local fitness is calculated for both scenarios: Local fitness of agent i when using local candidate solutions Local fitness when using candidate solution values ​​of agent j ; Local fitness of agent j when using local candidate solutions Local fitness when using candidate solution values ​​of agent i .

[0031] Finally, agent i generates a function containing... and The fitness vector generated by agent j contains and The fitness vector.

[0032] S2: Neighboring agents exchange fitness vectors, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector.

[0033] In a specific implementation, this step is the competition phase, used for global fitness calculation and consensus solution generation, as well as competition result generation.

[0034] First, agents i and j send each other the local fitness values ​​corresponding to each other's candidate solutions, i.e. and This is to achieve full sharing of assessment data.

[0035] Then, based on the exchanged fitness data, the global fitness of the candidate solutions for both sides is calculated for the shared variable d: Through formula Calculate the global fitness A of the candidate solutions for agent i; Through formula Calculate the global adequacy B of the candidate solutions for agent j.

[0036] Finally, the consensus solution is selected and the competition results are recorded: If the current problem is a maximization problem, determine whether A > B; if so, then take the candidate solution of agent i as the consensus solution. If not, then the candidate solution of agent j will be used as the consensus solution. ; If the current problem is a minimization problem, determine whether A < B; if so, then take the candidate solution of agent i as the consensus solution. If not, then the candidate solution of agent j will be used as the consensus solution. ; When the candidate solution of agent i is used as the consensus solution At that time, generate a competition result vector. ; When the candidate solution of agent j is used as the consensus solution At that time, generate a competition result vector. .

[0037] S3: Update the penalty switch based on the competition result vector, update the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and update the penalty basis objective function.

[0038] In a specific implementation, this step is a sharing phase, used for penalty switch updates, population optimization, and adjustment of the objective function.

[0039] First, based on the competition result vector Adjust the penalty switch : like ,but That is, if agent i wins the competition in variable d, no penalty is imposed.

[0040] like ,but That is, if agent i fails in the competition for variable d, a penalty is imposed.

[0041] Then, update the local candidate solution: if Then use consensus solution Update the candidate solution values ​​of agent i; that is, force the candidate solution values ​​of variable d in agent i's local subpopulation to be updated to the consensus solution. This guides the population to evolve toward consensus.

[0042] Finally, update the penalty basis objective function of agent i based on the current state:

[0043] in, Let N be the original objective function of agent i. i Let i be the set of neighboring agents of agent i. I i,j Let w be the subset of shared variables between agents i and j, and w be the penalty weight.

[0044] Penalty weights are determined by the formula. Adaptive computation.

[0045] Where λ is the preset scaling factor, and n is the total number of agents. Let be the consensus solution for agent k.

[0046] S4: Perform gradient estimation on the shared variables. Neighboring agents exchange gradient estimation results. If the gradient directions of the two agents on the shared variables are consistent, remove the penalty term constraint.

[0047] In a specific implementation, this step is the conflict detection stage, used for gradient estimation and dynamic removal of penalty terms.

[0048] For the shared variable d, through the formula Calculate the positive perturbation gradient of agent i using the formula Calculate the negative perturbation gradient of agent i; where, It is a d-dimensional unit vector. The perturbation step size; For the shared variable d, through the formula Calculate the positive perturbation gradient of agent j using the formula Calculate the negative perturbation gradient of agent j.

[0049] At this point, if the gradient estimation results of two adjacent agents exchanged satisfy: × >0, and × >0 Then it is determined that the gradient directions of both parties on variable d are consistent, there is no optimization conflict, and the settings are set. Remove the penalty constraint from the variable.

[0050] It is evident that through the cyclical execution of the above four stages, combined with formulaic computational logic and rule-based judgment criteria, multi-agent systems can achieve precise fusion of local information and global objectives. This not only ensures the global optimality of the consensus solution but also improves negotiation efficiency through dynamic penalties and conflict detection, ultimately achieving efficient collaborative decision-making.

[0051] In this embodiment, a data foundation is laid through candidate solution interaction and local fitness evaluation among agents. The global fitness is calculated by combining the evaluation results of both parties to select the optimal consensus solution. The penalty constraint state is dynamically adjusted and the population direction is updated based on the competition results. The group is guided to evolve towards consensus through an adaptive penalty mechanism. At the same time, redundant penalties of conflict-free variables are removed based on gradient direction consistency detection. Ultimately, the global optimal consideration, efficient consensus achievement, flexible collaborative adaptation and precise optimization control of multi-agent decision-making are realized, ensuring the quality and reliability of group collaborative decision-making.

[0052] like Figure 2 As shown, the following are embodiments of the multi-agent negotiation system provided in this disclosure. This system and the multi-agent negotiation methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the multi-agent negotiation system, please refer to the embodiments of the above multi-agent negotiation methods.

[0053] A multi-agent negotiation system, comprising: The evaluation module is used to send the local subpopulation optimal solution as a candidate solution to neighboring agents for each agent, and receive the candidate solutions of neighboring agents. It calculates the local fitness for each shared variable using the local candidate solution value and the candidate solution value of neighboring agents, and generates a fitness vector.

[0054] The competition module is used for neighboring agents to exchange fitness vectors, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector.

[0055] The shared module is used to update the penalty switch based on the competition result vector, update the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and update the penalty basis objective function.

[0056] The conflict detection module is used to perform gradient estimation on shared variables. Neighboring agents exchange gradient estimation results. If the gradient directions of the two agents on the shared variables are consistent, the penalty term constraint is removed. The multi-agent negotiation system provided in this embodiment avoids local decision-making bias by integrating global goals, achieves efficient group consensus and reduces conflicts by relying on clear rules, dynamically adjusts the constraint strength to balance autonomous optimization and collaborative needs, removes redundant constraints by judging the consistency of optimization direction to improve negotiation efficiency, and the standardized procedures of each link effectively ensure the stability and operability of the negotiation process, thus fully realizing high-quality, efficient and reliable multi-agent collaborative decision-making.

[0057] Figure 3 A schematic diagram of the hardware structure of an electronic device for implementing various embodiments of the present invention.

[0058] The multi-agent negotiation method provided in this application can be applied to electronic devices. Those skilled in the art will understand that the electronic device structure involved in the embodiments of this invention does not constitute a limitation on the electronic device. An electronic device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, the electronic device includes, but is not limited to, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0059] Electronic devices may include processors, external memory interfaces, internal memory, universal serial bus (USB) interfaces, charging management modules, power management modules, batteries, wireless communication modules, audio modules, speakers, microphones, sensor modules, buttons, cameras, displays, and SIM card interfaces, etc.

[0060] A processor may include one or more processing units, such as: a central processing unit (CPU), an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, memory, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural network processing unit (NPU). Different processing units may be independent devices or integrated into one or more processors.

[0061] The processor can serve as the nerve center and command center of an electronic device. The controller can generate operation control signals based on the instruction opcode and timing signals to control the fetching and execution of instructions.

[0062] The processor may also include memory for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can store instructions or data that the processor has just used or that are used repeatedly. If the processor needs to use the instruction or data again, it can retrieve it directly from this memory. This avoids repeated accesses, reduces processor latency, and thus improves system efficiency.

[0063] An external storage interface (ESI) can be used to connect external memory cards, such as microSD cards, to expand the storage capacity of electronic devices. The external memory card communicates with the processor through the ESI to perform data storage functions, such as saving music and video files on the external memory card.

[0064] Internal memory can be used to store computer executable program code, which includes instructions. The processor executes various functional applications and data processing of electronic devices by running the instructions stored in internal memory. Internal memory can include a program storage area and a data storage area. Internal memory can include high-speed random access memory, and can also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash storage (UFS), etc.

[0065] Wireless communication functionality in electronic devices can be achieved through antennas, wireless communication modules, modem processors, and baseband processors.

[0066] Wireless communication modules can provide solutions for wireless communication applications in electronic devices, including wireless local area networks (WLANs) (such as wireless fidelity (Wi-Fi) networks), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), and infrared (IR) technologies.

[0067] Electronic devices can implement audio functions through audio modules, speakers, receivers, microphones, headphone jacks, and application processors.

[0068] Electronic devices can achieve shooting functions through ISPs, cameras, video codecs, GPUs, displays, and application processors.

[0069] Electronic devices can achieve display functions through GPUs, displays, and application processors.

[0070] A GPU is a microprocessor for image processing, connected to the display screen and application processor. GPUs are used to perform mathematical and geometric calculations for graphics rendering. A processor may include one or more GPUs, which execute program instructions to generate or modify display information.

[0071] A display screen is used to display images, videos, etc. A display screen includes a display panel.

[0072] The aforementioned electronic device realizes the multi-agent negotiation method of this application by integrating global goals and avoiding local decision-making biases, efficiently achieving group consensus and reducing conflicts based on clear rules, dynamically adjusting the constraint strength to balance autonomous optimization and collaborative needs, removing redundant constraints by judging the consistency of optimization direction to improve negotiation efficiency, and ensuring negotiation stability and operability with standardized and normative processes, thus achieving the beneficial effects of high-quality, efficient and reliable multi-agent collaborative decision-making.

[0073] The storage medium provided in this application stores a program product capable of implementing a multi-agent negotiation method.

[0074] Multi-agent negotiation methods include: For each agent, the local subpopulation optimal solution is sent as a candidate solution to neighboring agents, and the candidate solutions of neighboring agents are received. For each shared variable, the local fitness is calculated when using the local candidate solution value and the candidate solution value of neighboring agents, and a fitness vector is generated. Neighboring agents exchange fitness vectors, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector. Update the penalty switch based on the competition result vector, update the candidate solutions that failed in the competition in the local subpopulation using the consensus solution, and update the penalty basis objective function; Gradient estimation is performed on the shared variables, and adjacent agents exchange gradient estimation results. If the gradient directions of the two agents on the shared variables are consistent, the penalty term constraint is removed. In some possible implementations, the multi-agent negotiation method of this disclosure can be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0075] The storage medium disclosed herein may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-agent negotiation method, characterized by, The method comprises the following steps: For each agent, the optimal solution of the local sub-population is sent to the adjacent agent as a candidate solution, and the candidate solution of the adjacent agent is received, and the local fitness is calculated when the local candidate solution value and the adjacent agent candidate solution value are used for each shared variable, and a fitness vector is generated; The adjacent agents exchange the fitness vectors, calculate the global fitness of each shared variable, and select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector; The penalty switch is updated according to the competition result vector, the consensus solution is used to update the candidate solution that fails in the competition in the local sub-population, and the penalty-based objective function is updated; Gradient estimation is performed on the shared variable, the adjacent agents exchange the gradient estimation results, and if the gradient directions of the two agents on the shared variable are consistent, the penalty term constraint is removed.

2. The multi-agent negotiation method of claim 1, wherein, The method comprises the following steps: Any agent i sends the optimal solution of the local sub-population to the adjacent agent j as a candidate solution, and synchronously receives the candidate solution sent by the agent j; For each shared variable d, the local fitness of the candidate solution value of agent i and the local fitness of the candidate solution value of agent j are calculated respectively. For each shared variable d, the local fitness of the candidate solution value of agent i is calculated and the local fitness of the candidate solution value of agent j is calculated For each shared variable d, the local fitness of the candidate solution value of agent i is calculated and the local fitness of the candidate solution value of agent j is calculated For each shared variable d, the local fitness of the candidate solution value of agent i is calculated and the local fitness of the candidate solution value of agent 3. The multi-agent negotiation method of claim 2, wherein, The adjacent agents exchange the fitness vectors, calculate the global fitness of each shared variable, and select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector; The method comprises the following steps: The global fitness A of the candidate solution of the agent i is calculated by the formula A = f (x) The global fitness B of the candidate solution of agent j is calculated by the formula B = f (xj) 4. The multi-agent negotiation method of claim 3, wherein, A and B are compared through the selection rule of the consensus solution; The penalty switch is updated according to the competition result vector, the consensus solution is used to update the candidate solution that fails in the competition in the local sub-population, and the penalty-based objective function is updated; If the current is a maximization problem, judge whether A > B; if yes, the candidate solution of agent i is taken as the consensus solution ; if not, the candidate solution of agent j is taken as the consensus solution ; If the current is a minimization problem, determine whether A < B; if yes, then the candidate solution of agent i is the consensus solution ; if no, then the candidate solution of agent j is the consensus solution ; When the candidate solution of the agent i is as the consensus solution , a competition result vector is generated ; When the candidate solution of the agent j is as the consensus solution , a competition result vector is generated.

5. The multi-agent negotiation method of claim 4, wherein, The penalty-based objective function is further updated as follows: Vector of competition results Update penalty switch ; if then ; If then ; If then update the candidate solution value of agent i with the consensus solution ; The method comprises the following steps: wherein, is the original objective function of agent i, N i is the set of neighboring agents of agent i, I i,j w is a penalty weight for a subset of shared variables of agent i and agent j.

6. The multi-agent negotiation method of claim 5, wherein, the penalty weight ; wherein λ is a preset scaling coefficient, n is the total number of agents, is the consensus solution of the agent k.

7. The multi-agent negotiation method of claim 6, wherein, The system adopts the multi-agent negotiation method according to any one of claims 1 to 7; For the shared variable d, the positive perturbation gradient of agent i is calculated by the formula The negative perturbation gradient of agent i is calculated by the formula ; wherein, is a d-dimensional unit vector, is a perturbation step size; For the shared variable d, the positive perturbation gradient for agent j is computed by the formula The negative perturbation gradient for agent j is computed by the formula​ If x > 0, and x > 0, then set .

8. A multi-agent negotiation system, characterized in that, The system comprises: An evaluation module is configured to, for each agent, send the optimal solution of the local sub-population to the adjacent agent as a candidate solution, receive the candidate solution of the adjacent agent, calculate the local fitness when the local candidate solution value and the adjacent agent candidate solution value are used for each shared variable, and generate a fitness vector; A competition module is configured to exchange the fitness vectors of the adjacent agents, calculate the global fitness of each shared variable, select the better candidate solution as the consensus solution by comparing the global fitness, and generate a competition result vector; A sharing module is configured to update the penalty switch according to the competition result vector, update the candidate solution that fails in the competition in the local sub-population by using the consensus solution, and update the penalty-based objective function; A conflict detection module is configured to perform gradient estimation on the shared variable, exchange the gradient estimation results of the adjacent agents, and remove the penalty term constraint if the gradient directions of the two agents on the shared variable are consistent. ​ 9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the multi-agent negotiation method as claimed in any one of claims 1 to 7 when executing the program.

10. A storage medium having stored thereon a computer program, characterized in that The computer program implements the steps of the multi-agent negotiation method as claimed in any one of claims 1 to 7 when executed by the processor.

Images