A digital twin application method for airports based on multi-agent model

Abstract

The present invention relates to the field of airport digital twin technology, and specifically, to an airport digital twin application method based on a multi-agent model, the steps of which include: defining a real agent set and a virtual agent set based on a digital twin platform; defining actions for the real agent set and the virtual agent set respectively, and connecting the actions defined by the real agent set with the actions defined by the virtual agent set based on a large language model to form a collaborative autonomous agent set; acquiring airport operation data, positively evaluating the airport operation data based on the collaborative autonomous agent set, and combining the positive evaluation results, constructing an optimization strategy through a large language model, and reversely inputting the optimization strategy into the digital twin platform for verification, and interacting with the airport operation data in real time to complete the construction of a multi-agent data twin closed-loop framework of the airport. The present invention effectively improves the operation efficiency and decision-making analysis accuracy of the airport.

Description

Technical Field

[0001] The present invention relates to the technical field of airport digital twins, and in particular to an airport digital twin application method based on a multi-agent model. Background Art

[0002] With the continuous expansion of airport business scale and the continuous improvement of complexity, traditional airport operation management methods can no longer meet the efficient operation needs of modern airports. First, various business systems are often in an information island state, lacking effective data sharing and coordination mechanisms, resulting in low overall operational efficiency; second, the decision-making process mainly relies on manual experience, lacks scientific quantitative analysis and prediction capabilities, and is difficult to cope with complex and changing operating environments; third, traditional management methods cannot achieve real-time dynamic optimization, the system response speed is slow, and it is difficult to handle emergencies in a timely manner; finally, there is a lack of effective simulation verification mechanism, and the implementation of new management strategies is high-risk, which may cause operational interruptions. Summary of the invention

[0003] The purpose of the present invention is to provide an airport digital twin application method based on a multi-agent model, which realizes the seamless connection and real-time interaction of the airport's virtual and real environments by constructing a closed-loop framework combining digital twins and multi-agents. At the same time, with the help of the intelligent decision-making ability of the large language model, it can independently optimize and verify strategies, greatly improving the airport's operational efficiency and decision-making accuracy. At the same time, the closed-loop framework ensures the reliability of the optimization strategy and the predictability of the implementation effect.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for applying airport digital twins based on a multi-agent model, the method comprising the following steps:

[0006] Define real agent sets and virtual agent sets based on the digital twin platform;

[0007] Define actions for the real agent set and the virtual agent set respectively, and concatenate the actions defined by the real agent set and the actions defined by the virtual agent set based on the large language model to form a collaborative autonomous agent set;

[0008] Acquire airport operation data, conduct positive evaluation on the airport operation data based on a set of collaborative autonomous agents, and build an optimization strategy through a large language model based on the positive evaluation results. The optimization strategy is then reversely input into the digital twin platform for verification, and interacts with the airport operation data in real time to complete the construction of the airport's multi-agent data twin closed-loop framework.

[0009] Optionally, the real agent set and the virtual agent set are defined based on the digital twin platform, which are specifically:

[0010] The real agent set is defined as a set of physical systems of the airport, and the operation data of the airport is output based on the operation of the real agent set;

[0011] The virtual agent set is defined as a simulated collaborative optimization agent set constructed to achieve the association and collaboration of airport operation data.

[0012] Optionally, the specific formation process of the collaborative autonomous agent set is as follows:

[0013] Data perception of a real set of intelligent agents and other sets of intelligent agents, where the other intelligent agents are specifically non-airport physical systems;

[0014] Based on the set of virtual agents, define the action plan of each agent in units;

[0015] Execute action planning, and memorize the perception, planning and action results of each agent, and connect data perception, planning, action and memory actions in series based on the large language model to form the collaborative autonomous agent set.

[0016] Optionally, the obtaining of airport operation data further includes cleaning and preprocessing the airport operation data to ensure data consistency.

[0017] Optionally, the positive evaluation of the airport operation data based on the collaborative autonomous agent set is specifically:

[0018] A single agent obtains the corresponding N business node data to form a single agent's perception data set, and performs a union operation on each agent to form a global situation;

[0019] Identify the airport's business bottlenecks based on the overall situation, analyze the causes of each bottleneck, and establish a corresponding relationship between the bottleneck and the cause;

[0020] An evaluation index is defined based on the bottleneck point, and an evaluation dimension is determined according to the cause to form an evaluation result set, which is characterized as a positive evaluation result.

[0021] Optionally, the optimization strategy is constructed by combining the positive evaluation results with a large language model, which is specifically:

[0022] Inputting the evaluation result set into the large language model to generate an analysis report, wherein the analysis report includes analysis results of the operating status, bottleneck points, and preliminary optimization suggestions;

[0023] Generate optimization strategies based on analysis reports, segment the optimization strategies through large language models, convert them into operation plans, and generate instructions that can be recognized by the digital twin platform.

[0024] Optionally, the optimization strategy is inputted back into the digital twin platform for verification, which is specifically:

[0025] Build a simulation model based on the digital twin platform, perform simulation runs on the optimization strategy, and collect simulation data;

[0026] Compare the differences between the simulation data and the actual data, and iteratively update the optimization strategy based on the comparison results until the maximum number of iterations is reached, and then output the optimal optimization strategy.

[0027] Optionally, the specific calculation formula for comparing the difference between the simulation data and the actual data is:

[0028]

[0029] Among them, A is the comparative analysis result, Compare is the comparative analysis function, S is the simulation data, For actual data.

[0030] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0031] The present invention achieves seamless connection and real-time interaction between the virtual and real environments of the airport by constructing a closed-loop framework that combines digital twins with multi-agents. At the same time, with the help of the intelligent decision-making ability of the large language model, it can independently optimize and verify strategies, greatly improving the airport's operational efficiency and decision-making accuracy. At the same time, the closed-loop framework ensures the reliability of the optimization strategy and the predictability of the implementation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic flow chart of an airport digital twin application method based on a multi-agent model provided by the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, what is described is only a part of the present invention, not all of it. Generally, the components of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0034] like Figure 1 As shown, the present invention provides one embodiment: an airport digital twin application method based on a multi-agent model, the steps of the method comprising:

[0035] Define real agent sets and virtual agent sets based on the digital twin platform;

[0036] Define actions for the real agent set and the virtual agent set respectively, and concatenate the actions defined by the real agent set and the actions defined by the virtual agent set based on the large language model to form a collaborative autonomous agent set;

[0037] Acquire airport operation data, conduct positive evaluation on the airport operation data based on a set of collaborative autonomous agents, and build an optimization strategy through a large language model based on the positive evaluation results. The optimization strategy is then reversely input into the digital twin platform for verification, and interacts with the airport operation data in real time to complete the construction of the airport's multi-agent data twin closed-loop framework.

[0038] In this embodiment, this embodiment is specifically based on the integration capability of digital twin technology to construct virtual and actual intelligent entities, which are mapped in real time in the twin world to comprehensively reflect the business operation status of the airport. Then, on the basis of multiple intelligent entities, the barriers between multiple business systems are broken down from the top level, and the multi-agents formed by the business system of the physical airport are defined and collaboratively planned using large model technology. Next, using the real-time simulation capability of the digital twin platform, a corresponding evaluation and analysis method is constructed to dynamically evaluate the actual operation status of the physical airport, and to achieve positive feedback and evaluation of the physical airport to the digital twin world. Finally, a closed-loop feedback decision-making mechanism combining large models and digital twins is constructed to support airport operation managers in dynamic decision optimization and auxiliary operation through real-time feedback and evaluation.

[0039] In this process, the specific steps provided in this embodiment are as follows:

[0040] Based on this, in the implementation of step (1):

[0041] In this embodiment, two major types of agents are first defined in the digital twin platform: A1, the existing business system agents of the airport, i.e., a set of real agents, and A2, the virtual agents, i.e., a set of virtual agents. Among them, A1 agents include: the airport's collaborative decision-making system, the airfield management system, the resource allocation system, the vehicle positioning system, the airside management system, the departure control system, the baggage handling system, the ground service system, etc. The A2 agent is a virtual role that mainly simulates manual collaboration, and mainly includes: command issuance, data aggregation, feedback control, simulation analysis, task decomposition and other functions. It is responsible for aggregating data from all business systems and transmitting it bidirectionally to the existing business system and the digital twin platform.

[0042] Then, the business system data is divided, and for a certain agent in A1 agent , where i represents the actual business system of the airport (such as the departure system), Data stored in Divided into internal system interactions and external influences ,The dataset of external influence represents the influence generated by the ,collaboration of intelligent agents and the planning simulation data generated in the ,digital twin platform, and is defined as follows:

[0043]

[0044]

[0045] in, is a specific internal data point, m is the number of internal data points, is a specific external data point, and n is the number of internal data points.

[0046] Next, this embodiment uses the A2 agent to realize the association of business data and the construction of logic. X(t) is used to represent the original business data summary result, which is defined as formula (3). G represents the business node network, where the node represents each component in the business system, and the edge represents the interaction relationship between the components, which is defined as shown in formula (4), where V is a node set, representing each component in the business system, and E is an edge set, representing the interaction relationship between the components.

[0047]

[0048]

[0049] Finally, an overall collaborative agent is constructed in the A2 agent to manage and summarize all agents, comprehensively associate and aggregate the defined data and business, and issue instructions. The issuance of instructions C can be expressed as a function, which generates specific instructions based on the business node network G and the current business data XX: ,in is a specific instruction, and k is the number of instructions.

[0050] In the implementation of step (2):

[0051] First, the actions of A1 and A2 agents are defined, including perception, planning, action, and memory. Perception is divided into data perception of the real-time operation process of the physical airport and perception of the actions of non-self agents. represents the real-time operation data of the airport perceived by agent i. The process is shown in formula (5), where Represents the data perceived by the agent at a certain perception point at time t, such as flight status, luggage location, passenger flow, etc. It represents the perception of the actions of other agents, and its definition is shown in formula (6), where Represents the action result data of other agents at time t.

[0052]

[0053] Secondly, after sensing all the data, action planning is required, which is achieved through the A2 agent. For each agent i, an action plan is defined as shown in formula (7), where represents the planning function of agent i, Represents the specific strategy of a certain action, there are k in total.

[0054]

[0055] Then, after each agent i receives the time series data from the airport business system, some actual actions and feedback results need to be saved. This embodiment uses a memory function to characterize the process, which is defined as shown in formula (8), where is the memory set of agent i, Indicates the length of T memory quantity at a time node.

[0056]

[0057] Next, the A2 virtual agent connects perception, planning, action, and memory in series to achieve overall collaborative autonomy. The complete process of an agent i at time t can be expressed as shown in formulas (9) and (10), where represents the result of the action at time t+1, according to The planning results are obtained. Indicates how the memory result is updated after the action is completed.

[0058]

[0059] Finally, the planning action memory results of the agent are optimized through collaborative decision-making, and distributed decision-making is carried out from the A2 agent, considering each agent's independent decision-making and the impact on other agents.

[0060] In the specific implementation of step (3):

[0061] First, based on the constructed intelligent agent, this embodiment collects all-field data and business operation status diagnosis. The data is cleaned and pre-processed to ensure data consistency. Then the global situation is perceived and used represents the data perceived by agent i at time t, and its definition is shown in formula (11), where It represents the memory data of the Nth business node in the agent at time t, which is the data obtained through the previous agent autonomy process. There are N business nodes in total. After obtaining the perception result of agent i, global diagnosis is performed. Its definition is shown in formula (12), where S(t) is the global operation status and Z represents the number of agents.

[0062]

[0063] Secondly, this embodiment performs business bottleneck analysis based on global perception. By analyzing the global situation model S(t), potential business bottlenecks are identified. Data analysis and machine learning methods can be used to perform time series analysis, cluster analysis, and anomaly detection. The definition of the bottleneck identification model is shown in formula (13), where B(t) is the set of business bottlenecks identified at time t, is a specific bottleneck point, and k represents the number of identified bottleneck points. Further, bottleneck cause analysis is performed based on the identified bottlenecks and matched with the business process. C(t) represents the set of bottleneck causes analyzed at time t, and its definition is shown in formula (14). is a specific bottleneck cause, and l is the number of bottleneck causes.

[0064]

[0065] Finally, the generated optimization plan is evaluated and verified on the digital twin platform to ensure the effectiveness and feasibility of the plan. Simulation tools and evaluation indicators can be used, such as performance indicators such as throughput, waiting time, etc., as well as cost, quality, resource utilization, and energy consumption. The definition of the evaluation model is shown in formula (15), where E(π) is the set of evaluation results for the action plan π, is a specific evaluation indicator, and m is the number of evaluation indicators. After the evaluation results are obtained, they are fed back to the actual business system to guide actual operations and decision-making. Visualization tools and reports can be used to help managers understand the effects of the optimization plan. F(E(π)) represents the function of feedback of the evaluation results to the actual business system, which is defined as shown in formula (16), fn is a specific feedback information, and n represents the number of feedback information. Through the above process, a multi-agent task action planning and evaluation method based on digital twins can be realized, which can improve the operating efficiency and management level of the airport business system.

[0066]

[0067] In the specific implementation of step (4):

[0068] First, this embodiment uses a large language model to analyze the data and generate preliminary evaluation results. The evaluation results include the current business operation status, potential bottlenecks, preliminary optimization suggestions, etc. The evaluation result input model definition is shown in formula (17), is the preliminary evaluation result, LLM is the large language model, and D is the input data set.

[0069]

[0070] Secondly, this embodiment is based on the preliminary evaluation results , using a multi-agent system to generate optimization strategies. Each agent is responsible for a specific business system or functional module, and a global optimization strategy is generated through the interaction between agents. The optimization strategy generation model is defined as formula (18), where π is the optimization strategy and MAS is the multi-agent system.

[0071]

[0072] Next, this embodiment refines the generated strategy and verifies it on the simulation platform. The generated optimization strategy is refined and explained using a large language model to ensure the feasibility and operability of the strategy. The refined strategy is obtained using formula (19): , and input it into the simulation model constructed in the digital twin platform. The construction of the simulation model is expressed by formula (20), where M is the simulation model and DT is the digital twin platform. The simulation model M is run on the digital twin platform to simulate the implementation effect of the optimization strategy, as shown in formula (21), and the simulation result S is obtained. RUM represents the simulation operation function.

[0073]

[0074] Then, the simulation result S is compared with the physical process A comparative analysis is performed to verify the effectiveness of the strategy. The comparative analysis model is defined as shown in formula (22), where A represents the comparative analysis result, Compare is the comparative analysis function, S is the simulation result, Represents physical operation data.

[0075]

[0076] Iterate the above process and evaluate the optimal result. Adjust the optimization strategy based on the comparative analysis result A. , and re-simulate and compare and analyze until a satisfactory optimization effect is achieved. The definition of the iterative optimization model is shown in formula (23), where Iterate is the iterative optimization function. is the current optimization strategy. Compare and analyze with the physical airport operation process, and output the optimal result. The evaluation process is shown in formula (24), where O is the optimal result evaluation.

[0077]

[0078] Finally, the optimal result is evaluated Feedback is provided to decision makers to provide auxiliary decision support. Visualization tools and reports can be used to help managers understand the effects of optimization solutions. Through the above steps, a closed-loop decision-making action method based on large language models and digital twins can be implemented to improve the operating efficiency and management level of airport business systems.

[0079] In summary, this embodiment, based on digital twin technology, proposes a multi-agent construction solution that integrates business logic and data flow, breaking the existing data silos and business barriers from the top level. Two types of agents are constructed using the virtual modeling and business analysis capabilities of digital twins. One type is the existing actual business system agent, which implements the construction of a relationship network of airport global data by implanting business logic into multi-agents, solving the data silo problem of the existing business system; the other type is a virtual collaborative process agent, which is responsible for the autonomous regulation of the actual business system agent, solving the collaborative barrier problem of the existing business system.

[0080] This embodiment is based on big model technology and proposes a cooperative multi-agent autonomy mechanism. On the basis of analyzing the existing business operation rules of the airport, the transfer learning ability of the big model is used to define core actions such as perception, planning, action and memory for the multi-agents in the digital twin, allowing the agents to decompose complex tasks into simple sub-tasks like humans, solving the problems of insufficient autonomy and complex management processes in existing business systems.

[0081] This embodiment combines digital twin and large model technology to propose a closed-loop evaluation and decision-making support framework based on real-time feedback. By perceiving the actual operation status of the airport, it is positively input into the constructed digital twin world, and the constructed business system agent and virtual collaborative process agent are used to evaluate the current operation plan or bottleneck. The large model technology is used to coordinate the business system agent and the virtual agent, decompose and plan the complex tasks, and act reversely on the physical airport operation. Utilizing this forward and reverse closed-loop framework, the physical airport operation process is dynamically and real-time monitored and assisted, effectively improving the airport's operating efficiency and decision-making analysis accuracy.

[0082] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An airport digital twin application method based on a multi-agent model, characterized in that: The steps of the method include: Define real agent sets and virtual agent sets based on the digital twin platform; Define actions for the real agent set and the virtual agent set respectively, and concatenate the actions defined by the real agent set and the actions defined by the virtual agent set based on the large language model to form a collaborative autonomous agent set; Acquire airport operation data, conduct positive evaluation of the airport operation data based on a collection of collaborative autonomous agents, and build an optimization strategy through a large language model based on the positive evaluation results. The optimization strategy is then input back into the digital twin platform for verification, and interacts with the airport operation data in real time to complete the construction of the airport's multi-agent data twin closed-loop framework. The real agent set and the virtual agent set are defined based on the digital twin platform, which are specifically: The real agent set is defined as a set of physical systems of the airport, and the operation data of the airport is output based on the operation of the real agent set; The virtual agent set defines a simulated collaborative optimization agent set for associating and coordinating airport operation data; The specific formation process of the collaborative autonomous agent set is as follows: Data perception of a real set of intelligent agents and other sets of intelligent agents, where the other intelligent agents are specifically non-airport physical systems; Based on the set of virtual agents, define the action plan of each virtual agent in units; Execute action planning and memorize the perception, planning and action results of each agent, and connect the data perception, planning, action and memory actions in series based on the large language model to form the collaborative autonomous agent set; The positive evaluation of airport operation data based on a collaborative autonomous agent set is specifically as follows: A single collaborative autonomous agent obtains the corresponding N business node data to form a set of perception data of a single collaborative autonomous agent, and performs a union operation on each collaborative autonomous agent to form a global situation; Identify the airport's business bottlenecks based on the overall situation, analyze the causes of each bottleneck, and establish a corresponding relationship between the bottleneck and the cause; Defining evaluation indicators based on bottleneck points, determining evaluation dimensions based on causes, and forming an evaluation result set, wherein the evaluation result set is characterized as a positive evaluation result; The optimization strategy is constructed by combining the positive evaluation results with the large language model, which is specifically: Inputting the evaluation result set into the large language model to generate an analysis report, wherein the analysis report includes analysis results of the operating status, bottleneck points, and preliminary optimization suggestions; Generate optimization strategies based on analysis reports, segment optimization strategies through large language models, convert them into operation plans, and generate instructions that can be recognized by the digital twin platform; The optimization strategy is reversely input into the digital twin platform for verification, which is specifically as follows: Build a simulation model based on the digital twin platform, perform simulation runs on the optimization strategy, and collect simulation data; Compare the differences between the simulation data and the actual data, and iteratively update the optimization strategy based on the comparison results until the maximum number of iterations is reached, and then output the optimal optimization strategy.

2. The airport digital twin application method based on the multi-agent model according to claim 1 is characterized in that: The obtaining of airport operation data also includes cleaning and preprocessing the airport operation data to ensure data consistency.

3. The airport digital twin application method based on the multi-agent model according to claim 2 is characterized in that: The specific calculation formula for comparing the difference between the simulation data and the actual data is: Among them, A is the comparative analysis result, Compare is the comparative analysis function, S is the simulation data, For actual data.

Images