A museum event handling platform based on multi-agent

Abstract

The present invention discloses a museum event handling platform based on multiple agents, which relates to the field of artificial intelligence technology and solves the problems of complicated risk event handling procedures in museums, inability to properly resolve risk events, and repeated occurrence of risk events. The museum event handling platform includes an environmental monitoring agent, an event warning agent, a resource scheduling agent, an emergency response agent, and a collaborative rehearsal agent. Through the effective collaboration of multiple agents, the complicated and cumbersome museum risk event handling procedures are simplified, thereby shortening the handling time of museum risk events and improving the handling efficiency of museum risk events.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a museum event handling platform based on multi-agents. Background Art

[0002] As a strategic technology leading the new round of scientific and technological revolution and industrial transformation, artificial intelligence has increasingly achieved innovative applications across fields, tasks, and modalities. With the deepening integration of digital technology and museums, leveraging artificial intelligence to enhance the efficiency of museums' digital provision and achieve high-quality development has important strategic value and practical significance.

[0003] Digital museum provision is a new model and form that leverages the internet and digital technologies to reshape core museum functions such as collections, research, exhibitions, and education. It also creates innovative products and services that integrate online and offline museums, enhancing their preservation, management, and service capabilities. The current digital museum provision system can be summarized into three tiers.

[0004] First, digital technology is used to digitize the collection of cultural relics. This involves collecting, storing, analyzing, and processing digital information about the cultural relics, and accurately reproducing the form, structure, and decoration of the buildings, landscapes, and environments, thereby providing the necessary conditions for their restoration, reproduction, sharing, and reuse.

[0005] Second, integrate, develop, and transform museum database resources. Use virtual reality, 3D vision, holographic imaging, and other technologies to build a virtual roaming system. This system provides visitors with readily accessible and ready-to-use services such as remote access, online tours, voice guides, and location-based navigation via the internet and various smart terminals, creating diverse scenarios for museum digital exhibitions, research, and educational activities.

[0006] Third, the construction and system integration of a digital museum management platform. Leveraging technologies such as big data, artificial intelligence, the Internet of Things, blockchain, digital twins, AR / VR / MR, image recognition, and knowledge graphs, we will build a digital operation and management platform covering everything from conservation and restoration to exhibition and display, education, and communication. This will empower museums with an "intelligent brain" and transform museum management from "experience-driven" to "data-driven."

[0007] AI agents, with large language models at their core, possess the ability to autonomously understand, perceive, plan, remember, and use tools, enabling them to automate and complete complex tasks. The success of AI agents provides strong support for the development of swarm intelligence. Multiple AI agents can collaborate and complement each other to accomplish higher-level, complex tasks beyond the capabilities of a single agent.

[0008] Patent No. CN2022104095185 discloses a museum cultural relic management system based on big data analysis and a method for use thereof, the system including a management platform, the management platform including a core control module, a data display module, and a platform operation module; the above invention also proposes a method for use of a museum cultural relic management system based on big data analysis, comprising the following steps: S1, setting a vehicle barrier at the vehicle entrance and exit, S2, setting a face recognition camera in the museum's display area, storage area, and other cultural relic storage areas, S3, setting a monitoring center in the museum, and the monitoring center is equipped with a behavior analysis camera. The present invention is provided with an intelligent recognition module, so that people, wheels, and cultural relics entering and leaving the museum can be effectively identified, recorded, and monitored in real time, thereby ensuring the safety inside and outside the museum. In addition, an intelligent early warning system is provided, which can analyze and warn of events occurring in the museum, achieving the purpose of timely early warning, thereby ensuring the safety of cultural relics.

[0009] Patent No. CN2019102129931 discloses a Raspberry Pi-based museum tour guide robot and its use method. The Raspberry Pi-based museum tour guide robot includes a main control robot base, visual sensors, speakers, a Raspberry Pi, an LCD, and a humanoid shell. The use method of the Raspberry Pi-based museum tour guide robot includes: starting the system; initialization; controlling the robot to move and create a map of the museum's indoor environment; loading the created two-dimensional map in the 3D visualization tool RVIZ and setting multiple destinations on the LCD; combining global path planning and local path planning for navigation to complete the navigation guidance task; after arriving at the destination, the Raspberry Pi plays the relevant audio through the speakers to complete the visitor's explanation task. The present invention realizes autonomous navigation of the robot, strengthens human-computer interaction functions, improves the ability to respond to emergencies, and realizes guidance functions during the navigation process.

[0010] However, the above patents cannot provide museums with corresponding handling strategies for risk events. For example, there is often no system in the handling process of risk events in the same museum, which leads to cumbersome risk event handling process, inability to properly resolve risk events and repeated risk events. Summary of the Invention

[0011] The purpose of this invention is to provide a museum event handling platform based on multiple agents, which can simplify the cumbersome and complex museum risk event handling process through effective collaboration of multiple agents, thereby effectively shortening the handling time of museum risk events and greatly improving the handling efficiency of museum risk events.

[0012] The present invention utilizes the following technical solutions:

[0013] A museum event handling platform based on multiple agents, including environmental monitoring agents, event warning agents, resource scheduling agents, emergency response agents and collaborative rehearsal agents;

[0014] The environmental monitoring agent is used to build a digital virtual museum based on the museum's architecture and exhibits. It also collects the museum's internal and surrounding comprehensive data in real time to obtain comprehensive museum monitoring data, and combines this with pre-set anomaly determination rules to obtain anomaly datasets. Furthermore, a dynamic virtual museum is obtained based on the digital virtual museum and the anomaly datasets.

[0015] Among them, internal comprehensive data includes display environment data and visitor quantity and behavior data; display environment data includes temperature, humidity, air quality, light intensity, oxygen concentration and noise decibel value; visitor quantity and behavior data includes crowd density and visitor behavior; surrounding comprehensive data includes weather data, public activity data and natural disaster data;

[0016] The event warning agent is used to simulate the risk anomalies of the abnormal data set based on the museum monitoring data and the abnormal data set, and obtain risk warning reports and security control information;

[0017] The resource scheduling agent is used to schedule the physical museum's security resources based on the risk warning report, and to allocate security resources based on the security control information, and obtain the resource allocation table;

[0018] The emergency response agent is used to simulate risk events in the dynamic virtual museum based on the proposed resource allocation table and the abnormal data set using spatiotemporal data analysis algorithms or game consensus algorithms, and obtain a multimodal emergency plan;

[0019] The collaborative rehearsal agent is used to coordinate the museum risk events through the above four agents according to the multimodal emergency plan and obtain the event emergency rehearsal report.

[0020] Preferably, the environmental monitoring agent uses a three-dimensional modeling algorithm to construct a digital virtual museum based on the museum's architecture and exhibition information;

[0021] On the other hand, the environmental monitoring agent first uses wireless sensor networks, intelligent monitoring cameras, data analysis tools and remote sensing meteorological satellites to collect the museum's internal and surrounding comprehensive data, and records the data collection time; then the environmental monitoring agent uses a data segmentation algorithm to divide the internal and surrounding comprehensive data into several original data blocks according to the data collection time; the environmental monitoring agent then uses a data cleaning algorithm to deduplicate the information data in each original data block, verify data consistency and standardize data, and obtain several standard data blocks; then, the environmental monitoring agent uses a data classification algorithm to cluster the information data in each standard data block according to the data type, and obtain several aggregate data blocks; then, the environmental monitoring agent uses a data association algorithm to associate and normalize the aggregate data blocks of the internal environment and the surrounding environment according to the data collection time, and obtain the museum's comprehensive monitoring data; finally, combined with the preset anomaly judgment rules and using the data anomaly detection algorithm, the abnormal information data in the aggregate data block is aggregated into an anomaly data set, that is, an anomaly data set is obtained; at that time, the environmental monitoring agent uses a data fusion algorithm to integrate the anomaly data set into the digital virtual museum to obtain a dynamic virtual museum.

[0022] Preferably, the event warning agent first extracts the tourist quantity behavior data from the museum monitoring data, and uses the human body reconstruction algorithm in combination with the cultural relics display information to reshape all tourists within the range of X1 to X2 centimeters near each cultural relics exhibition stand according to the characteristics of tourist behavior to generate virtual characters, and generate virtual visitors; then the event warning agent uses the motion redirection algorithm in combination with the tourist behavior to generate a set of tourist motion vectors for the virtual visitors; then the event warning agent simulates various risk events of the digital virtual museum based on the virtual visitors and the tourist motion vector set, and at the same time combines the exhibition hall distribution information and the display environment data in the museum monitoring data to predict the probability of occurrence of risk events, and then generate a risk warning report; finally, the event warning agent simulates and upgrades the alarm level of the risk event according to the risk warning report, and generates corresponding safety control information for the museum as the alarm level changes.

[0023] Preferably, the resource scheduling intelligent agent first classifies and counts the security resources to obtain a security resource list; then adaptively allocates various types of resources in the security resource list according to the security control information to obtain a proposed resource allocation table; then uses the expert prior model to correct the proposed resource allocation table according to the characteristics of each control node in the security control information to obtain a node resource allocation list; at the same time, uses the path planning algorithm to arrange and plan the resource allocation order of each control node according to the node resource allocation list to obtain a resource allocation path; finally, uses the redundant fusion algorithm to fuse and judge the node resource allocation list and the resource allocation path to generate a proposed resource allocation table.

[0024] Preferably, the emergency response intelligent agent includes a data extraction unit, an event simulation unit and a strategy handling unit; the data extraction unit uses an information extraction algorithm combined with a risk warning report to extract data from the museum monitoring data table according to the alarm level and event type, and obtains key feature data of the event; the event simulation unit uses a generative adversarial network model to generate pending risk events of different alarm levels in the dynamic virtual museum according to the key feature data of the event; the strategy handling unit uses an expert prior model combined with the handling methods and handling basis of historical risk events, and uses a game consensus algorithm to combine the pending risk events with the resource allocation table and the abnormal behavior information table to generate a corresponding event handling report.

[0025] Preferably, the policy handling unit first uses the expert prior model combined with the feature extraction layer to extract the core elements of the historical risk event data, and generates an event core list in combination with the event sequence code; the core elements include event type, time and place, target object, alarm level, event cause, scale assessment, handling measures and responsibility determination; then the expert prior model uses the context word vector function combined with the first learning and training branch to convert each core element in the event core list into a corresponding word vector, and at the same time uses the polymorphic fitting function to construct a number of functional utility functions and a number of event handling strategy functions according to the event type and the functional department; then the policy handling unit uses the game consensus algorithm combined with the second learning and training branch to determine the policy plan of each functional department based on the resource allocation table and the abnormal behavior information table; then the update output layer is used to optimize the policy plan of each functional department according to the processing time of various historical risk events, combined with the functional utility function and the event handling strategy function, to obtain the optimal policy plan of each functional department; finally, the optimal policy plan is converted into consensus based on the policy consensus point to obtain the final policy plan.

[0026] Preferably, the feature extraction layer includes 3 5X5 convolutional layers with a step size of 2, 2 3X3 convolutional layers with a step size of 1, and 2 batch normalization layers; the first learning and training branch includes 2 InceptionV3 blocks of different depths, 3 residual blocks, 2 inverted residual blocks, 3 batch normalization layers, and a MISH activation function; the second learning and training branch includes 2 CBR blocks, 2 residual blocks, 2 InceptionV4 blocks, 3 Transformer encoding blocks, 1 batch normalization layer, and a Silu activation function; the CBR block includes 3 5X5 convolutional layers with a step size of 1, 4 batch normalization layers, and a Relu activation function; the update output layer includes 3 fully connected layers, 2 random dropout layers, and a cross entropy loss function.

[0027] Preferably, the emergency response intelligent body also includes a plan generation unit and an emergency assistance unit; the plan generation unit uses a multimodal large model to convert the final strategy plan into a multimodal emergency plan; the multimodal emergency plan includes a text plan, an image plan and a video plan; the emergency assistance unit performs real-time correction and supervision of the risk event emergency process based on the real-time situation of the risk event in combination with the multimodal emergency plan.

[0028] Preferably, the collaborative rehearsal agent includes a communication interaction unit and an information sharing unit; the communication interaction unit is responsible for communication operations between multiple agents using a communication protocol; the information sharing unit uses an information tracking algorithm to identify and record information between multiple agents and share the information.

[0029] Preferably, the collaborative rehearsal agent also includes a collaborative rehearsal unit and an integrated management unit; the collaborative rehearsal unit uses a rehearsal decision algorithm to design and implement cooperation strategies between agents based on a multimodal emergency plan, and ensures smooth collaborative work between agents; the integrated management unit uses a state monitoring algorithm to manage the operating state and collaborative relationship of the agents, and ensures effective collaboration between agents to generate event emergency rehearsal reports.

[0030] The present invention simplifies the cumbersome and complex museum risk event handling process through effective collaboration of multiple intelligent agents, thereby shortening the handling time of museum risk events and improving the handling efficiency of museum risk events. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of the museum incident handling platform.

[0032] Figure 2 Schematic diagram of the principle of emergency response intelligent agent. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0034] like Figures 1 to 2 As shown, the museum event handling platform based on multiple agents described in the present invention includes an environment monitoring agent, an event warning agent, a resource scheduling agent, an emergency handling agent and a collaborative rehearsal agent; wherein,

[0035] The environmental monitoring agent is used to build a digital virtual museum based on the museum's architecture and exhibits. It also collects the museum's internal and surrounding comprehensive data in real time to obtain comprehensive museum monitoring data, and combines this with pre-set anomaly determination rules to obtain anomaly datasets. Furthermore, a dynamic virtual museum is obtained based on the digital virtual museum and the anomaly datasets.

[0036] Among them, internal comprehensive data includes display environment data and visitor quantity and behavior data; display environment data includes temperature, humidity, air quality, light intensity, oxygen concentration and noise decibel value; visitor quantity and behavior data includes crowd density and visitor behavior; surrounding comprehensive data includes weather data, public activity data and natural disaster data;

[0037] The event warning agent is used to simulate the risk anomalies of the abnormal data set based on the museum monitoring data and the abnormal data set, and obtain risk warning reports and security control information;

[0038] The resource scheduling agent is used to schedule the physical museum's security resources based on the risk warning report, and to allocate security resources based on the security control information, and obtain the resource allocation table;

[0039] The emergency response agent is used to simulate risk events in the dynamic virtual museum based on the proposed resource allocation table and the abnormal data set using spatiotemporal data analysis algorithms or game consensus algorithms, and obtain a multimodal emergency plan;

[0040] The collaborative rehearsal agent is used to coordinate the museum risk events through the above four agents according to the multimodal emergency plan and obtain the event emergency rehearsal report.

[0041] In this embodiment, the interior of a museum includes the main building and its ancillary facilities (such as underground storage, office areas, equipment rooms, etc.), as well as all indoor spaces such as exhibition halls, corridors, staircases, elevators, restrooms, and rest areas. The perimeter of a museum includes areas outside the main building, including surrounding green spaces, squares, parking lots, roads, etc.; it also includes other buildings adjacent to the museum (such as commercial areas, residential areas, schools) and public facilities (such as subway stations and bus stops).

[0042] In this embodiment, the display environment data has an impact on the preservation of exhibits:

[0043] Temperature and humidity are the most common environmental factors affecting exhibits. They have a significant impact on certain exhibits, such as cultural relics, paintings, and paper, as well as display cases and showcases. Excessive temperature and humidity can lead to bacterial growth, decomposition, and increased oxidation reactions, causing exhibits to rot, damage, and deteriorate.

[0044] Poor air quality can cause dust and dirt to adhere to the surface of exhibits, damaging the appearance and quality of the exhibits and accelerating the aging and decay of the exhibits. Harmful gases in the air can also cause damage to the exhibits.

[0045] Excessive or insufficient light intensity will not only affect the appearance and quality of the exhibits, but may also cause the colors of the exhibits to fade and be damaged;

[0046] Too much or too little oxygen concentration will affect the quality and appearance of the exhibits. When displaying certain cultural relics, paintings and other cultural works of art, too much or too little oxygen is not allowed, otherwise it will accelerate the aging and decay of the exhibits.

[0047] Noise can directly generate mechanical vibrations and sound waves that can harm exhibits. Physical damage to some ceramics, glass, and paper products is related to noise, and angular fiber materials can also break easily when stimulated.

[0048] Visitor quantity and behavior data can easily lead to damage to exhibits or cause stampedes and congestion:

[0049] Excessive crowd density and improper walking patterns can easily lead to friction and collisions between people and objects. This is especially true in busy areas of the exhibition area, such as entrances and exits, and intersections. Special attention should be paid to the protection of exhibits.

[0050] When there is too much traffic, carbon dioxide, ammonia, and volatile organic compounds are produced, which can easily reduce the air quality in the exhibition area. Exhibition air management is an important task, otherwise the exhibits may oxidize, mold, and discolor.

[0051] When tourists knock on some electromechanical exhibits in the museum due to extreme haste or crowded conditions, it is easy to cause damage to the exhibits;

[0052] Due to the dense crowds of visitors, unexpected changes may occur in the lighting time, light intensity, air conditioning positioning, and humidity stability of the exhibits while creating the atmosphere of the event. These changes may cause a certain degree of damage to the exhibits and need to be addressed in a timely manner;

[0053] Some visitors may touch or fiddle with the exhibits, which may cause minor damage to them.

[0054] Security resources include fire extinguishers, fire hydrants, smoke alarms, access control equipment, inspection equipment, and explosion-proof equipment;

[0055] In the present invention, the environmental monitoring agent, on the one hand, uses a three-dimensional modeling algorithm to construct a digital virtual museum based on the museum's architecture and exhibition information;

[0056] On the other hand, the environmental monitoring agent first uses wireless sensor networks, intelligent monitoring cameras, data analysis tools and remote sensing meteorological satellites to collect the museum's internal and surrounding comprehensive data, and records the data collection time; then the environmental monitoring agent uses a data segmentation algorithm to divide the internal and surrounding comprehensive data into several original data blocks according to the data collection time; the environmental monitoring agent then uses a data cleaning algorithm to deduplicate the information data in each original data block, verify data consistency and standardize data, and obtain several standard data blocks; then, the environmental monitoring agent uses a data classification algorithm to cluster the information data in each standard data block according to the data type, and obtains several aggregate data blocks; then, the environmental monitoring agent uses a data association algorithm to associate and normalize the aggregate data blocks of the internal environment and the surrounding environment according to the data collection time, and obtains the museum's comprehensive monitoring data; finally, combined with the preset anomaly judgment rules and using the data anomaly detection algorithm, the abnormal information data in the aggregate data block is aggregated into an anomaly data set, that is, an anomaly data set is obtained; at that time, the environmental monitoring agent uses a data fusion algorithm to integrate the anomaly data set into the digital virtual museum to obtain a dynamic virtual museum;

[0057] In this embodiment, the wireless sensor network includes a temperature sensor, a humidity sensor, a light sensor, a gas sensor, a noise sensor, an infrared pyroelectric sensor, a fiber grating sensor, a geomagnetic sensor, and a tilt sensor;

[0058] The temperature sensor, humidity sensor, light sensor, gas sensor and noise sensor collect temperature, humidity, air quality, light intensity, oxygen concentration and noise decibel value respectively;

[0059] Infrared pyroelectric sensors sense the movement of dynamic heat sources by detecting changes in infrared radiation emitted by objects (human bodies, animals, etc.); fiber grating sensors are used to sense low-frequency vibrations caused by earthquakes or construction, triggering cultural relic reinforcement mechanisms; geomagnetic sensors are used to monitor the position and movement of metal cultural relics (bronze and iron artifacts) and trigger anti-theft alarms; and tilt sensors are used to monitor geological activity.

[0060] Weather is a significant factor influencing public travel. Sunny weather generally increases visitors' willingness to travel, which in turn boosts museum attendance, promotes cultural dissemination, and boosts economic benefits. Extreme weather (such as high temperatures, heavy rainfall, and lightning) can damage museum collections and buildings. For example, high temperatures can cause the aging of cultural relics, while heavy rainfall can cause leaks or flooding, threatening the building structure.

[0061] By analyzing public activity data (such as holiday schedules, school holidays, large-scale events, etc.), museums can predict peak visitor periods and thus optimize resource allocation, such as adding tour guides and extending opening hours. By analyzing public activity data (such as holiday schedules, school holidays, large-scale events, etc.), museums can predict peak visitor periods and thus optimize resource allocation, such as adding tour guides and extending opening hours. High-frequency public activities may accelerate the aging or damage of cultural relics, especially in the absence of effective protection measures.

[0062] By analyzing natural disaster data (such as floods, earthquakes, and mudslides), museums can develop targeted emergency response plans in advance to reduce disaster losses. For example, they can install flood control facilities and conduct earthquake reinforcement.

[0063] Natural disasters are one of the greatest threats facing museums. For example, floods can destroy ancient buildings and submerge collections, while earthquakes can cause building collapses or damage artifacts. Natural disasters can also cause museum closures, disrupting exhibitions and educational activities while increasing restoration and reconstruction costs.

[0064] Spatiotemporal data analysis algorithms, data segmentation algorithms, data cleaning algorithms, data classification algorithms, data association algorithms, data fusion algorithms and three-dimensional modeling algorithms are all commonly used technical means by personnel in this field and will not be elaborated here.

[0065] In the present invention, the event warning agent first extracts the visitor quantity and behavior data from the museum monitoring data, and uses the human body reconstruction algorithm in combination with the cultural relics display information to reshape all the visitors within the range of X1 to X2 centimeters near each cultural relics exhibition stand according to the characteristics of the visitors' behavior to generate virtual characters, thereby generating virtual visitors; then the event warning agent uses the action redirection algorithm in combination with the visitor behavior to generate a set of visitor action vectors for the virtual visitors; then, the event warning agent simulates various risk events of the digital virtual museum based on the virtual visitors and the visitor action vector set, and at the same time, combines the exhibition hall distribution information and the display environment data in the museum monitoring data to predict the probability of occurrence of risk events, and then generates a risk warning report; finally, the event warning agent simulates and upgrades the alarm level of the risk event according to the risk warning report, and generates corresponding security deployment information for the museum as the alarm level changes.

[0066] In this embodiment, risk events include vandalism and graffiti, fire and explosion, entry of prohibited items, stampede, natural disasters, and security failures.

[0067] Derivative simulation refers to the process of assuming the occurrence of a risk event and deducing its possible development process, impact scope and ultimate consequences during the risk assessment process. This method helps decision makers to more comprehensively understand potential risks and formulate effective response measures.

[0068] "Simulation Upscaling" is the process of incrementally enhancing or optimizing a system or process within a hypothetical or virtual scenario. During this process, the system's or process's performance, functionality, or capabilities are gradually improved as simulated conditions change, assessing its performance and response capabilities under different circumstances.

[0069] In the present invention, the resource scheduling intelligent body first classifies and counts the security resources to obtain a security resource list; then, according to the security control information, various types of resources in the security resource list are adaptively allocated to obtain a resource proposed allocation table; then, the expert prior model is used to modify the resource proposed allocation table according to the characteristics of each control node in the security control information to obtain a node resource allocation list; at the same time, a path planning algorithm is used to arrange and plan the resource allocation sequence of each control node according to the node resource allocation list to obtain a resource allocation path; finally, a redundant fusion algorithm is used to fuse and judge the node resource allocation list and the resource allocation path to generate a resource proposed allocation table.

[0070] In this embodiment, the expert prior model, the redundancy fusion algorithm, and the path planning algorithm are all commonly used technical means by those skilled in the art and will not be described in detail here.

[0071] In the present invention, the emergency response intelligent body includes a data extraction unit, an event simulation unit, a strategy handling unit, a plan generation unit and an emergency assistance unit; the data extraction unit uses an information extraction algorithm in combination with a risk warning report to extract data from the museum monitoring data according to the alarm level and event type, and obtains key feature data of the event; the event simulation unit uses a generative adversarial network model to generate risk events to be processed with different alarm levels in a dynamic virtual museum according to the key feature data of the event; the strategy handling unit uses an expert prior model in combination with the handling method and handling basis of historical risk events, and uses a game consensus algorithm to combine the risk events to be processed with the resource allocation table and the abnormal data set to generate a corresponding event processing report; the plan generation unit uses a multimodal large model to convert the final strategy plan into a multimodal emergency plan; the multimodal emergency plan includes a text plan, an image plan and a video plan; the emergency assistance unit uses the multimodal emergency plan to perform real-time correction and supervision on the emergency process of the risk event according to the real-time status of the risk event;

[0072] In this embodiment, the characteristics of the control nodes include the first-level risk area (core exhibition area), the second-level risk area (general exhibition hall area), and the third-level risk area (auxiliary facilities area). The generative adversarial network model, game consensus algorithm, and multimodal large model are all commonly used technical means in this field and will not be described in detail here.

[0073] A written emergency plan is an emergency plan that is primarily expressed in text and is used to describe in detail key information such as response measures, division of responsibilities, and resource allocation when an emergency occurs.

[0074] Graphical plans are emergency plans presented in the form of pictures, charts, etc., which convey information in a visual way to facilitate quick understanding and implementation;

[0075] The video plan is an emergency plan in the form of dynamic video, which displays emergency procedures and measures through simulated scenarios and demonstration operations.

[0076] In the present invention, the policy handling unit first uses the expert prior model combined with the feature extraction layer to extract the core elements of the historical risk event data, and generates an event core list in combination with the event sequence code; the core elements include event type, time and place, target object, alarm level, event cause, scale assessment, handling measures and responsibility determination; then the expert prior model uses the context word vector function combined with the first learning and training branch to convert each core element in the event core list into a corresponding word vector, and at the same time uses the polymorphic fitting function to construct a number of functional utility functions and a number of event handling strategy functions according to the event type and the functional department; then the policy handling unit uses the game consensus algorithm combined with the second learning and training branch to determine the policy plan of each functional department according to the resource allocation table and the abnormal data set; then the update output layer is used to optimize the policy plan of each functional department according to the processing time of various historical risk events, combined with the functional utility function and the event handling strategy function, to obtain the optimal policy plan of each functional department; finally, the optimal policy plan is converted into consensus according to the policy consensus point to obtain the final policy plan.

[0077] The feature extraction layer includes three 5X5 convolutional layers with a stride of 2, two 3X3 convolutional layers with a stride of 1, and two batch normalization layers; the first learning and training branch includes two InceptionV3 blocks of different depths, three residual blocks, two inverted residual blocks, three batch normalization layers, and a MISH activation function; the second learning and training branch includes two CBR blocks, two residual blocks, two InceptionV4 blocks, three Transformer encoding blocks, one batch normalization layer, and a Silu activation function; the CBR block includes three 5X5 convolutional layers with a stride of 1, four batch normalization layers, and a Relu activation function; the update output layer includes three fully connected layers, two random dropout layers, and a cross-entropy loss function;

[0078] In this embodiment, the working principle of the combination of the expert prior model and the feature extraction layer is as follows:

[0079] Preprocess historical risk event data, such as cleaning data, removing noise and standardizing, to ensure data quality; use the feature extraction layer to process historical risk event data; extract feature vectors that can represent the core elements of the event through a combination of convolutional layers and batch normalization layers; combine the prior knowledge in the expert prior model with the feature vectors extracted by the feature extraction layer: select the most representative features based on prior knowledge; weight different features in the feature vector according to prior knowledge to highlight important features and suppress minor features; input the feature vectors output by the expert prior model and the feature extraction layer into a joint model to comprehensively consider the prior knowledge and extracted features to generate an event core list; based on the fused feature vectors and prior knowledge, generate an event core list containing event type, time and place, target object, alarm level, event cause, scale assessment, disposal measures and responsibility determination; deduplicate, sort and format the generated event core list to ensure the accuracy and readability of the event core list.

[0080] In this embodiment, the working principle of the expert prior model using the context word vector function combined with the first learning training branch is:

[0081] The contextual word vector function is used to convert each core element in the event core list into a corresponding word vector. These word vectors serve as input to the first learning and training branch. The first learning and training branch receives the word vectors as input and performs feature extraction and fusion through network layers such as InceptionV3 blocks, residual blocks, and inverted residual blocks. These network layers capture high-level features in the word vectors and fuse them into more representative feature representations. Using a polymorphic fitting function, several functional utility functions and event handling strategy functions are constructed based on the event type and functional departments. These functions can perform intelligent analysis and decision-making on events based on the extracted and fused feature representations. Based on the constructed functional utility functions and event handling strategy functions, intelligent analysis and decision-making are performed on events, and corresponding handling strategies or recommendations are output.

[0082] In this embodiment, the working principle of the game consensus algorithm combined with the second learning and training branch is:

[0083] The historical data, current status, and proposed resource allocation tables for each functional department are collected. After preprocessing, this data serves as input to the second learning and training branch. Feature extraction and representation of the data are performed using CBR blocks, residual blocks, InceptionV4 blocks, and Transformer encoding blocks. These model blocks extract rich feature information from the data and convert it into a form suitable for subsequent processing. Based on the extracted feature information, the game consensus algorithm uses a game theory model to generate preliminary strategic plans for each functional department. These plans are then optimized using optimization algorithms (such as gradient descent) to improve their performance and stability. A consensus algorithm is used to ensure that each functional department can reach consensus on the final strategic plan.

[0084] In this embodiment, the working principle of updating the output layer is:

[0085] The feature representation extracted from the second learning training branch is input to the first fully connected layer of the update output layer. The fully connected layer performs a linear transformation on the input features and introduces nonlinearity through activation functions (such as ReLU). Subsequent fully connected layers further transform the features to gradually approach the optimal policy. A random dropout layer is inserted between fully connected layers to randomly discard the output of some neurons. This helps prevent the model from overfitting the training data during training and improves its generalization ability. The output of the last fully connected layer serves as the predicted policy. The cross-entropy loss function is used to calculate the difference between the predicted policy and the actual optimal policy. The predicted policy is then optimized through backpropagation to more closely match the actual optimal policy.

[0086] In the present invention, the collaborative rehearsal agent includes a communication interaction unit, an information sharing unit, a cooperative rehearsal unit and an integrated management unit; the cooperative rehearsal unit uses a rehearsal decision algorithm to design and implement a cooperative strategy between agents for museum risk events based on a multimodal emergency plan, and ensures that the collaborative work between agents proceeds smoothly; the integrated management unit uses a state monitoring algorithm to manage the operating state and collaborative relationship of the agents, and ensures effective collaboration between the agents and generates an event emergency rehearsal report; the communication interaction unit uses a communication protocol to be responsible for communication operations between multiple agents; the information sharing unit uses an information tracking algorithm to identify and record information between multiple agents, and share the information;

[0087] In this embodiment, the preview decision algorithm, state monitoring algorithm and information tracking algorithm are all commonly used technical means by people in this field, and will not be described in detail here.

[0088] Example:

[0089] The environmental monitoring agent uses wireless sensor networks (temperature sensors, humidity sensors, gas sensors, noise sensors, infrared pyroelectric sensors, fiber Bragg grating sensors, geomagnetic sensors, and tilt sensors) and smart surveillance cameras to collect comprehensive internal data of the museum (temperature, humidity, air quality, light intensity, noise decibel level, crowd density, visitor behavior, and oxygen concentration). It also uses data analysis tools and remote sensing meteorological satellites to collect comprehensive surrounding data (wind speed, rainfall, lightning warnings, public security incidents, and geological activities), and records the time of data collection.

[0090] The environmental monitoring agent uses a data segmentation algorithm to divide the internal comprehensive data and the surrounding comprehensive data into several original data blocks according to the data collection time; the environmental monitoring agent uses a data cleaning algorithm to deduplicate the information data in each original data block, verify the data consistency and standardize the data to obtain several standard data blocks; the environmental monitoring agent uses a data classification algorithm to cluster the information data in each standard data block according to the data type to obtain several aggregate data blocks; the environmental monitoring agent uses a data association algorithm to associate and normalize the aggregate data blocks of the internal environment and the surrounding environment according to the data collection time to obtain museum monitoring data; at the same time, the data anomaly detection algorithm is used to aggregate the abnormal information data in the aggregate data block into an abnormal data set; the environmental monitoring agent uses a three-dimensional modeling algorithm to build a digital virtual museum based on the museum building information and cultural relics display information, combined with the museum monitoring data;

[0091] The event warning agent first extracts visitor quantity and behavior data from the museum's monitoring data. Combining this with information about the cultural relics display, the agent uses a human body reconstruction algorithm to reshape all visitors within a range of X1 to X2 centimeters near each cultural relics display stand based on their behavioral characteristics, generating virtual visitors. The agent then uses a motion redirection algorithm to combine the visitor's behavior with the virtual visitors to generate a set of visitor motion vectors. The agent then simulates various risk events in the digital virtual museum based on the virtual visitors and the set of visitor motion vectors. Combining the exhibition hall distribution information with the display environment data from the museum's monitoring data, the agent predicts the probability of risk events (vandalism and graffiti, fire and explosion, contraband entry, natural disasters, and security failures) and generates a risk warning report. Finally, the agent simulates and increases the alert level of the risk event based on the risk warning report, and generates corresponding security control information for the museum as the alert level changes.

[0092] The resource scheduling agent first classifies and counts security resources (fire extinguishers, fire hydrants, smoke alarms, access control equipment, inspection equipment, and explosion-proof equipment) to obtain a security resource list. It then adaptively allocates the various resources in the security resource list based on the security control information to obtain a proposed resource allocation table. It then uses an expert prior model to modify the proposed resource allocation table based on the characteristics of each control node in the security control information to obtain a node resource allocation list. It also uses a path planning algorithm to arrange and plan the resource allocation order for each control node based on the node resource allocation list to obtain a resource allocation path. Finally, a redundancy fusion algorithm is used to fuse the node resource allocation list and the resource allocation path to generate a proposed resource allocation table.

[0093] The data extraction unit of the emergency response agent uses information extraction algorithms combined with risk warning reports to extract museum monitoring data based on the alarm level and event type, and obtains key feature data of the event; the event simulation unit uses a generative adversarial network model to generate pending risk events of different alarm levels in the virtual museum based on the key feature data of the event; the strategy handling unit uses an expert prior model combined with the handling methods and handling basis of historical risk events, and uses a game consensus algorithm to combine the pending risk events with the proposed resource allocation table and the abnormal data set to generate a corresponding event handling report; the plan generation unit uses a multimodal large model to convert the final strategy plan into a multimodal emergency plan; the multimodal emergency plan includes text plans, image plans and video plans; the emergency assistance unit uses the real-time status of the risk event in combination with the multimodal emergency plan to revise and supervise the emergency process of the risk event in real time;

[0094] Among them, the strategy handling unit first uses the expert prior model combined with the feature extraction layer to extract the core elements of the historical risk event data, and generates an event core list in combination with the event sequence code; the core elements include event type, time and place, target object, alarm level, event cause, scale assessment, handling measures and responsibility determination; then the expert prior model uses the context word vector function combined with the first learning and training branch to convert each core element in the event core list into a corresponding word vector, and at the same time uses the polymorphic fitting function to construct several functional utility functions and several event handling strategy functions according to the event type and functional departments; then the strategy handling unit uses the game consensus algorithm combined with the second learning and training branch to determine the strategy plan of each functional department based on the resource allocation table and the abnormal data set; then the update output layer is used to optimize the strategy plan of each functional department according to the processing time of various historical risk events, combined with the functional utility function and event handling strategy function, to obtain the optimal strategy plan of each functional department; finally, the optimal strategy plan is converted into consensus based on the strategy consensus point to obtain the final strategy plan;

[0095] At the same time, the communication interaction unit of the collaborative rehearsal agent uses the communication protocol to be responsible for the communication operations between multiple agents; the information sharing unit uses the information tracking algorithm to identify and record the information between multiple agents, and share the information; the collaborative rehearsal unit uses the rehearsal decision algorithm to design and implement the cooperation strategy between agents based on the multimodal emergency plan for museum risk events, and ensures the smooth collaboration between agents; the comprehensive management unit uses the status monitoring algorithm to manage the operating status and collaborative relationship of the agents, and ensures effective collaboration between agents to generate event emergency rehearsal reports.

Claims

1. A multi-agent-based museum event handling platform, characterized by: It includes environmental monitoring agents, event warning agents, resource scheduling agents, emergency response agents and collaborative rehearsal agents; among them, The environmental monitoring agent is used to build a digital virtual museum based on the museum's architecture and exhibits. It also collects the museum's internal and surrounding comprehensive data in real time to obtain comprehensive museum monitoring data, and combines this with pre-set anomaly determination rules to obtain anomaly datasets. Furthermore, a dynamic virtual museum is obtained based on the digital virtual museum and the anomaly datasets. Among them, internal comprehensive data includes display environment data and visitor quantity and behavior data; display environment data includes temperature, humidity, air quality, light intensity, oxygen concentration and noise decibel value; visitor quantity and behavior data includes crowd density and visitor behavior; surrounding comprehensive data includes weather data, public activity data and natural disaster data; The event warning agent is used to simulate the risk anomalies of the abnormal data set based on the museum monitoring data and the abnormal data set, and obtain risk warning reports and security control information; The resource scheduling agent is used to schedule the physical museum's security resources based on the risk warning report, and to allocate security resources based on the security control information, and obtain the resource allocation table; The emergency response agent is used to simulate risk events in the dynamic virtual museum based on the proposed resource allocation table and the abnormal data set using spatiotemporal data analysis algorithms or game consensus algorithms, and obtain a multimodal emergency plan; Among them, the emergency response agent includes a data extraction unit, an event simulation unit, and a strategy handling unit; the data extraction unit uses an information extraction algorithm combined with a risk warning report to extract data from the museum monitoring data table according to the alarm level and event type, and obtains key feature data of the event; the event simulation unit uses a generative adversarial network model to generate pending risk events of different alarm levels in the dynamic virtual museum based on the key feature data of the event; the strategy handling unit uses an expert prior model combined with the handling methods and handling basis of historical risk events, and uses a game consensus algorithm to combine the pending risk events with the resource allocation table and the abnormal behavior information table to generate a corresponding event handling report; The strategy handling unit first uses the expert prior model combined with the feature extraction layer to extract the core elements of the historical risk event data, and generates an event core list in combination with the event sequence code; then the expert prior model uses the context word vector function combined with the first learning and training branch to convert each core element in the event core list into a corresponding word vector, and uses the polymorphic fitting function to construct several functional utility functions and several event handling strategy functions according to the event type and functional departments; then the strategy handling unit uses the game consensus algorithm combined with the second learning and training branch to determine the strategy plan of each functional department based on the resource allocation table and the abnormal behavior information table; then, the update output layer is used to optimize the strategy plan of each functional department according to the processing time of various historical risk events, combined with the functional utility function and the event handling strategy function, to obtain the optimal strategy plan of each functional department; finally, the optimal strategy plan is converted into consensus based on the strategy consensus point to obtain the final strategy plan; The collaborative rehearsal agent is used to coordinate the museum risk events through the above four agents according to the multimodal emergency plan and obtain the event emergency rehearsal report.

2. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The environmental monitoring agent, on the one hand, uses the museum's architecture and exhibition information to build a digital virtual museum; on the other hand, the environmental monitoring agent first collects the museum's internal and surrounding comprehensive data and records the data collection time; The environmental monitoring agent then divides the internal integrated data and the surrounding integrated data into several raw data blocks according to the data collection time. The environmental monitoring agent then performs data deduplication, data consistency verification, and data standardization on the information data in each raw data block to obtain several standard data blocks. Subsequently, the environmental monitoring agent clusters the information data in each standard data block according to the data type and obtains several aggregate data blocks; after that, the environmental monitoring agent associates and normalizes the aggregate data blocks of the internal environment and the surrounding environment according to the data collection time to obtain the comprehensive monitoring data of the museum; finally, the abnormal information data in the aggregate data block is aggregated into an abnormal data set, that is, an abnormal data set is obtained; at that time, the environmental monitoring agent integrates the abnormal data set into the digital virtual museum to obtain a dynamic virtual museum.

3. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The event warning agent first extracts visitor quantity and behavior data from the museum's monitoring data. It then uses a human body reconstruction algorithm based on visitor behavior characteristics, combined with information about cultural relics displays, to reshape all visitors within a range of X1 to X2 centimeters near each cultural relic display stand into virtual characters, generating virtual visitors. The event warning agent then uses a motion redirection algorithm combined with visitor behavior to generate a set of visitor motion vectors. The event warning agent then simulates various risk events in the digital virtual museum based on the virtual visitors and the set of visitor motion vectors. It also predicts the probability of risk events by combining exhibition hall distribution information and display environment data from the museum's monitoring data, thereby generating a risk warning report. Finally, the event warning agent simulates and upgrades the alarm level of the risk event based on the risk warning report, and generates corresponding security control information for the museum as the alarm level changes.

4. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The resource scheduling agent first classifies and counts the security resources to obtain a security resource list; then, according to the security control information, it adaptively allocates the various resources in the security resource list to obtain a proposed resource allocation table; Subsequently, the expert prior model is used to modify the proposed resource allocation table according to the characteristics of each control node in the security control information to obtain the node resource allocation list; at the same time, the path planning algorithm is used to arrange and plan the resource allocation order of each control node according to the node resource allocation list to obtain the resource allocation path; finally, the redundant fusion algorithm is used to fuse and judge the node resource allocation list and the resource allocation path to generate the proposed resource allocation table.

5. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The feature extraction layer includes three 5X5 convolutional layers with a step size of 2, two 3X3 convolutional layers with a step size of 1, and two batch normalization layers; the first learning and training branch includes two InceptionV3 blocks of different depths, three residual blocks, two inverted residual blocks, three batch normalization layers, and a MISH activation function; the second learning and training branch includes two CBR blocks, two residual blocks, two InceptionV4 blocks, three Transformer encoding blocks, one batch normalization layer, and a Silu activation function; the CBR block includes three 5X5 convolutional layers with a step size of 1, four batch normalization layers, and a Relu activation function; the updated output layer includes three fully connected layers, two random dropout layers, and a cross-entropy loss function.

6. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The emergency response agent also includes a plan generation unit and an emergency assistance unit; the plan generation unit uses a multimodal large model to convert the final strategy plan into a multimodal emergency plan; the multimodal emergency plan includes a text plan, an image plan, and a video plan; The emergency assistance unit conducts real-time correction and supervision of the risk event emergency process based on the real-time status of the risk event and the multimodal emergency plan.

7. The multi-agent-based museum event handling platform according to claim 1, characterized in that: The collaborative rehearsal agent includes a communication interaction unit and an information sharing unit; the communication interaction unit uses a communication protocol to be responsible for communication operations between multiple agents; the information sharing unit uses an information tracking algorithm to identify and record information between multiple agents and share the information.

8. The multi-agent-based museum event handling platform according to claim 1 is characterized by: The collaborative rehearsal agent also includes a collaborative rehearsal unit and an integrated management unit; the collaborative rehearsal unit uses a rehearsal decision algorithm to design and implement cooperation strategies between agents based on a multimodal emergency plan, and ensures that the collaborative work between agents proceeds smoothly; the integrated management unit uses a status monitoring algorithm to manage the operating status and collaborative relationship of the agents, and ensures effective collaboration between the agents to generate event emergency rehearsal reports.

Images