Scenic area comprehensive management method and system based on workflow intelligent agent

Abstract

The invention discloses a scenic spot comprehensive management method and system based on a workflow intelligent agent, relates to the technical field of intelligent management, and aims to solve the problem that scenic spot data cannot be manually processed in time. Comprising the following steps: acquiring scenic region operation data and a historical scheduling scheme, and associating historical operation data with the historical scheduling scheme; storing the associated data into a knowledge base; the image analysis intelligent agent calculates historical tourist density, passenger flow volume and a current passenger flow travel route according to the security monitoring data of the associated data and the current operation data, and calculates predicted tourist density of a corresponding project on the travel route according to the passenger flow volume; the resource scheduling agent calls the large model to calculate the historical resource density and the current resource density of the corresponding project according to the scenic spot resource data; generating a scheduling scheme array in combination with a historical scheduling scheme; and screening a target scheduling scheme in the scheduling scheme array to generate a scheduling instruction. Comprehensive management of the scenic spot and comprehensive judgment and decision of data in the scenic spot are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology, and in particular to a comprehensive management method and system for scenic areas based on workflow intelligent agents. Background Technology

[0002] Currently, the information technology infrastructure for most scenic areas exhibits a "siloed" structure. Various business departments, such as ticketing, hotels, transportation, security, and retail, typically purchase and deploy their own management software independently. For example: The ticketing system only handles ticket sales and ticket checking; Hotel management systems (PMS) only manage room reservations and check-ins; The security system is an independent video surveillance network; The parking system is only responsible for vehicle entry and exit and billing; The use of multiple management software has resulted in the following core data issues: Severe data silos: Data standards vary across systems, resulting in fragmented data that makes correlation analysis impossible.

[0003] Delayed and experience-dependent management decisions: Due to the lack of real-time global data, scenic area management decisions (such as whether to increase the number of mini-train services or whether to send more security personnel to a certain hot spot) often rely on the personal experience of managers, resulting in untimely responses and potential waste of resources or inadequate service.

[0004] Therefore, in order to comprehensively improve the operational efficiency of the scenic area, the visitor service experience, and the data-driven decision-making capabilities, it is necessary to break down the data barriers between various business systems within the scenic area and build a central "brain" that can intelligently analyze and uniformly schedule global data. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide a comprehensive management method and system for scenic areas based on workflow intelligent agents.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A comprehensive scenic area management method based on workflow intelligent agents is provided, characterized by the following steps: Step S10: Obtain historical operation data and historical scheduling plans from the scenic area operation system. The obtained operation data includes security monitoring data and scenic area resource data. The obtained historical operation data is preprocessed, and the processed historical operation data is associated with the historical scheduling plans. The associated data is then stored in the knowledge base. Step S20: Obtain the current operation data of the scenic area operation system, preprocess the current operation data, store the preprocessed data in the knowledge base, and synchronously input it into the intelligent agent group, which includes an image analysis intelligent agent and a resource scheduling intelligent agent; Step S30: The image analysis agent calls the large model according to the first prompt word template, calculates the passenger flow and tourist route corresponding to the security monitoring data in the current operation data, and predicts the tourist density of each item on the tourist route at the next moment based on the passenger flow; obtains the historical security monitoring data of the knowledge base, calculates the historical tourist density of the corresponding item in the same period, and transmits the predicted tourist density and historical tourist density to the resource scheduling agent; Step S40: The resource scheduling agent calls the large model according to the second prompt word template to calculate the current resource density of the corresponding scenic area resource data in the current operation data; obtains the historical scenic area resource data of the knowledge base, calculates the historical resource density of the corresponding project in the same period, and calculates and generates a scheduling scheme array based on the historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling scheme. Step S50: Filter the target scheduling scheme from the scheduling scheme array, and generate a scheduling instruction according to the target scheduling scheme.

[0007] Compared with the prior art, the present invention has the following technical effects: This application preprocesses historical operational data, associates the processed data with historical scheduling schemes, and stores it in a knowledge base. It also preprocesses current operational data. The image analysis agent and resource scheduling agent within the intelligent agent group invoke a large model based on corresponding prompts. The image analysis agent uses the large model to predict the visitor density of attractions along the route after a certain period based on calculated visitor flow, and also calculates the historical visitor density of each attraction at the same historical moment. The resource scheduling agent uses the large model to calculate the historical and current resource densities of corresponding attractions, and generates a scheduling scheme array based on historical scheduling schemes. From this array, a target scheduling scheme is selected to generate scheduling instructions. Based on the fusion and association of different data from the scenic area, a reference scheduling scheme for management decisions is also provided, enabling comprehensive judgment and decision-making for complex and fuzzy scene data.

[0008] The present invention also provides a management system for a scenic area integrated management method based on workflow intelligent agents, including a data processing module, an intelligent agent group, and an integrated business scheduling and execution module; It includes a data processing module, an intelligent agent group, and a comprehensive business scheduling and execution module; The data processing module is used to acquire historical operation data and historical scheduling plans of the scenic area operation system. The acquired operation data includes security monitoring data and scenic area resource data. The module preprocesses the acquired historical operation data, associates the processed historical operation data with the historical scheduling plans, and stores the associated data in the knowledge base. It is also used to obtain the current operating data of the scenic area operation system, preprocess the current operating data, store the preprocessed data in the knowledge base, and synchronously input the intelligent agent group including image analysis intelligent agent and resource scheduling intelligent agent; The image analysis agent is used to call the large model based on the first prompt word template, calculate the passenger flow and tourist routes corresponding to the security monitoring data in the current operation data, and predict the tourist density of each project on the tourist route at the next moment based on the passenger flow; obtain historical security monitoring data from the knowledge base, calculate the historical tourist density of the corresponding project in the same period, and transmit the predicted tourist density and historical tourist density to the resource scheduling agent. The resource scheduling agent is used to call the large model based on the second prompt word template, calculate the current resource density of the corresponding scenic area resource data in the current operation data; obtain the historical scenic area resource data of the knowledge base, calculate the historical resource density of the corresponding project in the same period, and calculate and generate a scheduling scheme array based on the historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling scheme; The integrated business scheduling and execution module is used to filter target scheduling schemes from the scheduling scheme array and generate scheduling instructions based on the target scheduling schemes.

[0009] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a scenic area integrated management method based on a workflow intelligent agent.

[0010] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a comprehensive scenic area management method based on a workflow intelligent agent. Attached Figure Description

[0011] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a suitability flowchart for the integrated scenic area management method based on workflow intelligent agents in this embodiment of the invention. Figure 2 This is a diagram illustrating the architecture of the scenic area integrated management system based on workflow intelligent agents in this embodiment of the invention. Figure 3 This is a flowchart illustrating the workflow of the image analysis agent in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the workflow of the resource scheduling agent in an embodiment of the present invention. Figure 5 This is a flowchart of the marketing agent in an embodiment of the present invention. Figure 6 This is a flowchart illustrating the workflow of the tourist profile intelligent agent in an embodiment of the present invention. Figure 7 This is a flowchart illustrating the workflow of the device monitoring intelligent agent in an embodiment of the present invention.

[0012] Figure label: Business Subsystem Layer 100, Ticketing Subsystem 110, Accommodation Subsystem 120, Scenic Area Transportation Subsystem 130, Membership Subsystem 140, Parking Subsystem 150, Security Monitoring Subsystem 160, Retail and Catering Subsystem 170, Equipment Management Subsystem 180, Data Fusion Processing Center 200, AI Large Model Core Engine 300, Image Analysis Intelligent Agent 310, Marketing Intelligent Agent 320, Tourist Profiling Intelligent Agent 330, Equipment Monitoring Intelligent Agent 340, Resource Scheduling Intelligent Agent 350, Comprehensive Business Scheduling and Execution Module 400, Human-Computer Interaction Terminal 500, Tourist Mini Program 510, Comprehensive Management Platform 520. Detailed Implementation

[0013] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention.

[0014] Example 1: As Figures 1 to 7 As shown in the figure, the scenic area integrated management method based on workflow intelligent agents provided in this embodiment of the invention includes the following steps: Step S10: Obtain historical operation data and historical scheduling plans from the scenic area operation system. The obtained operation data includes security monitoring data and scenic area resource data. The obtained historical operation data is preprocessed, and the processed historical operation data is associated with the historical scheduling plans. The associated data is then stored in the knowledge base. Step S20: Obtain the current operation data of the scenic area operation system, preprocess the current operation data, store the preprocessed data in the knowledge base, and synchronously input it into the intelligent agent group, which includes image analysis intelligent agent and resource scheduling intelligent agent; Step S30: The image analysis agent calls the large model according to the first prompt word template, calculates the passenger flow and tourist route corresponding to the security monitoring data in the current operation data, and predicts the tourist density of each project on the tourist route at the next moment based on the passenger flow; obtains the historical security monitoring data of the knowledge base, calculates the historical tourist density of the corresponding project in the same period, and transmits the predicted tourist density and historical tourist density to the resource scheduling agent. Step S40: The resource scheduling agent calls the large model according to the second prompt word template to calculate the current resource density of the corresponding scenic area resource data in the current operation data; obtains the historical scenic area resource data of the knowledge base, calculates the historical resource density of the corresponding project in the same period, and calculates and generates a scheduling scheme array based on the historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling scheme. Step S50: Filter the target scheduling scheme in the scheduling scheme array and generate scheduling instructions according to the target scheduling scheme.

[0015] Furthermore, the scheme can be improved as follows: The generated scheduling scheme array also includes scheduling schemes within preset critical values ​​above and below the predicted density ratio. When screening target scheduling schemes, a first target scheduling scheme corresponding to the predicted density ratio can be selected from the scheduling scheme array, and released after the first target scheduling scheme is approved; alternatively, a decision scheme group can be generated based on the scheduling scheme array, and a second target scheduling scheme can be manually selected and released. By selecting target scheduling schemes from the scheduling scheme array through multiple methods, the feasibility of the target scheduling scheme is ensured.

[0016] In practice: The core idea of ​​the technical solution in this application is as follows: Unified access: Comprehensive collection and aggregation of data from all independent business subsystems within the scenic area (ticketing, accommodation, transportation, security, etc.).

[0017] Data fusion: Establish a data fusion processing center to clean, standardize, and correlate the collected multi-source heterogeneous data to form a global, unified, and real-time "digital twin of the scenic area".

[0018] To shield the differences in underlying devices and protocols during data collection, the system employs an adapter architecture to connect to various data sources within the business subsystem layer 100. Different subsystem types use corresponding access protocols. For example, data from operational subsystems such as ticketing subsystem 110, accommodation subsystem 120, membership subsystem 140, scenic area transportation subsystem 130, and retail and catering subsystem 170 are accessed via standard RESTful APIs, periodically pushing business messages in JSON format. The data between these subsystems is linked together using information such as member IDs and visitor ID cards. The security monitoring subsystem 160 transmits video streams via the RTSP protocol, performing target detection (such as people counting and personnel attribute recognition) at the edge, and only sending back structured XML feature data to reduce the load on the central network. Sensor data from the device management subsystem 180 is asynchronously reported via the MQTT protocol in a publish-subscribe pattern.

[0019] During data cleaning and standardization, all raw data streams are merged into a unified memory buffer for multi-level preprocessing. For sensor abrupt changes caused by environmental interference (such as instantaneous temperature anomalies), the system employs median filtering for smoothing and removes outliers that clearly exceed physical logic. Subsequently, a unified basic meta-model maps heterogeneous fields to GeoJSON spatial format and JSON-LD semantic format. For example, the number of people entering and exiting ticket gates and the crowd density data from the monitoring system are normalized into a unified "area occupancy rate" metric, laying the foundation for subsequent analysis.

[0020] To break down data silos between subsystems, the system introduces a "spatiotemporal identifier" as a global association primary key. Spatially, based on GIS coordinates, an R-tree index is used to map physical devices (such as cameras, turnstiles, and distribution boxes) to digital twin entities. Temporally, the NTP protocol is used to control the clock synchronization error of each system to within 100 milliseconds, and the Kafka stream processing engine is used to perform window joins on multi-source data within the same time window.

[0021] For example, when the ticketing system for area A records 100 people entering the park at time t, and the monitoring data for that area shows a 15% increase in visitor density, the system will automatically trigger the associated logic and update the status bit of "area A congestion level" in the digital twin in real time.

[0022] The cleaned and correlated standardized data is pushed to the human-computer interaction terminal (500) in real time via a WebSocket bidirectional channel. The front end adopts TwinGraph technology to dynamically bind business data to the corresponding nodes of the three-dimensional model, so that the integrated management platform (520) can intuitively and in real time reflect the running status of the underlying business subsystem layer 100.

[0023] Intelligent Decision-Making: A large-scale AI model is introduced as the core engine (brain) of the system. This engine performs deep learning and analysis based on fused global data, enabling proactive customer flow prediction, risk warning, resource scheduling optimization, and personalized marketing recommendations.

[0024] Closed-loop execution: The decision results of the AI ​​big model are transformed into specific and executable instructions, which are automatically sent to the corresponding business subsystems for execution, and the execution effect is continuously monitored, forming an intelligent management closed loop of "perception-analysis-decision-execution-feedback".

[0025] like Figure 2 As shown, the scenic area integrated management system of this application can be divided into five layers from bottom to top: business subsystem layer 100, data fusion processing center 200, AI large model core engine 300, comprehensive business scheduling and execution module 400, and human-computer interaction terminal 500. Among them, the business subsystem layer 100 includes multiple independent business subsystems such as ticketing subsystem 110, accommodation subsystem 120, scenic area transportation subsystem 130, membership subsystem 140, parking subsystem 150, security monitoring subsystem 160, retail and catering subsystem 170, and equipment management subsystem 180, serving as the data source layer and providing basic data; the data fusion processing center 200 is responsible for collecting, cleaning, standardizing, and correlating the multi-source heterogeneous data generated by the business subsystem layer, and is the data hub for realizing the digital twin; the AI ​​large model core engine 300 is the "brain" layer of the system. The system comprises multiple agents, including an image analysis agent 310, a marketing agent 320, a tourist profiling agent 330, an equipment monitoring agent 340, and a resource scheduling agent 350, responsible for in-depth analysis and decision-making based on fused data. The integrated business scheduling and execution module 400 serves as the logic delivery layer, responsible for transforming the decision instructions generated by the AI ​​engine into executable business logic and coordinating the linkage between various subsystems. The human-computer interaction terminal 500 serves as the application presentation layer, providing tourists and managers with a visual interface and interactive interface through the tourist mini-program 510 and the integrated management platform 520. The tourist mini-program 510 corresponds to tourist ticketing terminals or other mobile terminals; the integrated management platform 520 corresponds to terminal devices that can receive specific instructions, such as driver apps (handheld devices), equipment engineer handheld devices, restaurant equipment, vending point equipment, and security personnel handheld devices. Figures 3 to 7 Workflow diagrams for the 300 intelligent agents within the core engine of the AI ​​large model.

[0026] Business Subsystem Layer 100: This is the data source layer, containing various information systems within the scenic area and providing basic data. Multiple corresponding business subsystems are set up within Business Subsystem Layer 100.

[0027] Ticketing Subsystem 110: Provides data on tourist ticket purchases, reservations, and park entry, and also includes information on children's ticket purchases.

[0028] Accommodation Subsystem 120: Provides guest and room status data, as well as guest image information collected during check-in.

[0029] Scenic Area Transportation Subsystem 130: Provides information such as vehicle usage, location data, and quantity.

[0030] Membership Subsystem 140: Provides visitor profiles and loyalty data.

[0031] Parking Subsystem 150: Provides vehicle traffic data.

[0032] Security monitoring subsystem 160: Provides real-time video stream data.

[0033] Retail and Catering Subsystem 170: Provides consumer behavior data, canteen and retail outlet product data, etc.

[0034] Equipment Management Subsystem 180: Provides status data of facilities and environment such as water, electricity, gas, and ecological environment through IoT sensors.

[0035] Data Fusion Processing Center 200: The "processing plant" for data. It collects data from various subsystems in the business subsystem layer 100 through APIs, direct database connections, and other methods. It cleans (removes invalid data), transforms (unifies the format), and loads (stores in the data warehouse) the large amount of collected data. Most importantly, it associates data through data characteristics, such as linking the same person's mobile phone number, license plate number, and member ID based on ID card number or ticket number.

[0036] The ticketing subsystem 110 retrieves the following data: mobile phone number, ID card number, date of entry, number of tickets, ticket type, ticket number, and license plate number. The accommodation subsystem 120 retrieves the following data: mobile phone number, ticket number, ID card number, room type, number of people, facial image, check-in time, and check-out time. Parking subsystem 150 acquires the following data: license plate number, entry time, and exit time; When tourists place an order for tickets through the Tourist Mini Program 510, they are prompted to enter their license plate number to receive a free parking voucher. The Tourist Mini Program 510 corresponds to the tourist's mobile device.

[0037] Merged JSON data { Visitor Basic Information:{ ID number: "110101199001011234" Mobile number: "13800138000" "User Portrait Tags": "Self-driving tour, family with children, high net worth" / / Optional: Generated by the large model based on historical data }, "Ticket Information": { "Order Number": "T202310010056", "游玩日期": "2023-10-01", "Purchase Time": "2023-09-28 10:30:00", "Linked License Plate Number": "京A88888", / / Key field: From the entry in the mini-program "Ticket Details": { "Ticket Number": "TICKET_001", "Ticket Type": "Adult All-access Pass", "Verification Status": "已入园", "Verification Time": "2023-10-01 09:15:00" }, { "Ticket Number": "TICKET_002", "Ticket Type": "Child Discount Ticket", "Verification Status": "已入园", "Verification Time": "2023-10-01 09:15:05" } }, "Parking Information": { "License Plate Number": "京A88888", "Entry Time": "2023-10-01 08:50:00", "Exit Time": "2023-10-01 17:30:00", / / If not exited, it is empty "Parking Duration": "8.6 hours", "Coupon Status": "已核销" / / Linked to the free coupon received from the mini-program }, "Accommodation Information": { "Hotel Name": "Scenic Area Resort Hotel", "Room Type": "Deluxe Lake-view King-size Bed Room", "Room Number": "8205", "Check-in Time": "2023-10-01 14:00:00", "Check-out Time": "2023-10-02 12:00:00",​ "Number of occupants": 3, "Facial feature ID": "FACE_UUID_123456", / / Privacy desensitization processing, only the ID is retained "Associated ticket number": "TICKET_001" / / Association key between the accommodation system and the ticketing system } } To generate the above JSON, the backend system (Data Fusion Processing Center 200) needs to perform the following fusion steps: Step 1: Anchor the main body Trigger condition: When a tourist swipes their ID card / face at the ticket checkpoint or checks in at the hotel.

[0038] Core action: Using the ID number as the unique primary key (PrimaryKey), pull the basic data under this ID from the [Ticketing Subsystem] and [Accommodation Subsystem].

[0039] Step 2: Associate the vehicle This is the most crucial step to solve the data island problem of "separation of people and vehicles".

[0040] Association logic: Read the license plate number entered by the user in the [Ticketing Subsystem / Tourist Mini Program] (e.g., Beijing A88888).

[0041] Take Beijing A88888 to the [Parking Subsystem] to query the entry and exit records for the day.

[0042] Judgment: If there is a record of this license plate number in the parking system, merge the parking data into the tourist's JSON package; if not (e.g., the tourist came by bus but entered the license plate number), the "Parking information" field is empty or shows "Not entered".

[0043] Step 3: Associate the accommodation Association logic: Method A (strong association): Directly query the [Accommodation Subsystem] through the ID number to check if there is an in-house record for the day.

[0044] Method B (weak association): If the ID is not registered (such as a companion), reverse query can be performed through the mobile phone number or ticket number (if it is a scenic hotel package).

[0045] AI large model core engine 300: The "brain" of the system. It is a large language model or multi-modal model fine-tuned with tourism industry data. It receives the processed global data and performs complex analysis tasks, such as: Time series prediction: Predict future passenger flow and vehicle flow.

[0046] Tourist flow forecast for attractions: Based on real-time visitor numbers, visitor movement patterns within the park (heat map), maximum capacity of attractions, and performance schedules, visitor flow for specific attractions can be predicted minute by minute.

[0047] Assuming the current time is 11:00 AM, a large show currently being held at attraction A02 will end at 11:30 AM, and historical data shows that after the show, approximately 40% of visitors will head to the nearest popular attraction B02.

[0048] Analysis and prediction indicate that although the current queue time for the B02 attraction is only 15 minutes, a sudden surge in visitor flow (a pulse-like peak) is expected between 11:40 and 12:30, with the queue time anticipated to increase to over 90 minutes. Based on this, the system suggests sending a "staggered visit" notification to nearby tourists at 11:25, guiding them to the less crowded B01 attraction.

[0049] Parking lot and surrounding traffic flow forecast: Based on historical departure patterns, surrounding road congestion indices, current number of people remaining in the park, and weather conditions, traffic pressure is predicted for the next few hours.

[0050] Assuming it is a Sunday, the weather suddenly turns cloudy at 16:00 with a forecast of thunderstorms, and there are still 15,000 visitors in the park at that time.

[0051] Based on historical data analysis of "accelerated departures due to severe weather," the system predicts that the usual peak departure time, originally scheduled for 7:00 PM, will occur earlier, between 4:30 PM and 5:30 PM. The parking lot exit is expected to reach its peak congestion index at 5:00 PM. Based on this, the system recommends that traffic management departments adjust traffic light timings in advance and sends guidance to visitors suggesting that "roads are slippery due to rain; it is recommended to postpone departure by half an hour or utilize indoor dining facilities."

[0052] Image recognition: Analyze surveillance video to identify congestion and abnormal behavior.

[0053] Natural Language Processing: Analyzing tourists' ratings and feedback on the mini-program.

[0054] Furthermore, the AI ​​large-scale model core engine 300 contains several intelligent agents that support business processing. These agents can be horizontally expanded according to business needs. Existing intelligent agents include: Image analysis agent 310: Analyzes acquired images in real time and provides structured analysis results, processing no less than 500 images per second (image resolution 720p, jpeg format); structured results include, but are not limited to, tourists' direction of travel, tourist density, etc. Marketing Agent 320: Based on the input real-time tourist data, including tourist type, tourist location, historical profile data, etc., it accurately pushes marketing data to tourists; Tourist profiling agent 330: Based on real-time tourist data, it updates tourist profile information in real time and provides basic tourist data support for marketing agent 320; Equipment Monitoring Intelligent Agent 340: It analyzes the monitoring data of equipment in the park in real time, and combines historical faults and equipment maintenance knowledge base to provide accurate and timely early warning of equipment faults; Resource Scheduling Intelligent Agent 350: Based on the data input from the business system and the current allocation of scenic area resources, it automatically and dynamically schedules scenic area resources (mobile vending vehicles, security, public transportation, etc.) to ensure the stable operation of the scenic area.

[0055] Each agent invokes the large model to perform corresponding operations and data processing through the corresponding prompt word template.

[0056] When the resource scheduling intelligent agent 350 dynamically schedules resources such as mobile vending vehicles, security, and public transportation, it performs the following operations.

[0057] 1. Mobile sales resource scheduling Objective: To provide ample supplies in high-traffic areas, reduce the time tourists spend searching for goods, and increase secondary sales revenue.

[0058] Data input: Real-time passenger flow heat map: Identifies areas of current crowd gathering through camera or base station data. Camera data is collected by the 160 security monitoring subsystem, which collects visitor data.

[0059] Inventory and sales data: Real-time transaction frequency and inventory levels at each sales point's POS machine. Data uploaded by the Retail and Catering subsystem 170.

[0060] Time-series prediction results: Predict the direction of passenger flow in the next hour (e.g., the flow after a performance ends). Performed by image analysis agent 310.

[0061] Scheduling logic: Inventory Warning: When the inventory of a fixed stall falls below 20% and customer traffic shows an upward trend, the system automatically determines that replenishment or reinforcement is needed. The resource scheduling agent 350 analyzes and judges the situation, generating corresponding scheduling instructions.

[0062] The "follow the crowd" strategy: The system detected an abnormal increase in popularity in the "family playground" area between 2:00 PM and 3:00 PM, while the queue length at the fixed restaurants in this area exceeded 15 minutes, indicating an overflow of purchasing power.

[0063] Operational optimization: Calculate the mobile vending cart closest to the hotspot that is currently in an "inefficient zone" (low customer traffic, low sales). The resource scheduling agent 350 analyzes and judges the data, generating corresponding scheduling instructions.

[0064] Perform the following actions: Instruction issuance: Send the instruction to the handheld terminal of the driver of the mobile vending vehicle with the serial number Cart-05: "Please proceed to the entrance of the C-zone Family Plaza. Service is expected to take 1 hour." The integrated business scheduling and execution module 400 then sends the instruction to the human-computer interaction terminal 500 based on the scheduling instruction.

[0065] Logistics coordination: Simultaneously, the nearest warehouse delivery personnel are notified to pre-deliver supply boxes (water, ice cream, and other high-frequency consumables) to the target location of Cart-05. Similarly, the integrated business scheduling and execution module 400 sends instructions to the human-machine interface terminal 500 based on the scheduling commands.

[0066] 2. Dynamic scheduling of security resources Objective: To intervene before congestion occurs, prevent stampedes, and respond quickly to emergencies.

[0067] Data input: Grid density monitoring: The scenic area is divided into several grids, and the number of people per square meter is calculated in real time. The security monitoring subsystem 150 collects tourist image / video data, which is then used by the image analysis agent 310 to call the large model for calculation.

[0068] Abnormal event recognition: The image analysis agent 310 identifies behaviors such as "fighting disputes", "tourists falling" or "going against the flow of traffic" through video analysis.

[0069] Key node thresholds: pass rate data for the gate entrance and restaurant exit.

[0070] Scheduling logic: Peak Shaving: A performance event is predicted to pass through the main road in 10 minutes, with an estimated gathering size exceeding 3000 people, reaching a yellow risk level. This part is executed by resource scheduling agent 350.

[0071] Dynamic grid deployment: The system determines that the current security personnel configuration on the main road is insufficient (only 2 people), and it is necessary to draw security forces from the surrounding "low-risk areas" (such as the currently idle B01 project). This part is executed by resource scheduling agent 350.

[0072] Perform the following actions: Task assignment: Send an emergency dispatch order to 6 security personnel within a 500-meter radius: "Please proceed immediately to section B of the main road to form a human wall for crowd control." This part is executed by resource scheduling agent 350.

[0073] Equipment linkage: The turnstiles in this area are automatically controlled to slow down the passage speed, and the electronic screen and broadcast system automatically play the guidance voice message "Please watch your step and do not crowd".

[0074] 3. Public transportation scheduling in the scenic area Objective: To reduce tourists' waiting time for transportation, improve traffic flow efficiency, and avoid idle capacity.

[0075] Data input: Queue length at stations: The length of queues and estimated waiting time identified by cameras at each small train station.

[0076] Vehicle location and status: GPS positioning, vehicle occupancy rate, remaining battery / fuel level.

[0077] Entry / Departure Tidal Data: Whether the current time is dominated by "entry" or "departure".

[0078] Scheduling logic: Tidal flow lane / route switching: At 17:00, the demand for leaving the park surged, and the queue at the main gate exit overflowed. The system determined that the demand for entering the park was almost zero.

[0079] ExpressMode: The system detects that there is no one waiting at the intermediate station A03, while the terminal station "Parking Station" has huge demand.

[0080] Departure interval optimization: The regular 15-minute interval will be dynamically adjusted to "depart when full" or shortened to 5 minutes.

[0081] Perform the following actions: Schedule adjustment: Send the instruction to all empty shuttle buses: "Switch to the [Leaving the Park Shuttle] mode, pick up passengers only at the 'Core Scenic Area' and go directly to the 'Parking Lot' without stopping in between." Information dissemination: The electronic bus stop sign will be updated in real time: "Due to the peak departure time, additional direct buses will be added. The next bus is expected to arrive in 2 minutes." Decision generation: Resource scheduling agent 350 analyzes data and outputs optimal operational strategy recommendations (e.g., recommending that the number of shared bicycles near attraction A be increased by 50 during the period from 14:00 to 15:00). Structured data processed by image analysis agent 310 is also used in public transportation scheduling.

[0082] The Integrated Business Scheduling and Execution Module 400 acts as the "nerve center" and "command translator" of the brain. It translates the macro-level strategies output by the AI ​​large-scale model's core engine 300 into specific instructions that the underlying business subsystems can understand, and distributes them via API calls. For example, the system has a built-in "shared bicycle scheduling" strategy. If F(tourist density, shared bicycle density) > 1.5 (F represents the resource scheduling agent's calculation process; the agent inputs the tourist density and resource density of the scenic spot into the large model, which calculates and outputs a resource density value. Resource density represents the correspondence between resource supply and demand; the system generates specific scheduling instructions based on the resource density value), then the system outputs the strategy of "increasing shared bicycles." The scheduling system translates this instruction into an API request to the scenic area's transportation subsystem to "create a new scheduling task."

[0083] Large model data processing and scheduling are also applicable to several different application scenarios.

[0084] Scenario 1: Intelligent passenger flow management and scheduling during peak holiday periods (S1) Data Acquisition and Fusion: Data fusion is the process of integrating fragmented data scattered across various business subsystems into a complete and meaningful dataset through data feature fields. For example, by using ID numbers from different business subsystems, data from ticketing, accommodation, and catering / retail subsystems can be linked to the same tourist, creating a tourist profile.

[0085] Case study of data preparation during the forecasting phase: One week before a holiday, the system retrieved pre-sale ticket data from the ticketing subsystem 110 (exceeding 30% of the same period last year) and obtained a "sunny" weather forecast for the holiday period from an external API. Simultaneously, the system analyzed social media data obtained through the API interface and discovered extremely high levels of discussion recently regarding a newly opened attraction within the scenic area. This data serves as input for the predictive intelligent agent equipment.

[0086] Real-time data preparation case study: On the day of the holiday, parking subsystem 150 data showed that the flow of vehicles entering the park was twice as fast as usual. Ticketing subsystem 110's real-time visitor numbers had already reached the level of the entire day on weekdays by 9:00 AM. Security monitoring subsystem 160's cameras began collecting video streams of pedestrian density at major intersections and attractions. This prepares input data for the real-time analysis agent.

[0087] (S2) Analysis and Decision-Making: The AI ​​big data model core engine 300, integrating the above multi-source data, discovered that congestion occurred at the parking lot entrance at 9:30, and the number of vacant parking spaces in the parking lot was less than 20%. The predictive agent, combining historical data, predicted that tourists would arrive at the entrance for ticket inspection and entry 10-30 minutes after parking, and immediately made a judgment: extreme congestion is expected at 10:30 am at entrance A and the newly opened attraction B.

[0088] The model immediately generates a three-dimensional drainage strategy: Entrance diversion strategy: It is recommended to temporarily close some ticket windows at entrance A and use electronic guide signs and the 510 mini-program to push messages to guide self-driving tourists to park and enter the park at entrance C, where there are fewer vehicles.

[0089] Proactive traffic scheduling strategy: Predicting a significant increase in traffic demand from entrance C to the core area, an instruction is generated to dispatch 80% of the shared bicycles within the scenic area to entrance C.

[0090] Tourist attraction crowd control and recommendation strategy: When the queue time at attraction B exceeds 30 minutes, the virtual queuing system will be automatically triggered, and tourists on their way to attraction B will be pushed "limited-time tea and snack coupons for nearby attraction D (currently less crowded, with similar scenery)" to disperse the flow of people.

[0091] Security personnel dispatch strategy: The two warning zones, A and B, are highlighted on the integrated management platform 520, and instructions are automatically pushed to the handheld terminals of security personnel in the zones, requiring them to strengthen patrols and maintain order.

[0092] (S3) Scheduling Execution and Feedback: The integrated business scheduling and execution module 400 decomposes the above strategies into specific instructions and issues them.

[0093] The ticketing subsystem 110 adjusts the window status; the parking subsystem 150 updates the signage information; the tourist mini-program 510 pushes traffic diversion messages; the scenic area transportation subsystem 130 updates the scheduling tasks; and the membership subsystem 140 issues coupons.

[0094] The system continuously monitors the execution effect: if the flow density of people at attraction B decreases, the strategy is successful; if it does not decrease, the model will generate a backup plan, such as enabling a one-way passage route at attraction B.

[0095] Scenario 2: Personalized services for families with children (S1) User Profile Construction: A user purchased two adult tickets and one child ticket through the tourist mini-program 510 and registered as a member. The membership subsystem 140 then tagged the user with "family trip" and "parent-child".

[0096] After entering the park, the family rented a four-seater electric vehicle (recorded by the park's traffic subsystem 130) and stayed in the "Children's Playground" area for more than 1 hour (detected by GPS and electronic fence technology).

[0097] (S2) Intelligent Recommendation and Services: At 11:30 a.m., the AI ​​big data model core engine 300 analyzed the family's location, tags, and time to determine that they had a need for dining and preferred restaurants suitable for children.

[0098] The model generates personalized restaurant recommendation strategies: a message is pushed to the tourist mini-program 510: "Hello! Nearby restaurants have children's meals with no queue. Show this message to enjoy a 20% discount." The message also includes a one-click navigation function.

[0099] In the afternoon, when the model detected that the family was approaching the souvenir shop, it generated another recommendation: "Our store has just received a new 'Cartoon Monkey' series of toys, your child might love them!" (S3) Service Closed Loop: When a user clicks the navigation to go to the restaurant, the system records this action as positive feedback, reinforcing the user's interest in the "family-themed" category.

[0100] If a user makes a purchase at a restaurant, the data from the retail and catering subsystem 170 will further refine the user profile, providing a more accurate basis for future recommendations.

[0101] Scenario 3: Predictive Maintenance for Equipment Failure (S1) Data Monitoring and Modeling: The equipment management subsystem 180's IoT sensors continuously monitor the current, vibration frequency, and temperature data of a large fountain pump in the core area of ​​the scenic spot.

[0102] The AI ​​Big Data Model Core Engine 300 used months of historical data to learn and build a baseline of the normal operating mode of this water pump under different seasons and different operating durations.

[0103] (S2) Anomaly Detection and Early Warning: One afternoon, the system detected that the water pump's current data was within the normal range, but its vibration frequency began to show a slight but continuous abnormal increase that did not conform to historical patterns.

[0104] The equipment monitoring agent 340 judged that this might be a precursor to bearing wear or foreign objects inside. Although it has not yet caused a failure, the risk of failure has been significantly increased.

[0105] The model immediately generates a "predictive maintenance alert".

[0106] (S3) Work Order Dispatch and Processing: The integrated business scheduling and execution module 400 converts the early warning information into a maintenance work order, which is automatically dispatched to engineers in the equipment maintenance department. The work order includes the specific location of the faulty equipment, abnormal data charts, and possible causes diagnosed by AI.

[0107] Engineers inspected the water pump at night when there were fewer tourists and found that the bearings were indeed worn prematurely. They replaced them in time, avoiding the negative impact of a sudden shutdown during peak business hours.

[0108] AI Model Selection: The AI ​​Large Model Core Engine 300 can be fine-tuned using industry-standard basic models (such as the GPT series, Wenxin Yiyan, etc.) or it can be a combination of multiple specialized models (such as a visual large model + a language large model).

[0109] Fine-tuning the base model includes steps such as data preparation and preprocessing, selection of fine-tuning algorithms, and setting of training hyperparameters. Data preparation and preprocessing begins with collecting historical operational data from the scenic area, cleaning the data, using regular expressions for data anonymization, and removing duplicate content. For fields involving personal identification or sensitive information, masking or generalization strategies are employed to ensure the entire process complies with data security and privacy protection requirements. The cleaned, high-quality text is then reconstructed into instruction-response pairs, transforming previously fragmented business knowledge into a question-and-answer format that the model can learn. For example, "{"instruction":"Recommended itinerary for family trips","input":"2 days 1 night, two adults and one child","output":"Step 1: Check in at the park..."}".

[0110] The fine-tuning algorithm employs LoRA (Low-Rank Adaptation) technology. For the base model layer, a Transformer-based model is selected, and its original parameters are frozen, with low-rank bypass matrices introduced only in each attention layer. The rank is set to 64, and the scaling factor α is 16; gradient updates are performed only on these newly added parameters. The optimization objective uses the standard cross-entropy loss function, minimizing the difference between the model's predicted tokens and the actual tokens. The cross-entropy loss formula is as follows:

[0111] in is a trainable low-rank parameter used to minimize the difference between the predicted tourism corpus tokens and the actual tokens.

[0112] During training, the learning rate was set to 5e-5, and a cosine annealing scheduling strategy was used to ensure smooth convergence; the batch size was 128; and the number of training epochs was controlled between 3 and 5, with the specific termination time determined by the model perplexity stabilizing on the validation set. This configuration effectively avoided the risk of overfitting while ensuring training efficiency.

[0113] After fine-tuning, the AI ​​large-scale model's core engine 300 was evaluated using a set of evaluation datasets in the tourism vertical industry, focusing on three dimensions: the rationality of route planning, the accuracy of attraction information, and the timeliness of ticket prices. If the overall accuracy does not meet expectations, the system will automatically activate the RAG (Retrieval Augmentation) mechanism, connecting the engine with the real-time database of scenic spots, using the latest data to supplement the model's knowledge gaps, and significantly improving the timeliness of the output content.

[0114] Data source expansion: In addition to data from within the scenic area, the system can also access external data sources, such as weather forecast APIs, social media sentiment data, and surrounding traffic data, to make more accurate predictions and decisions.

[0115] Deployment methods: The system can be deployed entirely in the cloud (SaaS model), or on the scenic area's local server (private deployment), or in a hybrid cloud model.

[0116] Hardware carriers: In addition to mini-programs and web management platforms, interactive terminals can also take the form of smart speakers, AR glasses, digital signage, and other forms.

[0117] Application Areas Expansion: The architecture and methodology of this system are not only applicable to tourist attractions, but with appropriate modifications, they can also be applied to other complex scenarios requiring comprehensive management, such as large shopping malls, theme parks, airports, and transportation hubs.

[0118] When using it, prompts based on the roles and tasks will be used to call up the large model. The prompt format and workflow are as follows.

[0119] Prompt words are used for setting up AI characters and tasks in large models.

[0120] Role: Intelligent Scenic Area Resource Scheduling Commander Task: Your core task is to generate a precise, efficient, and executable resource scheduling instruction based on the real-time input of visitor density and resource density for specific attractions, combined with your historical visitor data, historical resource scheduling plans, and real-time weather and temperature information.

[0121] Core Objectives: 1. Safety First: Absolutely guarantee the personal safety of tourists and employees.

[0122] 2. Experience First: Optimize the visitor experience and reduce congestion and waiting.

[0123] 3. Core of Efficiency: Achieving refined and highly efficient allocation of resources such as people, vehicles, and materials.

[0124] Workflow and instruction format: 1. Receiving Input: You will receive real-time data in the following format, which will be the signal that triggers you to generate scheduling instructions: 1) Project Name: [The specific project name that needs to be scheduled] 2) Predict tourist density: [e.g., low / medium / high / 85%] 3) Current resource density: [e.g., insufficient / sufficient / surplus / 90%] 2. Analysis and Decision: Before generating any instructions, you silently perform the following analysis in the background (calling and integrating MCP data): Historical data review: Analyze historical tourist flow patterns that are similar to the current date, time, and holiday.

[0125] Evaluate the effectiveness and inadequacy of dispatch schemes under similar density and weather conditions in history.

[0126] Environmental factor assessment: Get the current weather forecast (e.g., sunny, cloudy, rainy, snowy) and temperature for the next 2-4 hours.

[0127] Assess the potential impact of weather and temperature on tourist behavior (e.g., in hot weather, tourists will flock indoors or to shady areas; in rainy weather, emergency guidance and provision of shelter are required).

[0128] Comprehensive assessment: Based on all the above information, an overall assessment of the current situation is formed (e.g., "High tourist density, risk of localized congestion, and rain forecast, advance planning is necessary").

[0129] 3. Generate scheduling instructions (OutputGeneration): Your output should be a structured, clear, and unambiguous scheduling instruction. Please strictly adhere to the following format: Output content { "dispatchInstruction": { "targetInfo": { "attraction": "[Autofill attraction name]", "currentState": { "visitorDensity": "[e.g., high / 85%]", "resourceDensity": "[For example: less than 90%]"}, "assessment": "[Your brief assessment of the current situation, such as: high concentration of tourists, resource shortages, and potential safety hazards]" }, "decisionBasis": { "historicalReference": "[Briefly describe which historical models were referenced in this decision-making process, such as: referencing the scheduling plan for the same day and time of last year's National Day holiday]", "weatherFactor": "[Briefly describe the impact of weather on decision-making, such as: sunny weather, temperature 26°C, strong willingness of tourists to engage in outdoor activities]" }, "specificInstructions": { "personnel": [ { "id": 1, "instruction": "Two security personnel will be dispatched from the 'Visitor Center' to the entrance of the 'One Line Sky' to maintain order." "details": { "estimatedArrivalTime": "Within 10 minutes" } }, { "id": 2, "instruction": "Add one more guide to the 'A1 Project' to manage crowd flow and provide explanations." "details": null } ], "vehicles": [ { "id": 1, "instruction": "Increase the shuttle bus departure frequency of the "L1" line from 15 minutes / trip to 8 minutes / trip." "details": null } ], "materials": [ { "id": 1, "instruction": "Emergency replenishment of supplies for the 'Tourist Station'." "details": { "items": [ { "name": "Drinking Water", "quantity": "20 boxes" }, { "name": "raincoat", "quantity": "50 pieces" } ] } } ], "informationRelease": [ { "id": 1, "instruction": "Issue a warning for tourist diversion." "details": { "channels": ["Scenic Area Official APP", "Entrance LED Screen"], "content": "Tourists are advised to prioritize visiting more spacious attractions such as 'B2'." } } ] }, "contingencyPlan": { "triggerCondition": "If the density of tourists in 'A2' continues to increase by 5% within 15 minutes", "action": "Immediately activate temporary flow control measures at the entrance." } } } The instructions for resource allocation in the scenic area are as follows.

[0130] 1. Root object Table 1: Root Objects field name type describe dispatchInstruction Object The root object of the entire scheduling instruction contains all the information for this scheduling. 2. targetInfo (Target Information) This object contains the target of the scheduling instruction and the current state assessment.

[0131] Table 2: Target Information field name type describe Example attraction String The specific name of the attraction targeted by the instruction Project A currentState Object Describe the current density status of tourists and resources. visitorDensity String The current density of resources (such as staff and facilities) at the attraction; "High" or "85%" resourceDensity String AI's brief assessment of the current situation based on comprehensive information. "Insufficient" or "90%" assessment String AI's brief assessment of the current situation based on comprehensive information. "Tourists are highly concentrated, and resources are strained." 3. Decision Basis This object describes the key information that the AI ​​relied on to generate this scheduling instruction.

[0132] Table 3: Basis for Decision-Making field name type describe Example historical reference String Historical data models or scheduling plans referenced in decision-making. "Refer to the contingency plan for the same day and time during last year's National Day holiday." weatherFactor String The impact of weather and environmental factors on this decision. "Sunny weather, temperature 26°C, strong desire for outdoor activities." 4. Specific Instructions Table 4: Specific Scheduling Instructions field name type describe personnel Array Personnel Dispatch Instruction List vehicles Array Vehicle dispatch instruction list informationRelease Array List of material dispatch instructions materials Array List of information release instructions General structure of instruction objects (taking personnel as an example): Table 5: General Structure of Instruction Objects field name type describe Example id String Unique identifier of the instruction 10010 instruction String Clear and concise instruction content "Two security personnel will be transferred from the 'Visitor Center'..." details Object It contains additional details of the instruction; otherwise, it is null. estimatedArrivalTime String (Personnel Scheduling) Estimated Arrival Time Within 10 minutes items Array (Materials Dispatch) Specific list of materials to be dispatched name String Material Name "drinking water" quantity String Quantity and unit of materials 20 boxes channels Array List of channels for (information dissemination) ["Official Scenic Area App", "Entrance LED Screen"] content String (Information Release) Specific content to be released "We recommend that visitors prioritize the 'B1' area..." 5. Contingency Plan This object defines contingency plans for dealing with emergencies.

[0133] Table 6: Emergency Response Plan field name type describe Example triggerCondition String Triggering conditions for activating the emergency response plan If the visitor density of "Project A2" continues to increase by 5% within 15 minutes... action String Specific response measures to be taken after triggering "Immediately implement temporary flow control measures at the entrance." The prompts for the image analysis agent are as follows.

[0134] You are a tourist flow density analysis module for a scenic area. Please analyze the input image and the specified area, and **output only structured tourist density information**, without including any explanations, suggestions, warnings or additional fields.

[0135] #### **Input** - Image: Base64 encoded image data; - Region: One or more normalized coordinates `[x_min, y_min, x_max, y_max]` (range [0,1]); - Region identifier (e.g., "region_01").

[0136] #### **Processing Requirements** - Only analyze the specified region; - Accurately detect and count tourists in the area (including those partially obscured, those with their backs to the camera, those sitting, etc.); - Exclude non-human objects (such as statues, billboards, and animals); - If the image quality is poor or the region is invalid, a reasonable estimate should still be returned (the confidence level may be lowered).

[0137] #### **Output Format (Strict JSON, no extra characters)** json { "type": "object", "properties": { "regions": { "type": "array", "items": { "type": "object", "properties": { "region_id": { "type": "string"}, "visitor_count": { "type": "integer", "minimum": 0}, "density_score": { "type": "number", "minimum": 0, "maximum": 1} }, "required": ["region_id", "visitor_count", "density_score"] } } }, "required": ["regions"] } ``` #### **Field Description** - `region_id`: A unique identifier for the region provided by the user (if not provided, it can be automatically generated, such as `"region_0"`, `"region_1"`). - `visitor_count`: The number of visitors detected in this area (integer); - `density_score`: **Normalized density metric**, calculated as follows: \[ \text{density\_score} = \min\left(1.0,\ \frac{\text{visitor\_count}}{\text{region\_pixel\_area}} \times \alpha \right) \] Where \(\alpha\) is the normalization coefficient, ensuring that typical crowded scenes (such as 50 people / 10000 pixels) are mapped to the range of 0.8~1.0; This value reflects only **relative density intensity** and can be used for sorting, comparison, or as input to the function *F*(tourist density, resource density).

[0138] **Absolutely prohibited**: Outputting non-JSON content, Markdown, comments, natural language, blank lines, and extra fields (such as confidence, level, alert, etc.).

[0139] The prompt template for the visitor profile AI agent is as follows.

[0140] You are a **tourist profiling agent**, responsible for extracting, fusing, and inferring tourist characteristics from multi-source data of the scenic area to generate a **unified, structured, and privacy-compliant** tourist profile. Output only JSON and must not contain any explanations, comments, or additional text.

[0141] #### **1. Input Data (Partially Missing)** You will receive one or a combination of the following inputs: - **Ticketing Data**: Order type (adult / student / senior), ticketing channel, number of visits, number of accompanying persons; - **Verification Data**: Entry time, entry gate, and whether a discount pass was used; - **Trajectory Data**: UWB / Bluetooth / WiFi location point sequence (including timestamps and region IDs); - **Consumer Data**: Food / Goods / Rental Consumption Records (including category, amount, time, and location); - **Interaction Data**: Mini Program browsing behavior, voice assistant keyword search, and photo upload area; - **External Tags** (optional): De-identified interest tags from third-party platforms (such as "parenting", "outdoors", "culture").

[0142] All inputs are structured or semi-structured logs; no raw images / videos are included.

[0143] #### **2. Processing Rules** - **Integrate multi-source information** and process conflicting data according to the priority of "behavior > declaration > inference" (e.g., consumption records are more reliable than ticket age). - **Supports cold start:** If there is no historical data, a temporary profile will be generated based solely on the current visit to the park. - **Privacy Protection**: - Do not output PII information such as ID card number, mobile phone number, and facial features; - Age output should be in segments (e.g., "18-24"), and gender can be "unknown"; - **Dynamically Updated**: The profile reflects the latest status of the current trip, rather than being a long-term fixed tag.

[0144] #### **3. Output Format (Strict JSON Schema)** json { "type": "object", "properties": { "visitor_id": { "type": "string"}, "profile_generated_at": { "type": "string", "format": "date-time"}, "demographics": { "type": "object", "properties": { "age_group": { "type": ["string", "null"], "enum": ["<18", "18-24", "25-35", "36-45", "46-55", "56-65", ">65", null] }, "gender": { "type": ["string", "null"], "enum": ["male", "female", "unknown", null] }, "visitor_type": { "type": "string", "enum": ["solo", "couple", "family", "group", "school_trip", "tour_group"] } }, "required": ["visitor_type"] }, "behavioral_traits": { "type": "object", "properties": { "interests": { "type": "array", "items": { "type": "string", "enum": ["nature", "culture","photography", "food", "shopping", "adventure", "relaxation", "education"]}, "maxItems": 3 }, "pace": { "type": "string", "enum": ["slow", "medium", "fast"] }, "consumption_level": { "type": ["string", "null"], "enum": ["low", "medium", "high", null] }, "visit_pattern": { "type": "string", "enum": ["first_time", "repeat_visitor", "frequent_visitor"] } }, "required": ["interests", "pace", "visit_pattern"] }, "real_time_context": { "type": "object", "properties": { "current_zone": { "type": "string"}, "dwell_time_minutes": { "type": "number", "minimum": 0}, "next_likely_zone": { "type": ["string", "null"]}, "fatigue_level": { "type": "string", "enum": ["low", "medium", "high"] } }, "required": ["current_zone", "dwell_time_minutes", "fatigue_level"] } }, "required": ["visitor_id", "profile_generated_at", "demographics", "behavioral_traits", "real_time_context"] } ``` #### **4. Key Field Descriptions** - `visitor_type`: Inference based on the number of people traveling together, trajectory clustering, and consumption synergy; - `interests`: derived from the area of ​​stay (e.g., museum → culture), the category of consumer goods (coffee → relaxation), and the search keywords ("cherry blossom photography" → photography); - `pace`: Calculated based on the distance traveled and the frequency of stops per unit time (slow: ≤ 30m / min, medium: 30–60m / min, fast: > 60m / min); - `next_likely_zone`: Optional field, can be filled if the model has predictive capabilities (e.g., from "Entrance" → "Viewing Platform"); - `fatigue_level`: Infers based on continuous walking time, stay in sloped areas, visits to rest facilities, etc.

[0145] #### **5. Strict Constraints** - Only outputs the above JSON, without any extra characters; - If a field cannot be determined, use `null` or a reasonable default value (such as `gender: "unknown"`). - External assumptions (such as "women prefer shopping") must not be introduced; - All enumeration values ​​strictly match the schema and cannot be customized with tags.

[0146] --- The prompt template for the marketing agent is as follows.

[0147] You are a **marketing decision-making intelligence agent** deployed in a smart scenic area platform. Based on the **tourist's real-time location, historical movement path, and profile tags** provided by the user, and combined with the scenic area's **marketing knowledge base** (including structured information such as products, restaurants, activities, and coupons), generate a **personalized, scenario-based, and executable** marketing plan for the tourist.

[0148] #### **1. Input Standards** You will receive the following structured input: json { "visitor_id": "V20251223001", "current_position": { "x": 120.1234, "y": 30.5678, "zone": "Observation Deck Area A"}, "trajectory": [ { "zone": "South Gate Entrance", "timestamp": "2025-12-23T10:00:00"}, { "zone": "Cultural and Creative Store", "timestamp": "2025-12-23T10:15:00"}, { "zone": "Observation Deck Area A", "timestamp": "2025-12-23T10:30:00"} ], "profile": { "age_group": "25-35", "gender": "female", "interests": ["Cultural and Creative Industries", "Photography", "Light Meals"], "consumption_level": "medium_high", "visit_frequency": "first_time", "group_type": "Couple" } } ``` Note: All fields may be missing; fault tolerance is required.

[0149] #### **2. Knowledge Base Retrieval Rules** - Your built-in or accessible scenic area **marketing knowledge base** contains the following entities (structured): - **Product**: `{id, name, category, price, location_zone, tags}` - **Activity**: `{id, name, type, start_time, end_time, location_zone,participant_type}` - **Coupon**: `{id, title, discount_type, valid_zones, valid_to,target_profile}` - Only marketing resources that can be reached within the **two areas ahead of the traffic flow** (or the current area) will be recommended; - Prioritize projects that match tourist profile tags (such as `interests`, `spending level`) by ≥ 70%; - Avoid repeatedly recommending content from areas where tourists have already spent more than 10 minutes (e.g., if they have already stopped at a cultural and creative shop, do not recommend similar products again).

[0150] #### **3. Output Requirements** - Outputs only structured JSON, without any explanation, greetings, or additional text; - Marketing plans are sorted by **priority** (most relevant first); - Each solution includes the **content, type, location, and matching reason** that can be executed.

[0151] #### **4. Output Format (Strictly Follow)** json { "type": "object", "properties": { "visitor_id": { "type": "string"}, "recommendations": { "type": "array", "items": { "type": "object", "properties": { "rec_id": { "type": "string"}, "type": { "type": "string", "enum": ["product", "activity","coupon"]}, "name": { "type": "string"}, "location_zone": { "type": "string"}, "distance_from_current": { "type": "number", "description":"Unit: meters, estimated value"}, "match_reason": { "type": "string", "description": "A brief description of the matching criteria, such as 'interest tag matching: photography'" }, "priority_score": { "type": "number", "minimum": 0, "maximum": 1} }, "required": [ "rec_id", "type", "name", "location_zone", "match_reason", "priority_score" ] }, "maxItems": 5 } }, "required": ["visitor_id", "recommendations"] } ``` #### **5. Behavioral Constraints** - **No fabrication allowed:** All recommendations are based on entities that exist in the knowledge base; - **Prohibit the output of non-JSON content:** including spaces, comments, Markdown, and natural language; - **Privacy compliance**: Sensitive profile fields (such as ID card number, mobile phone number) must not be exposed in the output; - **Timeliness**: Only recommends activities or coupons that are valid at the current time or have not expired; - **Diversity**: If multiple recommendations are made, they should cover at least two types (e.g., products + events).

[0152] --- ### Example Output (for reference only; should not appear in the actual model response) json { "visitor_id": "V20251223001", "recommendations": [ { "rec_id": "ACT_0023", "type": "activity", "name": "Sunset Photography Spot on the Mountain Top Now Open", "location_zone": "Observation Deck Area B", "distance_from_current": 80, "match_reason": "Interest tag matches: Photography; Currently located in observation deck area A, near area B", "priority_score": 0.92 }, { "rec_id": "PROD_1056", "type": "product", "name": "Limited Edition Mountain View Journal", "location_zone": "Cultural and Creative Store", "distance_from_current": 150, "match_reason": "Interest tag matching: cultural and creative industries; consumption level support", "priority_score": 0.85 } ] } ``` The prompts for the device monitoring agent are as follows.

[0153] You are an intelligent agent deployed in a smart scenic area management platform, responsible for monitoring the health and energy efficiency of all types of equipment. Your task is: Based on real-time equipment operation data, historical baselines, fault knowledge dictionaries, and environmental context, anomaly detection is performed on electrical or amusement equipment. Simultaneously, it determines whether there are mechanical, electrical, control, or energy-related risks and outputs structured early warning results. #### **1. Input Standards** You will receive a JSON object containing the following fields: json { "device_id": "string (unique device ID)", "device_type": "string (such as 'water_pump', 'roller_coaster_motor')", "category": "string('electrical' or 'amusement')", "location": "string (device location region)", "timestamp": "ISO8601 timestamp", "current_readings": { / / Dynamic field, example: "voltage_V": 220.5, "current_A": 9.2, "power_W": 1950, "energy_kWh_today": 42.3, "temperature_C": 68.7, "vibration_mm_s": 2.9, "operational_mode": "running | standby | maintenance", "cycle_count_today": 94, "o2_concentration_pct": 92.1, / / Oxygen concentrator specific "battery_soc_pct": 95, / / UPS specific "light_output_lux": 8500 / / Specific to lighting systems }, "historical_baseline": { "avg_power_running_W": 1800, "avg_power_standby_W": 150, "avg_energy_per_cycle_kWh": 0.45, "maintenance_interval_hours": 1000, "runtime_total_hours": 2340 }, "environmental_context": { "ambient_temp_C": 6.5, "humidity_pct": 72, "last_maintenance_date": "2025-11-20" } } ``` Note: Some fields may be missing; fault tolerance is required.

[0154] --- #### **2. Analytical Logic** ##### **General Rules** - **Use only input data and the built-in knowledge dictionary**, and do not fabricate sensors or parameters; - **Environmental Compensation**: Higher starting current is allowed at low temperatures, and the temperature threshold is moderately relaxed in high humidity environments; - **Maintenance Exemption**: If `last_maintenance_date` is within the last 7 days, minor anomalies will not trigger high-priority alerts; - **Tidal Load Identification**: High power during performance / peak hours is considered normal.

[0155] ##### **Anomaly Detection Dimensions** | Type | Example of detection logic | |------|--------| | **Mechanical Abnormality** | Sudden increase in vibration + unchanged power → Bearing wear / imbalance | | **Electrical Abnormality** | Current > 1.3 times rated + Rapid temperature rise → Overload or insulation deterioration | | **Control Anomaly** | During non-business hours, `operational_mode = running` → Accidental startup | | **Abnormal Energy Consumption** | Standby power > 200% of baseline → Leakage or control board malfunction; Single cycle energy consumption > 130% of baseline → Decreased transmission efficiency | ##### **Fault Knowledge Dictionary (Built-in, Some Examples)** - `F-PUMP-CLOG`: Water pump clogged → `power_W ↓` & `vibration_mm_s ↑` - `F-RIDE-DRIVE-SLIP`: Amusement vehicle slippage → `vibration ↑` & `speed ↓` (If no inference can be made, it indicates a power anomaly) - `F-UPS-BAT-DEG`: UPS battery aging → `battery_soc_pct < 85%` after full charge - `F-LIGHT-CTRL-ERR`: Lighting system mistakenly activated → `light_output_lux > 0` during night-off schedule --- #### **3. Output Format (Strict JSON format, no extra content allowed)** json { "type": "object", "properties": { "device_id": { "type": "string"}, "device_type": { "type": "string"}, "category": { "type": "string", "enum": ["electrical", "amusement"]}, "analysis_time": { "type": "string", "format": "date-time"}, "status": { "type": "string", "enum": ["normal", "anomaly", "suspected_fault"] }, "alerts": { "type": "array", "items": { "type": "object", "properties": { "alert_id": { "type": "string"}, "alert_type": { "type": "string", "enum": ["mechanical", "electrical", "control", "energy"] }, "fault_code": { "type": ["string", "null"]}, "fault_name": { "type": ["string", "null"]}, "severity": { "type": "string", "enum": ["low", "medium", "high"]}, "triggered_parameters": { "type": "object", "description": "Actual parameter names and values that triggered the alert", "additionalProperties": { "type": ["number", "string"]} }, "confidence": { "type": "number", "minimum": 0, "maximum":1}}, "required": ["alert_id", "alert_type", "severity", "triggered_parameters", "confidence"] }, "maxItems": 3 } }, "required": ["device_id", "device_type", "category", "analysis_time", "status", "alerts"] } ``` **Example Output (For illustration only; should not actually occur)** If the water pump's standby power consumption is abnormal: json > { > "alert_type": "energy", > "fault_code": "F-PUMP-STANDBY-LEAK", > "fault_name": "Abnormal standby power consumption of water pump", > "triggered_parameters": { "power_W": 420, "operational_mode": "standby"}, > "severity": "medium", > "confidence": 0.82 >} > ``` --- #### **4. Strict Behavioral Constraints** - **Absolutely prohibited** from outputting any characters other than JSON (including spaces, newlines, comments, and natural language); - If there are no abnormalities, `alerts` will be `[]`, and `status` will be `"normal"`; - `fault_code` and `fault_name` are only filled in if they match the built-in knowledge dictionary; otherwise, they are null. - All values ​​retain their original units and precision; - Do not expose sensitive information such as device IP, manufacturer, and password.

[0156] --- When processing data, the large model performs calculations according to the principle that F(tourist density, resource density) > 1.5, and outputs the corresponding solutions. The tourist density and resource density in the formula can be adjusted to other parameters based on actual needs. The internal processing of the large model follows this procedure.

[0157] 1. Data Preprocessing and Normalization Input data acquisition: Obtain raw data of "tourist density" (denoted as D_v, unit: people / square kilometer) and "resource density" (denoted as D_r, unit: resource units / square kilometer) of the current area in real time from the security monitoring subsystem.

[0158] Time window weighting: In order to smooth short-term fluctuations and capture trends, a weighted moving average (WMA) or exponential moving average (EMA) is applied to D_v and D_r over the past N time points (e.g., 60 data points in the past hour) to generate smoothed densities D_v' and D_r'.

[0159] Historical baseline calibration: The smoothed densities D_v' and D_r' are compared with the baseline densities D_{v,base} and D_{r,base} of the same historical period (the average of the same date, time and weather conditions), and the relative densities D_v^{rel}=D_v' / D_{v,base} and D_r^{rel}=D_r' / D_{r,base} are calculated.

[0160] Normalization: The relative densities D_v^{rel} and D_r^{rel} are mapped to a dimensionless interval [0,C] (e.g., [0,10]), resulting in normalized inputs I_v (normalized visitor density) and I_r (normalized resource density). The normalization function Norm(x) can be a piecewise linear function or a sigmoid function to handle extrema.

[0161] 2. Feature Engineering and High-Dimensional Mapping Derivative Feature Extraction: Based on I_v and I_r, a series of derived features are calculated, for example: f_1 = I_vI_r (Visitor-Resource Interaction Item) f_2 = I_v^2 (Nonlinear term for tourist density) f_3=sqrt(I_r+epsilon) (The square root term of resource density, where epsilon is a small constant to prevent division by zero) f_4 = log(I_v + 1) (logarithmic term of tourist density) f_5=max(0,I_v-I_r) (Overload index) f_6 = I_v / (I_r + epsilon) (Tourist resource ratio) f_7=(I_r-I_v) / max(I_v,I_r,epsilon) (supply and demand difference rate) f_8=... (More complex features can be added as needed) High-dimensional vector construction: Combine all derived features into a high-dimensional feature vector V_{features}=[f_1,f_2,...,f_n], where n is the number of features.

[0162] 3. Large Model Agent Processing (LLMAgent Processing) Context building: The high-dimensional feature vector V_{features} and related metadata (such as geographic location, current time, weather conditions, special event information, etc.) are built into a structured text context Context.

[0163] Intelligent reasoning: Inputs context into a large model. Used to understand complex, multi-dimensional descriptions of environmental states.

[0164] Intermediate state simulation: During its internal processing, LLMAgent may simulate various "what-if" scenarios of resource scheduling, assess the potential impact of different decisions on s, and perform multi-step reasoning.

[0165] Output Parsing: LLMAgent outputs a text containing the intermediate inference process and the final judgment. From this text, a core, continuous numerical value, Score_{raw}, is parsed using predefined rules or regular expressions. This Score_{raw} represents the model's comprehensive assessment of the current resource availability and can range widely (e.g., from negative to positive).

[0166] 4. Non-linear Transformation and Normalization Activation function application: Apply one or more complex nonlinear activation functions (such as Swish, GELU, or custom combination functions) to Score_{raw} to enhance the model's expressive power.

[0167] Final scaling calculation: The transformed Score_{raw} is mapped to the final resource availability scaling s using a final, differentiable function G, which may contain multiple parameters. The function G can be designed as a small, fixed neural network layer or a complex mathematical expression to ensure that the value of s falls within a reasonable expected range (e.g., through a scaling and offset operation).

[0168] Stability smoothing: Apply a simple filter (such as a first-order hysteresis filter) to the final output s to reduce drastic changes in the output caused by small fluctuations in the input, thereby increasing the stability of the decision.

[0169] 5. Decision threshold judgment After calculating s, a judgment is made based on a preset threshold: If s > 1.5, it is determined that there are insufficient resources, triggering the instruction to "schedule resources for the current location".

[0170] If s < 0.5, it is determined that there is a resource surplus, triggering the instruction to "schedule resources to the outside".

[0171] If 0.5 <= s <= 1.5, then the resources are considered to be in a balanced state and will not be scheduled for the time being.

[0172] The comprehensive scenic area management method proposed in this application is a continuous, closed-loop process, and its execution steps are as follows: Data collection: The system automatically collects various operational data from all business subsystems 24 / 7.

[0173] Data fusion: Data enters the data fusion processing center, where it is processed and correlated in real time, updating the global data view.

[0174] Intelligent Analysis: The core engine of the AI ​​large model continuously analyzes the latest global data.

[0175] When collecting operational data, the system adopts an "edge filtering + central elastic retrieval" strategy. For example, for the high-bandwidth video source of the security monitoring subsystem 160, the front-end edge gateway directly completes image analysis and only transmits back key features (such as instantaneous number of people, warning level, and device status code) as KB-level structured text.

[0176] The ticketing subsystem 110, scenic area transportation subsystem 130, and equipment management subsystem 180 adopt an event-driven incremental synchronization mechanism. Full data alignment is only triggered when key events such as ticket transactions, vehicle scheduling, or equipment status changes occur, thereby reducing invalid data collection.

[0177] In addition, the data fusion processing center 200 is equipped with a distributed computing cluster for dynamic load balancing.

[0178] Furthermore, the system incorporates a dynamically adjusted data collection strategy to adapt to different business scenarios. High-priority data (such as pedestrian flow in core areas and security alarms) is consistently collected at a rate of seconds; other types of low-priority data are automatically reduced in frequency when computing power is strained. The system also sets multiple business thresholds. For example, when the parking subsystem's occupancy rate is below 80%, the collection interval is at the minute level; once the 90% threshold is exceeded, the system switches to high-frequency real-time scanning. For subsystems that remain silent for extended periods, the system enters a low-power "sleep" mode, resuming data collection only upon receiving status updates or operational commands.

[0179] Meanwhile, data collection is also divided into daytime and nighttime periods. During peak daytime hours, real-time analysis is prioritized; after the park closes at night, the focus shifts to background tasks such as cold data archiving and index rebuilding.

[0180] Case Study: Traffic Management: The model found that ① the ticketing subsystem showed a tour group of 30 people had just entered the park; ② the park's transportation subsystem showed they had purchased electric shuttle bus passes; ③ their electronic guide route's first stop was attraction B. The model inferred that in the next 5-10 minutes, traffic demand from the entrance to attraction B would surge.

[0181] Dispatch Execution: The AI ​​big data model core engine generates an optimized strategy of "immediately dispatching two electric vehicles to the scenic area entrance to stand by." After receiving this, the integrated business dispatch and execution module automatically sends instructions to the integrated business dispatch and execution module and pushes the task to the nearby drivers' apps.

[0182] This application can combine historical data, pre-sale ticket information, and real-time passenger flow trends to predict potential congestion points within the next half hour to one hour, thereby enabling proactive resource allocation and management. Through video analysis, it can proactively identify events such as abnormal crowd gatherings, tourists entering dangerous areas, and lost items, automatically generating warnings and transforming passive monitoring into proactive defense. Based on tourists' real-time location, interest tags (such as family, photography, and adventure), physical condition (estimated by walking speed), and current queue times at various attractions, this invention can dynamically generate and adjust personalized tour suggestions, providing each tourist with a unique and optimal experience. Furthermore, it can accurately push coupons for nearby shops when it detects tourists lingering at a particular attraction for an extended period, achieving a conversion rate far higher than indiscriminate, broadcast marketing. This invention can analyze real-time heat maps of overall passenger flow and the supply and demand distribution of transportation (such as electric vehicles and shared bicycles). The system intelligently dispatches transportation based on passenger density at different attractions, maximizing transportation efficiency and reducing tourist waiting time.

[0183] As one possible implementation, when a group of intelligent agents generates an array of scheduling schemes, the following steps are performed: The predicted tourist density and current resource density are processed by weighted moving average. The processed predicted tourist density and current resource density are compared with the historical tourist density and historical resource density of the same period in history. The relative tourist density and relative resource density are calculated respectively, and the relative density is normalized. Extract the derived features of normalized relative tourist density and relative resource density, and construct a high-dimensional feature vector; Based on high-dimensional feature vectors and historical scheduling schemes, the resource availability ratio of the target area is evaluated; The resource availability ratio is subjected to nonlinear transformation and normalization to calculate the predicted density ratio. A scheduling scheme is generated based on the predicted density ratio. The scheduling scheme includes feasible scheduling schemes that match the predicted density ratio and those within the range of the predicted density ratio. An array of scheduling schemes is constructed, and the predicted density ratio and the array of scheduling schemes are output.

[0184] By processing tourist density, resource density, and other parameters, the accuracy of the output predicted density value is ensured. The output scheduling scheme array contains multiple corresponding scheduling schemes, providing a basis for determining the target scheduling scheme.

[0185] As one possible implementation method, the management method further includes the following steps: The acquired operational data also includes ticketing information and parking lot vehicle flow rate. The historical ticketing information, historical parking lot vehicle flow rate and historical ticket gate scheduling scheme of the scenic area operation system are obtained. The historical ticketing information and historical parking lot vehicle flow rate are preprocessed, and the preprocessed data is associated with the historical ticket gate scheduling scheme. The associated data is stored in the knowledge base. The resource scheduling agent invokes the large model based on the third prompt word template, reads the associated data from the knowledge base, calculates the proportional relationship between the current ticketing information and real-time parking lot vehicle flow speed in the current operational data and the historical ticketing information and historical parking lot vehicle flow speed during the same period, predicts the passenger flow at the ticket gates, and generates a ticket gate scheduling plan based on historical ticket gate scheduling plans. When predicting passenger flow at the ticket gates, the time it takes for people to travel from parking lots to the ticket gates is also considered, allowing for the prediction of when congestion might occur at the ticket gate location.

[0186] By comparing ticket sales information and parking lot vehicle flow rate with historical data from the same period, the future passenger flow at the current ticket checkpoint can be predicted. Combined with historical ticket checkpoint scheduling plans, a real-time ticket checkpoint scheduling plan can be generated to ensure that tourists can pass through the ticket checkpoint smoothly and avoid congestion.

[0187] As one possible implementation method, the management method further includes the following steps: The acquired operational data also includes check-in information. The intelligent agent group's tourist profile intelligent agent calls the large model according to the fourth prompt word template to obtain the ticket sales data in the current operational data, builds a user profile for tourists who purchase tickets including children's tickets, associates the tourist images collected during check-in with the user profiles, and generates parent-child tour route plans by filtering projects with a high proportion of children playing. Family travel itineraries will be pushed to family travel users' mobile devices for purchasing tickets.

[0188] By constructing user profiles, we can accurately identify family travel users and develop professional travel plans for them.

[0189] As one possible implementation method, the management method further includes the following steps: The acquired operational data also includes canteen catering information and retail product information; the marketing intelligent agent of the intelligent agent group calls the large model according to the fifth prompt word template, and generates children's meal plan and cultural and creative recommendation plan based on canteen catering information and retail product information; The program pushes children's meal packages and cultural and creative product recommendations to the mobile ticketing terminals of family travel users.

[0190] As one possible implementation method, the management method further includes the following steps: The acquired operational data also includes equipment names and operating parameters. The equipment monitoring intelligent agent template of the intelligent agent group calls the large model according to the sixth prompt word, obtains the historical operating parameters of the knowledge base, constructs the normal operating benchmark of the equipment based on the historical operating parameters, and associates the normal operating benchmark with the fault type standard data. The received real-time operating parameters are compared with the historical operating parameters of the same equipment. When the real-time operating parameters deviate, the fault type is predicted by combining the fault type standard data stored in the knowledge base, and predictive maintenance warnings are generated.

[0191] By using a large-scale model to monitor equipment within the park, timely warnings can be issued when equipment malfunctions, ensuring the long-term stable operation of equipment within the park.

[0192] As one possible implementation, the preprocessing includes the following steps: The received historical or current operational data is cleaned to remove invalid data, resulting in cleaned data. The cleaned data is formatted and then feature-correlated.

[0193] By processing historical and current operational data, data consistency is ensured, making reading and processing more convenient and direct.

[0194] Example 2: Figure 2 As shown, based on the integrated management method of scenic area based on workflow intelligent agents, a comprehensive management system of scenic area based on workflow intelligent agents is also provided, including a data processing module, intelligent agent group and integrated business scheduling and execution module; The data processing module is used to acquire historical operation data and historical scheduling plans of the scenic area operation system. The acquired operation data includes security monitoring data and scenic area resource data. The module preprocesses the acquired historical operation data, associates the processed historical operation data with the historical scheduling plans, and stores the associated data in the knowledge base. It is also used to obtain the current operating data of the scenic area operation system, preprocess the current operating data, store the preprocessed data in the knowledge base, and synchronously input the intelligent agent group including image analysis intelligent agent and resource scheduling intelligent agent; The image analysis agent is used to call the large model based on the first prompt word template, calculate the passenger flow and tourist routes corresponding to the security monitoring data in the current operation data, predict the tourist density of each project on the tourist route at the next moment based on the passenger flow, obtain historical security monitoring data from the knowledge base, calculate the historical tourist density of the corresponding project in the same period, and transmit the predicted tourist density and historical tourist density to the resource scheduling agent. The resource scheduling agent is used to call the large model based on the second prompt word template, calculate the current resource density of the corresponding scenic area resource data in the current operation data; obtain the historical scenic area resource data of the knowledge base, calculate the historical resource density of the corresponding project in the same period, and calculate and generate a scheduling scheme array based on the historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling scheme; The integrated business scheduling and execution module is used to filter target scheduling schemes from the scheduling scheme array and generate scheduling instructions based on the target scheduling schemes.

[0195] This invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of a comprehensive scenic area management method based on a workflow intelligent agent.

[0196] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a comprehensive scenic area management method based on a workflow intelligent agent.

[0197] In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0198] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A comprehensive scenic area management method based on workflow intelligent agents, characterized in that, Includes the following steps: Step S10: Obtain historical operation data and historical scheduling plans from the scenic area operation system. The obtained operation data includes security monitoring data and scenic area resource data. The obtained historical operation data is preprocessed, and the processed historical operation data is associated with the historical scheduling plans. The associated data is then stored in the knowledge base. Step S20: Obtain the current operation data of the scenic area operation system, preprocess the current operation data, store the preprocessed data in the knowledge base, and synchronously input it into the intelligent agent group, which includes an image analysis intelligent agent and a resource scheduling intelligent agent; Step S30: The image analysis agent calls the large model according to the first prompt word template, calculates the passenger flow and tourist route corresponding to the security monitoring data in the current operation data, and predicts the tourist density of each item on the tourist route at the next moment based on the passenger flow; obtains the historical security monitoring data of the knowledge base, calculates the historical tourist density of the corresponding item in the same period, and transmits the predicted tourist density and historical tourist density to the resource scheduling agent; Step S40: The resource scheduling agent calls the large model according to the second prompt word template to calculate the current resource density of the corresponding scenic area resource data in the current operation data; Obtain historical scenic area resource data from the knowledge base, calculate the historical resource density of the corresponding project in the same historical period, and calculate and generate an array of scheduling schemes based on historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling schemes; Step S50: Filter the target scheduling scheme from the scheduling scheme array, and generate a scheduling instruction according to the target scheduling scheme.

2. The scenic area integrated management method based on workflow intelligent agents according to claim 1, characterized in that, The generation of the scheduling scheme array includes the following steps: The predicted tourist density and current resource density are processed by weighted moving average. The processed predicted tourist density and current resource density are compared with the historical tourist density and historical resource density of the same period in history. The relative tourist density and relative resource density are calculated respectively, and the relative density is normalized. Extract the derived features of normalized relative tourist density and relative resource density, and construct a high-dimensional feature vector; Based on the high-dimensional feature vector and historical scheduling scheme, the resource availability ratio of the target area is evaluated; The resource availability ratio is subjected to nonlinear transformation and normalization to calculate the predicted density ratio. A scheduling scheme is generated based on the predicted density ratio. The scheduling scheme includes feasible scheduling schemes that match the predicted density ratio and those within the range of the predicted density ratio. An array of scheduling schemes is constructed, and the predicted density ratio and the array of scheduling schemes are output.

3. The scenic area integrated management method based on workflow intelligent agents according to claim 1, characterized in that, The management method also includes the following steps: The acquired operational data also includes ticketing information and parking lot vehicle flow rate. The historical ticketing information, historical parking lot vehicle flow rate and historical ticket gate scheduling scheme of the scenic area operation system are obtained. The historical ticketing information and historical parking lot vehicle flow rate are preprocessed, and the preprocessed data is associated with the historical ticket gate scheduling scheme. The associated data is stored in the knowledge base. The resource scheduling agent calls the large model based on the third prompt word template, reads the associated data in the knowledge base, calculates the ratio of the current ticketing information and real-time parking lot vehicle flow rate in the current operation data to the historical ticketing information and historical parking lot vehicle flow rate in the same period, predicts the passenger flow at the ticket gate, and generates a ticket gate scheduling plan in combination with the historical ticket gate scheduling plan.

4. The scenic area integrated management method based on workflow intelligent agents according to claim 3, characterized in that, The management method also includes the following steps: The acquired operational data also includes check-in information. The intelligent agent group's tourist profile intelligent agent calls the large model according to the fourth prompt word template to obtain the ticket sales data in the current operational data, builds a user profile for tourists who purchase tickets including children's tickets, associates the tourist images collected during check-in with the user profiles, and generates parent-child tour route plans by filtering projects with a high proportion of children playing. Family travel itineraries will be pushed to family travel users' mobile devices for purchasing tickets.

5. The scenic area integrated management method based on workflow intelligent agents according to claim 4, characterized in that, The management method also includes the following steps: The acquired operational data also includes canteen catering information and retail product information; the marketing intelligent agent of the intelligent agent group calls the large model according to the fifth prompt word template, and generates children's meal plan and cultural and creative recommendation plan based on canteen catering information and retail product information; The program pushes children's meal packages and cultural and creative product recommendations to the mobile ticketing terminals of family travel users.

6. The scenic area integrated management method based on workflow intelligent agents according to claim 1, characterized in that, The management method also includes the following steps: The acquired operational data also includes equipment names and operating parameters. The equipment monitoring intelligent agent template of the intelligent agent group calls the large model according to the sixth prompt word, obtains the historical operating parameters of the knowledge base, constructs the normal operating benchmark of the equipment based on the historical operating parameters, and associates the normal operating benchmark with the fault type standard data. The received real-time operating parameters are compared with the historical operating parameters of the same equipment. When the real-time operating parameters deviate, the fault type is predicted by combining the fault type standard data stored in the knowledge base, and predictive maintenance warnings are generated.

7. The scenic area integrated management method based on workflow intelligent agents according to claim 1, characterized in that, The preprocessing includes the following steps: The received historical or current operational data is cleaned to remove invalid data, resulting in cleaned data. The cleaned data is formatted and then feature-correlated.

8. A management system based on the scenic area integrated management method based on workflow intelligent agents as described in any one of claims 1 to 7, characterized in that, It includes a data processing module, an intelligent agent group, and a comprehensive business scheduling and execution module; The data processing module is used to acquire historical operation data and historical scheduling plans of the scenic area operation system. The acquired operation data includes security monitoring data and scenic area resource data. The module preprocesses the acquired historical operation data, associates the processed historical operation data with the historical scheduling plans, and stores the associated data in the knowledge base. It is also used to obtain the current operating data of the scenic area operation system, preprocess the current operating data, store the preprocessed data in the knowledge base, and synchronously input the intelligent agent group including image analysis intelligent agent and resource scheduling intelligent agent; The image analysis agent is used to call the large model based on the first prompt word template, calculate the passenger flow and tourist routes corresponding to the security monitoring data in the current operation data, and predict the tourist density of each project on the tourist route at the next moment based on the passenger flow; obtain historical security monitoring data from the knowledge base, calculate the historical tourist density of the corresponding project in the same period, and transmit the predicted tourist density and historical tourist density to the resource scheduling agent. The resource scheduling intelligent agent is used to call the large model based on the second prompt word template to calculate the current resource density of the corresponding scenic area resource data in the current operation data. Obtain historical scenic area resource data from the knowledge base, calculate the historical resource density of the corresponding project in the same historical period, and calculate and generate an array of scheduling schemes based on historical tourist density, historical resource density, predicted tourist density, current resource density and historical scheduling schemes; The integrated business scheduling and execution module is used to filter target scheduling schemes from the scheduling scheme array and generate scheduling instructions based on the target scheduling schemes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the integrated scenic area management method based on workflow intelligent agents according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the scenic area integrated management method based on workflow intelligent agents according to any one of claims 1-7.

Images