Scenic area intelligent regulation and control method and system based on time-sharing reservation and space-time prediction

Abstract

The invention relates to the technical field of intelligent scenic spot management, in particular to a scenic spot intelligent regulation and control method and system based on time-sharing reservation and space-time prediction. The method comprises the following steps: constructing a dynamic layering-to-park rate table through time-sharing real-name reservation and multi-source real-time sensing data acquisition, and performing adaptive weighted fusion with real-time passenger flow observation of edge preprocessing to form fusion passenger flow estimation; constructing a space-time prediction model mixed by a graph neural network and a long-short-term memory network, and realizing short-term and medium-long-term passenger flow prediction of each region of the scenic region; a closed-loop decision-making system is formed through entrance microscopic release regulation and control, probabilistic path induction, capacity grading early warning and reservation quota dynamic callback, and passenger flow distribution is actively adjusted; and through a multi-dimensional effect evaluation and self-adaptive correction mechanism, a feedback reverse optimization prediction model and a distribution strategy are executed, and the reservation quota is adjusted in a linkage manner. According to the invention, full-process intelligent regulation and control from a reservation source to field management and control are realized, and the operation safety, balance and tourist experience of the scenic spot are improved.

Description

Technical Field

[0001] This invention relates to the field of smart scenic area management technology, specifically to a method and system for intelligent control of scenic areas based on time-sharing reservation and spatiotemporal prediction. Background Technology

[0002] With the continuous growth of national tourism consumption demand and the in-depth advancement of smart city construction, scenic area operation and management are rapidly developing towards digitalization, refinement, and intelligence. Time-sharing reservation systems have been widely applied in many scenic areas, providing a forward-looking data foundation for visitor flow management. At the same time, advancements in IoT sensing technology, edge computing, and artificial intelligence models have made it possible to acquire and process large-scale visitor flow spatiotemporal data in real time.

[0003] Chinese invention patent application CN118297310A discloses a method, device, equipment, and medium for intelligent passenger flow control. The method includes: acquiring passenger flow impact data values ​​corresponding to a target scenic area within a future predicted time interval; the passenger flow impact data values ​​include the values ​​of multiple target passenger flow impact characteristic variables; determining the predicted passenger flow data corresponding to the target scenic area within the predicted time interval based on the passenger flow impact data values ​​and a pre-established random forest regression model; and controlling passenger flow for the target scenic area according to a preset passenger flow control strategy and the predicted passenger flow data.

[0004] Against this backdrop, how to deeply integrate the prior information of the reservation system, real-time multi-dimensional sensing data and advanced spatiotemporal prediction models, and build an adaptive and closed-loop optimized dynamic control strategy system based on the prediction results, so as to realize intelligent management of the entire process from passenger flow early warning and on-site guidance to source scheduling, has become a key technological development direction for improving the scenic area's safety guarantee capabilities, resource utilization efficiency and tourist service quality. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a method and system for intelligent control of scenic spots based on time-sharing reservation and spatiotemporal prediction.

[0006] The technical solution of this invention: a smart control method for scenic spots based on time-sharing reservation and spatiotemporal prediction, comprising the following specific implementation steps: S1. A dynamic hierarchical arrival rate table is constructed by time-sharing real-name reservation and multi-source real-time perception data collection, and adaptively weighted and fused with real-time passenger flow observations preprocessed at the edge to form a fused passenger flow estimate. S2. Construct a spatiotemporal prediction model that combines graph neural networks and long short-term memory networks. Using the fused passenger flow estimate as input, the model can predict short-term and medium-to-long-term passenger flow in various areas of the scenic spot and output the prediction confidence level. S3. Using short-term and medium-to-long-term passenger flow forecasts as core inputs, a closed-loop decision-making system is formed through micro-level entry control, probabilistic path guidance, capacity-level early warning, and dynamic adjustment of reservation quotas to proactively regulate passenger flow distribution within the scenic area. S4. Conduct a multi-dimensional quantitative evaluation of the control effect of step S3, and based on the evaluation results, optimize the parameters and diversion strategy of the spatiotemporal prediction model in reverse through an adaptive correction mechanism, while simultaneously adjusting the reservation quota of the time-sharing reservation system.

[0007] Preferably, step S1 specifically includes: Collect tourists' real-name time-sharing reservation information, which includes at least the reserved entry time and points of interest preferences; stratify the reservation data according to the reservation advance, mode of transportation and date type characteristics, generate and dynamically update the reservation-to-park conversion rate table, and record the actual probability of arrival at the park for each reservation stratum in different time windows; Real-time passenger flow data is collected by at least two of the following: gate counters, video sensing devices, and wireless probes; real-time passenger flow data is aggregated, trajectory extracted, anomaly detected, and noise denoised in edge processing units deployed at entrances and key nodes to generate regional real-time passenger flow observations. The expected passenger flow prior to reservations is calculated based on the reservation-to-park conversion rate table. The fusion coefficient is dynamically calculated based on the confidence level of real-time observation data and the uncertainty of the reservation prior. The fusion coefficient is used to weight and fuse the expected passenger flow prior to reservations with the real-time passenger flow observations of the region, and the fused passenger flow estimate of each region is output. The reservation data and real-time data are mapped to a unified scenic area spatial grid or functional area for time alignment and anomaly correction. The reservation-to-park conversion rate table and the integrated visitor flow estimate are updated in real time to form an online data stream.

[0008] Preferably, step S2 specifically includes: The scenic area is divided into several functional area nodes. For each node, a feature vector is constructed that includes real-time fusion visitor flow estimation, reservation prior, area capacity, historical visitor flow patterns, and area type encoding. Based on the spatial proximity relationship between areas and the historical visitor movement probability, a spatial topology graph of the scenic area with edge weights is constructed. The feature vector sequence and the spatial topology map are input into a hybrid model consisting of a graph neural network and a first long short-term memory network. The graph neural network is used to aggregate the features of spatially adjacent nodes to capture spatial dependencies, and the first long short-term memory network is used to process the time series of node features to capture temporal evolution. The model outputs short-term passenger flow prediction values ​​and corresponding confidence levels for each region within the next 30 to 60 minutes. Based on the prior probability of reservations and historical passenger flow patterns provided by the reservation-to-park conversion rate table, a second long short-term memory network is used to model and predict the medium- and long-term passenger flow trends of various areas in the next few hours to days. Based on the confidence level of short-term passenger flow forecasts and the accuracy of historical forecasts, the weighting coefficients of short-term and medium-to-long-term forecast results are dynamically determined, and the two are weighted and merged to form the final regional forecast passenger flow matrix.

[0009] Preferably, the inlet micro-release control in step S3 specifically includes: Calculate the predicted arrival rate of each entrance in the future time window based on the short-term passenger flow forecast results; Establish an entrance queue dynamics model, combine the predicted arrival rate and real-time release rate, and calculate and update the real-time entrance queue length; The optimal passage rate of each entrance gate is dynamically calculated based on the preset maximum safe queuing number threshold, the predicted congestion penalty in the downstream area, and the expected average visitor waiting time. The gate is controlled to perform micro-release or batch release operations based on the optimal release rate.

[0010] Preferably, the probabilistic path induction in step S3 specifically involves: Tourists are dynamically segmented based on their interests, real-time location, and the context of their visit. For each group of tourists, several candidate tour routes are selected from the topology of the scenic area; Construct a path reward function that integrates at least the static experience utility of the path, the real-time congestion risk based on short-term predictions, and the path allocation balance index. The contextual bandit online learning model is adopted to dynamically update the selection probability of each candidate path for each tourist group based on the real-time calculated path reward. Route guidance information is disseminated to tourists through electronic displays, park announcements, or mobile applications, using a multiple-selection recommendation method.

[0011] Preferably, the capacity tiered early warning in step S3 specifically includes: Based on the final predicted passenger flow, the maximum safe carrying capacity of each region, and the prediction confidence level, the comprehensive overload risk index of each region is calculated. Set three levels of early warning thresholds: low risk, medium risk, and high risk. When the overall overload risk index reaches the low-risk threshold, a flexible reminder message is sent to tourists; when it reaches the medium-risk threshold, a route redirection instruction is triggered and linked with the probabilistic route guidance; when it reaches the high-risk threshold, a flow restriction instruction is triggered for the associated entrance or a temporary closure instruction for the target area.

[0012] Preferably, the dynamic callback of the reservation quota in step S3 is specifically as follows: Based on the medium- and long-term passenger flow trend forecast results, analyze the passenger flow carrying capacity pressure of different areas in different time periods in the future; With the goal of reducing the risk of regional overloading and balancing the overall spatial load, a reservation quota optimization model was established; The optimization model is used to dynamically adjust the upper limit of the number of reservation slots available in the time-sharing reservation system for each time period, each entrance, and each area of ​​interest in the future. The adjusted reservation quota will be synchronized to the reservation platform in real time, and the public reservation interface will be updated.

[0013] Preferably, the multi-dimensional quantitative evaluation in step S4 specifically includes: Within the preset evaluation time window, real-time changes in passenger flow density, average visitor stay duration, visitor route transfer ratio, and entrance queue length were collected for each controlled area. The above four variables are combined to form a multi-dimensional diversion execution effect vector, which is used to quantitatively characterize the comprehensive effect of the diversion strategy in terms of spatial load, time delay, path response and entry order.

[0014] Preferably, the adaptive correction mechanism in step S4 specifically includes: Calculate the deviation between the predicted and actual passenger flow values ​​for each area, and evaluate the prediction model error and strategy contribution by combining the implementation intensity of each diversion measure; Based on the deviation and prediction uncertainty, the parameters of the hybrid model of graph neural network and long short-term memory network are fine-tuned using gradient descent. Based on the strategy contribution and path reward feedback, local adjustments were made to the path selection probability distribution in probabilistic path induction and the parameters of the release rate calculation formula in the entry micro-release control. The historical risk levels of each region, obtained from long-term assessment and statistics, are fed back to the dynamic adjustment process of the reservation quota, which is used to dynamically correct the adjustment coefficient in the reservation quota calculation formula.

[0015] The technical solution of this invention: A scenic area intelligent control system based on time-sharing reservation and spatiotemporal prediction, which is used to execute the above-mentioned scenic area intelligent control method based on time-sharing reservation and spatiotemporal prediction, comprising: The time-sharing reservation enhancement and multi-source data acquisition module is used to collect and manage time-sharing real-name reservation information, and integrate multi-source real-time sensing data from turnstiles, videos and wireless probes to complete data fusion processing; The spatiotemporal prediction module, which integrates reservation data, is used to build and run a hybrid model of graph neural network and long short-term memory network. Based on the output of the time-sharing reservation enhancement and multi-source data acquisition module, it performs short-term and medium-to-long-term passenger flow prediction. The dynamic diversion strategy generation and execution module is used to generate an entrance release control strategy, a probabilistic path guidance strategy, and a capacity classification early warning instruction based on the prediction results of the spatiotemporal prediction module of the fused reservation data, and send them to the corresponding execution terminals. The execution feedback evaluation and reservation control self-optimization module is used to perform multi-dimensional quantitative evaluation of the effectiveness of the diversion strategy, adaptively correct the prediction model parameters and diversion strategy parameters based on the evaluation results, and adjust the reservation quota of the time-sharing reservation system in conjunction with the evaluation results.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs an intelligent scenic area control method and system based on time-sharing reservation and spatiotemporal prediction. By integrating prior information from time-sharing reservations with multi-source real-time observation data, it significantly improves the accuracy and foresight of scenic area visitor flow perception. This allows visitor flow estimation to not only rely on historical and real-time information but also reflect the visitor intentions of those who have made reservations, providing a more reliable data foundation for subsequent decision-making. The spatiotemporal prediction architecture, combining graph neural networks and sequence models, effectively captures the spatial correlations and dynamic evolution of visitor flow within the scenic area, achieving accurate predictions of short-term and medium-to-long-term visitor flow trends and outputting prediction confidence levels, thus enhancing the operability and risk foresight of management decisions. Based on this, the system constructs a closed-loop dynamic control system driven by prediction results. The system employs a series of collaborative strategies, including entrance release rate adjustment, probabilistic path guidance, capacity-level early warning, and dynamic allocation of reservation quotas. These strategies enable proactive, flexible, and tiered guidance of visitor flow, effectively alleviating regional congestion and balancing spatial load. This, in turn, enhances the overall visitor experience and scenic area operational efficiency while ensuring visitor safety. The system also incorporates a complete execution feedback and adaptive optimization mechanism. This mechanism allows for quantitative evaluation of the actual effects of control strategies and coordinated correction of prediction model parameters and diversion strategies. Furthermore, it feeds back operational experience to reservation quota adjustments, forming a continuous self-optimization closed loop from reservation source to on-site management and strategy iteration. This significantly enhances the system's robustness and long-term adaptability in handling different visitor flow scenarios, providing technical support for the refined, intelligent, and sustainable operation and management of scenic areas. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for intelligent regulation of scenic areas based on time-sharing reservation and spatiotemporal prediction proposed in this invention. Figure 2 This is a system architecture diagram of a scenic area intelligent control system based on time-sharing reservation and spatiotemporal prediction proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1As shown, the present invention proposes a smart scenic area control method based on time-sharing reservation and spatiotemporal prediction, the specific implementation steps of which are as follows: S1. By constructing prior information on future visitor flow and current status of the scenic area through time-sharing real-name reservations and multi-source real-time data collection, and combining dynamic tiered arrival rate (ART) and edge preprocessing mechanisms, the reservation and real-time flow are integrated. This provides high-precision, low-latency, and online-updable basic data input for subsequent short-term and medium-to-long-term forecasts and dynamic diversion strategies, forming a closed-loop adaptive data collection system. The specific implementation process is as follows: S11. In the stage of time-sharing real-name reservation collection and enhanced coding, tourists submit real-name reservation information on the official platform. A reservation-arrival conversion rate table (ART) is generated hierarchically based on multi-dimensional features such as reservation advance time, interests, and transportation methods. The arrival probability at each level is dynamically recorded to provide a prior probability distribution for short-term prediction and support subsequent strategy optimization. Specifically: Reservation information is collected. Tourists make real-name time-sharing reservations through the official reservation platform and the reservation information is recorded, including but not limited to: tourist ID (hash-processed to ensure privacy), reserved entry time slot, preferred points of interest / route preference, number of people in the group, and special permission marks (such as seniors, children, and disabled persons). The reservation data is stratified according to multidimensional features to form a reservation-arrival conversion rate stratification table (ART). The ART table records the arrival probability of each reservation stratum and is used to dynamically adjust the reservation prior. ; in, This represents the actual probability of the l-th tier of the reservation group arriving at the park within the time window h; This represents the actual number of visitors to the park at time window h in the l-th layer of history; This represents the number of reservations made by the l-th layer within time window h; This represents the overall prior attendance rate, which in this embodiment can be set as the historical overall average attendance rate. Indicates the smoothness intensity (controlling the dependence on priors); It should be noted that, based on the principle of hierarchical design, principal component decomposition was performed on factors that have historically significantly influenced park visit behavior, and a small number of key dimensions (such as advance booking, mode of transportation, weekday / weekend labels, whether it is a holiday, etc.) were selected to ensure that the number of strata is within a statistically feasible range (to avoid the sample size of each stratum being too small). S12. In the multi-source real-time sensing and acquisition and edge preprocessing stage, real-time passenger flow is collected through gate counting, video head counting, Wi-Fi / Bluetooth probes, and base station signaling. Edge nodes aggregate the data, extract trajectories, detect anomalies, and denoise, generating low-latency, high-reliability regional real-time passenger flow information. This provides basic observation data for fusion and prediction, specifically: The system uses gate counters to record the number of people entering the park and the flow of people entering and exiting in real time. It also uses video surveillance and AI head counting to achieve anonymized people and trajectory extraction through edge computing. In addition, it uses Wi-Fi / Bluetooth probes to capture signals from visitors' devices to help estimate crowd density and movement speed, and obtain coarse-grained location data. Edge processing units are deployed at each entry point and key node to process the raw data: Aggregation: Counts the number of people entering the park every 15 seconds or 1 minute; Trajectory extraction: Generate a sequence of movement trajectories mapped to region IDs; Anomaly detection: Real-time detection of abnormal events such as queue backlog or congestion; Data anonymization and denoising: protecting privacy and reducing noise interference; The edge performs time alignment and preliminary weighted fusion of the various sensor sources to obtain real-time observations of the region: ; ; ; in, S represents the real-time passenger flow density of region r at time t; r This represents the set of all sensor sources covering the coverage area r; This represents the number of people observing sensor s at time t; This indicates the sensor source weight, which is set based on historical accuracy, coverage, and latency performance. The confidence coefficient represents the reliability of sensor source s at time t, used to measure the real-time observations from that sensor source. Reliability at the current moment; This indicates the estimation accuracy of sensor s relative to the "true value" or a high-confidence source (such as a gate) within the most recent time window; This indicates a comprehensive set of indicators, including whether the sensor is currently online, signal quality, and response delay (1 for online, 0 for offline, or a quality value of 0-1). Indicates the weighting balancing factor; A lightweight anomaly detector runs at the edge, using both threshold and model-based judgments for metrics such as queue length, dwell time, and speed decrease. If an anomaly is detected (e.g., queue growth rate exceeding the threshold or dwell time increasing), the edge immediately reports the event type, geographic window ID, and confidence level, allowing dispatchers or automated strategies to trigger temporary control measures. Short trajectory segments are encoded in "path language" at the edge (e.g., encoding consecutively passed grid ID strings into short sequences) and uploaded to the cloud for behavior pattern learning, while protecting the privacy of the original trajectory. S13. In the stage of fusion of reservation data and real-time flow, the ART stratified arrival rate and real-time observation data are fused according to a dynamic weighting formula. The weights are adaptively adjusted to balance the reservation prior and the reliability of the real-time flow, and the fused regional passenger flow estimate is output. This provides a high-precision input for subsequent short-term forecasting and diversion strategy generation, and also supports anomaly risk assessment. Perform data fusion: ; ; Among them, L r This represents the set of reservation hierarchies for all possible access areas r; This indicates that integration serves as a short-term window for the future. (e.g., 15 minutes) Estimated number of visitors expected in area r; This indicates the number of reservations expected to flow to region r within layer l during the specified time period; This represents the expected value of the prior reservation mapped to a specific region r; This represents the fusion coefficient (0~1), used to balance the weights of the current observation and the reserved prior in different scenarios; This indicates the selection of point of interest p in history. i The probability that a tourist is observed to enter area r (derived from historical trajectory statistics); Dynamically set the fusion coefficient : ; in, Indicates the confidence level of real-time observations; This indicates the uncertainty of the prior knowledge required for the appointment; Indicates the equilibrium hyperparameters; Represents a small constant that prevents division by zero; S14. Map reservation and real-time data to a unified grid or functional area, align them by time, and perform anomaly correction. Update the ART table and fused passenger flow estimates in real time to ensure data structure, continuity, and availability. This provides an online, usable data stream for predictive models and strategy generation, forming a closed-loop data foundation. Map all reservation and real-time data to a unified grid or functional area ID, unify the time granularity of reservation and real-time data (e.g., 15 minutes), ensure consistency in the time window used for short-term forecasting and strategy generation, update the status of arrived tourists in ART in real time, smooth out observed anomalies, and have edge nodes quickly report abnormal events to the cloud to generate regional integrated passenger flow data. .

[0019] S2. Construct a spatiotemporal prediction model that integrates prior reservation data with multi-source real-time observation. Through a hybrid architecture of graph neural networks and time series models, achieve short-term dynamic visitor flow prediction and medium-to-long-term trend prediction for scenic spots, and output confidence scores. This provides an operable, continuous, and quantifiable predictive basis for subsequent dynamic diversion strategies. It also supports reservation quota optimization and capacity early warning. The specific implementation process is as follows: S21. Divide the scenic area into functional area nodes, generate fused feature vectors, including real-time observed visitor flow, prior reservations, area capacity, historical visitor flow patterns, and functional labels, construct spatial proximity relationships and flow direction weights between nodes, and form a graph structure suitable for GNN processing, providing a unified input basis for short-term and medium-to-long-term predictions, specifically: The scenic area is divided into N functional areas or grid nodes. Each node corresponds to a scenic spot, trail section, rest area, or entrance plaza within the scenic area; a feature vector is generated for each node. : ; in, Indicates the upper limit of regional carrying capacity, used for capacity constraints and safety early warning; It indicates historical passenger flow patterns, including periodic characteristics such as hourly, daily, and holiday periods; This indicates the area type code (such as entrance, core scenic spot, trail section, rest area, etc.); Construct an edge set E to represent the spatial proximity and tourist flow relationships between regions. The edge weights are calculated by combining historical movement probabilities and physical distances. ; in, Represents node v i With v j The edge weights between nodes represent the degree of influence that neighbors have on the current node. The language matrix representing the historical trajectory from v i Move to v j The probability of; Indicate physical distance to avoid long distances directly affecting predictions; and Indicates the balance coefficient; S22. By using a graph neural network to capture the spatial dependencies between nodes and then processing the time series evolution through LSTM, short-term passenger flow prediction for the next 30-60 minutes is achieved. The predicted passenger flow value and confidence level for each area are output, providing a refined decision-making basis for real-time entrance control and route guidance. At the same time, it can quickly respond to real-time fluctuations. Specifically: Construct a hybrid GNN+LSTM model, utilizing the GNN portion to capture spatial dependencies between nodes: ; Then, the LSTM part is used to capture the changes in the time series: ; Forming short-term forecast output: ; in, This represents the node feature vector after GNN aggregation; This represents the LSTM hidden state of node i at time t; Indicates the length of the time window used for short-term forecasting; Indicates that node i in the future Short-term passenger flow forecast within the area; This indicates the short-term forecast time step, such as 30 minutes or 60 minutes; This represents a short-sequence long short-term memory network used to capture short-cycle, real-time dynamic features; S23. Based on prior reservation probabilities and historical passenger flow patterns, a weighted time series model is constructed to predict the medium- to long-term passenger flow trends in various regions over the next few hours to days, namely: Wherein, UL represents the set of entry points for reservation sources or reservation time slots; Indicates that node i in the future Medium- to long-term passenger flow forecasts for different time periods; This represents a long short-term memory network used to capture long-term trends, cumulative effects, and process changes. S24. Short-term forecasts and medium-to-long-term trend forecasts are weighted and fused according to confidence level to form a continuous, robust, and quantifiable forecast matrix. The output includes region ID, forecast time, passenger flow value, and confidence level, providing actionable input for dynamic diversion strategy generation. This achieves forecast continuity, robustness, and quantitative risk management. Specifically: Combine the prediction results: ; The output format includes: region ID, time step, predicted passenger flow, confidence level, and prediction type (short-term / medium-to-long-term). in, This represents the final predicted value of node i, which combines short-term and medium-to-long-term forecasts. This represents the weighting coefficient for short-term and medium-to-long-term forecasts, which is dynamically adjusted based on the confidence level of real-time observation data at each node and the historical forecast accuracy.

[0020] S3. Using short-term and medium-to-long-term spatiotemporal forecasts as core inputs, a closed-loop decision-making system is constructed through four sub-processes: micro-level entry control, probabilistic path guidance, capacity early warning and tiered intervention, and dynamic adjustment of reservation quotas. With prediction-driven control as the main thread, it integrates queue dynamics, probabilistic guidance algorithms, and causal feedback mechanisms to achieve proactive, tiered, explainable, and correctable dynamic control of the population. This ensures that the scenic area maintains a safe, smooth, and comfortable experience even under fluctuating visitor flow. The specific implementation process is as follows: S31. Based on short-term passenger flow forecasts and real-time queuing data at the entrance, a queue dynamics model is constructed to calculate the optimal release rate and dynamically control the gates using two rhythms: micro-release and batch release. A downstream congestion penalty is introduced into the control process to ensure that entrance release is linked to the internal carrying capacity, avoiding situations where "the entrance is clear but the interior is congested." This achieves precise risk suppression and flow regulation at the entrance, specifically: Based on the short-term forecast results from step S2, the expected arrival rate of each entrance is calculated to form a short-term inflow forecast for the next 30-60 minutes: ; By simulating changes in the inbound queue using dynamic equations and combining arrival rate with release rate, the real-time queue length is calculated, providing real-time feedback for the release strategy. ; Based on the surrounding terrain, evacuation capacity, and scenic area safety regulations, a queue safety threshold is set. To ensure that queuing at the entrance does not pose a safety hazard: ; The release rate is dynamically adjusted based on the predicted queue length and the risk of congestion in downstream areas, while also taking into account the waiting time of tourists: ; in, Indicates that entry point e is in the future prediction time window. Short-term forecast of total park visitor flow; This indicates the time step used for passenger flow forecasting and control, typically set to 30 seconds to 2 minutes. This represents the predicted arrival rate of entrance e at the current time t, which is the expected number of tourists entering the queuing area per unit time. This represents the actual number of people queuing at entrance e at time t; This represents the actual passage rate of entrance e at time t, i.e., the number of tourists allowed to pass through the gate per unit time. This indicates the maximum safe queuing size threshold allowed at entrance e; This represents the safety correction factor for entry point e; This indicates the rated maximum carrying capacity of the downstream area connected to inlet e; This represents the optimal entry release rate calculated after considering both the comprehensive prediction results and safety constraints. This indicates the average waiting time for the set entry target; This represents the predicted congestion penalty term for the downstream region at time t, which is calculated based on the short-term predicted passenger flow density and capacity threshold of the downstream region. This represents the response coefficient for adjusting the entry release procedure; and These represent the minimum and maximum entry release rates, respectively; S32. Tourists are grouped according to contextual features such as interests and time budget. Multiple candidate routes are matched to different groups through a probability allocation mechanism. The contextual bandit online learning model is used to continuously update the allocation probability based on congestion risk, experience benefits, and guidance spillover effects, inducing a "multiple selection ratio recommendation" approach. This effectively avoids concentrated influx to a single route and achieves flexible, targeted dispersal of the crowd. Specifically: Tourists are grouped according to their interests, real-time location, visit duration, and physical condition. Each group corresponds to a set of available routes. This grouping can be dynamically updated to ensure that the diversion strategy adapts to the real-time situation. ; ; Assigning multiple path selection probabilities to each group ensures a balanced distribution of tourists and reduces congestion concentration. ; ; Taking into account route experience, congestion risk, and environmental factors, a comprehensive reward score is calculated for each route, which is used for dynamic probability updates. ; By using online learning (such as the Contextual Bandit algorithm) to adjust the allocation probability in real time according to the reward function, the path guidance strategy can be adaptively optimized according to the real-time status of the scenic area. in, This refers to a group of tourists, each group having similar sightseeing characteristics; This represents the k-th tourist group; This represents the set of candidate tour routes applicable to group g, which is dynamically selected based on the scenic area's topology and currently accessible routes. This represents the probability of group g choosing path k at time t; Representing a path The experience utility value is comprehensively evaluated based on factors such as attraction rating, landscape quality, and walking intensity. Representing a path The congestion risk value at time t is calculated based on the short-term predicted passenger flow density of the area along the route. An index representing the imbalance in path allocation; , and These represent experience weight, risk weight, and equilibrium weight, respectively, and are calibrated based on historical performance. It should be noted that contextual bandit is a reinforcement learning framework used to optimize action selection by incorporating environmental context information during the decision-making process to maximize long-term gains. In each round of decision-making, the algorithm observes a context, selects an action, and receives a reward signal. The goal is to minimize cumulative regret, i.e., the lost gains compared to the ideal policy, by balancing exploration (trying new actions to gather information) and exploitation (selecting the currently known best action). S33. Calculate the regional risk index based on predicted values ​​and confidence levels, and set progressively lower thresholds to trigger intervention measures such as soft prompts, route redirection, entrance flow control, and temporary area closures. The priority of each intervention is adjusted based on the causal effect backtracking of the scene memory database to ensure maximum risk reduction with minimal impact without disrupting the visitor experience, achieving verifiable and progressive dynamic safety control. Specifically: Risk coefficients are calculated based on predicted passenger flow and capacity thresholds, while also considering prediction confidence levels to quantify the overload probability in each area: ; Different intervention measures are triggered based on the risk level: flexible prompts are provided for low-risk situations, while restrictions or access controls are implemented for high-risk situations. ; Interventions are implemented by combining electronic screens, broadcasts, and app push notifications to ensure that tourists receive timely guidance, and the effectiveness of the implementation is continuously monitored; in, Indicates region i in the future time Predicted passenger flow within the area; This represents the maximum safe carrying capacity of area i, determined by scenic area planning, safety assessment, and historical operating experience. This represents the prediction uncertainty for region i within the prediction time. The comprehensive overload risk index for region i at future time is calculated by combining predicted passenger flow, capacity, and prediction uncertainty. and These represent risk classification thresholds, corresponding to low, medium, and high risk levels, and are determined through historical accident data and simulation analysis. S34. Utilizing medium- and long-term forecasts and historical diversion effectiveness, automatically redistribute reservation quotas for future time periods based on regional capacity, terrain characteristics, and load trends at different entrances; fine-tune the quotas by combining the effectiveness of past interventions, creating a closed loop between "coarse adjustments" at the reservation level and "fine adjustments" at the venue level, thereby reducing peak flows at the source and ensuring a more balanced and controllable passenger flow structure for future time periods. Specifically: Adjusting short-term quotas (RL) involves quickly calculating reservation limits for the next few time periods online based on the difference between forecasts and actual traffic, balancing security, balance, and fairness. ; Optimize long-term strategy LP by learning appointment allocation strategies based on historical traffic patterns, seasonal factors, and event impacts, and automatically optimize long-term traffic distribution rules and incentive schemes to improve overall efficiency; By combining short-term LP and long-term RL strategies to form a closed loop, the short-term strategy ensures safety and immediate optimization, while the long-term strategy accumulates experience and achieves continuous optimization. The calculation results are synchronized to the time-sharing reservation system to update the available reservation quota in real time and notify tourists on the APP and the scenic spot's official platform. Where N represents the total number of core areas that need to be managed within the scenic area; H represents the reservation quota for region i in the future time period h; H represents the predicted time span for reservation adjustment, which is set according to the operating cycle and prediction reliability. and These represent the weighting of overloading penalties and the weighting of fairness, respectively. The index representing the dispersion or balance of reservation quotas is calculated from the reservation volume distribution in each region.

[0021] S4. Based on the stable feature output and intermediate analysis results formed in step S3, a comprehensive processing mechanism for final result generation and credibility verification is constructed. Through a closed-loop process of result quantitative modeling, multidimensional consistency verification, adaptive correction, output expression and feedback, the preliminary analysis results are transformed into a final output that is directly applicable, interpretable, and stable. This ensures that the overall method not only provides results, but also that the results are reliable, reproducible, and have engineering applicability. The specific implementation process is as follows: S41. Based on core feature parameters and state discrimination results, a quantitative expression model for results is constructed, which maps discrete or semi-structured intermediate outputs into continuous and comparable quantitative results. By introducing weight constraints and scale normalization mechanisms, information from different sources and with different dimensions can work synergistically within the same evaluation space. Specifically: Using a fixed rolling time window (e.g., 5 minutes or 10 minutes) as the cycle, the following status data are re-collected for each controlled area: real-time number of visitors in the area, changes in visitor density per unit area, average stay time of visitors in the area, proportion of path transfers between areas, and trend of changes in entrance queue length; To avoid misjudgments due to relying on a single indicator, the above multi-dimensional states are organized into a unified execution effect vector: ; in, This represents the split execution effect vector of region i at time t, used to quantify the actual effect of each dimension after the strategy is implemented; This represents the change in passenger flow density, characterizing the change in passenger flow per unit area in region i before and after the implementation of the strategy. It is derived from gate counting, Wi-Fi / Bluetooth probes, and video analysis. This represents the change in average length of stay, reflecting the change in the time tourists spend in the area. It is derived from trajectory analysis or video monitoring to determine whether the length of stay has increased or decreased. It represents the change in the proportion of path shifts, measures tourists' response to path guidance, is derived from real-time trajectory analysis, and reflects the actual impact of the strategy on path distribution; This represents the change in the length of the entrance queue, which comes from the gate queuing monitoring system. It measures the effectiveness of the entrance access strategy to ensure both safety and user experience. S42. Based on the preliminary quantification results, a multi-dimensional consistency verification process is introduced to cross-validate the results from multiple dimensions such as temporal continuity, spatial adjacency, and logical constraints. The focus is on identifying abnormal jumps, unreasonable offsets, or significant discrepancies with historical patterns. Specifically: Calculate the fundamental deviation between the forecast and the actual: ; A strategy causal contribution evaluation mechanism is introduced to analyze the correlation between deviation and the intensity of diversion measures: ; Among them, K i This represents the total number of diversion measures in effect in region i during the current time period; This represents the prediction deviation of region i at time t, used to determine the difference between the prediction result and the actual passenger flow; This indicates the real-time actual passenger flow in region i; This represents the passenger flow predicted by the short-term forecasting model, used to compare the model's estimate with the actual flow. The strategy contribution coefficient for region i represents the degree to which the diversion strategy contributes to the actual effect in that region. The historical effectiveness weight represents the actual effectiveness of various measures in historical operation. It is automatically generated through long-term statistics and is used to balance the contribution of different measures to the diversion effect. This indicates the intensity of the implementation of the k-th type of diversion measures (entry control, path guidance, capacity intervention) in region i; S43. For deviations or unstable factors identified during consistency verification, an adaptive correction mechanism is activated. Based on the source and scope of the deviation, the weights of relevant parameters or the local structure of the model are flexibly adjusted to gradually bring the results back to a reasonable range. Specifically: After identifying the source of the deviation, the collaborative correction phase begins, in which the prediction model parameters and the diversion strategy parameters are simultaneously incorporated into the adaptive adjustment framework to ensure the consistency and stability of the overall system behavior. For the prediction model, the model parameters are flexibly adjusted based on the prediction confidence level and the magnitude of the bias: ; For the traffic diversion strategy, it is locally modified based on the strategy contribution coefficient, such as adjusting the path induction probability: ; in, This represents the updated prediction model parameters, which have undergone adaptive correction. This represents the core parameters of the original prediction model, such as spatiotemporal weights and regional node influence coefficients. Indicates the learning rate; It represents the uncertainty of prediction, characterizes the reliability of the prediction result, and is derived from the uncertainty index output by the prediction model; This indicates the corrected path guidance weight; This represents the probability or weight of the k-th type of guidance measure corresponding to the original path g; This indicates the step size for strategy adjustments, controlling the magnitude of strategy updates and preventing excessive oscillations. The path reward function is derived from the prediction and diversion results in step S3, representing the success rate of path diversion or the effectiveness of guidance. S44. After completing the correction and confirming the stability of the results, the final results are output in a structured manner with feedback annotations. This includes not only the final numerical value or judgment conclusion, but also the result formation path and credibility indicators, providing a basis for subsequent applications or system self-learning. This output is then recorded in the system operation log, forming a closed-loop data foundation to support the continuous optimization of the overall method. Specifically: After completing short-term operational adjustments, the results of multiple operational cycles will be further fed back to the time-sharing reservation system to reduce the probability of similar congestion in the future. Historical risk levels for each region during different reservation periods will be statistically analyzed, and reservation quotas will be dynamically adjusted accordingly. ; in, This indicates the adjusted maximum number of reservation slots, which is adjusted based on historical risk levels. This indicates the original maximum number of reservation slots for region i during time period h; This represents the historical average risk level, derived from statistics on the effectiveness of traffic diversion implementation, prediction deviations, and strategy responses, reflecting the degree of congestion or safety risk in the region during that period. This represents the reservation adjustment sensitivity coefficient, used to control the adjustment range and prevent large fluctuations in reservation quotas.

[0022] Example 2, as Figure 2 As shown, the present invention proposes a scenic area intelligent control system based on time-sharing reservation and spatiotemporal prediction, which is used to execute a scenic area intelligent control method based on time-sharing reservation and spatiotemporal prediction proposed in Embodiment 1. It includes: a time-sharing reservation enhancement and multi-source data acquisition module, a spatiotemporal prediction module for fusing reservation data, a dynamic diversion strategy generation and execution module, and an execution feedback evaluation and reservation control self-optimization module.

[0023] The time-sharing reservation enhancement and multi-source data collection module is responsible for collecting and managing tourists' time-sharing real-name reservation information. At the same time, it integrates real-time data from multiple sources within the scenic area to provide a comprehensive data foundation for prediction and control. The system receives tourists' entry time slot selection, identity information, and personalized preferences through the official reservation platform, and establishes a reservation database. Furthermore, it collects real-time data on the number, location, and movement trajectory of tourists entering the park through multiple channels such as scenic area gates, surveillance cameras, Wi-Fi / Bluetooth probes, and base station signaling to achieve a panoramic perception of the current visitor flow status. The spatiotemporal prediction module, which integrates reservation data, uses reservation data and real-time sensing data as inputs to construct short-term and medium-to-long-term passenger flow prediction models: The short-term prediction model is based on the current passenger flow dynamics and uses a hybrid graph neural network and sequence learning model to analyze the changes in passenger flow density in various areas in the next half hour to one hour, thus identifying potential congestion areas in advance; The medium-to-long-term prediction model uses time-sharing reservation data to make multi-period predictions, which can cover the scenic area's passenger flow trends from several hours to several days, providing a predictive basis for resource allocation and diversion strategy formulation. The dynamic diversion strategy generation and execution module generates and executes diversion strategies in real time based on prediction results. Strategy generation includes three functions: entrance access control, route guidance, and capacity warning intervention. Entrance access control dynamically adjusts the gate release speed based on real-time visitor numbers and reservation information to prevent instantaneous congestion at the entrance. Route guidance provides visitors with the optimal real-time tour route through electronic screens, app push notifications, and park announcements, guiding a balanced distribution of visitor flow. Capacity warning intervention automatically triggers measures when the predicted visitor flow approaches a safe threshold, including suspending guidance to that area, prompting detours, or adding temporary guidance personnel. The module is also responsible for sending diversion instructions to each execution terminal and collecting execution feedback. The execution feedback evaluation and reservation control self-optimization module evaluates the effectiveness of the implemented diversion strategy in real time. It monitors passenger flow density, dwell time, route adoption rate, and entrance queue changes in each area through multi-dimensional indicators to form a quantitative vector of diversion effect. The system analyzes the prediction deviation and the actual contribution of the strategy, identifies the causes of prediction errors or strategy failures, and performs collaborative adaptive correction of prediction model parameters and diversion strategy parameters. At the same time, the module applies short-term operating experience back to the reservation system, dynamically adjusting regional reservation quotas and time slot allocations to achieve a match between long-term reservations and passenger flow carrying capacity. The module ensures that the system forms a closed-loop self-optimization mechanism from reservation-prediction-diversion-feedback-re-reservation.

[0024] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for intelligent regulation of scenic areas based on time-sharing reservation and spatiotemporal prediction, characterized in that, The specific implementation steps include the following: S1. A dynamic hierarchical arrival rate table is constructed by time-sharing real-name reservation and multi-source real-time perception data collection, and adaptively weighted and fused with real-time passenger flow observations preprocessed at the edge to form a fused passenger flow estimate. S2. Construct a spatiotemporal prediction model that combines graph neural networks and long short-term memory networks. Using the fused passenger flow estimate as input, the model can predict short-term and medium-to-long-term passenger flow in various areas of the scenic spot and output the prediction confidence level. S3. Using short-term and medium-to-long-term passenger flow forecasts as core inputs, a closed-loop decision-making system is formed through micro-level entry control, probabilistic path guidance, capacity-level early warning, and dynamic adjustment of reservation quotas to proactively regulate passenger flow distribution within the scenic area. S4. Conduct a multi-dimensional quantitative evaluation of the control effect of step S3, and based on the evaluation results, optimize the parameters and diversion strategy of the spatiotemporal prediction model in reverse through an adaptive correction mechanism, while simultaneously adjusting the reservation quota of the time-sharing reservation system.

2. The intelligent scenic area control method based on time-sharing reservation and spatiotemporal prediction according to claim 1, characterized in that, Step S1 specifically includes: Collect tourists' real-name time-sharing reservation information, which includes at least the reserved entry time and points of interest preferences; stratify the reservation data according to the reservation advance, mode of transportation and date type characteristics, generate and dynamically update the reservation-to-park conversion rate table, and record the actual probability of arrival at the park for each reservation stratum in different time windows; Real-time passenger flow data is collected by at least two of the following: gate counters, video sensing devices, and wireless probes; real-time passenger flow data is aggregated, trajectory extracted, anomaly detected, and noise denoised in edge processing units deployed at entrances and key nodes to generate regional real-time passenger flow observations. The expected passenger flow prior to reservations is calculated based on the reservation-to-park conversion rate table. The fusion coefficient is dynamically calculated based on the confidence level of real-time observation data and the uncertainty of the reservation prior. The fusion coefficient is used to weight and fuse the expected passenger flow prior to reservations with the real-time passenger flow observations of the region, and the fused passenger flow estimate of each region is output. The reservation data and real-time data are mapped to a unified scenic area spatial grid or functional area for time alignment and anomaly correction. The reservation-to-park conversion rate table and the integrated visitor flow estimate are updated in real time to form an online data stream.

3. The intelligent scenic area control method based on time-sharing reservation and spatiotemporal prediction according to claim 2, characterized in that, Step S2 specifically includes: The scenic area is divided into several functional area nodes. For each node, a feature vector is constructed that includes real-time fusion visitor flow estimation, reservation prior, area capacity, historical visitor flow patterns, and area type encoding. Based on the spatial proximity relationship between areas and the historical visitor movement probability, a spatial topology graph of the scenic area with edge weights is constructed. The feature vector sequence and the spatial topology map are input into a hybrid model consisting of a graph neural network and a first long short-term memory network. The graph neural network is used to aggregate the features of spatially adjacent nodes to capture spatial dependencies, and the first long short-term memory network is used to process the time series of node features to capture temporal evolution. The model outputs short-term passenger flow prediction values ​​and corresponding confidence levels for each region within the next 30 to 60 minutes. Based on the prior probability of reservations and historical passenger flow patterns provided by the reservation-to-park conversion rate table, a second long short-term memory network is used to model and predict the medium- and long-term passenger flow trends of various areas in the next few hours to days. Based on the confidence level of short-term passenger flow forecasts and the accuracy of historical forecasts, the weighting coefficients of short-term and medium-to-long-term forecast results are dynamically determined, and the two are weighted and merged to form the final regional forecast passenger flow matrix.

4. The intelligent scenic area control method based on time-sharing reservation and spatiotemporal prediction according to claim 3, characterized in that, The specific details of the inlet micro-release control in step S3 are as follows: Calculate the predicted arrival rate of each entrance in the future time window based on the short-term passenger flow forecast results; Establish an entrance queue dynamics model, combine the predicted arrival rate and real-time release rate, and calculate and update the real-time entrance queue length; The optimal passage rate of each entrance gate is dynamically calculated based on the preset maximum safe queuing number threshold, the predicted congestion penalty in the downstream area, and the expected average visitor waiting time. The gate is controlled to perform micro-release or batch release operations based on the optimal release rate.

5. A method for intelligent regulation of scenic areas based on time-sharing reservation and spatiotemporal prediction according to claim 4, characterized in that, The probabilistic path induction in step S3 specifically involves: Tourists are dynamically segmented based on their interests, real-time location, and the context of their visit. For each group of tourists, several candidate tour routes are selected from the topology of the scenic area; Construct a path reward function that integrates at least the static experience utility of the path, the real-time congestion risk based on short-term predictions, and the path allocation balance index. The contextual bandit online learning model is adopted to dynamically update the selection probability of each candidate path for each tourist group based on the real-time calculated path reward. Route guidance information is disseminated to tourists through electronic displays, park announcements, or mobile applications, using a multiple-selection recommendation method.

6. The intelligent scenic area control method based on time-sharing reservation and spatiotemporal prediction according to claim 5, characterized in that, The capacity tiered early warning in step S3 is specifically as follows: Based on the final predicted passenger flow, the maximum safe carrying capacity of each region, and the prediction confidence level, the comprehensive overload risk index of each region is calculated. Set three levels of early warning thresholds: low risk, medium risk, and high risk. When the overall overload risk index reaches the low-risk threshold, a soft reminder message is sent to tourists. When the medium-risk threshold is reached, a path redirection instruction is triggered and linked with the probabilistic path induction; when the high-risk threshold is reached, a flow restriction instruction for the associated entry point or a temporary blockade instruction for the target area is triggered.

7. The intelligent scenic area control method based on time-sharing reservation and spatiotemporal prediction according to claim 6, characterized in that, The dynamic callback of the reservation quota in step S3 is as follows: Based on the medium- and long-term passenger flow trend forecast results, analyze the passenger flow carrying capacity pressure of different areas in different time periods in the future; With the goal of reducing the risk of regional overloading and balancing the overall spatial load, a reservation quota optimization model was established; The optimization model is used to dynamically adjust the upper limit of the number of reservation slots available in the time-sharing reservation system for each time period, each entrance, and each area of ​​interest in the future. The adjusted reservation quota will be synchronized to the reservation platform in real time, and the public reservation interface will be updated.

8. A method for intelligent regulation of scenic areas based on time-sharing reservation and spatiotemporal prediction according to claim 7, characterized in that, The multi-dimensional quantitative evaluation in step S4 specifically includes: Within the preset evaluation time window, real-time changes in passenger flow density, average visitor stay duration, visitor route transfer ratio, and entrance queue length were collected for each controlled area. The above four variables are combined to form a multi-dimensional diversion execution effect vector, which is used to quantitatively characterize the comprehensive effect of the diversion strategy in terms of spatial load, time delay, path response and entry order.

9. A method for intelligent regulation of scenic areas based on time-sharing reservation and spatiotemporal prediction according to claim 8, characterized in that, The adaptive correction mechanism described in step S4 is as follows: Calculate the deviation between the predicted and actual passenger flow values ​​for each area, and evaluate the prediction model error and strategy contribution by combining the implementation intensity of each diversion measure; Based on the deviation and prediction uncertainty, the parameters of the hybrid model of graph neural network and long short-term memory network are fine-tuned using gradient descent. Based on the strategy contribution and path reward feedback, local adjustments were made to the path selection probability distribution in probabilistic path induction and the parameters of the release rate calculation formula in the entry micro-release control. The historical risk levels of each region, obtained from long-term assessment and statistics, are fed back to the dynamic adjustment process of the reservation quota, which is used to dynamically correct the adjustment coefficient in the reservation quota calculation formula.

10. A scenic area intelligent control system based on time-sharing reservation and spatiotemporal prediction, used to execute the scenic area intelligent control method based on time-sharing reservation and spatiotemporal prediction as described in any one of claims 1 to 9, characterized in that, include: The time-sharing reservation enhancement and multi-source data acquisition module is used to collect and manage time-sharing real-name reservation information, and integrate multi-source real-time sensing data from turnstiles, videos and wireless probes to complete data fusion processing; The spatiotemporal prediction module, which integrates reservation data, is used to build and run a hybrid model of graph neural network and long short-term memory network. Based on the output of the time-sharing reservation enhancement and multi-source data acquisition module, it performs short-term and medium-to-long-term passenger flow prediction. The dynamic diversion strategy generation and execution module is used to generate an entrance release control strategy, a probabilistic path guidance strategy, and a capacity classification early warning instruction based on the prediction results of the spatiotemporal prediction module of the fused reservation data, and send them to the corresponding execution terminals. The execution feedback evaluation and reservation control self-optimization module is used to perform multi-dimensional quantitative evaluation of the effectiveness of the diversion strategy, adaptively correct the prediction model parameters and diversion strategy parameters based on the evaluation results, and adjust the reservation quota of the time-sharing reservation system in conjunction with the evaluation results.

Images