Method and system for monitoring and decision support of regional tourism industry operation

Abstract

The application discloses a regional tourism industry operation monitoring and decision support method and system, and belongs to the technical field of tourism industry operation management, and comprises the following steps: obtaining convergent concurrent load time sequence data of each regional monitoring node; obtaining physical passenger flow characteristic vectors of each regional monitoring node; generating a space structure characteristic matrix; detecting data deviation anomalies of the convergent concurrent load time sequence data and dynamically adjusting fusion weights, synthesizing comprehensive state characteristic vectors of each regional monitoring node; constructing a state conduction graph model and outputting prediction result data for triggering regional resource scheduling; and through the establishment of a base station-node mapping relationship and the introduction of a spatial overlap area weight distribution mechanism, the application realizes the accurate attribution of load data in a physical space. In combination with characteristic vectors constructed in dimensions such as spatial distribution entropy, the application can comprehensively depict the regional state from four dimensions of total amount, stability, dispersion and abnormal impact, and greatly reduces the system false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of tourism industry operation and management technology, and in particular to a method and system for monitoring and supporting the operation of regional tourism industries. Background Technology

[0002] The proposal of methods and systems for monitoring and decision support in the regional tourism industry stems from the significant complexity, dynamism, and spatial coupling characteristics exhibited in the tourism industry's operation. With the development of smart tourism and the digital economy, tourism activities are no longer limited to single scenic spots or service providers, but involve the coordinated operation of multiple sectors such as transportation, communication, accommodation, catering, and entertainment, forming a cross-regional, multi-node interconnected industrial network. Current technologies for monitoring the operational status of the tourism industry typically rely on statistical reports, sampling surveys, or analysis of single data sources, such as estimating visitor flow based on data from scenic spot gates, ticketing systems, or mobile signaling. These methods, on the one hand, are mostly post-hoc statistics at the time granularity, making it difficult to reflect real-time trends; on the other hand, they are often limited to single points or local areas at the spatial dimension, lacking the ability to depict the overall interconnected relationships within the region. Furthermore, differences in collection criteria and time series exist between different data sources, leading to information biases during multi-source data fusion, which in turn affects the accurate judgment of actual visitor flow and the operational status of the industry.

[0003] As the application of communication network data (such as base station concurrent load and channel utilization) in characterizing crowd activity deepens, using communication-side data to reflect regional passenger flow dynamics has become an important technical approach. However, existing methods mostly focus on load analysis of single base stations or simple spatial aggregation, lacking modeling of passenger flow relationships between base stations and failing to reveal the transmission paths and directions of passenger flow in space. Furthermore, existing technologies lack effective anomaly identification and dynamic weight adjustment mechanisms for potential temporal discrepancies between passenger flow data and service operation indicators, resulting in insufficient stability of the fusion results.

[0004] At the decision support level, traditional methods rely heavily on empirical rules or static thresholds to trigger resource scheduling, lacking the ability to predict future conditions and making it difficult to respond promptly to sudden changes in passenger flow or structural load imbalances. Especially during holidays, emergencies, or peak events, passenger flow within a region may exhibit rapid migration and concentrated bursts. If trend prediction and linkage analysis cannot be performed in advance, it can easily lead to local resource overload or insufficient service capacity.

[0005] For example, Chinese invention patent CN114723480B discloses a method for predicting passenger flow and a cargo dispatching system for rural tourism, relating to the field of rural tourism informatization. This invention estimates passenger flow through a scenic spot passenger flow estimation step based on short-term data, deployed using existing base station nodes or routing nodes, incurring no hardware deployment costs. It accurately obtains passenger flow information within a small area, with high accuracy in short-term data, thereby improving the accuracy of small-scale, short-term predictions. The invention combines the movement trajectories of various mobile devices; selects the frequency of occurrence of each monitoring node in the node switching sequence of each movement trajectory; designates the monitoring block containing the monitoring node with the highest frequency of occurrence as the location of the cargo distribution station, and establishes stations within this area. By using the passenger flow of each scenic spot as a weight value and constructing a weighted center offset formula, it selects the theoretically most suitable station for construction, achieving rapid cargo dispatching and reducing dispatching mileage.

[0006] For example, the tourism data survey and monitoring system disclosed in Chinese invention patent CN104809634B includes a media survey platform, a tourism economic operation data tracking subsystem, a tourism industry dynamic monitoring subsystem, a tourism financial information database, and a media sample library operation and management platform. The media survey platform provides a workflow platform for managing and publishing research projects for browsers and clients; the tourism economic operation data tracking subsystem collects and tracks online public information and proactive survey data; the tourism industry dynamic monitoring subsystem dynamically monitors the data; the tourism financial information database provides a retrieval database for data survey and monitoring; and the media sample library operation and management platform stores samples in the tourism survey sample library to support periodic and follow-up survey projects.

[0007] However, in the process of implementing the technical solutions in the embodiments of the present invention, it was found that the above-mentioned technology has at least the following technical problems:

[0008] In existing technologies, passenger flow monitoring methods based on communication-side data mostly remain at the level of load analysis at a single base station or simple spatial aggregation, focusing primarily on passenger flow monitoring in a single scenic area or local nodes. They treat tourism elements (transportation, accommodation, catering, etc.) as isolated statistical indicators, severely lacking the ability to depict the overall interrelationships within a region. Furthermore, when business data suffers from inaccurate reporting, equipment malfunctions, or inconsistent timing, there is a lack of dynamic evaluation methods for the reliability of various data sources, resulting in insufficient stability of the fusion results and an inability to accurately reflect the true operational status of the tourism industry. Summary of the Invention

[0009] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for regional tourism industry operation monitoring and decision support. The technical solution is as follows:

[0010] Methods for monitoring and supporting the operation of the regional tourism industry include:

[0011] Based on the preset geofence boundaries, regional monitoring nodes are delineated, concurrent load time series data of each base station within the target area are obtained, and a mapping relationship between the base station and the regional monitoring nodes is established according to the spatial affiliation relationship between each base station and the geofence. The concurrent load time series data of each base station is aggregated to the corresponding regional monitoring nodes to obtain the aggregated concurrent load time series data of each regional monitoring node.

[0012] The time-series data of the aggregated concurrent load from each monitoring node in each region are periodically decomposed to extract load distribution characteristics and combine them to form the physical passenger flow feature vector of each monitoring node in each region.

[0013] Based on the spatial correlation between the concurrent load time series data of each base station, a base station network topology map reflecting the direction and path of passenger flow is constructed, and the base station network topology map is aggregated and mapped to the regional monitoring node level to generate a spatial structure feature matrix.

[0014] The system calculates the temporal correlation between the derived operational indicator sequence extracted from the preset business system and the physical passenger flow feature vector, detects data deviation anomalies in the aggregated concurrent load time series data, dynamically adjusts the fusion weights based on the anomaly detection results, and synthesizes the comprehensive status feature vector of each regional monitoring node. The derived operational indicator sequence refers to the time series data of statistics extracted from the business system that can reflect the operational status of the tourism industry.

[0015] Using the spatial structure feature matrix and the comprehensive state feature vector as input, a state transmission graph model is constructed, multi-step temporal evolution prediction is performed, and prediction result data is output to trigger regional resource scheduling.

[0016] In addition, it provides a regional tourism industry operation monitoring and decision support system, including:

[0017] The aggregation module is used to delineate regional monitoring nodes based on preset geofence boundaries, acquire concurrent load time series data of each base station within the target area, and establish a mapping relationship between base stations and regional monitoring nodes based on the spatial affiliation relationship between each base station and the geofence. The concurrent load time series data of each base station is aggregated to the corresponding regional monitoring node to obtain the aggregated concurrent load time series data of each regional monitoring node.

[0018] The combination module is used to perform periodic decomposition processing on the aggregated concurrent load time series data of monitoring nodes in each region, extract load distribution characteristics, and combine them to form the physical passenger flow feature vector of each monitoring node in each region.

[0019] The matrix construction module is used to construct a base station network topology map that reflects the direction and path of passenger flow based on the spatial correlation between the concurrent load time series data of each base station, and to aggregate and map the base station network topology map to the regional monitoring node level to generate a spatial structure feature matrix.

[0020] The anomaly detection module is used to calculate the temporal correlation between the derived operation index sequence extracted from the preset business system and the physical passenger flow feature vector, detect data deviation anomalies in the aggregated concurrent load time series data, and dynamically adjust the fusion weights based on the anomaly detection results to synthesize the comprehensive status feature vector of each regional monitoring node. The derived operation index sequence refers to the time series data of statistics extracted from the business system that can reflect the operation status of the tourism industry.

[0021] The output module is used to construct a state transmission graph model with the spatial structure feature matrix and the comprehensive state feature vector as input, perform multi-step temporal evolution prediction, and output prediction result data for triggering regional resource scheduling.

[0022] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0023] 1. The regional tourism industry operation monitoring and decision support method provided by this invention establishes a base station-node mapping relationship through ray statistics of geofence boundaries and base station coordinates, and introduces a spatial overlap area weight allocation mechanism, effectively solving the problem of load aliasing caused by cross-regional base station signal coverage, and realizing the accurate physical spatial attribution of load data. Based on this, through daily, weekly, and yearly multi-level periodic stripping technology, regular background noise such as commuting peak hours and weekend travel is eliminated, enabling the residual signal to accurately pinpoint sudden surges in passenger flow. Combined with a "digital fingerprint" feature vector constructed from dimensions such as spatial distribution entropy, it can comprehensively characterize the regional state from four dimensions: total volume, stability, dispersion, and abnormal impacts, greatly reducing the system's false alarm rate.

[0024] 2. This invention, without acquiring any user privacy data such as IMSI / IMEI, utilizes the similarity of concurrent load waveforms of base stations and calculates the cross-correlation function under different hysteresis values ​​to identify the echo effect of crowds moving from location A to location B. This mechanism not only quantifies the intensity of passenger flow transmission but also accurately determines the flow direction by the order of peak activation, thus reversing the dynamic migration path of the crowd at the level of physical communication signals. The construction of this directed weighted base station network topology map extracts massive amounts of raw communication beacons into a minimally simplistic spatial structure matrix, achieving in-depth analysis of the logic of passenger flow transmission between regions while ensuring data security and compliance.

[0025] 3. This invention addresses the potential risks of response delays and time-series discrepancies between physical passenger flow and operational indicators (such as catering and ticketing) in the tourism industry by introducing a sliding window cross-correlation detection mechanism. The system can identify abnormal deviations of derived indicators relative to physical signals in real time based on historical benchmark intervals and dynamically adjust the fusion weights accordingly. When information deviations occur in the business system due to data entry delays, equipment malfunctions, or human error, the system automatically reduces the fusion weight of that indicator, ensuring that high-reliability physical passenger flow data is always used as the benchmark. This mechanism effectively enhances the system's data robustness and the accuracy of monitoring conclusions in complex operating environments.

[0026] 4. This invention simulates the spatiotemporal evolution of upstream node state fluctuations to downstream nodes through the transmission weights defined by the spatial structure feature matrix. Multi-step time-series prediction using an autoregressive iterative deduction strategy can predict passenger flow migration trends and load changes over the next few hours. This prediction data can directly trigger multi-dimensional resource coordination and scheduling across security, transportation, and infrastructure, shifting the management model from reactive emergency response to proactive prevention, effectively resolving the structural contradiction between local resource overload and service capacity imbalance during holidays and emergencies. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the regional tourism industry operation monitoring and decision support method provided in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of a regional tourism industry operation monitoring and decision support system module provided in an embodiment of the present invention.

[0030] Figure 3 This is a schematic diagram of the logical flow of the regional tourism industry operation monitoring and decision support system provided in an embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that embodiments of the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention.

[0032] It should be understood that the accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the description of the embodiments of the present invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "this embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects.

[0033] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0034] Please see Figure 1 The diagram shown is a flowchart of the regional tourism industry operation monitoring and decision support method in an embodiment of the present invention.

[0035] like Figure 3 The diagram shown illustrates the logical flow of regional tourism industry operation monitoring and decision support in this embodiment of the invention. It demonstrates how this embodiment achieves automated transformation from underlying raw data to high-level decision instructions through five core steps, specifically including:

[0036] Spatial mapping stage. The system collects concurrent load data from base stations and matches it with preset geofences. For base stations whose coverage spans multiple geofences, the system performs load splitting and weight allocation based on the proportion of signal overlap area, ensuring that each piece of communication data can be accurately assigned to the corresponding service area.

[0037] Feature extraction stage. The aggregated time-series data undergoes periodic stripping to filter out normal daily, weekly, and yearly patterns. In this way, the system can identify net anomalies after removing the normal background, thereby pinpointing passenger flow disturbances truly caused by sudden events or random factors.

[0038] In the topology modeling stage, the system filters candidate connections based on the coverage relationships and geographical locations of base stations, and uses the echo effect to calculate the cross-correlation of the load waveforms of each base station. By analyzing the order and similarity of waveform peaks, the system inverts the flow direction and transmission path of passenger flow without infringing on personal privacy, and finally generates a spatial structure feature matrix describing the degree of connection between areas.

[0039] The intelligent fusion process involves standardizing the metrics from the business system and comparing them with the previously extracted physical passenger flow characteristics. The system monitors in real time whether there are any temporal discrepancies between the two types of data and dynamically adjusts the fusion weights of the business data accordingly, ultimately synthesizing a multi-dimensional comprehensive state feature vector.

[0040] Predictive scheduling stage. The system inputs the spatial structure matrix and comprehensive state characteristics into the transmission model to perform multi-step time-series evolution simulation. By analyzing the magnitude and trend of passenger flow load changes over a future period, the system generates structured prediction results and directly triggers corresponding resource scheduling schemes, such as adjusting security deployments, increasing traffic frequency, or issuing diversion guidance instructions.

[0041] Methods for monitoring and supporting the operation of the regional tourism industry, specifically including:

[0042] Based on the preset geofence boundaries, regional monitoring nodes are delineated, concurrent load time series data of each base station within the target area are obtained, and a mapping relationship between the base station and the regional monitoring nodes is established according to the spatial affiliation relationship between each base station and the geofence. The concurrent load time series data of each base station is aggregated to the corresponding regional monitoring nodes to obtain the aggregated concurrent load time series data of each regional monitoring node.

[0043] Concurrent load time-series data includes direct load metrics and indirect load metrics. Direct load metrics include at least the number of simultaneously online device connections, physical resource block utilization, and uplink and downlink throughput.

[0044] Concurrent load time-series data refers to a set of multi-dimensional signals that can continuously reflect the resource occupancy status of communication networks over time. It consists of direct and indirect indicators and serves as the underlying data source for subsequent status analysis.

[0045] The number of simultaneously connected devices refers to the number of terminals served by a base station at a certain point in time, measured in units.

[0046] Physical resource block utilization refers to the proportion of wireless physical channels occupied, expressed as a percentage.

[0047] Uplink and downlink throughput refers to the amount of data transmitted uplink and downlink per unit time, measured in Mbps.

[0048] Indirect load metrics include at least the number of handovers, location update frequency, and channel quality metrics.

[0049] The number of handovers refers to the number of times a user switches from one base station to another, reflecting user mobility.

[0050] Location update frequency refers to how frequently a user's location information is refreshed within the network, and is used to analyze movement trajectories.

[0051] Channel quality metrics are signal quality parameters that measure the state of a wireless channel and affect throughput.

[0052] Direct load indicators are technical parameters that reflect the absolute consumption of hardware resources and network bandwidth, while indirect load indicators are derived parameters that reflect the mobility of the target group and the status of environmental signals.

[0053] The combination of these two factors enables concurrent load time-series data to fully describe the network state from both static consumption and dynamic trend dimensions.

[0054] The geographic coordinates of all base stations within the target area are obtained in advance. At the same time, several regional monitoring nodes are pre-defined according to business management requirements. Each regional monitoring node corresponds to a geofence polygon defined by boundary coordinates.

[0055] The regional monitoring node is a logical computing unit (corresponding to a real business or geographical block) set in the embodiments of the present invention. Its spatial range is abstracted in the system as a geofence polygon formed by a series of latitude and longitude coordinate points.

[0056] A ray is emitted from the base station coordinates in a fixed direction. The number of intersections between the ray and the fence boundary is counted. If the number of intersections is odd, the base station is determined to be inside the fence and assigned to the corresponding area monitoring node. If the number of intersections is even, the base station is determined to be outside the fence.

[0057] For a base station whose coverage spans multiple fence boundaries, calculate the spatial overlap area between the base station's coverage area and each fence polygon. Use the proportion of each overlap area to the total overlap area as the allocation weight, and then include the concurrent load time series data of the base station into the corresponding area monitoring nodes according to the allocation weight.

[0058] Taking a city operation monitoring scenario as an example, a 5G macro base station is deployed at the intersection of a scenic area and a surrounding commercial district in the city, with a total coverage area of ​​0.5 square kilometers. Spatial calculations show that 0.3 square kilometers of the base station's area falls within the scenic area monitoring node, and 0.2 square kilometers falls within the surrounding commercial district monitoring node, corresponding to allocation weights of 60% and 40%, respectively. When the base station detects 1200 online device connections and 450Mbps downlink throughput at a certain moment, the system will, according to weighted logic, assign 720 connections and 270Mbps of traffic to the scenic area node, and the remaining 480 connections and 180Mbps of traffic to the surrounding commercial district node.

[0059] For a base station whose coverage area falls entirely within a certain fence, its concurrent load data is completely attributed to the monitoring node in that area.

[0060] After determining the attribution, the concurrent load time series data of all attribution base stations within the same monitoring node are weighted and summed according to their respective weights to obtain the aggregated concurrent load time series data of that node.

[0061] This attribution method solves the problem of load aliasing caused by base station coverage across regions, and achieves accurate spatial allocation of load data.

[0062] The time-series data of the aggregated concurrent load from each monitoring node in each region are periodically decomposed to extract load distribution characteristics and combine them to form the physical passenger flow feature vector of each monitoring node in each region.

[0063] For the concurrent load time series data of each monitoring node, the daily cycle component, weekly cycle component and annual cycle component are identified and extracted in sequence. After subtracting the above cycle components from the original concurrent load time series data in sequence, the remaining residual signal is the net abnormal passenger flow component.

[0064] The daily cycle represents the daily routine of alternating day and night, the weekly cycle represents the rhythm difference between weekdays and rest days, and the annual cycle represents the changes in macroeconomic benchmarks brought about by seasonal changes or long holidays.

[0065] Net abnormal passenger flow component refers to the signal disturbances that remain after removing all regular fluctuations, which are purely caused by sudden events or random factors.

[0066] Based on the completion of the periodic decomposition, load distribution characteristics are extracted for each regional monitoring node. The load distribution characteristics include at least the periodic dimension characteristics that characterize the periodic operation pattern, the distribution dimension characteristics that characterize the statistical distribution characteristics of the data, and the anomaly dimension characteristics that characterize irregular fluctuations.

[0067] The cycle dimension features are specifically daily cycle components, weekly cycle components, and annual cycle components.

[0068] The specific distribution dimension features are absolute load level, load volatility, and spatial distribution entropy.

[0069] Absolute load level represents the average passenger flow intensity in a region. Load volatility represents the intensity of volatility.

[0070] Spatial distribution entropy is a numerical value used to measure the uniformity of load distribution within a region. This value is high when the load is evenly distributed throughout the region; it decreases significantly when the load is highly concentrated in a few hotspot areas.

[0071] The specific abnormal dimension features are net abnormal passenger flow component and load duration.

[0072] The load distribution characteristics are spliced ​​together in a preset order to obtain the physical passenger flow characteristic vector of each monitoring node in the current time window. This vector can represent a set of standard values ​​of the current state of the area, similar to the digital fingerprint of the area.

[0073] A subtractive coupling exists between the original signal and the three periodic components. This stripping operation is the inverse of each other, ensuring the conservation of data during the decomposition process and enabling the residual signal to accurately represent irregular fluctuations. Distributional and periodic characteristics together constitute the static background of the region. Absolute load level provides a scale for the total load, load volatility provides a measure of stability, and spatial distribution entropy extends the perspective from time to space, compensating for the geographical dispersion of the load. These three dimensions corroborate each other, distinguishing between load increases caused by overall population growth and load fluctuations caused by localized population clustering. Anomaly characteristics (such as residual intensity and duration) are merged with the aforementioned periodic and distributional characteristics in a predetermined order. This arrangement ensures that when the computer reads the data, it can simultaneously grasp the background patterns, current statistical distribution, and instantaneous abnormal shocks of the region, thereby achieving comprehensive state perception.

[0074] By stripping away daily, weekly, and yearly cycles, the system can effectively filter out false anomalies caused by normal behaviors such as commuting rush hours and weekend outings. This preprocessing method greatly reduces the system's false alarm rate, ensuring that genuine sudden risks are exposed against a clean background.

[0075] Based on the spatial correlation between the concurrent load time series data of each base station, a base station network topology map reflecting the direction and path of passenger flow is constructed, and the base station network topology map is aggregated and mapped to the regional monitoring node level to generate a spatial structure feature matrix.

[0076] Based on the spatial correlation between the concurrent load time-series data of each base station, a base station network topology map reflecting the direction and path of passenger flow is constructed, specifically including:

[0077] Each base station within the target area is considered a node in the base station network topology.

[0078] The base station network topology consists of two basic elements: nodes and edges. A node is an abstract mapping of a real physical entity. In this embodiment of the invention, each mobile communication base station actually deployed within the target area corresponds to an independent node in the base station network topology. Nodes themselves do not carry directional attributes; they only serve as the basic carrier for subsequent edge construction and attribute assignment.

[0079] The system acquires the geographical location information of each base station within the target area, retrieves the latitude and longitude data of all base stations from the database, calculates the Euclidean distance between base stations, and identifies base station pairs with a geographical distance less than a preset distance threshold as initial adjacent base stations. Initial edges connecting these pairs are then established to form a candidate edge set. Euclidean distance refers to the straight-line distance between two points in a two-dimensional plane coordinate system. In geographic applications, it is calculated by projecting the latitude and longitude coordinates of the base stations onto a Cartesian coordinate system, and the result is approximately equal to the actual horizontal distance between two points on the Earth's surface. It is important to note that the initial edges established here are undirected and unweighted candidate connections, and do not yet possess direction or strength attributes.

[0080] A base station network topology diagram is a structural model that simulates the logical relationships between base stations. It is important to emphasize that the base station network topology diagram simulates the logical relationships between base stations, not the physical cabling or communication protocol connections.

[0081] The preset distance threshold is set based on the average coverage radius of the base stations. The preset process consists of three stages: data acquisition, parameter estimation, and scene classification. In the data acquisition stage, the system extracts engineering parameters such as antenna height, transmit power, antenna gain, and downtilt angle of all base stations within the target area from the operator's network planning database. These parameters collectively determine the actual signal coverage range of the base stations. In the parameter estimation stage, the effective service radius of each base station is calculated based on traditional wireless signal propagation models (such as the Okumura model or the COST-Hata model), and the average effective service radius of all base stations within the target area is used as the reference value for the average coverage radius of that area. In the scene classification stage, the system identifies the urban / rural attributes of the target area: for densely built-up urban environments with small base station spacing, the coverage areas of adjacent base stations already overlap significantly, and setting the threshold to approximately 500 meters can effectively cover most of the actual adjacency relationships; for sparsely built-up suburban environments with large base station spacing, the base station coverage radius is larger, and the threshold is increased to one kilometer to ensure the integrity of the candidate edge set. If the coverage areas of two base stations do not intersect spatially at all, tourists will inevitably be associated with several intermediate base stations during their movement between the two locations. Therefore, there is no direct passenger flow connection between the two, and the corresponding candidate edges do not need to be established.

[0082] Since not all base station pairs have direct passenger flow relationships, candidate edges are first filtered based on geographical adjacency: candidate connections are established only between base station pairs whose coverage areas overlap or are geographically adjacent, excluding base station pairs that are geographically far apart and have an extremely low probability of direct passenger flow. After the above filtering, a set of candidate edges for the base station network topology is obtained.

[0083] Geographical overlap refers to the spatial intersection of signals emitted by two base stations. Geographical adjacency refers to the proximity of the signal edges of two base stations, forming a continuous coverage chain. The determination of screening criteria relies on the engineering parameters of the base stations. By analyzing the antenna downtilt angle and transmission power, the effective service radius of each base station is calculated, thereby accurately defining the overlapping area. Even if the Euclidean distance between two base stations meets the distance threshold requirement, it does not necessarily mean that there is an effective passenger flow transmission relationship between them, because physical barriers such as mountains, rivers, and building complexes may exist in the actual terrain, preventing tourists from traveling directly between the two locations despite their short straight-line distance. By introducing coverage overlap and geographical adjacency as a composite screening criterion, such physically inaccessible base station pairs can be further eliminated, making the candidate edge set closer to the real-world population mobility network.

[0084] For each pair of adjacent base stations in the candidate edge set, the cross-correlation function curve between their concurrent load time series data under different lags is calculated. Specifically, the cross-correlation coefficients of the concurrent load time series data of the two base stations under different time lags are calculated to obtain the cross-correlation function curve of the base station pair. The cross-correlation function curve is used to measure the similarity between two time series at different time offsets. In this embodiment of the invention, it is used to find the echo effect between load fluctuations of two base stations.

[0085] The cross-correlation function is a classic mathematical tool in signal processing used to measure the degree of linear correlation between two time series as time shifts. In this embodiment of the invention, its calculation process can be understood as follows: the load time series curve of base station B is gradually shifted to the right on the time axis with a fixed step size. At each shift position, the Pearson correlation coefficient between the shifted base station B curve and the stationary base station A curve is calculated. Connecting the correlation coefficients corresponding to all shift positions forms the cross-correlation function curve of the base station pair. The cross-correlation coefficient ranges from negative one to positive one. The closer the value is to positive one, the more synchronized the fluctuation patterns of the two time series curves are under that lag.

[0086] The echo effect refers to the phenomenon in tourist traffic scenarios where, after a group of people moves from the coverage area of ​​base station A to the coverage area of ​​base station B, the load waveform of base station B will reproduce the fluctuation characteristic of the load waveform of base station A after a certain time delay, much like the echo produced by sound propagating in space. This echo effect is manifested on the cross-correlation function curve as a peak value significantly higher than the background noise level at the lag corresponding to the actual movement time of the crowd. The discovery of the echo effect is the core technical principle behind this method for inferring the movement path of crowds from communication signals.

[0087] Hysteresis represents the offset of one signal relative to another on the time axis. In a passenger flow scenario, it represents the average time required for a group of people to move from base station A to base station B. The unit of measurement for hysteresis is consistent with the sampling interval of concurrent load time-series data. If the system collects base station load data with a sampling granularity of five minutes, then each unit of hysteresis corresponds to five minutes in reality. The hysteresis corresponding to the peak value of the cross-correlation function directly reflects the typical time required for a group of people to move between the coverage areas of two base stations. This time comprehensively reflects the influence of multiple factors such as walking distance between the two locations, mode of transportation, route terrain, and internal organization and guidance within the scenic area.

[0088] If the cross-correlation function curve has a significant peak at a certain lag, it is determined that there is a statistically significant passenger flow correlation between the two base stations, and the candidate edge is retained in the base station network topology graph.

[0089] The presence of a significant peak indicates a stable correlation between the load fluctuations of the two base stations, exceeding the level of random background noise. This stable correlation is driven by genuine directional flow behavior of people. If there is no regular flow of people between the two areas, the load time-series data of the two base stations should be nearly statistically independent at all lags, and the cross-correlation function curves will fluctuate randomly around zero at each lag, without forming a stable peak at any specific location. Conversely, if there is regular passenger flow, the corresponding peak will repeat in multiple sampling periods, and its amplitude will significantly exceed the upper limit of random background noise.

[0090] If the cross-correlation function has no significant peak value under all lags, it is determined that there is no obvious passenger flow correlation between the two base stations, and the candidate edge is deleted.

[0091] The criteria for determining a significant peak value are:

[0092] A peak value is defined as a significant peak value when the magnitude of the cross-correlation function exceeds a preset significance threshold.

[0093] The process of setting the significance threshold is a statistical calibration process based on historical data. Specifically, it involves selecting time periods from the historical database where there are no large-scale passenger flows in the target area, such as late nights when the scenic area is closed or weekdays with extremely low passenger flow outside of statutory holidays. The concurrent load time series data of base stations during these periods are then extracted as background noise samples.

[0094] For each candidate base station pair in the background noise sample, calculate the cross-correlation function, collect all cross-correlation values, and construct its statistical distribution.

[0095] The upper quantile of this statistical distribution (usually the 95th or 99th quantile, the specific value of which is determined based on the trade-off between target precision and recall) is taken as the upper limit estimate of background random association.

[0096] Set this upper limit value as the significance threshold and write it into the system parameter library.

[0097] For each retained base station network topology edge, the activation order of the peak concurrent load time-series data of the two base stations is analyzed to obtain the direction of passenger flow. Specifically, if the peak concurrent load of base station A is earlier than that of base station B, it is determined that the passenger flow is from the coverage area of ​​base station A to the coverage area of ​​base station B, and the direction of this edge in the base station network topology diagram is set from A to B. The time lead amount on which the above judgment is based is the lag amount corresponding to the peak value of the cross-correlation function, which also reflects the time required for passenger flow to flow between the coverage areas of the two base stations.

[0098] After determining the direction of the edge, the magnitude of the peak value of the cross-correlation function is used as the weight of the edge to quantify the passenger flow transmission intensity represented by the edge; the higher the peak value, the closer the passenger flow linkage between the two base stations, and the greater the weight of the corresponding edge; the lower the peak value, the weaker the passenger flow linkage, and the smaller the weight of the corresponding edge.

[0099] After the above processing, each base station network topology edge carries two types of attributes: flow direction and weight, which together form a directed weighted base station network topology graph.

[0100] The base station network topology is aggregated and mapped to the regional monitoring node level to generate a spatial structure feature matrix. The specific process includes:

[0101] Each base station node in the base station network topology is assigned to its corresponding regional monitoring node, and internal edges between base stations within the same regional monitoring node are ignored during the aggregation process.

[0102] The mapping relationship between base station nodes and regional monitoring nodes is a predefined static configuration that is automatically completed based on the inclusion relationship between the geographical coordinates of the base station and the geofence of the scenic area's functional zones. That is, it determines whether the coordinates of each base station fall within the polygonal fence of a certain functional zone.

[0103] Passenger flow between base stations within the same monitoring node only reflects the micro-movement behavior of tourists within that zone, such as movement between different exhibition halls within the same park. This type of micro-flow has limited significance for analyzing the overall passenger flow situation across regions, and it will generate self-loops (i.e., diagonal elements of the matrix) after matrixing, introducing unnecessary numerical interference. By ignoring internal edges in the aggregation layer, the system focuses its analysis on the cross-node flow relationships between regions, allowing the off-diagonal elements of the spatial structure feature matrix to purely reflect the passenger flow connections between regions.

[0104] For cross-node directed edges connecting monitoring nodes in different regions to their respective base stations, the weights of the edges in the set of edges are weighted and averaged using the proportion of the concurrent load value of each starting base station to the total load of the monitoring nodes in its respective region as the aggregation coefficient, so as to obtain the node edge weights of the directed connections between the monitoring nodes in the corresponding regions.

[0105] If a monitoring node in a certain area contains three base stations, with their concurrent loads accounting for 60%, 30%, and 10% of the node's total load, respectively, then when calculating the edge weights from this node to adjacent nodes, the cross-node edge connected to the base station with the highest load percentage will contribute the most to the final node edge weight. A higher load percentage for a base station at a given time indicates a larger number of tourists gathered within its coverage area, and a greater outflow of tourists to adjacent areas. Therefore, the weight of its corresponding directed edge should receive a higher weight during aggregation.

[0106] Since the concurrent load values ​​of each base station change over time, the aggregation coefficient is not a fixed constant, but a weight vector that is dynamically updated with the time step. This ensures that the calculation of the node edge weights always reflects the real passenger flow distribution at the current moment.

[0107] When there are two directed connections between monitoring node pairs in the same area, the weights of the two directions and their corresponding node edges are retained respectively.

[0108] In the actual operation of scenic areas, there are bidirectional flows of people between many adjacent functional areas. For example, tourists flow from the entrance area to the core scenic area in the morning and from the core scenic area back to the entrance area in the afternoon. The flow scale and time distribution in the two directions are different. Forcibly merging bidirectional connections into undirected connections or taking an average will result in the loss of reverse flow information, making it impossible to support the directional analysis of net inflow and net outflow. At the level of expressing the spatial structure feature matrix, retaining bidirectional connections means that the two elements at symmetrical positions in the matrix (the weight of the flow from regional monitoring node A to regional monitoring node B, and the weight of the flow from regional monitoring node B to regional monitoring node A) can take different values. That is, the matrix is ​​an asymmetric matrix, which is fundamentally different from the symmetric matrix corresponding to an undirected network.

[0109] A spatial structure feature matrix is ​​constructed using the regional monitoring nodes as row and column indices. The value of each element in the matrix is ​​the weight of the node edge connecting the corresponding regional monitoring node pairs. The value of the corresponding element for node pairs without directed connections is set to zero.

[0110] The spatial structure feature matrix is ​​a square matrix with the number of rows and columns equal to the number of regional monitoring nodes. The row indices represent the departure regional monitoring nodes, and the column indices represent the arrival regional monitoring nodes. The element values ​​at the intersection of rows and columns represent the edge weights of the nodes in the corresponding directions. For regional monitoring node pairs without directed connections, their matrix elements are set to zero instead of being missing. This process ensures the integrity and regularity of the matrix, allowing it to be directly used as the adjacency matrix input for graph neural networks.

[0111] The zeroing process also implies a sparsity assumption: in actual scenic areas, not every pair of functional zones has a direct flow of visitors. The matrix element values ​​between most pairs of monitoring nodes are zero, which makes the matrix usually appear as a sparse matrix.

[0112] Without accessing users' mobile phone privacy (such as IMEI or phone number), the algorithm deduces crowd movement paths solely from the waveform similarity of base station loads, resolving the conflict between data security and monitoring accuracy. By compressing the complex relationships of thousands of base stations into a spatial structure feature matrix, the algorithm extracts massive amounts of raw communication signals into extremely simplified regional dynamic indicators, enabling it to process large-scale city-level regional monitoring nodes in real time.

[0113] The system calculates the temporal correlation between the derived operational indicator sequence extracted from the preset business system and the physical passenger flow feature vector, detects data deviation anomalies in the aggregated concurrent load time series data, and dynamically adjusts the fusion weights based on the anomaly detection results to synthesize the comprehensive status feature vector of each regional monitoring node.

[0114] The process of detecting data deviation anomalies in aggregated concurrent load time-series data includes:

[0115] The derived operational indicator sequences corresponding to the monitoring nodes in each region are extracted from the preset business system, and the extracted derived operational indicator sequences are normalized to form standardized derived operational indicator sequences.

[0116] Pre-configured business systems refer to various tourism industry-related data platforms that are pre-configured and connected by the operation and management party during the system deployment phase. The scope of these systems can be flexibly set according to the specific business model of the scenic spot. Typical access objects include scenic spot ticketing management system, catering cashier system, parking management system, cable car and sightseeing vehicle operation system, hotel check-in registration system, etc.

[0117] The extraction process of derived operational indicator sequences involves cross-system data interface integration. The data reporting frequency of each business system may differ. For example, the ticketing system may report the number of redemptions in real time at the minute level, while the catering POS system may summarize the turnover at the hour level. Therefore, after extraction, it is necessary to unify the time granularity and resample all indicator sequences to the same time resolution as the physical passenger flow feature vector.

[0118] Normalization is a necessary step to eliminate differences in the dimensions of different indicators. It usually adopts a standard score transformation method based on historical mean and standard deviation to convert each indicator sequence into a standardized sequence with a mean of zero and a standard deviation of one. This makes indicator sequences from different sources and with different dimensions comparable on a numerical scale, laying the foundation for subsequent unified cross-correlation calculations.

[0119] Derivative operational indicator sequences refer to time-series data of statistics extracted from business systems that reflect the operational status of the tourism industry. These are non-communication-related business statistical data.

[0120] The physical passenger flow characteristic vector is directly calculated from the concurrent load signals of the base station. Its data source is the objectively existing physical communication signals, which do not rely on any manual input or business system operation. Therefore, it has high objectivity and anti-interference ability. The derived operational indicator sequence comes from the statistical summary of the business system. Its values ​​involve multiple links that may introduce errors or distortions in the entire chain from actual business events to system records, such as equipment collection, manual input, system transmission, and database storage. Therefore, it has a relatively low inherent reliability guarantee.

[0121] In this embodiment, the statistically aggregated data from the business system includes, but is not limited to, ticket verification records, catering transaction records, parking entry and exit records, cable car and sightseeing vehicle ride records, and hotel check-in registration information. Derivative operational indicators are obtained by performing a preset statistical transformation on the statistically aggregated data from the business system. Specifically, the preset statistical transformation involves calculating the ratio between the statistically aggregated data from the business system and pre-stored verification data in the database. The numerical changes are driven by the usage status of scenic area resources and exhibit synchronous or lagging response relationships with changes in passenger flow, thus serving as a proxy observation of passenger flow status for subsequent correlation analysis. In other embodiments, the effectiveness of the indicators can be screened by calculating the correlation coefficient between the derived operational indicator sequence and the physical passenger flow feature vector, retaining only indicator sequences with correlation coefficients higher than a preset threshold for subsequent analysis.

[0122] For each monitoring node in the region, the sliding window cross-correlation method is used to calculate the cross-correlation coefficient between the standardized derived operation index sequence and each component of the physical passenger flow feature vector within a preset lag range. The maximum value of the cross-correlation coefficient is taken as the time-series correlation measure between the two.

[0123] The timeline is divided into several overlapping time windows, and the cross-correlation coefficient is calculated independently within each window. This yields a dynamic sequence of correlation evolution over time, rather than a single static correlation value for the entire historical period. In practice, the window length is typically set to cover several complete passenger flow fluctuation cycles. For example, if sampling is done in one-hour increments, the window length can be set to 24 to 72 hours.

[0124] Changes in derived operational indicators may lag systematically with respect to changes in physical passenger flow. For example, after tourists arrive in a certain area, their dining consumption may only occur within one to two hours of arrival. Therefore, the time-series data of catering revenue in the same area will naturally lag behind the time-series data of base station concurrent load by one to two hours. The preset lag range refers to the maximum time offset that the system allows to be considered in cross-correlation calculations. Its upper bound should cover all reasonable response delays that may exist between various derived indicators and passenger flow signals. It is usually set according to the business logic characteristics of each indicator. For example, the lag range for ticket verification is set to 0 to 30 minutes, and the lag range for catering revenue is set to 0 to 2 hours. Taking the maximum value of the cross-correlation coefficient within the preset lag range is to eliminate the interference of natural response delays on the correlation measurement results and ensure that the correlation measurement value reflects the signal correlation strength between the two, rather than the time alignment accuracy.

[0125] Based on historical data, establish the normal correlation benchmark interval for each monitoring node in the region. Compare the time-series correlation metric calculated within the current time window with the normal correlation benchmark interval. If the current time-series correlation metric is lower than the lower bound of the normal correlation benchmark interval, it is determined that there is a data deviation anomaly in the monitoring node of that region.

[0126] The normal correlation benchmark interval is the core reference standard for judging data deviation anomalies. Its establishment process includes: selecting time-series correlation measurement value sequences of each regional monitoring node from the historical database during historical operating periods with confirmed good data quality; fitting the statistical distribution of this historical sequence; extracting its mean and standard deviation; and using the mean minus a certain multiple of the standard deviation as the lower bound of the benchmark interval. The selection of the lower bound multiple reflects the system's trade-off between false alarm rate and false negative rate: the larger the multiple, the lower the lower bound, the higher the system tolerance, and the higher the false negative rate; the smaller the multiple, the higher the lower bound, the higher the system sensitivity, and the higher the false alarm rate. In practice, this multiple is usually customized based on the historical data quality of each scenic area and the sensitivity requirements of operation and management for anomaly response. The lower bound of the normal correlation benchmark interval is specific to each regional monitoring node; that is, each regional monitoring node independently establishes its benchmark interval based on its own historical correlation distribution, rather than using a uniform threshold shared by all regional monitoring nodes.

[0127] The fusion weights are dynamically adjusted based on the anomaly detection results to synthesize a comprehensive state feature vector for each monitoring node in each region, specifically including:

[0128] The comprehensive state feature vector is a multi-dimensional regional state description formed by organically integrating the physical passenger flow feature vector with the derived operation index sequence.

[0129] The standardized derived operation index sequences of each regional monitoring node are assigned fusion weights. The fusion weight of regional monitoring nodes that have not detected data deviation anomalies is taken as a preset normal value. The fusion weight of regional monitoring nodes that have detected data deviation anomalies is reduced proportionally according to the degree to which the time series correlation metric is lower than the lower bound of the normal correlation benchmark interval. The greater the deviation, the lower the fusion weight.

[0130] The preset normal value is the baseline for the fusion weights assigned to the derived operational indicator sequence when no anomaly detection results are found. This value reflects the system's assessment of the incremental information contribution of business data relative to physical passenger flow characteristics under normal conditions. The optimal value is determined through ablation experiments on historical datasets, i.e., comparing the performance of the comprehensive state feature vector for subsequent prediction tasks under different normal value settings, and selecting the value corresponding to the optimal prediction accuracy. The scaling-down function describes the rate at which the fusion weights decrease as the temporal correlation metric further decreases from its lower bound. When the correlation metric is exactly equal to the lower bound, the fusion weights remain at the preset normal value; when the correlation metric drops to a very low level (e.g., close to zero), the fusion weights drop to close to zero; the intermediate state between these two corresponds to the intermediate weight value of linear interpolation.

[0131] Based on the physical passenger flow feature vectors of monitoring nodes in each region, the standardized derived operation index sequence is weighted according to the corresponding fusion weights and then concatenated with the physical passenger flow feature vectors to form the comprehensive status feature vectors of monitoring nodes in each region.

[0132] The concatenation method preserves the independent information of the two types of features and does not introduce linear mixing between features, enabling subsequent machine learning models or decision rules to learn or utilize the unique patterns of the two types of features respectively.

[0133] Using the spatial structure feature matrix and the comprehensive state feature vector as input, a state transmission graph model is constructed, multi-step temporal evolution prediction is performed, and prediction result data is output to trigger regional resource scheduling.

[0134] Using the monitoring nodes of each region as nodes in the graph model, a node set is constructed so that each monitoring node in a region corresponds one-to-one with a unique node in the graph model.

[0135] Once constructed, the node set is fixed as a static structure, its size equal to the total number of regional monitoring nodes within the target tourism area, and it does not increase or decrease over time during system operation. At this stage, the node set does not carry any state information; it is merely a set of abstract topological placeholders. Its actual content will be filled in subsequent steps by loading a comprehensive state feature vector.

[0136] The directed connection relationship between nodes is defined by the spatial structure feature matrix: the starting node of the directed connection is used as the row index and the ending node is used as the column index to locate the corresponding element in the spatial structure feature matrix.

[0137] The convention of the starting node corresponding to the row index and the ending node corresponding to the column index establishes a clear one-to-one mapping relationship between matrix elements and directed edges. That is, the element in the matrix located in a certain row and a certain column corresponds to the directed connection from the monitoring node of the region represented by the certain row to the monitoring node of the region represented by the certain column.

[0138] When the value of this element is not zero, it is determined that there is a directed connection between the two nodes, and this element value is used as the transmission weight of the corresponding directed connection to characterize the state transmission strength between regional monitoring nodes.

[0139] Transmission weight is the strength of the influence of the current state of a monitoring node on the future state of its downstream neighboring nodes. The magnitude of the transmission weight originates from the value of the corresponding element in the spatial structure feature matrix, which in turn traces back to the amplitude of the peak value of the cross-correlation function between the time-series data of concurrent load at base stations. Therefore, the transmission weight fundamentally reflects the tightness of the population flow connection between two regions in historical passenger flow data. The transmission weight ranges between a significance threshold and one. The closer the value is to one, the stronger and more regular the influence of the upstream node's state fluctuations on the downstream node. For example, there is a high transmission weight between the main entrance area and the core scenic spot area of ​​a scenic area because most tourists must pass through the entrance area before reaching the core scenic spot. The closer the value is to the significance threshold, the weaker the state transmission relationship between the two areas, and perhaps only a few tourists will move directly between them. An element value of zero means that there is no effective passenger flow transmission relationship in the corresponding direction. In the graph model, this is represented by no directed edge connection between the two nodes. These nodes are independent in the state transmission calculation and do not produce cross-node influence.

[0140] The comprehensive state feature vectors of monitoring nodes in each region are mapped to the corresponding nodes, so that each node carries the comprehensive state feature vector of the monitoring nodes in the corresponding region.

[0141] Loading the comprehensive state feature vector transforms each node from an abstract placeholder without content into a meaningful entity carrying rich regional state information.

[0142] The state transmission graph model is composed of a set of nodes, directed connections, transmission weights, and a comprehensive state feature vector. The node set provides the basic carrier of the spatial topology; directed connections define the paths and directions of state transmission between regions; transmission weights quantify the strength of influence on each transmission path; and the comprehensive state feature vector provides a multi-dimensional state description of each node at the current moment. These four elements have a strict hierarchical dependency: the node set is the foundation, directed connections exist with the node set as an index, transmission weights are defined based on directed connections, and the comprehensive state feature vector is assigned values ​​based on the node set. Mathematically, the state transmission graph model belongs to the category of directed weighted attribute graphs, which is fully compatible with the standard graph structure definition in the field of graph neural networks. It can directly apply mature graph learning operators such as graph convolution and graph attention for state transmission calculation, or it can be deduced using rule-based iterative propagation methods.

[0143] After the state transmission graph model is constructed, based on the directed connection relationship between nodes and the corresponding transmission weight defined by the spatial structure feature matrix, cross-node state transmission calculation is performed on the comprehensive state feature vector of each monitoring node. The comprehensive state feature vector of each monitoring node and the comprehensive state feature vector of its neighboring nodes are weighted and fused according to the corresponding transmission weight to obtain the comprehensive state feature vector of each monitoring node at the next time step.

[0144] For each regional monitoring node, its comprehensive state feature vector at the next moment is formed by fusing two parts of information: the first part is the comprehensive state feature vector of the node itself at the current moment, which represents the state continuity within the region; the second part is the comprehensive state feature vector of all adjacent upstream nodes that terminate at this node at the current moment. The contribution of each upstream node is weighted according to the transmission weight of the corresponding directed connection, which represents the impact of passenger flow state input from the upstream region on the future state of this node.

[0145] By using the comprehensive state feature vector of the next moment as the current input, the cross-node state propagation calculation is repeatedly performed to gradually obtain the comprehensive state feature vectors of multiple future moments, thereby realizing multi-step temporal evolution prediction.

[0146] Multi-step time series evolution prediction adopts an autoregressive iterative deduction strategy, which means that the output of each step of the propagation calculation is directly used as the input of the next step of the calculation. By repeatedly executing the same propagation calculation rules, it gradually extends into the future on the time axis.

[0147] Time series analysis is performed on the comprehensive state feature vectors of multiple future moments to extract the state change magnitude and trend of monitoring nodes in each region within the preset prediction time range.

[0148] For each regional monitoring node, its multi-step prediction results within a preset prediction time range constitute a time series of a multi-dimensional feature vector. The magnitude of state change reflects the width of the extreme value range of the comprehensive state feature vector of the regional monitoring node within the prediction time range, that is, the maximum fluctuation range of the state within the prediction interval. The larger the magnitude, the more drastic the expected load changes in the region in the future period, and the higher the urgency of resource scheduling. The trend of state change reflects the monotonic trend of the comprehensive state feature vector as the number of prediction steps increases, specifically manifested as continuous rise, continuous fall, rise followed by fall, or fall followed by rise, etc. Identifying the trend is crucial for judging the direction of regional load evolution. An upward trend indicates that the management should prepare sufficient resources in advance to cope with the upcoming peak, while a downward trend indicates that resources can be released in an orderly manner to avoid redundant occupation.

[0149] The preset forecast time range is the time boundary parameter for time series analysis. Its setting must match the advance requirements of scenic area resource scheduling. For example, if the deployment of scenic area security personnel needs to be completed two hours in advance, the preset forecast time range should cover at least the next two hours.

[0150] Based on the magnitude and trend of status changes of monitoring nodes in each region, predictive data characterizing the degree of change in regional operating load are constructed.

[0151] The forecast results data is a standardized data object formed by structuring and encapsulating the output of time series analysis. Its structural design needs to meet the data format requirements of the subsequent resource scheduling triggering module. The forecast results data uses regional monitoring nodes as the basic organizational unit. Each regional monitoring node corresponds to a set of structured fields, including node identifier, forecast timestamp, state change magnitude value, state change trend category label, and necessary confidence assessment indicators.

[0152] The confidence level assessment index reflects the impact of the cumulative error on the reliability of the prediction conclusion under the current prediction steps. It is usually estimated based on the prediction error distribution under similar historical conditions. The more prediction steps, the lower the confidence level. The system can use this to transmit uncertainty information to the downstream scheduling module, so that the scheduling decision can reasonably consider the reliability of the prediction conclusion.

[0153] There is a synergistic interpretation between the magnitude and trend of state changes: high magnitude combined with an upward trend corresponds to the highest level of resource scheduling warning, indicating that a significant load surge is imminent in the region; low magnitude combined with a downward trend corresponds to the lowest level of scheduling priority, indicating that the load in the region is stable and trending towards easing. The construction of the prediction results data integrates this multi-dimensional information into a unified structured output, avoiding the analytical complexity caused by downstream modules directly processing the original vector sequence.

[0154] Output the prediction results data to trigger regional resource scheduling.

[0155] The output of the prediction results serves as the interface between the entire state transition diagram model prediction process and the external resource scheduling system. The mechanism for triggering regional resource scheduling typically employs a combination of threshold triggering and tiered response design.

[0156] The system pre-sets corresponding scheduling response schemes for different levels of state change magnitude and trend combinations. When the values ​​of each field of the prediction result data reach or exceed the trigger threshold of the corresponding level, it automatically sends a structured instruction containing the location of the regional monitoring node, the warning level, and the suggested response measures to the resource scheduling platform.

[0157] In this embodiment of the invention, the types of resource scheduling that can be triggered include at least the regional deployment adjustment of security personnel, the adjustment of the overtime frequency of tour vehicles, the preparation for temporary expansion of food stalls, the increase in the cleaning frequency of sanitation facilities, and the push of diversion instructions from the broadcast guidance system.

[0158] Please see Figure 2 As shown in the figure, this embodiment of the invention also provides a regional tourism industry operation monitoring and decision support system, including:

[0159] The aggregation module is used to delineate regional monitoring nodes based on preset geofence boundaries, acquire concurrent load time series data of each base station within the target area, and establish a mapping relationship between base stations and regional monitoring nodes based on the spatial affiliation relationship between each base station and the geofence. The concurrent load time series data of each base station is aggregated to the corresponding regional monitoring node to obtain the aggregated concurrent load time series data of each regional monitoring node.

[0160] The combination module is used to perform periodic decomposition processing on the aggregated concurrent load time series data of monitoring nodes in each region, extract load distribution characteristics, and combine them to form the physical passenger flow feature vector of each monitoring node in each region.

[0161] The matrix construction module is used to construct a base station network topology map that reflects the direction and path of passenger flow based on the spatial correlation between the concurrent load time series data of each base station, and to aggregate and map the base station network topology map to the regional monitoring node level to generate a spatial structure feature matrix.

[0162] The anomaly detection module is used to calculate the temporal correlation between the derived operation index sequence extracted from the preset business system and the physical passenger flow feature vector, detect data deviation anomalies in the aggregated concurrent load time series data, and dynamically adjust the fusion weights based on the anomaly detection results to synthesize the comprehensive status feature vector of each regional monitoring node. The derived operation index sequence refers to the time series data of statistics extracted from the business system that can reflect the operation status of the tourism industry.

[0163] The output module is used to construct a state transmission graph model with the spatial structure feature matrix and the comprehensive state feature vector as input, perform multi-step temporal evolution prediction, and output prediction result data for triggering regional resource scheduling.

[0164] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the above functions can be divided into different functional modules to complete all or part of the functions described above.

[0165] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units through some interfaces, and may be electrical, mechanical, or other forms.

[0166] 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 within the technical scope disclosed in the present invention should be covered 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 method for monitoring and decision support of regional tourism industry operation, characterized in that: include: Based on the preset geofence boundary, regional monitoring nodes are defined, concurrent load time series data of each base station in the target area are obtained, and a mapping relationship between the base station and the regional monitoring node is established according to the spatial affiliation relationship between each base station and the geofence. The concurrent load time series data of each base station is aggregated to the corresponding regional monitoring node to obtain the aggregated concurrent load time series data of each regional monitoring node. The time-series data of the aggregated concurrent load from each monitoring node in each region are periodically decomposed to extract load distribution characteristics and combine them to form the physical passenger flow feature vector of each monitoring node in each region. Based on the spatial correlation between the concurrent load time series data of each base station, a base station network topology map reflecting the direction and path of passenger flow is constructed, and the base station network topology map is aggregated and mapped to the regional monitoring node level to generate a spatial structure feature matrix. The time-series correlation between the derived operation index sequence extracted from the preset business system and the physical passenger flow feature vector is calculated. Data deviation anomalies of the aggregated concurrent load time-series data are detected. Based on the anomaly detection results, the fusion weight is dynamically adjusted to synthesize the comprehensive status feature vector of each regional monitoring node. The derived operation index sequence refers to the time-series data of statistics extracted from the business system that can reflect the operation status of the tourism industry. Using the spatial structure feature matrix and the comprehensive state feature vector as input, a state transmission graph model is constructed, multi-step temporal evolution prediction is performed, and prediction result data is output to trigger regional resource scheduling.

2. The method for monitoring and decision support of regional tourism industry operation as claimed in claim 1, wherein: The obtained concurrent load time-series data from the monitoring nodes in each region specifically includes: Concurrent load time-series data includes direct load metrics and indirect load metrics. Direct load metrics include at least the number of simultaneously online device connections, physical resource block utilization, and uplink and downlink throughput. Indirect load metrics include at least the number of handovers, location update frequency, and channel quality metrics. The geographic coordinates of all base stations within the target area are obtained in advance, and several regional monitoring nodes are pre-defined. Each regional monitoring node corresponds to a geofence polygon defined by the boundary coordinates. A ray is emitted from the base station coordinates in a fixed direction. The number of intersections between the ray and the fence boundary is counted. If the number of intersections is odd, the base station is determined to be inside the fence and assigned to the corresponding area monitoring node. If the number of intersections is even, the base station is determined to be outside the fence. For a base station whose coverage spans multiple fence boundaries, calculate the spatial overlap area between the base station's coverage area and each fence polygon, and use the proportion of each overlap area to the total overlap area as the allocation weight. Then, the concurrent load time series data of the base station is allocated to the corresponding area monitoring node according to the allocation weight. For a base station whose coverage area falls entirely within a certain fence, its concurrent load data is completely attributed to the monitoring node in that area; After determining the attribution, the concurrent load time series data of all attribution base stations within the same monitoring node are weighted and summed according to their respective weights to obtain the aggregated concurrent load time series data of that node.

3. The method for monitoring and decision support of regional tourism industry operation as claimed in claim 1, wherein: The formation of physical passenger flow feature vectors for each regional monitoring node specifically includes: For the concurrent load time series data of each monitoring node, the daily cycle component, weekly cycle component and annual cycle component are identified and extracted in sequence. After subtracting the above cycle components from the original concurrent load time series data in sequence, the remaining residual signal is the net abnormal passenger flow component. Based on the completion of the periodic decomposition, load distribution characteristics are extracted for each regional monitoring node. The load distribution characteristics include at least the periodic dimension characteristics that characterize the periodic operation pattern, the distribution dimension characteristics that characterize the statistical distribution characteristics of the data, and the anomaly dimension characteristics that characterize irregular fluctuations. The load distribution characteristics are spliced ​​together in a preset order to obtain the physical passenger flow characteristic vector of each monitoring node in the current time window.

4. The method for monitoring and decision support of regional tourism industry operation as described in claim 1, characterized in that: The construction of the base station network topology map reflecting the direction and path of passenger flow specifically includes: Each base station within the target area is considered a node in the base station network topology. Obtain the geographical location information of each base station in the target area, calculate the Euclidean distance between the base stations, determine the base station pairs with a geographical distance less than a preset distance threshold as initial adjacent base stations, and establish an initial edge connecting the two to form a candidate edge set. For each pair of adjacent base stations in the candidate edge set, calculate the cross-correlation function curve between their concurrent load time series data under different hysteresis. If the cross-correlation function curve has a significant peak at a certain lag, it is determined that there is a statistically significant passenger flow correlation between the two base stations, and the candidate edge is retained in the base station network topology graph. If the cross-correlation function has no significant peak value under all lags, it is determined that there is no obvious passenger flow correlation between the two base stations, and the candidate edge is deleted. The criteria for determining the significant peak value are: A peak value exceeding a preset significance threshold is considered a significant peak value. For each retained base station network topology edge, analyze the activation order of the peak concurrent load time series data of the two base stations to obtain the direction of passenger flow. After determining the direction of the edge, the magnitude of the peak value of the cross-correlation function is used as the weight of the edge to quantify the passenger flow transmission intensity represented by the edge. After the above processing, each base station network topology edge carries two types of attributes: flow direction and weight, which together form a directed weighted base station network topology graph.

5. The regional tourism industry operation monitoring and decision support method as described in claim 1, characterized in that: The specific process of generating the spatial structure feature matrix includes: Each base station node in the base station network topology is assigned to its corresponding regional monitoring node, and the internal edges between base stations within the same regional monitoring node are ignored during the aggregation process. For cross-node directed edges connecting monitoring nodes in different regions to their respective base stations, the weights of the edges in the set of edges are weighted and averaged using the proportion of the concurrent load value of each starting base station to the total load of the monitoring nodes in its respective region as the aggregation coefficient, so as to obtain the node edge weights of the directed connections between the monitoring nodes in the corresponding regions. When there are two directed connections between monitoring node pairs in the same area, retain the weights of the two directions and their corresponding node edges respectively. A spatial structure feature matrix is ​​constructed using the regional monitoring nodes as row and column indices. The value of each element in the matrix is ​​the weight of the node edge connecting the corresponding regional monitoring node pairs. The value of the corresponding element for node pairs without directed connections is set to zero.

6. The regional tourism industry operation monitoring and decision support method as described in claim 1, characterized in that: The specific process for detecting data deviation anomalies in the aggregated concurrent load time-series data includes: Extract the derivative operation indicator sequence corresponding to each regional monitoring node from the preset business system, and normalize the extracted derivative operation indicator sequence to form a standardized derivative operation indicator sequence. For each monitoring node in the region, within the preset lag range, the cross-correlation coefficient between the standardized derived operation index sequence and each component of the physical passenger flow feature vector is calculated, and the maximum value of the cross-correlation coefficient is taken as the time-series correlation measure between the two. Based on historical data, establish the normal correlation benchmark interval for each monitoring node. Compare the time-series correlation metric calculated within the current time window with the normal correlation benchmark interval. If the current time-series correlation metric is lower than the lower bound of the normal correlation benchmark interval, it is determined that the node has a data deviation anomaly.

7. The method for monitoring and decision support of regional tourism industry operation as described in claim 1, characterized in that: The synthesized comprehensive state feature vector of each monitoring node in each region specifically includes: The standardized derived operation index sequences of each regional monitoring node are assigned fusion weights. The fusion weight of regional monitoring nodes that have not detected data deviation anomalies is taken as a preset normal value. The fusion weight of regional monitoring nodes that have detected data deviation anomalies is reduced proportionally according to the degree to which the time series correlation metric is lower than the lower bound of the normal correlation benchmark interval. Based on the physical passenger flow feature vectors of monitoring nodes in each region, the standardized derived operation index sequence is weighted according to the corresponding fusion weights and then concatenated with the physical passenger flow feature vectors to form the comprehensive status feature vectors of monitoring nodes in each region.

8. The method for monitoring and decision support of regional tourism industry operation as described in claim 1, characterized in that: The specific process of constructing the state transmission graph model includes: Using the monitoring nodes of each region as nodes in the graph model, a node set is constructed so that each monitoring node in a region corresponds one-to-one with a unique node in the graph model. The directed connection relationship between nodes is defined by the spatial structure feature matrix: the starting node of the directed connection is used as the row index and the ending node is used as the column index to locate the corresponding element in the spatial structure feature matrix; When the value of this element is not zero, it is determined that there is a directed connection between the two nodes, and this element value is used as the transmission weight of the corresponding directed connection to characterize the state transmission strength between regional monitoring nodes. The comprehensive state feature vectors of each monitoring node in each region are mapped to the corresponding nodes, so that each node carries the comprehensive state feature vector of the monitoring node in the corresponding region. The state transmission graph model is composed of a set of nodes, directed connections, transmission weights, and a comprehensive state feature vector.

9. The method for monitoring and decision support of regional tourism industry operation as described in claim 1, characterized in that: The specific process of performing multi-step temporal evolution prediction and outputting prediction result data for triggering regional resource scheduling includes: After the state transmission graph model is constructed, based on the directed connection relationship between nodes and the corresponding transmission weight defined by the spatial structure feature matrix, cross-node state transmission calculation is performed on the comprehensive state feature vector of each monitoring node. The comprehensive state feature vector of each monitoring node and the comprehensive state feature vector of its neighboring nodes are weighted and fused according to the corresponding transmission weight to obtain the comprehensive state feature vector of each monitoring node at the next time step. Using the comprehensive state feature vector of the next moment as the current input, the cross-node state propagation calculation is repeatedly executed to gradually obtain the comprehensive state feature vectors of multiple future moments, thereby realizing multi-step temporal evolution prediction. Time series analysis is performed on the comprehensive state feature vectors of multiple future moments to extract the state change magnitude and trend of monitoring nodes in each region within the preset prediction time range; Based on the magnitude and trend of status changes of monitoring nodes in each region, predictive data characterizing the degree of change in regional operating load are constructed. The predicted data is output to trigger regional resource scheduling.

10. A regional tourism industry operation monitoring and decision support system, characterized by: The aggregation module is used to delineate regional monitoring nodes based on preset geofence boundaries, obtain concurrent load time series data of each base station in the target area, and establish a mapping relationship between base stations and regional monitoring nodes based on the spatial affiliation relationship between each base station and the geofence, and aggregate the concurrent load time series data of each base station to the corresponding regional monitoring node to obtain the aggregated concurrent load time series data of each regional monitoring node. The combination module is used to perform periodic decomposition processing on the aggregated concurrent load time series data of monitoring nodes in various regions, extract load distribution characteristics, and combine them to form the physical passenger flow feature vector of each monitoring node in various regions. The matrix construction module is used to construct a base station network topology map that reflects the direction and path of passenger flow based on the spatial correlation between the concurrent load time series data of each base station, and to aggregate and map the base station network topology map to the regional monitoring node level to generate a spatial structure feature matrix. The anomaly detection module is used to calculate the temporal correlation between the derived operation index sequence extracted from the preset business system and the physical passenger flow feature vector, detect data deviation anomalies in the aggregated concurrent load time series data, and dynamically adjust the fusion weight based on the anomaly detection results to synthesize the comprehensive status feature vector of each regional monitoring node. The derived operation index sequence refers to the time series data of statistics extracted from the business system that can reflect the operation status of the tourism industry. The output module is used to construct a state transmission graph model with the spatial structure feature matrix and the comprehensive state feature vector as input, perform multi-step temporal evolution prediction, and output prediction result data for triggering regional resource scheduling.

Images