Shared bicycle multi-dimensional risk data early warning management method and system

Abstract

The application provides a shared bicycle multi-dimensional risk data early warning management method and system, relates to the technical field of data processing, and the method comprises the following steps: step 1, acquiring riding data, vehicle positioning data and environment data, and performing pretreatment to generate fusion data with geographic coordinates; step 2, based on the fusion data, a geographic boundary box covering all vehicle positions is constructed as a virtual reference plane, the virtual reference plane is divided into four sub-zones according to the distribution density characteristics of the vehicles on the virtual reference plane, the number change rate of the vehicles in each sub-zone and the space-time stability index are calculated, and a sub-zone adjustment coefficient is obtained. The application realizes early warning and dynamic scheduling of the shared bicycle vehicle accumulation and shortage risk, and improves the operation management efficiency and the user bicycle experience.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for managing multi-dimensional risk data for shared bicycles. Background Technology

[0002] In the daily operation of shared bicycles, accurate early warning of vehicle congestion or shortage at electronic fence stations is a key link in ensuring user experience and operational efficiency. Currently, some existing early warning management methods may mainly rely on time series analysis of the historical vehicle quantity data of a single electronic fence station. This type of local data analysis method centered on the station may, to some extent, ignore the spatial correlation of vehicle distribution between the station and its surrounding geographical area. The flow of vehicles in the city is a continuous spatial process. The change in the status of a station is mostly closely related to the dynamics of vehicle aggregation and diffusion in a larger surrounding area. If these broader spatial distribution characteristics are not effectively integrated, the judgment basis of the early warning model may not be comprehensive enough.

[0003] For example, during the morning rush hour, the vehicle inflow at an electronic fence station at a subway station exit depends not only on the number of vehicles before the station, but also on the rate and pattern of vehicles converging towards the station from communities and streets within a radius of several hundred meters. Some existing methods may struggle to quantify this spatial convergence effect, leading to delays in early warning of vehicle congestion risks or requiring improvements in warning accuracy. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method and system for managing multi-dimensional risk data for shared bicycles, so as to achieve the rational allocation and efficient circulation of resources.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a multi-dimensional risk data early warning management method for shared bicycles, the method comprising:

[0007] Step 1: Acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates;

[0008] Step 2: Based on the fused data, construct a geographic bounding box covering all vehicle locations as a virtual reference plane. Divide the virtual reference plane into four zones according to the distribution density characteristics of vehicles on the virtual reference plane. Calculate the vehicle number change rate and spatiotemporal stability index for each zone to obtain the zone adjustment coefficient.

[0009] Step 3: Apply a partition adjustment coefficient to the fused data for weighted correction to obtain the corrected fused data; calculate the vehicle inflow rate, outflow rate and instantaneous holding capacity of the preset electronic fence station in a continuous time slice based on the corrected fused data, and perform feature aggregation to obtain the spatiotemporal feature tensor.

[0010] Step 4: Input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network to analyze the spatial geometric features of vehicle distribution and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period.

[0011] Step 5: Based on the probability of vehicle congestion risk or the probability of vehicle shortage risk and the corresponding risk level, trigger the corresponding risk level vehicle shortage warning or congestion warning, and generate a dynamic vehicle dispatching strategy based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

[0012] Secondly, the shared bicycle multi-dimensional risk data early warning management system includes:

[0013] The fusion module is used to acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates.

[0014] The partitioning module is used to construct a geographic bounding box covering all vehicle locations as a virtual reference plane based on fused data. According to the distribution density characteristics of vehicles on the virtual reference plane, the virtual reference plane is divided into four partitions. The vehicle number change rate and spatiotemporal stability index of each partition are calculated to obtain the partition adjustment coefficient.

[0015] The aggregation module is used to perform weighted correction processing on the fused data using partition adjustment coefficients to obtain corrected fused data; based on the corrected fused data, the vehicle inflow rate, outflow rate and instantaneous holding capacity of the preset electronic fence stations in a continuous time slice are calculated, and feature aggregation is performed to obtain the spatiotemporal feature tensor.

[0016] The analysis module is used to input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network, analyze the spatial geometric features of vehicle distribution, and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period.

[0017] The scheduling module is used to trigger vehicle shortage warnings or congestion warnings based on the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level, and to generate dynamic vehicle scheduling strategies based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

[0018] Thirdly, a computing device includes:

[0019] One or more processors;

[0020] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0021] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0022] The above-described solution of the present invention has at least the following beneficial effects:

[0023] By integrating multi-source data from cycling, location, and environment, and combining it with the spatial characteristics of vehicle distribution across the entire region, risk assessment no longer relies on isolated historical data from individual stations. It covers the continuous spatial process of vehicle flow, making the judgment basis more complete. Through geographical zoning and weighted correction of zoning adjustment coefficients, the relevance of data in key areas is enhanced, and irrelevant interference is reduced. By constructing spatiotemporal feature tensors and analyzing spatial geometric features, the distribution patterns of vehicle aggregation or dispersion can be identified, effectively avoiding problems such as spatial aggregation effects that are difficult to quantify, reducing early warning bias, and making the determination of risk type and level more accurate. Through deep mining of multi-dimensional spatiotemporal features using spatiotemporal recurrent neural networks, the development trend of vehicle congestion or shortage can be predicted in advance, avoiding potential delays in early warning and allowing operators sufficient response time to help deploy relevant measures in advance. Dynamic scheduling strategies generated based on risk level, geographical location, and real-time traffic status achieve reasonable allocation and efficient flow of resources, alleviating parking pressure caused by vehicle congestion and solving user difficulties caused by vehicle shortages. At the same time, scheduling paths and priorities are optimized to improve operational efficiency. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the multi-dimensional risk data early warning management method for shared bicycles provided in an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of a shared bicycle multi-dimensional risk data early warning management system provided in an embodiment of the present invention. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] like Figure 1As shown, embodiments of the present invention propose a multi-dimensional risk data early warning management method for shared bicycles, the method comprising the following steps:

[0028] Step 1: Acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates;

[0029] Step 2: Based on the fused data, construct a geographic bounding box covering all vehicle locations as a virtual reference plane. Divide the virtual reference plane into four zones according to the distribution density characteristics of vehicles on the virtual reference plane. Calculate the vehicle number change rate and spatiotemporal stability index for each zone to obtain the zone adjustment coefficient.

[0030] Step 3: Apply a partition adjustment coefficient to the fused data for weighted correction to obtain the corrected fused data; calculate the vehicle inflow rate, outflow rate and instantaneous holding capacity of the preset electronic fence station in a continuous time slice based on the corrected fused data, and perform feature aggregation to obtain the spatiotemporal feature tensor.

[0031] Step 4: Input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network to analyze the spatial geometric features of vehicle distribution and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period.

[0032] Step 5: Based on the probability of vehicle congestion risk or the probability of vehicle shortage risk and the corresponding risk level, trigger the corresponding risk level vehicle shortage warning or congestion warning, and generate a dynamic vehicle dispatching strategy based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

[0033] In this embodiment of the invention, by integrating multi-source data from cycling, positioning, and environment, and combining it with the spatial characteristics of vehicle distribution across the entire region, risk assessment no longer relies on isolated historical data from individual stations. It covers the continuous spatial process of vehicle flow, making the judgment basis more complete. Through geographical partitioning and weighted correction of partition adjustment coefficients, the relevance of key area data is enhanced, reducing irrelevant interference. By constructing spatiotemporal feature tensors and analyzing spatial geometric features, it identifies vehicle aggregation or dispersion patterns, effectively avoiding problems such as difficult-to-quantify spatial aggregation effects, reducing early warning bias, and making the determination of risk type and level more accurate. Through deep mining of multi-dimensional spatiotemporal features using spatiotemporal recurrent neural networks, it can predict the development trend of vehicle congestion or shortage in advance, avoiding potential delays in early warning, allowing operators sufficient response time, and facilitating the early deployment of relevant measures. Based on risk level, geographical location, and real-time traffic status, a dynamic scheduling strategy is generated to achieve reasonable allocation and efficient flow of resources, alleviating parking pressure caused by vehicle congestion and solving user difficulties caused by vehicle shortages. Simultaneously, it optimizes scheduling paths and priorities, improving operational efficiency.

[0034] In a preferred embodiment of the present invention, step 1 involves acquiring cycling data, vehicle positioning data, and environmental data, and preprocessing them to generate fused data with geographic coordinates; step 2 includes:

[0035] Step 200: Based on the maximum and minimum latitude and longitude coordinates of the vehicle locations in the fused data, determine a geographic bounding box covering all vehicle locations, and use this bounding box as a virtual reference plane. Specifically, this involves: real-time collection of riding data, vehicle location data, and environmental data through interfaces with shared bicycle onboard IoT terminals, the operation backend server, and open data interfaces from transportation departments, meteorological service platforms, and public transportation operators. Riding data includes the vehicle's unique identifier, start and end times, and start and end location information; vehicle location data includes real-time latitude and longitude and location timestamps; and environmental data includes real-time regional traffic flow, weather conditions, and passenger density at public transportation stations. The system first uses a built-in data cleaning function to automatically filter and remove records with missing core fields, duplicate records, and abnormal data that are outside the city's geographical range or outside the operating hours of 6:00 to 24:00. Then, through cloud geocoding services, the location data from different sources are uniformly converted into a unified geodetic coordinate system to eliminate the impact of coordinate system differences. Finally, an associated index is established based on the vehicle's unique identifier and timestamp, and the cycling data, location data, and environmental data of the corresponding spatiotemporal dimensions are intelligently matched and integrated to generate standardized fused data for each record, which includes geographic coordinates, vehicle usage status, time information, and environmental information. This data is then synchronously stored in a cloud-based distributed database.

[0036] The system extracts the location coordinates of all vehicles from the fused data in batches from the distributed database. The maximum and minimum values ​​of latitude and longitude are then filtered out using the cloud spatial computing engine. The points corresponding to the minimum longitude and minimum latitude are used as the lower left corner vertices, and the points corresponding to the maximum longitude and maximum latitude are used as the upper right corner vertices. A rectangular geographic bounding box is automatically constructed, which can completely cover the distribution range of all vehicles. This geographic bounding box is directly used as a virtual reference plane for subsequent spatial analysis and is synchronously stored in the cloud spatial database.

[0037] Step 201 involves projecting the vehicle position coordinates from the fused data onto a virtual reference plane to obtain vehicle distribution data on the virtual reference plane. Specifically, this includes: using a plane mapping tool to project the latitude and longitude coordinates of each vehicle from the fused data onto the virtual reference plane, with the lower left corner of the virtual reference plane as the origin, the longitude extension direction as the x-axis, and the latitude extension direction as the y-axis. The difference between the latitude and longitude of each vehicle and the origin is calculated, and the longitude difference is directly mapped to the x-axis coordinate value, and the latitude difference is directly mapped to the y-axis coordinate value. Finally, two-dimensional plane coordinates corresponding to each vehicle are generated in the cloud. The set of plane coordinates of all vehicles constitutes the vehicle distribution data on the virtual reference plane, which is synchronized to the analysis engine in real time.

[0038] Step 202: Based on the vehicle distribution data on the virtual reference plane, calculate the vehicle density value at each location on the virtual reference plane. Specifically, this includes: according to the urban area range and early warning accuracy requirements, preset the side length of the square grid unit to 100 meters in the configuration interface, divide the virtual reference plane into several equally sized grid units through the grid partitioning function, count the number of vehicles in each grid unit in the cloud distributed computing nodes, and then divide the number of vehicles in the grid unit by the area of ​​the grid unit (10,000 square meters) to obtain the vehicle density value of each grid unit. The calculation logic is: vehicle density value = number of vehicles in the grid unit ÷ 10,000. The density values ​​of all grid units are summarized in real time to form a vehicle density distribution heat map dataset of the virtual reference plane.

[0039] Step 203: Based on the spatial distribution characteristics of vehicle density values, a density-based spatial partitioning algorithm is used to divide the virtual reference plane into a first region, a second region, a third region, and a fourth region. The first region is the region with the highest vehicle density value and is located at the center of the virtual reference plane. The second region surrounds the first region, with a vehicle density value lower than that of the first region. The third region surrounds the second region, with a vehicle density value lower than that of the second region. The fourth region is located at the outermost edge of the virtual reference plane, with a vehicle density value lower than that of the third region. Specifically, this includes: sorting the density values ​​of all grids from high to low, and using statistical analysis tools combined with the density distribution patterns in historical operational data to determine four fixed density boundary thresholds, where the first boundary threshold is 8 vehicles / square meter, the second boundary threshold is 5 vehicles / square meter, the third boundary threshold is 5 vehicles / square meter, the fourth boundary threshold is 5 vehicles / square meter, the fifth boundary threshold is 5 vehicles / square meter, the sixth boundary threshold is 5 vehicles / square meter, the seventh boundary threshold is 5 vehicles / square meter, the eighth boundary threshold is 5 vehicles / square meter, the ninth boundary threshold is 5 vehicles / square meter, the eleventh ... The density thresholds are set at 2 bikes / square meter for the first region and 0.5 bikes / square meter for the second region, corresponding to the density cutoff values ​​for the four regions (these values ​​are based on statistics of the normal density of shared bikes in different areas of the city). Using the geometric center of the virtual reference plane as the reference point, the grid cells with a density value higher than 8 bikes / square meter are designated as the first region, ensuring that they are located at the center of the plane. The grid cells surrounding the first region with a density value between 5 and 8 bikes / square meter are designated as the second region. The grid cells surrounding the second region with a density value between 2 and 5 bikes / square meter are designated as the third region. Finally, the grid cells located on the outermost edge of the virtual reference plane with a density value lower than 2 bikes / square meter but not lower than 0.5 bikes / square meter are designated as the fourth region, ultimately forming four concentric ring-shaped continuous zones with successively decreasing densities.

[0040] Step 204: Calculate the vehicle quantity change rate index and the spatiotemporal stability index for each partition within a preset time window. The spatiotemporal stability index reflects the spatiotemporal fluctuation characteristics of vehicle distribution. Specifically, it includes: setting a fixed time window of 15 minutes, using a time slicing tool to calculate the average number of vehicles in each partition within the current 15-minute time window, and then calculating the average number of vehicles in the adjacent 15-minute time window. The cloud computing node calculates the vehicle quantity change rate according to the following logic: Vehicle quantity change rate = (Average number of vehicles in the current time window - Average number of vehicles in the previous time window) ÷ Average number of vehicles in the previous time window. The rate of change in the number of vehicles in each zone is obtained. This indicator reflects the dynamic increase or decrease trend of the number of vehicles in the zone in real time. Within a preset 15-minute time window, three time slices are automatically extracted at 5-minute intervals. The average vehicle density value in the zone under each time slice is calculated to form a set of density data sequences. The standard deviation of this density data sequence is calculated by cloud statistical analysis tools. Then, the average vehicle density value of the zone within the time window is calculated. The spatiotemporal stability index is calculated according to the following logic: Spatiotemporal stability index = 1 - (standard deviation ÷ average vehicle density value). The larger the value of this index, the smaller the spatiotemporal fluctuation of the vehicle distribution in the zone and the stronger the stability.

[0041] Step 205: Based on the vehicle quantity change rate index and spatiotemporal stability index of the four partitions, the partition adjustment coefficient for each partition is obtained through weighted fusion calculation. Specifically, based on the verification results of historical operating data, the weight of the vehicle quantity change rate index is preset to 0.6, and the weight of the spatiotemporal stability index is preset to 0.4, with the sum of the two weights being 1, in order to balance the impact of dynamic change trend and distribution stability on subsequent analysis; for each partition, the cloud computing node multiplies its vehicle quantity change rate by 0.6 and its spatiotemporal stability index by 0.4, and then adds the two products together. The calculation logic is: partition adjustment coefficient = vehicle quantity change rate × 0.6 + spatiotemporal stability index × 0.4. Finally, the partition adjustment coefficients corresponding to the four partitions are obtained and stored in real time in the cloud analysis result database.

[0042] This embodiment integrates key information on cycling, positioning, and environment through multi-source data fusion and preprocessing. It also removes abnormal data and unifies the coordinate system to ensure the integrity and consistency of the fused data, avoiding distortion of analysis results due to data deviation. Based on vehicle density characteristics, it divides the data into four concentric ring-shaped zones, which not only conforms to the actual distribution pattern of vehicle density decreasing from the center to the periphery in the city, but also breaks the local perspective of a single site and fully considers the spatial correlation of vehicle distribution. By calculating the rate of change of vehicle quantity and spatiotemporal stability index, it quantifies the dynamic change trend and distribution fluctuation characteristics of each zone. Then, through weighted fusion, it obtains the zone adjustment coefficient, which can reflect the differentiated characteristics of each zone.

[0043] In a preferred embodiment of the present invention, step 3 includes:

[0044] Step 300 involves applying weighted corrections to the vehicle data in the corresponding regions of the fused data using the zoning adjustment coefficients for the first, second, third, and fourth regions, respectively, to obtain the corrected fused data. Specifically, this includes: first, retrieving the boundary coordinate ranges of the first to fourth regions defined in step 203 (including the latitude and longitude boundary thresholds of each region and the inner and outer coordinate boundaries of the ring-shaped distribution); performing region attribution determination on each piece of fused data before correction; extracting the vehicle geographic coordinates (latitude and longitude) from each piece of data; and comparing them one by one with the boundary coordinate ranges of the four regions. If the vehicle coordinates fall within the boundary range of a certain region, then the data is determined to belong to that region. First, all data is categorized by region. Then, the specific types of vehicle data to be corrected are defined, including core data items such as the number of vehicle location records per unit time, the number of times a vehicle enters or leaves a corresponding region, and the duration of stay within the region. For all vehicle data within each region, a weighted correction is performed by applying the corresponding zoning adjustment coefficient. The correction calculation logic for a single data point is: Corrected single vehicle data point = Original value of the data point × Zoning adjustment coefficient of its region. This method gives higher weight to vehicle data from high-density, highly dynamic regions, strengthening the impact of key region data on subsequent analysis. Finally, all weighted corrected vehicle data from the four regions are integrated in a unified format, and invalid redundant records generated during the correction process are removed, resulting in a structurally standardized and accurate corrected fused data set.

[0045] Step 301: Based on the corrected and fused data, calculate the vehicle inflow rate, outflow rate, and instantaneous vehicle inventory for each preset electronic fence station within a continuous time slice. Specifically, this includes: based on the corrected and fused data, first defining the calculation rules; the preset continuous time slice duration is 5 minutes, and each time slice is numbered sequentially according to time sequence; the fixed geographical boundary of each preset electronic fence station is determined using a closed polygon area defined by latitude and longitude coordinates, based on the station's functional positioning (e.g., surrounding bus stops, community entrances / exits, pedestrian walkways in commercial areas) and planned parking capacity. Specifically, the boundary is defined by using the latitude and longitude of the station's planned center point as a reference, and determining 4 to 6 vertices based on the available parking space in the surrounding area. The vertices are arranged clockwise to form a closed boundary, with the boundary area matching the planned parking capacity (corresponding to space for 20 to 50 shared bicycles). The latitude and longitude coordinates of all vertices are pre-entered into the system as fixed spatial boundaries for indicator statistics. For each electronic fence station, core indicators are calculated segment by segment according to time slices. The calculation of vehicle inflow rate first extracts the location records of all vehicles from the corrected fused data within the current time slice. By comparing the positional relationship between the vehicle's latitude and longitude coordinates and the closed polygon boundary of the station, two types of vehicles are selected: first, vehicles whose coordinates are outside the boundary at the start of the time slice and inside the boundary at the end of the time slice; second, vehicles whose location records within the time slice... For vehicles crossing from outside the boundary to inside the boundary, the total number of these two types of vehicles is counted, which is the total number of vehicles flowing in within the time slice. This total number is then divided by the time slice duration (5 minutes). The calculation logic is: Vehicle inflow rate = Total number of vehicles flowing in within the time slice ÷ 5, and the result is the number of vehicles flowing in per unit time. For vehicle outflow rate, within the current time slice, the positional relationship between the vehicle's location coordinates and the station boundary is compared to filter out two types of vehicles: first, vehicles whose coordinates are inside the boundary at the start of the time slice and outside the boundary at the end of the time slice; second, vehicles whose location records show them crossing from inside the boundary to outside the boundary within the time slice. The total number of these two types of vehicles is counted, which is the total number of vehicles outflowing within the time slice. This total number is then used to calculate the outflow rate. Divided by the time slice duration (5 minutes), the calculation logic is as follows: vehicle outflow rate = total number of vehicles outflowing within the time slice ÷ 5, the result being the number of vehicles outflowing per unit time; instantaneous inventory is calculated using the end time of each time slice as the statistical node. The initial inventory of the first time slice is determined by filtering and correcting the data to find the number of vehicles whose starting time coordinates are within the station boundary and have no starting time departure records; the instantaneous inventory of subsequent time slices is calculated as follows: instantaneous inventory = instantaneous inventory at the end of the previous time slice + total number of vehicles flowing in within the current time slice - total number of vehicles outflowing within the current time slice, the result being the total number of vehicles actually parked within the station boundary at the statistical node time.

[0046] Step 302 involves organizing the vehicle inflow rate, outflow rate, and instantaneous vehicle inventory of each preset electronic fence station within consecutive time slices into multi-dimensional time series features. These multi-dimensional time series features are then aggregated to generate a spatiotemporal feature tensor. Specifically, this includes: first, organizing the indicator data for each preset electronic fence station according to the order of time slice numbers, that is, extracting the vehicle inflow rate, outflow rate, and instantaneous vehicle inventory of that station in all consecutive time slices sequentially, and organizing them into a two-dimensional data sequence (row dimension: total number of time slices; column dimension: 3 core indicators) according to the correspondence between time slice number and indicator type, forming the multi-dimensional time series features of a single station. Subsequently, feature aggregation processing is initiated, using all... The unique ID of each electronic fence station is used as the spatial dimension (arranged in ascending order of ID). The time slice number is used as the time dimension (arranged in chronological order). The vehicle inflow rate, outflow rate, and instantaneous vehicle inventory are used as feature dimensions (the order of these indicators is fixed). The two-dimensional time series features of all stations are stacked one by one according to the spatial dimension. Specifically, the two-dimensional sequence of each station is used as a spatial slice of a tensor. All slices are combined in the order of station ID to finally construct a three-dimensional spatiotemporal feature tensor of total number of stations × total number of time slices × 3. In this tensor, each element precisely corresponds to the specific value of a certain station, a certain time slice, and a certain core indicator, thus completely preserving the temporal evolution of vehicle dynamics at each station and the spatial distribution correlation information between different stations.

[0047] This embodiment differentiates vehicle data in different regions by adjusting the partitioning coefficients, allowing the data to more accurately reflect the vehicle distribution characteristics of each region and reducing interference from non-critical area data. It focuses on the inflow and outflow rates and instantaneous inventory of electronic fence stations, comprehensively depicting the real-time changes in vehicle supply and demand at the stations, thus overcoming the limitations of focusing only on a single quantitative indicator. It aggregates multi-dimensional indicators into feature tensors according to the spatiotemporal dimension, which not only preserves the temporal evolution of vehicle data but also takes into account the spatial correlation between different stations, helping to avoid judgment biases caused by local data analysis.

[0048] In a preferred embodiment of the present invention, step 4 includes:

[0049] Step 400: Input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network. By analyzing the spatial geometric features of vehicle distribution, identify the geometric morphological features of vehicle cluster areas. Specifically, in step 400a, input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network to extract the latitude and longitude coordinates of all shared bicycles in each electronic fence station within a preset time period to form a set of vehicle distribution location points. The preset time period is set to the next 30 minutes. The spatiotemporal recurrent neural network construction and training process is as follows: First, collect riding data, vehicle positioning data, and environmental related data from the past 6 months as the basic training dataset. Perform preprocessing on this dataset consistent with the corrected fused data (remove outliers, unify coordinate system, and perform association matching). Then, divide the data into 80% as the training set and 20% as the validation set. The network structure adopts an architecture combining convolutional layers and recurrent layers. The convolutional layers are used to extract the spatial dimension of station distribution features, with their weights initialized to random values ​​in the range [-0.05, 0.05] and their biases initialized to 0. The recurrent layers are used to capture the temporal dimension of traffic flow patterns, with their weights initialized to random values ​​in the range [-0.05, 0.05] and their biases initialized to 0.01. The training objective is to minimize the error in predicting vehicle distribution, and the mean squared error is selected as the loss function. The network weights and biases are iteratively adjusted through a gradient descent optimizer. After each training round, the model's prediction accuracy is verified using a validation set. The preset validation set error threshold is 0.02. When the validation set error tends to stabilize and is below this threshold for 5 consecutive rounds, training stops, ultimately forming a spatiotemporal recurrent neural network with accurate analysis of spatiotemporal feature correlation information.

[0050] The three-dimensional spatiotemporal feature tensor is input into the trained neural network. The network first parses the spatial dimension (site number), temporal dimension (time slice number), and feature dimension association information in the tensor. Based on the preset future 30-minute time range, the tensor data for the corresponding time period is selected. Based on the identification information of each site in the tensor, the vehicle location records in the data are associated, corrected, and fused. The latitude and longitude coordinates of all shared bicycles in each electronic fence site within the next 30 minutes are extracted one by one to ensure that each vehicle that stops, enters, or leaves the site corresponds to a complete and continuous coordinate record, avoiding the omission of key location information. Finally, the latitude and longitude coordinates of vehicles at all sites are summarized and integrated to form a set of vehicle distribution location points covering the next 30 minutes and containing all relevant vehicle location information.

[0051] Step 400b involves calculating the minimum circumcircle of the vehicle distribution location set, obtaining the center coordinates and radius of the minimum circumcircle. Specifically, this includes: firstly, using a local Cartesian coordinate transformation method, converting all latitude and longitude coordinates in the vehicle distribution location set to Cartesian coordinates to eliminate spherical deviations in geographic coordinates. The specific transformation logic is as follows: firstly, select the center latitude and longitude of the vehicle distribution area as a reference point, and convert the latitude and longitude of this reference point to radians. The conversion method is: longitude radians equal to the reference point's longitude multiplied by π and then divided by 180; latitude radians equal to the reference point's latitude multiplied by π and then divided by 180. Next, calculate the latitude and longitude distance conversion factor for the local area. The actual distance corresponding to each degree of latitude is fixed at 111319.9 meters, and the actual distance corresponding to each degree of longitude is 111319.9 meters multiplied by the cosine of the reference point's latitude radians. Because longitude distance varies with latitude, and the latitude of the reference point within the local area can represent the overall latitude level, the conversion accuracy can be guaranteed. Subsequently, the conversion factor is... For each vehicle, its latitude and longitude coordinates are calculated, and the difference between its latitude and longitude and the reference point is calculated. The longitude difference is the vehicle's longitude minus the reference point's longitude, and the latitude difference is the vehicle's latitude minus the reference point's latitude. Finally, the latitude and longitude differences are converted to planar coordinates. The x-axis coordinate (corresponding to the longitude direction) is equal to the longitude difference multiplied by the distance per degree in the longitude direction, and the y-axis coordinate (corresponding to the latitude direction) is equal to the latitude difference multiplied by the distance per degree in the latitude direction. The reference point's planar coordinates are set to (0,0). This yields the x-axis and y-axis Cartesian coordinates of each vehicle's position. After the coordinate transformation, all transformed planar coordinates are iterated, and the maximum and minimum values ​​of the x-axis (corresponding to the longitude direction) and y-axis (corresponding to the latitude direction) are recorded. These four extreme values ​​define the initial rectangular boundary of the point set. Then, the center coordinates of this initial boundary are calculated: the x-axis center is the sum of the x-axis maximum and minimum values ​​divided by 2, and the y-axis center is the sum of the y-axis maximum and minimum values ​​divided by 2. This center is used as the temporary center of a circle.

[0052] Using a temporary center as a reference, calculate the Euclidean distance from each point in the point set to each coordinate. The calculation method is to add the square of the difference between the location point and the temporary center along the x-axis to the square of the difference along the y-axis, then take the square root. Record the maximum value among all distances as the temporary radius. To find the circumcircle with the smallest radius, set an iterative adjustment step size of 0.1 meters. Alternately adjust the coordinates of the temporary center along the positive and negative x-axis and y-axis, changing only one axis coordinate each time while keeping the other axis unchanged. After adjustment, recalculate the Euclidean distance to all points and update the maximum distance. When the change in the temporary radius is less than 0.01 meters after two consecutive iterations, stop the iteration. The circle obtained at this point is the circumcircle. The smallest circumcircle with the smallest radius that completely covers all coordinates of the point set is determined. After determining the planar coordinate parameters of the smallest circumcircle, the Cartesian coordinates of the circle's center are converted back to latitude and longitude coordinates through the inverse operation of the above transformation method. First, the difference between the plane coordinates of the circle's center and the plane coordinates of the reference point is calculated, i.e., the x-axis difference is the x-axis coordinate of the circle's center, and the y-axis difference is the y-axis coordinate of the circle's center. Then, this plane coordinate difference is divided by the corresponding distance per degree in the latitude and longitude directions to obtain the longitude difference and latitude difference. Finally, the corresponding difference is added to the latitude and longitude degrees of the reference point (the longitude difference corresponds to the longitude degree of the reference point, and the latitude difference corresponds to the latitude degree of the reference point), which gives the latitude and longitude coordinates of the circle's center. Since the coordinate transformation error in the local area is extremely small, the radius calculated from the plane coordinates can be directly equated to the actual geographical distance, thus finally determining the latitude and longitude coordinates of the center of the smallest circumcircle and its geographical radius.

[0053] Step 400c: Based on the distances between all locations in the vehicle distribution location set and the center of the circle, calculate the standard distance of the vehicle distribution location set. Specifically, this includes: using the Cartesian coordinates of the center of the smallest circumcircle after the transformation in step 400b, and the Cartesian coordinates of each vehicle location, calculate the Euclidean distance between each location point and the center of the circle, forming a dataset containing the distances from all locations to the center of the circle. Sum all the values ​​in this dataset and then divide by the total number of locations. The calculation logic is: Standard distance = Sum of all distances from all locations to the center of the circle ÷ Total number of locations. The result is presented in actual geographical distance units (such as meters), which is the standard distance of the vehicle distribution location set.

[0054] Step 400d: Based on the center and standard distance, determine the equivalent circle of the vehicle distribution location set and obtain the area of ​​the equivalent circle. Specifically, this includes: using the center of the smallest circumcircle obtained in step 400b (latitude and longitude coordinates after inverse projection) as the center of the equivalent circle, and using the standard distance calculated in step 400c as the radius of the equivalent circle, the area of ​​the circle is calculated using the formula for calculating the area of ​​a circle: equivalent circle area = fixed constant π × standard distance × standard distance (π is taken as 3.1416). The unit of the result corresponds to the unit of the standard distance. If the standard distance is meters, then the unit of the equivalent circle area is square meters, ensuring that the area calculation matches the actual geographic space.

[0055] Step 400e: Calculate the ratio of the area of ​​the equivalent circle to the area of ​​the minimum circumcircle to obtain the geometric attribute parameter representing the spatial clustering degree of the vehicle distribution point set. The smaller the value of the geometric attribute parameter, the higher the clustering degree. Specifically, this includes: first, calculating the area of ​​the minimum circumcircle according to the same calculation rule as the area of ​​the equivalent circle: minimum circumcircle area = fixed constant π × minimum circumcircle radius × minimum circumcircle radius (π is 3.1416, and the radius is the geographical radius obtained in step 400b); then, dividing the area of ​​the equivalent circle by the area of ​​the minimum circumcircle to obtain the geometric attribute parameter. The calculation logic is: geometric attribute parameter = equivalent circle area ÷ minimum circumcircle area. This parameter is a dimensionless value, ranging from 0 to 1. The smaller the value, the more concentrated the vehicle distribution and the higher the spatial clustering degree.

[0056] Step 400f: Compare the geometric attribute parameters with a first preset threshold to obtain a comparison result. Based on the comparison result, determine whether the vehicle distribution pattern is a clustered distribution pattern or a dispersed distribution pattern. Based on the judgment result, generate the geometric morphological features of the vehicle cluster area. The geometric morphological features include the coordinates of the center point of the cluster area, the radius of the equivalent circle representing the vehicle distribution range, and the density change trend of the vehicle distribution location point set. Specifically, based on the historical operating data of the past 6 months, statistically analyze the geometric attribute parameters of all vehicle cluster cases (with subsequent risk of congestion) and dispersed cases (with subsequent risk of shortage). Determine the first preset threshold as 0.3. This threshold has been verified multiple times and can distinguish the two distribution patterns to the greatest extent. The geometric attribute parameters of clustered cases are all ≤0.3, and those of dispersed cases are all >0.3. Compare the currently calculated geometric attribute parameters with 0.3. If the parameter is ≤0.3, it is determined to be a clustered distribution pattern; if the parameter is >0.3, it is determined to be a dispersed distribution pattern.

[0057] Based on the judgment results, geometric morphological features are generated. Specifically, the center point coordinates of the cluster area are directly taken from the latitude and longitude center coordinates of the smallest circumscribed circle; the equivalent circle radius is the standard distance calculated in step 400c (retaining two decimal places); the density change trend of the vehicle distribution location point set is calculated by statistically analyzing the point set density of six 5-minute time slices within the next 30 minutes (number of vehicles in each time slice ÷ area of ​​the equivalent circle), calculating the density difference between two adjacent time slices (density of the later slice minus density of the earlier slice), and determining the density growth or decline trend based on the positive or negative value of the difference, forming a continuous density increase and decrease curve, thereby quantifying the evolution of density over time.

[0058] Step 401: Based on geometric morphological features, determine whether the vehicle distribution exhibits a clustered distribution pattern. When a clustered distribution pattern is identified, calculate the probability of vehicle congestion risk at each electronic fence station within a preset time period. When a dispersed distribution pattern is identified, calculate the probability of vehicle shortage risk at each electronic fence station within a preset time period. Specifically, this includes: reconfirming the vehicle distribution pattern based on the generated geometric morphological features, calculating the risk probability according to the corresponding pattern, and if it is a clustered distribution pattern, calculating the probability of vehicle congestion risk. First, obtain the maximum carrying capacity per unit area of ​​the electronic fence station, and then calculate the risk probability according to the station planning. The parking capacity (e.g., 50 vehicles) is divided by the site boundary area (e.g., 100 square meters) to calculate the risk. Then, the frequency weight of historical congestion occurrences is obtained. This involves statistically analyzing the number of times the site experienced congestion risk during the same time period (e.g., weekday morning rush hour) and under similar environmental conditions (e.g., sunny days, normal traffic flow) over the past 6 months, dividing this number by the total number of days to obtain the frequency. This frequency is multiplied by 1.0 as the weight (range 0.7 to 1.2). Combined with the instantaneous parking volume trend within the next 30 minutes (1.0 for growth, 0.8 for stability, and 0.5 for decline), the calculation logic is as follows: Congestion risk... Rate = (Current vehicle density within the equivalent circle ÷ Maximum carrying capacity density per unit area of ​​the station) × (1 ÷ Geometric attribute parameter) × Historical consolidation frequency weight × Instantaneous inventory trend coefficient, with the result rounded to three decimal places; If the distribution pattern is dispersed, the probability of vehicle shortage risk is calculated by first obtaining the predicted vehicle demand value of the surrounding area, i.e., combining the average historical outflow rate of the three adjacent stations and the current environmental data (a coefficient of 1.2 is used for large traffic flow, 1.0 for normal flow, and 0.8 for small flow) for weighted estimation; then, the historical shortage frequency weight is obtained, i.e., the past 6 The frequency of shortage risk at a station during the same period of the month and under similar environmental conditions (such as sunny days and normal traffic flow) is calculated by dividing the number of times the station experiences shortage risk by the total number of days. The frequency is then multiplied by 1.0 as a weight (with a value range of 0.7 to 1.2). Combined with the trend of vehicle outflow rate changes in the next 30 minutes (a coefficient of 1.2 is used for growth, 1.0 for stability, and 0.8 for decline), the calculation logic is as follows: Shortage risk probability = (Current station vehicle outflow rate ÷ Forecast value of vehicle demand in the surrounding area) × Geometric attribute parameter × Weight of historical shortage occurrence frequency in the same period × Outflow rate trend coefficient. The result is rounded to three decimal places.

[0059] Step 402: Based on the probability of vehicle congestion or vehicle shortage, determine the corresponding risk level and generate a risk assessment result including risk type, risk probability, and risk level. Specifically, this includes: preset risk probability grading intervals (set based on operational safety thresholds and user experience requirements): low risk (0 ≤ probability < 0.3), low-to-medium risk (0.3 ≤ probability < 0.5), medium risk (0.5 ≤ probability < 0.7), and medium-to-high risk (0.7 ≤ probability < 0.5). For high risk (0.9 ≤ probability ≤ 1.0), the calculated probability of vehicle congestion or vehicle shortage is substituted into the interval for comparison. If the probability falls within the corresponding interval, it is determined to be the corresponding risk level. For example, if the probability is 0.62, it is determined to be medium risk. Key information is integrated to generate structured risk assessment results. Each record includes the unique identifier of the electronic fence station, the station's geographical location (latitude and longitude range), the preset time period (future 30 minutes), the risk type (congestion or shortage), the risk probability (retaining three decimal places), the risk level (low / low-medium / medium / medium-high / high), and a summary of geometric features (center point coordinates, equivalent circle radius, density change trend). All assessment records are sorted from high to low risk level and stored synchronously in the analysis results database.

[0060] This embodiment extracts vehicle location points and calculates geometric features to transform the spatial state of vehicle distribution into quantifiable parameters, identifying clustering or dispersion patterns and overcoming the difficulty in characterizing spatial relationships. The risk probability calculation integrates spatial geometric attributes, site operation characteristics, and historical data weights, avoiding the limitations of relying solely on local quantitative data. The assessment results include risk type, probability, and level, along with geometric morphological features and density change trends. By combining the analysis of global spatial features with local site status, it effectively avoids risk misjudgments caused by local data analysis.

[0061] In a preferred embodiment of the present invention, step 5 includes:

[0062] Step 500: Based on the risk probability and risk level in the risk assessment results, when the risk probability exceeds the second preset threshold, a vehicle shortage warning or vehicle congestion warning for the corresponding risk level is triggered. The warning includes the warning type, risk level, and information on affected electronic fence sites. Specifically, it includes: First, based on historical warning response data, operational service quality targets, and user experience thresholds, a second preset threshold is set. The second preset thresholds for shortage risk and congestion risk are uniformly set according to the risk level. Low risk (0 ≤ probability < 0.3) and low-to-medium risk (0.3 ≤ probability < 0.5) have no second preset threshold (no warning is triggered); medium risk (0.5 ≤ probability < 0.5) has no second preset threshold. The second preset threshold corresponding to 0.7 is 0.55; the second preset threshold corresponding to medium-high risk (0.7≤probability<0.9) is 0.75; and the second preset threshold corresponding to high risk (0.9≤probability≤1.0) is 0.9. This threshold has been verified by 6 months of operational data and can balance the timeliness of early warning and scheduling costs, avoiding excessive or delayed early warnings. The risk assessment results generated in step 402 are retrieved, and the risk probability of each electronic fence site is compared with the second preset threshold of the corresponding risk level according to the risk type (shortage or congestion). If the risk probability exceeds the threshold, the corresponding type of early warning is triggered immediately. The warning content is generated according to a fixed structure, that is, the warning type directly matches the risk type (vehicle shortage warning or vehicle congestion warning). The risk level adopts the level in the assessment results. The information of the affected electronic fence sites includes the unique identifier of the site, the latitude and longitude location range, the current risk probability, and the estimated impact range corresponding to the risk level. The estimated impact range is calculated according to the following logic: first, the actual service radius of the preset fixed site is determined according to the functional positioning of the site. Specifically, the coverage range is set as follows: 600 meters for bus hubs, 500 meters for core business districts, 400 meters for community entrances and exits, and 450 meters for schools / hospitals. These values ​​are based on actual surveys of user travel radius in different scenarios to cover the travel needs of regular users in the surrounding areas. The impact range coefficient is adjusted by the difference between the risk probability and the threshold. The coefficient is calculated as (risk probability - second preset threshold) ÷ (1 - second preset threshold), with a range of 0 to 1. The closer the risk probability is to 1, the more urgent the risk and the wider the impact. The final impact range is quantified by the radius of influence. The calculation logic is: radius of influence = actual service radius of the station × (1 + impact range coefficient). For example, if the actual service radius of the station is 500 meters and the impact range coefficient is 0.3, the radius of influence = 500 meters × (1 + 0.3) = 650 meters. This means that the affected area of ​​the warning is expanded by 30% from the original 500 meters, ultimately covering an area within 650 meters, ensuring that the warning information accurately defines the affected geographical area.

[0063] Step 501: Based on the risk level in the risk assessment results, the geographical location information of the electronic fence station, and the surrounding traffic status information in the environmental data, a dynamic vehicle dispatching strategy is generated. This strategy includes the number of dispatched vehicles, dispatching routes, dispatching priorities, and estimated dispatching times. The dispatching routes are optimized based on the surrounding traffic status information, and the dispatching priorities are determined according to the risk level. Specifically, this includes: first, setting a risk level coefficient: 1.2 for high risk, 1.0 for medium-high risk, and 0.8 for medium risk. This coefficient is set based on the urgency of the risk and the redundancy requirements of dispatching resources. If there is a vehicle shortage risk, first obtain the maximum demand value per unit time for the electronic fence station (set in conjunction with the station's functional positioning and historical operating data: 30 vehicles / hour for stations around bus hubs, 25 vehicles / hour for stations in the core area of ​​commercial districts, 15 vehicles / hour for stations at community entrances and exits, and 20 vehicles / hour for stations around schools / hospitals). The system calculates the number of vehicles to be dispatched based on the following parameters: (maximum demand per unit time - current instantaneous inventory) × risk level coefficient, rounded up to ensure coverage of demand gaps at stations while reserving a small amount of redundancy to avoid shortages due to demand fluctuations. For the risk of vehicle congestion, the system obtains the station's maximum capacity (calculated based on the station's planned area and parking density standards: 40 vehicles for stations around bus hubs, 35 vehicles for stations in core business districts, 25 vehicles for stations at community entrances / exits, and 30 vehicles for stations around schools / hospitals, with parking density calculated according to the industry standard of 1.5 vehicles / square meter), the current instantaneous inventory, and the corresponding risk level coefficient. The calculation logic is: number of vehicles to be dispatched = (current instantaneous inventory - maximum capacity) × risk level coefficient, rounded up to ensure the removal of redundant vehicles exceeding the capacity, preventing vehicle accumulation from affecting normal station operation and surrounding traffic.

[0064] Retrieve surrounding traffic status information from environmental data, including real-time traffic flow, congested road segment distribution, and traffic efficiency coefficients (1.0 for smooth traffic, 0.8 for light congestion, 0.5 for moderate congestion, and 0.3 for heavy congestion). Using affected stations as destinations (shortage risk) or origins (congestion risk), and combining the current locations of available dispatch vehicles within a 3-kilometer radius, generate 3 to 5 candidate routes. For each candidate route, calculate the total travel time by weighting the road segment traffic efficiency coefficients. The logic is as follows: first, for each road segment in the route, divide the distance of the segment by the real-time average speed of the segment, and then multiply by the traffic efficiency coefficient of the segment to obtain the weighted travel time of a single segment. Sum the weighted travel times of all single segments to obtain the total travel time of the candidate route. Select the route with the shortest total travel time as the optimized dispatch route to avoid congested road segments and improve dispatch efficiency.

[0065] Dispatch priorities are assigned based on risk level: high risk corresponds to Level 1 priority, medium-high risk to Level 2 priority, and medium risk to Level 3 priority, with priorities ordered from highest to lowest. When multiple dispatch sites exist, dispatch resources are allocated according to priority order, with Level 1 priority sites receiving priority access to available dispatch vehicles. If resources are insufficient, allocation follows a principle of prioritizing nearby sites to ensure high-risk areas receive the first response. The estimated dispatch time consists of two parts: vehicle assembly time and route travel time. Vehicle assembly time is determined based on the current density of dispatched vehicles (≥3 vehicles / km² = 10 minutes, 1 to 3 vehicles / km² = 10 minutes). (15 minutes per kilometer, 20 minutes for less than 1 vehicle per square kilometer); Route travel time is the total travel time of the optimized dispatch route, calculated as: Estimated dispatch time = Vehicle assembly time + Route travel time, accurate to the minute. This provides a time reference for dispatch execution and user notification. Integrating the above calculation results with the set information, a complete dynamic vehicle dispatch strategy is generated, including the number of dispatched vehicles, the optimized dispatch route (including road segment details and travel time), the dispatch priority determined by risk level, and the estimated dispatch time accurate to the minute. This strategy is simultaneously pushed to the dispatch management system and the terminals of relevant dispatch personnel.

[0066] This embodiment triggers early warnings based on a scientific comparison of risk probability and preset thresholds, clearly identifying affected site information and helping operators quickly locate risk areas and proactively address potential problems. By combining risk level, geographical location, and traffic status to generate strategies, the number of dispatched vehicles closely matches actual needs, routes avoid congestion, and priorities are clearly defined, improving dispatch execution efficiency. Dispatch resources are allocated according to priority to avoid waste, ensuring high-risk areas receive priority responses and balancing overall operational needs. Through timely early warnings and dynamic dispatching, vehicle shortages or congestion are effectively alleviated, reducing user inconvenience while lowering operating costs and improving overall operational service levels.

[0067] like Figure 2 As shown, embodiments of the present invention also provide a multi-dimensional risk data early warning management system for shared bicycles, including:

[0068] The fusion module is used to acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates.

[0069] The partitioning module is used to construct a geographic bounding box covering all vehicle locations as a virtual reference plane based on fused data. The virtual reference plane is divided into four partitions according to the distribution density characteristics of vehicles on the virtual reference plane. The vehicle number change rate and spatiotemporal stability index of each partition are calculated to obtain the partitioning adjustment coefficient.

[0070] The aggregation module is used to perform weighted correction processing on the fused data using partition adjustment coefficients to obtain corrected fused data; based on the corrected fused data, the vehicle inflow rate, outflow rate and instantaneous holding capacity of the preset electronic fence stations in a continuous time slice are calculated, and feature aggregation is performed to obtain the spatiotemporal feature tensor.

[0071] The analysis module is used to input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network, analyze the spatial geometric features of vehicle distribution, and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period.

[0072] The scheduling module is used to trigger vehicle shortage warnings or congestion warnings based on the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level, and to generate dynamic vehicle scheduling strategies based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

[0073] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0074] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0075] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0076] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A multi-dimensional risk data early warning management method for shared bicycles, characterized in that, The method includes: Step 1: Acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates; Step 2: Based on the fused data, construct a geographic bounding box covering all vehicle locations as a virtual reference plane. Divide the virtual reference plane into four zones according to the distribution density characteristics of vehicles on the virtual reference plane. Calculate the vehicle number change rate and spatiotemporal stability index for each zone to obtain the zone adjustment coefficient. Step 3: Apply a partition adjustment coefficient to the fused data for weighted correction to obtain the corrected fused data; calculate the vehicle inflow rate, outflow rate and instantaneous holding capacity of the preset electronic fence station in a continuous time slice based on the corrected fused data, and perform feature aggregation to obtain the spatiotemporal feature tensor. Step 4: Input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network to analyze the spatial geometric features of vehicle distribution and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period. Step 5: Based on the probability of vehicle congestion risk or the probability of vehicle shortage risk and the corresponding risk level, trigger the corresponding risk level vehicle shortage warning or congestion warning, and generate a dynamic vehicle dispatching strategy based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

2. The shared bicycle multi-dimensional risk data early warning management method according to claim 1, characterized in that, Step 2 includes: Based on the maximum and minimum latitude and longitude values ​​of the vehicle location coordinates in the fused data, a geographic bounding box covering all vehicle locations is determined, and the geographic bounding box is used as a virtual reference plane. The vehicle position coordinates information in the fused data are projected onto a virtual reference plane to obtain the vehicle distribution data on the virtual reference plane. Based on the vehicle distribution data on the virtual reference plane, calculate the vehicle density value at each location on the virtual reference plane; Based on the spatial distribution characteristics of vehicle density values, a density-based spatial partitioning algorithm is used to divide the virtual reference plane into four continuous partitions. The vehicle density characteristics within each partition are the same, and there are differences in the vehicle density characteristics between the partitions. The rate of change of vehicle quantity in each partition within a preset time window and the spatiotemporal stability index of each partition within a preset time window are calculated respectively. The spatiotemporal stability index reflects the spatiotemporal fluctuation characteristics of vehicle distribution. Based on the vehicle quantity change rate index and spatiotemporal stability index of the four zones, the zone adjustment coefficient of each zone is obtained through weighted fusion calculation.

3. The shared bicycle multi-dimensional risk data early warning management method according to claim 2, characterized in that, Based on the spatial distribution characteristics of vehicle density values, a density-based spatial partitioning algorithm is used to divide the virtual reference plane into four continuous partitions. Each partition contains the same vehicle density characteristics, but the vehicle density characteristics differ between partitions, including: Based on the spatial distribution characteristics of vehicle density values, a density-based spatial partitioning algorithm is used to divide the virtual reference plane into a first region, a second region, a third region, and a fourth region. The first region is the region with the highest vehicle density value and is located at the center of the virtual reference plane. The second region surrounds the first region and has a lower vehicle density value than the first region. The third region surrounds the second region and has a lower vehicle density value than the second region. The fourth region is located at the outermost edge of the virtual reference plane and has a lower vehicle density value than the third region.

4. The shared bicycle multi-dimensional risk data early warning management method according to claim 3, characterized in that, Step 3 includes: The vehicle data in the corresponding regions of the fused data are weighted and corrected by using the partition adjustment coefficients of the first region, the second region, the third region and the fourth region respectively, to obtain the corrected fused data; Based on the corrected fused data, the vehicle inflow rate, outflow rate and instantaneous vehicle holding capacity of each preset electronic fence station are calculated in a continuous time slice. The vehicle inflow rate, outflow rate, and instantaneous vehicle inventory of each preset electronic fence station within a continuous time slice are organized into multi-dimensional time series features according to the time series, and the multi-dimensional time series features are aggregated to generate a spatiotemporal feature tensor.

5. The shared bicycle multi-dimensional risk data early warning management method according to claim 4, characterized in that, Step 4 includes: The spatiotemporal feature tensor is input into the spatiotemporal recurrent neural network, and the geometric features of vehicle clusters are identified by analyzing the spatial geometric features of vehicle distribution. Based on geometric morphological features, determine whether the vehicle distribution exhibits a clustered distribution pattern. When a clustered distribution pattern is identified, calculate the probability of vehicle congestion risk at each electronic fence station within a preset time period. When a dispersed distribution pattern is identified, calculate the probability of vehicle shortage risk at each electronic fence station within a preset time period. Based on the probability of vehicle congestion or vehicle shortage, the corresponding risk level is determined, and a risk assessment result containing the risk type, risk probability, and risk level is generated.

6. The shared bicycle multi-dimensional risk data early warning management method according to claim 5, characterized in that, By inputting spatiotemporal feature tensors into a spatiotemporal recurrent neural network, and analyzing the spatial geometric features of vehicle distribution, the geometric morphological features of vehicle cluster areas are identified, including: The spatiotemporal feature tensor is input into the spatiotemporal recurrent neural network to extract the latitude and longitude coordinates of all shared bicycles in each electronic fence station within a preset time period, forming a set of vehicle distribution location points; Calculate the minimum circumcircle of the set of vehicle distribution points, and obtain the coordinates of the center and radius of the minimum circumcircle; Calculate the standard distance of the vehicle distribution location set based on the distances between all location points in the vehicle distribution location set and the center of the circle; Based on the center and standard distance, determine the equivalent circle of the vehicle distribution location point set, and obtain the area of ​​the equivalent circle; The ratio of the area of ​​the equivalent circle to the area of ​​the smallest circumcircle is calculated to obtain the geometric attribute parameter representing the degree of spatial clustering of the vehicle distribution location point set. The smaller the value of the geometric attribute parameter, the higher the degree of clustering. The geometric attribute parameters are compared with a first preset threshold to obtain a comparison result. Based on the comparison result, it is determined whether the vehicle distribution pattern is a clustered distribution pattern or a dispersed distribution pattern. Based on the judgment result, the geometric morphological features of the vehicle cluster area are generated. The geometric morphological features include the coordinates of the center point of the cluster area, the radius of the equivalent circle representing the vehicle distribution range, and the density variation trend of the vehicle distribution location point set.

7. The shared bicycle multi-dimensional risk data early warning management method according to claim 6, characterized in that, Step 5 includes: Based on the risk probability and risk level in the risk assessment results, when the risk probability exceeds the second preset threshold, a vehicle shortage warning or vehicle congestion warning of the corresponding risk level is triggered. The warning includes the warning type, risk level and information on the affected electronic fence sites. Based on the risk level in the risk assessment results, the geographical location information of the electronic fence station, and the surrounding traffic status information in the environmental data, a dynamic vehicle dispatching strategy is generated. The dynamic vehicle dispatching strategy includes the number of dispatched vehicles, the dispatching route, the dispatching priority, and the expected dispatching time. The dispatching route is optimized according to the surrounding traffic status information, and the dispatching priority is determined according to the risk level.

8. A multi-dimensional risk data early warning management system for shared bicycles, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The fusion module is used to acquire cycling data, vehicle location data, and environmental data, and preprocess them to generate fused data with geographic coordinates. The partitioning module is used to construct a geographic bounding box covering all vehicle locations as a virtual reference plane based on fused data. According to the distribution density characteristics of vehicles on the virtual reference plane, the virtual reference plane is divided into four partitions. The vehicle number change rate and spatiotemporal stability index of each partition are calculated to obtain the partition adjustment coefficient. The aggregation module is used to perform weighted correction processing on the fused data using partition adjustment coefficients to obtain the corrected fused data; Based on the corrected fused data, the inflow rate, outflow rate and instantaneous holding capacity of vehicles at the preset electronic fence stations within a continuous time slice are calculated, and feature aggregation is performed to obtain the spatiotemporal feature tensor. The analysis module is used to input the spatiotemporal feature tensor into the spatiotemporal recurrent neural network, analyze the spatial geometric features of vehicle distribution, and generate the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level for each electronic fence station within a preset time period. The scheduling module is used to trigger vehicle shortage warnings or congestion warnings based on the probability of vehicle congestion risk or vehicle shortage risk and the corresponding risk level, and to generate dynamic vehicle scheduling strategies based on the risk level, the geographical location of the electronic fence station and traffic status information in the environmental data.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images