A method, system, and storage medium for predicting traffic flow

Abstract

The application relates to the technical field of intelligent transportation, and discloses a method, system and storage medium for predicting traffic flow. The method comprises the following steps: acquiring multi-source dynamic data of a mountainous road network, constructing a road network connection relationship model, extracting traffic transfer characteristics and transfer dynamic fluctuation characteristics between nodes, performing time series trend analysis on a historical data sequence of a traffic transfer path, and predicting a transfer influence range caused by a road network state change; when the transfer influence range exceeds a preset range threshold, filtering a road subset affected by the transfer influence range, generating a traffic transfer space distribution map, updating weight parameters of the road network connection relationship model according to the traffic transfer space distribution map, performing traffic deviation simulation, determining an adjusted road network connection state characteristic, performing geographical space mapping on the road network connection state characteristic, extracting a traffic flow quantitative index, and outputting a traffic flow prediction result when a preset congestion risk condition is met. The application can improve the accuracy and pertinence of mountainous road network traffic flow prediction.

Description

Technical Field

[0001] This application relates to the field of intelligent transportation technology, and in particular to a method, system and storage medium for predicting traffic flow. Background Technology

[0002] With the development of intelligent transportation systems, traffic flow prediction technology has been widely applied in scenarios such as road operation monitoring, traffic organization optimization, travel guidance, and congestion early warning. Early research on traffic flow prediction mainly relied on empirical models based on historical statistical patterns, estimating future traffic conditions by statistically analyzing parameters such as road cross-section flow, vehicle speed, and occupancy. Subsequently, with the development of sensor networks, positioning technology, and multi-source data acquisition technology, traffic flow prediction gradually evolved from single-section data analysis to multi-source traffic information fusion, enabling the comprehensive use of road network topology, vehicle trajectory data, geospatial data, and real-time monitoring data to model traffic conditions. In recent years, with the development of graph models, time series analysis methods, and spatial correlation analysis methods, traffic flow prediction has further developed towards networking, dynamism, and refinement, focusing not only on traffic changes in individual road segments but also on the transfer and propagation patterns of traffic flow between adjacent nodes and roads, as well as the impact of local traffic disturbances on the overall road network operation.

[0003] Existing traffic flow prediction methods have achieved certain results in relatively regular road environments such as urban plains. However, for mountainous road networks, due to complex road alignments, numerous slopes and curves, significant fluctuations in traffic conditions, and susceptibility to factors such as weather, geological conditions, and the spread of local congestion, the transfer paths and evolution processes of traffic flow within the road network exhibit greater dynamism and uncertainty. Existing methods typically focus on extrapolating trends from historical traffic data to target road segments or performing conventional state predictions based on static road network structures. These methods struggle to effectively characterize the characteristics of traffic flow transfer between nodes in mountainous road networks, the dynamic fluctuations of transfer, and their impact on the expansion of the affected area. Furthermore, they lack an integrated modeling, mapping, and prediction mechanism for the spatial diffusion of traffic flow within affected road subsets, changes in road network connectivity, and the evolution of congestion risk. Consequently, the prediction results fail to accurately reflect the dynamic evolution of traffic conditions in mountainous road networks.

[0004] Therefore, how to construct a traffic flow prediction method that can accurately predict the dynamic changes in traffic flow, taking into account multi-source dynamic data, road network connectivity, traffic flow transfer characteristics, spatial distribution evolution, and congestion risk status, has become a key technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This application provides a method, system, and storage medium for predicting traffic flow, aiming to solve the technical problems in the prior art that are insufficient in characterizing the dynamic changes of traffic flow in mountainous road network scenarios, difficult to accurately reflect the laws of traffic flow transfer and propagation, and unable to effectively combine changes in road network connectivity for traffic flow prediction, resulting in low prediction accuracy and applicability.

[0006] In a first aspect, this application provides a method for predicting traffic flow, the method comprising:

[0007] S1. Obtain multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer characteristics between nodes, and extract dynamic fluctuation characteristics of traffic transfer based on the changes of traffic transfer characteristics in the time dimension.

[0008] S2. Perform time-series trend analysis on the historical data sequence of traffic transfer paths, and combine the dynamic fluctuation characteristics of transfer to predict the scope of the transfer impact caused by changes in road network status;

[0009] S3. When the scope of the transfer impact exceeds the preset threshold, the affected road subset is selected based on the road network topology, and a traffic transfer spatial distribution map is generated based on the transfer traffic density index corresponding to the road subset.

[0010] S4. Update the weight parameters of the road network connection relationship model according to the spatial distribution map of traffic transfer, and perform traffic offset simulation to determine the adjusted road network connection status characteristics.

[0011] S5. Perform geospatial mapping on the road network connectivity characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution.

[0012] S6. Determine whether the traffic flow quantitative indicators meet the preset congestion risk conditions. If so, output the traffic flow prediction results.

[0013] Secondly, this application provides a system for predicting traffic flow, the system comprising:

[0014] The feature extraction module is used to acquire multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer features between nodes, and extract dynamic fluctuation features of traffic transfer based on the changes of traffic transfer features in the time dimension.

[0015] The predictive analysis module is used to perform time-series trend analysis on historical data sequences of traffic transfer paths, and combined with the dynamic fluctuation characteristics of transfer, to predict the scope of the transfer impact caused by changes in road network status.

[0016] The filtering and mapping module is used to filter the affected road subsets based on the road network topology when the scope of the transfer impact exceeds a preset threshold, and generate a traffic transfer spatial distribution map based on the transfer traffic density index corresponding to the road subsets.

[0017] The iterative adjustment module is used to update the weight parameters of the road network connection relationship model based on the spatial distribution map of traffic transfer, and to perform traffic offset simulation to determine the characteristics of the adjusted road network connection status.

[0018] The spatial mapping module is used to perform geospatial mapping on the road network connectivity status characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution.

[0019] The prediction output module is used to determine whether the traffic flow quantification index meets the preset congestion risk conditions. If so, it outputs the traffic flow prediction result.

[0020] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned method for predicting traffic flow.

[0021] The beneficial effects of the technical solution of this application are at least as follows:

[0022] By acquiring multi-source dynamic data of mountain road networks and extracting traffic transfer features and dynamic fluctuation features based on road network connectivity models, the dynamic propagation process of traffic flow in the road network can be characterized from the traffic flow transfer relationships between nodes, thereby improving the ability to represent changes in traffic conditions in complex mountain road networks. By performing time-series trend analysis on historical data sequences of traffic transfer paths and combining them with dynamic fluctuation features to predict the scope of transfer impact, the propagation impact of road network condition changes on surrounding road segments and nodes can be identified more accurately, improving the foresight and accuracy of traffic flow prediction. Furthermore, by filtering the affected road subset and generating a traffic transfer spatial distribution map based on the transfer traffic density, the weight parameters of the road network connectivity model can be updated and traffic flow biased based on this distribution map. Traffic flow simulation can dynamically reflect the migration and redistribution process of traffic flow in local road networks, thereby improving the simulation accuracy of traffic flow evolution. By geospatial mapping of the adjusted road network connectivity characteristics, quantitative indicators of traffic flow such as average travel time and path delay increment are extracted, and a visual distribution map of traffic flow evolution is generated. This enables quantitative analysis and spatial representation of traffic conditions, improving the intuitiveness and interpretability of traffic flow prediction results. By judging congestion risk conditions based on quantitative traffic flow indicators and outputting traffic flow prediction results when the risk conditions are met, it can promptly reflect the expected traffic flow and congestion spread trend within a preset time period, thus providing more reliable technical support for traffic operation monitoring, congestion early warning, and traffic organization management in mountainous road networks. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a method for predicting traffic flow according to this application;

[0025] Figure 2 This is a schematic diagram of the local road network state before traffic diversion according to an embodiment of this application;

[0026] Figure 3 This is a schematic diagram of the local road network state after traffic diversion according to an embodiment of this application;

[0027] Figure 4 This is a schematic diagram of the structure of a system for predicting traffic flow according to this application. Detailed Implementation

[0028] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] For ease of understanding, the specific process of the embodiments of this application is described below. Figure 1 The diagram shows a flowchart of a method for predicting traffic flow provided by the present invention. The flowchart specifically includes the following steps:

[0030] S1. Obtain multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer characteristics between nodes, and extract dynamic fluctuation characteristics of traffic transfer based on the changes of traffic transfer characteristics in the time dimension.

[0031] In one specific embodiment, S1 involves acquiring multi-source dynamic data of the mountain road network and constructing a road network connectivity model, including:

[0032] By using satellite imagery and ground sensor networks, latitude and longitude coordinates and traffic status data of mountain road networks are obtained, which are defined as multi-source dynamic data;

[0033] Noise filtering and timestamp labeling are performed on multi-source dynamic data to obtain preprocessed data;

[0034] Traffic transfer paths are identified based on the time sequence of latitude and longitude coordinates in the preprocessed data, and the transfer probability distribution is determined based on the historical frequency of each traffic transfer path. At the same time, the initial time delay distribution corresponding to each traffic transfer path is determined based on the time difference in the change sequence.

[0035] Based on preprocessed data, transition probability distribution, and initial time delay distribution, a dynamic dataset of mountain road network is generated.

[0036] Based on the dynamic dataset of mountain road networks, the graph structure analysis method is used to model the connection relationships between road nodes and road connecting edges in the mountain road network, resulting in a road network connection relationship model.

[0037] Specifically, mountain roads are characterized by tunnels, bridges, continuous curves, and limited local alternative routes, making it difficult to characterize the traffic flow transfer process between adjacent roads using single-section detection data. Therefore, using the mountain road network base map as the basic spatial reference, intersections, ramp entrances, tunnel entrances, both ends of bridges, toll stations, service area entrances and exits, and the start and end points of continuous long downhill slopes are designated as road nodes, and the passable road sections between adjacent road nodes are designated as road connection edges.

[0038] Latitude and longitude coordinates and traffic status data of mountain road networks are acquired through satellite imagery and ground sensor networks, and the acquired data is defined as multi-source dynamic data. Satellite imagery is used to provide spatial location of roads, changes in road appearance, and spatial trajectory segments of vehicle targets, while the ground sensor network is used to provide cross-sectional traffic flow, average vehicle speed, occupancy, queue length, lane access status, and event alarm information. Multi-source dynamic data includes at least the latitude and longitude coordinates of road nodes, the geometric alignment of road connecting edges, the acquisition time, cross-sectional traffic flow, average vehicle speed, road occupancy, traffic status markers, and event markers.

[0039] Noise filtering is performed according to the data source. For the vehicle target center point sequence extracted from satellite imagery, target detection and association are performed first, and then map matching is performed based on the road centerline to map continuous trajectory points to the corresponding road connection edges. Trajectory points that deviate from the road red line and whose speed continuity with previous and subsequent times is inconsistent are identified as outliers and deleted. For the traffic flow, average vehicle speed, and occupancy sequences output from ground sensors, sliding window mid-range filtering is used to remove glitch values. After noise processing, the acquisition time is uniformly added to all types of data, and timestamp correction is performed according to a unified clock. When the satellite image refresh cycle is inconsistent with the ground sensor sampling cycle, all data is resampled to a unified time granularity, while retaining the original sampling source identifier. The preprocessed data includes at least road node identifiers, road connection edge identifiers, latitude and longitude coordinates, matched road location, acquisition time, traffic flow, average vehicle speed, occupancy, and traffic status.

[0040] A traffic flow transfer path is defined as a sequence of road connection edges formed when traffic flow is redistributed among road connection edges. Specifically, at a uniform time granularity, a state change sequence is constructed for each road connection edge. When the average vehicle speed of a road connection edge drops below a preset threshold or its occupancy rises above a preset threshold, and adjacent road connection edges with a topological connection to this road connection edge experience an increase in traffic flow, an increase in occupancy, or a decrease in average vehicle speed within a lag time window, this state combination is identified as a traffic flow transfer event. To avoid misjudging regular tidal fluctuations as traffic flow transfer events, it is also required that the target road connection edge and the original traffic path satisfy an alternative traffic relationship, and the travel time increment of the alternative path does not exceed a preset allowable range. Based on the above judgment conditions, each traffic flow transfer event is represented as an event record consisting of the starting road connection edge (the road connection edge that serves as the source of traffic flow transfer in the traffic flow transfer event), the target road connection edge (the road connection edge that serves as the recipient of traffic flow transfer in the traffic flow transfer event), the time of occurrence, and the delay duration. The number of events occurring within the historical observation period where the same starting road connection edge points to each target road connection edge is statistically analyzed, and the transition probability is calculated: the transition probability of a starting road connection edge pointing to a target road connection edge is equal to the number of flow transfer events occurring from that starting road connection edge to that target road connection edge, divided by the total number of all outflow transfer events from that starting road connection edge. This yields the transition probability distribution for each starting road connection edge. For each identified flow transfer path, the time difference between the moment the starting road connection edge experiences state deterioration and the moment the target road connection edge experiences flow response is calculated. This time difference is then statistically distributed in the historical event set, serving as the initial time delay distribution for that flow transfer path. The initial time delay distribution can be characterized using the mean, quantiles, and dispersion, or it can be stored using binned frequency.

[0041] The dynamic dataset for mountain road networks is established according to the data organization method required for graph structure analysis. For each sampling time, each road node, and each road connection edge, corresponding record items are created. Each record item must include at least the latitude and longitude coordinates, road grade, number of lanes, slope type, traffic volume, average speed, occupancy rate, traffic status, event marker, associated traffic transfer path identifier, corresponding transfer probability value, and corresponding initial delay statistics. When the same road connection edge corresponds to multiple possible target road connection edges, a transfer relationship record is created for each target road connection edge. The dynamic dataset for mountain road networks can be divided into a node attribute table, an edge attribute table, and a transfer relationship table. The node attribute table stores the geographical location, node type, and node identifier of road nodes; the edge attribute table stores the start and end nodes, length, road grade, design speed, traffic capacity, and real-time operating status of road connection edges; the transfer relationship table stores the starting road connection edge, target road connection edge, associated traffic transfer path identifier, transfer probability value, and initial delay distribution parameters.

[0042] Road nodes are defined as node elements in the graph, and passable road segments between adjacent road nodes are defined as directed edge elements. For bidirectional roads, two directed edges in opposite directions are established. Each node element records at least its latitude and longitude coordinates, node type, and set of adjacent edges; each directed edge records at least its starting node, target node, length, road grade, design speed, capacity, current traffic flow, current average speed, current traffic state, and traffic resistance value. The traffic resistance value is calculated based on the length of the road connection edge, average speed, and state penalty term, representing the travel cost of the corresponding road connection edge under the current traffic state. A transfer relationship layer is further superimposed between directed edges on the basic topological relationship (static reachability relationship between nodes). For starting and target road connection edges with traffic transfer event statistics, corresponding transfer relationship connections are established between them, and transfer probability values ​​and initial time delay distribution parameters are assigned to the transfer relationship connections. Therefore, the road network connectivity model simultaneously includes the basic road topology, real-time road operating status, and traffic flow transfer and propagation relationships, and can output an adjacency list, a weighted adjacency matrix, and a transition relation matrix. The adjacency list is used for topology search; the weighted adjacency matrix is ​​used to characterize the connectivity relationships between nodes and the corresponding traffic resistance; and the transition relation matrix is ​​used to characterize the flow transfer intensity and propagation delay characteristics between road connection edges.

[0043] In one specific embodiment, in S1, traffic transfer features between nodes are extracted, and dynamic fluctuation features of the transfer are extracted based on the changes in traffic transfer features over time, including:

[0044] Based on the road network connection relationship model and the transfer probability distribution corresponding to each traffic transfer path, the state transfer relationship between nodes is determined, and a traffic transfer probability matrix is ​​constructed as a traffic transfer feature.

[0045] Set a sliding time window, extract time series data of the transfer probability between each node in the flow transfer probability matrix within the sliding time window, calculate the statistical standard deviation of the time series data, and obtain the dynamic fluctuation characteristics of the transfer.

[0046] Specifically, the road network connectivity model includes road nodes, road connection edges, and topological reachability relationships between nodes. The transition probability distribution for each traffic transfer path includes the starting road node, the target road node, the corresponding transition probability, and the corresponding initial delay distribution. State transition relationships refer to the traffic flow redistribution from the source road node to the target road node via the corresponding road connection edge within a given time granularity. For each identified traffic transfer path, its starting road node, target road node, and corresponding transition probability are read, and it is determined whether the starting road node and the target road node have a direct connection relationship or satisfy an alternative reachability relationship within a preset adjacency level in the road network connectivity model. If so, the node pair is determined as a valid state transition relationship; otherwise, it is not included in the subsequent matrix construction. For target paths that are directionally restricted, restricted, closed, or lack practical detour conditions, no corresponding state transition relationship is established.

[0047] The traffic flow transfer probability matrix has rows corresponding to source road nodes and columns corresponding to target road nodes. Each element in the matrix represents the conditional probability of traffic flow transferring from a source road node to a target road node. Historical observation data is segmented according to a uniform time granularity. Within each time segment, the number of effective traffic flow transfer events from each source road node to each target road node is counted. The sum of all outflow transfer events for the same source road node is then calculated, and the number of transfer events from that source road node to a target road node is divided by the total number of outflow transfer events to obtain the transfer probability from that source road node to that target road node. These transfer probabilities are then arranged in order of target road nodes to form row vectors representing the transfer probabilities of the corresponding source road nodes. Combining these row vectors in node order yields the traffic flow transfer probability matrix for the corresponding time segment. If no outflow transfer event occurs at a source road node within a given time segment, all elements in the row corresponding to that source road node are set to zero, or the probability value from the previous adjacent time segment is retained with a missing value marker. The traffic flow transfer probability matrix characterizes the direction and intensity of traffic flow transfer between nodes.

[0048] To characterize the change in traffic transfer intensity between nodes over time, traffic transfer probability matrices are constructed for multiple consecutive time segments, resulting in a sequence of traffic transfer probability matrices arranged in chronological order. Each matrix is ​​bound to the start and end times of the corresponding time segment and is associated with the traffic status, average vehicle speed, occupancy rate, and event markers of the same time segment.

[0049] The sliding time window is a fixed-length statistical interval that moves continuously along the time axis. Its length is determined based on the sampling period, the propagation duration of traffic disturbances, and the forecast period requirements. The window step size is determined based on the update frequency requirements. For each valid node pair in the traffic transition probability matrix, the transition probability values ​​corresponding to each time segment within the coverage of the sliding time window are read sequentially, forming the time series data of that node pair within that sliding time window. After the sliding time window moves one step along the time axis, the same extraction process is repeated to obtain the transition probability time series of that node pair under different sliding time windows. If the number of valid observations within a certain sliding time window is lower than a preset lower limit, the corresponding fluctuation results are not output, or the neighboring window completion method is used for processing and a completion mark is added.

[0050] The statistical standard deviation is used to characterize the dispersion of the transition probability of the same node pair relative to the window mean within the corresponding sliding time window. After calculating the statistical standard deviation for all node pairs, a fluctuation feature matrix corresponding to the dimension of the flow transition probability matrix is ​​obtained, or a dynamic fluctuation feature vector is formed by combining the node pairs in order. The larger the statistical standard deviation, the more obvious the change in the flow transition intensity of the node pair within the corresponding sliding time window; the smaller the standard deviation, the smoother the change in the flow transition intensity of the node pair within the corresponding sliding time window. In the scenario of mountainous road networks, in order to avoid the distortion of fluctuation features caused by sporadic observations, the validity screening of the transition probability sequence within the sliding time window is also performed before calculating the statistical standard deviation. When a node pair has only one isolated non-zero transition probability within the entire sliding time window, while the probability is zero at all other times, the corresponding statistical standard deviation is not directly output. Instead, it is judged whether the minimum number of valid observations for the node pair within the preset observation period has reached the lower limit; only when the lower limit is reached is it recorded as a valid fluctuation feature. When there is a known sampling delay in satellite imagery data or ground sensor data, time shift correction can be performed on the corresponding time series based on the initial time delay distribution, and then the statistical standard deviation can be calculated to align the source node state changes with the target node flow response in time.

[0051] S2. Perform time-series trend analysis on the historical data sequence of traffic transfer paths, and combine the dynamic fluctuation characteristics of transfer to predict the scope of transfer impact caused by changes in road network status.

[0052] In one specific embodiment, the process of performing step S2 may specifically include the following steps:

[0053] Obtain the historical data sequence corresponding to the traffic transfer path;

[0054] The time-series trend test method is used to perform trend analysis on historical data sequences, and combined with the dynamic fluctuation characteristics of the transfer, the trend direction and trend significance of the flow transfer changes in the historical data sequences are determined;

[0055] Based on the trend direction and trend significance, potential affected road nodes or road segments associated with traffic transfer paths are identified, and the scope of the transfer impact caused by changes in road network status is predicted accordingly.

[0056] Specifically, the historical data sequence corresponding to the traffic flow transfer path is extracted using a single traffic flow transfer path as the organizational unit. Each traffic flow transfer path is uniquely identified by the starting road node, the target road node, and the sequence of road connecting edges linking the two. For each traffic flow transfer path, continuous observation records within a preset historical period are extracted from the mountainous road network dynamic dataset and the traffic flow transfer probability matrix sequence, and sorted according to a uniform time granularity to form the corresponding historical data sequence. The uniform time granularity is consistent with the time granularity used when constructing the preceding traffic flow transfer probability matrix. The historical data sequence includes at least the transfer probability value of the traffic flow transfer path in each time segment, the traffic status data of the starting road node or starting road segment, the traffic status data of the target road node or target road segment, event markers, and the transfer dynamic fluctuation characteristic value of the corresponding time segment. Traffic status data includes at least one of average vehicle speed, occupancy rate, queuing status, or traffic status markers. If there are missing observations in a certain time segment, they are filled in using adjacent valid observations, or the missing markers are retained but the time is not included in the trend significance calculation. The historical data sequence can continuously reflect the change process of the transfer intensity of the same traffic flow transfer path within a historical period.

[0057] The preferred time-series trend testing method is a non-parametric trend testing method to adapt to non-normal and non-stationary time-series data in mountainous road networks caused by holiday traffic fluctuations, severe weather, accident control, or partial traffic closures. For example, the Mann-Kendall trend test is used to determine trend significance, and Sen slope estimation is used to determine the trend direction and magnitude of change. For each traffic transfer path, the transfer probability sequence from its historical data series is extracted as the main test sequence, and a pre-defined trend testing method (such as the Mann-Kendall trend test) is used to perform a trend test on the main test sequence. When the test result shows a positive change and the corresponding significance probability is not greater than the pre-defined significance level threshold, the traffic transfer path is recorded as an upward trend path; when the test result shows a negative change and the corresponding significance probability is not greater than the pre-defined significance level threshold, the traffic transfer path is recorded as a downward trend path; when the corresponding significance probability is greater than the pre-defined significance level threshold, the traffic transfer path is recorded as a path with no significant trend.

[0058] The system reads the dynamic fluctuation characteristics of the traffic transfer path within the same time range. If the traffic transfer path is an upward trend path and its dynamic fluctuation characteristics are within a preset stable range, it is marked as a continuously expanding path. If the traffic transfer path is an upward trend path but its dynamic fluctuation characteristics exceed a preset fluctuation threshold, it is marked as a volatile expanding path. If the traffic transfer path is a downward trend path, it is marked as a weakening path. For example, if the transfer probability of a traffic transfer path continuously increases over multiple consecutive time segments, and the corresponding standard deviation remains in the low to medium range, the path is considered to have a stable expansion trend. If the transfer probability increases but the corresponding standard deviation remains high, the path is considered a fluctuating expansion path. Based on this, the trend direction, trend significance, and fluctuation degree are unified into the same judgment chain, avoiding the misjudgment of short-term shocks as stable diffusion based solely on the trend direction.

[0059] When predicting the scope of the impact of traffic transfer, the trend test results are not directly used as the spatial scope of impact. Instead, a topological association search is performed on the road network connectivity model, starting with traffic transfer paths exhibiting a significant upward trend. For each continuously expanding or volatile expanding path, its starting road node, target road node, and path edge sequence are read, and the road segment containing the target road node or the target road connecting edge is determined as the initial impact object. Using the initial impact object as the core, an association search is performed in the road network connectivity model along the downstream propagation direction and adjacent substitution direction to obtain a set of candidate affected road nodes and a set of candidate affected road segments. To prevent unconstrained expansion of the impact scope, a search depth is set for the topological search. The search depth can be dynamically set according to the transfer probability intensity or trend significance level of the corresponding traffic transfer path. For example, a larger search depth is used for traffic transfer paths with higher transfer intensity or stronger trend significance, while a smaller search depth is used for traffic transfer paths with lower transfer intensity or weaker trend significance. For candidate affected road nodes or segments, the following criteria are used for screening: whether there is a continuous topological reachability relationship between them and the initial affected object; whether there is a decrease in average vehicle speed, an increase in occupancy, or a deterioration in traffic conditions within the corresponding time window; and whether there is a continuously non-zero transfer probability association between them and the traffic transfer path. Only when the preset screening criteria are met is the road node or segment identified as a potential affected object. If there is a parallel path with an alternative travel relationship to the target road and the travel time increment is within a preset allowable range, the relevant road nodes or segments on the parallel path are included in the candidate set and screened according to the same rules.

[0060] For all traffic transfer paths showing a significant upward trend, the aforementioned identification of potentially affected objects is performed. The sets of potentially affected road nodes and road segments corresponding to each path are then merged and deduplicated to form the output result of the transfer impact range. The transfer impact range is preferably output in the form of sets of affected road node identifiers and sets of affected road segment identifiers, and an index is established to correlate these with the trend direction, trend significance, and fluctuation classification results of the corresponding traffic transfer path. When spatial representation is required, the sets of affected road nodes and affected road segments can be projected onto a mountain road network base map to form a spatial mapping result of the impact range.

[0061] S3. When the affected area exceeds the preset threshold, the affected road subset is selected based on the road network topology, and a traffic transfer spatial distribution map is generated based on the traffic transfer density index corresponding to the road subset.

[0062] In one specific embodiment, the process of performing step S3 may specifically include the following steps:

[0063] Determine whether the affected area exceeds a preset threshold. If so, perform a topology search within the affected area based on the road network topology represented by the road network connection model to select the affected road subset.

[0064] Based on the traffic transfer characteristics, the transfer probability between each node within the road subset is extracted, and a local traffic transfer probability matrix corresponding to the road subset is constructed.

[0065] Based on the local traffic flow transfer probability matrix and the current traffic flow of the road subset, the transfer flow of each road segment within the road subset is determined, and the transfer flow density index corresponding to each road segment is calculated by combining the capacity coefficient of each road segment.

[0066] The topological spatial location of a subset of roads is spatially mapped to the traffic flow density index to generate a traffic flow spatial distribution map.

[0067] Specifically, the preset range threshold is used to determine whether the set of affected objects obtained in the previous step has reached the level requiring local fine-grained analysis. It can be characterized by at least one of the following: the number of affected road nodes, the number of affected road segments, the total length of affected road segments, or the coverage radius of the affected area. Preferably, a combined threshold of the number of affected road nodes and the total length of affected road segments is used for judgment. When either indicator exceeds the corresponding threshold, it is determined that the transferred influence range exceeds the preset range threshold. This avoids the subsequent local subnet analysis being directly triggered by isolated nodes, short-term fluctuations, or local anomalies.

[0068] When the scope of the transfer exceeds a preset threshold, a subset of affected roads is selected. A graph search is performed starting from the core road nodes or core road connecting edges within the transfer scope. Core road nodes are preferably target road nodes corresponding to significant upward trend paths in trend analysis, or highly correlated nodes intersecting with multiple significant upward trend paths. Breadth-first search or hierarchical adjacency search is performed along the adjacency relationships in the road network connection model, visiting one-hop adjacency nodes, two-hop adjacency nodes, and other adjacency nodes within the preset level layer by layer, to obtain an initial set of candidate nodes and an initial set of candidate road connecting edges. To avoid unconstrained expansion of the search scope, for each visited node, the correlation degree between the node and the core road nodes needs to be calculated; the correlation degree is determined by the topological distance and the path distance calculated along the actual travel direction of the road network, representing the degree of spatial separation between the two; the topological distance is the minimum number of road connecting edges traversed between the core road node and the candidate node, and the path distance is the cumulative road length along the actual traversable paths of the road network. Due to the presence of tunnels, winding mountain roads, and terrain barriers in mountainous road networks, Euclidean distance is not used to directly replace the actual path distance. Nodes with a correlation value not higher than the preset correlation threshold and their connected road edges are retained as a subset of affected roads. The smaller the correlation value, the stronger the correlation between the candidate node and the core road node.

[0069] If the probability of local migration is high or the trend is significant, a larger search level is used for its corresponding core node; if the probability of local migration is low, the search level is reduced, and only the core node and its directly connected road edges are retained.

[0070] Extract the node pair transfer probability records where both the starting node and the target node are located within the road subset from the global flow transfer probability matrix sequence obtained in the previous steps. If there are node pairs within the road subset that are not significantly represented individually in the global matrix but have continuous transfer observations within a local area, then re-count the number of local flow transfer events by combining the traffic status sequence of the road subset in the current analysis period. For each source node in the road subset, count the number of effective flow transfer events from it to each target node within the current statistical time window, and divide the number of flow transfer events corresponding to a target node by the total number of outflow transfer events of the source node to obtain the local transfer probability from the source node to the corresponding target node. Then, arrange them according to the target node order to form the local probability row vector of the source node, and combine them according to the source node order to obtain the local flow transfer probability matrix corresponding to the road subset. If a source node does not have any effective outflow transfer events within the current statistical time window, then set the corresponding row of the source node to zero, or inherit the local probability value of the previous statistical time window and add a missing flag. For transition relationships pointing from the boundaries of a road subset to nodes outside the subset, since these external nodes are not within the current local analysis scope, they are not included as local target nodes in the local matrix. If necessary, the probability terms retained within the subset can be renormalized to ensure that the probabilities of corresponding rows for the same source node in the local matrix meet the local analysis requirements. The local flow transition probability matrix can more accurately represent the local relationships of traffic flow redistribution from one node to adjacent nodes within the affected road subset.

[0071] The actual traffic flow of each source node or starting road connection edge within the road subset is read in the current time segment. The actual traffic flow is obtained by dividing the number of vehicles by the statistical time. The current traffic flow of the source node or starting road connection edge is multiplied by the local transfer probability pointing to each target node or target road connection edge to obtain the transfer flow between the corresponding node pairs. Based on the local shortest feasible path or the identified traffic transfer path between each node pair, the transfer flow of the node pair is distributed to each road connection edge that makes up the path to obtain the transfer flow carried by each road segment in the current statistical time segment. The transfer flow represents the actual traffic flow allocation applied to each road connection edge due to local traffic redistribution under the current traffic conditions, rather than simply the event frequency. For each road connection edge in the road subset, the corresponding capacity coefficient is determined according to its number of lanes, road grade, design speed, or saturation capacity. The capacity coefficient preferably represents the theoretical capacity of the road connection edge per unit time. Dividing the transfer flow rate of a road connector by its capacity coefficient yields the transfer flow rate density index (a dimensionless quantity), representing the proportion of transfer flow rate to the available capacity of that connector. A higher transfer flow rate density index indicates a stronger impact of local traffic transfer on the connector. If considering the statistical duration, the cumulative transfer flow rate within that duration can be calculated first, then divided by the product of the capacity coefficient and the statistical duration; the result is also dimensionless. Even with similar absolute transfer flow rates, roads of different grades can still reflect different degrees of local pressure through differences in capacity.

[0072] The system reads the latitude and longitude coordinates of each road node and the geometric shape of each road connecting edge within the road subset. It projects the topology of the road subset onto a mountain road network base map and maps the transfer flow density index corresponding to each road connecting edge to a visual attribute on the road layer. For example, road connecting edges can be rendered in hierarchical levels according to the magnitude of the transfer flow density index, or the transfer flow density index can be continuously mapped to color depth and line width. For node pairs in the local flow transfer probability matrix that are non-zero and exceed a preset display threshold, directed line segments or arrows are overlaid to display their transfer direction and transfer probability intensity. The arrow direction corresponds to the flow transfer direction, and the arrow thickness corresponds to the local transfer probability magnitude. If necessary, the average vehicle speed reduction rate, occupancy increase rate, or congestion frequency label of key nodes can also be overlaid on the spatial distribution map. The spatial distribution map of flow transfer can characterize where traffic flow within the affected road subset is transferred from and to, which road connecting edges bear higher migration loads, and where local diffusion hotspots are located.

[0073] S4. Update the weight parameters of the road network connection relationship model based on the spatial distribution map of traffic transfer, and perform traffic offset simulation to determine the adjusted road network connection status characteristics.

[0074] In one specific embodiment, the process of performing step S4 may specifically include the following steps:

[0075] Based on the spatial distribution map of traffic transfer, the incremental traffic resistance of each road segment in the road subset and the congestion frequency of each node are determined.

[0076] By combining the increase in traffic resistance and the frequency of congestion, the weight parameters of the corresponding node connection edges in the road network connectivity model are updated;

[0077] Traffic offset simulation is performed based on the updated road network connectivity model to obtain the traffic distribution deviation of the road network under traffic transfer conditions;

[0078] Based on the traffic distribution deviation, the adjusted road network connectivity characteristics are determined.

[0079] Specifically, the road network connectivity model adopts a weighted graph structure, including road nodes and road connection edges. The edge weight parameter is used to characterize the passage cost of the corresponding road connection edge. This step is used to address the problem that in mountainous road networks, the original road network connectivity model still uses static edge weights after local traffic shifts occur, making it difficult to reflect the increased local traffic resistance, node queue propagation, and secondary traffic flow shifts.

[0080] The traffic flow transfer spatial distribution map includes the transfer traffic flow density index, traffic flow transfer direction, and local congestion status of key nodes for each road connection edge within the road subset, serving as input for calculating resistance increment and congestion frequency. For any road connection edge in the road subset, the basic traffic resistance value, current average vehicle speed, and corresponding transfer traffic flow density index of that road connection edge are read. The basic traffic resistance value is preferably the travel time of that road connection edge under the baseline state, which can be the historical average state during periods without significant traffic flow transfer. The current travel time is calculated based on the current average vehicle speed, and the state penalty time is determined in conjunction with the transfer traffic flow density index of that road connection edge. The state penalty time represents the additional delay caused by local traffic flow migration pressure. Preferably, the state penalty time is determined according to the segmented intervals of the transfer traffic flow density index: when the transfer traffic flow density index is below a first preset threshold, the state penalty time is zero; when the transfer traffic flow density index is between the first and second preset thresholds, the state penalty time increases linearly with the increase of the transfer traffic flow density index; when the transfer traffic flow density index is above the second preset threshold, the state penalty time is a preset upper limit value. In this way, the state penalty time maintains a monotonic correspondence with the local traffic migration pressure, and the penalty time is prevented from increasing indefinitely. The current travel time is added to the state penalty time to obtain the current travel resistance value; the difference between the current travel resistance value and the basic travel resistance value is the travel resistance increment of the road connection edge, which represents the increased travel cost under local traffic transfer conditions. The node congestion frequency is calculated by taking each road node in the road subset as the statistical object, counting the number of congestion events occurring at that node within a preset statistical time window, and dividing by the total number of observations within that statistical time window. A congestion event includes at least one of the following: queue overflow at the node's entrance lane, the node's average vehicle speed being lower than a preset threshold, the node's occupancy rate being higher than a preset threshold, or the critical road connection edge connected to the node being in an interrupted state. Preferably, when multiple consecutive sampling times meet the congestion event conditions, the consecutive time periods are combined and recorded as a single congestion event to avoid repeated counting of the same persistent congestion state. The travel resistance increment characterizes the degree of traffic deterioration at the road connection edge level, and the node congestion frequency characterizes the persistence of the bottleneck at the node level.

[0081] The weight parameters of the corresponding node connection edges in the road network connectivity model are updated. The updated weight parameters describe the passage cost of the road connection edge under the current conditions. Preferably, the weight parameters are defined as time-based edge weights. For any road connection edge in the road network connectivity model, the original edge weight is read, followed by its corresponding traffic resistance increment, and the congestion frequencies of the starting and ending nodes of the road connection edge. The traffic resistance increment and node congestion frequencies are mapped to edge weight corrections and superimposed with the original edge weights to obtain the updated edge weight parameters. When writing the node congestion frequency into the update formula, it is first multiplied by a congestion penalty time coefficient to ensure consistency of the dimensions of the components involved in the superposition. The congestion penalty time coefficient is preferably pre-calibrated based on at least one of the node's historical average queuing dissipation time, signal control cycle length, or average node control delay, and is used to convert the node congestion frequency into the corresponding time-based penalty. For example, the average congestion frequencies of the starting and ending nodes can be multiplied by the congestion penalty time coefficient to obtain the additional time penalty for node congestion on that edge. This additional time penalty is added together with the traffic resistance increment to the original edge weight to obtain the updated edge weight parameters. Thus, the road network connectivity model can reflect the changes in the expected toll costs of different road connection edges under local traffic transfer conditions.

[0082] Traffic flow offset simulation is used to characterize the redistribution of traffic flow in road subsets and their associated road networks after edge weight updates. The traffic flow state of the current time segment is used as the simulation input, which includes at least the current traffic flow of each starting road node or starting road connecting edge, the local traffic flow transition probability matrix, and the updated edge weight parameters. Then, path selection and traffic flow allocation calculations are performed on the updated road network connection model. Preferably, a traffic flow offset simulation method combining weighted shortest path and local probability allocation is adopted: for each source node, several feasible paths are first determined based on the updated edge weights, and then the current traffic flow of the source node is allocated to feasible paths based on the local traffic flow transition probability matrix. This process is then accumulated segment by segment to obtain the predicted traffic flow of each road connecting edge under simulation conditions. Feasible paths are preferably determined using the k-shortest path algorithm under updated edge weight constraints, or by a set of candidate paths that satisfy preset path length and edge weight constraints. The number of candidate paths is preferably a preset finite value to avoid including obviously unreasonable detour paths in the traffic flow allocation process. To obtain the deviation, a control simulation was also set up. The control simulation used the same source node traffic flow input and the same local flow transfer probability matrix as the previous simulation, but used the original edge weight parameters before the update. For each road connection edge in the road subset, the predicted flow obtained from the updated simulation and the reference flow obtained from the control simulation were read respectively. The difference between the two was taken as the edge-level flow deviation of that road connection edge; if necessary, this difference could also be divided by the reference flow to obtain the relative flow deviation. Based on this, the redistribution effect caused by the edge weight update is separated from the original flow background, thus more clearly characterizing the degree of local flow shift. Here, the "deviation" is not measured by simple spatial distance, but by edge-by-edge flow difference and its normalization result, avoiding the significant changes in individual key narrow road sections in mountainous road networks being masked by the overall scale.

[0083] The adjusted road network connectivity characteristics refer to a set of node and edge attributes used to characterize the current changes in the local road network connectivity, after considering the increase in traffic resistance, node congestion frequency, updated edge weight parameters, and simulated traffic redistribution results. First, the edge-by-edge traffic deviation, updated edge weight parameters, and node congestion frequency are jointly processed. Then, a corresponding edge status record is generated for each road connection edge. The record includes at least the road connection edge identifier, updated edge weight, simulated traffic, reference traffic, edge-level traffic deviation, and the corresponding traffic status level. The traffic status level corresponding to the edge can be determined based on the ratio of simulated traffic to capacity or based on the comparison between the updated edge weight and a preset threshold. For road nodes, a corresponding node status record is generated. The record includes at least the node identifier, node congestion frequency, the average weight of the edge associated with the node, and the average traffic deviation of the edges connected to that node. All edge state records and node state records can be organized into an adjusted road network connection state feature dataset, or represented as a weighted graph structure with node attributes and edge attributes; edge attributes include at least the updated toll cost and edge-level traffic deviation, and node attributes include at least the congestion frequency and local congestion status marker.

[0084] S5. Perform geospatial mapping on the road network connectivity characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution.

[0085] In one specific embodiment, the process of performing step S5 may specifically include the following steps:

[0086] By combining the geographical coordinates of the corresponding road segments with the preset regional influence weights, the road network connection status characteristics are geospatially mapped.

[0087] Traffic flow quantitative indicators are extracted based on spatial mapping results. These indicators include at least average travel time and path delay increment.

[0088] Based on the spatial mapping results, a visual distribution map reflecting the dynamic evolution of road network traffic is generated.

[0089] Specifically, the latitude and longitude coordinates and geometric shapes of each road node and each road connecting edge established in S1 are read, and the node attributes and edge attributes are projected onto the mountain road network base map, so that each road connecting edge corresponds to the actual road segment location. Regional influence weights characterize the importance of different regions in the overall traffic operation and should not be statically set based solely on administrative divisions, but should be determined based on road network functions and traffic demand distribution. Preferably, the mountain road network is divided into multiple grid units or road influence zones according to the geographic information system. Then, for each grid unit, its historical daily average traffic flow, road grade weighted value, and traffic attraction source quantitative indicators are statistically analyzed. The traffic attraction source quantitative indicators are determined based on the quantity, distribution density, or traffic attraction intensity of at least one of scenic spots, town centers, toll stations, emergency lane nodes, and freight distribution points. Subsequently, the historical daily average traffic flow, road grade weighted value, and traffic attraction source quantitative indicators are normalized respectively, and regional influence weights are determined based on the normalization results. Then, according to the grid unit to which the road node or road connecting edge belongs, the regional influence weights are associated with the corresponding node or edge. Road nodes connecting the main entrance of a scenic area, the confluence of main roads, or emergency access routes have a higher regional impact weight than ordinary mountain branch road nodes.

[0090] The spatial mapping result is obtained by fusing regional influence weights with geographic coordinates and road network connectivity features. For each road connection edge in the connectivity features, its updated edge weight, simulated flow, edge-level flow deviation, and traffic status level are read, and the corresponding regional influence weights are read from the nodes at both ends of the edge. To avoid directly and mechanically superimposing node weights with edge status, it is preferable to first construct a representative edge-level status value and then fuse it with the regional influence weights. The updated edge weight and edge-level flow deviation can be combined into a representative edge-level status value according to a preset ratio. The updated edge weight represents the traffic cost, and the edge-level flow deviation represents the degree of flow deviation. The average of the regional influence weights of the nodes at both ends of the edge is used as the edge-level regional influence weight. The edge-level representative status value and the edge-level regional influence weight are then fused by product or weighted summation to obtain the spatial mapping value corresponding to the road connection edge. In this way, the spatial mapping value can simultaneously reflect the current connectivity status change of the road segment and the importance of the area where the road segment is located. If a product method is used, road segments with higher regional influence weights will have higher spatial response values ​​under the same state change conditions; if a weighted summation method is used, the excessive amplification of local states by extreme weights can be avoided. Either method can be chosen in the same embodiment to maintain consistency in the calculation method.

[0091] Based on the spatial mapping results, a set of preset key origin-destination pairs are selected. These pairs preferably cover major entrance / exit nodes, scenic area nodes, township center nodes, toll station nodes, and emergency lane nodes in the mountainous road network, ensuring that quantitative indicators reflect the overall road network traffic characteristics. The spatial mapping results are primarily used for spatial filtering, importance identification, and result display of the key origin-destination pairs. For calculating the average travel time, the adjusted road network connectivity model is preferred, using the updated edge weights as path search weights to calculate the theoretical travel time for each origin-destination pair. The theoretical travel time can be determined by summing the updated edge weights of all road connecting edges on the optimal path. The average travel time of all origin-destination pairs is then calculated, representing the absolute traffic efficiency under the current prediction state.

[0092] For path delay increments, it is preferable to use the same set of origin-destination pairs as the calculation object. The baseline travel time for this set of origin-destination pairs is calculated under the baseline model; the baseline model uses the road network connectivity model corresponding to the period without significant traffic shifts, or the road network connectivity model before the update. Then, the adjusted travel time is calculated using the updated edge weights of the corresponding paths in the adjusted road network connectivity model, and this adjusted travel time is subtracted from the baseline travel time one by one to obtain the path delay change for each origin-destination pair. The average of all path delay changes is then calculated to obtain the path delay increment, which represents the relative degree of change compared to the baseline state. A positive path delay increment indicates an increase in overall travel delay under the adjusted state; a negative path delay increment indicates a decrease in overall delay. Spatial mapping results are used for spatial representation and key origin-destination pair selection. The average travel time and path delay increment are preferably calculated based on the updated edge weights to ensure that the traffic physical meaning of the travel time index is not directly affected by regional influence weights.

[0093] The visualization distribution map in S5 is used to represent the overall evolution of the road network connectivity status in geospatial space after the weight update of the road network connectivity model and the simulation of traffic offset. Each road connection edge in the spatial mapping result is rendered hierarchically according to its spatial mapping value; this can be achieved using varying color depths, line widths, or a combination of both. Color represents the spatial mapping value or traffic status level, while line width represents the simulated traffic flow or edge-level traffic deviation. The map overlays statistical legends corresponding to key node locations, regional influence weight levels, average travel time, and path delay increments. If a dynamic evolution process needs to be represented, the spatial mapping results can be drawn frame-by-frame according to continuous time segments, forming a time-series distribution layer or a multi-frame dynamic image, allowing continuous display of the status changes of the same road segment within different time segments. In this way, both high-pressure road segments in local areas and the overall road network evolution process over time can be observed.

[0094] S6. Determine whether the traffic flow quantitative indicators meet the preset congestion risk conditions. If so, output the traffic flow prediction results.

[0095] In one specific embodiment, the process of performing step S6 may specifically include the following steps:

[0096] Based on traffic flow quantification indicators, calculate the comprehensive congestion risk index and traffic flow saturation of the corresponding road segment;

[0097] Determine whether the comprehensive congestion risk index and / or traffic flow saturation exceed the corresponding preset congestion risk threshold. If so, based on the spatial mapping results and road network connection status characteristics, predict the expected traffic flow and congestion spread trend of the corresponding road segment in the future preset time period.

[0098] The expected traffic volume and congestion spread trend will be used as the result of traffic flow forecasting.

[0099] Specifically, the key road connecting edges in the selected spatial mapping results for the corresponding road segment are characterized by high spatial mapping values, large edge-level flow deviations, high updated edge weights, or large path delay increments. The comprehensive congestion risk index represents the overall risk level of congestion or further deterioration of the corresponding road segment under the current predicted state, while traffic flow saturation represents the degree to which traffic demand occupies the available capacity of the corresponding road segment. For each corresponding road segment, at least several of the following are read: path delay increment, spatial mapping value, edge-level flow deviation, current travel time corresponding to the updated edge weight, node congestion frequency, and traffic status level. Each indicator is normalized separately, and then weighted and summed according to preset weights to obtain the comprehensive congestion risk index for the corresponding road segment. The preset weights are set according to the degree of influence of different factors on congestion formation in mountainous road network scenarios. If a road segment is located on a main access road to a scenic area, a township entrance or exit, or an emergency access road, the weight of the spatial mapping value and / or node congestion frequency in the comprehensive congestion risk index can be appropriately increased.

[0100] Traffic flow saturation is determined by the ratio of the current simulated or predicted traffic flow to the capacity of the corresponding road segment. Preferably, the traffic flow saturation of the corresponding road segment is obtained by dividing the simulated traffic flow of the corresponding road segment at the current moment by the capacity of the road segment. If it is necessary to consider the short-term trend in the future, the combined value of the current simulated traffic flow and the average expected traffic flow over a preset period in the future can also be used as the numerator, but the numerator should be kept consistent in the same embodiment.

[0101] A preset congestion risk threshold is used to define the conditions under which future time period predictions should be output. A risk threshold is set for the comprehensive congestion risk index, and a saturation threshold is set for traffic flow saturation. When either indicator exceeds its corresponding threshold, the corresponding road segment is determined to meet the preset congestion risk conditions. Preferably, a tiered determination method is adopted: when the comprehensive congestion risk index exceeds the risk threshold but the traffic flow saturation does not exceed the saturation threshold, the road segment is recorded as a risk warning state; when both the comprehensive congestion risk index and traffic flow saturation exceed their respective thresholds, the road segment is recorded as a high-risk state; when only the traffic flow saturation exceeds the threshold but the comprehensive congestion risk index does not exceed the threshold, it is recorded as a high-load state.

[0102] The preferred future time period is a short-term prediction window with the same granularity as the preceding statistical time. The simulated traffic flow, updated edge weights, edge-level traffic deviation, spatial mapping value, regional influence weight, and the status of upstream, downstream, and alternative path nodes directly connected to the corresponding road segment at the current time are read. Then, the transfer probability records associated with the corresponding road segment are read from the local traffic transfer probability matrix. Combined with the updated road network connectivity model, the traffic migration within the future time period is recursively calculated. Preferably, a rolling prediction method based on the current traffic status, local transfer probability, and edge weight constraints is adopted: In the first future time segment, the current simulated traffic flow is used as the initial traffic flow, and the predicted traffic flow into and out of the road segment is calculated based on the upstream inflow probability and downstream diversion probability associated with the road segment; in the second and subsequent future time segments, the expected traffic flow of the previous time segment is used as input, and the same calculation is repeated. In this way, the expected traffic flow of each corresponding road segment within the future time period can be obtained segment by segment. For example, the projected traffic flow for a given road segment in a future time segment can be updated using a conservation approach. This involves adding the upstream inflow forecast to the previous time segment's projected flow and subtracting the downstream outflow forecast. The upstream inflow forecast is determined based on the previous time segment's projected flow and corresponding local transfer probabilities of the upstream road connection edges linked to the road segment. The downstream outflow forecast is determined based on the local transfer probabilities of the road segment pointing to downstream nodes or downstream road segments and the feasible path allocation results under the updated edge weight constraints. If it is necessary to reflect the impact of the current local flow offset trend on future flow, edge-level flow deviation can be written as a correction term into the projected flow, but this should remain within the aforementioned flow conservation framework and should not be amplified independently outside of network flow balance.

[0103] Instead of directly outputting abstract probability labels, the congestion spread trend is determined by combining the road network topology and the state evolution of adjacent road segments. The corresponding road segment meeting the preset congestion risk conditions is designated as the core congestion road segment. The expected traffic flow, traffic saturation, updated edge weights, and spatial mapping values ​​of its upstream and downstream adjacent road segments, as well as parallel road segments with alternative traffic relationships, are read over a preset time period. The system then determines whether, in consecutive future time segments, there is a continuous increase in the expected traffic flow, traffic saturation, updated edge weights, or spatial mapping values ​​of adjacent road segments. If these conditions occur consecutively in multiple adjacent road segments along the main propagation direction, the congestion spread trend is determined to be one or more of the following: downstream expansion, upstream spillover, or diffusion to alternative paths. If these conditions are limited to the core road segment and adjacent road segments do not show continuous deterioration, the congestion spread trend is determined to be locally sustained. If the expected traffic flow and traffic saturation of the core road segment and adjacent road segments decrease synchronously, the congestion spread trend is determined to be weakening. The propagation direction here is consistent with the direction of the flow transfer path and the direction of the spatial distribution map formed in S2 and S3.

[0104] To avoid misjudging short-term isolated fluctuations as congestion spread, a continuity constraint can be added when assessing spread trends. Specifically, adjacent road segments are only considered as spreading targets if they consistently meet the conditions for increased expected traffic volume or elevated traffic saturation over several consecutive time segments. If an abnormal increase occurs only in a single time segment, it is retained as a short-term fluctuation and not directly included in the congestion spread trend. If the corresponding road segment is located within a grid cell with a high regional influence weight, the threshold for adjacent road segments to enter the spread assessment can be appropriately lowered to reflect the propagation sensitivity of main corridors and key areas; however, this adjustment remains within the preset rules.

[0105] Traffic flow forecast results are preferably output by corresponding road segments. Each record should include at least the road segment identifier, the expected traffic flow for each time segment within the preset future time period, the current traffic flow saturation, the comprehensive congestion risk index, and the corresponding congestion spread trend label. If necessary, the regional influence weight of the road segment, the start and end node identifiers, and the set of adjacent road segments involved in the spread can also be added.

[0106] In a preferred embodiment, when the traffic flow quantification index meets the preset congestion risk conditions, S6 also generates traffic diversion suggestion information, including the following steps:

[0107] The spatiotemporal correlation coefficient is used to determine the correlation influence weight of each node and trigger an adaptive adjustment mechanism.

[0108] The traffic light sequence of each node is reconfigured according to the associated influence weights;

[0109] The overall traffic distribution is simulated based on the configuration results, and traffic management suggestions are generated based on the simulation results.

[0110] Specifically, based on the identified traffic transfer paths and local traffic transfer probability matrices, spatial location pairs to be analyzed are determined; these pairs are preferably combinations of upstream and downstream nodes with significant traffic transfer probabilities. For each spatial location pair, the time series of traffic flow saturation for the road segments associated with the upstream node and the time series of changes in the comprehensive congestion risk index for the road segments or nodes associated with the downstream node are extracted and aligned on a unified time axis. The correlation coefficient between the two time series is calculated under multiple preset time delays, and the correlation coefficient with the largest absolute value and satisfying the forward propagation logic is selected as the spatiotemporal correlation coefficient of the spatial location pair. The corresponding time delay is recorded as the propagation lag of the node pair. The spatiotemporal correlation coefficient characterizes the propagation intensity of upstream load changes on downstream risk changes, and the propagation lag characterizes the time lag of this influence. Preferably, within a fixed statistical time window, a series of several consecutive time segments can be used as the calculation object, and the correlation coefficient can be calculated iteratively for multiple delay values, retaining the maximum correlation result and its corresponding delay.

[0111] After obtaining the spatiotemporal correlation coefficients of each spatial location pair, the association influence weight of each node is further determined. For any node, the spatiotemporal correlation coefficients of all spatial location pairs with that node as an upstream or downstream node are summarized. Combined with the node's regional influence weight in the spatial mapping results, the node's congestion frequency, and the edge-level flow deviation of its associated edges, the association influence weight of that node is calculated, characterizing the degree of influence of that node on the spread of congestion and traffic control adjustments in the surrounding area. The spatiotemporal correlation coefficients, regional influence weights, node congestion frequency, and edge-level flow deviations can be normalized first, and then combined according to preset weights to obtain the node-level association influence weights.

[0112] The adaptive adjustment mechanism is triggered based on the correlation influence weight and preset triggering rules. Preferably, a dual-threshold triggering method is adopted, that is, when the proportion of node pairs with high spatiotemporal correlation coefficients to all analyzed node pairs exceeds a preset proportion threshold, and the correlation influence weight of the corresponding node reaches the node adjustment threshold, the adaptive adjustment mechanism is activated.

[0113] After triggering the adaptive adjustment mechanism, the traffic light sequences corresponding to road nodes with signal control capabilities and identified as highly correlated influence nodes are reconfigured. The signal control range requiring reconfiguration is determined; this range consists of signal control nodes associated with nodes having high spatiotemporal correlation coefficients, their directly connected signal control intersections, and adjacent signal control nodes with alternative passage relationships. For each signal control node within the range, its current signal light control parameters, as well as the corresponding entry lane's transfer flow density index, node congestion frequency, current queue length, current traffic flow saturation, and relevant records in the local flow transfer probability matrix are read. The traffic light sequence includes at least the signal cycle length, the green light duration allocation ratio for each phase, the phase switching order, and the phase clearing time. Based on the node's correlation influence weight, the signal cycle and the green light duration for each phase are reconfigured; for entry directions with high flow transfer probability, high transfer flow density, high current traffic flow saturation, and high correlation influence weight, their green light duration ratio is increased; for directions with lower flow demand and weaker propagation influence, their green light duration ratio is correspondingly reduced. If a control node simultaneously performs the functions of main channel traffic release and alternative route diversion, it is preferable to prioritize ensuring the release capacity of the main propagation direction and the high-risk spillover direction, and then consider the alternative route direction. To avoid signal timing imbalance, it is preferable to keep the total duration of a single cycle constant, or adjust the cycle length only within a preset fluctuation range; at the same time, the minimum green light duration and clearing time of each phase should not be lower than the traffic control safety requirements.

[0114] For example, the control priority for each inbound direction is first calculated for each control node. The control priority is determined based on the associated impact weight, transfer flow density index, node congestion frequency, and current traffic flow saturation for that inbound direction. The green light duration is then redistributed according to the control priority. If an inbound direction has a higher control priority, its green light duration is increased while still meeting the minimum clearance requirements; if an inbound direction has a lower priority, its green light duration is shortened. If the propagation delay between a node and a downstream high-risk node is short, the clearance priority for that node's direction towards the downstream high-risk node is increased to reduce upstream vehicle accumulation and decrease congestion spillover intensity.

[0115] The reconfigured signal control parameters are input into the road network connectivity model updated via S4. The current traffic flow state is loaded into the model as initial conditions, and overall traffic distribution simulation is performed over several consecutive statistical time segments. The simulation process is preferably based on a graph-structured traffic flow allocation mechanism: for any starting node, based on the updated signal light sequence, local traffic transition probability matrix, and edge weight constraints, the waiting time, passage time, and traffic flow allocated to downstream nodes are calculated progressively at each node. By recursively allocating traffic flow to all starting nodes, the simulated traffic distribution results of the entire affected area and its associated road network under reconfiguration control conditions are obtained. To facilitate comparison of the degree of improvement before and after control, a control simulation can be set up; the control simulation uses the original signal light sequence, while keeping other input conditions consistent. By comparing the changes in average travel time, path delay increment, comprehensive congestion risk index of key nodes, expected traffic flow on key road sections, queue length, and number of congested road sections in the simulation results before and after adjustment, it can be determined whether the new signal timing scheme has actually improved the local road network operation.

[0116] If the simulation results show a decrease in overall average travel time, a reduction in path delay increment, a suppression of the growth trend of the comprehensive congestion risk index at key nodes, a decrease in the expected traffic flow and traffic saturation on key road sections, or a shift in the congestion spread trend from expansion to local maintenance or weakening, then the current traffic light configuration parameters and the corresponding simulation results will be organized into traffic management suggestion information. Traffic management suggestion information should include at least: signal control node identifier, suggested signal cycle length for the corresponding node, suggested green light duration for each phase, suggested execution time period, a list of congested road sections expected to be alleviated, expected reduction in average travel time, expected reduction in path delay increment, and expected suppression of congestion spread direction. If necessary, the associated influence weight of the corresponding node, the spatiotemporal correlation coefficient triggering the suggestion, and the propagation hysteresis can also be added to determine the scope and priority of the suggestion.

[0117] Figure 2 and Figure 3 This is a comparison diagram of the road network status before and after the traffic diversion in this application. Figure 2 This diagram illustrates the local road network state before traffic diversion, drawn based on the traffic flow offset simulation results in S4. It represents the local road network connectivity state before the traffic diversion plan was applied. Arrows indicate the direction of travel along road connections, and nodes represent road nodes. Statistical results show that before diversion, the average travel time was 45 minutes, the path delay increment was 12 minutes, and there were 5 congested road segments. Figure 3This diagram illustrates the local road network status after traffic diversion. It is drawn based on the overall traffic distribution simulation results after reconfiguring signal control parameters in the preferred embodiment of S6, representing the local road network connectivity status after applying the traffic diversion scheme. Corresponding statistical results show that the average travel time after diversion is 32 minutes, the path delay increment is 3 minutes, and the number of congested road segments is 1. The comparison before and after diversion shows that the traffic conditions of the local road network have improved, with a decrease in both average travel time and path delay increment, and a reduction in the number of congested road segments.

[0118] The above describes a method for predicting traffic flow according to an embodiment of this application. The following describes a system for predicting traffic flow according to an embodiment of this application. Please refer to [link / reference]. Figure 4 This application provides a schematic diagram of a system for predicting traffic flow, the system comprising:

[0119] The feature extraction module 10 is used to acquire multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer features between nodes, and extract transfer dynamic fluctuation features based on the changes of traffic transfer features in the time dimension.

[0120] The predictive analysis module 20 is used to perform time-series trend analysis on the historical data sequence of traffic transfer paths, and combined with the dynamic fluctuation characteristics of transfer, to predict the scope of the transfer impact caused by changes in road network status.

[0121] The filtering and mapping module 30 is used to filter the affected road subsets based on the road network topology when the transfer impact range exceeds the preset range threshold, and generate a traffic transfer spatial distribution map based on the transfer traffic density index corresponding to the road subsets.

[0122] The iterative adjustment module 40 is used to update the weight parameters of the road network connection relationship model according to the spatial distribution map of traffic transfer, and to perform traffic offset simulation to determine the adjusted road network connection status characteristics.

[0123] The spatial mapping module 50 is used to perform geospatial mapping on the road network connection status characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution.

[0124] The prediction output module 60 is used to determine whether the traffic flow quantitative indicators meet the preset congestion risk conditions. If so, it outputs the traffic flow prediction results.

[0125] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the method for predicting traffic flow.

[0126] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting traffic flow, characterized in that, The method includes: S1. Obtain multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer characteristics between nodes, and extract transfer dynamic fluctuation characteristics based on the changes of the traffic transfer characteristics in the time dimension. S2. Perform time-series trend analysis on the historical data sequence of traffic transfer paths, and combine the dynamic fluctuation characteristics of the transfer to predict the scope of the transfer impact caused by changes in road network status; S3. When the range of influence of the transfer exceeds the preset range threshold, the affected road subset is screened based on the road network topology, and a traffic transfer spatial distribution map is generated according to the transfer traffic density index corresponding to the road subset. S4. Update the weight parameters of the road network connection relationship model according to the traffic transfer spatial distribution map, and perform traffic offset simulation to determine the adjusted road network connection status characteristics. S5. Perform geospatial mapping on the road network connection status characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution. S6. Determine whether the traffic flow quantification index meets the preset congestion risk conditions. If so, output the traffic flow prediction result. In step S1, multi-source dynamic data of the mountain road network is acquired, and a road network connection relationship model is constructed. This includes: acquiring latitude and longitude coordinates and traffic status data of the mountain road network through satellite imagery and ground sensor networks, defining this as multi-source dynamic data; performing noise filtering and timestamp annotation on the multi-source dynamic data to obtain preprocessed data; identifying traffic transfer paths based on the time-varying sequence of latitude and longitude coordinates in the preprocessed data, determining the transfer probability distribution based on the historical frequency of each traffic transfer path, and simultaneously determining the initial time delay distribution corresponding to each traffic transfer path based on the time difference in the changing sequence; generating a mountain road network dynamic dataset based on the preprocessed data, the transfer probability distribution, and the initial time delay distribution; and modeling the connection relationships of road nodes and road connecting edges in the mountain road network using graph structure analysis methods based on the mountain road network dynamic dataset to obtain the road network connection relationship model. In S1, traffic transfer features between nodes are extracted, and dynamic fluctuation features of the transfer are extracted based on the changes of the traffic transfer features in the time dimension. This includes: determining the state transfer relationship between nodes based on the road network connection relationship model and the transfer probability distribution corresponding to each traffic transfer path, and constructing a traffic transfer probability matrix as the traffic transfer feature; setting a sliding time window, extracting the time series data of the transfer probability between each node in the traffic transfer probability matrix within the sliding time window, calculating the statistical standard deviation of the time series data, and obtaining the dynamic fluctuation features of the transfer. S3 includes: determining whether the transfer impact range exceeds a preset range threshold; if so, performing a topology search within the transfer impact range based on the road network topology structure represented by the road network connection relationship model to filter out the affected road subset; extracting the transfer probability between each node within the road subset according to the traffic transfer characteristics, and constructing a local traffic transfer probability matrix corresponding to the road subset; determining the transfer traffic of each road segment within the road subset based on the local traffic transfer probability matrix and the current traffic flow of the road subset, and calculating the transfer traffic density index corresponding to each road segment in conjunction with the capacity coefficient of each road segment; and spatially mapping the topological spatial location of the road subset with the transfer traffic density index to generate the traffic transfer spatial distribution map.

2. The method according to claim 1, characterized in that, S2 include: Obtain the historical data sequence corresponding to the traffic transfer path; The historical data sequence is analyzed using a time-series trend test method, and combined with the dynamic fluctuation characteristics of the transfer, the trend direction and significance of the flow transfer changes in the historical data sequence are determined. Based on the trend direction and the significance of the trend, potential affected road nodes or road segments associated with the traffic transfer path are identified, and the scope of the transfer impact caused by changes in road network status is predicted accordingly.

3. The method according to claim 1, characterized in that, S4 includes: Based on the spatial distribution map of traffic transfer, the incremental traffic resistance of each road segment in the road subset and the congestion frequency of each node are determined. By combining the traffic resistance increment and the congestion frequency, the weight parameters of the corresponding node connection edges in the road network connection relationship model are updated; Traffic offset simulation is performed based on the updated road network connectivity model to obtain the traffic distribution deviation of the road network under traffic transfer conditions; Based on the traffic distribution deviation, the adjusted road network connection status characteristics are determined.

4. The method according to claim 1, characterized in that, S5 include: By combining the geographical coordinates of the corresponding road segments with the preset regional influence weights, the road network connection status characteristics are geospatially mapped. Traffic flow quantification indicators are extracted based on spatial mapping results. These traffic flow quantification indicators include at least average travel time and path delay increment. Based on the spatial mapping results, a visual distribution map reflecting the dynamic evolution of road network traffic is generated.

5. The method according to claim 1, characterized in that, S6 include: Based on the aforementioned traffic flow quantification indicators, the comprehensive congestion risk index and traffic flow saturation of the corresponding road segment are calculated. Determine whether the comprehensive congestion risk index and / or the traffic flow saturation exceed the corresponding preset congestion risk threshold. If so, predict the expected traffic flow and congestion spread trend of the corresponding road segment in the future preset time period based on the spatial mapping result and the road network connection status characteristics. The projected traffic flow and the congestion spread trend are used as the traffic flow prediction results.

6. A system for predicting traffic flow, for implementing the method as described in any one of claims 1 to 5, characterized in that, The system includes: The feature extraction module is used to acquire multi-source dynamic data of the mountain road network, construct a road network connection relationship model, extract traffic transfer features between nodes, and extract transfer dynamic fluctuation features based on the changes of the traffic transfer features in the time dimension. The predictive analysis module is used to perform time-series trend analysis on the historical data sequence of traffic transfer paths, and, in combination with the dynamic fluctuation characteristics of the transfer, predict the range of impact of the transfer caused by changes in the road network status. The filtering and mapping module is used to filter the affected road subset based on the road network topology when the range of the transfer impact exceeds a preset range threshold, and generate a traffic transfer spatial distribution map according to the transfer traffic density index corresponding to the road subset. The iterative adjustment module is used to update the weight parameters of the road network connection relationship model according to the traffic transfer spatial distribution map, and to perform traffic offset simulation to determine the adjusted road network connection state characteristics. The spatial mapping module is used to perform geospatial mapping on the road network connection status characteristics, extract traffic flow quantitative indicators based on the spatial mapping results, and generate a visual distribution map of traffic flow evolution. The prediction output module is used to determine whether the traffic flow quantification index meets the preset congestion risk conditions. If so, it outputs the traffic flow prediction result.

7. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement a method for predicting traffic flow as described in any one of claims 1 to 5.

Images