A city traffic heat resilience evaluation method and system based on multi-system feature fusion and machine learning

Abstract

This invention discloses a method and system for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning, belonging to the field of data assessment technology. The method includes: defining traffic sensitivity based on the road segment baseline capacity of urban traffic; defining grid-layer thermal indices and calculating heat load; identifying heat events based on a heat load sequence list; using a machine learning algorithm to learn and train based on neighborhood state vectors and feature-fused state vectors, outputting local decision vectors, and constructing exit allocation coefficients; weighting and summing the exit allocation coefficients with the effective capacity superimposed on accidents and heat events to obtain the service capacity value of the intersection; defining the convergence inflow volume derived from upstream exit flow on the traffic flow topology map, forming a performance sequence derived from the traffic flow topology map; and transforming the thermal impact from spatial diffusion to structured transmission along traffic reachable channels through linear attenuation of traffic travel distance.

Description

Technical Field

[0001] This invention relates to the field of data assessment technology, and in particular to a method and system for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning. Background Technology

[0002] Urban transportation systems face multiple pressures under high-temperature scenarios, including decreased operational efficiency, disruptions in travel demand, and shifts in travel modes. Traditional urban transportation research and management practices mainly focus on indicators such as traffic flow, capacity, congestion level, and travel efficiency, and often use traffic demand models, network distribution models, or classical statistical regression models to characterize travel behavior under normal weather conditions. These methods have a certain explanatory power for traffic operation under normal temperature conditions.

[0003] However, it is difficult to reflect the dynamic evolution characteristics of traffic resilience. First, many methods treat the thermal impact as a uniform background and often use simple spatial proximity or administrative area average to distribute thermal exposure. Second, in the actual process of considering thermal resilience, it is easy not to convert the thermal load into a clear constraint on the traffic supply side. Moreover, when accidents and high temperatures are superimposed, there is a lack of a unified capacity superposition variable, which makes it difficult to reflect the nonlinear amplification and cascading congestion of the compound disturbance. Therefore, the assessment often stays at the level of predicting the probability of congestion and it is difficult to further quantify the resilience dimensions such as maximum performance loss, recovery speed and instability of the recovery process. As a result, when facing real urban scenarios with frequent high temperatures and superimposed disturbances, existing technologies often cannot simultaneously take into account accuracy, interpretability and feasibility. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning. It addresses the shortcomings of existing technologies that treat thermal impact as a uniform background, often using simple spatial proximity or administrative district averages to distribute heat exposure. Furthermore, in practical considerations of thermal resilience, it is easy to fail to translate heat load into a clear constraint on the traffic supply side. Moreover, when accidents and high temperatures overlap, there is a lack of a unified capacity superposition variable, making it difficult to reflect the nonlinear amplification and cascading congestion of the combined disturbances. Consequently, assessments often remain at the level of predicting congestion probability, making it difficult to further quantify resilience dimensions such as maximum performance loss, recovery speed, and the instability of the recovery process. This makes it difficult for existing technologies to simultaneously address accuracy, interpretability, and feasibility when facing real-world urban scenarios with frequent high temperatures and overlapping disturbances.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning, comprising,

[0008] Collect urban traffic data and regional meteorological information to form a heterogeneous node set;

[0009] Analyze wind-driven anisotropy and calculate the directional consistency of different edges. Combine the network temperature gradient with meteorological information to define the thermal retention trigger index. Define traffic sensitivity based on the road segment baseline capacity of urban traffic. Combine traffic sensitivity, thermal retention trigger index and directional consistency to form the thermal impact edge weight.

[0010] Based on the heat impact edge weights and the corresponding traffic structure edge sets, the grid layer heat index is defined and the heat load is calculated. The heat impact edge weights of grid intersections and grid road segments are considered to form a heat load sequence table, and heat events are identified based on the heat load sequence table.

[0011] Define the capacity reduction ratio for each road segment, calculate the effective traffic thermal capacity after the impact of heat, define the multi-system feature fusion state vector of the intersection, and use machine learning algorithms to learn and train based on the neighborhood state vector and the feature fusion state vector to output the local decision vector and construct the exit allocation coefficient.

[0012] The service capacity value of the intersection is obtained by weighting and summing the effective capacity superimposed with the exit allocation coefficient and the accident and heat events. This enables the clearing capacity of the road segment under the capacity constraint for accidents and heat events. Based on the queue length of the current road segment, the arrival volume at the intersection, and the service capacity value, the convergence inflow volume derived from the upstream exit flow on the traffic flow topology map is defined, forming the derivation performance sequence of the traffic flow topology map.

[0013] Urban traffic thermal resilience is assessed by defining comparison thresholds for thermal resilience assessment items.

[0014] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the following steps are included: collecting urban traffic data and meteorological regional information, collecting urban traffic map data and meteorological regional grid information, mapping urban traffic segments and intersections and meteorological regional grid data to form a multi-system heterogeneous node set, constructing adjacency edges between intersections based on the heterogeneous node set, and simultaneously constructing the edge topology of intersections and road segments to jointly form a traffic structure topology graph.

[0015] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the step of analyzing wind-driven anisotropy and calculating the directional consistency of different sides includes obtaining the corresponding wind direction data of the meteorological regional grid through urban traffic meteorological information data, and combining the heterogeneous node set to analyze wind-driven anisotropy and calculate directional consistency.

[0016] By using meteorological information data of urban traffic, the amplitude of network temperature gradient is calculated and a high gradient network threshold is set. A heat retention trigger index is defined, and the quantile of the lowest nighttime temperature in the city is set as the nighttime threshold. This threshold, together with the high gradient network threshold, forms a heat source trigger term.

[0017] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the comprehensive traffic sensitivity, thermal retention trigger index, and directional consistency together constitute the thermal impact edge weight, which includes analyzing meteorological regional grid information, extracting the m intersections closest to the grid node, determining the propagation neighborhood based on the maximum allowable m number, extracting the traffic step distance of the grid to different intersections, and propagating the grid thermal impact to the intersection through traffic channels with linear attenuation. Based on the endpoint data of different road segments, the minimum channel distance from the grid to the intersections at both ends of the road segment is extracted as a channel gating.

[0018] Based on the baseline capacity analysis of urban traffic road segments, the sensitivity of different road segments and the sensitivity of road segment intersection capacity are analyzed. Based on the queue length of the node intersection, the oscillation susceptibility analysis is carried out, and traffic sensitivity is defined. The thermal influence edge weight is composed of traffic sensitivity, channel gating, thermal retention trigger index and directional consistency.

[0019] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the step of defining a capacity reduction ratio for each road segment and calculating the effective traffic thermal capacity after thermal impact includes defining a grid layer thermal index through a thermal index, and calculating the thermal load by weighted aggregation based on the thermal impact edge weights and the corresponding traffic structure edge sets, and calculating the intersection thermal load and road segment thermal load together with the thermal impact edge weights of the grid intersections and the grid road segments, thus forming a thermal load sequence table.

[0020] Define a capacity reduction ratio for each road segment, limit the capacity reduction ratio using a truncation function, calculate the heat-induced effective capacity of traffic after the impact of heat based on the capacity reduction ratio, and calculate the capacity change based on the baseline effective capacity of urban road segments.

[0021] For each heat event, the set of intersections is associated with the road segment where the event occurred, and a set of diffusion intersections is defined based on the propagation neighborhood. Traffic accident status is monitored at any intersection in the set of intersections to form an accident status matrix. The road segments where the accident occurred are selected, and the effective capacity of the superposition of the accident and heat event is calculated.

[0022] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the heat load sequence table includes the definition of a multi-system feature fusion state vector for intersections based on the accident state matrix, the effective capacity of the superposition of accidents and heat events, and the traffic intersection queue observation queue and road segment flow extracted from urban traffic data.

[0023] Based on the propagation neighborhood and the feature fusion state vector, state mapping is performed through aggregation calculation to obtain the neighborhood state vector of the intersection. Through the machine learning module, the neighborhood state vector and the feature fusion state vector are used for learning and training, outputting the local decision vector, and constructing the exit allocation coefficient.

[0024] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the inferred performance sequence constituting the traffic flow topology map includes: obtaining the service capacity value of the intersection by weighting and summarizing the effective capacity superimposed with the exit allocation coefficient and the accident and heat events, thereby realizing the clearing capacity of the road segment under the capacity constraint for accidents and heat events; and defining the convergence inflow volume inferred from the upstream exit flow on the traffic flow topology map based on the queue length of the current road segment, the arrival volume at the intersection, and the service capacity value, thus forming the inferred performance sequence of the traffic flow topology map.

[0025] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the thermal resilience assessment item comprises: averaging the pre-event window based on the projected performance sequence and referring to the thermal event as the baseline projected performance value; quantifying the maximum performance loss during the event; comparing and normalizing the minimum value as the disturbance rejection reduction magnitude; and quantifying the recovery instability from the start of the event to the recovery time as the event oscillation penalty, which together with the disturbance rejection reduction magnitude constitutes the thermal resilience assessment item.

[0026] As a preferred embodiment of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning described in this invention, the urban traffic thermal resilience assessment includes: calculating the sum of the mean and twice the standard deviation of the disturbance rejection reduction magnitude and the oscillation penalty data based on historical traffic flow topology data in the thermal resilience assessment items, respectively, as the disturbance rejection threshold and the penalty threshold; if the disturbance rejection reduction magnitude is less than the disturbance rejection threshold or the oscillation penalty data is less than the penalty threshold, it is judged to be of poor thermal resilience.

[0027] Secondly, this invention provides an urban traffic thermal resilience assessment system based on multi-system feature fusion and machine learning, comprising,

[0028] Heterogeneous node construction module: Constructs a heterogeneous node set including meteorological grids, intersections, and road segments;

[0029] Thermal impact edge weight calculation module: Based on the consistency of wind direction anisotropy, network temperature gradient and nighttime heat retention trigger index, and combined with traffic sensitivity, calculate the thermal impact edge weight from the grid to the intersection / road segment.

[0030] Heat load calculation module: The heat index of the grid layer is weighted and aggregated according to the edge weight of heat impact and the edge set of traffic structure to form a heat load sequence table;

[0031] The heat-induced effective capacity calculation module: Based on the heat load sequence list, it identifies heat events during abnormal heat periods, calculates the capacity reduction ratio for each road segment according to the intensity of the heat event, and obtains the heat-induced effective capacity;

[0032] Exit allocation coefficient generation module: Based on the propagation neighborhood aggregation to obtain the neighborhood state vector, the fused state vector and the neighborhood state vector are used as input for learning and training, and the local decision vector is output to construct the allocation coefficient of each exit of the intersection.

[0033] Traffic simulation module: Calculates intersection service capacity based on exit allocation coefficient and accident heat superposition effective capacity, and simulates convergence inflow and queue evolution on traffic flow topology map to generate simulation performance sequence;

[0034] Thermal resilience assessment module: Based on preset comparison thresholds, the module judges the inferred performance sequence and outputs the urban traffic thermal resilience assessment results.

[0035] The beneficial effects of this invention are as follows: By using the linear decay of traffic topology distance as channel gating and constraining the propagation from the grid to the road segment with the minimum channel distance at the road segment endpoints, the thermal impact is transformed from spatial diffusion to structured transmission along traffic-accessible channels. This reduces false impact edges that cross topological barriers and improves the ability to characterize the cascading propagation of congestion. By using the thermal impact edge weights and traffic structure edge sets together for the topological weighted aggregation of grid-layer thermal indices and simultaneously forming intersection thermal load and road segment thermal load sequences, the same meteorological thermal field is consistently mapped into two types of operable disturbances: nodes and edges. By establishing baseline extrapolation performance through the average of the pre-event window aligned with thermal events and normalizing the disturbance resistance reduction magnitude with the minimum value during the event, and simultaneously characterizing recovery instability by accumulating the absolute values ​​of adjacent differences in the queue, the depth loss and recovery quality under the same disturbance are decoupled and quantified. This enables the identification of vulnerable patterns with relatively fast surface recovery but repeated congestion resurgence, thus making the thermal resilience assessment not only reflect the magnitude and duration of the drop but also the stability of the recovery path. Attached Figure Description

[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning in Example 1.

[0038] Figure 2 This is a schematic diagram of the heterogeneous node construction and thermal impact propagation of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning in Example 1.

[0039] Figure 3 This is a flowchart of the machine learning module and exit allocation coefficient generation process of the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning in Example 1.

[0040] Figure 4 This is a structural diagram of the urban traffic thermal resilience assessment system based on multi-system feature fusion and machine learning in Example 1. Detailed Implementation

[0041] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0043] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0044] Example 1, referring to Figure 1 and Figure 4 This is the first embodiment of the present invention, which provides a method for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning, including the following steps:

[0045] S1 collects urban traffic data and regional meteorological information to form a heterogeneous node set;

[0046] S1.1 Collect urban traffic map data and meteorological area grid information, map urban traffic segments and intersections and meteorological area grid data to form a multi-system heterogeneous node set, construct the adjacency edge between intersections based on the heterogeneous node set, and construct the edge topology between intersections and road segments to form a traffic structure topology graph.

[0047] Specifically, by using urban traffic map data, the traffic network topology is obtained, including the intersection set, road segment set, and the start and end points of the road segments. The meteorological regional grid information of the urban area is obtained, including the grid cell set and the grid center coordinates. Based on the urban traffic map data, spatial mapping relationship data is obtained, including intersection coordinates and road segment center coordinates.

[0048] The transportation network topology, meteorological regional grid information, and spatial mapping relationship data are combined to form a multi-system heterogeneous node set.

[0049] S2 analyzes the wind-driven anisotropy and calculates the directional consistency of different edges. It combines the network temperature gradient of meteorological information, defines the thermal retention trigger index, defines traffic sensitivity based on the road segment baseline capacity of urban traffic, and combines traffic sensitivity, thermal retention trigger index and directional consistency to form the thermal impact edge weight.

[0050] S2.1: Obtain the corresponding wind direction data of the meteorological area grid through urban traffic meteorological information data, and combine the heterogeneous node set to analyze the anisotropy driven by wind direction and calculate the direction consistency.

[0051] The anisotropic calculation direction consistency includes defining the displacement vectors and included angles of heterogeneous nodes, determining the downwind direction consistency based on the included angles, defining wind speed gating based on the city-wide quantiles for the month, defining the distance term based on linear radius attenuation, and synthesizing the directional propagation intensity value, expressed as:

[0052]

[0053]

[0054]

[0055]

[0056]

[0057]

[0058] in, Represents spatial coordinate distance. and Let x and g represent the spatial coordinates of nodes respectively. This indicates the angle between the direction the grid points towards the target and the wind direction. Indicates the wind direction of grid g. Represents the unit vector of wind direction. Indicates the spatial decay scale of thermal effects. Indicates the minimum value. This indicates that the wind direction is observable. Indicates wind speed gating. Represents the wind speed at node g. and These represent the 10th and 90th percentiles of the city's wind speed for the month, respectively. Represents the distance term. Indicates the directional propagation intensity value;

[0059] Using meteorological information data of urban traffic, the amplitude of network temperature gradient is calculated and a high gradient network threshold is set. A heat retention trigger index is defined, and the quantile of the lowest nighttime temperature in the city is set as the nighttime threshold. This threshold, together with the high gradient network threshold, forms a heat source trigger term.

[0060] It should be noted that, based on the network temperature gradient magnitude and the 90th percentile of all current grid gradients, the heat source triggering term can be represented as follows:

[0061]

[0062]

[0063]

[0064] in, Indicates the heat source trigger item. This indicates the gradient triggering indicator variable. This indicates the nighttime heat retention trigger index. Indicates the magnitude of the temperature gradient. This represents the 90th percentile of the temperature gradient magnitude across all grid cells. This indicates the lowest temperature recorded at night within the grid. This represents the 75th percentile of the lowest nighttime temperature across all grid cells;

[0065] By simultaneously incorporating the directional consistency of wind anisotropy, wind speed quantile gating, and distance linear attenuation into the edge weight calculation, the thermal impact is amplified only along spatial channels with downwind conditions and wind speeds within the effective propagation range, while other directions are systematically suppressed. This results in a distinguishable order of propagation strength at the same Euclidean distance, avoiding the misjudgment of static proximity as the actual heat transport path.

[0066] S2.2 Analyze the meteorological area grid information, extract the m intersections closest to the grid node, and extract the traffic traversal distances between the grid and different intersections. The grid thermal impact is propagated to the intersections through traffic channels with linear attenuation. Based on the endpoint data of different road segments, the minimum channel distance from the grid to the intersections at both ends of the road segment is extracted as channel gating.

[0067] It should be noted that the linear attenuation effect of the thermal influence of the grid propagating through traffic channels to the intersection can be expressed as:

[0068]

[0069] in, Indicates channel gating, Let g represent the traffic traverse distance from grid g to intersection i, and K represent the maximum allowable number of meters.

[0070] Similarly, the propagation neighborhood is determined based on the maximum allowed number of m, which can refer to all nodes that can be reached from a given node through the edges in the node graph, and the number of hops of the shortest path does not exceed the maximum allowed number of m.

[0071] Based on the baseline capacity analysis of urban traffic road segments, the sensitivity of different road segments and the sensitivity of road segment intersection capacity are analyzed, and the oscillation susceptibility analysis is performed according to the queue length of node intersections to define traffic sensitivity.

[0072] It should be noted that the queue length is used for oscillation analysis. The variance of the queue length at different nodes at different times can be used as the oscillation value. The sum of the mean and variance of the sensitivity values ​​of all nodes is used as the discrimination threshold. Sensitivity values ​​greater than the discrimination threshold are determined as discrimination indicators (1 or 0) through an indicator function. The sum of the mean and variance of the oscillation values ​​of all nodes is used as the discrimination threshold. The oscillation values ​​greater than the discrimination threshold are determined as analysis indicators (1 or 0) through an indicator function. The discrimination indicators and analysis indicators are combined to form the traffic sensitivity level value.

[0073] The thermal impact weight is composed of comprehensive traffic sensitivity, channel gating, thermal retention triggering index, and directional consistency.

[0074] It should be noted that the thermal impact edge weights can be jointly calculated using geometric mean for comprehensive traffic sensitivity, channel gating, heat source triggering term, and directional consistency, expressed as:

[0075]

[0076]

[0077] in, Indicates the heat-affected edge weights at grid intersections. Indicates the heat impact edge weight of the grid segment. This indicates that the wind direction is observable. Indicates the directional propagation intensity value. Indicates the traffic sensitivity level value;

[0078] By using linear attenuation of traffic topology distance as channel gating and constraining the propagation of the grid to the road segment by the minimum channel distance at the road segment endpoints, the thermal impact is transformed from spatial diffusion to structured transmission along traffic-accessible channels. This reduces false impact edges crossing topological barriers and improves the ability to characterize the cascading propagation of congestion. By combining baseline capacity sensitivity and queue variance oscillation to form a traffic sensitivity level, and then combining it with heat source triggering, channel gating, and directional consistency through geometric averaging, a multiplicative gating effect dominated by the weakest link is formed. That is, the edge weights are significant only when the heat source is strong, the direction and channel are accessible, and the node vulnerability is high. This automatically focuses meteorological shocks on key intersections and key road segments that are more likely to cause systemic instability, achieving an interpretable mapping of thermal exposure to traffic degradation risk and improving the nonlinear discriminative power of the assessment results.

[0079] S3. Based on the heat impact edge weights and the corresponding traffic structure edge sets, define the grid layer heat index and calculate the heat load. Consider the heat impact edge weights of grid intersections and grid road segments to form a heat load sequence table. Identify heat events based on the heat load sequence table, define the capacity reduction ratio for each road segment, calculate the effective traffic heat-induced capacity after heat impact, define the intersection multi-system feature fusion state vector, and use machine learning algorithms to learn and train based on the neighborhood state vector and the feature fusion state vector, output the local decision vector, and construct the exit allocation coefficient.

[0080] S3.1 Define the grid layer thermal index using the thermal index HI (which can be calculated in the form of temperature and humidity), and calculate the thermal load by weighted aggregation based on the thermal impact edge weights and the corresponding traffic structure edge sets. The thermal load of the grid intersections and the thermal load of the grid road segments are calculated together with the thermal impact edge weights of the grid intersections and the thermal load of the road segments to form a thermal load sequence table.

[0081] The calculation of heat load is expressed as follows:

[0082]

[0083]

[0084]

[0085] in, Indicates the heat load at the intersection. Indicates the heat load of the road section. Represents the set of edges of the traffic structure at the intersection. Represents the set of edges of the road segment traffic structure. Indicates the thermal index of the grid layer. Temperature data representing time t, Indicates relative humidity;

[0086] The grid layer thermal index at each time step is defined and identified using an indicator function. The identification threshold is defined based on the sum of the mean and twice the standard deviation of the grid layer thermal index data of all grids, and grids whose grid layer thermal index exceeds the identification threshold are identified as thermal events.

[0087] A capacity reduction ratio is defined for each road segment. This ratio is then limited by a cutoff function. Based on this ratio, the heat-induced effective capacity of traffic after thermal effects is calculated. Furthermore, based on the baseline effective capacity of urban road segments, the capacity change is calculated, which can be expressed as:

[0088]

[0089]

[0090]

[0091] in, Indicates the capacity reduction ratio. Indicates the recognition threshold. This indicates the maximum discount limit. Indicates thermally induced effective capacity. Indicates the change in capacity;

[0092] For each heat event, the set of intersections associated with the road segment where the event occurred is defined, and a set of diffusion intersections is defined based on the propagation neighborhood. Traffic accident status monitoring is performed on any intersection in the set of intersections to form an accident status matrix. Road segments where accidents occurred are then selected, and the effective capacity of the superposition of accidents and heat events is calculated, expressed as:

[0093]

[0094] in, This represents the effective capacity for the superposition of accident and heat events, where k represents the event index. This represents the heat-induced effective capacity of event index k. Indicates the percentage of congestion caused by accidents;

[0095] By using the thermal impact edge weights and traffic structure edge sets together for topological weighted aggregation of grid-layer thermal indices and simultaneously forming intersection thermal load and road segment thermal load sequences, the same meteorological thermal field is consistently mapped into two types of operable disturbances: nodes and edges. This transforms spatial thermal exposure into a temporal input that can directly drive traffic supply degradation and control decisions, enabling simultaneous identification of heat source location, propagation path, and differences in the heating of traffic components. By superimposing the heat-induced effective capacity and the accident road segments selected by the accident state matrix with the congestion ratio, a composite effective capacity is obtained. The propagation neighborhood limits the scope of accident diffusion, ensuring that the composite disturbance only generates nonlinear amplification on topologically reachable links, reducing far-field false injections and enhancing the characterization of congestion cascading trigger conditions.

[0096] S3.2, Based on the accident state matrix, the effective capacity of the superposition of accidents and heat events, and the traffic intersection queue observation queue and road segment flow extracted from urban traffic data, define the intersection multi-system feature fusion state vector;

[0097] Using traffic accident status monitoring data, the ratio of traffic flow to effective capacity of the approach road segment is used to calculate saturation. By considering the effective capacity resulting from the superposition of accidents and heat events, a decrease in capacity is immediately reflected in the status characteristics. The traffic flow data item in the feature fusion status vector is updated and replaced by the saturation calculation, as shown below:

[0098]

[0099] in, Indicates the traffic saturation at the intersection. Indicates the collection of road sections at the intersection entrance. Indicates the inflow traffic volume to the road segment;

[0100] Based on the propagation neighborhood and the feature fusion state vector, state mapping is performed through aggregation calculation to obtain the neighborhood state vector of the intersection. Then, the machine learning module learns and trains using the neighborhood state vector and the feature fusion state vector, outputting a local decision vector and constructing the exit allocation coefficient, expressed as:

[0101]

[0102]

[0103] in, Represents the local decision vector. This represents the output of the machine learning module. Represents the feature fusion state vector. Represents the neighborhood state vector. Represents the export allocation coefficient. Indicates the set of road segments at the intersection exits. Indicates the segment index of non-x segment;

[0104] By using composite effective capacity in the feature fusion state to participate in saturation calculation and writing back to replace the flow term, the demand intensity and the disturbed supply capacity can be compared in real time on the same feature dimension. This makes it easier for learning and training to capture the critical saturation transition and queuing surge inflection point caused by capacity compression. By jointly inputting the aggregation mapping of the neighborhood state vector and the local fusion state into the machine learning module and outputting the local decision vector, and then normalizing it to generate the exit allocation coefficient, the control allocation has the adaptive flow guidance capability of neighborhood cooperation under the constraints of thermal load and composite capacity. Thus, without relying on global control, the limited release capacity is prioritized to the direction that can better suppress thermal propagation loss, thereby improving the system-level thermal resilience maintenance and recovery efficiency.

[0105] S4 constructs the service capacity of different intersections by superimposing the exit allocation coefficient with the effective capacity of accidents and thermal events, and defines the convergence inflow volume derived from the upstream exit flow on the traffic flow topology map to form the derivation performance sequence of the traffic flow topology map. The maximum performance loss during the event and the recovery instability between recovery times are quantified by the derivation performance sequence to form the thermal resilience assessment item.

[0106] S4.1, by weighting and summing the effective capacity superimposed with the exit allocation coefficient and the accident and heat events, the service capacity value of the intersection is obtained, thereby realizing the clearing capacity of the road segment under the capacity constraint for accidents and heat events. Based on the queue length of the current road segment, the arrival volume of the intersection, and the service capacity value, the convergence inflow volume derived from the upstream exit flow on the traffic flow topology map is defined, forming the derivation performance sequence of the traffic flow topology map.

[0107] It should be noted that in calculating the service capacity value, the weighted sum of the exit capacity is defined as the service capacity value by considering the allocation coefficient of each exit of the intersection exit segment set, as well as the effective capacity superimposed by accidents and heat events, and is expressed as:

[0108]

[0109]

[0110]

[0111] in, Indicates the service capability value. Indicates the number of arrivals at the intersection. Indicates the current queue at the intersection. Indicates the deduction queue;

[0112] By weighting and summing the exit allocation coefficient with the effective capacity superimposed on accidents and heat events to construct the intersection service capacity, the learned local diversion decisions are forcibly projected onto the disturbed supply boundary. This binds the control strategy to the physical clearance capacity, avoiding evaluation biases where decisions are feasible but capacity is infeasible under extreme heat and accident scenarios, and making the assessment results more sensitive to actual supply collapse. By constraining queue evolution with service capacity on the traffic flow topology map and using upstream exit flow to extrapolate convergence inflow to form a closed loop, the impact of disturbances is gradually transmitted along the topological links with a conservation relationship. This naturally presents the diffusion and back-transmission effect of cascading congestion in the performance sequence, avoiding misjudging local capacity decline as global degradation or the opposite.

[0113] S4.2, based on the projected performance sequence, the average value before the event is taken as the baseline projected performance value, and the maximum performance loss during the event is quantified. The minimum value is compared and normalized to be used as the disturbance rejection reduction magnitude. At the same time, the recovery instability from the start of the event to the recovery time is quantified as the event oscillation penalty, and together with the disturbance rejection reduction magnitude, they form the thermal toughness evaluation term.

[0114] It should be noted that the oscillation penalty is calculated by accumulating the absolute values ​​of adjacent differences based on the derivation queue, thereby obtaining the oscillation penalty term;

[0115] By establishing a baseline performance projection using the pre-event window averaging aligned with thermal events, and normalizing the disturbance rejection rate using the minimum value during the event period, recovery instability is characterized by accumulating the absolute values ​​of adjacent differences in the queue. This decouples and quantifies the depth loss and recovery quality under the same disturbance, enabling the identification of vulnerable modes with rapid surface recovery but recurring congestion and resurgence. Thus, thermal resilience assessment reflects not only the magnitude and duration of the drop but also the stability of the recovery path. By combining the disturbance rejection rate with the oscillation penalty to form a thermal resilience assessment term, the assessment becomes more sensitive to nonlinear amplification and secondary instability under complex disturbances. This allows for the differentiation between two different mechanisms of resilience degradation: slow thermal capacity decay and instantaneous bottleneck triggering caused by accident superposition. This provides a usable classification index for subsequent strategy optimization and vulnerable chain location.

[0116] S5, assessing urban traffic thermal resilience by defining comparison thresholds for thermal resilience assessment items;

[0117] S5.1 According to the thermal resilience assessment item, the sum of the mean and twice the standard deviation of the data for the disturbance rejection reduction and the oscillation penalty is calculated based on the historical traffic flow topology data. These are used as the disturbance rejection threshold and the penalty threshold, respectively. If the disturbance rejection reduction is less than the disturbance rejection threshold or the data for the oscillation penalty is less than the penalty threshold, then the thermal resilience is judged to be poor.

[0118] This embodiment also provides an urban traffic thermal resilience assessment system based on multi-system feature fusion and machine learning, including,

[0119] Heterogeneous node construction module: Constructs a heterogeneous node set including meteorological grids, intersections, and road segments;

[0120] Thermal impact edge weight calculation module: Based on the consistency of wind direction anisotropy, network temperature gradient and nighttime heat retention trigger index, and combined with traffic sensitivity, calculate the thermal impact edge weight from the grid to the intersection / road segment.

[0121] Heat load calculation module: The heat index of the grid layer is weighted and aggregated according to the edge weight of heat impact and the edge set of traffic structure to form a heat load sequence table;

[0122] The heat-induced effective capacity calculation module: Based on the heat load sequence list, it identifies heat events during abnormal heat periods, calculates the capacity reduction ratio for each road segment according to the intensity of the heat event, and obtains the heat-induced effective capacity;

[0123] Exit allocation coefficient generation module: Based on the propagation neighborhood aggregation to obtain the neighborhood state vector, the fused state vector and the neighborhood state vector are used as input for learning and training, and the local decision vector is output to construct the allocation coefficient of each exit of the intersection.

[0124] Traffic simulation module: Calculates intersection service capacity based on exit allocation coefficient and accident heat superposition effective capacity, and simulates convergence inflow and queue evolution on traffic flow topology map to generate simulation performance sequence;

[0125] Thermal resilience assessment module: Based on preset comparison thresholds, the module judges the inferred performance sequence and outputs the urban traffic thermal resilience assessment results.

[0126] This embodiment also provides a computer device applicable to the urban traffic thermal resilience assessment method and system based on multi-system feature fusion and machine learning, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the urban traffic thermal resilience assessment method and system based on multi-system feature fusion and machine learning proposed in the above embodiment.

[0127] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0128] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the urban traffic thermal resilience assessment method and system based on multi-system feature fusion and machine learning as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0129] In summary, this invention uses linear attenuation of traffic topology distance as channel gating and constrains the propagation of grid to road segment by minimum channel distance at road segment endpoints. This transforms the spatial diffusion of thermal impact into a structured transmission along accessible traffic channels, reducing false impact edges that cross topological barriers and improving the ability to characterize congestion cascading propagation. By using thermal impact edge weights and traffic structure edge sets together for topological weighted aggregation of grid-layer thermal indices and simultaneously forming intersection thermal load and road segment thermal load sequences, the same meteorological thermal field is consistently mapped into two types of operable perturbations: nodes and edges. Baseline performance is established by averaging pre-event windows aligned with thermal events, and the perturbation resistance reduction is obtained by normalizing with the minimum value during the event. At the same time, recovery instability is characterized by accumulating the absolute values ​​of adjacent differences in the queue. This decouples and quantifies the depth loss and recovery quality under the same perturbation, enabling the identification of vulnerable patterns with rapid surface recovery but recurring congestion resurgence. Thus, thermal resilience assessment not only reflects the magnitude and duration of the drop but also the stability of the recovery path.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning, characterized in that: include, Collect urban traffic data and regional meteorological information to form a heterogeneous node set; Analyze wind-driven anisotropy and calculate the directional consistency of different edges. Combine the network temperature gradient with meteorological information to define the thermal retention trigger index. Define traffic sensitivity based on the road segment baseline capacity of urban traffic. Combine traffic sensitivity, thermal retention trigger index and directional consistency to form the thermal impact edge weight. Based on the heat impact edge weights and the corresponding traffic structure edge sets, the grid layer heat index is defined and the heat load is calculated. The heat impact edge weights of grid intersections and grid road segments are considered to form a heat load sequence table, and heat events are identified based on the heat load sequence table. Define the capacity reduction ratio for each road segment, calculate the effective traffic thermal capacity after the impact of heat, define the multi-system feature fusion state vector of the intersection, and use machine learning algorithms to learn and train based on the neighborhood state vector and the feature fusion state vector to output the local decision vector and construct the exit allocation coefficient. The service capacity value of the intersection is obtained by weighting and summing the effective capacity superimposed with the exit allocation coefficient and the accident and heat events. Based on the queue length of the current road segment, the arrival volume of the intersection, and the service capacity value, the convergence inflow volume derived from the upstream exit flow on the traffic flow topology map is defined to form the derivation performance sequence of the traffic flow topology map. Urban traffic thermal resilience is assessed by defining comparison thresholds for thermal resilience assessment items.

2. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 1, characterized in that: The collection of urban traffic data and meteorological regional information includes collecting urban traffic map data and meteorological regional grid information, mapping urban traffic segments and intersections and meteorological regional grid data to form a multi-system heterogeneous node set, constructing adjacency edges between intersections based on the heterogeneous node set, and constructing edge topology between intersections and road segments to jointly form a traffic structure topology graph.

3. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 2, characterized in that: The analysis of wind-driven anisotropy and calculation of directional consistency of different sides includes obtaining the corresponding wind direction data of the meteorological area grid through urban traffic meteorological information data, and combining the heterogeneous node set to analyze wind-driven anisotropy and calculate directional consistency. By using meteorological information data of urban traffic, the amplitude of network temperature gradient is calculated and a high gradient network threshold is set. A heat retention trigger index is defined, and the quantile of the lowest nighttime temperature in the city is set as the nighttime threshold. This threshold, together with the high gradient network threshold, forms a heat source trigger term.

4. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 3, characterized in that: The comprehensive traffic sensitivity, heat retention trigger index, and directional consistency together constitute the thermal impact edge weight. This includes analyzing meteorological regional grid information, extracting the m intersections closest to the grid node, determining the propagation neighborhood based on the maximum allowable m number, extracting the traffic step distance between the grid and different intersections, and linearly attenuating the grid thermal impact to the intersections through traffic channels. Based on the endpoint data of different road segments, the minimum channel distance from the grid to the intersections at both ends of the road segment is extracted as channel gating. Based on the baseline capacity analysis of urban traffic road segments, the sensitivity of different road segments and the sensitivity of road segment intersection capacity are analyzed. Based on the queue length of the node intersection, the oscillation susceptibility analysis is carried out, and traffic sensitivity is defined. The thermal influence edge weight is composed of traffic sensitivity, channel gating, thermal retention trigger index and directional consistency.

5. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 4, characterized in that: The process of defining a capacity reduction ratio for each road segment and calculating the effective traffic heat-induced capacity after being affected by heat includes defining a grid layer heat index through a heat index, and calculating the heat load by weighted aggregation based on the heat impact edge weights and the corresponding traffic structure edge sets. The heat load at the grid intersections and the heat load on the road segments are calculated together with the heat impact edge weights of the grid intersections and the heat impact edge weights of the grid road segments to form a heat load sequence table. Define a capacity reduction ratio for each road segment, limit the capacity reduction ratio using a truncation function, calculate the heat-induced effective capacity of traffic after the impact of heat based on the capacity reduction ratio, and calculate the capacity change based on the baseline effective capacity of urban road segments. For each heat event, based on the set of intersections associated with the road segment where the event occurred, and the set of diffusion intersections defined according to the propagation neighborhood, traffic accident status monitoring is performed on any intersection in the set of intersections to form an accident status matrix, the road segments where the accident occurred are screened, and the effective capacity of the superposition of the accident and heat event is calculated.

6. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 5, characterized in that: The heat load sequence table includes the definition of a multi-system feature fusion state vector for intersections based on the accident state matrix, the effective capacity of the superposition of accidents and heat events, and the traffic intersection queue observation queues and road segment traffic extracted from urban traffic data. Based on the propagation neighborhood and the feature fusion state vector, state mapping is performed through aggregation calculation to obtain the neighborhood state vector of the intersection. Through the machine learning module, the neighborhood state vector and the feature fusion state vector are used for learning and training, outputting the local decision vector, and constructing the exit allocation coefficient.

7. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 6, characterized in that: The performance sequence of the traffic flow topology map includes: weighting and summarizing the effective capacity by combining the exit allocation coefficient with the overlay of accident and heat events to obtain the service capacity value of the intersection, thereby realizing the clearing capacity of the road segment under capacity constraints for accidents and heat events; and defining the convergence inflow volume derived from the upstream exit flow on the traffic flow topology map based on the queue length of the current road segment, the arrival volume at the intersection, and the service capacity value, thus forming the performance sequence of the traffic flow topology map.

8. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 7, characterized in that: The thermal toughness assessment item includes: averaging the pre-event window based on the projected performance sequence and referring to the thermal event as the baseline projected performance value; quantifying the maximum performance loss during the event; comparing and normalizing the minimum value as the disturbance rejection reduction magnitude; and quantifying the recovery instability from the start of the event to the recovery time as the event oscillation penalty, which together with the disturbance rejection reduction magnitude constitutes the thermal toughness assessment item.

9. The urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in claim 8, characterized in that: The urban traffic thermal resilience assessment includes calculating the sum of the mean and twice the standard deviation of the disturbance rejection reduction and oscillation penalty data based on historical traffic flow topology data, respectively, as the disturbance rejection threshold and penalty threshold. If the disturbance rejection reduction is less than the disturbance rejection threshold or the oscillation penalty data is less than the penalty threshold, the thermal resilience is judged to be poor.

10. A system for assessing urban traffic thermal resilience based on multi-system feature fusion and machine learning, based on the urban traffic thermal resilience assessment method based on multi-system feature fusion and machine learning as described in any one of claims 1 to 9, characterized in that: include, Heterogeneous node construction module: Constructs a heterogeneous node set including meteorological grids, intersections, and road segments; Thermal impact edge weight calculation module: Based on the consistency of wind direction anisotropy, network temperature gradient and nighttime heat retention trigger index, and combined with traffic sensitivity, calculate the thermal impact edge weight from the grid to the intersection / road segment. Heat load calculation module: The heat index of the grid layer is weighted and aggregated according to the edge weight of heat impact and the edge set of traffic structure to form a heat load sequence table; The heat-induced effective capacity calculation module: Based on the heat load sequence list, it identifies heat events during abnormal heat periods, calculates the capacity reduction ratio for each road segment according to the intensity of the heat event, and obtains the heat-induced effective capacity; Exit allocation coefficient generation module: Based on the propagation neighborhood aggregation to obtain the neighborhood state vector, the fused state vector and the neighborhood state vector are used as input for learning and training, and the local decision vector is output to construct the allocation coefficient of each exit of the intersection. Traffic simulation module: Calculates intersection service capacity based on exit allocation coefficient and accident heat superposition effective capacity, and simulates convergence inflow and queue evolution on traffic flow topology map to generate simulation performance sequence; Thermal resilience assessment module: Based on preset comparison thresholds, the module judges the inferred performance sequence and outputs the urban traffic thermal resilience assessment results.

Images