Method for dynamically allocating comprehensive traffic capacity based on multi-source data fusion

Abstract

The application constructs a comprehensive traffic capacity dynamic allocation method based on multi-source data fusion, comprising: acquiring multi-source heterogeneous traffic data and performing space-time reference unified processing; extracting asynchronous response time sequence characteristics of each data source, and identifying demand mutation event attributes accordingly; according to the demand mutation event attributes, combining the pre-constructed hub space resource occupation relationship, evaluating the actual available capacity of different traffic modes in the affected transfer hub; under the condition of considering the self-interference effect of the allocated vehicles, generating a capacity allocation scheme for maximizing the demand evacuation quantity; using the response time constant of each traffic mode, converting the allocation scheme into a multi-mode dynamic allocation instruction sequence and outputting. The application can be used to realize early high-confidence determination of demand mutation, quantify capacity attenuation caused by spatial competition, and suppress capacity oscillation in the allocation process, thereby improving the evacuation efficiency under traffic interruption events.

Description

Technical Field

[0001] This invention belongs to the field of urban traffic operation management technology, and in particular relates to a method for dynamic allocation of comprehensive transportation capacity based on multi-source data fusion. Background Technology

[0002] In urban integrated transportation systems, rapid response and dynamic balance of transport capacity are essential to cope with sudden surges in passenger flow or traffic disruptions. Identifying sudden changes in traffic demand and allocating multi-modal transportation resources in real time can alleviate passenger congestion at transfer hubs and improve the overall efficiency of the road network. Researching transport capacity allocation mechanisms can ensure the robustness and operational safety of urban transportation systems, and is conducive to optimizing resource allocation and reducing the risk of systemic congestion.

[0003] Current traffic capacity scheduling primarily relies on sensor data from a single source or static scheduling plans from operating departments. When dealing with multimodal transportation systems, public transportation, taxis, and ride-hailing services are often treated as independent supply units, with their rated capacity calculated separately. When sudden changes in traffic demand occur, response procedures are often only initiated after multiple monitoring indicators trigger warnings.

[0004] In summary, existing technologies struggle to balance the proactiveness of demand identification with the accuracy of resource allocation when facing sudden traffic events. In multimodal transportation environments, the relationship between capacity supply, spatial constraints, and the dynamic evolution of the road network urgently needs to be analyzed. Therefore, it is necessary to research a method that can improve the sensitivity of response to sudden changes in traffic demand and achieve refined dynamic allocation of multimodal transport capacity. Summary of the Invention

[0005] Purpose of the invention: This application designs a method for dynamic allocation of integrated transportation capacity based on multi-source data fusion, in order to solve the above-mentioned problems in the prior art.

[0006] Beneficial effects: This invention can be used to achieve early high-confidence determination of demand mutations, quantify capacity attenuation caused by spatial competition, suppress capacity oscillations during the allocation process, and improve evacuation efficiency under traffic disruption events.

[0007] Technical solution: This application provides a method for dynamic allocation of integrated transportation capacity based on multi-source data fusion, including:

[0008] Acquire multi-source heterogeneous traffic data and perform unified spatiotemporal benchmark processing;

[0009] Extract asynchronous response time-series features from various data sources in multi-source heterogeneous traffic data, and identify the attributes of demand mutation events accordingly;

[0010] Based on the attributes of demand mutation events and the relationship of hub space resource occupancy, assess the actual available capacity of different transportation modes in the affected transfer hubs;

[0011] Taking into account the self-interference effect of the allocated vehicles, a capacity allocation scheme is generated to maximize the demand dispersal volume.

[0012] By utilizing the response time constants of each transportation mode, the capacity allocation plan is transformed into a multi-mode dynamic allocation instruction sequence and output. Attached Figure Description

[0013] Figure 1 A flowchart illustrating an example of a dynamic allocation method for integrated transportation capacity based on multi-source data fusion, provided in this application embodiment.

[0014] Figure 2 This is a flowchart for extracting asynchronous response timing features from various data sources in multi-source heterogeneous traffic data, as provided in an embodiment of this application.

[0015] Figure 3 This is a flowchart illustrating the identification of demand mutation event attributes based on asynchronous response timing characteristics, provided in an embodiment of this application.

[0016] Figure 4 The flowchart illustrates the process of obtaining the probability distribution of response delay, the conditional distribution of anomaly intensity, and the cumulative distribution of response delay, as provided in the embodiments of this application.

[0017] Figure 5 A flowchart for determining the occupancy relationship of hub space resources provided in this application embodiment.

[0018] Figure 6 This is a flowchart provided for embodiments of the present application, which assesses the actual available capacity of different transportation modes in affected transfer hubs by combining the spatial resource occupancy relationship of the hub. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a predetermined order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in sequences other than those illustrated or described herein. Furthermore, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes 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.

[0021] To address the aforementioned issues, the applicant conducted in-depth searches and analyses, and discovered:

[0022] Most existing dispatching algorithms assume that network impedance is only affected by background traffic flow. They lack in-depth correlation analysis of multi-source heterogeneous data, resulting in significant lag in the identification of sudden events in their early stages. Furthermore, due to the failure to fully consider the spatial coupling and competition between different transport modes, the actual carrying capacity within transfer hubs is often overestimated, making it easy for dispatching schemes to induce secondary congestion or fluctuations in transport capacity during implementation.

[0023] To solve these problems, combined with Figures 1 to 6 The present invention will be specifically described through the following embodiments.

[0024] In this application, the preset and pre-set data can be set according to the specific requirements of the actual application scenario.

[0025] Abnormal samples refer to time period records with a missing rate exceeding a set threshold or sample data whose spatial coordinate drift exceeds a reasonable range;

[0026] A unified spatiotemporal reference multi-source data matrix can be organized according to a unified time window and a unified spatial coding.

[0027] The M / M / 1 queuing model can be called a Markov-type single-server queuing model.

[0028] In some scenarios, the asynchronous response time sequence fingerprint feature vector is equivalent to the asynchronous response time sequence feature.

[0029] As an example, a method for dynamic allocation of integrated transportation capacity based on multi-source data fusion includes the following steps:

[0030] Step 101: Acquire multi-source heterogeneous traffic data and perform spatiotemporal benchmark unification processing;

[0031] In the following text, multi-source heterogeneous traffic data refers to real-time data streams accessed from the urban integrated transportation information platform, with data sources having different physical attributes and collection frequencies.

[0032] For example, it may specifically include route replanning request logs from navigation applications, real-time order and vehicle location data from ride-hailing platforms, public transport IC card transaction records, mobile communication operator base station signaling data, and real-time text streams containing traffic event keywords from social media platforms.

[0033] After obtaining the raw records, the records from each channel are uniformly packaged according to the source type identifier, the original collection timestamp, and the original spatial coordinates, and aggregated to form a multi-source heterogeneous traffic raw dataset.

[0034] Furthermore, the original traffic datasets from multiple heterogeneous sources are read, and the timestamps of each data source are uniformly calibrated to the satellite timing reference to eliminate clock deviations between different systems.

[0035] Furthermore, spatial information from various data sources is uniformly mapped to a pre-defined traffic analysis zone coding system.

[0036] Point data is mapped according to the assigned analysis cell, while trajectory data is mapped according to the sequence of analysis cells it passes through.

[0037] After the mapping is completed, outlier samples will be removed to obtain a unified spatiotemporal reference multi-source data matrix.

[0038] In this matrix, rows correspond to different time windows, and columns correspond to the observations of each data source in each analysis cell.

[0039] It should be understood that this step can lay the foundation for the fusion of data from different dimensions.

[0040] Step 102: Extract the asynchronous response time series characteristics of each data source in the multi-source heterogeneous traffic data, and identify the attributes of demand mutation events based on these (asynchronous response time series characteristics);

[0041] This includes extracting the asynchronous response time-series features of each data source in multi-source heterogeneous traffic data, including:

[0042] The normalized anomaly intensity of each data source is determined by comparing the observations of multi-source heterogeneous traffic data within the current time window with the pre-stored historical benchmark.

[0043] Based on the number of time windows in which the normalized anomaly intensity continuously exceeds a preset threshold, the activation status of each data source is determined, and the duration of the data source in the activated state is recorded.

[0044] The normalized anomaly intensity, data source activation state, and duration are combined to construct the asynchronous response timing feature.

[0045] Accordingly, the differences in response time to the same traffic demand surge event from different traffic data sources are utilized. In practical applications, when a traffic surge occurs, fast-response data sources, such as route replanning data from navigation software, typically show rapid fluctuations, while slow-response data sources, such as base station signaling or social media text, exhibit relatively delayed responses.

[0046] The extraction process specifically includes calculating the normalized anomaly intensity of each data source within the current time window and maintaining the activation status flag of each data source.

[0047] The asynchronous response time-series fingerprint feature vector is obtained by combining the status, anomaly intensity, and duration since activation of all data sources.

[0048] In some embodiments, the attributes of demand mutation events are identified based on this (asynchronous response timing characteristics), specifically using a joint likelihood inference method, including:

[0049] For the preset candidate event types and occurrence times, based on the pre-calibrated response delay probability distribution and anomaly intensity conditional distribution, the first conditional probability of the normalized anomaly intensity observed by the data source in the activated state for the duration is calculated.

[0050] Based on the pre-calibrated cumulative distribution of response delay, calculate the second conditional probability that a data source in an inactive state will not respond for a corresponding duration.

[0051] The joint likelihood value is obtained by multiplying the first conditional probabilities of all activated data sources with the second conditional probabilities of all unactivated data sources.

[0052] Select the candidate combination that maximizes the joint likelihood value, and use the event type identifier, estimated occurrence time, and severity level corresponding to it as the attribute of the demand mutation event.

[0053] In some embodiments, the corresponding duration can be the duration from the candidate occurrence time to the current time.

[0054] Based on this, the attributes of demand mutation events are identified, that is, the attribute information of mutation events is determined through the joint likelihood inference framework.

[0055] For example, the demand mutation event attribute is specifically represented as a demand mutation event attribute triple, which includes an event type identifier, an estimated time of occurrence, and a severity level.

[0056] Based on this, a high-confidence determination can be made in the early stages of a mutation event, that is, when most data sources have not yet changed.

[0057] Step 103: Based on the attributes of the demand mutation event and combined with the pre-built hub space resource occupancy relationship, assess the actual available capacity of different transportation modes in the affected transfer hub;

[0058] In this step, the physical layout data and operational parameters of the affected transfer hubs are read, and a mode resource occupancy relationship matrix is ​​established;

[0059] The total overflow demand into the hub is estimated based on the severity level, and a cross-modal spatial interference matrix is ​​constructed by combining the occupancy of shared resources by each mode and the congestion sensitivity weight of the resources; this matrix is ​​used to quantify how an increase in demand for one mode of transportation leads to a decrease in the effective capacity of another mode through spatial competition.

[0060] Based on this, the coupling equilibrium between demand distribution and effective capacity is solved by the fixed-point iterative algorithm to obtain the actual available capacity vector of each mode of the hub, which includes the actual available capacity of each mode.

[0061] It can be used to avoid overestimation of capacity caused by independently assessing the rated capacity of each mode.

[0062] In the above text, the assessment process is used to quantify the cross-capacity attenuation effect of various alternative transportation modes within the affected transfer hub due to the sharing of limited space resources;

[0063] Hub space resources mainly include the number of bus bays, the capacity of curbside passenger drop-off and pick-up areas, the width and throughput of pedestrian walkways, and the area of ​​non-motorized vehicle parking areas;

[0064] The relationship of hub space resource occupancy reflects the proportion of physical resources consumed by each mode of transportation to the total capacity of that resource in order to complete a standard service.

[0065] Operational parameters include the standard stopping time for buses, the average dwell time for ride-hailing vehicles to pick up passengers, and the length of roadside they occupy.

[0066] Step 104: Based on the attributes of the demand mutation event and the actual available transport capacity, and taking into account the self-interference effect of the allocated vehicles, generate a transport capacity allocation plan to maximize the demand evacuation volume.

[0067] Accordingly, the generation process involves developing a detailed plan for allocating transport vehicles from surrounding stations to the affected hubs.

[0068] Furthermore, a three-body coupled optimization framework for allocation decision-making, road network traffic, and demand dispersal was established.

[0069] In practice, the road network topology and normal traffic flow of the affected area are read, and the evacuation traffic and detour traffic caused by the sudden change are superimposed to obtain the dynamic impedance function of the road network under the influence of the interruption.

[0070] Next, an optimization model is established with the goal of maximizing the effective service demand of the affected hubs, where the objective function is to minimize the total unmet demand of all affected hubs.

[0071] When solving the model, a reverse contraction solution strategy is adopted, that is, starting from the upper bound of the full-load allocation scheme of each station, and gradually identifying and reducing invalid route vehicles with negative marginal contribution under the most severe congestion state.

[0072] The above method can directly locate the self-interference saturation point, enabling the output of a globally optimal feasible solution while meeting the real-time allocation calculation time limit requirements.

[0073] In the above text, the self-interference effect refers to the feedback loop, that is, when the number of dispatched vehicles increases, the additional traffic generated after they enter the road network will aggravate the congestion of the road segment, thereby prolonging the travel time of all vehicles passing through the shared road segment, and may even cause some dispatched vehicles to become invalid because their arrival time exceeds the service time window.

[0074] In other words, the act of replenishing transport capacity itself may worsen the conditions for replenishing transport capacity.

[0075] The capacity allocation plan can include the number of vehicles allocated from each station to each hub, the specific route number, and the estimated arrival time.

[0076] Step 105: Using the pre-determined response time constants of each traffic mode, the capacity allocation plan is transformed into a multi-mode dynamic allocation instruction sequence and output.

[0077] In this step, the response time constant reflects the median of the sample delay from receiving the instruction to the actual activation of capacity for different traffic modes.

[0078] For example, ride-hailing services typically have a small time constant, belonging to the fast response layer;

[0079] The time constant for bus and shared bicycle dispatching is usually large, belonging to the slow response layer.

[0080] Because the adjustment speeds of different modes are heterogeneous, if instructions are issued synchronously without distinction, it is easy to cause oscillations in the adjustment. Therefore, a hierarchical feedback control mechanism is adopted in the conversion process.

[0081] Accordingly, the slow response layer uses the capacity allocation plan as the baseline plan and updates and evaluates it at time points that are integer multiples of its time constant;

[0082] The fast response layer is responsible for making up for the instantaneous capacity gaps in the slow response layer during the arrival process.

[0083] Optionally, a predetermined control law is further designed so that the allocation amount of the fast response layer is driven by the total supply and demand gap, and is limited to a fixed fraction of the gap that the slow response layer has not yet filled.

[0084] Optionally, a structural demand correction mechanism is also introduced, using a mode switching elasticity coefficient to compensate for the elastic passenger flow demand that is transferred due to the temporary intervention of the fast response layer, so that the allocation plan of the slow response layer is not reduced due to the temporary mode switching of passengers.

[0085] Based on this, the output multi-mode dynamic allocation instruction sequence encapsulates the capacity increase / decrease instructions, execution path and expected service recovery time in chronological order, and directly connects to the interface of the operating entity of each transportation mode.

[0086] Another example is that the pre-calibrated probability distribution of response delay, conditional distribution of anomaly intensity, and cumulative distribution of response delay can be obtained through the following pre-construction steps:

[0087] Step 201: Obtain historical operational interruption event records and historical observation time-series data from various data sources within the affected area;

[0088] Accordingly, multi-source heterogeneous traffic data includes at least two of the following data sources:

[0089] The data sources include route replanning request logs from navigation applications, real-time order and vehicle location data from ride-hailing platforms, public transport card transaction records, mobile communication base station signaling data, and real-time text streams containing traffic incident keywords from social media platforms.

[0090] Among them, the historical operation interruption event records correspond to past operation anomaly information extracted from the database archived by the urban traffic management department, including the confirmed occurrence time and geographical impact range of each event.

[0091] Historical observation time series data from various data sources can be used to construct response benchmarks for different types of traffic data sources to different types of sudden events, thereby providing statistical support for likelihood function calculation in the real-time monitoring phase.

[0092] Step 202: Based on the first time window when the historical observations of each data source exceed the daily fluctuation range, and combined with the confirmation time of the corresponding event in the historical operation interruption event record, extract response delay samples, and fit the response delay samples with a log-normal distribution to obtain the response delay probability distribution and the response delay cumulative distribution.

[0093] In the above text, the daily fluctuation range can be determined based on historical observation time series data.

[0094] In this embodiment, the time series within the affected analysis cell is traced back from the historical observation time series data, and the time window in which the observed value of a certain data source first exceeds three times the standard deviation of its daily fluctuation is defined as the activation time.

[0095] Furthermore, the time difference between the activation time and the event confirmation time is calculated and extracted as a response delay sample.

[0096] All historical events are grouped by event type, and the median, first quartile, and third quartile of the response delay samples for each data source under each event type are extracted.

[0097] Based on the fact that the response delay of traffic data sources is naturally positive and exhibits a right-skewed shape, this method uses a log-normal distribution as the fitting model.

[0098] The estimation logic for the parameters of the log-normal distribution can be expressed as follows:

[0099] μ _k_j =ln(m _k_j );

[0100] σ _k_j =(ln(Q3 _k_j )-ln(Q1 _k_j )) / (2×0.6745);

[0101] Where, μ _k_j Let m be the location parameter of the log-normal distribution. _k_j Let σ be the median of the response delay samples for the k-th data source under event type j. _k_j Q3 is the scaling parameter of the log-normal distribution. _k_j Q1 is the corresponding third quartile. _k_j is the corresponding first quartile, ln represents the natural logarithm, and 0.6745 is the 75th percentile value of the standard normal distribution.

[0102] Based on this, the standard distribution function can be used to generate the probability distribution of response delay and the cumulative distribution of response delay for the corresponding event type and data source.

[0103] In some scenarios, for data sources with hard physical lower bounds, such as bus card transaction records constrained by vehicle departure intervals, a three-parameter log-normal distribution can be used instead of a two-parameter log-normal distribution.

[0104] For example, introducing an additional position translation parameter into the model as the physical lower bound time of the response delay and performing an iterative algorithm to obtain this parameter can further reduce the model's fitting error on such discrete and constrained data.

[0105] Step 203: Based on historical observation time series data, calculate the normalized anomaly intensity of each data source in historical events;

[0106] Using the elapsed time after activation as a conditional variable, the normalized anomaly intensity of each data source in historical events is fitted with a gamma distribution to obtain the conditional distribution of anomaly intensity.

[0107] Optionally, a probabilistic model conditioned on activation duration can be established;

[0108] The specific implementation method is to record the normalized anomaly intensity value of the first time window and subsequent time windows after the activation of the data source that has been activated in each historical event.

[0109] The duration after activation is extracted as a condition variable, and the shape and rate parameters of the gamma distribution are solved from the normalized anomaly intensity samples using the maximum likelihood estimation method.

[0110] In the following text, the shape parameter characterizes the underlying bias morphology of the normalized anomaly intensity distribution;

[0111] The rate parameter is set to a function that changes dynamically with the activation duration.

[0112] Furthermore, the rate parameter can be specifically represented by a piecewise linear function to approximate the nonlinear evolution of anomaly intensity over time.

[0113] Accordingly, by substituting the calculated shape parameters and piecewise linear rate parameters into the gamma distribution density function, the final anomaly intensity conditional distribution can be output.

[0114] This step can be used to characterize the probability density of anomaly intensity observations over a given time span, thus providing a parameter matrix for joint likelihood calculation.

[0115] Another example further illustrates the process of identifying sudden changes in demand; namely:

[0116] Step 301: Extract the normalized anomaly intensity and duration of each data source, and distinguish between activated and inactive data source states.

[0117] In this embodiment, a unified spatiotemporal reference multi-source data matrix is ​​read window by window using a rolling time window.

[0118] For each type of data source within each analysis cell, perform an anomaly strength extraction operation.

[0119] Accordingly, the calculation of the normalized anomaly intensity can be as follows:

[0120] The difference between the current observation value and the pre-stored historical baseline for the same time period is calculated, and then divided by the standard deviation of the historical baseline; that is:

[0121] a _k (t)=(v _k (t)-v _k_hist ) / σ _k_hist ;

[0122] Among them, a _k (t) represents the normalized anomaly intensity of the k-th data source at time t, v _k (t) represents the real-time observation value of the k-th data source at time t, v _k_hist Let σ be the baseline mean of the k-th data source over the corresponding historical period. _k_hist Let be the standard deviation of the k-th data source in the corresponding historical period.

[0123] Accordingly, the activation status flags of each data source will be maintained synchronously.

[0124] As an example, the activation determination logic is as follows:

[0125] If the normalized anomaly intensity of a data source exceeds a preset threshold for several consecutive time windows, the data source is determined to be activated, and the activation start time of its first trigger threshold is recorded.

[0126] The difference between the current time and the activation start time is defined as the duration.

[0127] Furthermore, at each monitoring moment, the status, anomaly intensity, and duration of all data sources are encapsulated into an asynchronous response time-series fingerprint feature vector.

[0128] For example, the feature vector has a fixed-length segmented structure, with the normalized anomaly intensity of each data source arranged sequentially in the first half, and 0 filled in the corresponding position for data sources that have not yet been activated.

[0129] The latter half lists the duration of each data source from the activation time to the current time, with inactive data filled with 0. This provides data support for introducing the non-responding probability as diagnostic information.

[0130] Step 302 introduces the non-response probability of the data source that has not yet been activated as a joint penalty term in the inference likelihood function.

[0131] In this scheme, the attributes of demand mutation events are determined using a joint likelihood inference framework based on the asynchronous response time-series fingerprint feature vector and a pre-calibrated set of response delay feature parameters for each data source. This step considers the silent state of slow-response data sources in the early stages of mutation as a likelihood contribution with diagnostic value.

[0132] Accordingly, the calculation of the joint likelihood value can be described as follows:

[0133] L(j,t')=(PI _k∈A(t) (g _k_j (τ _k -t')×h _k_j (a _k (t)|t-τ _k )))×(PI _k∈U(t) (1-G _k_j (tt�)));

[0134] Where L(j,t') is the joint likelihood value of the candidate event type j and the candidate occurrence time t', A(t) is the set of data sources that are currently active, U(t) is the set of data sources that are not yet active, and τ _k For the activation time of the k-th type of data source, g _k_j For the response delay probability density function, h _k_j For the distribution density under abnormal intensity conditions, G _k_j For the cumulative distribution function of response delay, PI is the multiplication operator.

[0135] For data sources that have not yet been activated, their cumulative distribution function of response delay can be used to calculate the probability that the data source is still unresponsive at the current time under the assumed conditions, i.e., 1-G. _k_j (tt'); This constitutes the penalty term of the joint likelihood function.

[0136] For example, if a high-severity event hypothesis is true, the penalty term will have a lower value if the data source that should have responded quickly is still silent at the current moment, thereby reducing the joint likelihood of the false hypothesis.

[0137] Conversely, the joint likelihood value under the true assumption will be relatively more prominent.

[0138] During periods of low data activity, such as late night or early morning, the latency distribution of slow response sources may exhibit heavy-tailed characteristics, naturally reducing the discriminative power of the penalty term. Consequently, it automatically reverts to a regular detection mode that primarily relies on observations from activated data sources. This mechanism prevents the system from introducing additional false alarm risks under extreme external conditions.

[0139] Furthermore, the combination that maximizes the joint likelihood value is selected in the two-dimensional search space, and the triplet of the demand mutation event attribute is output.

[0140] The two-dimensional search space consists of all candidate event types and all candidate occurrence times traced back from the current time.

[0141] As an optional implementation method, the pre-constructed hub space resource occupancy relationship can be determined in the following way:

[0142] Step 401: Obtain the total capacity of various spatial resources within the affected transfer hub, as well as the consumption of each spatial resource by each transportation mode to complete a single standard service.

[0143] Extract the total capacity parameters of various resources from the physical layout database of the affected transfer hubs, such as the total length of bus bays, the maximum number of vehicles that can be accommodated in the curbside passenger drop-off and pick-up areas, and the design width of pedestrian walkways.

[0144] Simultaneously, operational parameters for each transportation mode are obtained to determine the spatial resource consumption of a single standard service. For example, the bay occupancy length and time required for a bus to complete one stop and passenger pick-up / drop-off, the average curb occupancy length required for a ride-hailing vehicle to stop and pick up passengers, and the area space required for a single pickup / drop-off of a shared bicycle.

[0145] Accordingly, spatial resources include, but are not limited to, bus bays, curbside drop-off and pick-up areas, pedestrian walkways, and non-motorized vehicle parking areas.

[0146] Step 402: Calculate the proportion of consumption of each transportation mode to the total capacity, and obtain the resource occupation ratio of each transportation mode on each spatial resource.

[0147] In this embodiment, the consumption is divided by the corresponding total capacity; the resource consumption in different dimensions is converted into a unified dimensionless proportional value; this is beneficial for the consistency of values ​​in subsequent matrix operations.

[0148] Step 403: Based on the resource occupancy ratio of each transportation mode on each spatial resource, the spatial resource occupancy relationship of the hub is constructed.

[0149] Accordingly, all resource occupancy ratios are organized in a matrix structure, and the matrix of patterns and resource occupancy relationships is constructed as an entity representation of the resource occupancy relationships in the hub space.

[0150] In this matrix, rows correspond to different traffic mode indices, and columns correspond to various spatial resource indices;

[0151] The specific element values ​​within the matrix represent the proportion of each mode's resource usage.

[0152] In some embodiments, the actual available capacity of different transportation modes in the affected transfer hub is assessed by combining pre-built hub space resource occupancy relationships, including:

[0153] Optionally, the congestion-sensitive weights of various spatial resources can be obtained.

[0154] Congestion sensitivity weights are used to characterize the service efficiency degradation characteristics of predetermined spatial resources under congestion conditions. In the specific configuration process, the service efficiency of each resource at different utilization levels is calibrated based on measured data extracted from historical operational data. The specific calculation relationship is as follows:

[0155] ω _r =(S _free -S _cong ) / S _free ;

[0156] Where, ω _r S represents the congestion-sensitive weight of spatial resource r. _free S represents the average service efficiency under free-flow conditions. _cong This represents the average service efficiency under congested conditions.

[0157] If the historical database lacks actual service efficiency samples under high congestion conditions, theoretically designed traffic capacity curves can be used for alternative calculations.

[0158] For example, for pedestrian access resources, a standard service level model can be used to estimate the rate of decline, and corresponding congestion-sensitive weights can be set accordingly.

[0159] Optionally, for any two traffic modes, the resource occupancy ratios of the two modes on the same spatial resource are multiplied to calculate the product, and the resulting product is weighted and summed with the congestion sensitivity weight of the corresponding spatial resource to construct a cross-mode spatial interference matrix that reflects the intensity of resource competition between modes.

[0160] Furthermore, the element values ​​in the cross-mode spatial interference matrix are calculated using the following relationship:

[0161] Г _m_n =Σ(O _m_r ×O _n_r ×ω _r );

[0162] Among them, Г _m_n Let O be the cross-modal spatial interference coefficient between traffic mode m and traffic mode n. _m_r O represents the resource occupancy ratio of pattern m to spatial resource r. _n_r ω represents the resource occupancy ratio of pattern n to spatial resource r. _r Let Σ be the congestion-sensitive weight of spatial resource r, and let Σ be the summation operation over all shared spatial resource items.

[0163] Through calculation, a dimensionless matrix is ​​obtained in which the diagonal elements reflect the self-interference intensity of the mode and the off-diagonal elements reflect the cross-interference intensity between modes; the unit demand mutual interference coefficient generated by different traffic modes through shared resources is directly quantified.

[0164] Optionally, the actual available capacity of different transportation modes can be evaluated based on the cross-modal spatial disturbance matrix and demand mutation event attributes.

[0165] In the following text, data convergence can represent the current demand of each transportation mode, where the rate of change between two adjacent iterations is lower than a preset convergence threshold;

[0166] Based on the cross-modal spatial disturbance matrix, the actual available capacity of different transportation modes is evaluated, specifically including the following two-stage iterative process executed alternately until data convergence:

[0167] On the one hand, the current demand of each transportation mode is determined based on the attributes of demand mutation events. The current demand is combined with the cross-mode spatial interference matrix to calculate the capacity attenuation coefficient of each transportation mode. The pre-configured rated capacity of each mode is multiplied by the capacity attenuation coefficient to obtain the actual available capacity.

[0168] In this embodiment, the total overflow demand into the hub is extracted based on the severity level of the mutation event, and the current demand of each traffic mode is set according to the initial sharing ratio.

[0169] Further, the following calculation is performed to determine the capacity attenuation coefficient:

[0170] α _m =max(0,1-Σ(Г _m_n ×d _n / c _n ));

[0171] Where, α _m Let Γ be the capacity attenuation coefficient for traffic mode m. _m_n Let be the cross-mode spatial interference coefficient between mode m and mode n, Σ be the summation operation over all traffic modes, max be the maximum value function, and d be the maximum value function. _n Let c be the current demand for transportation mode n, i.e., the current demand capacity. _n The pre-configured rated capacity for traffic mode n.

[0172] On the other hand, the expected waiting time for each mode of transportation is estimated based on the actual available capacity and current demand, and the current demand for each mode of transportation is updated based on the difference in expected waiting time.

[0173] As the actual available capacity of each mode decreases due to spatial disruptions, passenger queue times will change.

[0174] The new expected waiting time is calculated using a queuing theory model, and this waiting time is used as an input parameter to trigger subsequent demand transfer and update mechanisms.

[0175] In other scenarios, queuing theory models can also be used to estimate the expected waiting time for each mode of transportation.

[0176] Accordingly, for each mode of transportation, the ratio of its current demand to its actual available capacity is used as the service intensity, which is then substituted into the M / M / 1 queuing model or other queuing theory models selected based on the actual service process to calculate the expected waiting time.

[0177] In some embodiments, updating the current demand for each transportation mode based on the difference in expected waiting times specifically includes:

[0178] Optionally, the expected waiting time of each transportation mode is substituted into the pre-built Logit mode selection model to recalculate the demand sharing ratio of each transportation mode in the total overflow demand.

[0179] The total overflow demand can be determined based on the attributes of demand mutation events.

[0180] In this embodiment, the pre-built Logit mode selection model uses expected waiting time as the core utility variable. The calculated expected waiting time for each mode is substituted into the model's utility function to calculate the probability of each traffic mode being selected, thereby determining the updated demand sharing ratio.

[0181] Optionally, the current demand for each mode of transportation can be updated by combining the total overflow demand and demand sharing ratio at the transfer hub.

[0182] The system multiplies the fixed total overflow demand by the demand sharing ratio to obtain a new demand vector distribution. This updated current demand will be re-entered to initiate a new round of capacity attenuation calculations.

[0183] On the other hand, after the iteration converges, the final actual available capacity is output.

[0184] Convergence is determined between two adjacent iterations; the specific convergence metric is calculated based on the rate of change of the maximum component of the demand vector, i.e.:

[0185] Δd=max(abs(d _n_new -d _n_old ) / d _n_old );

[0186] Where Δd is the rate of change of the maximum component of the demand vector, abs is the absolute value function, and d _n_new Let d be the current demand for the current traffic mode n in the current round. _n_old This represents the current demand for the previous traffic mode n.

[0187] The iteration is considered to have converged when the calculated maximum rate of change of the component continuously decreases and falls below the preset convergence threshold. After the iteration terminates, the output state vector serves as the actual available capacity for each mode of the hub.

[0188] As an optional implementation, taking into account the self-interference effect of dispatched vehicles, a capacity dispatching scheme for maximizing demand dispersal is generated, including:

[0189] Step 501: Based on the attributes of the demand mutation event, determine the evacuation superposition flow of each road segment of the affected road network;

[0190] Vehicles on each candidate transfer route leading to the affected transfer hub will be converted into equivalent superimposed traffic injected into the road network within a predetermined time slice.

[0191] Alternatively, based on the pre-acquired information on the topology of the affected road network and the distribution of stations, each candidate transfer path leading to the affected transfer hub is determined; the transfer vehicles on each candidate transfer path are converted into equivalent superimposed traffic injected into the road network within the corresponding time slice.

[0192] Accordingly, the time-varying spatial diffusion pattern of evacuation passenger flow is extracted from the spatiotemporal distribution feature database of outbound passenger flow in similar historical events. Based on the severity level in the identified demand mutation event attributes, the diffusion pattern is linearly scaled and superimposed on the normal traffic flow of each road segment to obtain the background evacuation traffic flow of the road network after the event.

[0193] Simultaneously, the allocation scheme vectors on each candidate allocation path are obtained. Since allocation schemes are typically measured in units of vehicle numbers, a conversion factor for the equivalent passenger car volume of large buses is introduced to maintain consistency with the background traffic volume. The calculation process for the equivalent superimposed traffic volume is described below:

[0194] v _p_equiv =x _p ×η / Δt;

[0195] Among them, v _p_equiv Inject the equivalent superimposed traffic of the road network into path p, x _p Let η be the number of vehicles transferred on path p, η be the car equivalent conversion factor, and Δt be the time discretization granularity corresponding to the transfer operation.

[0196] Step 502: Combine the pre-acquired normal traffic flow, evacuation superimposed traffic flow, and equivalent superimposed traffic flow of the road segment to construct the allocation time correlation relationship including self-interference term;

[0197] Among them, the travel time of any candidate allocation route is dynamically affected by the equivalent superimposed flow of other candidate allocation routes on the same travel segment, so that the route travel time exhibits an endogenous nonlinear delay as the total allocation volume increases.

[0198] "Same travel segment" can refer to the road segment that is traveled together.

[0199] In this embodiment, the road network load resulting from the superposition of multiple traffic sources is mapped to driving impedance using a road resistance function. The constructed driving time model including self-interference terms considers the endogenous feedback of allocation decisions on the road network state. For any road segment in the road network, the driving time including self-interference terms is calculated as follows:

[0200] t _a (x)=t _a0 ×(1+μ _BPR ×((v _a0 +v _a_evac +Σ p (x _p ×η / Δt) / C _a ) γ );

[0201] Among them, t _a (x) represents the travel time of road segment a under allocation scheme x, including self-interference, t. _a0 v represents the free-flow travel time of the road segment. _a0 For the normal traffic flow of the road section, v _a_evac C represents the evacuation and superimposed flow rate for road segment a. _a Let μ be the traffic capacity of road segment a. _BPR γ and x are preset parameters for the path resistance function. _p Let η be the number of vehicles transferred on path p, η be the car equivalent conversion factor, and Δt be the time discretization granularity corresponding to the transfer operation.

[0202] The total travel time is obtained by summing the travel times of all road segments along the route.

[0203] It depends not only on its own allocation volume, but also on the allocation volume of all other routes that share the same road segment with it;

[0204] This model is used to describe the phenomenon of self-interference changes when the scale of traffic allocation gradually increases, the addition of new vehicles aggravates road congestion, and the arrival time of vehicles is later than the service cutoff time.

[0205] Furthermore, a reverse contraction strategy can be adopted to generate a capacity allocation plan, namely:

[0206] The upper limit of the full-load allocation scheme is to allocate the available vehicles of each station in advance to the shortest path.

[0207] Based on the time correlation of allocation with self-interference term, the marginal contribution of each active path is evaluated by first-order sensitivity analysis. The marginal contribution is determined by the difference between the direct capacity benefit brought by retaining the allocated vehicles of the path and the indirect road network congestion loss caused by it.

[0208] The allocation of vehicles on paths with negative marginal contributions is reduced round by round until the marginal contributions of all retained paths are non-negative, and a preliminary capacity allocation plan is output.

[0209] The shortest path can be determined based on the combined evacuation traffic flow and the normal traffic flow of the road segment.

[0210] Furthermore, this application proposes a reverse contraction optimization strategy. This strategy starts from the initial state of most severe congestion and optimizes the solution by gradually identifying and eliminating invalid paths that lead to a decrease in overall efficiency.

[0211] In practice, first-order sensitivity analysis is used to calculate the marginal contribution of each active path. The calculation logic for the marginal contribution is described as follows:

[0212] MC _p =Δ p_dir -Δq _ind ;

[0213] Among them, MC _p Δq represents the marginal contribution of active path p. _ind Let Δp be the total indirect road network improvement benefit resulting from the reduction of travel time on other shared road segments after reducing the number of vehicles on path p. _dir The reduction in the number of vehicles on route p directly leads to a decrease in the allocated transport capacity, resulting in a loss of transport capacity.

[0214] As an example, in each iteration, all active paths with a transfer amount greater than or equal to 1 are evaluated.

[0215] If the marginal contribution of a certain path is negative, it means that the indirect losses caused by the vehicle in exacerbating congestion have exceeded its direct capacity contribution, that is, the presence of the vehicle reduces the overall dispatch efficiency.

[0216] Select the path with the largest absolute value of the negative marginal contribution for vehicle reduction;

[0217] When the marginal contribution of all remaining paths is non-negative, it means that the direct capacity benefits of retaining these vehicles are sufficient to offset their congestion costs. The algorithm then terminates and outputs a preliminary capacity allocation plan.

[0218] Optionally, the time limit determination in the marginal contribution calculation can be approximated using a sigmoid function, i.e., a smoothing function.

[0219] The smoothing parameter can be dynamically set based on the standard deviation of the path travel time obtained in the first round of calculation to ensure the effectiveness of the sensitivity gradient in the numerical search process and avoid the gradient vanishing problem caused by the step function.

[0220] This ensures that the system avoids self-interference saturation points within the calculation time limit and outputs the allocation plan.

[0221] In other possible scenarios, the marginal contribution can also be determined by the difference between the direct capacity gains from reserving vehicles along the route and the indirect network congestion losses it causes.

[0222] On the one hand, after outputting the initial capacity allocation plan, it also includes a consistency verification step between allocation and hub balancing, specifically:

[0223] Step 601: Convert the vehicles expected to arrive at the hub in the preliminary capacity allocation plan into additional resource occupancy of the hub's space resources.

[0224] Accordingly, the number of various types of vehicles to be dispatched from each station to the affected transfer hubs was extracted from the preliminary capacity allocation plan. Considering that the dispatched vehicles are mainly large buses or emergency shuttle buses, the vehicles need to stop to pick up and drop off passengers after arriving at the hub, which will consume the parking spaces of the bus bay, the length of the roadside passenger pick-up and drop-off area, and the space of the internal passageway.

[0225] By utilizing the pre-acquired data on the consumption of various spatial resources for a single standard service across different transportation modes, the extracted number of dispatched vehicles is mapped to specific resource consumption values.

[0226] Based on this, the proportion of this resource consumption value to the total capacity of this type of spatial resource is calculated, thereby obtaining the additional resource usage. This step realizes the equivalent transformation of scheduling instructions in the spatial dimension.

[0227] Step 602: Feedback the additional resource usage to the step of assessing the actual available capacity of different transportation modes in the affected transfer hub for reassessment, and obtain the updated actual available capacity;

[0228] Alternatively, the additional resource usage can be fed back to the capacity assessment stage to recalculate the available capacity of each mode of transportation within the affected transfer hub.

[0229] In this scheme, the calculated additional resource occupancy will be added to the original hub space resource occupancy relationship.

[0230] That is, by adding the additional resource usage amount to the resource usage ratio of the corresponding dimension of the allocated buses, a revised pattern and resource usage relationship matrix is ​​formed.

[0231] Based on the corrected matrix, the construction operation of the cross-mode spatial interference matrix is ​​re-executed, that is, the product of the occupancy ratios of each item and the weighted sum of the congestion sensitivity weights are recalculated.

[0232] Furthermore, the updated cross-mode spatial interference matrix is ​​substituted into the capacity balancing solution module to trigger a new round of capacity attenuation calculation and demand redistribution iteration process.

[0233] After the fixed-point iteration process converges again, the capacity supply of each mode in the converged state is extracted as the updated actual available capacity.

[0234] Step 603: Compare the deviation of the actual available capacity before and after the update. If the deviation exceeds the preset threshold, the reverse contraction strategy is re-executed with the updated actual available capacity as a constraint, and the final capacity allocation plan is output.

[0235] If the deviation does not exceed the preset threshold, the preliminary capacity allocation plan will be output as the final capacity allocation plan.

[0236] The following section compares the actual available capacity obtained from the initial assessment with the updated actual available capacity obtained in this round of verification, and calculates the maximum relative deviation rate.

[0237] For example, D _max =max(abs(E _new -E _old ) / E _old );

[0238] Among them, D _max The maximum relative deviation rate is given by E, where abs is the absolute value function. _new E represents the updated actual available capacity for a certain transportation mode. _old This represents the actual available capacity of this transportation mode in the initial assessment.

[0239] Based on this, determine whether the calculated maximum relative deviation rate is greater than a preset threshold;

[0240] If the maximum relative deviation rate is not greater than the preset threshold, it means that the spatial interference caused to the hub by the preliminary capacity allocation plan is within the tolerable range, and the preliminary capacity allocation plan is directly confirmed as the final capacity allocation plan.

[0241] If the maximum relative deviation rate is greater than the preset threshold, it indicates that the influx of a large number of transferred vehicles has caused a serious secondary decline in the hub's carrying capacity, and the original plan is no longer feasible.

[0242] At this point, the updated actual available capacity replaces the upper limit of the receiving capacity constraint in the original optimization model, and the preliminary capacity allocation scheme is used as the initial solution to re-trigger the reverse contraction strategy.

[0243] Since the initial solution of the second run of the reverse contraction algorithm is close to the optimal solution region, only a small number of correction iterations are needed to converge. Thus, under the premise of meeting the real-time calculation requirements, the final capacity allocation scheme that is fully compatible with the dynamic physical carrying capacity limit of the hub is output.

[0244] To address the capacity oscillation problem caused by inconsistent response speeds among different modes in a mixed transportation system, one possible implementation involves utilizing the response time constants of each transportation mode to transform the capacity allocation scheme into a multi-mode dynamic allocation command sequence, specifically including:

[0245] Step 701: Divide each traffic mode into a fast response layer and a slow response layer according to the response time constant;

[0246] The response time constant represents the time delay between the system issuing a dispatch instruction for the planned transportation mode and the arrival of the transport entity in the designated area, generating actual supply capacity.

[0247] Optionally, a time separation boundary value can be set. For example, this time separation boundary value can be set to 10 minutes, which is just an example and not the only limitation; other examples are also possible.

[0248] For traffic modes whose response time constant is less than or equal to the boundary value, they are classified as fast response layers. The specific traffic modes can be ride-hailing or shared bicycles.

[0249] For traffic modes with a response time constant greater than this boundary value, they are classified as slow response layers. The specific traffic modes can be regular buses or emergency shuttle buses.

[0250] In this logic, the hybrid capacity pool is decoupled into two control dimensions with different dynamic adjustment characteristics.

[0251] Step 702: Use the capacity allocation plan as the baseline plan for the slow response layer;

[0252] Furthermore, the global capacity allocation scheme obtained from the optimization algorithm is directly mapped to the slow response layer for execution. Since the scheduling in slow response mode involves the physical movement of large vehicles and depot departure instructions, its plan remains locked within a given scheduling period. Therefore, the slow response layer, as the main body of capacity, is not affected by short-term high-frequency data fluctuations, which is beneficial to the stable operation of the system.

[0253] Step 703: Obtain the total demand gap of the affected transfer hubs in real time at the current moment, as well as the actual unmet demand gap of the slow response layer;

[0254] Alternatively, the total demand gap can be determined based on real-time collected current passenger flow data of affected transfer hubs and currently available capacity;

[0255] The actual unmet demand in the slow response layer can be determined based on the difference between the planned available capacity at the current moment and the actual available capacity of slow response vehicles obtained in real time, as shown in the baseline plan for the slow response layer.

[0256] The following calculation, using passenger flow counting sensors and queue monitoring equipment deployed at transfer hubs, calculates the difference between the current total number of stranded passengers and the current total available capacity, thus obtaining the total demand gap;

[0257] Based on the real-time location coordinates of slow-response vehicles and the road network impedance model, the estimated arrival time of each slow-response vehicle is calculated. The difference between the planned capacity to arrive at the current time in the slow-response layer baseline plan and the actual capacity that has arrived is calculated to obtain the actual unmet demand of the slow-response layer.

[0258] Step 704: Compare the total demand gap with the actual unmet demand gap limited by a preset ratio upper limit, take the smaller value of the two (total demand gap and actual unmet demand gap) and determine the fast response layer transition supplement amount under the non-negative boundary limit condition;

[0259] By limiting the intervention depth of the fast response layer through control laws, it is made to only play a supplementary role in the transition phase.

[0260] In other scenarios, the total demand gap can be treated with a proportional gain factor.

[0261] For example, the transition supplement amount of the fast response layer can be calculated using the following formula:

[0262] u _fast =max(0,min(K _f ×gap _total ,λ×gap _slow ));

[0263] Among them, u _fast The fast response layer transition supplement amount, max is the maximum value operation, min is the minimum value operation, K _f That is, the proportional gain coefficient, gap _total That is, the total demand gap, where λ is the preset upper limit coefficient, gap. _slow This refers to the actual gap in the slow response layer.

[0264] In the above, the operation min ensures that the allocation amount of the fast response layer is subject to dual constraints, that is, it will not exceed the total gap requirement, nor will it exceed the set proportion of the unfulfilled amount of the slow response layer.

[0265] The operation max constitutes the non-negative boundary limiting condition.

[0266] Therefore, when the slow response layer arrives abnormally early, resulting in a negative gap in actual arrival, this limiting condition will intercept the negative output command generated by the calculation, preventing the system from issuing incorrect capacity withdrawal actions to the fast response layer, which is beneficial to maintaining the boundary security of control commands.

[0267] The value of the proportional gain coefficient must ensure the stability of the control loop.

[0268] It can be determined by performing closed-loop simulation on historical data of typical sudden operating conditions, and selecting the value that prevents overshoot in the system's capacity supply curve and has the shortest adjustment time.

[0269] Step 705: Combine the slow response layer baseline plan with the fast response layer transitional supplement to generate a multi-mode dynamic allocation instruction sequence;

[0270] The slow-response layer instructions belonging to static scheduling and the fast-response layer instructions belonging to dynamic compensation are combined and aligned on the time axis to form a complete instruction set covering all relevant traffic modes.

[0271] In other embodiments, step 705 further includes structural requirement adjustments to incorporate passenger mode switching flexibility, namely:

[0272] Step 7051: Obtain the pre-configured mode switching elasticity coefficient representing the passenger transfer triggered by changes in the fast response mode capacity;

[0273] Because the temporary increase in capacity in the fast response layer will shorten queuing time, some passengers who were originally waiting in the slow response mode will change their travel decisions and switch to the fast response mode.

[0274] In the above text, the mode switching elasticity coefficient reflects the number of passengers originally belonging to the slow response mode that are absorbed for each additional unit of fast response capacity.

[0275] This coefficient can be obtained by calculating the partial derivative of the historical pattern selection probability; it can then be pre-configured offline in the system parameter library.

[0276] Step 7052: Based on the fast response layer transition replenishment amount and mode switching elasticity coefficient, calculate the elastic demand replenishment amount temporarily transferred from the slow response layer to the fast response layer;

[0277] Alternatively, the elastic demand replenishment amount can be calculated using the following expression:

[0278] d _elastic =β×(u _fast -u _base );

[0279] Where, d _elastic β is the elastic demand replenishment amount, and u is the mode switching elasticity coefficient. _fast For the transition supplement of the fast response layer, u _base This is the basic allocation level for the fast response layer when the system is in a normal state without sudden events.

[0280] It should be understood that this step calculates the amount of derived passenger flow transfers caused by temporary scheduling behavior of the system.

[0281] Step 7053: The elastic demand replenishment amount is superimposed with the slow response layer observation demand collected in real time to obtain the slow response layer structural demand.

[0282] In this solution, the number of people waiting in the slow response layer is currently collected by the physical sensors, and this number is added to the elastic demand replenishment amount; this restores the potential passenger flow demand that is masked by the siphon effect of the fast response mode.

[0283] Step 7054: Use the structural demand of the slow response layer to replace the actual observed demand as the capacity assessment input when updating the baseline plan of the slow response layer.

[0284] In subsequent time windows, the apparent queuing data will no longer be used; instead, the structural demand of the slow response layer will be used to verify whether the capacity is sufficient.

[0285] This operation avoids the risk of secondary congestion and disruption caused by the misleading reduction of slow-response capacity and the subsequent withdrawal of fast-response capacity.

[0286] In some embodiments, the multi-mode dynamic allocation instruction sequence is encapsulated into an execution interface format for each mode of transportation operation entity;

[0287] The execution interface formats, in chronological order, include:

[0288] The capacity increase / decrease instructions issued to the predetermined transportation mode can be used to define specific order dispatch requirements or increase / decrease the number of departures.

[0289] The execution path or target area information corresponding to the capacity increase / decrease command can be used to define the spatial driving range of the vehicle or the coordinates of the vertex of the service polygon.

[0290] The anticipated service recovery time calculated based on the capacity allocation plan can provide time boundary information for operators to arrange capacity turnover.

[0291] In this solution, the generated multi-mode dynamic allocation instruction sequence is converted into a standardized message format that can be parsed by various operation and scheduling systems;

[0292] The message structure includes action variables and spatial constraint variables;

[0293] In one exemplary application scenario, suppose a subway line in a city experiences a section of operation disruption due to equipment failure, and the affected hubs experience a large number of stranded passengers within several minutes after the event.

[0294] By employing the method of this invention, joint likelihood inference of asynchronous response timing characteristics is performed, and event attribute determination is completed in the early stage when the path replanning quantity of the navigation application first shows anomalies. Compared with the method of waiting for multiple data sources to trigger the threshold simultaneously, the warning time is advanced.

[0295] In the capacity assessment phase, an iterative equilibrium calculation of the cross-mode spatial interference matrix is ​​performed. The actual available capacity of each mode obtained from the assessment is lower than the result of independently superimposing the rated capacity, reflecting the carrying capacity limit under the hub spatial constraints.

[0296] In the allocation phase, the reverse contraction strategy outputs an allocation plan after considering the self-interference effect of the allocated vehicles, thus avoiding secondary congestion on the road network caused by excessive allocation.

[0297] In the instruction generation stage, the hierarchical feedback control mechanism and the flexible demand replenishment strategy suppress the oscillation of transport capacity supply during the alternation of fast and slow response modes.

[0298] It is understandable that the specific advance warning range, capacity assessment deviation, and evacuation efficiency improvement may vary depending on the city size, hub layout, event type, and data source access conditions.

[0299] In this scheme, asynchronous response timing features are introduced, and the inactive state of slow-response data sources is used as a penalty term in the joint likelihood function. This breaks the logic that multi-source data must wait for consensus, and advances the warning time of sudden events from the abnormal outbreak period of most data sources to the initial outbreak period of a single source, thus achieving high-confidence early judgment.

[0300] In this method, a cross-modal spatial interference matrix is ​​constructed, which quantifies the competition between different transport capacities for spaces such as harbors and road edges into cross-attenuation coefficients. Through supply and demand equilibrium iteration, the transport capacity assessment results are made to fit the hub space limit.

[0301] In the capacity allocation stage, in response to the endogenous nonlinear delay caused by the influx of a large number of rescue vehicles exacerbating road network congestion, a reverse contraction optimization strategy can be adopted. Starting from the upper limit of full-load allocation, the path vehicles with negative marginal contribution are reduced round by round. This avoids the trial of forward search, locates the self-interference saturation point, eliminates the spatiotemporal disconnect in the closed-loop verification, and ensures the overall evacuation efficiency is maximized.

[0302] Based on this, in order to eliminate the overcapacity and repeated fluctuations caused by the different response speeds of multi-mode scheduling, a hierarchical feedback control mechanism and a flexible demand replenishment strategy were designed. The fast-response capacity is used as a transitional compensation within the limited boundary, and the implicit demand caused by the temporary switching of passenger flow is actively superimposed, so as to ensure the transition of the multi-mode mixed capacity supply system under the sudden change.

[0303] The order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments.

[0304] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0305] The above description is only an optional implementation of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for dynamic allocation of integrated transportation capacity based on multi-source data fusion, characterized in that, include: Acquire multi-source heterogeneous traffic data and perform unified spatiotemporal benchmark processing; Extract asynchronous response time-series features from various data sources in multi-source heterogeneous traffic data, and identify the attributes of demand mutation events based on these asynchronous response time-series features; Based on the attributes of demand mutation events and the relationship of hub space resource occupancy, assess the actual available capacity of different transportation modes in the affected transfer hubs; Taking into account the self-interference effect of the allocated vehicles, a capacity allocation scheme is generated to maximize the demand dispersal volume. By utilizing the response time constants of each transportation mode, the capacity allocation plan is transformed into a multi-mode dynamic allocation instruction sequence and output.

2. The method of claim 1, wherein, Extracting asynchronous response time-series features from various data sources in multi-source heterogeneous traffic data, including: The normalized anomaly intensity of each data source is determined by comparing the observations of multi-source heterogeneous traffic data within the current time window with the pre-stored historical benchmark. Based on the number of time windows in which the normalized anomaly intensity continuously exceeds the threshold, the activation status of each data source is determined, and the duration of the data source in the activated state is recorded. The normalized anomaly intensity, data source activation state, and duration are combined to construct the asynchronous response timing feature.

3. The method of claim 2, wherein, The attributes of demand mutation events are identified based on the timing characteristics of asynchronous responses. Specifically, this identification is performed using a joint likelihood inference method, including: For candidate event types and occurrence times, based on the response delay probability distribution and anomaly intensity conditional distribution, calculate the first conditional probability of a data source in an activated state observing normalized anomaly intensity over a given duration. Based on the cumulative distribution of response delay, calculate the second conditional probability that a data source in an inactive state will not respond within a corresponding duration. The joint likelihood value is obtained by multiplying the first conditional probabilities of all activated data sources with the second conditional probabilities of all unactivated data sources. Select the candidate combination that maximizes the joint likelihood value, and use its corresponding event type identifier, estimated occurrence time, and severity level as the attributes of the demand mutation event.

4. The method of claim 3, wherein, The probability distribution of response delay, the conditional distribution of anomaly intensity, and the cumulative distribution of response delay are obtained through the following pre-construction steps: Acquire historical operational disruption event records and historical observation time-series data from various data sources within the affected area; Response delay samples are extracted based on the first time window in which the historical observations of each data source exceed the daily fluctuation range. Log-normal distribution is fitted to the response delay samples to obtain the probability distribution and cumulative distribution of response delay. Using the elapsed time after activation as a conditional variable, the normalized anomaly intensity of each data source in historical events is fitted with a gamma distribution to obtain the conditional distribution of anomaly intensity.

5. The method of claim 1, wherein, The allocation of hub space resources is determined in the following ways: Obtain the total capacity of various spatial resources within the affected transfer hub, as well as the consumption of various spatial resources by each transportation mode to complete a single standard service. Calculate the proportion of consumption of each transportation mode to the total capacity to obtain the resource occupation ratio of each transportation mode for each spatial resource. Based on the resource occupancy ratio of each transportation mode on various spatial resources, the spatial resource occupancy relationship of the hub is constructed.

6. The method of claim 1, wherein, Based on the spatial resource occupancy relationship of the hub, assess the actual available capacity of different transportation modes in the affected transfer hubs, including: Obtain the congestion-sensitive weights of various spatial resources in their pre-configuration; For any two traffic modes, the resource occupancy ratios of the two modes on the same spatial resource are multiplied to calculate the product. The resulting product is then weighted and summed with the congestion sensitivity weight of the corresponding spatial resource to construct a cross-mode spatial interference matrix that reflects the intensity of resource competition between modes. Based on the cross-modal spatial interference matrix, the actual available capacity of different transportation modes is evaluated.

7. The method of claim 1, wherein, Taking into account the self-interference effect of dispatched vehicles, the capacity dispatch plan generated to maximize demand evacuation includes flow analysis, specifically: Based on the attributes of demand mutation events, determine the evacuation superposition flow of each road segment in the affected road network; Vehicles on each candidate transfer route leading to the affected transfer hub will be converted into equivalent superimposed traffic injected into the road network within a predetermined time slice.

8. The method of claim 1, wherein, By utilizing the response time constants of each transportation mode, the capacity allocation plan is transformed into a multi-mode dynamic allocation instruction sequence, specifically including: Based on the response time constant, each traffic mode is divided into a fast response layer and a slow response layer; The capacity allocation plan is used as the baseline plan for the slow response layer. Obtain the total demand gap of the affected transfer hubs in real time at the current moment, as well as the actual unmet demand gap of the slow response layer; Compare the total demand gap with the actual unmet demand gap limited by a preset ratio upper limit, take the smaller of the two values, and determine the fast response layer transition supplement amount under non-negative boundary limiting conditions; By combining the baseline plan of the slow response layer with the transitional supplement of the fast response layer, a multi-mode dynamic allocation instruction sequence is generated.

9. The method of claim 1, wherein, Multi-source heterogeneous traffic data includes at least two of the following data sources: The data sources include route replanning request logs from navigation applications, real-time order and vehicle location data from ride-hailing platforms, public transport card transaction records, mobile communication base station signaling data, and real-time text streams containing traffic incident keywords from social media platforms.

10. The method of claim 1, wherein, The multi-mode dynamic allocation instruction sequence is encapsulated into an execution interface format for operators of various transportation modes; The execution interface format includes, in chronological order: the capacity increase / decrease instruction issued to the predetermined transportation mode, the execution path or target area information corresponding to the capacity increase / decrease instruction, and the expected service recovery time.

Images