Line intelligent optimization system based on traffic big data
By predicting travel time and monitoring deviations in real time using multi-source data, an antifragile cost function is constructed, which solves the problem of insufficient vulnerability assessment of routes in traditional solutions, realizes robust optimization of intelligent transportation routes, and improves the safety and punctuality of escort missions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NINGBO XINYI SMART TECH CO LTD
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-30
AI Technical Summary
Existing traffic big data route planning solutions rely heavily on prediction models, which may lead to systematic deviations between predicted and actual travel times. They also cannot quantify the degree to which routes depend on the uncertainty of prediction models, resulting in reduced on-time performance and safety.
By predicting road network travel time using multi-source heterogeneous traffic data, generating baseline routes by combining rigid constraints of escort missions, calculating prediction dependence and total vulnerability index, monitoring prediction deviations in real time, constructing an antifragile cost function, and replanning suboptimal robust paths.
It enables quantitative assessment of the vulnerability of the baseline line, accurately identifies resonance zones, automatically balances efficiency and safety, and improves the safety and punctuality of escort missions.
Smart Images
Figure CN121936689B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation and big data technology applications, specifically to a route intelligent optimization system based on traffic big data. Background Technology
[0002] In the field of route planning based on traffic big data, especially for tasks with rigid constraints such as escorting passengers, existing solutions focus on planning the optimal baseline route based on predicted travel time. However, such planning is highly dependent on traffic prediction models, which themselves have inherent uncertainties. This uncertainty can lead to systematic deviations between predicted and actual travel times, i.e., algorithmic resonance, and traditional solutions cannot quantify the degree of dependence or vulnerability of a route to this uncertainty. When the vulnerability of the optimal route is too high, or when real-time traffic conditions deviate significantly from the prediction, adhering to the route will reduce the on-time performance and safety of the task. Therefore, how to assess and avoid highly vulnerable routes during planning, monitor the risk of prediction failure in real time during execution, and construct a decision-making closed loop that can proactively select suboptimal and robust routes is an urgent technical problem to be solved. Summary of the Invention
[0003] To address the aforementioned technical problems, this invention provides a route intelligent optimization system based on traffic big data. Specifically, the technical solution of this invention includes:
[0004] The data acquisition module is used to acquire multi-source heterogeneous traffic data and rigid constraints of the escort mission; among which, the rigid constraints of the escort mission include the mission security level and service time window.
[0005] The first processing module is used to predict the travel time of each road segment in the road network based on multi-source heterogeneous traffic data, and to plan and generate a baseline route by combining the predicted travel time with the rigid constraints of the escort mission.
[0006] The second processing module is used to calculate the predicted dependency of each segment in the baseline route, and combine the predicted dependency, the preset topological complexity and the task security level to calculate the total vulnerability index of the baseline route.
[0007] The third processing module is used to obtain the actual travel time of the road segment where the vehicle is located, compare the actual travel time with the predicted travel time, and calculate the prediction deviation factor. The third processing module is also used to determine whether the prediction deviation factor continuously exceeds the preset resonance threshold within the preset time threshold. If it continuously exceeds the threshold, the resonance area is identified and marked.
[0008] The route correction module is used to determine whether the total vulnerability index exceeds the preset vulnerability tolerance threshold and whether the baseline route geographically overlaps with the resonance area. If the total vulnerability index exceeds the vulnerability tolerance threshold or the baseline route geographically overlaps with the resonance area, an antifragile cost function is constructed, and a suboptimal robust path is replanned and generated based on the antifragile cost function.
[0009] Preferably, the first processing module predicts the travel time, including:
[0010] Based on multi-source heterogeneous traffic data, we determine the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight.
[0011] The predicted travel time is calculated based on the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight.
[0012] Preferably, the first processing module plans and generates a baseline route, including:
[0013] The predicted travel time is used as the road segment cost, and the service time window is used as a constraint.
[0014] Invoke the vehicle routing problem solver with time windows to generate the baseline route with the minimum total predicted travel time cost.
[0015] Preferably, the second processing module calculates the predicted dependency, including:
[0016] The system invokes the real-time congestion factor, predicted congestion factor, real-time data weight, and predicted data weight determined by the first processing module.
[0017] Based on the above factors and weights, the total congestion gain factor and the congestion gain contributed by the predicted congestion factor are determined.
[0018] The ratio of the congestion gain contributed by the predicted congestion factor to the total congestion gain factor is calculated to obtain the prediction dependence.
[0019] Preferably, the second processing module calculates the total vulnerability index, including:
[0020] The comprehensive vulnerability score of each segment in the baseline route is determined based on the product of prediction dependency, preset topological complexity, and task security level.
[0021] Obtain the comprehensive vulnerability score of the most vulnerable section of the baseline route;
[0022] Obtain the total accumulated vulnerabilities of the baseline line;
[0023] Based on the score of the most vulnerable road segment and the cumulative sum of vulnerabilities, the total vulnerability index is determined through nonlinear amplification.
[0024] Preferably, the third processing module identifies and marks the resonance region, including:
[0025] Continuously aggregate prediction bias factors from multiple road segments within a specific geographical area;
[0026] Calculate the average prediction bias factor for a specific geographical region;
[0027] Determine whether the average prediction deviation factor continuously exceeds the resonance threshold within a preset time threshold.
[0028] If so, the specific geographical area will be marked as a resonance zone;
[0029] If not, then a specific geographical area will not be marked as a resonance zone.
[0030] Preferably, the antifragile cost function is a weighted multi-objective optimization function, and the antifragile cost function includes:
[0031] Total predicted travel time cost of the route;
[0032] Vulnerability time penalty is determined based on the total vulnerability index and the vulnerability time penalty coefficient;
[0033] And a penalty for the real-time resonance zone.
[0034] Preferably, the real-time resonance region penalty term includes:
[0035] Determine whether the road segment is located in the identified resonance zone;
[0036] If the road segment is located in the resonance zone, a preset maximum time penalty value is applied to the road segment;
[0037] If the road segment is not located in the resonance zone, the real-time resonance zone penalty is zero.
[0038] Preferably, the route correction module re-plans and generates a suboptimal robust path, including:
[0039] Minimize the antifragile cost function to determine the suboptimal robust path;
[0040] If the correction is triggered because the total vulnerability index exceeds the vulnerability tolerance threshold, then the suboptimal robust path will be used as the final execution plan.
[0041] If the correction is triggered by coinciding with the resonance region, the suboptimal robust path will be issued as a dynamic optimization instruction.
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] 1. This system calculates the prediction dependency and combines it with preset topology complexity and task security level to achieve a quantitative assessment of the inherent vulnerability of the baseline line, solving the problem that traditional solutions cannot quantify the degree of dependence of the line on the uncertainty of the prediction model.
[0044] 2. By comparing the actual travel time with the predicted travel time in real time and introducing dual threshold judgments in time and space dimensions, this system can accurately identify the resonance zone where the prediction model fails systematically rather than accidentally, thus solving the problem that traditional solutions cannot detect the silent failure of the prediction model.
[0045] 3. This system constructs an antifragile cost function, which explicitly quantifies the invisible vulnerability risk and the real-time resonance zone risk into equivalent time costs, enabling the system to automatically balance efficiency and security when making decisions, and realizing an intelligent transformation from choosing the optimal to choosing the robust.
[0046] 4. This system achieves dual risk avoidance by constructing a technical closed loop of prediction-evaluation-monitoring-correction: it can proactively identify and abandon theoretically optimal but overly vulnerable routes during the planning stage, and monitor and detour around resonance areas where the prediction model has failed during the execution stage, which greatly improves the safety and punctuality of escort missions. Attached Figure Description
[0047] The present invention will be further explained below with reference to the accompanying drawings and embodiments:
[0048] Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation
[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0050] Example 1:
[0051] Please see Figure 1 A route intelligent optimization system based on traffic big data includes:
[0052] The data acquisition module is used to acquire multi-source heterogeneous traffic data and rigid constraints of the escort mission; among which, the rigid constraints of the escort mission include the mission security level and service time window.
[0053] The first processing module is used to predict the travel time of each road segment in the road network based on multi-source heterogeneous traffic data, and to plan and generate a baseline route by combining the predicted travel time with the rigid constraints of the escort mission.
[0054] The second processing module is used to calculate the predicted dependency of each segment in the baseline route, and combine the predicted dependency, the preset topological complexity and the task security level to calculate the total vulnerability index of the baseline route.
[0055] The third processing module is used to obtain the actual travel time of the road segment where the vehicle is located, compare the actual travel time with the predicted travel time, and calculate the prediction deviation factor. The third processing module is also used to determine whether the prediction deviation factor continuously exceeds the preset resonance threshold within the preset time threshold. If it continuously exceeds the threshold, the resonance area is identified and marked.
[0056] The route correction module is used to determine whether the total vulnerability index exceeds the preset vulnerability tolerance threshold and whether the baseline route geographically overlaps with the resonance area. If the total vulnerability index exceeds the vulnerability tolerance threshold or the baseline route geographically overlaps with the resonance area, an antifragile cost function is constructed, and a suboptimal robust path is replanned and generated based on the antifragile cost function.
[0057] This embodiment discloses a route intelligent optimization system based on traffic big data. This system, as a complete closed-loop technology of prediction, evaluation, monitoring, and correction, includes the following modules:
[0058] The data acquisition module aims to provide basic data input for subsequent prediction, planning, and evaluation. This data input includes multi-source heterogeneous traffic data and rigid constraints on the escort mission. The multi-source heterogeneous traffic data integrates data streams from various sources and formats, covering real-time floating car data streams, fixed road sensor data, and historical traffic flow datasets. The rigid constraints on the escort mission are the boundary conditions for mission execution; in this embodiment, they specifically include the geographical location of the financial outlets and their service time windows. ,in, For service time windows, For service start time, The service end time, which must be reached between the start and end times, and the mission security level. , a dimensionless coefficient used to quantify risk tolerance;
[0059] The first processing module aims to generate a theoretically optimal baseline route based on current data and a prediction model. In this embodiment, this module performs two consecutive steps: based on multi-source heterogeneous traffic data, it processes each road segment in the road network... Predicting future travel times yields the predicted travel time. ; and combined with predicted travel time Using cost and escort mission rigid constraints as boundary conditions, a route is planned and generated that satisfies all constraints and has a total predicted travel time. Minimum reference line ;
[0060] The second processing module aims to quantize the reference circuit. The inherent vulnerability arising from the uncertainty of the prediction model; in this embodiment, the module performs a series of calculations: calculating the baseline line various sections of the road Predictive dependency This indicator is used to quantify road segments. Predicted travel time To what extent does it depend on uncertain predictive congestion factors? This was only achieved; subsequently, it was combined with predictive dependency. The preset topological complexity obtained from static map data and the task security level defined by the task parameters. The overall vulnerability score for each road segment is calculated. Ultimately, it converges from all sections of the route. The score is used to calculate the overall vulnerability index of the entire baseline line using a nonlinear amplification model. ;
[0061] The third processing module aims to monitor the actual performance of the prediction model in real time and identify the resonance zone where the prediction model experiences algorithmic resonance. In this embodiment, this module performs the following operations: for vehicles performing tasks, it continuously acquires the actual travel time of the road segment where the vehicle is located. Actual travel time The first processing module predicts the travel time for this road segment. Perform real-time comparisons and calculate prediction bias factors. The third processing module is also used to determine specific geographical areas. Average prediction bias factor within Is it within the preset time threshold? The resonance threshold is continuously exceeded within the range of the preset threshold. If the threshold is consistently exceeded, it indicates that the predictive model for that region has systematically, rather than randomly, failed, and the system identifies and labels that geographic region. This is the resonance region;
[0062] The purpose of the line correction module is to build a decision-making closed loop, enabling the system to automatically abandon a path when it identifies a high vulnerability or high real-time risk. Instead, a more robust solution is implemented. In this embodiment, the module continuously checks the following triggering conditions; the condition triggered during planning is: determining the total vulnerability index. Does it exceed the preset vulnerability tolerance threshold? ; As a vulnerability tolerance threshold, it refers to a core operational security parameter. Its value is determined by backtesting historical failure cases, and it can be set as the minimum threshold that can identify 95% of such failure cases in advance. Value; the condition triggered during execution is: determining the baseline line. Whether the untraveled road sections resonate with the areas identified by the third processing module Geographical overlap occurs;
[0063] If any of the above conditions are triggered, the line correction module constructs an antifragile cost function. Based on this antifragile cost function, a suboptimal robust path is replanned and generated. ;
[0064] This embodiment constructs a complete closed loop of prediction, evaluation, monitoring, and correction through the collaborative work of the above modules. It can not only plan the theoretically most efficient route, but also actively identify and quantify the vulnerability and real-time risks of the route during planning and execution. When the risk exceeds the tolerance level, the system can automatically abandon the optimal solution in the data and instead execute a more robust suboptimal solution. Thus, in the face of the inherent uncertainty of traffic big data prediction models, it greatly improves the safety and punctuality of escort missions and realizes antifragile intelligent route optimization.
[0065] Example 2:
[0066] The first processing module predicts the travel time, including:
[0067] Based on multi-source heterogeneous traffic data, we determine the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight.
[0068] The predicted travel time is calculated based on the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight.
[0069] As a specific implementation of the first processing module of Example 1, its function of predicting travel time is achieved through a dynamic cost function that integrates multidimensional data;
[0070] In this embodiment, the first processing module determines the following parameters based on multi-source heterogeneous traffic data:
[0071] Real-time congestion factor, dimensionless, defined as follows: The data originates from real-time sensor or floating car data acquired by the data acquisition module.
[0072] : Congestion prediction factor, dimensionless, defined as It originates from the calculation results of a time-series prediction model based on historical data; as a specific implementation method, the real-time congestion factor... It can be calculated using the following formula: ,in, It is the instantaneous passage time of the current road segment obtained through real-time sensor or floating car data. This serves as the free-flow reference time for this road segment;
[0073] The time-series prediction model on which it is based can specifically adopt the well-known Long Short-Term Memory Network model in the field. This model predicts the congestion gain factor in the future time window by inputting the historical traffic time series, day of the week, and whether it is a holiday or not.
[0074] The weights are real-time data weights, dimensionless, and are system hyperparameters.
[0075] The data weights are dimensionless and are system hyperparameters;
[0076] To establish a basic travel time prediction model that reflects the dynamic characteristics of traffic, this model will serve as the cost function for all subsequent route planning. Based on the aforementioned real-time congestion factor, predicted congestion factor, real-time data weight, and predicted data weight, the predicted travel time is calculated. The calculation formula is as follows:
[0077] ;
[0078] in, For road section The predicted travel time, measured in units of time, is the output of this formula. For road section The free-flow reference time, measured in units of time, originates from static map data acquired by the data acquisition module; , , , All are as defined above;
[0079] In this embodiment, and satisfy To illustrate the calibration process, in a calibration dataset, define: This refers to the actual historical passage time. These correspond to the historical predicted travel times; the values of these weights are determined by analyzing multiple sets of data. and Regression analysis using historical datasets can be performed using the least squares method to determine the parameters; during periods of stable traffic, the accuracy can be increased. Trust prediction models are used; however, during periods of sudden traffic disruptions, the [data / increase] is increased. To respond quickly to real-time traffic conditions;
[0080] Through the above method, the predicted travel time in this embodiment is obtained. It is no longer a static value, but a dynamic fusion of data from three dimensions: static baseline, real-time traffic conditions, and historical predictions; this design enables... It can more accurately reflect the future toll costs of a road segment, allowing the first processing module to plan a more realistic baseline route. It provides highly accurate cost input.
[0081] Example 3:
[0082] The first processing module plans and generates the baseline route, including:
[0083] The predicted travel time is used as the road segment cost, and the service time window is used as a constraint.
[0084] Invoke the vehicle routing problem solver with time windows to generate the baseline route with the minimum total predicted travel time cost.
[0085] As a further refinement of the first processing module in Embodiment 1, it plans and generates a reference line. Its functionality is implemented based on the Vehicle Routing Problem (VRP) solver;
[0086] For specific details, the first processing module will use the predicted travel time calculated in Example 2. As part of the road segment cost, the service time window obtained by the data collection module will also be considered. As a constraint, the system invokes the VRPTW solver for the vehicle routing problem with a time window, using the start and end points of the escort mission and the financial outlets it must pass through as nodes to solve the problem. Under constraints, the total cost The shortest path;
[0087] The path output by the solver is the baseline route with the minimum total predicted travel time cost. ;
[0088] This embodiment uses the VRPTW solver to ensure the accuracy of the generated baseline circuit. Not only is it optimal in terms of total time in terms of prediction, but it also strictly satisfies the rigid business constraints of the escort mission in terms of logic, thus generating a theoretically optimal and business-feasible benchmark solution as the object of subsequent vulnerability assessment.
[0089] Example 4:
[0090] The second processing module calculates the predicted dependency, including:
[0091] The system invokes the real-time congestion factor, predicted congestion factor, real-time data weight, and predicted data weight determined by the first processing module.
[0092] Based on the above factors and weights, the total congestion gain factor and the congestion gain contributed by the predicted congestion factor are determined.
[0093] The ratio of the congestion gain contributed by the predicted congestion factor to the total congestion gain factor is calculated to obtain the prediction dependence.
[0094] As a specific implementation of the second processing module of Example 1, it calculates the prediction dependency. This function is one of the core steps in the present invention to achieve vulnerability quantification; its purpose is to identify those theoretically optimal factors that heavily depend on uncertain predicted congestion factors. Only sections of road can be completed;
[0095] In this embodiment, the second processing module calls the real-time congestion factor determined by the first processing module. Predicting congestion factors Real-time data weighting and prediction data weights ;
[0096] Based on the above factors and weights, the module performs the following calculations: Determines the total congestion gain factor, which in this embodiment is defined as... The total congestion gain term in the formula, i.e. The congestion gain contributed by the predicted congestion factor is determined; in this embodiment, the gain is defined as the predicted congestion factor. and their weights The absolute value of the contributed gain, i.e. Finally, the ratio of the congestion gain contributed by the predicted congestion factor to the total congestion gain factor is calculated to obtain the prediction dependence. The calculation formula is as follows:
[0097] ;
[0098] in, For road section The predictive dependency is a dimensionless parameter, used as the output of this formula, and its value ranges from 0 to 1; other parameters... , , , All are inherited from the first processing module. Calculation process;
[0099] Using the above formula, this embodiment quantifies a key indicator. ;when When it approaches 1, it indicates that the road segment Efficient travel time is highly dependent on uncertain predicted congestion factors. Therefore, it is highly vulnerable; The value is used for subsequent calculation of overall vulnerability. It provides core, quantifiable risk inputs.
[0100] Example 5:
[0101] The second processing module calculates the overall vulnerability index, including:
[0102] The comprehensive vulnerability score of each segment in the baseline route is determined based on the product of prediction dependency, preset topological complexity, and task security level.
[0103] Obtain the comprehensive vulnerability score of the most vulnerable section of the baseline route;
[0104] Obtain the total accumulated vulnerabilities of the baseline line;
[0105] Based on the score of the most vulnerable road segment and the cumulative sum of vulnerabilities, the total vulnerability index is determined through nonlinear amplification.
[0106] As a further refinement of the second processing module in Example 1, it calculates the total vulnerability index. The function of this invention is the core step in quantifying the weakest link effect;
[0107] The second processing module calculates the baseline. various sections of the road Overall vulnerability score This is based on predictive dependency. Preset topological complexity and mission security level It is determined by the product; It is topological complexity, which is a dimensionless risk coefficient with a configurable range of values. It is derived from static map data and pre-calibrated based on factors such as the number of intersections and the complexity of traffic light timing. It refers to the task safety level, a dimensionless parameter defined by the task parameters acquired by the data acquisition module; as a specific implementation, The value can be determined by looking up a table in the data acquisition module based on the attributes of the escort mission. For example, for regular-level missions, For important-level tasks, For the highest level mission, ;
[0108] The calculation formula is as follows:
[0109] ;
[0110] Obtain all sections of the route. Then, the second processing module determines the total vulnerability index through nonlinear amplification. Specifically: the system obtains the comprehensive vulnerability score of the most vulnerable section of the baseline route, i.e. This represents the weakness of the line; the system obtains the total accumulated vulnerabilities of the baseline line, that is... This represents the total risk exposure of the route; ultimately, based on the score of the most vulnerable section and the cumulative sum of vulnerabilities, a non-linear amplification is performed through multiplication to determine the total vulnerability index. :
[0111] ;
[0112] in, This is the route currently being evaluated; this embodiment uses... The formula will reduce the uncertainty risk of the model. Static topology risks and business security risks Combining them, we obtained a comprehensive result. More importantly, through The formula, this invention achieves nonlinear amplification of vulnerability; a circuit with obvious shortcomings and constant minor problems, its This will increase exponentially, thus being severely punished in subsequent corrective decisions, ensuring the system's ability to avoid high-risk routes.
[0113] Example 6:
[0114] The third processing module identifies and marks the resonance region, including:
[0115] Continuously aggregate prediction bias factors from multiple road segments within a specific geographical area;
[0116] Calculate the average prediction bias factor for a specific geographical region;
[0117] Determine whether the average prediction deviation factor continuously exceeds the resonance threshold within a preset time threshold.
[0118] If so, the specific geographical area will be marked as a resonance zone;
[0119] If not, then a specific geographical area will not be marked as a resonance zone.
[0120] As a specific implementation of the third processing module of Example 1, it identifies and marks the resonance region. Its function is designed to capture systematic rather than accidental prediction failures;
[0121] The third processing module continuously aggregates data from specific geographic areas in the background. Prediction bias factor fed back from multiple road segments ϵ(s) is the prediction bias factor, a dimensionless parameter, obtained through the formula... The calculation yields Ta(s) as the actual travel time and Tp(s) as the predicted travel time from Module 1; furthermore, the module calculates the travel time for a specific geographical area. Average prediction bias factor The specific geographic region Z can be divided using Geohash encoding, which is well-known in the field. For example, the map can be pre-divided into a fixed-precision Geohash grid, with each grid representing a geographic region Z. During aggregation, all road segments falling within the same grid Z will be included. Prediction bias factor The average prediction bias factor for the region is obtained by performing a weighted average. ;
[0122] Determine the average prediction bias factor Is it within the preset time threshold? The internal resonance threshold is continuously exceeded. ; The resonance threshold is determined based on the analysis of historical prediction data. The 95th percentile is used to determine this, and it can be set to 20% to filter out normal random fluctuations in traffic. As a time threshold, the value of this threshold is determined to ensure that the capture of continuous failures can be set to 10 minutes to filter out instantaneous traffic disturbances.
[0123] If so, then the specific geographical area Mark as a resonance zone; otherwise, do not mark a specific geographical area as a resonance zone.
[0124] This embodiment, by introducing dual thresholds of time and space, can accurately identify the resonance zone where the traffic big data prediction model is experiencing algorithmic resonance from a large amount of data. This provides the route correction module with real-time, high-confidence risk input, preventing vehicles from being navigated to an area that appears clear on the data but is actually congested.
[0125] Example 7:
[0126] The antifragile cost function is a weighted sum multi-objective optimization function, and includes the following:
[0127] Total predicted travel time cost of the route;
[0128] Vulnerability time penalty is determined based on the total vulnerability index and the vulnerability time penalty coefficient;
[0129] And a penalty for the real-time resonance zone.
[0130] As the core mechanism of the line correction module in Example 1, the antifragile cost function It is a weighted and multi-objective optimization function; its purpose is to replace the original single time cost when a correction is triggered. Instead, it aims to achieve a balance between efficiency, static risk, and real-time risk.
[0131] The antifragile cost function Used to evaluate the entire path The complete formula for the overall cost is as follows:
[0132] ;
[0133] This function consists of three parts:
[0134] Among them, the total predicted travel time cost This item represents the efficiency target, inherited from the first processing module, and is a measure of the path. All sections of the road Predicted travel time Summation;
[0135] Total vulnerability time penalty This item represents a static risk assessment; among which, For the line The overall vulnerability index is inherited from the second processing module; The vulnerability time penalty coefficient, with time as its dimension, is a core operational security parameter. Its physical meaning is to convert the dimensionless total vulnerability index into a fixed value. Converted to equivalent time cost; The calibration method involves performing regression analysis on historical datasets to solve a... Value, that The value makes Minimum, of which For the sample index in the historical dataset, Indicates the first The actual driving route of this historical mission. Indicates the first The actual passage time for this historical mission. Indicates the first The sum of the predicted travel times for each historical mission. Indicates the first The equivalent vulnerability time penalty for the secondary historical task, i.e., fitting historical data using the least squares method, ensures that the vulnerability penalty term best explains the systematic gap between predicted and actual travel times; total real-time resonance zone penalty. This item represents real-time risk avoidance, inherited from the third processing module, and is related to path... All sections of the road Resonance punishment Summation, where The resonant region has been identified.
[0136] This embodiment constructs... Functions reveal vulnerabilities that were previously invisible. The risk of resonance is explicitly quantified as a time cost; this makes the path planner more efficient in solving problems. In this way, it can automatically weigh the following: although a solution has a short predicted travel time, if its vulnerability is high or it passes through a resonance zone, its total cost will be higher. It can also become extremely high, thus being automatically abandoned by the planner, realizing a decision shift from choosing the optimal to choosing the robust.
[0137] Example 8:
[0138] The real-time resonance region penalty includes:
[0139] Determine whether the road segment is located in the identified resonance zone;
[0140] If the road segment is located in the resonance zone, a preset maximum time penalty value is applied to the road segment;
[0141] If the road segment is not located in the resonance zone, the real-time resonance zone penalty is zero.
[0142] As a response to the antifragile cost function in Example 7 A key component, the real-time resonance zone penalty term Used to achieve factual avoidance of the resonance region;
[0143] Its implementation logic is as follows: determine the road segment Is it located in the resonance region? Inside; if road section Located in the resonance region Then a preset maximum time penalty value is applied to the road segment. ; As the maximum time penalty value, its dimension is time; its value is based on an engineering setting, and can be set as follows: The time in seconds is set far beyond the total time of any normal path to ensure it is effectively prohibited in path planning; if the road segment is not located in the resonance zone, the real-time resonance zone penalty is zero.
[0144] therefore, The definition is: if , ;like , ;
[0145] This embodiment sets a maximum penalty value. This ensures that any region identified as a resonance area by the third processing module is protected. ,exist The costs in the cost function all approach infinity; this makes it difficult for the path planning solver to minimize... At that time, these road sections will be automatically avoided 100% of the time, thus achieving real-time and mandatory risk avoidance during execution.
[0146] Example 9:
[0147] The route correction module re-plans and generates suboptimal robust paths, including:
[0148] Minimize the antifragile cost function to determine the suboptimal robust path;
[0149] If the correction is triggered because the total vulnerability index exceeds the vulnerability tolerance threshold, then the suboptimal robust path will be used as the final execution plan.
[0150] If the correction is triggered by coinciding with the resonance region, the suboptimal robust path will be issued as a dynamic optimization instruction.
[0151] As the execution logic of the route correction module in Example 1, it re-plans and generates a suboptimal robust path. The process is as follows:
[0152] The route correction module solves the antifragile cost function defined in Examples 7 and 8. Minimize the problem, i.e., solve the formula: ,in, This represents the set of all feasible paths that satisfy the rigid constraints of the escort mission. The set of independent variables that minimizes the objective function is obtained by solving the problem. This is the determined suboptimal robust path;
[0153] The system differentiates the handling of Probust based on the specific triggering scenario; if the correction is due to the discovery of the total vulnerability index during planning... Exceeding the vulnerability tolerance threshold When triggered, the system will This will be issued as the final implementation plan; at this time... It is a more robust route that proactively avoids highly vulnerable road sections; the correction was due to the discovery of a resonance zone between the route and the real-time conditions during operation. If the system is triggered by overlap, then the system will... As a dynamically optimized instruction, it is issued; due to In This ensured The resonance zone has been automatically avoided. This instruction will guide vehicles to avoid the risk area in real time;
[0154] This embodiment achieves differentiated correction strategies by distinguishing between two triggering scenarios: planning and execution. During planning, it provides proactive risk avoidance; during execution, it provides reactive risk avoidance. This dual mechanism ensures that the system can generate and execute a more robust path, whether the risk is predicted in the planning stage or monitored in real time in the execution stage, thus achieving an antifragile closed loop throughout the entire process.
[0155] 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.
Claims
1. A route intelligent optimization system based on traffic big data, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous traffic data and rigid constraints of the escort mission; among which, the rigid constraints of the escort mission include the mission security level and service time window. The first processing module is used to predict the travel time of each road segment in the road network based on multi-source heterogeneous traffic data, and to plan and generate a baseline route by combining the predicted travel time with the rigid constraints of the escort mission. The first processing module predicts the travel time by: determining the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight based on multi-source heterogeneous traffic data; and calculating the predicted travel time based on the real-time congestion factor, the predicted congestion factor, the real-time data weight, and the predicted data weight. The second processing module is used to calculate the predicted dependency of each segment in the baseline route, and to calculate the total vulnerability index of the baseline route by combining the predicted dependency, the preset topology complexity, and the task security level. The second processing module calculates the predicted dependency by: calling the real-time congestion factor, predicted congestion factor, real-time data weight, and predicted data weight determined by the first processing module; determining the total congestion gain factor and the congestion gain contributed by the predicted congestion factor based on the above factors and weights; and calculating the ratio of the congestion gain contributed by the predicted congestion factor to the total congestion gain factor to obtain the predicted dependency. The second processing module calculates the total vulnerability index, including: determining the comprehensive vulnerability score of each segment in the baseline route based on the product of prediction dependency, preset topological complexity and task security level; obtaining the comprehensive vulnerability score of the most vulnerable segment in the baseline route; obtaining the cumulative vulnerability sum of the baseline route; and determining the total vulnerability index based on the score of the most vulnerable segment and the cumulative vulnerability sum through nonlinear amplification. The third processing module is used to obtain the actual travel time of the road segment where the vehicle is located, compare the actual travel time with the predicted travel time, and calculate the prediction deviation factor. The third processing module is also used to determine whether the prediction deviation factor continuously exceeds the preset resonance threshold within the preset time threshold. If it continuously exceeds the threshold, the resonance area is identified and marked. The route correction module is used to determine whether the total vulnerability index exceeds a preset vulnerability tolerance threshold and whether the baseline route geographically overlaps with a resonance zone. If the total vulnerability index exceeds the vulnerability tolerance threshold, or the baseline route geographically overlaps with a resonance zone, an antifragile cost function is constructed, and a suboptimal robust path is replanned and generated based on the antifragile cost function. The antifragile cost function is a weighted multi-objective optimization function, which includes: the total predicted travel time cost of the route; a vulnerability time penalty determined based on the total vulnerability index and the vulnerability time penalty coefficient; and a real-time resonance zone penalty term. The real-time resonance zone penalty term includes: determining whether the road segment is located in an identified resonance zone; if the road segment is located in a resonance zone, a preset maximum time penalty value is applied to the road segment; if the road segment is not located in a resonance zone, the real-time resonance zone penalty term is zero.
2. The intelligent route optimization system based on traffic big data according to claim 1, characterized in that, The first processing module plans and generates a baseline route, including: The predicted travel time is used as the road segment cost, and the service time window is used as a constraint. Invoke the vehicle routing problem solver with time windows to generate the baseline route with the minimum total predicted travel time cost.
3. The intelligent route optimization system based on traffic big data according to claim 1, characterized in that, The third processing module identifies and marks the resonance region, including: Continuously aggregate prediction bias factors from multiple road segments within a specific geographical area; Calculate the average prediction bias factor for a specific geographical region; Determine whether the average prediction deviation factor continuously exceeds the resonance threshold within a preset time threshold. If so, the specific geographical area will be marked as a resonance zone; If not, then a specific geographical area will not be marked as a resonance zone.
4. The intelligent route optimization system based on traffic big data according to claim 1, characterized in that, The route correction module re-plans and generates a suboptimal robust path, including: Minimize the antifragile cost function to determine the suboptimal robust path; If the correction is triggered because the total vulnerability index exceeds the vulnerability tolerance threshold, then the suboptimal robust path will be used as the final execution plan. If the correction is triggered by coinciding with the resonance region, the suboptimal robust path will be issued as a dynamic optimization instruction.