A multimodal transport highway-railway path dynamic programming method

Abstract

The present application relates to a kind of multimodal transport public road and rail path dynamic programming method, through real-time acquisition highway, railway, link point multi-source dynamic data, through the data of specified time period in the future and pre-processing, generate dynamic database;Build the time-space coupling target function with shortest total transport time, lowest total transport cost, highest public road and rail link efficiency as collaborative target;Introduce fitness dynamic calibration mechanism to correct target function value;With current time as starting point, set decision window, solve local optimal path, generate initial scheme containing highway real-time navigation path, railway dynamic class matching, link point resource reservation time slice;Based on the value of prediction, set three-level early warning threshold, trigger hierarchical adjustment mechanism to generate adjustment scheme;Evaluate the time / cost variation rate and execution feasibility of adjustment scheme, execute after reaching the standard.Effectively improve the foresight, dynamic adaptability and landing reliability of path planning, effectively reduce transport cost and delay, significantly improve the efficiency of public road and rail transport.

Description

Technical Field

[0001] This invention belongs to the field of route planning technology, and in particular relates to a dynamic planning method for multimodal transport routes (road and rail). Background Technology

[0002] As the logistics industry transforms towards higher efficiency, lower cost, and greener practices, multimodal transport (road-rail intermodal transport) has become a core mode for cross-regional freight transport due to its combination of the door-to-door flexibility of road transport and the long-distance, high-capacity economy of rail transport. However, existing multimodal transport route planning technologies still have multidimensional shortcomings and are difficult to adapt to the dynamic and complex nature of transportation scenarios. Specific problems are as follows: Existing planning methods rely heavily on static basic data and fail to effectively integrate real-time dynamic information. On the one hand, they cannot capture sudden scenarios such as highway congestion, railway delays, and equipment failures at connection points in real time, resulting in a serious disconnect between the planning scheme and the actual transportation process. On the other hand, they lack the ability to predict future transportation conditions and plan global routes based solely on "current instantaneous data," making it difficult to cope with short-term scenario changes.

[0003] Existing methods ignore key impacts of road-rail connections, such as freight trucks waiting for available loading and unloading equipment after arriving at the connection point, or delays caused by mismatched railway schedules. They only optimize independent indicators for road and rail sections, resulting in low overall transportation efficiency.

[0004] Existing solutions often adopt a "one-size-fits-all" approach, initiating a global replanning regardless of the magnitude of the deviation, leading to resource waste and delayed adjustments. For example, if the congestion coefficient of a certain highway section increases slightly, the entire route is recalculated, increasing computational costs and scheduling complexity. Furthermore, when faced with severe risks such as railway trunk line shutdowns or complete failures at connection points, the lack of tiered response strategies results in slow development of adjustment plans, causing the best response opportunity to be missed.

[0005] Therefore, this invention aims to provide a path planning method that can integrate dynamic prediction data, construct a spatiotemporal coupling model, realize dynamic algorithm adaptation, hierarchical adjustment, and ensure feasibility, in order to solve the technical problems existing in the prior art. Summary of the Invention

[0006] The purpose of this invention is to provide a dynamic planning method for multimodal transport routes (rail-road) to solve the technical problems existing in the prior art.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A dynamic programming method for multimodal transport routes (road and rail) includes the following steps: S1: Real-time acquisition of dynamic data on highways, railways, and connecting points; prediction of various data within a specified time period using an LSTM model, generating a dynamic database containing the current state and predicted values, and performing data preprocessing; S2: Construct a spatiotemporally coupled objective function. The objective function takes the shortest total transportation time, the lowest total transportation cost, and the highest efficiency of road-rail connection as the collaborative objectives, and the objective weights are dynamically adjusted according to real-time data. S3: Introduce a fitness dynamic calibration mechanism based on a dynamic database. Based on a genetic algorithm, the objective function value is adjusted in each generation of the population according to the latest prediction data. S4: Starting from the current moment, take a specified future time period as the decision window, solve the local optimal path based on the predicted data within the window, and update the window over time to generate an initial path planning scheme that includes real-time road navigation path, dynamic railway schedule matching, and time slot reservation for connection point resources. S5: Based on the predicted values ​​of the dynamic database, a three-level early warning threshold is set. When the monitored data reaches the corresponding early warning level, the graded adjustment mechanism is triggered to generate an adjustment path planning scheme. S6: Conduct an impact assessment on the adjusted path planning scheme, including time / cost change rate and feasibility assessment. Once all assessment results meet the corresponding requirements, execute the adjusted path planning scheme.

[0008] Preferably, the specific process of data preprocessing in step S1 is as follows: S11: Define the scenario constraint boundaries for each data type, that is, define the absolutely reasonable range and fluctuation warning range for each parameter as the basis for anomaly identification; S12: Perform a two-dimensional screening by combining statistical patterns and scenario constraints, and mark preliminary abnormal data; S13: For the initial abnormal data, perform secondary verification by calling the multi-source verification data source and dynamically adjusted verification rules according to the scenario to which it belongs, distinguish between true anomalies and false anomalies, correct the true abnormal data, and retain and mark the false abnormal data.

[0009] Preferably, the specific expression of the spatiotemporal coupling objective function F in step S2 is as follows: F= ω t (t)· T 总 ( s,t )+ ω c (t)· C 总 ( s,t )+ ω e ( s,t )·(1−E 衔接 ( s,t )); Where F is the overall target value of spatiotemporal coupling; the smaller the value, the better the scheme. ω t (t): Time weight, dynamically adjusted with time t; ω c (t): Cost weight, varying with resource prices during transportation periods; ω e ( s,t ): Connection efficiency weight, which is adjusted according to the real-time resource load of connection point s and time t; T 总 ( s,t Total transit time; C 总 ( s,t Total transportation cost; E 衔接 ( s,t ): The efficiency of the connection between road and rail depends on the connection point s and time t, and the value range is [0,1].

[0010] Preferably, in the spatiotemporal coupling objective function F, the total transportation time T 总 ( s,t The specific expression is: ; in, T 公路 ( L i ,v i ( t ) ,t (This refers to the time for the highway section.) i For road segment numbers; T 公路 ( L i ,v i ( t ) ,t )= L i / v i ( t )·(1+ γ 拥 ( t )); in, γ 拥 ( t () represents the congestion factor. L i For spatial length, v i ( t (This refers to the real-time speed.) T 铁路 ( D j ,f j ( t ) ,t (This refers to the railway section time.) T 铁路 ( D j ,f j ( t ) ,t )= D j / f j (t)+ max (0,t 班计划 -t 到达 ); in, t 班计划 For the scheduled departure time of the train, t 到达 The time it takes for the truck to arrive at the connection point. D j For spatial distance, f j ( t () represents the train's real-time speed; T 中转 ( s k ,q k ( t ) ,t (This refers to the transit time.) T 中转 ( s k ,q k ( t ) ,t )= Q / r k ( t )+ q k (t)· τ k ; Where Q is the quantity of goods. r k ( t This refers to the real-time loading and unloading efficiency at the connection point. q k ( t () is a real-time job queue. τ k This represents the average processing time for a single batch of goods. k The connection point number; C 总 ( s,t The total transportation cost is: ; in, C 公路 ( L i ,p 油 ( t ) , t) represents the highway cost: C 公路 ( L i ,p 油 ( t ) , t)= L i ·( a·p 油 ( t )+ b · δ ( t )); in, a Fuel consumption per unit distance (L / km) b The toll fee is calculated per unit distance (yuan / km). δ (t) represents the time-period premium coefficient, during peak hours. δ =1.2; C 铁路 ( Q,u j ( t ) ,t For railway costs: C 铁路 ( Q,u j ( t ) ,t )=Q· u j( t )·(1+ θ j ( t )); in, Q For the quantity of goods, u j ( t This is the real-time unit freight rate. θ j ( t () represents the capacity tension coefficient; C 中转 ( Q,s k ,t k () represents the transit cost, indicating the cost of the first... k Cost at each connection point, and the volume of goods. Q Connection point s k Resource prices and occupancy time t k Related: C 中转 ( Q,s k ,t k )= Q·c k + max (0, t k −t 基 )· d k ; in c k Loading and unloading fee per unit (yuan / ton). t 基 Based on free time (hours). d k Overtime fee; Efficiency of road-rail connection E 衔接 ( s,t ): E 衔接 ( s,t )= α / (1+ T 等待 ( s,t ))+(1− α )·(1− ρ k ( t )); in,T 等待 ( s,t () represents the waiting time (in hours) at the connection point; ρ k ( t () represents the real-time load rate of the connection point equipment. α These are the weighting coefficients.

[0011] Preferably, the specific process of step S3 is as follows: S31: Population individual coding, i.e., the gene sequence that defines the route plan: converting the road-rail intermodal transport route plan into GA-identifiable individuals, with each individual corresponding to a complete transport plan; S32: Initial fitness calculation, i.e. the first evaluation based on the current data: After population initialization, the initial fitness of each individual is calculated based on the current data in the dynamic database as the starting point for iteration; S33: Dynamic data-triggered calibration. After each selection-crossover-mutation to generate a new population, GA first checks whether the dynamic database has been updated. If the latest predicted data exists, fitness calibration is triggered. S34: Objective function parameter correction, updating the core indicators of individuals: For the population that triggered calibration, extract the genetic information of each individual, i.e. path and time, and combine it with the latest prediction data to correct the key parameters of its objective function.

[0012] Preferably, the specific process of step S34 is as follows: S341: Correction to total transport time Ttotal: Highway segment time correction: updating driving speed based on predicted congestion coefficient; Railway time adjustment: Update train delays based on predicted punctuality rates; Transfer time correction: Update loading and unloading efficiency based on predicted equipment load at connection points; S342: Correction to total transportation cost Ctotal: Highway costs: adjusted based on predicted fuel prices and congestion period premiums; Railway costs: Adjusting capacity premiums based on predicted carriage availability; Transit costs: Timeout fees adjusted based on predicted waiting time; S343: Adjust connection efficiency based on predicted waiting time and device load.

[0013] Preferably, the specific process of step S4 is as follows: S41: Decision window initialization: Define the spatiotemporal boundaries of local optimization: Starting from the current moment, clarify the time range of the window and the core decision objective, and define the boundaries of local optimization for subsequent solutions; S42: Predictive data extraction and adaptation within the window: Extract the predicted data within the window from the dynamic database and adapt it according to the categories of highway-railway-connection point; S43: Local optimal path solution, i.e., selecting the optimal solution within the window based on the predicted data; S44: Based on the optimal solution, integrate the executable information of navigation, schedule, and reservation to generate the initial route plan.

[0014] Preferably, the specific process of step S43 is as follows: S431: Abstract the transportation scenario within the window into a node-edge network and clarify the relationships between the elements; S432: Encode the highway-connection point-railway combination within the window to generate 100-200 candidate solutions; S433: Using dynamic programming to calculate the target value for each candidate, with the shortest total time and the highest resource reservation success rate as the local objectives; S434: Sort by target value and select the top 20% of candidates as the local optimum within the window.

[0015] Preferably, the specific process of step S5 is as follows: S51: Warning threshold setting: Define the deviation risk level based on the predicted value; S52: Real-time data monitoring and deviation calculation: Continuously compare predicted values ​​with actual values; S53: Warning level determination: The level is determined based on the deviation rate and parameter characteristics; S54: Tiered adjustment implementation: Match the minimum impact adjustment range according to the warning level.

[0016] Preferably, the specific process of step S6 is as follows: S61: Determine the evaluation criteria, scope, and standards for the adjusted route planning scheme; S62: Calculation of time and cost change rates, quantifying the benefits and costs of adjustments; S63: Conduct an execution feasibility assessment and determine whether the adjusted route planning scheme is "compliant and executable". If yes, execute the adjusted route planning scheme; otherwise, repeat steps S2-S6.

[0017] The beneficial effects of this invention include: 1. By combining multi-source dynamic data acquisition, LSTM prediction, and scenario-based anomaly filtering, the system addresses the issues of static and low reliability of existing data: Real-time collection of dynamic data from highways, railways, and connecting points, coupled with LSTM model prediction, generates a dynamic database of current state and predicted values. This allows route planning to not only be based on current facts but also adapt to future changes. 2. An adaptive anomaly filtering mechanism based on transportation scenario characteristics, compared to fixed threshold filtering, effectively improves data quality, avoids the loss of valid data, and provides accurate input for subsequent planning.

[0018] 2. By employing a spatiotemporally coupled objective function, this approach addresses the issues of singular objectives and spatiotemporal disconnect in existing solutions. It collaboratively optimizes three main objectives: total transportation time, total transportation cost, and rail-road connection efficiency. Connection efficiency is incorporated as a core optimization dimension, resulting in improved overall transportation efficiency compared to single-objective planning. Objective weights adaptively adjust based on dynamic scenarios and support three preset modes: time-first, cost-first, and balanced optimization. This adapts to different transportation needs, such as emergency supplies, regular goods, and fresh produce. Each parameter in the objective function is linked to spatial paths and time, accurately reflecting the spatiotemporal coupling characteristics and ensuring the optimization direction better aligns with actual transportation scenarios, thus improving the match between the solution and actual needs.

[0019] 3. By using a fitness dynamic calibration mechanism, the problems of rigidity and poor adaptability of existing algorithms are solved: first, global candidate solutions are generated quickly, and then local connection details are optimized. The efficiency of candidate solution generation is improved, and the accuracy of local detail optimization is improved, avoiding the mismatch between highway arrival time and railway departure time. The fitness of GA is calibrated based on the latest prediction data of the dynamic database in each generation, ensuring that the convergence direction of the algorithm always fits the real-time scenario and avoiding failure caused by evaluating dynamic solutions with static data.

[0020] 4. By implementing tiered early warning adjustments and impact assessments, the existing adjustments are addressed as being crude and lacking feasibility: Firstly, a three-tiered early warning system is established based on the prediction deviation rate, with differentiated adjustment ranges matched. Compared to a one-size-fits-all approach, this reduces computational costs and improves response speed. After adjustment, calculations based on time / cost change rates, combined with multi-dimensional feasibility assessments, ensure the solution is implementable while maintaining controllable costs, avoiding risks associated with blind adjustments and increasing the success rate of solution implementation. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the dynamic planning method for multimodal transport routes (rail-road) of the present invention.

[0022] Figure 2 This is a schematic diagram illustrating the logical principle of step S3 of the present invention. Detailed Implementation

[0023] The following is in conjunction with the appendix Figures 1-2 The present invention will be further described in detail below: Example 1 See appendix Figure 1 As shown, the dynamic planning method for multimodal transport routes (road and rail) includes the following steps: S1: Real-time acquisition of dynamic data on highways, railways, and connecting points; prediction of various data within a specified time period using an LSTM model, generating a dynamic database containing the current state and predicted values ​​for the next 1-3 hours; and preprocessing of abnormal data using an adaptive filtering mechanism based on transportation scenario characteristics.

[0024] The highway dynamic data includes real-time traffic conditions, congestion coefficients, and timely speed limits. The real-time traffic data comes from: Fixed monitoring equipment: Road cameras and microwave radars deployed by traffic management departments, which are acquired in real time through traffic API.

[0025] Floating car data: Real-time location, speed, and driving trajectory data uploaded by the vehicle's onboard terminal (GPS / BeiDou).

[0026] Real-time traffic conditions are described as "road segment traffic status," divided into four levels: "smooth traffic, light congestion, moderate congestion, and severe congestion," based on the following indicators: If the average vehicle speed of a road segment is ≥80% of the free-flow speed, it is considered to be smooth. The free-flow speed is the maximum design speed of the road segment when there is no congestion, such as 120km / h for highways. (60%-80%) free-flow velocity is considered mild congestion; (30%-60%) free-flow velocity is judged as moderate congestion; A free-flow velocity of ≤30% is considered severe congestion.

[0027] The congestion coefficient is calculated based on the raw real-time traffic data described above, without the need for additional data collection. The calculation formula is as follows: Congestion coefficient = 1 − average speed of vehicles on the current road segment / free-flow speed of the road segment.

[0028] For example, if the free-flow speed on a certain highway section is 100 km / h and the current average vehicle speed is 40 km / h, then the congestion coefficient = 1 - 40 / 100 = 0.6, which is considered moderate congestion. If there are no vehicles on the road section, to avoid a negative coefficient, it is forcibly set to 0.05, which is approximately considered smooth traffic.

[0029] The data for time-varying speed limits comes from the legally mandated speed limit database for roads, such as 120 km / h for highways and 80 km / h for national highways. It is also obtained through the temporary speed limit notification interface of the traffic management department, such as the speed limit being reduced to 80 km / h during rainfall and 40 km / h for construction sections. The data is further adjusted in conjunction with real-time weather data from the meteorological API, such as speed limit reductions triggered by heavy rain or fog.

[0030] The calculation formula is: The time-varying speed limit is the dynamically adjusted speed limit value; otherwise, the base speed limit is used. For example, on a sunny day with no construction, the time-varying speed limit on a highway is 120 km / h; in the event of a sudden rainstorm and a temporary traffic management notice, the time-varying speed limit is updated to 80 km / h.

[0031] The railway dynamic data includes on-time performance, available carriages, and capacity scheduling information.

[0032] The on-time rate of the train service is obtained from the railway freight dispatching system by acquiring the historical 30-day departure / arrival records of the target train service (including planned time and actual time) and the real-time status of the current train service (including departure / not yet departed and current delay duration).

[0033] On-time performance is a metric for measuring the reliability of train services, with a value ranging from 0 to 1, where 1 represents perfect on-time performance. It includes both historical on-time performance and current predicted on-time performance. Historical punctuality rate: This is calculated as the percentage of times a particular service is on time over the past 30 days. The specific expression is: Historical on-time rate = Number of on-time departures or arrivals in 30 days / Total number of trips in 30 days. Departure delays of ≤10 minutes and arrival delays of ≤15 minutes are considered on-time.

[0034] Current predicted on-time rate: If the bus has not yet departed, the historical on-time rate is adjusted based on real-time data. For example, if the previous stop was delayed by 20 minutes, and historical data shows that "when the previous stop was delayed by 20 minutes, the on-time rate of this bus decreased by 30%", then the current predicted on-time rate = historical on-time rate × 70%.

[0035] The car's reserve capacity is pushed in real time via data from the railway freight platform interface. The car's reserve capacity represents the maximum weight it can currently carry, measured in tons. The specific expression is: Car reserve capacity = Car's rated load capacity - Current actual load. For example, a freight car with a rated load capacity of 60 tons, currently carrying 45 tons, has a reserve capacity of 15 tons.

[0036] Transport capacity scheduling information is obtained directly from the real-time command interface of the railway dispatching system, including: Routine scheduling: increases or decreases in service frequency, route adjustments; Emergency dispatch: service suspension, carriage swapping (removing a carriage due to malfunction), priority adjustment (disaster relief supplies trains will be dispatched first).

[0037] The capacity scheduling information is structured and transformed, for example, "T3 service is suspended due to equipment failure" is converted into the label "[Service ID=T3, Scheduling type=Suspension, Effective time=2025-11-03 14:00]", so that the service can be excluded in subsequent route planning.

[0038] Dynamic data at connection points includes transit resource occupancy rate, real-time load of loading and unloading equipment, and warehouse turnover efficiency.

[0039] Transit resources include loading and unloading spaces, storage locations, forklifts / cranes, and other equipment, which are collected through the IoT system at the connection point (freight station). Parking / cargo space: RFID tag identification (occupied / vacant), video surveillance image recognition (number of vehicles parked in the parking space); Equipment: Sensors collect data on whether the system is currently in operation and the duration of continuous operation.

[0040] The transit resource utilization rate is the ratio of used resources to total resources, with a value between 0 and 1, and is expressed as follows: Resource utilization rate = Number of currently occupied resources / Total number of resources of that type; For example: If a freight station has 10 loading and unloading bays and 6 are currently occupied, then the bay occupancy rate is 6 / 10 = 0.6; if 3 out of 5 forklifts are in operation, then the forklift occupancy rate is 3 / 5 = 0.6.

[0041] The real-time load of loading and unloading equipment is obtained by acquiring raw data from the torque sensor (monitoring the lifting weight) and motion sensor (monitoring the operation frequency: times / hour) of the loading and unloading equipment (forklift, crane), and then uploaded in real time through the edge computing gateway.

[0042] Calculation process: Load is the ratio of the current workload of the equipment to its rated capacity, with a value of 0-1, where 1 represents full load. The expression is: Real-time load = Actual workload in the current hour / Rated hourly workload of the equipment.

[0043] For example, if a forklift is rated to move 20 tons per hour, and has moved 15 tons in the current hour, then the real-time load = 15 / 20 = 0.75.

[0044] Warehouse turnover efficiency is obtained from the warehouse management system (WMS) at the connection point and is calculated by combining the "inbound and outbound volume in the past hour" with warehouse capacity data (total capacity and current inventory).

[0045] Calculation process: Turnover efficiency reflects the speed of warehouse processing, unit: tons / hour, expression: Warehouse turnover efficiency = Inbound volume + Outbound volume in the past hour / 1 hour; For example, if 30 tons are received and 20 tons are shipped out within one hour, the turnover efficiency is 50 tons / hour.

[0046] S2: Construct a spatiotemporally coupled objective function. The objective function takes the shortest total transportation time, the lowest total transportation cost, and the highest efficiency of road-rail connection as its collaborative goals. The objective weights are dynamically adjusted based on real-time data. When the road congestion coefficient is ≥0.7, the time weight is automatically increased to 0.5 or higher. When the railway carriage capacity is ≤120% of the demand, the connection efficiency weight is automatically increased to 0.3 or higher. At the same time, dynamic constraints are configured, including time-limited road traffic restriction constraints (based on real-time traffic management notification updates), dynamic priority constraints for railway train schedules (ranked in real-time according to punctuality rate), and resource conflict constraints at connection points (to avoid equipment preemption through real-time load data).

[0047] S3: Introduce a "fitness dynamic calibration mechanism" based on a dynamic database. Based on a genetic algorithm, the objective function value is adjusted in each generation of the population according to the latest prediction data.

[0048] S4: Starting from the current moment, the next 3 hours are used as the decision window. The local optimal path is solved based on the predicted data within the window. The window is updated over time, and finally an initial path planning scheme is generated, which includes "real-time road navigation path - dynamic railway schedule matching - connection point resource reservation time slice".

[0049] S5: Based on the predicted values ​​from the dynamic database, three levels of early warning thresholds are set: minor warning is a prediction deviation of 10-20%; moderate warning is a prediction deviation of 20-30%; and severe warning is a prediction deviation of ≥30%. When the monitored data reaches the corresponding warning level, a graded adjustment mechanism is triggered: for minor warnings, only the parameters of the current road segment are optimized; for moderate warnings, the local path is reconstructed, which may replace the connection point; and for severe warnings, global replanning is initiated to generate an adjusted path planning scheme.

[0050] S6: Conduct an impact assessment on the adjusted path planning scheme, including time / cost change rate and feasibility assessment. Once all assessment results meet the corresponding requirements, execute the adjusted path planning scheme.

[0051] The specific process of data preprocessing for abnormal data in step S1 using an adaptive filtering mechanism based on transportation scenario characteristics is as follows: S11: Define the scenario constraint boundaries for each data type, that is, delineate the absolutely reasonable range and fluctuation warning range for each parameter, as the basis for anomaly identification. See Table 1 below for details: Table 1. Scenario Constraint Boundaries for Each Data Type S12: Perform a two-dimensional screening by combining statistical regularities and scenario constraints, and mark preliminary abnormal data: For each parameter, calculate its historical data distribution for the same period over the past 7 days, including the mean μ and standard deviation σ. If the current data meets one of the following conditions, it is marked as a statistical anomaly: Exceeding the range of [μ-3σ, μ+3σ], i.e., significantly deviating from historical fluctuations; The rate of change over three consecutive data collection periods is ≥50%, such as a sudden increase in the highway congestion coefficient from 0.2 to 0.5, which represents a rate of change of 150%. For example: The historical average congestion coefficient of Highway A at 8:00 AM is μ=0.3, σ=0.15, and the current value is 0.8 (exceeding the upper limit of 3σ 0.75), which is marked as a statistical anomaly.

[0052] Check if the data violates the scene constraint boundaries. If any of the following conditions are met, mark it as a scene anomaly: Exceeding the absolutely reasonable range, such as a congestion coefficient of 1.2, violates the constraint ∈ [0, 1]. This violates business logic. For example, if the railway T3 train has been suspended, but the remaining capacity of the carriages shows 10 tons, then the suspended train has no actual load, and the remaining capacity should be invalid.

[0053] For example, if the forklift load rate at the connection point is 1.2, it violates the constraint ∈[0,1] → mark the scenario as abnormal.

[0054] By merging statistical anomalies and scenario anomalies, preliminary anomaly data is obtained, which awaits further verification to avoid misjudgment based on a single dimension. For example, the highway congestion coefficient may exceed 3σ during heavy rain, but this may be a true reflection of the situation.

[0055] S13: For preliminary abnormal data, perform secondary verification by calling multi-source verification data sources and dynamically adjusted verification rules according to the scenario to distinguish between true anomalies (data errors) and false anomalies (real scenario changes). The verification rules are dynamically adjusted according to the scenario.

[0056] Highway data anomaly verification: Verification data sources: floating car GPS speed, adjacent road segment data, traffic event notifications (such as construction, accidents), and meteorological data; Adaptive verification rule: If the congestion coefficient is abnormal (e.g., statistical anomaly): compare with the floating car average speed. If the speed is ≤30km / h and the weather shows heavy rain, it is judged as real congestion (pseudo-anomaly); if the speed is ≥60km / h and there is no traffic event, it is judged as data error (real anomaly).

[0057] If the time-varying speed limit is abnormal, such as exceeding the legal limit: call the traffic management department's temporary speed limit filing interface. If there is a special filing, such as for a test section, it is determined to be a legal adjustment (pseudo-abnormality); otherwise, it is an interface data error (true abnormality).

[0058] Railway data anomaly verification: Verify the data source: data from previous stations (such as delay records of the T3 train at previous stations), status of other trains on the same line, carriage sensor logs (whether there is a fault), and dispatching instruction history.

[0059] Adaptive verification rule: If the remaining capacity of the carriage is abnormal (e.g., negative): check the data of the previous 10 minutes in the same carriage. If it is continuously negative and the sensor log shows a fault alarm, it is determined to be a sensor error (true anomaly); if only the current value is negative and the dispatching instruction shows temporary loading of goods, it is corrected to actual remaining capacity = 0 (false anomaly).

[0060] If the on-time rate is abnormal (such as a sudden drop): compare the delay time of the previous station. If the previous station has been delayed by 60 minutes (far exceeding the historical average), it is determined to be a true decrease in the on-time rate (false anomaly); otherwise, it is a statistical error (true anomaly).

[0061] Connection point data anomaly verification: Verify data sources: equipment operation logs (such as forklift work records), video surveillance images (manual verification), and work queue data (quantity of goods to be loaded and unloaded).

[0062] Adaptive verification rule: If the load rate of the loading and unloading equipment is abnormal, such as =1.2, check the equipment log. If it shows that the sensor is being calibrated, it is determined to be a calibration error (true anomaly); if the video monitoring shows that the equipment is running at full load and has an overload alarm, then correct the load rate to =1.0 (false anomaly).

[0063] If warehouse turnover efficiency is abnormal, such as far exceeding the equipment limit: check the work queue. If the actual inbound and outbound quantities are consistent with the data, and the scheduling record shows that additional equipment was temporarily dispatched, then it is determined to be a real efficiency improvement (pseudo-anomaly); otherwise, it is a WMS system counting error (real anomaly).

[0064] Adaptive handling: Dynamically handling exceptions based on exception type and scenario. Based on the verification results, differentiated handling strategies are adopted for true anomalies and false anomalies, and the handling methods are dynamically adjusted according to the transportation scenario (such as emergency freight / regular freight): True outlier data (data errors / missing data): Correction strategy: For short-term missing data (≤3 collection periods): use time-series interpolation + scenario adaptation for completion. For example, if the highway congestion coefficient is missing, use the average of the previous two periods combined with the current congestion trend (increasing / decreasing) of adjacent road segments for interpolation; Sensor failure (continuous abnormality): Use a backup data source as a substitute. For example, when the railway car balance sensor fails, estimate the load by subtracting the booked freight volume from the rated load. Obvious errors, such as negative values: force correction to reasonable limits, for example, if the car body margin is -5 tons, it will be corrected to 0 tons (meaning there is no margin).

[0065] Adaptive adjustment: In emergency freight scenarios, data from nearest neighbor devices is prioritized for completion, such as the real-time speed of adjacent trucks, to ensure data timeliness; for regular freight, more accurate historical average interpolation can be used to balance efficiency and accuracy.

[0066] Pseudo-anomaly data (real-world scenario changes): The original data is not modified, but extreme scenario markers are added. For example, a sudden accident causes the highway congestion coefficient to be 1.0 (exceeding 3σ), and it is marked "[Extreme event: road closure due to accident]", so that the subsequent path planning model can focus on it, such as prioritizing avoiding this road segment.

[0067] Dynamically update thresholds: If similar pseudo-anomalies occur frequently in a certain scenario, such as the highway congestion coefficient often exceeding 3σ during rainstorms, the statistical threshold for that scenario is automatically increased, such as from 3σ to 4σ, to reduce future misjudgments.

[0068] The specific expression for the spatiotemporal coupling objective function F in step S2 is as follows: F= ω t (t)· T 总 ( s,t )+ ω c (t)· C 总 ( s,t )+ ω e ( s,t )·(1− E 衔接 ( s,t )); Where F is the overall target value of spatiotemporal coupling; the smaller the value, the better the scheme. ω t (t): Time weight, which is dynamically adjusted with time t, such as ωt increasing during peak periods; ω c (t): Cost weight, adjusted according to resource prices during transportation periods, such as peak railway freight rates; ω e ( s,t ): Connection efficiency weight, varying with the connection point. s Real-time resource load and time t Adjustments can be made, such as when there is congestion at connection points. ω e Increase; T 总 ( s,t Total transport time, depending on spatial path s and time t; C总 ( s,t Total transportation cost, depending on spatial path s and time t; E 衔接 ( s,t ): The efficiency of the connection between road and rail depends on the connection point s and time t, and the value ranges from [0,1]. The larger the value, the higher the efficiency. In the objective function, "1-E connection" represents the "efficiency loss", which needs to be minimized.

[0069] Total transport time in the spatiotemporal coupling objective function F T 总 ( s,t The details are as follows: The total transportation time is the sum of the road segment, the rail segment, and the transfer time at the connecting points. The time for each link changes dynamically with the spatial path (segment, station) and the time (congestion periods, train schedules). The specific expression is as follows: ; in, T 公路 ( L i ,v i ( t ) ,t () represents the highway segment time, indicating the first... i The transport time of a section of highway depends on its spatial length. L i (km) and real-time speed v i ( t (km / h, varying with congestion level at time t).

[0070] T 公路 ( L i ,v i ( t ) ,t )= L i / v i ( t )·(1+ γ 拥 ( t )); in γ 拥 ( t ) is the congestion factor, calculated from the real-time congestion factor. For example, when the congestion factor is 0.6, γcongestion = 0.6.

[0071] T铁路 ( D j ,f j ( t ) ,t () represents the railway segment time: indicating the first... j The transport time of a railway segment depends on spatial distance. D j (km) and real-time train speed f j ( t (km / h, varying with the on-time rate fj(t)), including waiting time, is expressed as follows: T 铁路 ( D j ,f j ( t ) ,t )= D j / f j (t)+ max (0,t 班计划 -t 到达 ); in, t 班计划 For the scheduled departure time of the train, t 到达 The arrival time of the truck at the connection point is determined by its spatial location, reflecting the spatiotemporal coupling.

[0072] T 中转 ( s k ,q k ( t ) ,t () represents the transit time: indicating the first... k Connection points s k The transit time depends on the spatial resources (number of devices, etc.) at the connection point and the real-time job queue. q k ( t (Vehicles / batch, as needed) t change).

[0073] T 中转 ( s k ,q k ( t ) ,t )=Q / r k ( t )+ q k (t)· τ k ; Where Q represents the quantity of goods (tons). r k ( t This represents the real-time loading and unloading efficiency at the connection point, expressed in tons per hour, and varies with equipment load. τ k This represents the average processing time for a single batch of goods, expressed in hours per batch.

[0074] Total transportation cost C 总 ( s,t This is the sum of road, rail, and transfer costs. The cost is strongly coupled with the resource consumption of spatial routes (such as mileage) and time-sensitive pricing (such as peak-hour surcharges). The specific expression is as follows: ; in, C 公路 ( L i ,p 油 ( t ) , t) represents the highway cost, indicating the cost of the first... i Highway cost, including fuel costs (and spatial mileage) L i and real-time oil prices p 油 ( t Related to road segment type and time period t, such as the increase of highway toll during peak hours.

[0075] C 公路 ( L i ,p 油 ( t ) , t)= L i ·( a·p 油 ( t )+ b · δ ( t )); in, a Fuel consumption per unit distance (L / km) b The toll fee is calculated per unit distance (yuan / km). δ (t) represents the time-period premium coefficient, during peak hours. δ =1.2.

[0076] C 铁路 ( Q,u j ( t ) ,t (This refers to railway costs, in relation to freight volume) Q Real-time unit freight rate u j ( t (Unit: Yuan / ton; price varies depending on the remaining amount in the carriage; lower remaining amount will result in lower prices.) u j ( t (Increase) is related to.

[0077] C 铁路 ( Q,u j ( t ) ,t )=Q· u j ( t )·(1+ θ j ( t )); θ j ( t () represents the capacity tension coefficient, when the available carriage space is ≤ 120% of the demand. θ j ( t =0.1.

[0078] C 中转 ( Q,s k ,t k () represents the transit cost, indicating the cost of the first... k Cost at each connection point, and the volume of goods. Q Connection point s k Resource prices (such as loading and unloading fees) and occupation time t k (Hours, extra charges apply for exceeding the time limit) Related.

[0079] C 中转 ( Q,s k ,t k )= Q·c k + max (0,t k −t 基 )· d k ; in c k Loading and unloading fee per unit (yuan / ton). t 基 Based on free time (hours). d k Overtime fee (RMB / hour).

[0080] Efficiency of road-rail connection E 衔接 ( s,t ): Reflects the smoothness of road-rail transition, and is strongly coupled with the utilization rate of spatial resources at the connection point (equipment load, etc.) and the time matching degree (the connection gap between freight cars and trains). Expression: E 衔接 ( s,t )= α / (1+ T 等待 ( s,t ))+(1− α )·(1− ρ k ( t )); in, T 等待 ( s,t ) represents the waiting time (in hours) at the connection point, which is the interval between the arrival of the freight car and the departure of the railway (the spatial path determines the arrival time, which is coupled with the train schedule). ρ k ( t The real-time load rate of the connection point equipment is 0-1, varying with time. t change; α α is the weighting coefficient, 0 < α < 1, balancing time matching degree and resource utilization rate.

[0081] Manifestation of spatiotemporal coupling characteristics: The effect of time on space: the time T of the same spatial path (such as highway A→B) at different times t (peak / off-peak). 公路 And cost C 公路 different.

[0082] The constraints of space on time: The choice of spatial route (such as choosing connection point C or connection point D) directly determines the transfer time T and the waiting time for train services.

[0083] Spatiotemporal dynamics of weights: ωt (t), ω c (t), ω e ( s,t ) at any time t (e.g., during emergency transport periods) and connection points s (For example, at connection points where resources are scarce) Dynamic adjustments are made to adapt the objective function to the real-time scenario.

[0084] This formula quantifies the spatiotemporal interaction, avoiding the shortcomings of the separation of time and space in traditional static models. It ensures that the objective function can truly reflect the dynamic relationship between route selection, time consumption, and cost expenditure in multimodal transport, providing a precise optimization direction for subsequent route solving.

[0085] Example 2 Based on Example 1, see Figure 2 As shown, the specific process of step S3 is as follows: S31: Population individual encoding, i.e., the gene sequence defining the path scheme: Transform the rail-road intermodal transport route scheme into a chromosome that can be recognized by GA, i.e., an individual. Each individual corresponds to a complete transport scheme. The code must include key information on spatial route selection and time node matching to ensure that it can be associated with parameters in the dynamic database.

[0086] Encoding rules: The chromosome structure is: [Highway segment 1 ID, estimated departure time of highway segment 1, connection point 1 ID, railway train number 1 ID, estimated departure time of railway train number 1, connection point 2 ID, ..., highway segment n ID, estimated arrival time of the destination]; Example: An individual code is [G101, T0+1h, Station C, T1, T0+5h, Station D, G202, T0+24h], which means: from the starting point via G101 highway (departing at T0+1h) → Station C → take train T1 (departing at T0+5h) → Station D → via G202 highway → destination (expected arrival at T0+24h).

[0087] The association between coding and dynamic data: Each gene fragment (such as "G101 highway", "T1 train", "C station") can be directly mapped to the corresponding parameters in the dynamic database (such as the predicted congestion coefficient of G101, the predicted on-time rate of T1, and the predicted equipment load of C station), providing a data interface for subsequent calibration.

[0088] S32: Initial fitness calculation, i.e., the first evaluation based on the current data: After population initialization (randomly generating N individuals, N=50-200), the initial fitness of each individual is calculated based on the current time data (non-predicted values) of the dynamic database, which serves as the starting point for iteration.

[0089] Initial fitness function: The fitness value is negatively correlated with the spatiotemporal coupling objective function F; the smaller the objective function value, the higher the fitness. The expression is: Fit 初始 ( x )=1 / (1+F( x ,t0)); in, x For an individual (path scheme), t0 is the current time, F( x , t 0) is the objective function value (including total time, cost, and connection efficiency) calculated based on the data at time t0.

[0090] Calculation example: For individuals coded as [G101, T0+1h, Station C, T1, T0+5h, ...], retrieve data from the dynamic database. t Given the congestion coefficient of G101 at time 0 (0.3), the on-time rate of T1 (0.9), and the equipment load of station C (0.5), and calculating F=7500, the initial fitness Fitinitial=1 / (1+7500)≈0.00013.

[0091] S33: Dynamic data-triggered calibration, i.e., data freshness check before each iteration: After each "selection-crossover-mutation" process to generate a new population (offspring), GA does not immediately enter the next iteration. Instead, it first checks if the dynamic database has been updated. If the latest prediction data exists, including predictions for the next 1-3 hours, such as the congestion coefficient prediction for G101 at T0+1h and the delay prediction for T1 at T0+5h, then fitness calibration is triggered.

[0092] Triggering conditions: Time-triggered: Each generation of the population is iterated at intervals of ≥30 minutes, synchronized with the dynamic database update cycle; Event triggering: When key parameters in the dynamic database change significantly, such as railway train delays of ≥30 minutes or a sudden increase in highway congestion coefficient of ≥0.3, the event is triggered immediately, without being limited by the iteration interval. Data acquisition: Extract the latest predicted data related to individuals in the current population through the database interface; For example, for individuals involved in the G101 highway, the predicted congestion coefficient for their expected travel time (T0+1h to T0+3h) is obtained; for individuals involved in the T1 train, the predicted on-time rate for their departure time (T0+5h) is obtained.

[0093] S34: Objective function parameter correction, updating individual core metrics: For the population that triggered calibration, the genetic information (path, time) of each individual was extracted, and its objective function F was revised by combining it with the latest prediction data. x , t The key parameters (total time, cost, and connection efficiency) in the data reflect the dynamic changes in spatiotemporal coupling.

[0094] The specific process of step S34 is as follows: S341: Correction to total transport time Ttotal: Highway travel time correction: Travel speed is updated based on predicted congestion coefficients. For example, if the original data shows a congestion coefficient of 0.3 (speed 60km / h) for G101 at T0+1h, and the latest prediction shows it has risen to 0.6 (speed 40km / h), then the highway travel time will be corrected accordingly. T 公路 修正 =(L G101 / 40)·(1+0.6), the original calculated value is (L G101 / 60)·(1+0.3); Railway time adjustment: Train delays are updated based on predicted punctuality rates. For example, if the original predicted punctuality rate for train T1 was 0.9 (no delay), and the latest prediction drops to 0.5 (30-minute delay), then the train time will be adjusted accordingly. T 铁路 修正 =Original travel time + 30 minutes; Transfer time correction: Based on the predicted equipment load at the connection point, the loading and unloading efficiency is updated. For example, if the original predicted load for station C was 0.5 (efficiency 20 tons / hour), and the latest prediction rises to 0.8 (efficiency drops to 15 tons / hour), then the transfer time will be corrected. T 中转 修正 =Q / 15 + predicted queue length τ (The original calculated value is Q / 20 + original queue length τ).

[0095] S342: Correction to total transportation cost Ctotal: Highway costs: Based on predicted fuel prices and congestion period premiums, for example, if the predicted fuel price increase of 5% in T0+1h is made, the highway fuel cost will be adjusted to the original cost × 1.05; Railway costs: The capacity premium is adjusted based on the predicted carriage capacity. For example, if the original predicted capacity of the T1 train was 50 tons (no premium), and the latest prediction has dropped to 30 tons (10% premium), then the railway cost is adjusted to the original cost × 1.1. Transfer cost: The overtime fee is adjusted based on the predicted waiting time. For example, if the original predicted waiting time for Station C was 1 hour (no overtime), and the latest predicted waiting time is 2.5 hours (1.5 hours overtime), then the transfer cost will increase by 1.5 × d k (Overtime fee).

[0096] S343: Correction of connection efficiency E connection: Based on the predicted waiting time and equipment load adjustments, for example, if the original predicted waiting time for station C was 0.5 hours and the load was 0.5, and the latest predicted waiting time is 1.5 hours and the load is 0.8, then: E 衔接 修正 =α / (1+1.5)+(1−α)·(1−0.8)(The original calculated value is α / (1+0.5)+(1−α)·(1−0.5)).

[0097] The specific process of step S4 is as follows: S41: Decision window initialization, defining the spatiotemporal boundary of local optimization: Starting from the current time (t0), clarify the time range and core decision objective of the window, and define the boundary of "local optimization" for subsequent solutions to avoid computational inefficiency or prediction deviation due to excessive range.

[0098] Time window definition: Basic window: fixed as the current time t0 → the next 3 hours (t0+3h). For example, if the current t0 = 8:00, then the window is 8:00-11:00.

[0099] Flexible window adjustment: If the forecast data shows that there are sudden events in the next hour, such as road closures or train cancellations, the window will be temporarily narrowed to t0→t0+2h to prioritize short-term emergency decisions; if the scenario is stable, such as no fluctuations in road conditions or train schedules, the window can be extended to t0+4h, but not exceeding 4h, to avoid a decrease in forecast accuracy.

[0100] Spatial window definition: Scope: Centered on the current transportation location (e.g., the truck's current latitude and longitude, and the connection points already reached), filter spatial nodes that can be reached within t0+3 hours, including: Highway segment: All highway segments that can be traveled within 3 hours from the current location, with the maximum distance calculated based on the predicted speed limit. For example, if the speed limit is 60km / h, the range is within 180km. Connection points: Road-rail connection points that can be reached within 3 hours, such as freight stations and railway marshalling yards; Railway train schedules: Trains departing from the connecting point within 3 hours of t0 and with available capacity (excluding those that are sold out or delayed by more than 1 hour).

[0101] Decision objectives within the window: Core objective: Focusing on "local optimum within 3 hours" while also considering its connection to the subsequent global path, specifically: Highway section: predicted to have the lowest congestion coefficient, ensuring smooth driving; Connection point: Available resources, including loading and unloading equipment and storage space, can be booked within 3 hours; Railway schedules: The time difference between departure and arrival of freight trains at the connecting point should be ≤30 minutes to reduce waiting time and avoid missing trains; Constraint: Transportation costs within 3 hours should be ≤ 30% of the current budget to avoid excessive local costs.

[0102] S42: Extraction and adaptation of predicted data within the window to match the data requirements of local decision-making.

[0103] From the dynamic database, the predicted data within the window (t0→t0+3h) is extracted and adapted according to the categories of highway-railway-connection point to ensure that the data can be directly used for local optimum solutions: Highway prediction data extraction: Extracted content: Predicted congestion coefficients for all accessible highway segments within the window (updated every 15 minutes), predicted time-varying speed limits, and construction / accident warnings.

[0104] Adaptation processing: The data is bound to specific road segments, generating a road segment ID-time slice-predicted status lookup table. See Table 2 below for an example: Table 2. Road Segment ID-Time Slice-Predicted Status Comparison Table Road segment ID <![CDATA[Time slice (t0→t0+3h)]]> Predicted congestion coefficient Predicted speed limit Alert status G101 8:00-8:30 0.3 (Unimpeded) 100km / h none G101 8:30-9:00 0.6 (Moderate congestion) 80km / h Construction on the road G202 8:00-11:00 0.2 (Unobstructed) 90km / h none Railway forecast data extraction: Extracted content: "Railway train ID, predicted departure time, predicted carriage availability, and predicted on-time rate" at the connection point within the window.

[0105] Allocation processing: Sort by connection point - departure time, filter out shifts with remaining capacity < cargo demand and on-time rate < 60%, and generate a list of connection points - available shifts, such as Table 3 below: Table 3. Road Segment ID-Time Slice-Predicted Status Comparison Table Connection point Class ID Predicted departure time Forecast margin On-time prediction rate C Station T1 8:45 40 tons 92% C Station T3 9:30 60 tons 88% D Station T2 10:15 35 tons 75% Data extraction for connection point prediction: Extracted content: "Predicted equipment load rate, predicted idle time slice, and predicted loading and unloading efficiency" of the connection points within the window.

[0106] Adaptation processing: Divide the device's idle time into "15-minute / segment" reservable units, and generate a connection point-device ID-reservable time segment table, such as Table 4 below: Table 4: Generation Connection Point - Device ID - Scheduled Time Slots Connection point Device ID Scheduled film Predicted load factor Predicting loading and unloading efficiency C Station Loading and unloading No. 3 8:30-9:00 0.4 (Idle) 25 tons / hour C Station Loading and unloading No. 3 9:00-9:30 0.6 (Busy) 20 tons / hour D Station Loading and unloading No. 1 9:15-9:45 0.3 (Idle) 30 tons / hour S43: Local optimal path solution, i.e., selecting the optimal solution within 3 hours based on the predicted data: An improved adaptive DP-GA algorithm is used to solve for the local optimal path within the spatiotemporal window boundary. The core is to match the time and resources of highway-connection point-railway.

[0107] S431: Abstract the transportation scenario within the window into a node-edge network, clarifying the relationships between each element: Nodes: Current location (S0), reachable connection points within the window (Station C, Station D), and train departure points (e.g., the departure node of train T1 at Station C). Edge: Highway edge: S0→C station (corresponding to G101 section, weight is "predicted travel time + predicted cost"), S0→D station (corresponding to G202 section, weight is the same as above). Transit route: Station C → T1 (weighted by "predicted transit time + resource reservation cost", and must meet the condition that "truck arrival time at Station C ≤ T1 departure time - 30 minutes"). Railway side: T1 train → next station (weighted by "predicted railway travel time", but only "match" needs to be determined, no need to calculate the whole journey, as subsequent windows will cover it); S432: GA generates a candidate route set by encoding the "highway-connection point-railway" combination within the window, generating 100-200 candidate solutions (individuals), for example... Candidate 1: S0→G101→C Station (arrives at 8:30)→Schedule loading and unloading No. 3 (8:30-9:00)→Take the T1 train (departs at 8:45); Candidate 2: S0→G202→D station (arrival at 9:15)→Reserved loading and unloading No. 1 (9:15-9:45)→Take the T2 train (departure at 10:15).

[0108] S433: DP calculates local objective values ​​with "shortest total time within the window and highest resource reservation success rate" as the local objective, and uses dynamic programming to calculate the objective value for each candidate: For candidate 1: Calculate the travel time of G101 (40 minutes, 8:00-8:40? This needs to be corrected. Departing at 8:00, the predicted speed limit of G101 is 60km / h. If it's 100km, it would take 1 hour and 40 minutes? No, the previous spatial window was within 3 hours, so the distance needs to match. Assuming that the distance from S0 to station C via G101 is 80km, the predicted average speed is 60km / h, and the travel time is 1 hour and 20 minutes, departing at 8:00 and arriving at 9:20, then it will miss the connecting T1 train departing at 8:45. Therefore, GA will eliminate candidates with such mismatched times). Revised Candidate 1: S0→G101 (8:00-9:00, 90km, average speed 60km / h)→C Station arrives at 9:00→Reserve loading and unloading No. 3 (9:00-9:30)→Take the T3 train (departs at 9:30), target value: total time 90 minutes (8:00-9:30), resource reservation success rate 100% (loading and unloading No. 3 is available).

[0109] S434: Filter local optima by sorting by "target value" and select the top 20% of candidates as the local optimum within the window. For example, the target value of candidate 2 is: travel time 1 hour 15 minutes (8:00-9:15), transfer time 30 minutes (9:15-9:45), total time 1 hour 45 minutes, and resource success rate 100%. If candidate 1 has a shorter total time (1 hour 30 minutes), then candidate 1 is the local optimum.

[0110] S44: Initial route plan generation: Integrating executable information from navigation, schedules, and reservations. The abstract results of the local optimal path are transformed into a structured solution that can be directly implemented, including specific execution information for highways, railways, and connection points, ensuring that dispatchers and equipment can use it directly.

[0111] Highway execution information: Navigation instructions with real-time warnings for road sections and lanes; for example: Main route: S0 → G101 (prioritize the leftmost lane; there is construction on section K10-K20 from 8:00 to 8:30, so you need to merge into the middle lane in advance) → North entrance of Station C (arrive before 9:00). Alternative route: If the congestion coefficient of G101 exceeds 0.7 (predicted to occur after 8:30), then switch to G102 (predicted congestion coefficient 0.2), and the navigation instructions will be updated accordingly.

[0112] Railway operation information: Detailed information on the primary and alternative train services, including delay response strategies, such as... Main train service: Station C, T3 (departs at 9:30, carriage number 3, remaining capacity 60 tons, predicted on-time rate 88%). Alternative train schedule: Station C, T4 (departs at 10:00, with a remaining capacity of 50 tons and a 95% on-time rate). If T3 is delayed by more than 15 minutes (predicted to be confirmed before 9:15), the train will automatically switch to T4. Reservation information: 30 tons of capacity (cargo requirement 30 tons) has been reserved through the railway system for train T3, reservation number RT20251103001.

[0113] Connection point execution information: Reservation voucher for equipment-time-operation requirements, such as: Reservation Resources: Loading and unloading equipment No. 3 at Station C (9:00-9:30, 30 minutes); Operational requirements: Trucks must park at Platform 2 of Station C (before 9:00 AM) and load / unload 30 tons of cargo (a cargo list must be submitted to the Station C system in advance). Emergency backup: If loading / unloading station 3 malfunctions, it will be automatically reassigned to loading / unloading station 4 (available from 9:00 to 9:30), and the reservation system will synchronize in real time; The initial path scheme output example is shown in Table 5 below: Table 5 Initial Path Scheme Output Table unit Specific content Execution deadline Emergency backup plan Highway navigation G101 (8:00-9:00) → Platform 2, Station C 9:00 Switch to G102 during traffic jams; the navigation terminal updates in real time. Railway train schedules T3 (Departs at 9:30, Station C → Station B, Carriage 3) 9:25 Delays exceeding 15 minutes will result in switching to the T4 train. Connection point reservation Loading and unloading station 3 (9:00-9:30, Platform 2, Station C), handling 30 tons. 9:30 Switch to loading / unloading No. 4 in case of malfunction Example 3 Based on Example 1 or Example 2, the specific process of step S5 is as follows: S51: Warning threshold setting: Define the deviation risk level based on the predicted value.

[0114] First, based on the core parameters of highways, railways, and connecting points, and combined with the business impact of the predicted values, the prediction deviation rate is converted into a three-level early warning threshold.

[0115] Deviation rate calculation: The core indicator for early warning is the deviation rate between actual data and predicted values. The formula is: Deviation rate = |(Actual value collected in real time − Predicted value from dynamic database) / Predicted value from dynamic database × 100%; For example: if the predicted congestion coefficient for highway section A is 0.5, and the actual value collected in real time is 0.6, then the deviation rate = |(0.6-0.5) / 0.5|×100%=20%; Three-level threshold settings for key parameters: Different parameters have varying degrees of impact on transportation and require fine-tuning based on the general three-level range (mild 10-20%, moderate 20-30%, severe ≥30%) to ensure the rationality of the warning. Specific parameter thresholds are shown in Table 6 below: Table 6 Parameter Threshold Table Data types Key monitoring parameters Minor warning (deviation rate) Moderate warning (deviation rate) Severe warning (deviation rate) Special Notes (Business Impact and Adaptation) Highway dynamic data Congestion coefficient 10%-20% 20%-30% ≥30% Congestion directly affects travel time, and the threshold is strictly implemented according to the general range. Highway dynamic data Time-varying speed limit 15%-20% 20%-35% ≥35% The impact of speed limit deviation is small (e.g., from 120 to 100, the deviation is 16.7%), with a slight increase in the threshold. Railway dynamic data On-time rate 10%-20% 20%-30% ≥30% Low on-time performance directly leads to missed trains; thresholds are strictly enforced. Railway dynamic data Carriage slack 15%-20% 20%-40% ≥40% A slightly low margin can be addressed through loading optimization, while increasing the heavy-duty threshold. Connection point data Loading and unloading equipment load rate 10%-20% 20%-30% ≥30% High load causes relay delays; thresholds are strictly enforced. Connection point data Warehouse turnover efficiency 15%-20% 20%-35% ≥35% The impact of efficiency deviation is small (e.g., from 20 to 17 tons / hour, deviation is 15%), with a slight increase in the threshold. S52: Real-time Data Monitoring and Deviation Calculation: Continuously Comparing Predicted Values ​​with Actual Values Based on the update frequency of the dynamic database, emergency freight is updated every 5-10 minutes, and regular freight is updated every 30-60 minutes. The actual values ​​of key parameters are collected in real time, and the deviation rate is calculated with the corresponding predicted values ​​to provide data input for early warning judgment.

[0116] Monitoring objects and frequency matching: High-frequency monitoring parameters (every 5-10 minutes): Highway congestion coefficient, railway on-time rate (affects travel time and train schedule connection, requires real-time tracking); Medium frequency monitoring parameters (30 minutes / time): Time-varying speed limit, load rate of equipment at connection point (changes are slow, no need for excessively frequent monitoring); Low-frequency monitoring parameters (60 minutes / time): Carriage capacity, warehouse turnover efficiency (small fluctuations in a short period of time, such as the carriage capacity not dropping from 50 tons to 20 tons within 10 minutes).

[0117] Deviation calculation process: Extract the predicted value corresponding to the current monitoring period from the dynamic database: for example, at 9:00, extract the predicted congestion coefficient of highway section A for the "9:00-10:00" window as 0.5; The actual congestion coefficient of Highway A at 9:00 AM is collected through the sensor / system interface: for example, the actual value is 0.6. The deviation rate is calculated using the general formula: |(0.6-0.5) / 0.5|×100%=20%; Record deviation data: associate with "parameter type, predicted value, actual value, deviation rate, and monitoring time", and store it in the deviation log library (for subsequent algorithm optimization).

[0118] S53: Warning Level Determination: The level is determined based on the deviation rate and parameter characteristics. Based on the deviation rate range and the parameter's differential threshold, determine the warning level for a single parameter; if multiple parameters deviate simultaneously, determine the final warning level according to the "highest level superposition" principle to avoid missing high-risk scenarios. See Table 7 below for examples of warning level determination: Table 7 Example of Early Warning Level Determination parameter Predicted value actual value Deviation rate Corresponding threshold range Single parameter warning level Congestion coefficient of Highway Section A 0.5 0.6 20% Moderate alert (20%-30%) moderate On-time rate of railway T3 train 0.9 0.72 20% Moderate alert (20%-30%) moderate Load rate of equipment at connection point 3 0.6 0.78 30% Severe warning (≥30%) Severe The core rule for multi-parameter overlay judgment is to choose the highest level over the lowest: when multiple parameters deviate simultaneously, the highest warning level is taken as the final level, and all deviation parameters are marked to facilitate subsequent targeted adjustments.

[0119] For example, the congestion coefficient of highway section A is 20% (moderate), the on-time rate of railway T3 train is 15% (slight), and the load rate of connecting points is 30% (severe). Final warning level: Severe (highest level). Labeling deviation parameters: Highway congestion (moderate), railway punctuality rate (slight), and connection point load (heavy).

[0120] S54: Tiered adjustment implementation: Match the minimum impact adjustment range according to the warning level.

[0121] The core principle is to adjust parameters for minor warnings, adjust local parameters for moderate warnings, and adjust global parameters for severe warnings, in order to avoid wasting resources due to large adjustments for small deviations, while ensuring thorough optimization of the solution in high-risk scenarios.

[0122] Minor warning: Only optimizes the parameters of the current road segment, without changing the path structure.

[0123] Triggering conditions: Single parameter deviation rate of 10%-20% or multiple parameters are slight and do not affect core nodes of the path, such as connection points or railway train schedules; Adjustment logic: Only adjust the time / cost parameters of the current route segment, without changing the route segment, schedule, or connection point; For example, scenario: the congestion coefficient of highway section A is predicted to be 0.5, the actual value is 0.55, the deviation is 10%, a slight warning is issued, and the original planned travel time is 1 hour; Adjustment: Corrected driving time = 1 hour × (1 + 10%) = 1.1 hours (i.e., 1 hour and 6 minutes). Fuel consumption calculation is updated accordingly; increased congestion → fuel consumption increases by 5%. No adjustments will be made: Highway section A will not be replaced with another section, and the plans for subsequent connecting point C station and railway T3 train service will remain unchanged.

[0124] Moderate warning: Reconstruct local paths, replace some nodes, and retain the core structure; Triggering conditions: Single parameter deviation rate of 20%-30% or multiple parameters with moderate deviation, affecting local path nodes (such as a section of highway or a connection point), but not affecting the global structure; Adjustment logic: Replace the local nodes affected by the deviation, and retain other nodes, such as replacing the congested road segment with the alternative road segment, or replacing the connection point with the nearby backup point; For example, scenario: the on-time rate of railway train T3 is predicted to be 0.9, but is actually 0.72, a deviation of 20%, which is a moderate warning. The original plan was to take train T3 from station C (departing at 9:30). Adjustment: The initial route "Starting point → Highway A section → Station C" is retained (no deviation); Replacement train schedule: Station C T3 (deviation) → Station C T4 (departs at 10:00, on-time rate 0.95, no deviation); Synchronous adjustment of connection point resources: The original reservation for loading and unloading at Station C from 9:00 to 9:30 has been adjusted to 9:30 to 10:00 (matching the departure time of the T4 train). No adjustments will be made: the starting point and ending point will not be changed, and the A section and C stations of the highway will not be replaced (the core connection points will be retained).

[0125] Severe alert: Initiate global replanning, reconstruct the entire path, and cover all nodes; Triggering conditions: Single parameter deviation rate ≥ 30% or multiple parameters with severe deviation, and affecting core nodes of the global path such as failure of major connection points or suspension of main railway lines.

[0126] Adjust the logic: Re-execute the entire process of S2 (spatiotemporal coupling objective function) - S3 (fitness calibration) - S4 (rolling window solution) to generate a brand new path scheme; For example, scenario: equipment failure at connection point C station, load rate prediction 0.6, actual 1.2, deviation 100%, severe warning, original plan start point → section A → station C → T3 → station D → end point; Adjustment: Global replanning trigger: Abandon the original plan and rebuild the objective function based on the latest data (C station is faulty, alternative D station is available).

[0127] New route solution: Solve using the S4 rolling window (current time t0→t0+3h) to generate a new solution: starting point → B segment (alternative highway) → E station (alternative connection point) → T5 bus (departure from E station) → F station → destination.

[0128] Full resource synchronization: The new plan is synchronized to freight navigation (updating the B section route), railway system (booking T5 trains), and E-station (reserving loading and unloading times).

[0129] The specific process of step S6 is as follows: S61: Determine the evaluation criteria and scope: Evaluation benchmark: Using the original route plan before adjustment as the benchmark, extract its core indicators as the basis for comparison, including: Total time of the original plan ( T 原 ), time for each stage (highway) T 原 Railway T 原铁 transit T 原中 ); Total cost of the original plan ( C 原 Costs at each stage (highway) C 原公 ,railway C 原铁 transit C 原中 ); The original plan relied on core resources (such as the original highway sections, original railway schedules, and original connecting point equipment).

[0130] Evaluation object: The adjusted new path plan, which needs to have its corresponding indicators (T) extracted in advance. 新 C 新 To ensure consistency in indicator dimensions, if the original plan calculates the total time, the new plan must also calculate it using the same caliber.

[0131] Define the scope of the evaluation: Focus on adjusting the core aspects, avoid redundant evaluation of unchanged parts, and improve efficiency: For minor warning adjustments (optimizing only highway parameters): only assess changes in highway time / cost, not railway or transfer links; If the adjustment is a moderate warning (replacing railway train services): assess the time / cost changes of the railway segment and related transfer links; the unchanged highway segment is not assessed. If it is a severe early warning adjustment (overall replanning): assess the time and cost changes of the entire process of highways, railways and transfers, and the feasibility of all new resources; Set evaluation qualification standards (adapted to business needs): Set differentiated standards based on transportation type (emergency / routine). The standards need to be embedded into the system in advance to avoid manual adjustments. See the table below: Table 8 Evaluation Qualification Standards Evaluation Dimensions Regular freight (cost priority) Urgent shipping (time priority) illustrate Time variation rate ≤15% (acceptable delay) ≤5% (Strictly control delays) Change rate = (T new - T original) / T original × 100%. For urgent freight shipments, excessive time increases are not permitted. Cost change rate ≤10% (Strict cost control) ≤20% (cost can be negotiated) Change rate = (Cnew - Coriginal) / Coriginal × 100%. For regular freight transport, excessive cost overruns are not permitted. Feasibility score ≥70 points (out of 100) ≥80 points (out of 100) If the score is below the standard, the plan needs to be readjusted. S62: Calculation of time / cost change rate, quantifying the benefits and costs of adjustment: Based on the evaluation benchmark and the object, the total change is broken down into changes in each link, which not only calculates the overall change rate, but also locates the source of the change, making it easier to determine whether the overall overspending is caused by local adjustments.

[0132] Calculation of the rate of change over time: First, calculate the time difference of each stage of change, then sum them up to obtain the total time difference, and finally calculate the rate of change. The formula is consistent as follows: The rate of change of time in a certain stage = (the new time of that stage) T 新环节 - Original time of this step T 原环节 ) / T 原环节 ×100%; Total time change rate = (new total time) T 新 - Original total time T 原 ) / T 原 ×100%; Table 9. Examples of Adjustments to Moderate Warning Levels Link <![CDATA[Original solution (T 原 )]]> <![CDATA[New solution (T 新 )]]> Changes in process Link change rate Total time change rate railway section 18 hours (T3) 19 hours (T4) +1 hour +5.6% - transit 1.5 hours (Station C) 2 hours (Station C) +0.5 hours +33.3% - Highway section (unchanged) 3 hours 3 hours 0 0% - Total Time 22.5 hours 24 hours +1.5 hours +6.7% +6.7% Analysis of the reasons for the changes: The source of the changes should be marked simultaneously. For example, the railway section time is increased because the T4 train has a longer travel distance; the transfer time is increased because the T4 train departs late and the waiting time at the connection point needs to be extended, so as to provide a basis for decision-making.

[0133] Cost change rate calculation (same time logic, by stage + overview): Calculation logic: Consistent with the formula structure of the rate of change over time, only replaced with a cost indicator: Cost change rate at a certain stage = (new cost at that stage) C 新环节 - Original cost of this stage C 原环节 ) / C 原环节 ×100%; Rate of change in total cost = (new total cost) C 新 - Original total cost C 原 ) / C 原 ×100%; Table 10 Cost Change Rate Calculation Table Link <![CDATA[Original solution (C 原 )]]> <![CDATA[New solution (C 新 )]]> Changes in process Link change rate Total cost change rate railway section 4000 yuan (T3) 4200 yuan (T4) +200 yuan +5% - transit 800 yuan (C station) 900 yuan (C station) +100 yuan +12.5% - Highway section (unchanged) 1200 yuan 1200 yuan 0 0% - Total cost 6000 yuan 6300 yuan +300 yuan +5% +5% S63: Feasibility Assessment: Verify the feasibility of the solution from multiple dimensions. Meeting time / cost targets does not guarantee the feasibility of the plan (e.g., there are available tickets on trains but no equipment at connecting points). The plan needs to be scored from four core dimensions: "capacity, resources, collaboration, and risk," with a maximum score of 100. The weight of each dimension is adjusted according to the type of transportation ("capacity matching" has a higher weight for emergency freight).

[0134] Overall feasibility score calculation: The scores from each dimension are weighted and summed, expressed as: Overall score = ∑(score of a certain dimension × weight of that dimension); Table 11 Example of Comprehensive Feasibility Calculation Overall score: 28.5 + 22.5 + 20 + 17 = 88 points, meeting the standard of ≥70 points for regular freight and ≥80 points for emergency freight.

[0135] Comprehensive decision-making: Determining whether the plan "meets the standards and is executable". By combining time, cost variation rate and feasibility score, a decision to pass or fail is made based on the evaluation criteria.

[0136] The decision-making rules are shown in the table below: Table 12 Decision Rules Assessment Result Combination Decision Conclusion Handling method The rate of change meets the standard + the feasibility score meets the standard. pass Entering the solution execution phase, synchronous cross-system... Change rate meets target + feasibility score fails to meet target Not approved Feedback was sent to the adjustment phase, where resources were re-optimized (such as replacing equipment at connection points or adding alternative shifts). Change rate not up to standard + feasibility score up to standard Not approved Feedback is sent to the adjustment phase, where routes are readjusted (e.g., choosing routes / frequencies with shorter travel times / lower costs). Neither of them met the standards. Not approved Revert to S5 to re-trigger the adjustment mechanism, or re-execute global replanning (in case of severe alerts).

Claims

1. A dynamic planning method for multimodal transport routes (road and rail), characterized in that, Includes the following steps: S1: Real-time acquisition of dynamic data on highways, railways, and connecting points; prediction of various data within a specified time period using an LSTM model, generating a dynamic database containing the current state and predicted values, and performing data preprocessing; S2: Construct a spatiotemporally coupled objective function. The objective function takes the shortest total transportation time, the lowest total transportation cost, and the highest efficiency of road-rail connection as the collaborative objectives, and the objective weights are dynamically adjusted according to real-time data. S3: Introduce a fitness dynamic calibration mechanism based on a dynamic database. Based on a genetic algorithm, the objective function value is adjusted in each generation of the population according to the latest prediction data. S4: Starting from the current moment, take a specified future time period as the decision window, solve the local optimal path based on the predicted data within the window, and update the window over time to generate an initial path planning scheme that includes real-time road navigation path, dynamic railway schedule matching, and time slot reservation for connection point resources. S5: Based on the predicted values ​​of the dynamic database, a three-level early warning threshold is set. When the monitored data reaches the corresponding early warning level, the graded adjustment mechanism is triggered to generate an adjustment path planning scheme. S6: Conduct an impact assessment on the adjusted path planning scheme, including the time / cost change rate and execution feasibility assessment. Once all assessment results meet the corresponding requirements, execute the adjusted path planning scheme. The specific expression for the spatiotemporal coupling objective function F in step S2 is as follows: F= ω t (t)· T 总 ( s,t )+ ω c (t)· C 总 ( s,t )+ ω e ( s,t )·(1− E 衔接 ( s,t )); Where F is the overall target value of spatiotemporal coupling; the smaller the value, the better the scheme. ω t (t): Time weight, which is dynamically adjusted with time t, increasing during peak periods; ω c (t): Cost weight, adjusted according to resource prices during transportation periods and peak railway freight rates; ω e ( s,t ): Connection efficiency weight, varying with the connection point. s Real-time resource load and time t Adjustments made when connection points are congested. ω e Increase; T 总 ( s,t Total transport time, depending on spatial path s and time t; C 总 ( s,t Total transportation cost, depending on spatial path s and time t; E 衔接 ( s,t ): The efficiency of the connection between road and rail depends on the connection point s and time t, and the value ranges from [0,1]. The larger the value, the higher the efficiency. In the objective function, "1-E connection" represents the "efficiency loss", which needs to be minimized. Total transport time in the spatiotemporal coupling objective function F T 总 ( s,t The details are as follows: The total transportation time is the sum of the road segment, the rail segment, and the transfer time at the connecting points. The time of each link changes dynamically with the spatial path and time. The specific expression is as follows: ； in, T 公路 ( L i ,v i ( t ) ,t () represents the highway segment time, indicating the first... i The transport time of a section of highway depends on its spatial length. L i and real-time speed v i ( t ); T 公路 ( L i ,v i ( t ) ,t )= L i / v i ( t )·(1+ γ 拥 ( t ))； in γ 拥 ( t The congestion factor is calculated from the real-time congestion factor. T 铁路 ( D j ,f j ( t ) ,t () represents the railway segment time: indicating the first... j The transport time of a railway segment depends on spatial distance. D j and real-time train speed f j ( t (Including waiting time for each shift), the specific expression is: T 铁路 ( D j ,f j ( t ) ,t )= D j / f j (t)+ max (0,t 班计划 −t 到达 ); in, t 班计划 For the scheduled departure time of the train, t 到达 The arrival time of the truck at the connection point is determined by its spatial location, reflecting the spatiotemporal coupling. T 中转 ( s k ,q k ( t ) ,t () represents the transit time: indicating the first... k Connection points s k The transit time depends on the spatial resources of the connection point and the real-time job queue. q k ( t ); T 中转 ( s k ,q k ( t ) ,t )= Q / r k ( t )+ q k (t)· τ k ； Where Q is the quantity of goods. r k ( t The real-time loading and unloading efficiency at the connection point varies with the equipment load. τ k This represents the average processing time for a single batch of goods. Total transportation cost C 总 ( s,t This represents the sum of road, rail, and transfer costs. The cost is strongly coupled with resource consumption and time-sensitive pricing along spatial routes. The specific expression is as follows: ； in, C 公路 ( L i ,p 油 ( t ) , t) represents the highway cost, indicating the cost of the first... i The cost of the highway segment includes fuel costs and toll fees; C 公路 ( L i ,p 油 ( t ) , t)= L i ·( a·p 油 ( t )+ b · δ ( t ))； in, a Fuel consumption per unit mileage b Toll fee per unit distance δ (t) represents the time-period premium coefficient; C 铁路 ( Q,u j ( t ) ,t (This refers to railway costs, in relation to freight volume) Q Real-time unit freight rate u j ( t Related; C 铁路 ( Q,u j ( t ) ,t )=Q· u j ( t )·(1+ θ j ( t ))； θ j ( t The capacity tension coefficient is represented by ( ). C 中转 ( Q,s k ,t k () represents the transit cost, indicating the cost of the first... k Cost at each connection point, and the volume of goods. Q Connection point s k Resource prices and occupancy time t k Related; C 中转 ( Q,s k ,t k )= Q·c k + max (0, t k −t 基 )· d k ； in c k Loading and unloading fees for the unit. t 基 Based on free time, d k Overtime fee; Efficiency of road-rail connection E 衔接 ( s,t ): Reflects the smoothness of the transition between road and rail transport, and is strongly coupled with the spatial resource utilization and time matching degree of the connection point. Expression: E 衔接 ( s,t )= α / (1+ T 等待 ( s,t ))+(1− α )·(1− ρ k ( t ))； in, T 等待 ( s,t () refers to the waiting time at the connection point, i.e., the interval between the arrival of the freight car and its departure from the railway. ρ k ( t The real-time load rate of the connection point equipment is 0-1, varying with time. t change; α α is the weighting coefficient, 0 < α < 1, balancing time matching degree and resource utilization rate.

2. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 1, characterized in that, The specific process of data preprocessing in step S1 is as follows: S11: Define the scenario constraint boundaries for each data type, that is, define the absolutely reasonable range and fluctuation warning range for each parameter as the basis for anomaly identification; S12: Perform a two-dimensional screening by combining statistical patterns and scenario constraints, and mark preliminary abnormal data; S13: For the initial abnormal data, perform secondary verification by calling the multi-source verification data source and dynamically adjusted verification rules according to the scenario to which it belongs, distinguish between true anomalies and false anomalies, correct the true abnormal data, and retain and mark the false abnormal data.

3. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 1, characterized in that, The specific process of step S3 is as follows: S31: Population individual coding, i.e., the gene sequence that defines the route plan: converting the road-rail intermodal transport route plan into GA-identifiable individuals, with each individual corresponding to a complete transport plan; S32: Initial fitness calculation, i.e. the first evaluation based on the current data: After population initialization, the initial fitness of each individual is calculated based on the current data in the dynamic database as the starting point for iteration; S33: Dynamic data-triggered calibration. After each selection-crossover-mutation to generate a new population, GA first checks whether the dynamic database has been updated. If the latest predicted data exists, fitness calibration is triggered. S34: Objective function parameter correction, updating the core indicators of individuals: For the population that triggered calibration, extract the genetic information of each individual, i.e. path and time, and combine it with the latest prediction data to correct the key parameters of its objective function.

4. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 3, characterized in that, The specific process of step S34 is as follows: S341: Correction to total transport time Ttotal: Highway segment time correction: updating driving speed based on predicted congestion coefficient; Railway time adjustment: Update train delays based on predicted punctuality rates; Transfer time correction: Update loading and unloading efficiency based on predicted equipment load at connection points; S342: Total Transportation Costs C 总 Correction: Highway costs: adjusted based on predicted fuel prices and congestion period premiums; Railway costs: Adjusting capacity premiums based on predicted carriage availability; Transit costs: Timeout fees adjusted based on predicted waiting time; S343: Adjust connection efficiency based on predicted waiting time and device load.

5. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 1, characterized in that, The specific process of step S4 is as follows: S41: Decision window initialization: Define the spatiotemporal boundaries of local optimization: Starting from the current moment, clarify the time range of the window and the core decision objective, and define the boundaries of local optimization for subsequent solutions; S42: Predictive data extraction and adaptation within the window: Extract the predicted data within the window from the dynamic database and adapt it according to the categories of highway-railway-connection point; S43: Local optimal path solution, i.e., selecting the optimal solution within the window based on the predicted data; S44: Based on the optimal solution, integrate the executable information of navigation, schedule, and reservation to generate the initial route plan.

6. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 5, characterized in that, The specific process of step S43 is as follows: S431: Abstract the transportation scenario within the window into a node-edge network and clarify the relationships between the elements; S432: Encode the highway-connection point-railway combination within the window to generate 100-200 candidate solutions; S433: Using dynamic programming to calculate the target value for each candidate, with the shortest total time and the highest resource reservation success rate as the local objectives; S434: Sort by target value and select the top 20% of candidates as the local optimum within the window.

7. The method for dynamic planning of multimodal transport routes (road and rail) according to claim 1, characterized in that, The specific process of step S5 is as follows: S51: Warning threshold setting: Define the deviation risk level based on the predicted value; S52: Real-time data monitoring and deviation calculation: Continuously compare predicted values ​​with actual values; S53: Warning level determination: The level is determined based on the deviation rate and parameter characteristics; S54: Tiered adjustment implementation: Match the minimum impact adjustment range according to the warning level.

8. The method for dynamic planning of multimodal transport routes (rail-road) according to claim 1, characterized in that, The specific process of step S6 is as follows: S61: Determine the evaluation criteria, scope, and standards for the adjusted route planning scheme; S62: Calculation of time and cost change rates, quantifying the benefits and costs of adjustments; S63: Conduct an execution feasibility assessment and determine whether the adjusted route planning scheme meets the requirements for execution. If so, execute the adjusted route planning scheme; otherwise, repeat steps S2-S6.

Images