Dynamic disturbance-oriented emergency material multimodal transport intelligent scheduling method

Abstract

The invention relates to the technical field of emergency logistics engineering and intelligent traffic control, and relates to a dynamic disturbance-oriented emergency material multimodal transport intelligent scheduling method, which comprises the following steps of: A, constructing a multimodal transport dynamic space-time network based on physical and logic mapping, and opening up a data space in a memory by a processor based on a time extension graph technology; b, building an emergency material multimodal transport dynamic scheduling optimization model under a rolling time domain framework, building a continuous dynamic scheduling model based on the rolling time domain framework, and decomposing long-period global optimization into a series of short-period sub-problems which are mutually associated; c, designing a double-layer decision-making framework according to a self-adaptive large-field search algorithm fused with reinforcement learning under a rolling time domain framework: processing dynamic disturbance by adopting the rolling time domain framework on an outer layer, and designing a self-adaptive large-field search algorithm fused with a Q learning mechanism on an inner layer; the method has the advantage of self-adaptive evolution capability with higher and higher disaster evolution decision speed and higher and higher quality.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of emergency logistics engineering and intelligent traffic control, and particularly relates to an emergency material multimodal transport intelligent scheduling method for dynamic disturbance. BACKGROUND

[0002] The instantaneousness, high destructiveness and strong uncertainty of the emergency event pose a severe test to the urban traffic lifeline, especially in the "golden window" period of rescue, the efficiency and quality of emergency decision-making are directly related to life and property safety. Therefore, efficient and reliable scheduling of emergency materials is a key technical link to ensure the success of rescue. The traditional scheduling of emergency materials usually relies on a single transportation mode, however, in extreme disaster situations, the single mode of transportation system is easily damaged, the network topology is interrupted, forming a traffic "island", which leads to the failure of timely delivery of emergency materials and affects the efficiency of rescue.

[0003] To solve the problem of low reliability of single transportation mode, the existing research adopts the multimodal transport mode, which integrates the advantages of various transportation modes such as railway, highway and aviation, and utilizes the complementary advantages between various modes to build a more reliable comprehensive transportation network system to provide protection for emergency rescue.

[0004] However, in the application of multimodal transport to emergency rescue in actual combat, the existing technology faces a serious "calculation-execution space-time dislocation" technical problem, mainly reflected in the following two aspects: First, the transience of physical topology and the hysteresis of the model. The existing multimodal transport model is mostly a static network, which abstracts the transportation network as a fixed topology structure. When the road in the physical world collapses, the transport capacity is damaged or the node is paralyzed, the model cannot perceive and "prune" the corresponding infeasible logical link in real time, resulting in that the generated scheduling instructions are physically unexecutable.

[0005] Second, the "cold start" bottleneck of rolling time domain calculation. The existing scheduling model often uses the rolling time domain framework, but at each decision window update, each time window is usually considered as an independent problem to calculate from zero "cold start", and the calculation state of the previous window is discarded. In the scene of rapid spread of disaster and dynamic change of information, such repeated calculation will lead to "strategy forgetting" (unable to inherit the search experience of the previous time domain) and "response lag" (time-consuming of repeated calculation), so that the material transport vehicles miss the best rerouting opportunity, resulting in resource mismatch and time delay.

[0006] Chinese patent 202411086330.7 discloses an emergency material multimodal transport method based on sudden situation. In the face of sudden situation, the demand of disaster point, traffic condition has high uncertainty, and the demand is often large and urgent, and the traffic network may be interrupted. A single transport tool is difficult to meet the requirements of multi-transport in sudden situation. An emergency material multimodal transport method based on sudden situation is proposed to solve the problem that the optimization solving method of the multimodal transport scheme in the prior art does not highlight the uncertainty of the sudden situation and the urgency of the emergency material demand, resulting in low transportation efficiency and high cost.

[0007] However, in the face of the core challenge of dynamic disturbance of emergency material multimodal transport in sudden events, the modeling paradigm of existing research often assumes that the transport network topology is static and unchangeable, which is difficult to cope with dynamic disturbances such as demand dynamic emergence, road network interruption and transport capacity damage.

[0008] The existing method DOI: 10.27790 / d.cnki.gahgy.2021.000467 discloses "Emergency Material Dispatching Optimization Based on Improved Genetic Algorithm", in which scholar Yang Bing proposes an improved two-stage algorithm based on the idea of complete feasible region of recovery problem for multi-supply point dispatching optimization problem with inconsistent multiple target attributes in emergency material dispatching. In the face of emergency material multi-supply point problem, instead of simplifying it into multiple single-supply point problems, an idea of considering the complete feasible region of the problem is proposed, and a variable-length genotype improved genetic algorithm is designed to solve it.

[0009] The existing method DOI: 10.27322 / d.cnki.gsgyu.2024.000290 discloses "Research on Emergency Material Dispatching with Time Window under Demand Uncertainty", in which scholar Zhang Zhixin constructs a three-level dispatching network of emergency materials with time window under demand uncertainty, establishes an emergency material dispatching model with multimodal transport, and solves it by using NSGA multi-objective optimization algorithm.

[0010] Existing research often abstracts complex emergency dispatching as a static single optimization problem, i.e. planning a full-cycle scheme based on known information at the initial stage of emergency. However, this static assumption is seriously inconsistent with the dynamic disturbance characteristics of road network interruption and demand emergence in emergency practice. In addition, when solving in a rolling time domain framework, each time window is usually treated as an independent problem and calculated from zero, which leads to the technical bottlenecks of "strategy forgetting" (unable to inherit the search experience of the previous time domain) and "response lag" (repeated calculation time-consuming). SUMMARY

[0011] In view of the above problems, the purpose of the present application is to provide an emergency material multimodal transport intelligent scheduling method for dynamic disturbance, which is used for mapping the dynamic disturbance of the physical road network into the data model topology constraint in real time, and using the historical time domain search experience to accelerate the dynamic scheduling method of the current decision, so as to overcome the shortcomings of the prior art.

[0012] The present application provides an emergency material multimodal transport intelligent scheduling method for dynamic disturbance, comprising the following steps: A. Constructing a multimodal transport dynamic space-time network based on physical and logical mapping, The processor opens a data space in the memory based on the time expansion graph technology, constructs the data structure of the multimodal transport space-time network, and includes establishing the mapping between the physical entity and the data object; B. Construction of an emergency material multimodal transport dynamic scheduling optimization model under a rolling time domain framework, A continuous dynamic scheduling model based on a rolling time domain framework is constructed, the global optimization of a long period is decomposed into a series of interrelated short period sub-problems, the real-time updated space-time network state is used as input, and the dynamic optimization and real-time correction of the emergency material transport process are realized; C. Adaptive large domain search algorithm based on reinforcement learning under a rolling time domain framework, A double-layer decision-making framework is designed: the outer layer uses the rolling time domain framework to process dynamic disturbance, and the inner layer designs an adaptive large neighborhood search algorithm combined with a Q learning mechanism, through an intelligent operator selection strategy based on environment feature perception and cross-time domain learning, the cross-time domain inheritance of optimization experience and the adaptive evolution of decision-making strategy are realized.

[0013] As a preferred embodiment of the present application, the step A further comprises the following steps: A1. Construction of an emergency material multimodal transport static network, An emergency multimodal transport static network is generated: For the multimodal transport network , Indicates the set of all city nodes including the starting point, the ending point and the transfer point, Indicates the set of transportation arcs in the network, Indicates the set of transportation modes, and there are multiple transportation modes between adjacent two city nodes; in order to represent the three-level network attribute, among them Is divided into three mutually exclusive subsets: Indicates the set of supply points, Indicates the set of transfer centers, Indicates the set of demand points, and different functional attributes are given, As a set of supply points of emergency materials, it has the attributes of material inventory and transportation capacity; As a set of material demand points, it has the attributes of material demand quantity and the time window for the arrival of required materials; As a collection of transit centers, it has attributes such as material inventory, transportation capacity, and transshipment capability.

[0014] As a preferred embodiment of the present invention, step A further includes the following step: A2. Construction of a spatiotemporal network model for multimodal transport of emergency supplies under dynamic disturbances. A21. Time discretization processing Discretizing time is a prerequisite for constructing a spatiotemporal network. It requires dividing the entire timeframe of emergency dispatch into a series of discrete time points, defining the timeframe as... ; Select a time step Representing the set of time points as And denoted as the set of integers. ,common Each time point; A22. Expand nodes, Copy each physical node in the static network. The spatiotemporal nodes in the secondary spatiotemporal network are used to assign spatiotemporal attributes, in addition to the original static network. , This represents the physical location of a node, including supply points, demand points, and transit centers; within an extended spatiotemporal network. ,in It is a tuple , representing at a moment physical nodes ; A23. Spatiotemporal network power arc instantiation and dynamic mapping mechanism The arcs in the original physical network are transformed into arcs representing activities in the spatiotemporal network. Arcs in the spatiotemporal network are divided into three types: traveling arcs, waiting arcs, and transit arcs. (1) Travel arc: The processor generates spatiotemporal nodes that connect different physical locations. The data link is used to characterize the spatial displacement of supplies, representing the location of emergency supplies. From the node The journey begins, using transportation methods. ,spend Time, in Delivered to the node in real time ,in ; Dynamic processing for road condition / weather disturbances: The processor is equipped with a time-varying impedance update interface. When it receives road sensor or meteorological data indicating that a certain road segment is experiencing [unclear - possibly a problem], it updates the interface accordingly. When the traffic capacity decreases at any given time, the processor updates the travel time attribute corresponding to that road segment. If an interruption signal indicating a road collapse is received, the processor will update all start times. Furthermore, the weight of the travel arc passing through this road segment is marked as infinite, which is used to block path search at the logic layer; Dynamic handling of timetable disturbances: The processor performs conditional instantiation based on real-time timetable data for transportation vehicles, and only instantiates vehicles if the timetable data indicates that the vehicle is at a certain time. When available transportation capacity, the processor generates the corresponding travel arc object in the memory; if a trip cancellation signal is received, the processor immediately deletes the corresponding travel arc from the data structure using an index. (2) Waiting arc: The processor generates nodes that connect adjacent time points at the same physical location. The data link is used to characterize the storage status of supplies, indicating the location of emergency supplies. The time period stays at the node Place; Dynamic handling of node storage damage disturbances: The processor reads the storage monitoring data of each node in real time. If it receives a node... When a signal of warehouse damage or capacity saturation occurs, the processor locates the waiting arc in the spatiotemporal network starting from that node and forcibly resets its capacity to 0, preventing the transportation plan from choosing to store materials at that node. (3) Transfer arc: The processor generates connections between nodes at different times within the same transfer center. The data link is used to characterize the operation process across transportation modes, with a transfer arc connecting the same physical transfer center node. Two spatiotemporal nodes at different times are used to represent the location of emergency supplies. During the specified time period, transshipment will take place at the transfer center, by the mode of transportation. Convert to transportation mode ; Dynamic handling of node operation capacity disturbances: The processor establishes a real-time linkage mechanism between transfer arc attributes and transfer center equipment status. 1) When a power outage or loading / unloading equipment failure signal is received, the processor identifies the affected time window and lowers or sets the transfer capacity attribute of the transfer arc within the corresponding time period of the time window to zero. 2) When a congestion queuing signal is received, the time cost attribute of the transfer arc is dynamically increased using the Road Impedance Function (BPR) logic. At the model level, the transfer delay is simulated to guide subsequent route planning to avoid highly congested nodes.

[0015] As a preferred embodiment of the present invention, step B further includes the following step: B1. Objective function In emergency rescue scenarios, the primary goal is to minimize the weighted total completion time of all tasks, i.e. the material delivery time, in order to maximize rescue efficiency. B2. Constraints It is used to ensure that the generated scheduling scheme is feasible in the physical network, holds true in time logic, and can respond to dynamic disturbances in real time; (1) Task flow conservation and path connectivity constraints, Used to accomplish each emergency task In the spatiotemporal network, a continuous and complete transportation chain is formed, starting from the supply point and finally reaching the demand point, to prevent materials from being lost or interrupted at intermediate nodes and to force each task to be executed completely. (2) Dynamic resource capacity coupling constraints, This is used to limit the total amount of material transported by all tasks in any spatiotemporal arc to no more than the real-time available capacity of that arc, so that when dynamic disturbances cause the capacity to decrease or return to zero, the unreachable transport path is directly blocked at the logical level. (3) Task indivisibility constraint, This is used to limit the transportation of a standard mission package to a single stage, without splitting it into multiple shipments. (4) Path uniqueness and state mutual exclusion constraints, For the same task, it can only be in one specific physical state at any given time; (5) Hard time window constraint, Used to explicitly exclude all path schemes that exceed the deadline.

[0016] As a preferred embodiment of the present invention, step C further includes the following step: C1. Rolling Time Domain System Initialization and Status Acquisition C11. System initialization, Set the execution time window length for the rolling time domain And specify the planning time domain as Load the multimodal transport spatiotemporal network model constructed in step A and initialize the global Q-value table. It is used for long-term memory of intelligent operator selection strategies across time domains; C12. Dynamic status information acquisition, (1) Basic state update: at the current moment of each decision cycle The rolling time-domain system collects real-time data through an IoT interface and maps it to a spatiotemporal network state. ,include: 1) Requirement status update: Identify unfinished tasks from the previous time domain, add newly emerging tasks in the current time domain to the task pool, and identify the dynamic urgency of each task. ; 2) Resource status update: Real-time inventory of materials at each node, current spatiotemporal coordinates of transportation vehicles, and remaining transport capacity; 3) Network status update: Based on road collapse or traffic control information, the impedance of the corresponding travel arc in the spatiotemporal network is set to infinity; based on the node congestion, the travel time of the transfer arc is dynamically updated using the road resistance function BPR. (2) Environmental characteristic calculation: Constructing three-dimensional characteristic indicators, 1) Task Scale Characteristics : Calculate the ratio of the total number of tasks to be scheduled in the current task pool to the total number of available transportation vehicles in the network, reflecting the load pressure of the current scheduling problem; 2) Supply freedom characteristics : Traverse the tasks to be scheduled and calculate the average number of supply points that each task can reach in the current residual network. The lower the average number of supply points, the more significant the single-source strong constraint. 3) Resource Conflict Index A quick rehearsal is performed based on the spatiotemporal decoder to count the number of overlaps of the shortest paths of all tasks on the spatiotemporal arcs, and to calculate the proportion of the number of congested arcs to the total number of used arcs; the higher this proportion, the more intense the right-of-way competition among multiple tasks.

[0017] As a preferred embodiment of the present invention, step C further includes the following step: C2. Use the adaptive large-domain search algorithm to solve the dynamic scheduling model. C21. Encoding and Shortest Path Decoding Based on Spatiotemporal Networks (1) Indirect encoding: Using priority-based integer encoding, a task sequence The service priority order representing all tasks to be scheduled; (2) Spatiotemporal decoding: Design a greedy construction algorithm based on spatiotemporal state updates. State synchronization: Initialize the spatiotemporal network and synchronize the current resource usage of each node; Sequential planning: based on the task sequence In the order of extraction, tasks are retrieved from the spatiotemporal network. ; Optimal mapping: for the task Considering the current network's remaining resource constraints, a dynamic Dijkstra algorithm with a time window is used to search for a generalized minimum-cost path from the supply point to the demand point in the spatiotemporal network. Resource deduction: After the path is determined, based on the dynamic resource capacity coupling constraint, the remaining capacity of the arc segments along the path is immediately updated in the temporary spatiotemporal network. This is used to ensure that subsequent low-priority tasks meet the constraints. C22. Initial solution generation, (1) First time domain: weighted according to the urgency of each task. Generate an initial priority sequence from largest to smallest; (2) Subsequent time domain: Design a hot start strategy, Solution inheritance and pruning: Reading the previous decision cycle The optimal scheduling scheme, eliminating those already in use. Task instructions that are to be completed within a specified time period; Spatiotemporal state projection: Projecting the positions and resource states of vehicles with remaining unexecuted tasks onto the current moment. On the spatiotemporal network nodes, it serves as the skeleton of the current solution; Incremental insertion repair: For newly emerging tasks in the current time domain, a greedy insertion strategy is used to insert them into the idle time slots of the inherited solution as the initial solution in the current time domain; C23. Setting up a library of destruction / repair operators based on perturbation feature awareness. (1) Destruction operator, Random Removal: Random Selection One task removed; Delay / Worst Time Removal: Remove the task with the latest completion time or the highest delay penalty; Bottleneck section / node perception and removal: Real-time reading of the transit arc congestion coefficient and driving arc impedance value in the spatiotemporal network, and targeted removal of transportation tasks containing high impedance sections or congested nodes; (2) Repair operator, Greedy insertion: Insert the removal task at the position that minimizes the increment of the objective function; Regretful insertion: Prioritize inserting the task with the maximum regret value, which is defined as the difference between the objective function value when the task is inserted at the current best position and the second-best position; Modal perturbation insertion: When re-inserting a task, the repair operator forcibly disables the main transportation mode used in the previous iteration of the task. By leveraging the modal complementarity advantage of multimodal transport, it can quickly find an alternative mode when the physical path is interrupted. C24. Intelligent operator selection strategy based on environmental feature perception and cross-temporal learning. Design an environment-aware reinforcement learning mechanism and use a processor to maintain a global operator utility database stored in non-volatile memory; (1) Definition of multidimensional state space: Construct a composite state vector containing the search internal state and the features of the external environment. ; Search internal status : Describes the current convergence behavior of the adaptive large-domain search algorithm, discretized into the exploration phase, development phase, and stagnation phase; External environmental characteristics Discretize and map the extracted three-dimensional feature indicators: Task size Low load, high load; Supply freedom Strong constraints, weak constraints; Resource Conflict Index No conflict, high congestion; (2) Action definition: The action is to select a pair of operators. ; (3) The reward function is defined as follows: ; in ; (4) Update the Q-value in the Q-learning mechanism. (5) Cross-temporal inheritance of knowledge and its relation to the environment. In the global operator utility database, the Q value is determined not only by the operator but also by the environmental features index; Cross-time domain migration mechanism: when time changes from Scroll to At that time, the rolling time-domain system directly reads the historically accumulated high Q-value strategy; C25. Main algorithm flow, (1) Initialization: Based on the initial solution generated in step C22, load... ; (2) Main loop: State recognition: Determine whether the current state is exploration, development, or stagnation; Operator selection: Select operator pairs based on the Q-value table. ; Neighborhood search: Performs destruction and repair operations to obtain new chromosomes. ; Decoding evaluation: Calling the spatiotemporal decoder to calculate The objective function value; Acceptance Criterion: The simulated annealing criterion is used to determine whether to accept a new solution; Knowledge update: Calculate the reward based on the result and update the Q value in the corresponding state; (3) Termination: If the time or iteration limit is met, the optimal scheduling scheme is output, and the current Q value table is saved, thus completing the main flow of the scheduling plan algorithm.

[0018] As a preferred embodiment of the present invention, step C further includes the following step: C3. Implementation of the plan and monitoring of disturbances. The scheduling plan generated in step C25 is executed and dynamic disturbance information of external environmental characteristics is continuously received.

[0019] As a preferred embodiment of the present invention, step C3 further includes the following step: If new emergency needs arise or there are sudden changes in the road network condition, wait until the current decision-making cycle. At the end, time advances to the beginning of the next decision-making cycle. The updated rolling time-domain system state and Q-value table are returned to step C12 to enter the next rolling time-domain decision loop, thereby achieving continuous adaptation and optimized scheduling to the dynamic environment.

[0020] The beneficial effects of this invention are as follows: 1. This invention achieves real-time physical response: By establishing a real-time mapping mechanism (dynamic blocking / degradation) between physical disturbance events and spatiotemporal network topology parameters, infeasible paths are directly "physically cut off" at the logical layer, solving the technical problem that static networks cannot respond to dynamic emergencies such as road network interruptions and capacity loss, and ensuring that the generated scheduling instructions are physically executable.

[0021] 2. This invention achieves high efficiency and continuity in decision-making: By utilizing spatiotemporal state projection technology, the effective path of the previous time domain is transformed into the initial skeleton of the current time domain, realizing the "hot start" of the algorithm and avoiding the "cold start" and repeated calculation problems in traditional rolling time domain calculation, thus significantly reducing decision delay.

[0022] 3. This invention achieves environmentally adaptive evolutionary capability: By constructing a global Q-value table, cross-temporal knowledge transfer is realized. The system can learn "experience in dealing with congestion or modal disruption" in historical periods and directly transfer it to similar future scenarios, enabling the system to have an adaptive evolutionary capability that makes decisions faster and with higher quality as disasters evolve.

[0023] 4. This invention achieves deep modal collaboration: The designed "modal disturbance insertion operator" can force tasks to switch between different modes of transportation, giving full play to the modal complementarity advantage of multimodal transport when some road networks are paralyzed, and improving the overall robustness of the emergency logistics system. Attached Figure Description

[0024] Other objects and results of the invention will become more apparent and readily understood with reference to the following description taken in conjunction with the accompanying drawings. In the drawings: Figure 1 This is a flowchart of the scheduling method according to an embodiment of the present invention. Detailed Implementation

[0025] Example 1

[0026] See Figure 1 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] This invention provides an intelligent scheduling method for multimodal transport of emergency supplies oriented towards dynamic disturbances, comprising the following steps: A. Construct a multimodal transport spatiotemporal network based on physical and logical mapping. The processor, based on time-spread graph technology, allocates data space in memory and constructs the data structure of the multimodal transport spatiotemporal network, including establishing the mapping between physical entities and data objects. A1. Construction of a static network for multimodal transport of emergency supplies Generate an emergency multimodal transport static network: For multimodal transport networks , This represents the set of all city nodes, including the origin, destination, and transfer points. Represents the set of transport arcs in the network. This represents the set of various transportation modes, and there are multiple transportation modes between two adjacent city nodes; to characterize the attributes of the three-level network, it is divided into... It is split into three mutually exclusive subsets: Represents the set of supply points. Indicates a collection at a transit center. This represents a set of requirements and assigns different functional attributes to them. As a collection of emergency supplies supply points, it has attributes such as material inventory and transportation capacity; As a set of material demand points, it has the attributes of material demand quantity and the time window for the arrival of required materials; As a collection of transit centers, it has attributes such as material inventory, transportation capacity, and transshipment capability.

[0028] A2. Construction of a spatiotemporal network model for multimodal transport of emergency supplies under dynamic disturbances. A21. Time discretization processing Discretizing time is a prerequisite for constructing a spatiotemporal network. It requires dividing the entire timeframe of emergency dispatch (e.g., the critical 72 hours for rescue) into a series of discrete time points, defining the timeframe as... ; Select a time step Representing the set of time points as And denoted as the set of integers. ,common Each time point; A22. Expand nodes, Copy each physical node in the static network. The spatiotemporal nodes in the secondary spatiotemporal network are used to assign spatiotemporal attributes, in addition to the original static network. , This represents the physical location of a node, including supply points, demand points, and transit centers, but does not include time-related information; in the extended spatiotemporal network... ,in It is a tuple , representing at a moment physical nodes For example, (transfer center X, t=5) and (transfer center X, t=6) are two completely different nodes in the spatiotemporal network. A23. Spatiotemporal network power arc instantiation and dynamic mapping mechanism The arcs in the original physical network are transformed into arcs representing activities in the spatiotemporal network. Arcs in the spatiotemporal network are divided into three types: traveling arcs, waiting arcs, and transit arcs. (1) Travel arc: The processor generates spatiotemporal nodes that connect different physical locations. The data link is used to characterize the spatial displacement of supplies, representing the location of emergency supplies. From the node The journey begins, using transportation methods. ,spend Time, at every moment, is delivered to the node. ,in ; Dynamic processing for road condition / weather disturbances: The processor is equipped with a time-varying impedance update interface. When it receives road sensor or meteorological data indicating that a certain road segment is experiencing [unclear - possibly a problem], it updates the interface accordingly. When the traffic capacity decreases at any given time, the processor updates the travel time attribute corresponding to that road segment. If an interruption signal indicating a road collapse is received, the processor will update all start times. Furthermore, the weight of the travel arc passing through this road segment is marked as infinite, which is used to block path search at the logic layer; Dynamic handling of timetable disturbances: The processor instantiates conditions based on real-time timetable data of transportation vehicles (such as trains and flights), and only instantiates conditions when the timetable data indicates that the timetable is affected. When available transportation capacity, the processor generates the corresponding travel arc object in the memory; if a trip cancellation signal is received, the processor immediately deletes the corresponding travel arc from the data structure using an index. (2) Waiting arc: The processor generates nodes that connect adjacent time points at the same physical location. The data link is used to characterize the storage status of supplies, indicating the location of emergency supplies. The time period stays at the node Place; Dynamic handling of node storage damage disturbances: The processor reads the storage monitoring data of each node in real time. If it receives a node... exist If a signal of warehouse damage or capacity saturation occurs at any time, the processor locates the waiting arc in the spatiotemporal network starting from that node and forcibly resets its capacity to 0, preventing the transportation plan from choosing to store materials at that node. (3) Transfer arc: The processor generates connections between nodes at different times within the same transfer center. The data link is used to characterize the operation process across transportation modes, with a transfer arc connecting the same physical transfer center node. Two spatiotemporal nodes at different times are used to represent the location of emergency supplies. During the specified time period, transshipment will take place at the transfer center, by the mode of transportation. Convert to transportation mode ; Dynamic handling of node operation capacity disturbances: This is a key step in solving the core bottleneck of multimodal transport. The processor establishes a real-time linkage mechanism between the transfer arc attributes and the status of transfer center equipment. 1) When a power outage or loading / unloading equipment failure signal is received, the processor identifies the affected time window and lowers or sets the transfer capacity attribute of the transfer arc within the corresponding time period of the time window to zero. 2) When a congestion queuing signal is received, the time cost attribute of the transfer arc is dynamically increased using the Road Impedance Function (BPR) logic. At the model level, the transfer delay is simulated to guide subsequent route planning to avoid highly congested nodes.

[0029] B. Construction of an emergency material multimodal dynamic scheduling optimization model under a rolling time domain framework: A continuous dynamic scheduling model based on a rolling time domain framework is constructed, which decomposes the long-cycle global optimization into a series of interrelated short-cycle sub-problems. The real-time updated spatiotemporal network state is used as input to realize dynamic optimization and real-time correction of the emergency material transportation process. B1. Objective function In emergency rescue scenarios, the primary goal is to minimize the weighted total completion time of all tasks, i.e. the material delivery time, in order to maximize rescue efficiency. ; in: This represents the weighted total completion time for all tasks; This represents the set of emergency tasks to be scheduled. For task indexing; Indicates task The urgency weight; This represents the set of all arcs (including traveling arcs, waiting arcs, and transit arcs) in a spatiotemporal network. For arc segment index; Representing spacetime arc Time cost; As a decision variable, if the task The transportation path includes a spacetime arc. If the value is 1, then the value is 1; otherwise, it is 0.

[0030] B2. Constraints The optimization model constructed in this scheme uses the following five sets of constraints to ensure that the generated scheduling scheme is feasible in the physical network, holds true in time logic, and can respond to dynamic disturbances in real time. (1) Task flow conservation and path connectivity constraints, This is the core constraint of the model, used to achieve each emergency task. In the spatiotemporal network, a continuous and complete transportation chain is formed, starting from the supply point and finally reaching the demand point, to prevent materials from being lost or interrupted at intermediate nodes and to force each task to be executed completely. ; in: Representing a spatiotemporal network Any spacetime node in the universe; Represented by node The set of outflow arcs starting from; Represented by node The set of inflow arcs ending at; Indicates task The set of starting nodes in a spatiotemporal network; Indicates task Arrive at the set of nodes in the spatiotemporal network; (2) Dynamic resource capacity coupling constraints, This is used to limit the total amount of material transported by all tasks in any spatiotemporal arc to no more than the real-time available capacity of that arc, so that when dynamic disturbances cause the capacity to decrease or return to zero, the unreachable (infeasible) transport path is directly blocked at the logical level. ; in: Indicates the first The quantity of resources included in each task; Representing spacetime arc The remaining available capacity represents the remaining transport capacity for the travel arc, the remaining storage capacity for the waiting arc, and the remaining transit throughput capacity for the transit arc.

[0031] (3) Task indivisibility constraint, To ensure the completeness of emergency supplies and improve the efficiency of multimodal transport transshipment, this model is used to limit the transport of a standard mission package to a single stage without splitting it up. ; (4) Path uniqueness and state mutual exclusion constraints, For the same task, you can only be in one specific physical state at any given time (e.g., you cannot be in two different locations at the same time, or use two different modes of transportation at the same time). ; (5) Hard time window constraint, Although the objective function already includes time costs, it is used to explicitly exclude all path options that exceed the deadline in order to meet the stringent time requirements of emergency rescue.

[0032] ; in: Representing spacetime arc The endpoint of spacetime; Representing spatiotemporal nodes The corresponding time attribute; Indicates task The latest allowed delivery time, this constraint indicates that any arc selection that would cause a task delay exceeding the threshold will have its decision variable locked to 0.

[0033] C. An Adaptive Large Neighborhood Search Algorithm (QL-ALNS) Integrating Reinforcement Learning within a Rolling Temporal Framework: A two-layer decision architecture is designed. The outer layer uses a rolling temporal framework to handle dynamic perturbations, while the inner layer designs an adaptive large neighborhood search algorithm (QL-ALNS) integrating a Q-learning mechanism. Through an intelligent operator selection strategy based on environmental feature perception and cross-temporal learning, the system achieves cross-temporal inheritance of optimization experience and adaptive evolution of decision strategies, thereby endowing the rolling temporal system with the ability to continuously learn at an accelerating pace as disasters evolve.

[0034] C1. Rolling Time Domain System Initialization and Status Acquisition C11. System initialization, Set the execution time window length for the rolling time domain And specify the planning time domain as Load the multimodal transport spatiotemporal network model constructed in step A and initialize the global Q-value table. It is used for long-term memory of intelligent operator selection strategies across time domains; C12. Dynamic status information acquisition, (1) Basic state update: at the current moment of each decision cycle The rolling time-domain system collects real-time data through an IoT interface and maps it to a spatiotemporal network state. ,include: 1) Requirement status update: Identify unfinished tasks from the previous time domain, add newly emerging tasks in the current time domain to the task pool, and identify the dynamic urgency of each task. ; 2) Resource status update: Real-time inventory of materials at each node, current spatiotemporal coordinates of transportation vehicles, and remaining transport capacity; 3) Network status update: Based on road collapse or traffic control information, the impedance of the corresponding travel arc in the spatiotemporal network is set to infinity; based on the node congestion, the travel time of the transfer arc is dynamically updated using the road resistance function BPR. (2) Environmental Feature Calculation: To support intelligent decision-making in C24, a three-dimensional feature index is constructed. 1) Task Scale Characteristics : Calculate the ratio of the total number of tasks to be scheduled in the current task pool to the total number of available transportation vehicles in the network, reflecting the load pressure of the current scheduling problem; 2) Supply freedom characteristics : Traverse the tasks to be scheduled and calculate the average number of supply points that each task can reach in the current residual network. The lower the average number of supply points, the more significant the single-source strong constraint. 3) Resource Conflict Index A quick preview is performed based on the C21 spatiotemporal decoder to count the number of overlaps of the shortest paths of all tasks on spatiotemporal arcs, and to calculate the proportion of congested arcs to the total number of used arcs; the higher this proportion, the more intense the right-of-way competition among multiple tasks.

[0035] C2. Use the Adaptive Large Neighbor Search Algorithm (QL-ALNS algorithm) to solve the dynamic scheduling model. C21. Encoding and Shortest Path Decoding Based on Spatiotemporal Networks (1) Indirect encoding: Using priority-based integer encoding, a task sequence The service priority order representing all tasks to be scheduled; (2) Spatiotemporal decoding: In order to transform the priority sequence into a specific scheduling scheme, this invention designs a greedy construction algorithm based on spatiotemporal state updates. State synchronization: Initialize the spatiotemporal network and synchronize the current resource usage of each node; Sequential planning: based on the task sequence In the order of extraction, tasks are retrieved from the spatiotemporal network. ; Optimal mapping: for the task Considering the constraints of the remaining network resources (capacity, inventory, and transshipment capacity), the dynamic Dijkstra algorithm with time window (a single-source shortest path algorithm based on a greedy strategy) is used to search for a generalized minimum cost path from the supply point to the demand point in the spatiotemporal network. Resource deduction: After the path is determined, based on the dynamic resource capacity coupling constraint, the remaining capacity of the arc segments along the path is immediately updated in the temporary spatiotemporal network. This is used to ensure that subsequent low-priority tasks meet the constraints. C22. Initial solution generation, (1) First time domain: weighted according to the urgency of each task. Generate an initial priority sequence from largest to smallest; (2) Subsequent Time Domains: To avoid response delays caused by starting calculations from zero in each time domain, a warm-start strategy is designed. Solution inheritance and pruning: Reading the previous decision cycle The optimal scheduling scheme, eliminating those already in use. Task instructions that are to be completed within a specified time period; Spatiotemporal state projection: Projecting the positions and resource states of vehicles with remaining unexecuted tasks onto the current moment. On the spatiotemporal network nodes, it serves as the skeleton of the current solution; Incremental insertion repair: For newly emerging tasks in the current time domain, a greedy insertion strategy is used to insert them into the idle time slots of the inherited solution as the initial solution in the current time domain; C23. Setting up a library of destruction / repair operators based on perturbation feature awareness. (1) Destroy Operators. Random Removal: Random Selection One task removed; Delay / Worst Time Removal: Remove the task with the latest completion time or the highest delay penalty; Bottleneck section / node perception and removal: Real-time reading of the transit arc congestion coefficient and driving arc impedance value in the spatiotemporal network, and targeted removal of transportation tasks containing high impedance sections or congested nodes; (2) Repair Operators. Greedy insertion: Insert the removal task at the position that minimizes the increment of the objective function; Regret-k insertion: Prioritize inserting the task with the maximum regret value, which is defined as the difference between the objective function value of the task when it is inserted at the current best position and the second best position; Modal perturbation insertion: To address the problem that multimodal transport is prone to getting trapped in local optima of a single mode of transport (such as over-reliance on roads), this repair operator forcibly disables the main mode of transport used in the previous iteration of the task when re-inserting the task (e.g., forcibly switching from road to rail / air). By taking advantage of the modal complementarity of multimodal transport, it can quickly find alternative modes when physical paths are interrupted, rather than searching ineffectively in a single network that has been interrupted. C24. Intelligent operator selection strategy based on environmental feature perception and cross-temporal learning. To enable the algorithm to intelligently match the best search strategy based on different disaster scenarios (such as "congestion period" or "sparse period") and the state of the search process, an environment-aware reinforcement learning mechanism is designed, and a global operator utility database stored in non-volatile memory is maintained by the processor. (1) Definition of multidimensional state space: Construct a composite state vector containing the search internal state and the features of the external environment. ; Search internal status : Describes the current convergence behavior of the adaptive large-domain search algorithm, discretized into the exploration phase, development phase, and stagnation phase; External environmental characteristics Discretize and map the extracted three-dimensional feature indicators: Task size Low load, high load; Supply freedom Strong constraints (few choices), weak constraints (many choices); Resource Conflict Index No conflict, high congestion; Technical effects: For example, when the state is identified as [stagnation stage, high congestion], the system can learn that the "bottleneck removal operator" and "regretful insertion operator" should be called first to break out of the infinite loop; (2) Action definition: The action is to select a pair of operators. ; (3) The reward function is defined as follows: ; in ; (4) Update the Q-value in the Q-learning mechanism. ; in Includes the current search status and environmental characteristics. The learning rate is determined based on the current search state when selecting operators. The following combinations of operators Value adopted The -greedy strategy is selected, and after each time-domain solution is completed, the updated Q-value table is saved as... For use in the next time domain; (5) Cross-temporal inheritance of knowledge and its relation to the environment. In the global operator utility database Q-Table, the Q value is determined not only by the operator, but also by the environment feature index; Cross-time domain migration mechanism: when time changes from Scroll to Even if the task changes, if the environmental characteristics are similar (e.g., both are in a state of "high congestion and strong constraints"), the rolling time-domain system can directly read the high Q-value strategies accumulated in the past. This gives the algorithm the ability to identify scenarios, that is, the experience of dealing with congestion learned in the past can be directly transferred to similar scenarios in the future. C25. Main algorithm flow, (1) Initialization: Based on the initial solution generated in step C22, load... ; (2) Main loop: State recognition: Determine whether the current state is exploration, development, or stagnation; Operator selection: Select operator pairs based on the Q-value table. ; Neighborhood search: Performs destruction and repair operations to obtain new chromosomes. ; Decoding evaluation: Calling the spatiotemporal decoder to calculate The objective function value; Acceptance Criterion: The simulated annealing criterion is used to determine whether to accept a new solution; Knowledge update: Calculate the reward based on the result and update the Q value in the corresponding state; (3) Termination: If the time or iteration limit is met, the optimal scheduling scheme is output, and the current Q value table is saved, thus completing the main flow of the scheduling plan algorithm.

[0036] C3. Execution of the plan and monitoring of disturbances: Execute the scheduling plan generated in step C25 and continuously receive dynamic disturbance information of external environmental characteristics.

[0037] If new emergency needs arise or there are sudden changes in the road network condition, wait until the current decision-making cycle. At the end, time advances to the beginning of the next decision-making cycle. The updated rolling time-domain system state and Q-value table are returned to step C12 to enter the next rolling time-domain decision loop, thereby achieving continuous adaptation and optimized scheduling to the dynamic environment.

[0038] Example 2 To fully demonstrate the dynamic scheduling capabilities in scenarios with multiple concurrent tasks and limited resources, this embodiment constructs a scheduling scenario containing three differentiated emergency tasks, and explains in detail how the system in this embodiment performs multi-task collaboration and dynamic resource allocation.

[0039] 1. Scene initialization and task definition, 1.1 Physical network configuration, The local physical network constructed in this embodiment contains 4 key nodes. : (1) All of these are supply points. .

[0040] (2) Core transshipment node It has the capability for "road-rail" transshipment, but its maximum transshipment throughput is limited, set at 200 tons / hour.

[0041] (3) Demand points in severely affected areas .

[0042] (4) Collection of transportation modes Includes: highways (medium speed, prone to obstruction), railways (high speed, fixed timetable), and air travel (fastest speed, but extremely limited capacity).

[0043] 1.2 Set of tasks to be scheduled Define three emergency tasks with different characteristics. To test the priority sorting and resource allocation logic of the model: Task , Supplies contents: Emergency blood plasma; Origin and end points: arrive ; Material quantity 50 tons; Deadline T=8; Urgency weight :10 (Extremely high, must be delivered as quickly as possible, no delays allowed).

[0044] Task , Contents of the supplies: Drinking water; Origin and end points: arrive ; Material quantity 150 tons; Deadline T=12; Urgency weight 5 (Medium, high demand, high transportation capacity).

[0045] Task , Items included: disaster relief tents; Origin and end points: arrive

[0046] Material quantity 100 tons; Deadline : ; Urgency weight :1 (lower, less time-sensitive, can give way to high-priority tasks when necessary).

[0047] 1.3 Spatiotemporal network resource constraints setting, (1) Road section - (Highway): The normal travel time is 3 hours, and the transport capacity is sufficient.

[0048] (2) Road section - (Railway): Only and Trains depart regularly, with a journey time of 2 hours. The maximum capacity of a single train is limited to 100 tons.

[0049] (3) Road section - (Railway main line): Sufficient transport capacity, travel time is 3 hours.

[0050] Initial scheduling scheme generation (time T=0). At the initial moment, the system runs the QL-ALNS algorithm in step C of Example 1. Under the premise of satisfying all constraints, with the objective of minimizing the weighted completion time, the following initial scheme is generated: (1) Task (Emergency plasma) dispatch plan: path: --[Railway (t=1 train trip)]--> --[Costume Change]--> --[Railway]--> .

[0051] Reasons for the decision: Railways are fast, and The quantity of supplies is small (50 tons), but it can be made to... The early morning train (the train had 50 tons of remaining capacity, which was just enough).

[0052] Expected results: Arrive (satisfy) ).

[0053] (2) Task (Drinking water) scheduling plan: path: --[Highway]--> --[Costume Change]--> --[Railway]--> .

[0054] Reasons for the decision: The railway freight capacity has been It occupies 50 tons, and the remaining transport capacity is insufficient to load 150 tons. According to the "task cannot be indivisible constraint", Road transport is the only option.

[0055] Expected results: Arrive (satisfy) ).

[0056] (3) Task (Tent) Dispatch Plan: path: --[Highway]--> --[Costume Change]--> --[Railway]--> .

[0057] Expected results: Arrive (satisfy) ).

[0058] 3. Dynamic disturbance occurs ( time), The system runs to At that moment, all tasks had just begun. Suddenly, a secondary disaster occurred, and the monitoring module reported the following disturbances: (1) Disturbance events: to The railway embankment was damaged, and train services were cancelled; meanwhile, to The highway experienced severe congestion, with travel time surging from 3 hours to 6 hours.

[0059] (2) Model Response: The processor triggers the physical-logical mapping, which will... time to Railway travel arc capacity Set to 0; to Generalized cost of highway driving arc Updated to 6.

[0060] 4. Multi-task collaborative rescheduling in the rolling time domain. System entry At the decision-making window, due to railway disruptions and highway congestion, the original plan is no longer feasible or will cause severe delays. The system triggers rescheduling, and the core algorithm embodies the intelligent decision-making logic of "prioritizing higher-priority services and abandoning lower-priority ones." (1) Conflict identification and destruction. For the task (High Efficiency): The original train service has been cancelled, and a new route must be chosen. If the route is changed to a congested highway, the estimated arrival time will become 1 (current) + 6 (road) + 1 (transfer) + 3 (rail) = 11, exceeding the deadline. This is unacceptable.

[0061] For the task (Zhongyou): Originally, we planned to take the highway, but now the highway travel time is longer, which will also cause delays.

[0062] (2) Intelligent Repair and Resource Preemption The algorithm invokes the smart operator in step C24 of Example 1, and performs the following adjustments based on the experience of the global Q-value table: Strategy Adjustment A (for) ): The algorithm search found a small rescue helicopter (with a carrying capacity of only 60 tons) available from... direct flight However, the cost is extremely high. Because... Extremely high weight The objective function tends to use high-cost resources to ensure timeliness.

[0063] New decision: Abandon ground transportation and switch to direct air transport .

[0064] Strategy Adjustment B (for) and Resource game): This freed up ground resources. But when Reaching the node via congested highway Time (expected) Arrival), exactly with from The task brought in They clashed in terms of timing.

[0065] Node resource conflict: Time, node The ability to change outfits is limited. If processed simultaneously... (150 tons) and (100 tons), the total of 250 tons will exceed the upper limit of 200 tons.

[0066] Global collaborative optimization: According to weight (Drinking water is more urgent than tents), the algorithm determines the priority. .

[0067] Decision outcome: exist Immediately change clothes; and the mission Forced at the node Stop for one time step on the "waiting arc" to change clothes at off-peak times.

[0068] 5. Finally, revise the scheduling instructions. After calculation, the system outputs the updated scheduling instruction Plan_002: (1) Task (Plasma): Perform a mode switch. Immediately switch to direct helicopter flight. .

[0069] Execution result: Arrived. Successfully avoided congestion, meeting extremely high urgency requirements.

[0070] (2) Task (Water): Path maintained, accepting partial delays. Continue on congested roads, but at the junction... They have priority in changing clothes.

[0071] Execution result: Arrived. Although later than originally planned, the deadline constraint was still met. ).

[0072] (3) Task (Tent): Voluntarily gave way. Arrived via the original route. Afterwards, wait in the warehouse for 1 hour. Change clothes first.

[0073] Execution result: Arrived. Deadline constraint satisfied. ).

[0074] 6. Implementation effect analysis, This embodiment clearly demonstrates the three core advantages of this method: First, multimodal robustness. When high-priority tasks face the dilemma of both public transportation and rail disruptions or severe congestion, the algorithm can automatically activate backup modes (such as aviation) to avoid such situations.

[0075] Second, multi-task collaboration. When node resources are limited, the algorithm strictly follows the principle of "weight-based global optimality," sacrificing the time of low-priority tasks (such as allowing...). Waiting) to ensure high-priority tasks (such as...) ) circulation efficiency.

[0076] Third, constraint adaptability. The entire rescheduling process strictly adheres to physical constraints such as the indivisibility of resources and the upper limit of node capacity, resulting in a highly executable solution.

[0077] 7. Simulation experiments and performance evaluation, 7.1. Parameter Settings Network size: Set the number of nodes Number of tasks .

[0078] Dynamic disturbance scenario: simulation in Road disruptions and demand surges can occur randomly at any time.

[0079] 7.2. Comparison of experimental results Table 2. Performance comparison of different algorithms in dynamic scenarios Rating indicator GA Traditional ALNS QL-ALNS Average solution time (sec) 42.5 15.8 7.6 Weighted total completion time (hours) 2840.5 2415.2 2108.6 Task on-time delivery rate (%) 66% 73% 88% Average disturbance response speed (sec) 35.2 8.4 1.8 7.3. Results Analysis (1) Solution efficiency analysis: As can be seen from Table 2, this embodiment, due to the introduction of the rolling time domain "solution inheritance and pruning" mechanism (hot start), has a much faster replanning speed (1.8 seconds) when dynamic disturbances occur than the traditional algorithm. This means that in actual disaster relief, the command center can obtain the corrected solution almost in real time.

[0080] (2) Optimization capability analysis: Thanks to the intelligent selection of operators by reinforcement learning (Q-table mechanism), the algorithm in this embodiment can automatically select more destructive operators to escape local optima according to the "high congestion" environment. Therefore, the final weighted total completion time is shorter and the task delivery rate is the highest.

[0081] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent scheduling of emergency supplies multimodal transportation oriented to dynamic disturbance, characterized in that, Comprising the following steps: A. Constructing a multimodal dynamic space-time network based on physical and logical mapping, The processor opens up data space in the memory based on time expansion graph technology, constructs the data structure of the multimodal dynamic space-time network, including the establishment of the mapping between physical entities and data objects; B. Construction of an emergency material multimodal dynamic scheduling optimization model under a rolling horizon framework, A continuous dynamic scheduling model based on a rolling horizon framework is constructed, which decomposes the long-period global optimization into a series of interrelated short-period sub-problems, uses the real-time updated space-time network state as input, and realizes dynamic optimization and real-time correction of the emergency material transportation process; C. Adaptive large-domain search algorithm integrating reinforcement learning under a rolling horizon framework, A double-layer decision-making framework is designed: the outer layer uses a rolling horizon framework to handle dynamic disturbances, and the inner layer designs an adaptive large-neighborhood search algorithm integrating a Q-learning mechanism, which realizes the cross-time-domain inheritance of optimization experience and the adaptive evolution of decision-making strategies through intelligent operator selection strategies based on environmental feature perception and cross-time-domain learning. 2.The method of claim 1, wherein, In step A, the following steps are also included: A1. Construction of an emergency material multimodal static network, Generate emergency multimodal transport static network: For multimodal transport network , denotes the set of all city nodes including the starting point, the ending point and the transfer point, denotes the set of transport arcs in the network, denotes the set of transport modes, and there are multiple transport modes between adjacent city nodes; To represent the properties of the three-level network, among them is divided into three mutually exclusive subsets: denotes the set of supply points, denotes the set of transfer centers, denotes the set of demand points, and has different functional attributes, as the supply point set of emergency supplies, has the attributes of material inventory and transport capacity; as a set of material demand points, has the attributes of material demand and required material arrival time window; as a set of transfer centers, has the attributes of material inventory, transport capacity and transfer capacity. 3.The method of claim 2, wherein, In step A, the following steps are also included: A2. Construction of an emergency material multimodal dynamic space-time network model for dynamic disturbances, A21. Time discretization processing, Discrete time is the prerequisite of constructing space-time network, which needs to divide the whole time range of emergency dispatch into a series of discrete time points, and the time range is defined as ; Select a time step The set of time points is denoted as and as the set of integers , with time nodes; A22. Extended node, copying each physical node in a static network spatial node in a spatio-temporal network, for attributing spatio-temporal attributes, in the original static network , represents a physical location of a node, including supply points, demand points and transit centers; in the extended spatio-temporal network wherein is a tuple representing the physical node at time ;​ A23. Space-time network arc instantiation and dynamic mapping mechanism, Convert the original physical network arc segment into an arc representing activity in the space-time network. The arcs in the space-time network are divided into three types: travel arcs, waiting arcs, and transfer arcs; (1) Travel arc: The processor generates spatiotemporal nodes that connect different physical locations. The data link is used to characterize the spatial displacement of materials, representing the location of emergency supplies. From the node The journey begins, using transportation methods. ,spend Time, in Delivered to the node in real time ,in ; Dynamic processing for road condition / weather disturbance: the processor is configured with a time-varying impedance update interface, when receiving road sensor or meteorological data indicating that the road section is in the moment of reduced traffic capacity, the processor updates the travel time attribute corresponding to the road section , if the interruption signal of road collapse is received, the processor marks the weight of all starting time and travel arcs passing through the road section as infinite, to block the path search at the logical layer; Dynamic handling of schedule perturbation: the processor instantiates a condition based on the real-time schedule data of the transportation means, and only when the schedule data indicates that the transportation means is available at the time The processor generates the corresponding travel arc object in the memory of the storage only when the transportation means is available at the time; if a cancellation signal is received, the processor immediately deletes the corresponding travel arc from the data structure by indexing. (2) Waiting arc: The processor generates nodes that connect adjacent time points at the same physical location. The data link is used to characterize the storage status of supplies, indicating the location of emergency supplies. The time period stays at the node Place; Dynamic processing of node storage damaged disturbance: the processor reads the storage monitoring data of each node in real time, and if it receives a node In case of warehouse damage or capacity saturation signal, the processor locates the waiting arc in the space-time network starting from the node, and forcibly resets its capacity attribute to 0, forcing the transportation scheme to be unable to choose to stay at the node. (3) Transfer Arc: the processor generates data links connecting nodes at different times in the same transfer center , used to represent the operation process across transport modes, a transfer arc connects two space-time nodes at different times in the same physical transfer center node , used to represent the reloading operation of emergency materials at the transfer center within a certain period of time , converted from transport mode to transport mode ; Dynamic handling of node operation capacity disturbance: the processor establishes a real-time linkage mechanism between the transfer arc attribute and the transfer center equipment state: 1) When receiving a power interruption or loading and unloading equipment failure signal, the processor identifies the affected time window and downgrades or sets to zero the transfer capacity attribute of the transfer arc in the time window corresponding period; 2) When receiving a congestion queuing signal, dynamically increase the time cost attribute of the transfer arc using the road resistance function BPR logic to simulate the delay of transfer and guide subsequent path planning to avoid high congestion nodes.

4. The method of claim 1, wherein, In step B, the following steps are also included: B1. Objective function, In the emergency rescue scenario, the primary goal is to minimize the weighted total completion time of all tasks, i.e., the material delivery time, to maximize rescue efficiency; B2. Constraint conditions, Used to make the generated scheduling scheme feasible in the physical network and logically valid in time, and to respond to dynamic disturbances in real time; (1) Task flow conservation and path connectivity constraints, To implement each emergency task In the space-time network, a continuous and complete transport chain is formed from the supply point to the demand point, preventing the loss or interruption of materials at intermediate nodes, and forcing each task to be completed. (2) Dynamic resource capacity coupling constraints, Used to limit the total amount of material transportation of all tasks on any space-time arc segment to be less than the real-time available capacity of the arc segment, so that when dynamic disturbance causes capacity to decrease or zero, the transportation path that cannot be reached is directly blocked at the logical level; (3) Task indivisibility constraints, Used to limit the transportation of a standard task package within a single transportation phase, (4) Path uniqueness and state mutual exclusion constraints, Used to limit a task to be in only one determined physical state at the same time; (5) Hard time window constraints, For explicitly excluding all paths that exceed the cut-off time.

5. The dynamic-disturbance-oriented intelligent scheduling method for emergency material multimodal transport according to claim 1, characterized in that, In step C, the following steps are also included: C1. Rolling horizon system initialization and state collection, C11. System initialization, Setting the execution time window length of the rolling time domain , and the planning time domain is , load the multimodal transport space-time network model constructed in step A, initialize the global Q value table , for storing the long-term memory of intelligent operator selection strategy across time domains; C12. Dynamic state information collection, (1) Basic state update: At each decision cycle current time , the rolling horizon system collects real-time data through the Internet of Things interface and maps it into a space-time network state , including: 1) Demand state update: identify the unfinished tasks from the previous time horizon and add the newly emerged tasks in the current time horizon to the task pool, and identify the dynamic urgency of each task ; 2) Resource state update: real-time inventory of each node, current spatiotemporal coordinates of transportation tools, and remaining capacity; 3) Network state update: according to road collapse or traffic control information, set the corresponding travel arc impedance in the spatiotemporal network to infinity; according to the node congestion situation, use the road resistance function BPR to dynamically update the travel time of the transfer arc; (2) Environmental feature calculation: construct three-dimensional feature indicators, 1) Task size feature : Calculate the ratio of the total number of tasks to be scheduled in the current task pool and the total number of transport tools available in the network, reflecting the load pressure of the current scheduling problem; 2) Supply freedom feature : traverse the tasks to be scheduled, calculate the average value of the number of supply points that each task can reach in the current residual network, the lower the average value, the more significant the single-source strong constraint; 3) Resource conflict index : Based on the space-time decoder, a fast rehearsal is performed once, the number of overlaps of the shortest paths of all tasks on the space-time arc segments is counted, and the proportion of the number of congested arc segments to the total number of used arc segments is calculated; the higher the proportion value, the more intense the road right competition among multiple tasks.

6. The dynamic-disturbance-oriented intelligent scheduling method for emergency material multimodal transport according to claim 5, characterized in that, In step C, the following steps are also included: C2. Call adaptive large-domain search algorithm to solve dynamic scheduling model, C21. Encoding and decoding of shortest path based on spatiotemporal network, (1) Indirect encoding: priority-based integer encoding is used, and a task sequence represents the service priority order of all tasks to be scheduled; (2) Spatiotemporal decoding: design a greedy construction algorithm based on spatiotemporal state update, State synchronization: initialize the spatiotemporal network and synchronize the resource occupation of each node at the current time; Sequential planning: tasks are extracted from the space-time network in the order of the task sequence ;​ Optimal mapping: for tasks Under the consideration of the current network remaining resource constraints, a dynamic Dijkstra algorithm with time window is used to search a generalized cost minimum path from the supply point to the demand point in the space-time network. Resource deduction: after the path is determined, the residual capacity of the arc segments along the path is updated in the temporary space-time network according to the dynamic resource capacity coupling constraint for satisfying the constraint limit of the subsequent low-priority task; C22. Initial solution generation, (1) The first time domain: adopt according to the urgency weight of each task Generate an initial priority sequence from large to small; (2) Subsequent time domain: design a warm start strategy, Inheritance and pruning of solution: reading the last decision cycle optimal scheduling scheme, rejecting task instructions that have already been executed during the time period ​ Space-time state projection: project the vehicle positions and resource states of the remaining unexecuted tasks onto the space-time network nodes of the current time step as the skeleton of the current solution ; Incremental insertion repair: for newly emerged tasks in the current time domain, use a greedy insertion strategy to insert them into the free time slots of the inherited solution as the initial solution of the current time domain; C23. Destruction / repair operator library setting based on disturbance feature perception, (1) Destruction operator, Random removal: randomly selected Task removal; Delay / worst time removal: remove the task with the latest completion time or the highest delay penalty; Bottleneck section / node-aware removal: real-time read the transfer arc congestion coefficient and travel arc impedance value in the spatiotemporal network, and remove the transportation task containing high impedance road sections or congested nodes; (2) Repair operator, Greedy insertion: insert the removed task into the position that minimizes the increment of the objective function; Regret insertion: preferentially insert tasks with the largest regret value, which is defined as the difference between the objective function values when the task is inserted at the current optimal position and the suboptimal position; Modal disturbance insertion: when reinserting a task, the repair operator forcibly disables the primary transportation mode used in the previous iteration, taking advantage of the complementary advantages of multimodal transportation to quickly find alternative modes when the physical path is interrupted; C24. Intelligent operator selection strategy based on environmental feature perception and cross-time domain learning, Design an environment-aware reinforcement learning mechanism and maintain a global operator utility database in the processor's non-volatile memory; (1) Multi-dimensional state space definition: Construct a composite state vector containing both the search internal state and the external environment features ; Search internal state : describes the current convergence behavior of the adaptive large domain search algorithm, discretized into an exploration phase, a development phase, a stagnation phase; External environmental characteristics : Discretization mapping of the extracted three-dimensional feature indicators: Task size Low load, high load; Supply freedom Strong constraint, weak constraint; Resource conflict index No conflict, high congestion; (2) Action definition: Action is selecting a pair of operator combinations ; (3) The reward function is defined as follows: ； wherein ; (4) Update the Q value in the Q learning mechanism, (5) Cross-time domain inheritance and environmental association of knowledge, In the global operator utility database, Q values are determined not only by operators but also by environmental feature indices; Cross-time domain migration mechanism: when time is rolled from to , the high Q value strategy accumulated in history is directly read by the rolling time domain system; C25. Algorithm main process, (1) Initialization: Based on the initial solution generated in step C22, load ; (2) Main loop: State recognition: determine whether the current state is exploration, development, or stagnation; Operator selection: select operator pair according to Q-value table ; Neighborhood search: Perform destruction and repair operations to get new chromosomes ; Decode evaluation: Call spatiotemporal decoder to compute objective function value; Acceptance criterion: use simulated annealing criteria to decide whether to accept the new solution; Knowledge update: calculate the reward based on the results and update the Q value in the corresponding state; (3) Termination: meet the time or iteration limit, output the optimal scheduling scheme, and save the current Q value table, completing the algorithm main process of the scheduling plan.

7. The dynamic-disturbance-oriented intelligent scheduling method for emergency material multimodal transport according to claim 6, characterized in that, In step C, the following steps are also included: C3. Scheme execution and disturbance monitoring, The scheduling plan generated in step C25 is executed and dynamic perturbation information of external environment characteristics is continuously received.

8. The dynamic-disturbance-oriented intelligent scheduling method for emergency material multimodal transport according to claim 7, characterized in that, The following steps are also included in step C3: If new emergency demand or mutation of road network state occurs, the time is advanced to the starting time of the next decision cycle when the current decision cycle ends ; the updated rolling horizon system state and Q-value table are returned to step C12, and the next rolling horizon decision cycle is entered to realize continuous adaptation to dynamic environment and optimal scheduling.

Images