A hierarchical multi-agent collaborative scheduling system and method for urban multi-modal freight transfer

Abstract

This invention relates to the field of smart logistics technology and discloses a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections. The system includes: a multimodal coordination agent configured to switch from a normal mode to a collaborative mode in response to a collaborative activation signal triggered by an observation vector, generating a scheduling intention; an urban freight agent configured to receive the scheduling intention and generate an execution quota based on the scheduling intention and a two-way adjustment mechanism; and a logistics agent configured to receive the execution quota and generate control instructions for sub-agents based on the execution quota. The solution of this invention utilizes the multimodal coordination agent, the urban freight agent, and the logistics agent to establish a collaborative scheduling system for the entire process. By using the observation vector as the trigger condition for the collaborative activation signal, it establishes monitoring of changes in passenger and freight demand, thereby utilizing the established two-way adjustment mechanism to form a response mechanism to passenger and freight demand.

Description

Technical Field

[0001] This invention relates to the field of smart logistics technology, specifically to a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections. Background Technology

[0002] In recent years, with the rapid development of e-commerce and the acceleration of urbanization, smart logistics and multimodal transportation have become a global research and industrial application hotspot. To solve the problem of high costs in last-mile delivery, the industry is actively exploring the collaborative potential of heterogeneous transportation capacity such as rail transit, ground trucks, and drones.

[0003] In related technologies, existing scheduling systems are mostly single-modal optimizations or weakly coupled integrations, focusing primarily on short-term rapid delivery and often separating rail freight planning, cargo flow in transfer areas, and ground-to-air delivery routes. This type of scheduling approach is highly susceptible to scheduling chaos when dealing with surges in passenger and freight demand, severely impacting the system's operational resilience and the stability of subsequent logistics services. Therefore, existing scheduling systems lack a comprehensive collaborative scheduling mechanism, rely excessively on a single mode of transportation, and neglect the impact of passenger and freight demand on collaborative mechanisms. Summary of the Invention

[0004] In view of this, the present invention provides a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections, in order to solve the technical problems of full-process collaborative scheduling and the lack of response mechanisms to passenger and freight demand.

[0005] In a first aspect, the present invention provides a hierarchical multi-agent collaborative scheduling system for urban multimodal freight connections. The system includes: a multimodal coordination agent, an urban freight agent, and a logistics agent. The multimodal coordination agent is configured to switch from a normal mode to a collaborative mode in response to a collaborative activation signal triggered by an observation vector, and to generate a scheduling intention. The urban freight agent is configured to receive the scheduling intention and generate an execution quota based on the scheduling intention and a two-way adjustment mechanism. The logistics agent is configured to receive the execution quota and generate control instructions for the sub-agents based on the execution quota.

[0006] In conjunction with the first aspect, in one possible implementation of the first aspect, the multi-mode coordinated intelligent agent includes: a spatiotemporal topology perception unit, a passenger-cargo coupling evaluation unit, a resource allocation optimization unit, and a macro-strategy encapsulation unit. The spatiotemporal topology perception unit is configured to determine the dynamic spatiotemporal network topology based on logistics network hubs and corresponding operating parameters. The passenger-cargo coupling evaluation unit is configured to determine the connection area inventory state vector and the rail network capacity state tensor based on rail capacity and cargo demand, and to determine whether a collaborative activation signal is triggered based on the connection area inventory state vector and the rail network capacity state tensor. The resource allocation optimization unit is configured to determine a capacity balancing plan based on the dynamic spatiotemporal network topology, the connection area inventory state vector, and the rail network capacity state tensor. The macro-strategy encapsulation unit is configured to encode the capacity balancing plan into an executable strategy tensor as the scheduling intention.

[0007] In conjunction with the first aspect, in one possible implementation of the first aspect, the urban freight intelligent agent includes: an urban rail freight intelligent agent and a ground low-altitude logistics intelligent agent, wherein the urban rail freight intelligent agent is configured to determine and send cargo loading and unloading information to the ground low-altitude logistics intelligent agent, and receive scheduling intentions and client demand information sent by the ground low-altitude logistics intelligent agent; the ground low-altitude logistics intelligent agent is configured to determine and send client demand information to the urban rail freight intelligent agent, and receive scheduling intentions and cargo loading and unloading information sent by the urban rail freight intelligent agent.

[0008] In conjunction with the first aspect, in one possible implementation of the first aspect, the ground-based low-altitude logistics intelligent agent is further configured to generate loading and unloading execution quotas based on scheduling intentions and client demand; or, the ground-based low-altitude logistics intelligent agent is further configured to optimize loading and unloading execution quotas based on scheduling intentions and client demand in response to the triggering of the two-way adjustment mechanism.

[0009] In conjunction with the first aspect, in one possible implementation of the first aspect, the ground-based low-altitude logistics intelligent agent is further configured to generate delivery execution quotas based on scheduling intentions and cargo loading and unloading status; or, the ground-based low-altitude logistics intelligent agent is further configured to optimize delivery execution quotas based on scheduling intentions and cargo loading and unloading status in response to the triggering of the two-way adjustment mechanism.

[0010] In conjunction with the first aspect, in one possible implementation of the first aspect, the logistics intelligent agent includes: a connecting station sub-intelligent agent and a ground-low-altitude sub-intelligent agent, wherein the connecting station sub-intelligent agent includes a corresponding first transportation equipment, and the ground-low-altitude sub-intelligent agent includes a second transportation equipment composed of ground transportation equipment and low-altitude transportation equipment. The first transportation equipment is used for loading, unloading and transferring goods in the urban rail freight intelligent agent; the second transportation equipment is used to complete logistics distribution.

[0011] In conjunction with the first aspect, in one possible implementation of the first aspect, the connecting station sub-agent is configured to generate control instructions for the first transport equipment based on loading and unloading execution quotas.

[0012] In conjunction with the first aspect, in one possible implementation of the first aspect, the ground-low-altitude sub-agent is configured to generate control instructions for the second transport equipment based on delivery execution quotas.

[0013] Secondly, the present invention provides a hierarchical multi-agent cooperative scheduling method for urban multimodal freight connections. The method includes: switching from a normal mode to a cooperative mode in response to a cooperative activation signal triggered by an observation vector, and generating a scheduling intention; generating an execution quota based on the scheduling intention and a two-way adjustment mechanism; and generating control instructions for sub-agents based on the execution quota.

[0014] In conjunction with the second aspect, in one possible implementation of the second aspect, the bidirectional adjustment mechanism includes, in response to the bidirectional adjustment mechanism being triggered, performing the following steps: generating a counter-pressure signal based on a digital twin model, and optimizing the loading and unloading execution quota based on the counter-pressure signal; generating an order reminder signal based on the digital twin model, and optimizing the delivery execution quota based on the order reminder signal.

[0015] The technical solution of this invention has the following advantages: This invention provides a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections. The system includes a multimodal coordination agent, an urban freight agent, and a logistics agent. The multimodal coordination agent is configured to switch from a normal mode to a collaborative mode in response to a collaborative activation signal triggered by an observation vector, generating a scheduling intention. The urban freight agent is configured to receive the scheduling intention and generate an execution quota based on the intention and a two-way adjustment mechanism. The logistics agent is configured to receive the execution quota and generate control instructions for sub-agents based on the quota. In the overall system, the multimodal coordination agent performs global coordination, the urban freight agent performs collaboration, and the logistics agent performs distributed execution, thus establishing a collaborative scheduling system for the entire process. From a global perspective, the system uses the observation vector as the trigger condition for the collaborative activation signal to monitor changes in passenger and freight demand, thereby enabling the switching between normal and collaborative modes, making the overall system more adaptable to changes in passenger and freight demand. Furthermore, from a local perspective, the system utilizes the established two-way adjustment mechanism to adjust for changes in passenger and freight demand leading to delivery changes, forming a response mechanism to passenger and freight demand. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the structure of a hierarchical multi-agent collaborative scheduling system for urban multimodal freight connection provided by an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a multi-mode coordinated intelligent agent provided according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating another hierarchical multi-agent collaborative scheduling method for urban multimodal freight connections provided by an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] According to an embodiment of the present invention, a hierarchical multi-agent cooperative scheduling method for urban multimodal freight connections is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0020] This embodiment provides a hierarchical multi-agent collaborative scheduling system for urban multimodal freight connections, such as... Figure 1 As shown, the system includes: a multi-modal coordination agent, an urban freight agent, and a logistics agent, among which, A multi-mode coordinating agent is configured to switch from a normal mode to a cooperative mode in response to a cooperative activation signal triggered by an observation vector, and to generate a scheduling intention. An urban freight agent is configured to receive the scheduling intention and, based on the scheduling intention and a two-way adjustment mechanism, generate an execution quota. A logistics agent is configured to receive the execution quota and, based on the execution quota, generate control instructions for the sub-agents. The triggering conditions for the cooperative activation signal and the recovery conditions for the normal mode are detailed in subsequent embodiments.

[0021] Specifically, the hierarchical multi-agent collaborative scheduling system for urban multimodal freight connections includes a normal mode and a collaborative mode. The normal mode is the regular operation mode, in which each agent and sub-agent executes corresponding actions according to preset strategies. The collaborative mode is the full-process collaborative scheduling mode, in which each agent and sub-agent executes the corresponding issued strategies.

[0022] Specifically, the multi-modal coordination agent is connected to the urban freight agent and configured to collect multi-source capacity data and external environmental data from the urban rail network, ground transportation network, and low-altitude airspace. It identifies and calculates rail passenger flow, carriage load, road congestion, low-altitude airspace status, weather or traffic restrictions, etc., as traffic data tensors, determines the target connection area set and target route set, and encapsulates these into scheduling intentions. Among these, multi-source capacity data, external environmental data, and traffic data tensors can all serve as concrete representations of observation vectors. If the global state tensor of the system at time t is defined as... S t Then it can be expressed by the following formula. S t :

[0023] in, C t This represents the capacity state tensor of the rail network. I t The tensor representing the inventory status of the docking area. G t This represents the dynamic spatiotemporal network topology.

[0024] Specifically, S t The data collected serves as the input to the multimodal coordinating agent, where... C t It includes real-time data such as carriage load rate and passenger flow. I t It includes the real-time cargo backlog at each transfer station. G t It includes real-time traffic flow weights for ground transportation networks and low-altitude airspace.

[0025] Specifically, the observation vector is determined based on the global state tensor and is expressed by the following formula:

[0026] in, Let f(·) represent the observation vector, and let f(·) represent the fusion mapping function for multi-source heterogeneous data. Based on this, and according to the quantitatively assessed multi-source logistics impact situation, the system dynamically formulates a coordinated response and adaptively adjusts the scheduling intent of the lower-level intelligent agents, thereby guiding the entire network to comprehensively reduce inventory pressure, transportation delays, and the risk of physical constraint violations. The scheduling intent refers to macro-control data, specifically manifested as task constraints and configuration parameters, such as capacity allocation matrices, priorities, time window parameters, and mode switching identifiers.

[0027] Specifically, the urban freight intelligent agent is configured to receive scheduling intentions and, based on these intentions, generate execution quotas for lower-level intelligent agents to execute. Alternatively, the urban freight intelligent agent is configured to receive scheduling intentions and, after the bidirectional adjustment mechanism is triggered, optimize the execution quotas for lower-level intelligent agents to execute. The execution quotas include loading / unloading execution quotas and delivery execution quotas. Loading / unloading execution quotas are sent to the connecting station sub-intelligent agents to constrain the corresponding train-level loading / unloading tasks of that sub-intelligent agent group. Delivery execution quotas are sent to the ground-low-altitude sub-intelligent agents to constrain the corresponding delivery tasks of that sub-intelligent agent group. The triggering conditions for the bidirectional adjustment mechanism are detailed in subsequent embodiments.

[0028] Specifically, the logistics intelligent agent is configured to receive and execute quotas, and based on the execution quotas, generate control instructions for sub-intelligent agents. These control instructions for sub-intelligent agents include control instructions for a first transport device and control instructions for a second transport device.

[0029] This invention provides a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections. The system includes a multimodal coordination agent, an urban freight agent, and a logistics agent. The multimodal coordination agent is configured to switch from a normal mode to a collaborative mode in response to a collaborative activation signal triggered by an observation vector, generating a scheduling intention. The urban freight agent is configured to receive the scheduling intention and generate an execution quota based on the intention and a two-way adjustment mechanism. The logistics agent is configured to receive the execution quota and generate control instructions for sub-agents based on the quota. In the overall system, the multimodal coordination agent performs global coordination, the urban freight agent performs collaboration, and the logistics agent performs distributed execution, thus establishing a collaborative scheduling system for the entire process. From a global perspective, the system uses the observation vector as the trigger condition for the collaborative activation signal to monitor changes in passenger and freight demand, thereby enabling the switching between normal and collaborative modes, making the overall system more adaptable to changes in passenger and freight demand. Furthermore, from a local perspective, the system utilizes the established two-way adjustment mechanism to adjust for changes in passenger and freight demand leading to delivery changes, forming a response mechanism to passenger and freight demand.

[0030] In one alternative implementation, a multi-modal coordinating agent is used to assess the passenger flow margin of urban rail transit and, combined with the ground road network and low-altitude airspace conditions, generate a dynamic spatiotemporal network topology, thereby forming a scheduling intention. For example... Figure 2 As shown, the multi-modal coordinated intelligent agent includes: a spatiotemporal topology perception unit, an object-object coupling evaluation unit, a resource allocation optimization unit, and a macroscopic policy encapsulation unit, wherein... The spatiotemporal topology sensing unit is configured to determine the dynamic spatiotemporal network topology based on logistics network hubs and corresponding operating parameters; the passenger-cargo coupling evaluation unit is configured to determine the connection area inventory state vector and the rail network capacity state tensor based on rail capacity and cargo demand, and determine whether to trigger a collaborative activation signal based on the connection area inventory state vector and the rail network capacity state tensor; the resource allocation optimization unit is configured to determine the capacity balancing plan based on the dynamic spatiotemporal network topology, the connection area inventory state vector, and the rail network capacity state tensor; and the macro-strategy encapsulation unit is configured to encode the capacity balancing plan into an executable strategy tensor as the scheduling intent.

[0031] Specifically, the spatiotemporal topology sensing unit is configured to construct a directed topological graph describing the overall situation of urban logistics, denoted as... G =( V , L ), where the node set V Defined as key physical hubs in the urban logistics network, such as rail connection stations, ground transfer centers, and last-mile distribution points. Each node ∈ It correlates in real time with its inventory backlog status, processing rate, and physical throughput limits. (Directed edge set) L Defined as a dynamic transfer link between nodes, including rail freight corridors, ground road delivery routes, and low-altitude drone flight paths. Each directed edge l ij ∈ L It incorporates time-dependent operating parameters, such as real-time traffic flow weights, link capacity bandwidth, and transportation mode attributes. The spatiotemporal topology sensing unit, through streaming processing of multi-source heterogeneous data, achieves a digital spatial mapping of logistics resources, providing causal-directional topological constraints for subsequent macro-level scheduling, ensuring that scheduling instructions can propagate along effective logistics paths.

[0032] Specifically, the passenger-freight coupling assessment unit is configured to perform a dynamic matching assessment of rail capacity margin and freight demand. This is achieved by collecting real-time passenger flow load data within the railcars. P load ( t (in combination with the train's maximum physical load capacity) C maxThe capacity margin tensor characterizing the available freight space on the track side was calculated. C margin ( t The capacity margin tensor is represented by the following formula:

[0033] Furthermore, based on the cargo scanner in the docking buffer zone, the inventory status matrix of the docking area is calculated in real time. I buf ( t This vector contains the current inventory level, inflow rate, and expected outflow rate. By scalarizing the inventory status vector, the inventory pressure in the transfer area is monitored in real time, and the elastic potential modulus of the inventory in the transfer area is calculated. E buf ( t The modulus of elastic potential energy is expressed by the following formula:

[0034] in, I buf ( t ) express t A matrix of inventory status in the connection area at any given time.

[0035] Specifically, I buf ( t The row dimension corresponds to different product types / priorities, and the column dimension corresponds to different storage units. The Frobenius norm is used to aggregate the two-dimensional inventory status matrix into a single scalar indicator of stress intensity. E buf ( t The value of the norm directly reflects the degree of risk of cargo accumulation currently faced by the receiving area. The larger the norm, the higher the square root of the sum of squares of the matrix elements, meaning the greater the overall deviation of the inventory level of each storage unit from the baseline, and the greater the inventory pressure. Therefore, when E buf ( t The data continues to rise and exceeds the preset physical carrying capacity thresholds for the integrated transfer area, including physical storage capacity, cargo turnover rate, and carrying capacity redundancy rate. θ cap At this point, it is believed that the system faces the risk of node liquidation. This triggers a collaborative activation signal, causing the system to switch from normal mode to collaborative mode, denoted as . E buf ( t > θ cap .

[0036] The system faces the risk of node-level inventory overflow, which can be represented as the inventory deviation in the transfer area. △It The inventory deviation in the transfer area is expressed by the following formula:

[0037] in, I ref Indicates the historical baseline allowed for the connection area, The symbol represents norm operations, used to scalarize inventory pressure, determine the level of deviation, and output the adjustment tensor.

[0038] Specifically, the resource allocation optimization unit performs a global-dimensional cross-modal capacity dynamic balance calculation on top of the constructed dynamic spatiotemporal network topology. This is achieved by receiving the capacity margin tensor and the docking area inventory pressure index from the passenger-cargo coupling evaluation unit. C margin ( t )and E buf ( t This paper proposes a mathematical programming model to optimize the allocation of freight traffic across three modes: urban rail, ground roads, and low-altitude airspace. The model aims to achieve optimal overall logistics timeliness, reduced inventory pressure, decreased transportation delays, and a lower proportion of non-executable quotas, while also considering operating costs. During the calculation, the resource allocation optimization unit must fully consider the flow conservation constraints between physical nodes, the rigid timetable constraints of rail transit, and the physical load limits of heterogeneous ground and air transport capacity. By solving this planning model, the unit can dynamically derive the optimal flow allocation path between hub nodes to alleviate freight congestion bottlenecks at connecting stations.

[0039] Specifically, the macro-strategy encapsulation unit uses parameterized mapping technology to transform and encode macro-resource allocation strategies into strategy tensors that can be read and executed by lower-level intelligent agents. The macro-strategy encapsulation unit deconstructs the macro-level capacity balancing plan into track-side capacity adjustment instructions, ground-air-side capacity activation thresholds, and counter-pressure signals and expedited order signals corresponding to the bidirectional adjustment mechanism. The counter-pressure signals constrain train-level loading and unloading quotas, reducing the inflow rate of goods into the transfer area, while the expedited order signals adjust the activation quantity of ground-air-side capacity and the compression ratio of delivery time windows, increasing the outflow rate of goods from the transfer area.

[0040] Furthermore, during the encapsulation process, the macro-strategy encapsulation unit adopts a standardized data structure to ensure that global decision information is not distorted during cross-level transmission and can be logically aligned with the optimization objective functions of each agent.

[0041] In one alternative implementation, the multi-mode coordinating agent is also configured to switch from a collaborative mode to a normal mode in response to a recovery signal triggered by inventory deviation in the docking area.

[0042] Specifically, when the inventory deviation in the transfer area decreases significantly and meets the preset range for a preset time, it is determined that the inventory pressure in the transfer area has met the recovery conditions, i.e., the system switches from collaborative mode to normal mode. The recovery conditions are expressed by the following formula: Δ I t < and maintain continuously H Each time step, among which... Indicates the preset range. H This indicates a preset time. In this embodiment, there are no specific limitations on the preset range and preset time; they can be selected according to the actual working conditions.

[0043] In one alternative implementation, the urban freight intelligent agent includes: an urban rail freight intelligent agent and a ground-based low-altitude logistics intelligent agent, wherein... The urban rail freight intelligent agent is configured to determine and send cargo loading and unloading information to the ground low-altitude logistics intelligent agent, and receive scheduling intentions and client demand information sent by the ground low-altitude logistics intelligent agent; the ground low-altitude logistics intelligent agent is configured to determine and send client demand information to the urban rail freight intelligent agent, and receive scheduling intentions and cargo loading and unloading information sent by the urban rail freight intelligent agent.

[0044] Specifically, the urban rail freight intelligent agent comprises a distributed group of train-level intelligent agents, such as train intelligent agent 1, train intelligent agent 2, ..., train intelligent agent N, corresponding to train-level loading and unloading tasks. The urban rail freight intelligent agent improves the overall network transfer efficiency by optimizing the allocation of freight resources for each train in real time.

[0045] Specifically, the ground-based low-altitude logistics intelligent agent comprises multiple route intelligent agents, such as: route intelligent agent 1, route intelligent agent 2, ..., route intelligent agent N. Each route intelligent agent has corresponding connected sub-intelligent agents within the logistics intelligent agent, corresponding to ground-to-air transport capacity. The sub-intelligent agents include: connecting station sub-intelligent agents and ground-to-low-altitude sub-intelligent agents. The ground-based low-altitude logistics intelligent agent is responsible for configuring the ground and low-altitude transport capacity topology.

[0046] In one alternative implementation, the ground-based low-altitude logistics agent is further configured to generate loading and unloading execution quotas based on scheduling intentions and client demand; or, the ground-based low-altitude logistics agent is further configured to optimize loading and unloading execution quotas based on scheduling intentions and client demand in response to the triggering of a two-way adjustment mechanism.

[0047] Specifically, the urban rail freight intelligent agent is based on a distributed train-level intelligent agent swarm, for each train i Determine the train i Local observation state vector:

[0048] in, Represents the local observation state vector. Indicates train i Current carriage load factor This indicates the current progress of the unloading operation. Indicates train i Real-time inventory pressure at the connecting stations.

[0049] The local observation state vector is used to represent the real-time state of the line-level train agent in its physical space and task dimension. By dynamically monitoring the carriage load rate, the system can accurately perceive the remaining carrying capacity of the transportation unit, thereby determining whether it has the physical redundancy to continue accepting goods. The introduction of unloading operation progress transforms discrete loading and unloading tasks into continuous time functions, providing a quantitative basis for the agent to evaluate task completion and predict future surplus moments. By introducing real-time inventory pressure at connecting stations, the train agent is endowed with environmental perception capabilities, establishing logical coupling between transportation units and storage nodes.

[0050] Furthermore, to overcome the curse of dimensionality in the action space caused by the large number of trains and the complex coupling of right-of-way and capacity in urban rail transit networks, mean-field theory is introduced to construct a neighborhood state aggregation vector:

[0051] in, Represents the neighborhood state aggregation vector. Indicates in t At that moment, the j Local state vectors of neighborhood agents N i Indicates with train i A set of neighboring trains that have topological connections.

[0052] Specifically, the neighborhood state aggregation vector represents the average freight loading and unloading urgency of neighboring trains coupled with the current train. By aggregating the states of heterogeneous units within the topological neighborhood of the target train using a mean-based approach, the point-to-point game is simplified into a binary interaction between an individual and the neighborhood mean field. The distributed train-level intelligent agent swarm utilizes the neighborhood state aggregation tensor to perceive the surrounding operational situation, ensuring that each decision-making unit can achieve efficient real-time response that balances local optima and global stability while satisfying physical constraints.

[0053] Specifically, the urban rail freight intelligent agent is configured to generate loading and unloading execution quotas based on scheduling intentions and client demand. During this process, a policy network is used to nonlinearly fuse local observation state vectors with neighborhood state aggregation vectors to generate line-level freight transport quotas, which include loading and unloading execution quotas.

[0054] Specifically, the urban rail freight intelligent agent is configured to respond to the triggering of the two-way adjustment mechanism. Based on the scheduling intention and the client demand, during the process of optimizing the loading and unloading execution quota, the agent uses a policy network to nonlinearly fuse the local observation state vector with the neighborhood state aggregation vector. Using the two-way adjustment mechanism, the agent sends a back pressure signal to the distributed train-level intelligent agent group, thereby constraining the unloading quota or unloading time window and forming an optimized delivery execution quota.

[0055] Specifically, the policy network, through a neural network architecture, nonlinearly fuses the local observed state vector with the neighborhood state aggregation vector. After determining whether to trigger the bidirectional adjustment mechanism, it generates a line-level freight transport quota, which includes loading / unloading execution quota and delivery execution quota. The line-level freight transport quota is represented by the following formula:

[0056] in, This indicates the line-level freight transport quota. π θ This represents the trained policy network function. load i,t and unload i, t These represent the cargo loading and unloading status for the corresponding train and the client's demand, respectively, i.e., the loading quota and unloading quota for the corresponding train.

[0057] Specifically, the loading and unloading quota is represented by the following formula:

[0058] in, Indicates the first m The first transport equipment in t Loading and unloading quotas at specific times and These represent the coordinates for picking up goods in the station's transfer area and the coordinates for loading and unloading goods on the connecting ground transportation equipment, respectively. This indicates the time constraint required to complete the corresponding operational task.

[0059] In one alternative implementation, when the local inventory potential modulus exceeds a macroeconomic threshold, i.e. At that time, the intelligent agent triggers bidirectional adjustment actions according to preset control rules: on the one hand, it sends a back pressure signal to the track side to suppress the unloading rate; on the other hand, it sends an order reminder signal to the ground low-altitude logistics intelligent agent to improve the connection and transportation capacity, thereby restoring inventory balance within the physical constraints, that is, optimizing loading and unloading quotas and optimizing delivery execution quotas.

[0060] In one alternative implementation, the ground-based low-altitude logistics agent is further configured to generate delivery execution quotas based on scheduling intentions and cargo loading and unloading status; or, the ground-based low-altitude logistics agent is further configured to optimize delivery execution quotas based on scheduling intentions and cargo loading and unloading status in response to the triggering of a two-way adjustment mechanism.

[0061] Specifically, the ground-based low-altitude logistics intelligent agent is configured to generate delivery execution quotas based on scheduling intentions and cargo loading and unloading status. During this process, a policy network is used to nonlinearly fuse local observation state vectors with neighborhood state aggregation vectors to generate line-level cargo transportation quotas, which include delivery execution quotas.

[0062] Specifically, based on the dynamic spatiotemporal network topology, and using spatial and temporal windows as benchmarks, the delivery task set is analyzed. D Task clustering performed .in, D k For the first k Each line agent is responsible for a cluster of tasks.

[0063] Specifically, the spatial topological relationship features of the road network are extracted through spatiotemporal coupling, and the road state vector is output. = ( ).in, ( This is a graph neural network encoder. Subsequently, by combining road state vectors and task cluster features, the optimal spatiotemporal waypoint sequence for ground transportation equipment and low-altitude transportation equipment is gradually deduced and generated. = ( , By combining ground-based low-altitude transportation capacity, a final delivery execution quota is generated. The delivery execution quota is expressed by the following formula:

[0064] in, This indicates that the delivery quota is being executed. route k This represents a specific three-dimensional spatial and temporal waypoint sequence. time kThis represents the time window requirements corresponding to each waypoint. The highly nonlinear urban road network congestion, weather interference, no-fly altitudes, restricted flight zones, and low-altitude right-of-way constraints are mapped in a high dimension into a state vector containing spatiotemporal dependencies.

[0065] Specifically, the ground-based low-altitude logistics intelligent agent is configured to respond to the triggering of the two-way adjustment mechanism. Based on scheduling intentions and cargo loading / unloading status, during the process of optimizing delivery execution quotas, a policy network is used to nonlinearly fuse the local observed state vector with the neighborhood state aggregation vector. The two-way adjustment mechanism then sends an order reminder signal to the ground-based low-altitude logistics intelligent agent, thereby dynamically compressing the delivery time window and replanning routes to improve connection and distribution capacity, thus forming an optimized delivery execution quota. This process is the same as in the above embodiment and will not be repeated here.

[0066] In one optional embodiment, the logistics intelligent agent includes: a connecting station sub-intelligent agent and a ground-low-altitude sub-intelligent agent, wherein the connecting station sub-intelligent agent includes a corresponding first transportation device, and the ground-low-altitude sub-intelligent agent includes a second transportation device composed of ground transportation device and low-altitude transportation device. The first transportation device is used for loading, unloading and transferring goods in the urban rail freight intelligent agent; the second transportation device is used to complete logistics distribution.

[0067] Specifically, the first type of transportation equipment includes: shuttle transport vehicles, trucks, and other transportation equipment; ground transportation equipment includes trucks, trains, and other transportation equipment; and low-altitude transportation equipment includes: drones and other transportation equipment.

[0068] In one alternative implementation, the docking station sub-agent is configured to generate control commands for the first transport equipment based on loading and unloading execution quotas.

[0069] Specifically, the sub-agent at the docking station is used to construct a micro-execution situation vector by combining the loading and unloading execution quota with the pose data collected in real time by the sensors of the first transportation equipment, the motor feedback status, and information on surrounding obstacles. After verifying the physical constraints through built-in kinematic deduction, preliminary motor control commands are generated, namely the control commands for the first transport device:

[0070] in, For the target linear velocity, For the target angular velocity, To estimate motor torque and meet AGV kinematic and collision avoidance constraints, such as speed, acceleration, angular velocity, turning radius, minimum safety distance, path occupancy, and occupancy status of pick-up and drop-off points, i.e. .

[0071] In one alternative implementation, the ground-low-altitude sub-agent is configured to generate control commands for the second transport equipment based on delivery execution quotas.

[0072] Specifically, ground-low-altitude sub-agents are used to combine optimal spatiotemporal waypoint sequences. With delivery execution quota This is converted into coordinated control commands for the second transportation equipment, i.e., control commands for the second transportation equipment:

[0073] in, and These represent the acceleration and steering angle commands for ground transportation equipment, respectively. and These are the lift control signal for low-altitude transport equipment and the differential speed control signal for the quadcopter. Specifically, if the physical constraint verification fails, the system returns to the upper layer to apply a negative reward and regenerates the action.

[0074] This embodiment provides a hierarchical multi-agent cooperative scheduling method for urban multimodal freight connections, such as... Figure 3 As shown, the method includes the following steps: S101. In response to the observation vector triggering the cooperative activation signal, the system switches from the normal mode to the cooperative mode and generates a scheduling intention. For details, please refer to the relevant descriptions in the above embodiments; they will not be repeated here.

[0075] S102. Based on the scheduling intent and the two-way adjustment mechanism, an execution quota is generated. For details, please refer to the relevant descriptions in the above embodiments; they will not be repeated here.

[0076] S103. Based on the execution quota, generate control instructions for the sub-agent. For details, please refer to the relevant descriptions in the above embodiments; they will not be repeated here.

[0077] This invention provides a hierarchical multi-agent collaborative scheduling system and method for urban multimodal freight connections. The method includes: switching from a normal mode to a collaborative mode in response to a collaborative activation signal triggered by an observation vector, and generating a scheduling intention; generating an execution quota based on the scheduling intention and a two-way adjustment mechanism; and generating control instructions for sub-agents based on the execution quota. In this process, a multimodal coordinating agent performs global coordination, an urban freight agent performs collaboration, and a logistics agent performs distributed execution, thereby establishing a collaborative scheduling system for the entire process. From a holistic perspective, by using the observation vector as the trigger condition for the collaborative activation signal, monitoring changes in passenger and freight demand is established, thereby completing the switch between the normal mode and the collaborative mode, making the overall system more adaptable to changes in passenger and freight demand. Furthermore, from a local perspective, the established two-way adjustment mechanism is used to adjust for changes in delivery caused by changes in passenger and freight demand, forming a response mechanism to passenger and freight demand.

[0078] In one alternative implementation, the bidirectional adjustment mechanism includes, in response to the bidirectional adjustment mechanism being triggered, performing the following steps: generating a counter-pressure signal based on a digital twin model, and optimizing loading and unloading execution quotas based on the counter-pressure signal; generating an order reminder signal based on the digital twin model, and optimizing delivery execution quotas based on the order reminder signal.

[0079] Specifically, in a digital twin environment, when a back pressure signal is triggered, a negative reward is applied to the track-side strategy to reduce the unloading rate, while when an order reminder signal is triggered, a time penalty is applied to the ground-to-air strategy to accelerate delivery. This, through cross-agent collaborative rewards, shapes the dynamic equilibrium behavior of heterogeneous transport capacity in the junction area. For details, please refer to the above embodiments, which will not be repeated here.

[0080] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A hierarchical multi-agent collaborative scheduling system for urban multimodal freight connections, characterized in that, The system includes: a multi-modal coordination agent, an urban freight agent, and a logistics agent, wherein... The multi-mode coordinating agent is configured to switch from normal mode to cooperative mode in response to a cooperative activation signal triggered by an observation vector, and to generate a scheduling intention. The urban freight intelligent agent is configured to receive the scheduling intention and generate an execution quota based on the scheduling intention and a two-way adjustment mechanism. The logistics intelligent agent is configured to receive the execution quota and generate control instructions for the sub-intelligent agents based on the execution quota.

2. The system according to claim 1, characterized in that, The multi-modal coordinated intelligent agent includes: a spatiotemporal topology perception unit, an object-object coupling evaluation unit, a resource allocation optimization unit, and a macro-policy encapsulation unit, wherein... The spatiotemporal topology sensing unit is configured to determine the dynamic spatiotemporal network topology based on logistics network hubs and corresponding operating parameters; The passenger-cargo coupling evaluation unit is configured to determine the inventory state vector of the connection area and the rail network capacity state tensor based on rail transport capacity and cargo demand, and to determine whether to trigger a collaborative activation signal based on the inventory state vector of the connection area and the rail network capacity state tensor. The resource allocation optimization unit is configured to determine a capacity balance plan based on the dynamic spatiotemporal network topology, the connection area inventory state vector, and the rail network capacity state tensor. The macro-strategy encapsulation unit is configured to encode the capacity balancing plan into an executable strategy tensor as the scheduling intent.

3. The system according to claim 1, characterized in that, The urban freight intelligent agent includes: an urban rail freight intelligent agent and a ground low-altitude logistics intelligent agent, wherein, The urban rail freight intelligent agent is configured to determine and send cargo loading and unloading information to the ground low-altitude logistics intelligent agent, and receive the scheduling intention and the client demand sent by the ground low-altitude logistics intelligent agent. The ground-based low-altitude logistics intelligent agent is configured to determine and send the client's demand to the urban rail freight intelligent agent, and to receive the scheduling intention and the cargo loading and unloading information sent by the urban rail freight intelligent agent.

4. The system according to claim 3, characterized in that, The ground-based low-altitude logistics intelligent agent is also configured to generate loading and unloading execution quotas based on the scheduling intent and the client's demand; or, The ground-based low-altitude logistics intelligent agent is also configured to optimize loading and unloading execution quotas based on the scheduling intention and the client demand in response to the triggering of the bidirectional adjustment mechanism.

5. The system according to claim 4, characterized in that, The ground-based low-altitude logistics intelligent agent is also configured to generate delivery execution quotas based on the scheduling intent and the cargo loading and unloading status; or, The ground-based low-altitude logistics intelligent agent is also configured to optimize the delivery execution quota based on the scheduling intention and the cargo loading and unloading status in response to the triggering of the two-way adjustment mechanism.

6. The system according to claim 5, characterized in that, The logistics intelligent agent includes: a connecting station sub-intelligent agent and a ground-low-altitude sub-intelligent agent, wherein the connecting station sub-intelligent agent includes corresponding first transportation equipment, and the ground-low-altitude sub-intelligent agent includes second transportation equipment composed of ground transportation equipment and low-altitude transportation equipment. The first transportation equipment is used for loading, unloading and transferring goods in the urban rail freight intelligent system; The second transport equipment is used to complete logistics distribution.

7. The system according to claim 6, characterized in that, The sub-agent at the connection station is configured to generate control commands for the first transport equipment based on the loading and unloading execution quota.

8. The system according to claim 6, characterized in that, The ground-low-altitude sub-agent is configured to generate control commands for the second transport equipment based on the delivery execution quota.

9. A hierarchical multi-agent cooperative scheduling method for urban multimodal freight connections, applied to the hierarchical multi-agent cooperative scheduling system for urban multimodal freight connections as described in any one of claims 1 to 8, characterized in that, The method includes: In response to the observation vector triggering the cooperative activation signal, the system switches from normal mode to cooperative mode and generates scheduling intent; Based on the aforementioned scheduling intent and two-way adjustment mechanism, an execution quota is generated; Based on the execution quota, control instructions for the sub-agent are generated.

10. The method according to claim 9, characterized in that, The bidirectional adjustment mechanism includes, in response to the bidirectional adjustment mechanism being triggered, performing the following steps: Based on the digital twin model, a back pressure signal is generated, and based on the back pressure signal, the loading and unloading execution quota is optimized; Based on the digital twin model, a reminder signal is generated, and based on the reminder signal, the delivery execution quota is optimized.

Images