CN-121998390-A - Port transportation-oriented space-time task arrangement method and device and electronic equipment

CN121998390ACN 121998390 ACN121998390 ACN 121998390ACN-121998390-A

Abstract

The invention relates to a space-time task scheduling method, a space-time task scheduling device and electronic equipment for port transportation, which comprise the steps of obtaining a plurality of transportation tasks to be executed, dynamically screening a plurality of candidate stocking positions for each transportation task based on the residual capacity of the stocking positions and a working distance threshold to form an alternative stocking position set, analyzing the expanded transportation tasks in a full-shore range, identifying a forward-travel task and a return-travel task meeting the conditions of time window matching and path accessibility as a engageable task pair, modeling a vehicle transportation task scheduling process as a Markov decision process, adopting a reinforcement learning model to carry out collaborative optimization decision, outputting actions comprising the tasks, the stocking positions and the paths, generating a vehicle transportation task scheduling scheme according to the optimization decision result, issuing and executing, and dynamically adjusting the engagement relation and the travel path of the subsequent tasks in the execution process. The invention reduces the idle driving distance and waiting time of the vehicle and improves the utilization rate of vehicle resources and the operation efficiency of full shoreline operation.

Inventors

FU CHAO
Man Xintai
WANG XINGYU
HU JIA
ZHAO NING
LIU XIANGWEI
LAI JINTAO
SHEN YANG
ZHANG YUJIE
HE MENGJIE

Assignees

上海海事大学
同济大学

Dates

Publication Date: 20260508
Application Date: 20260410

Claims (10)

1. A method for scheduling space-time tasks for port-oriented transportation, the method comprising: acquiring a plurality of transport tasks to be executed, wherein each transport task comprises a determined starting point, an original target stacking position and a task execution time window; Dynamically screening a plurality of candidate stocking positions for each transportation task based on the residual capacity of the stocking positions and the operation distance threshold value between the residual capacity of the stocking positions and the starting point of the task to form a candidate stocking position set of the task so as to expand the transportation task from a single target position to a target position with multiple candidates; Analyzing the expanded transportation task in the whole shoreline range based on the alternative stacking position set, identifying a forward-travel task and a backward-travel task which meet the condition of time window matching and path accessibility as a connectable task pair, and forming a forward-travel task connection candidate set, wherein the condition of time window matching and path accessibility refers to the sum of the estimated completion time of a front-end task and the empty travel transfer time from the task end position to the starting point of a subsequent task, the sum is not more than the cut-off time of a subsequent task execution time window, and a feasible path for connecting the two task positions exists; Modeling a vehicle transportation task scheduling process as a Markov decision process, and adopting a reinforcement learning model based on an Actor-Critic architecture to carry out collaborative optimization decision, wherein the reinforcement learning model outputs actions for a vehicle according to the current system state comprising a task to be executed, a vehicle state and a work network state, wherein the actions comprise selecting a task from the task to be executed, selecting a stocking position from an alternative stocking position set of the task and planning a driving path; And in the execution process, when detecting deviation or environmental change, dynamically adjusting the connection relation of the follow-up tasks and the driving path.
2. The method for scheduling space-time tasks for port-oriented transportation according to claim 1, wherein the acquiring a plurality of transportation tasks to be performed further comprises: Abstracting a wharf operation area into a directed operation network diagram: Wherein, the The system comprises an operation node set, a control unit and a control unit, wherein the operation node set comprises a shore bridge operation point, a box area field bridge operation point, a vehicle waiting point and a road crossing node; For a set of directed paths that a vehicle can travel, each path An attribute vector representing the path length, the expected transit time, and the congestion factor is associated.
3. The method for scheduling space-time tasks for port-oriented transportation according to claim 1, wherein the dynamically screening a plurality of candidate stocking locations for each transportation task comprises: for any transportation task Its original attribute is Wherein As a starting point for the task, For the original target stowage position, Executing a time window for the task; To construct an alternative stack position set for the same The definition is: Wherein, the For a set of job nodes, Indicating the stacking position Is used for the remaining capacity of the (c), Representing the starting point To a piling position Is used for the distance of (a), For a preset maximum working distance threshold value by introducing To transport tasks Expanded to 。
4. The port transportation-oriented space-time task orchestration method according to claim 3, wherein the identifying the forward task and the return task that meet the time window matching and path reachability condition as engageable task pairs is specifically: Defining tasks And tasks Can be connected when meeting the conditions: And there are slave tasks Ending position to task Starting point Is a feasible path of (a) Wherein, the method comprises the steps of, For the task Is used to determine the expected completion time of the (c) for the (c) process, For vehicles slave tasks The end position is driven to the starting point The required idle driving transfer time, all task pairs meeting the above conditions Constructing the candidate set of the return-to-return task engagement 。
5. The port transportation oriented space-time task orchestration method according to claim 1, wherein the modeling of the vehicle transportation task scheduling process as a markov decision process, in particular comprises: At the time of decision State of system The definition is as follows: Wherein In order for a set of transportation tasks to be performed, As a set of vehicle states, The reinforcement learning model is the motion of the vehicle output The definition is as follows: Wherein Is the slave Is used for the transportation of the selected tasks, To be from the task Alternative stack location set of (c) Is arranged in the storage position selected by the user, For a planned vehicle travel path.
6. The port transportation oriented space-time task orchestration method according to claim 5, wherein the multi-objective rewards function is specifically defined as: Wherein, the Indicating the empty range of the vehicle within the decision step, Indicating the waiting time for the vehicle to operate, Indicating the amount of task delay, Is a preset weight coefficient, and the reinforcement learning model is used for maximizing expected accumulated rewards To update its decision strategy.
7. The port transportation-oriented space-time task orchestration method according to claim 6, wherein the collaborative optimization decision is performed by adopting an reinforcement learning model based on an Actor-Critic architecture, specifically: Modeling each transport vehicle as independent agents, wherein each agent is provided with a neural network model based on an Actor-Critic architecture, and the Actor network is used for controlling the system according to the current system state Output action For selecting tasks, heap locations and paths, critic network evaluation states For guiding policy updates of the Actor network; by the multi-objective rewarding function And co-optimizing strategies for all vehicle agents to achieve global transport efficiency optimization.
8. The method for scheduling space-time tasks for port transportation according to claim 1, wherein the dynamically adjusting the connection relation and the driving path of the following tasks comprises: Triggering dynamic adjustment when detecting that the vehicle execution progress deviates from the expected, the capacity of the target stocking position suddenly changes, or the critical path of the operation network is jammed; Based on the current system state And making a decision for the affected vehicle again through the reinforcement learning model, calculating a new task engagement relationship and a running path, and updating a task arrangement scheme issued to the corresponding vehicle so as to maintain the continuity of the transportation chain.
9. A port transportation oriented space-time task orchestration device, the device comprising: The transport task acquisition module is used for acquiring a plurality of transport tasks to be executed, wherein each transport task comprises a determined starting point, an original target stacking position and a task execution time window; the candidate stacking position screening module is used for dynamically screening a plurality of candidate stacking positions for each transportation task based on the remaining capacity of the stacking positions and the operation distance threshold value between the remaining capacity of the stacking positions and the starting point of the task, and forming a candidate stacking position set of the task so as to expand the transportation task from a single target position to a position with multiple candidate targets; The system comprises a connectable task pair identification module, a storage module and a storage module, wherein the connectable task pair identification module is used for analyzing an expanded transportation task in a full shorelin range based on the alternative storage position set, identifying a forward task and a backward task which meet the condition of time window matching and path accessibility as a connectable task pair and forming a forward task connection candidate set, wherein the condition of time window matching and path accessibility refers to the sum of the estimated completion time of a front task and the empty driving transfer time from the task end position to the starting point of a subsequent task, the sum does not exceed the cut-off time of a subsequent task execution time window, and a feasible path for connecting the two task positions exists; The decision process modeling module is used for modeling a vehicle transportation task scheduling process into a Markov decision process and adopting a reinforcement learning model based on an Actor-Critic architecture to carry out collaborative optimization decision, wherein the reinforcement learning model outputs actions for a vehicle according to the current system state comprising a task to be executed, a vehicle state and a work network state, the actions comprise selecting a task from the tasks to be executed, selecting a stocking position from a candidate stocking position set of the task and planning a driving path; The task arrangement scheme execution module is used for generating a vehicle transportation task arrangement scheme comprising a task sequence, a selected stacking position and a driving path according to an optimization decision result, and issuing the vehicle transportation task arrangement scheme to a corresponding vehicle for execution, and dynamically adjusting the connection relation of the subsequent tasks and the driving path when deviation or environmental change is detected in the execution process.
10. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the port transportation oriented spatiotemporal task orchestration method according to any one of claims 1 to 8 when the computer program is executed.

Description

Port transportation-oriented space-time task arrangement method and device and electronic equipment Technical Field The invention relates to the technical field of intelligent traffic systems and port logistics, in particular to a space-time task arrangement method and device for port transportation and electronic equipment. Background In an automated container terminal, a horizontal transport system is responsible for transferring containers between a quay bridge and a yard, and the operating efficiency of the horizontal transport system directly affects port throughput capacity. The core challenge is how to minimize the empty mileage of the transportation vehicle to increase the utilization of the capacity resources and energy. For this reason, many intelligent scheduling methods have been developed in the industry, and the common idea is to use an optimization algorithm to allocate appropriate transportation tasks to vehicles in real time and plan a driving path so as to realize "unequal vehicles and unequal vehicles". However, the prior proposal is deeply analyzed to find that a fundamental constraint factor is not solved effectively, namely, the rigid binding relation between the transportation task and the operation position. In the mainstream scheduling model, each task is assigned a unique and determined job location (storage location or container extraction location) at the time of generation, whether it is a ship-off task (from a quay to a specified yard) or a ship-on task (from a specified yard to a quay). The dispatch system performs task matching and path planning under this rigid constraint, i.e., the vehicle must travel to the fixed location to complete the task. This mode results in severely limited flexibility in system optimization. Specifically, since the task target position is single and fixed, the system has extremely narrow selection space when searching for multiple tasks of continuous and forward path for the vehicle. For example, after a vehicle completes a ship unloading task, the start of its next task is locked in a fixed yard position immediately after the unloading is completed. The system can only attempt to find a time window matching shipping mission around the fixed point. If there is no suitable mission nearby, or if the path to the fixed point is temporarily congested, the vehicle is often forced to travel to other areas, breaking the continuous cargo work chain. Vice versa for the loading task. This fixed-position-based task engagement is essentially "near match" logic, which, while reducing partial dead drive, does not allow for the active construction of a closed-loop transport chain at the global level formed by the natural coupling of the outbound and inbound tasks. Because the key to making a closed loop is the high degree of forward-path and close-in-time engagement of the two tasks on the spatial path, a single fixed location greatly limits the probability that such forward-path combinations will be found and successfully matched. Therefore, the existing scheme fails to fundamentally break through the rigidity limit of the task model, and the scheduling optimization is performed in a narrow solution space. This results in the phenomenon of empty vehicles not being completely eliminated, and the overall system transportation efficiency is improved by bottlenecks. Disclosure of Invention Based on the above, it is necessary to provide a space-time task scheduling method, device and electronic equipment capable of enhancing flexibility of task execution from the source, thereby creating a possible port-oriented transportation space-time task scheduling method, device and electronic equipment for dynamically scheduling seamless continuous transportation chains in a full shoreline range. The invention provides a space-time task arrangement method for port transportation, which comprises the following steps: acquiring a plurality of transport tasks to be executed, wherein each transport task comprises a determined starting point, an original target stacking position and a task execution time window; Dynamically screening a plurality of candidate stocking positions for each transportation task based on the residual capacity of the stocking positions and the operation distance threshold value between the residual capacity of the stocking positions and the starting point of the task to form a candidate stocking position set of the task so as to expand the transportation task from a single target position to a target position with multiple candidates; Analyzing the expanded transportation task in the whole shoreline range based on the alternative stacking position set, identifying a forward-travel task and a backward-travel task which meet the condition of time window matching and path accessibility as a connectable task pair, and forming a forward-travel task connection candidate set, wherein the condition of time window matching and path accessibility refers to the sum of