CN-122022224-A - Berth allocation optimization method considering berthing of dry bulk cargo wharf

CN122022224ACN 122022224 ACN122022224 ACN 122022224ACN-122022224-A

Abstract

The application provides a berth allocation optimization method considering migration berthing of a dry bulk cargo wharf, which belongs to the technical field of berth allocation and comprises the steps of acquiring ship data, wharf berth attribute information and environment data, and constructing a mixed integer linear programming model based on the acquired ship data, wharf berth attribute information and environment data; the method comprises the steps of generating an initial solution by adopting a dynamic priority heuristic initialization strategy, extracting a state feature vector of the initial solution, pre-training a depth Q network model, inputting the state feature vector of the initial solution into the trained depth Q network model, selecting a neighborhood operator with an optimal Q value by utilizing the depth Q network model, and minimizing the total harbor service time of all ships so as to optimize a berth allocation scheme.

Inventors

ZHANG JIE
LIU CHANGHUI
XUE SONG
ZHAO DONG
WANG HENGSEN
WANG YUQI
ZHANG WEI
ZHAO WEILI
WU YUZHEN
JIA PENG
ZHANG LEI
QI LIANG
YIN ZHANQIANG
GUO HONGYUE

Assignees

青岛港国际股份有限公司
青岛港国际股份有限公司前港分公司

Dates

Publication Date: 20260512
Application Date: 20251204

Claims (10)

1. A method for optimizing berth allocation for a dry bulk terminal with respect to berthing, the method comprising: acquiring ship data, dock berth attribute information and environment data, and constructing a mixed integer linear programming model based on the acquired ship data, dock berth attribute information and environment data; generating an initial solution by adopting a dynamic priority heuristic initialization strategy; Extracting a state feature vector of an initial solution; Pre-training a deep Q network model; and inputting the state feature vector of the initial solution into a trained deep Q network model, selecting a neighborhood operator with an optimal Q value by using the deep Q network model, and minimizing the total harbor service time of all ships so as to optimize a berth allocation scheme.
2. The method of claim 1, wherein obtaining vessel data comprises obtaining an estimated arrival time, a hull length, a draft, a loaded cargo weight for each vessel in the collection of vessels to be dispatched; the method comprises the steps of obtaining dock berth attribute information of each dock berth, and specifically obtaining the berth length, the maximum allowable draft, the operation efficiency and the berth type of each dock berth, wherein the berth type comprises deep water berths and shallow water berths; The acquisition of environmental data specifically includes acquisition of a tidal window.
3. The method of claim 1, wherein the mixed integer linear programming model includes the following constraints: Minimizing the total on-port service time of all vessels expressed as: Wherein, the Representing a ship Is used for the time of the in-port service, Representing a collection of all of the vessels, , , Representing the number of vessels; the first berthing start time of each ship is not earlier than its expected arrival time, expressed as: Wherein, the Representing a ship Is the first berthing start time; Representing a ship Is determined by the time of arrival of the mobile station; If the ship needs to be berthed, the berthing starting time must be expressed as follows after the first berthing operation of the ship is completed: Wherein, the Representing a ship Is used for the transfer start time of the (c), ; Indicating if the ship The need for transfer, then equals 1, otherwise equal to 0; Representing a sufficiently large positive number; The ship service time when the ship does not need to be berthed is expressed as follows: Wherein, the Representing a ship Waiting time before first berthing; Representing a ship The weight of the loaded cargo; Representing berths Is provided; Representing a slave berth The time to leave the dock; Representing a ship Whether to first berth to the berth ; Representing a deep water berth set, wherein, Representing the number of deep water berths; Representing a shallow water berth set, wherein, Is the number of shallow water berths; The ship service time when the ship needs to be berthed is expressed as follows: Wherein, the Representing a ship Is used for reducing the load of the vehicle; Representing a slave berth To berth Is used for the transfer time of the (a), , ; Representing berths Is provided; Representing a slave berth The time to leave the dock; Representing a ship Whether to move to the berth ; For each ship Modeling the load shedding amount selection of the model, and the expression is as follows: Wherein, the Representing a ship Whether to select a load shedding ratio; For each ship Modeling the load shedding amount of the model, and defining the actual load shedding amount as a weighted sum of binary selection variables, wherein the expression is as follows: Wherein, the E {1, 2..10 } is 0.1 x h, if y i,h =1, it means that the ship i is 0.1 x h; the total waiting time of the ship satisfies: Wherein, the Representing a ship At berth Is a work completion time of (a); Ensuring each ship Is precisely allocated an initial berth, and the expression is as follows: when the ship needs to be berthed, corresponding berthing positions are allocated to the ship, and the expression is as follows: Berth for berths The length of (2) is as follows: Wherein, the Representing berths Is a length of (2); Representing a ship Is provided with a plurality of grooves for the length of the hull, ; Berth for berths The length of (2) is as follows: Berth for berths The draft of (2) satisfies: Wherein, the Representing berths Is set at the maximum allowable draft of (2); Berth for berths The draft of (2) satisfies: Wherein, the Representing berths Is set at the maximum allowable draft of (2); Representing a ship Draft upon arrival; Representing a ship No-load draft; The tide rise constraint of the ship meets the following conditions: Wherein, the Representing a ship Whether the first berthing start time of (a) is at the first Within a flood tide window; Represent the first Start time of each climax window; Represent the first The end time of the individual climax windows; representing a tidal window set; the ship allocated to the same berth, the latter ship can only start berthing after the former ship leaves, and the expression is: Wherein, the Representing a ship At berth Is a work completion time of (a); Representing a ship Whether to move to the berth ; Representing a ship Whether or not it is a ship At berth Is a ship in front of the right-hand ship , , ; Limiting the range of decision variables, expressed as: 。
4. the method of claim 3, wherein generating an initial solution using a dynamic priority heuristic initialization strategy comprises: The method comprises the steps of adopting a 4 Xn integer matrix to represent an initial solution, wherein a first row represents the number of each ship and the number of each ship is numbered according to the expected arrival time of the ship, a second row represents the berth of each ship where the ship is berthed for the first time, a third row represents the load reduction amount of each ship, and a fourth row represents the berth of each ship; dynamically adjusting the priority of each ship according to the predicted arrival time of the ship and the congestion degree of berth resources; Dividing ship types based on draft, wherein ships with draft greater than or equal to the maximum allowable draft are defined as large ships and ships with draft less than the maximum allowable draft are defined as small ships; setting the initial available time of each berth to be 0; for the ship with the highest priority, generating a feasible berth set of the ship with the highest priority based on the type and the length of the ship, wherein the feasible berth set is a berth with the length of alloted berth bits being more than or equal to the length of the ship and the berth depth being more than or equal to the draft of the ship; if the ship with the highest priority is a large-sized ship and the current deep water berth resource crowding degree is lower than a threshold value, the ship is distributed to the deep water berth with the earliest available time, otherwise, load shedding and berthing operation is triggered, load shedding and berthing are determined, and load shedding and berthing are carried out; If the ship with the highest priority is a small ship, the ship is preferentially distributed to the shallow water berth, and when the crowding degree of the shallow water berth resource is higher than a threshold value, the berthing operation is carried out on the ship; updating the available time of each berth; and returning to continuously executing the dynamic adjustment of the priority of each ship until all the ships obtain the initially feasible berths, namely the initial solution.
5. The method of claim 4, wherein the expression of the congestion level of the berth resources is: Wherein τ represents the congestion degree of the berth resource; representing the proportion of the large ship to the whole ship set V, reflecting the required pressure of the deep water berth, wherein ρ represents the relative deviation of the average finishing time of the deep water berth and the shallow water berth, and the expression is as follows: Wherein, the Representing berths Is the earliest time of availability of (2); Representing berths Is the earliest time of availability of (2); When the ship is a large ship and the current berth resource crowding degree tau > lambda, the calculation formula of the load reduction of the large ship is as follows: Wherein, the Representing large vessels Is used for reducing the load of the vehicle; Representing a ship No-load draft; Representing a ship Draft upon arrival; Representing berths Is set at the maximum allowable draft of (2); The expression of the berth is: Wherein, the Representing a ship At the end of the job at the head-on berth.
6. The method of claim 5, further comprising constructing a neighborhood operator for a variable neighborhood search algorithm prior to pre-training the deep Q network model, comprising: Reassigning shallow water berths with the crowding degree of the shallow water berth resources being higher than a threshold value, selecting a shallow water berth with the latest finishing time, randomly selecting a ship from a service queue of the shallow water berth, randomly selecting the shallow water berth from a feasible berth set of the ship, and assigning the ship to the selected shallow water berth; randomly selecting a deep water berth, selecting a small ship from a service queue of the deep water berth, randomly selecting a shallow water berth from a feasible berth set of the small ship, and distributing the small ship to the selected shallow water berth; Selecting a ship needing to be berthed randomly, wherein the load shedding amount of the ship is increased by 0.1 or reduced by 0.1, selecting a berth with the latest finishing time in a shallow water berth set, randomly selecting a ship which is not arranged with berthing operation from the berthing service queue, reallocating the ship to a deep water berth, and determining the load shedding amount of the ship, thereby executing berthing; randomly selecting one berth and then selecting two vessels from the service queues thereof The positions of the service sequences are not more than 3 bits apart, and the service sequences are exchanged; And randomly selecting two berths of the same type, generating 0-1 exchange state feature vectors on the service queues, and exchanging the ship marked as 1 between the two service queues if the draft of the ship is smaller than or equal to the depth of the allocated berths and the hull length of the ship is smaller than or equal to the berth length of the allocated berths.
7. The method of claim 6, wherein pre-training the deep Q network model specifically comprises: training a deep Q network model by designing state feature vectors, action spaces and rewarding functions -Greedy strategy selects a neighborhood operator and directs deep Q network model training by the reward function; Inputting the state feature vector of the initial solution into a trained deep Q network model, selecting a neighborhood operator with an optimal Q value by using the deep Q network model, and minimizing the total harbor service time of all ships so as to optimize a berth allocation scheme, wherein the method specifically comprises the following steps of: loading a trained deep Q network model, and inputting the state feature vector of the initial solution into the trained deep Q network model; In each iteration, the trained depth Q network model evaluates the Q value of each neighborhood operator, selects the neighborhood operator with the optimal Q value, and executes local search in the neighborhood operator; And repeatedly iterating until the termination condition is met, outputting an optimal neighborhood operator, and obtaining a berth allocation scheme corresponding to the optimal neighborhood operator.
8. The method of claim 6, wherein the expression for the state feature vector is: Wherein, the A state feature vector representing a decision point t; Representing access frequency of each neighborhood operator of the decision point t; A mean value of load shedding amount of the decision point t is represented; representing the variance of the load shedding amount of the decision point t; Representing the utilization rate of each berth of the decision point t; Representing the total waiting time of each berth vessel at decision point t; representing the total service time of all vessels at decision point t; the total waiting time of all vessels representing the decision point t; representing the duty cycle of a large ship; The estimated arrival time sparse coefficient of the ship is represented; the expression of the mean value of the load reduction amount of the decision point t is as follows: the variance of the subtracted amount at decision point t is expressed as: Wherein R (t) represents a set of vessels that need to be moored at a decision point t; the expression of each berth utilization rate of the decision point t is as follows: Wherein m 1 represents the number of deep water berths, and m 2 represents the number of shallow water berths; For berths The utilization rate of the method at the decision point t is as follows: Wherein, the Representing berths at decision point t Is used for the utilization rate of the (a); Representing the total time that berth j is occupied by the vessel at decision point t, Maximum completion time for all berths; the expression of the total waiting time of each berth vessel at decision point t is: For berths The expression of the total vessel waiting time of berth j at decision point t is: Where V j (t) represents the set of vessels allocated to berth j at decision point t, Representing the operation ending time of the ship i at the first berth; the expression of the large ship duty ratio is: the expression of the sparse coefficient is: 。
9. The method of claim 8, wherein the action space is a six-dimensional discrete set corresponding to six neighborhood operators { N 1 , N 2 , …, N 6 }, expressed as: A={1,2,...,6} a represents an action space; At decision point t, estimating a state-action value { Q (S (t), a; θ) } a =1 by propagating a current state feature vector S (t) through a deep Q network model; By using -Greedy policy with probability Selecting a random action, otherwise, selecting an action a (t) that maximizes the state-action value Q (S (t), a; θ); determining a neighborhood operator N a(t) corresponding to the action a (t); And carrying out neighborhood search on the current solution X (t) by adopting a neighborhood operator N a(t) .
10. The method of claim 9, wherein the reward function consists of a neighborhood value reward r 1 , a target value reduction reward r 2 , and a neighborhood exploration reward r 3 ; the total reward of the decision point t is R (t) =r 1 +r 2 +r 3 ; Let the current solution of decision point t be X (t), the target value be J (X (t)); applying a neighborhood operator N a(t) to generate a candidate solution set { X (1) (t), X (2) (t), …, X (L) (t) } by L times of local perturbation, setting Γ to represent an index set of solutions that reduce the target value, Γ= { k|J (X (k) (t)) < J (X (t)) }; The expression of the neighborhood value prize r 1 is: The target value reduction reward is used to measure the average magnitude of the reduction based on the candidate solution set and normalize it using the difference between the best target value currently found J (X best ) and the initial target value of the current solution J (X init ); the expression for the target value reduction prize r 2 is: the neighborhood exploration rewards guide the intelligent agent to explore different neighborhoods more in the early stage, so that premature convergence to local optimum is avoided; Let C a (t) be the total number of times action a is performed before decision point t, then the expression for neighborhood exploration prize r 3 is: 。

Description

Berth allocation optimization method considering berthing of dry bulk cargo wharf Technical Field The invention belongs to the technical field of berth allocation, and particularly relates to a berth allocation optimization method considering berthing of a dry bulk cargo wharf. Background The dry bulk cargo wharf is taken as an important link in a logistics supply chain, and the berth distribution efficiency directly influences the overall operation cost of the port and the ship turnover rate. Along with the trend of large ships, berth resources are increasingly tense, and how to scientifically and reasonably allocate berths becomes a core problem of dock operation management. Particularly in dry bulk terminals, the berthing operation is a common scenario due to differences in factors such as ship draft, cargo loading, etc., which makes berthing allocation problems more complex. In the prior art, aiming at the problem of berth allocation, a mathematical programming method or a meta-heuristic algorithm is generally adopted for optimization. Integer linear programming models are used to handle constraints in berth assignments, while heuristic methods such as genetic algorithms, simulated annealing, etc., search the solution space to find near optimal solutions. The method can reduce the time of the ship in the harbor to a certain extent and improve the berth utilization rate. However, the prior art has high calculation complexity when processing a berthing scene of a dry bulk cargo wharf, is difficult to cope with large-scale real-time scheduling requirements, has low initial solution quality and influences subsequent optimization effects, and meanwhile, the prior method cannot adaptively select an optimal search operator and is easy to sink into a local optimal solution. Disclosure of Invention The invention provides a berth allocation optimization method considering berthing of a dry bulk cargo wharf, which at least solves the problems that an optimal search operator cannot be selected in a self-adaptive mode and a local optimal solution is easy to fall into in the prior art. The embodiment of the application provides a berth allocation optimization method considering berthing of a dry bulk cargo wharf, which comprises the following steps: acquiring ship data, dock berth attribute information and environment data, and constructing a mixed integer linear programming model based on the acquired ship data, dock berth attribute information and environment data; generating an initial solution by adopting a dynamic priority heuristic initialization strategy; Extracting a state feature vector of an initial solution; Pre-training a deep Q network model; and inputting the state feature vector of the initial solution into a trained deep Q network model, selecting a neighborhood operator with an optimal Q value by using the deep Q network model, and minimizing the total harbor service time of all ships so as to optimize a berth allocation scheme. Further, acquiring ship data, which specifically comprises the steps of acquiring the estimated arrival time, the hull length, the draft and the loaded cargo weight of each ship in a ship set to be scheduled; the method comprises the steps of obtaining dock berth attribute information of each dock berth, and specifically obtaining the berth length, the maximum allowable draft, the operation efficiency and the berth type of each dock berth, wherein the berth type comprises deep water berths and shallow water berths; The acquisition of environmental data specifically includes acquisition of a tidal window. Further, the mixed integer linear programming model includes the following constraints: Minimizing the total on-port service time of all vessels expressed as: Wherein, the Representing a shipIs used for the time of the in-port service,Representing a collection of all of the vessels,,,Representing the number of vessels; the first berthing start time of each ship is not earlier than its expected arrival time, expressed as: Wherein, the Representing a shipIs the first berthing start time; Representing a ship Is determined by the time of arrival of the mobile station; If the ship needs to be berthed, the berthing starting time must be expressed as follows after the first berthing operation of the ship is completed: Wherein, the Representing a shipIs used for the transfer start time of the (c),;Indicating if the shipThe need for transfer, then equals 1, otherwise equal to 0; Representing a sufficiently large positive number; The ship service time when the ship does not need to be berthed is expressed as follows: Wherein, the Representing a shipWaiting time before first berthing; Representing a ship The weight of the loaded cargo; Representing berths Is provided; Representing a slave berth The time to leave the dock; Representing a ship Whether to first berth to the berth;Representing a deep water berth set, wherein,Representing the number of deep water berths; Representing a s