CN116402173A

CN116402173A - Intelligent algorithm for distributing container areas and container positions of ship unloading container based on machine learning

Info

Publication number: CN116402173A
Application number: CN202211099026.7A
Authority: CN
Inventors: 王文渊; 彭云; 郭子坚; 刘华锟
Original assignee: Dalian University of Technology
Current assignee: Dalian University of Technology
Priority date: 2022-09-06
Filing date: 2022-09-06
Publication date: 2023-07-07

Abstract

The invention belongs to the field of container yard container area and container position allocation, and provides an intelligent ship unloading container area and container position allocation algorithm based on machine learning. Collecting historical data of container terminal ship unloading operation and yard states, calculating the yard states, and extracting typical yard states based on an unsupervised learning algorithm and a clustering algorithm; analyzing the influence factors of container unloading box area selection, extracting the selection rules of the unloading box storage box area based on machine learning, and designing an intelligent algorithm for allocation of the unloading box area and an heuristic intelligent algorithm for allocation of the unloading box position based on the rules. The invention considers the influence of container characteristics, operation characteristics and yard status characteristics on the allocation decision of the container areas in the actual operation of the wharf, utilizes machine learning based on massive historical operation data to effectively extract high-quality scheduling experience, provides decision support for the operation scheduling of the container yard, is beneficial to relieving the traffic jam of the yard, improving the production operation efficiency of the container yard and promoting the intelligent construction of the container yard.

Description

Intelligent algorithm for distributing container areas and container positions of ship unloading container based on machine learning

Technical Field

The invention relates to the field of container yard container area and container allocation, in particular to an intelligent algorithm for unloading container areas and container allocation based on machine learning.

Background

Along with the continuous increase of the throughput of the container terminal, the operation pressure of the terminal is increased, and the allocation strategy of the container areas and the container positions of the container terminal yard is formulated reasonably, so that the method has important significance for improving the utilization rate and turnover rate of the container yard and operation facilities, relieving the traffic jam of the container yard and improving the production operation efficiency of the terminal. The strong coupling relation between container yard operation and ship to port operation and land side suitcase operation is shown in Jiang, X.J., jin, J.G,2017.A branch-and-price method for integrated yard crane deployment and container allocation in transshipment yards, transport. Res. Part B98,62-75, "Wang K, zhen L, wang S.A., et al,2018.Column Generation for the Integrated Berth Allocation,Quay Crane Assignment,and Yard Assignment Problem.Transp.Sci.52 (4), 812-834," and "Yang, L.Y., ng, T.S., lee, L.H.2022.A robust approximation for yard template optimization under uncertaintry. Transport. Res. Part B160,21-53," which are obviously affected by randomness and uncertainty factors, the traditional mathematical modeling and operational optimization technology is difficult to accurately describe the yard operation process and effectively solve, and the yard planning difficulty of real scale is large in advance. In the actual operation process, a dispatcher often needs to make an on-site decision according to the real-time state of a storage yard and combines scheduling experience formed by accumulation of daily and monthly accumulation, so that the on-site decision is easily interfered by subjective and objective factors, and the stability of decision quality is difficult to ensure. How to fully utilize the massive historical data of container yard operations, deeply mine high-quality scheduling experience, and effectively apply to making yard operation decisions, and become an important problem facing intelligent construction of container terminals.

Disclosure of Invention

In order to solve the problems, the invention provides an intelligent algorithm for distributing the boxes and the boxes of the ship unloading container based on machine learning. Considering box allocation influence factors in the actual operation process of the container terminal, under the condition that the historical operation data of the ship unloading box, the container attributes and the storage yard state are known, deep mining is carried out on the historical operation data of the ship unloading box, implicit association rules between allocation decisions of the ship unloading box and the container attribute features, the operation features and the storage yard state features are extracted by utilizing an unsupervised learning concept, a random forest algorithm, big data mining and a statistical theory in machine learning, a box allocation heuristic algorithm is designed according to the extracted rules, the effectiveness of the corresponding index evaluation rule extraction and allocation algorithm is selected, and decision support is provided for real-time scheduling of port production operation.

The technical scheme of the invention is as follows:

an intelligent algorithm for distributing boxes and positions of ship unloading containers based on machine learning comprises the following steps:

step one, collecting historical data of container terminal ship unloading operation and storage yard state, wherein the historical data comprise container terminal plane layout data, ship unloading operation data and storage yard state data;

The plane layout data of the container terminal comprises the number of berths, the layout form of a storage yard and the number of boxes; the ship unloading operation data comprise ship unloading time, ship unloading berth, box dropping area and ship unloading box data; the ship unloading box data comprise an inlet and outlet type and a container owner; the yard status data comprises the piled up quantity, the operation number, the bridge number and the piled up box operation proportion of each box area in the yard at each time period; wherein the piled up quantity of the box region comprises the piling up number of each shellfish position and each stack of the box region; the operation number of the box area comprises the box stacking operation number, the box taking operation number and the box reversing operation number; the number of field bridges in the box area is the total number of field bridges with operation records in the box area in each time period; the box stacking operation proportion of the box area is the proportion of the box stacking operation number of the box area to the total operation number;

step two, calculating a storage yard state according to historical data based on an unsupervised learning algorithm, and extracting a typical storage yard state by using a clustering algorithm;

the yard state comprises a piled stock state, an operation number state, a yard bridge number state and a piling box operation proportion state; wherein, a certain state of the storage yard is represented by a vector formed by corresponding state values of all the box areas, and the length of the vector is equal to the number of the box areas in the storage yard;

The corresponding state of each box area is the relative size of each parameter in the storage yard state data of the current box area in 8 adjacent box areas; the parameters are the stock quantity/the operation number/the bridge number/the stacking box operation proportion;

the calculation method comprises the following steps: comparing the storage yard state data of any box area with adjacent box areas in pairs in sequence, and when the box area parameter value is greater than that of a certain adjacent box area, adding a corresponding state value of the box area to +1; when the box parameter value is smaller than a certain adjacent box, the corresponding state value of the box is-1; when the box parameter value is equal to a certain adjacent box, keeping the corresponding state value unchanged; after the comparison with all 8 adjacent box areas is finished, the corresponding state value of the accumulated box areas is obtained, and the value range is [ -8,8]; and the larger the state value is, the larger the parameters such as the stock quantity/operation number/bridge number/box stacking operation proportion of the box region are in the adjacent box region.

The state value vectors of the piled up quantity, the operation number, the field bridge number, the piled up box operation proportion and the like of each box area in the piled up field in different time periods are different, and meanwhile, the state value vectors in adjacent time periods have similarity. After the storage yard state value vectors in different time periods are obtained through calculation, an unsupervised learning algorithm in machine learning is utilized to construct a storage yard state value vector clustering model based on K-means, the distances between the center points of the clusters and the samples of different cluster numbers are utilized to represent the clustering accuracy, and the clustering accuracy is higher as the number of clusters is larger, and the clustering efficiency is lower. And comprehensively considering the clustering efficiency and the clustering accuracy, and selecting a proper clustering cluster number for clustering.

Constructing a K-means-based storage yard state value vector clustering model, and respectively clustering four types of storage yard states, namely the stored quantity of each box area, the operation number of each box area, the field bridge number of each box area, the stacking operation proportion of each box area and the like; in each type of storage yard state, taking the average value of the storage yard state value vectors of each cluster as the typical storage yard state of the cluster sample, and extracting to obtain the typical storage yard state, wherein the number of the typical storage yard states is equal to the number of the clusters.

Thirdly, analyzing influence factors selected in a container unloading area;

obtaining container unloading box area selection influencing factors including container attribute characteristics, unloading characteristics and storage yard state characteristics through statistics of box area selection conditions of different types of unloading containers in historical data; the container attribute features include container import and export types; the ship unloading characteristics comprise a ship unloading berth number; the storage yard status features comprise a stored quantity type, an operation number type, a field bridge number type and a stacking box operation proportion type of each box area;

extracting a selection rule of a ship unloading box stockpiling box region based on machine learning;

and (3) obtaining a typical storage state cluster with corresponding quantity of stored quantity states, operation number states, field bridge number states, box stacking operation proportion and the like in each hour by using a storage state clustering method in the step two, and digging a storage box region selection rule of the ship unloading box in the historical data on the basis.

Based on a supervised learning concept in machine learning, when the state value vectors of the stored quantity, the operation number, the field bridge number and the stacking operation proportion of each box region in a storage yard respectively belong to a corresponding type of typical storage yard state, counting box region allocation features under the storage yard state combination in historical data; the bin area allocation feature includes: a stored quantity state value, an operation number state value, a field bridge number state value, and a stacking operation proportion state value of the allocated bin area; counting the number of the box areas with different state values of the piled quantity, the operation number, the field bridge number and the piling operation proportion under each storage yard state combination, calculating the occurrence probability, and taking the box area selection probability with each state value under each storage yard state combination as a ship unloading box piling box area selection rule;

step five, designing an intelligent algorithm for the ship unloading box area allocation based on rules, wherein the intelligent algorithm comprises feature importance calculation, heuristic comparison rule design and algorithm verification;

in order to determine the influence degree of different influence factors on the allocation of the ship unloading box areas, firstly, a machine learning model is constructed by adopting a random forest algorithm with a disordered sequence, the importance of the allocation characteristics of the box positions of the ship unloading box areas is calculated by adopting the random forest algorithm with the disordered sequence, the allocation characteristics comprise the storage quantity state, the operation number state, the field bridge number state and the box piling operation proportion state of each box area of a storage yard, the selectable box dropping box area set of the ship unloading box is generated by utilizing a ship unloading box area allocation intelligent algorithm based on rules on the basis, and the specific implementation method of the ship unloading box area allocation intelligent algorithm based on the rules is as follows:

(1) The ship unloading box area distribution intelligent algorithm based on the rule comprises the following specific steps:

first, the ship unloading time t of the ship unloading box i is initialized _i Storage yard status

Storage yard state rule set { storage quantity state rule C, operation number state rule J, field bridge number state rule Y, stacking box operation proportion state rule R }, and alternative box region set B _feasible The minimum number M of the candidate box regions, and in four types of rule sets C, J, Y, R, the corresponding types of rules are respectively ordered according to the sequence from high to low of the box region selection probability;

secondly, sequentially calling RULE sets corresponding to the corresponding states according to the order of the importance of the four types of storage yard states from high to low to generate a selectable box area set, wherein the calling order is expressed as RULE ₁ 、RULE ₂ 、RULE ₃ 、RULE ₄ ；

According to the ship unloading operation sequence, for each ship unloading box i, firstly calculating the storage yard state s in the current storage yard time period ₁ Selecting RULE in order from high to low ₁ If the box with the highest probability state exists, putting into an alternative set B ₁ And enter the second layer of screening, if there is no box area of this type, continue to select RULE ₁ Bin with next highest probability state in (B) until alternative set B ₁ Is not an empty set;

in the second layer screening, the yard state s is calculated ₂ From high according to selection probabilityTo a low order, sequentially select RULE ₂ Bin with higher probability state up to alternative set B ₁ And B ₂ The intersection of (1) is not an empty set, and entering a third layer of screening;

in the third layer screening, the storage yard state s is calculated ₃ Sequentially selecting RULE according to the order of the selection probability from high to low ₃ Bin with higher probability state up to alternative set B ₁ 、B ₂ And B ₃ The intersection of (1) is not an empty set, and entering a fourth layer of screening;

in the fourth layer screening, the yard state s is calculated ₄ Sequentially selecting RULE according to the order of the selection probability from high to low ₄ Bin with higher probability state up to alternative set B ₁ 、B ₂ 、B ₃ And B ₄ Is not an empty set and the alternative set B ₁ 、B ₂ 、B ₃ And B ₄ Adding the intersection of the alternative sets B _feasible ；

Judgment B _feasible Whether the minimum number M of the candidate boxes is satisfied, if so, outputting a candidate set B _feasible ；

If the requirements are not met, starting from the fourth layer of screening, continuing to select the highest probability state box area in the residual rules to add the alternative set B _feasible Stopping screening until meeting the requirement;

if the requirements are not satisfied yet, returning to the third layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B ₃ Repeating the fourth layer screening process until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

If the requirements are not satisfied, returning to the second layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B ₂ Repeating the third and fourth screening processes until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

If the requirements are not satisfied, returning to the first layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B ₁ Repeating the second and third stepsAnd (3) a four-layer screening process, namely stopping screening until meeting the requirement, and outputting an alternative set B _feasible 。

(2) Heuristic contrast rules are used to determine the set B of candidate bins _feasible Determining a final box falling area, wherein the final box falling area comprises a random principle, a nearby principle and an idle principle; wherein the random principle is at B _feasible The final box falling area is selected randomly, and the principle nearby is that B is _feasible Selecting the nearest box area to the unloading berth of the container in the current task, wherein the idle principle is that in B _feasible Selecting a box area with the least number of owners or piled quantity; wherein, the box areas piled up with the cargo owner containers are preferentially selected.

(3) The algorithm verifies the tank area allocation effect of the intelligent algorithm for verifying the tank area allocation of the ship unloading tank based on the rule, and in the tank piling process, the larger the operation amount in a single tank area is, the larger the operation pressure of a field bridge is, so that the longer the time for waiting for the field bridge operation by the collecting card is; meanwhile, in the future suitcase process, the more the number of owners in a single suitcase area is, the more the number of the suitcases is, and the higher the probability of generating traffic jam is. Therefore, the evaluation index used includes maximum, minimum, average and variance of the work load and the number of owners of each box in the yard.

Step six, design of a ship unloading box position distribution heuristic intelligent algorithm, which comprises setting of a box position distribution principle, design and verification of the ship unloading box position distribution heuristic algorithm;

and D, according to comparison results of different box region selection principles in the fifth step, selecting the optimal principle to determine the final box region. After the box falling area of the ship unloading box is determined, the box position distribution is carried out according to the following principle.

The ship unloading box position allocation heuristic intelligent algorithm follows the box position allocation principle which comprises the following steps: (1) The containers of the same cargo owner are stacked as much as possible with the same or adjacent shellfish positions, and the containers of different cargo owners are stacked as much as possible in a scattered manner;

(2) When selecting the box positions in the shellfish positions, selecting stack positions with fewer residual box positions as far as possible;

(3) When different cargo owner containers are piled up in the same bin, different stacking positions are selected as far as possible to respectively pile up the cargo owner containers.

Determining a box falling area of any ship unloading box and judging; the logic of the heuristic algorithm for distributing the ship unloading boxes is as follows;

judging whether containers with the same cargo owner exist in the current box area or not;

1) When the current box area does not have containers with the same cargo owner, further judging whether the current box area has a shellfish position which is not occupied by other cargo owners;

1.1 When the shellfish level occupied by other cargo owners exists, selecting the shellfish level which is the farthest from the shellfish levels occupied by other cargo owners from the feasible shellfish levels, and selecting the stack with the least stacking quantity for stacking;

1.2 When all the shellfish positions are occupied by other cargo owners, selecting the shellfish position with the lowest cargo owner number from the feasible shellfish positions, and selecting the stack with the largest stacking amount for stacking;

2) When the current container area has containers with the same cargo owner, judging whether the container is occupied by the current cargo owner container or not;

2.1 If there is any bin occupied by the current cargo owner container, further judging whether there is any empty bin; 2.1.1 If the spare box is available, selecting a stack with the least remaining box and the spare box for piling up; 2.1.2 If there is no empty bin, further judging whether there is a situation that the bin is not occupied by other cargo owners;

2.1.2.1 When other shellfish positions which are not occupied by other cargo owners exist, selecting the shellfish position closest to the shellfish position occupied by the cargo owner, and selecting a stack with the least number of piled cargo owners and the least residual box position for piling;

2.1.2.2 When each bin is occupied by other owners, further judging whether the bin stored in the current owner container exists or not;

2.1.2.2.1 When the stock level of the current cargo main container is stored, selecting a stack with the least cargo main number and the least residual container level from the stock level of the current cargo main container to store;

2.1.2.2.2 When the stock level of the current cargo main container is not stored, selecting the stock level closest to the cargo main stock level from all possible stock levels, and selecting a stack with the minimum cargo main number and the minimum residual box level for storage;

2.2 When the current bin area has only the bin positions occupied by the current cargo owner and other cargo owners at the same time or the bin positions occupied by other cargo owners at the same time, further judging whether the bin positions containing the cargo owner container have spare bin positions or not;

2.2.1 When spare boxes exist in the boxes containing the cargo main container, selecting the box with the least cargo main number from all possible boxes containing the cargo main container, and selecting a stack with the least cargo main number and the least remaining boxes for stacking;

2.2.2 When the container containing the cargo owner has no spare container, selecting the container closest to the cargo owner container, and selecting a stack with the least cargo owner number and the least remaining container for stacking;

and after the box bit is determined, verifying the box bit distribution result. When the boxes are taken in the box area, if the containers are stored together with the cargo owner, the movement of the field bridge in the box area is reduced, and the box taking operation efficiency of the field bridge is improved; if the containers are stacked in a dispersed manner with the cargo owner, empty vehicle movement of a field bridge along the shellfish position direction in the process of carrying the containers can be increased, the operation efficiency is reduced, the queuing time of taking the containers and collecting the cards is further prolonged, and traffic jam between the containers is caused. Considering the distribution condition of the containers of the same cargo owner in a box area, verifying a ship unloading box position distribution heuristic algorithm, wherein the stacking density of the containers of the same cargo owner is used as a box position distribution effect evaluation index, and the calculation method is shown in the following formula;

Wherein n is _i The bin number f occupied by the owner i in the current bin _i The forefront bin number, l, occupied by the owner i in the current bin _i The rearmost shell number occupied by the cargo owner i in the current box area is numbered;

the range of the stacking density of the containers of the same cargo owner is (0, 1), and the closer the index is to 0, the more dispersed the stacking positions of the containers of the same cargo owner are, the closer to 1, and the more dense the stacking positions of the containers of the same cargo owner are.

And setting the cluster number to 8, and extracting 8 types of typical storage yard states.

The invention has the beneficial effects that: the invention considers the influence of container characteristics, operation characteristics and yard status characteristics on the allocation decision of the container in the actual operation process of the wharf, extracts association rules between the container allocation and the container characteristics, the operation characteristics and the yard status characteristics in the historical data by using a data mining method based on massive historical operation data, and designs an intelligent algorithm for the container allocation and the allocation of the container. The method effectively utilizes massive historical operation data of the container terminal, considers container characteristics, operation characteristics and yard status characteristics, deeply digs and extracts the allocation rule of the ship unloading box areas, optimizes the allocation decision of the box areas and the box bits on the basis, intelligently adjusts the operation decision according to the operation status of the terminal, can provide reasonable and effective allocation decision of the ship unloading box areas and the box bits for a port manager in real time, is beneficial to relieving the traffic jam of the yard, improving the production operation efficiency of the container terminal and promoting intelligent construction of the container terminal.

Drawings

FIG. 1 is a schematic diagram of a dock floor plan;

FIG. 2 is a conceptual diagram of a near box area;

FIG. 3 is a schematic diagram of two-dimensional bin status-one-dimensional yard status transitions;

FIG. 4 is a graph of cluster center point to sample distance for different cluster numbers;

FIG. 5 is a representative status clustering result of the stocked volume of the yard; (a) - (h) respectively correspond to the typical yard state 1-the typical yard state 8;

FIG. 6 is a typical status line diagram of the piled up volume of clusters;

FIG. 7 is a plot selection count for different yard states; (a) a bin piled up volume status; (b) the operation number state of the box area; (c) the box stacking operation proportion in the box area;

FIG. 8 is a calculation of the importance of the tank allocation features of the ship unloading tank area;

FIG. 9 is a logic flow diagram of a heuristic intelligent algorithm for unloading tank bit allocation.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

The invention takes 2019 Spearkakorons harbor CICT dock container ship unloading area and box position allocation as an embodiment.

Step one: and collecting historical data of container terminal ship unloading operation and storage yard states.

The collected data comprises container terminal plane layout related data, ship unloading operation data and yard state data.

The dock floor layout related data includes the number of berths, the storage yard layout form, the number of boxes, and the like. In this case, there are 4 berths, the yard adopts a horizontal arrangement form, there are 38 heavy box areas, 3 empty box areas and 2 cold box areas, and the berths and the yard are arranged as shown in fig. 1.

The ship unloading operation data comprise ship unloading time, ship unloading berth, box dropping area and ship unloading box data. The ship unloading box data comprises an import and export type and a container owner. In this case, the ship unloading case includes import case, transfer case and three types of cabinet case that falls, and wherein, import case need carry the case by outer integrated circuit card, and cabinet case that falls only needs interior integrated circuit card to transport in the storage yard, and transfer case includes at present wharf shipment and at other wharf shipment two kinds of circumstances, and the former operation flow is similar with cabinet case that falls, and the latter operation flow is similar with import case. The method comprises the steps that a company to which a suitcase collecting card belongs is used as cargo owner information of a ship unloading box, for import boxes and transfer boxes loaded on other wharfs, the cargo owner is a code of the company to which the suitcase collecting card belongs, and for a reverse cabinet box and the transfer box loaded on the current wharf, the cargo owner is the current wharf (CICT).

The yard status data includes the amount of piled up, the number of jobs, the number of bridges and the proportion of piled up boxes in each time period in each heavy box area in the yard. The piled up quantity of a certain box area comprises the piling up quantity of each bin position and each stack of the box area, the operation quantity comprises the piling up quantity, the box taking operation quantity and the box inverting operation quantity, the field bridge quantity is the total number of field bridges with operation records in the box area in each time period, and the piling up operation proportion is the proportion of the piling up quantity of the box area to the total operation quantity.

Aiming at the collected container terminal ship unloading operation data and the yard state data, the association relation among the container attributes, the yard state and the box area selection is analyzed by utilizing a data mining method, and the ship unloading box dropping box area selection rule is extracted from massive historical data.

Step two: based on an unsupervised learning algorithm, a yard state is calculated according to historical data, and a clustering algorithm is utilized to extract a typical yard state.

In the first step, the state of the stock, the state of the number of operations, the state of the number of bridges and the state of the proportion of operations of the stacking box in each time period are selected to describe the state of the stacking field in the current time period. The adjacent 8 bins near bin i are defined as the adjacent bin j of bin i (j=1, 2, and 8), as shown in FIG. 2.

Taking the piled-up quantity state as an example, comparing the piled-up quantity of the current bin with the piled-up quantity of the adjacent bin in turn, and recording the relative relation as v _i,j J=1, 2, the contents of (8), if the current bin piled up amount is larger than the adjacent bin piled up amount, v _i,j =1, if smaller than the piled up amount in the adjacent bin, v _i,j = -1, if the same as the adjacent bin has been piled up, v _i,j =0, the piled-up quantity state of the bin

V _i ∈[-8,8]。V _i The larger indicates that the larger the stocked volume of the bin is in the adjacent bin; v (V) _i The smaller indicates that the amount of stocked in the bin is relatively smaller in the adjacent bin.

After the box state representing method is determined, the two-dimensional distribution box state is converted into one-dimensional storage field state, a storage field state value vector is obtained, and the conversion process is shown in fig. 3.

The method comprises the steps of taking an hour as a unit, calculating a storage quantity state value vector of a storage yard of 8760 hours in the whole year in 2019 according to the method, obtaining storage quantity state value vector broken lines of the storage quantity state value vector of each hour, constructing a storage quantity state value vector clustering model based on K-means by using an unsupervised learning algorithm in machine learning, clustering the storage quantity state value vector of the storage yard of each hour according to the similarity among samples, and firstly calculating the distances between cluster center points of different clustering cluster numbers and the samples, as shown in figure 4.

The distance between the center point of each cluster and the sample point gradually decreases as the number of clusters increases, the clustering accuracy gradually increases, and the clustering efficiency decreases as the number of clusters increases. The clustering efficiency and the clustering accuracy are comprehensively considered, the clustering cluster number is set to be 8, the state value vectors of the piled quantity of the piled fields are clustered, the cluster class of the state value vector of the piled quantity of the piled field in each hour is obtained, in each class of the state value vector clusters of the piled field, the average value of the state value vectors of the piled fields of each cluster is taken as the typical state of the piled field of the sample of the cluster, and the typical state of the piled quantity of the piled field of 8 classes is obtained through extraction altogether, as shown in figure 5.

The black broken lines in fig. 5 represent typical states of the stocked amounts of the yards of the clusters, and the average stocked amounts of the states are extracted for comparison, as shown in fig. 6.

As can be seen by comparing FIG. 6, the partial bins remain relatively high or relatively low in the amount of piled up throughout the year (e.g., 18, 33, 35 bins, etc.), while the amount of piled up partial bins differ significantly in different clusters (e.g., 5, 14, 21 bins, etc.).

Step three: and (5) analyzing the influence factors selected in the ship unloading area of the container.

As can be seen from FIG. 1, the yard is divided into 4 areas (01-04), wherein the 01, 03 and 04 stacking areas are respectively provided with 10 heavy box areas (rows A-K), and the 02 area is provided with 8 heavy box areas (rows A-H). Based on a statistical theory, the influence of container attribute characteristics (import and export types), ship unloading characteristics (berth numbers) and storage yard state characteristics (storage quantity type, operation number type, field bridge number type and stacking operation proportion type) on the distribution of the container areas is analyzed by sorting the container area selection conditions of various types of ship unloading containers in different areas and different rows in the historical data.

(1) Import and export type

As described in step one, the ship unloading box in this case includes a transfer box and a reverse box which need to be loaded at the dock, and an inlet box and a transfer box which need to be lifted, so according to the ship unloading box, the positions of the ship unloading box storage box areas which need to be loaded and lifted in the historical data (2019) are counted respectively, and the statistics result is shown in the following table:

TABLE 1 rank number of ship unloading and loading box storage box area where loading is required

Table 2 number of row where ship unloading and storage areas of the suitcase are located

As is clear from the statistics, the ship unloading boxes requiring loading are mostly piled up in the box area near the shore side, and the ship unloading boxes requiring the suitcase are mostly piled up in the box area near the land side. The results show that in the ship unloading box area allocation process of the dock of the case, the ship unloading box needing to be loaded is prone to be allocated to the box area close to the shore side, so that subsequent ship loading operation is facilitated, and the ship unloading box needing to be loaded is prone to be allocated to the box area close to the land side, so that subsequent box loading operation is facilitated.

(2) Berth numbering

The areas of the container stacking areas of the unloading ships at different berths in the historical data (2019) are respectively counted, and the statistical results are shown in the following table:

TABLE 3 areas where ship unloading and stockpiling areas of different berths are located

As is clear from the statistical data, since the number of the boxes (8) in the 02 area is less than that in the 01, 03 and 04 areas (10), the number of the ship unloading boxes in each berth is small in the 02 area, and the number of the ship unloading boxes in each berth is relatively large in the area close to the ship unloading berth. The results show that the dock in the case has a tendency to select a box area which is closer to the berth of the ship during the process of carrying out the box area allocation of the ship.

(3) Storage yard status

As described in step 1, the piled up quantity state, the job number state and the piled up box job proportion of each box area in each period are selected to describe the piling up field state, wherein the representation method of the piled up quantity state and the job number state is as described in step two. The number of ship unloading areas with different piled up quantity states, different job number states and different pile box job proportions are respectively counted in historical data (2019), and the result is shown in fig. 7.

According to the statistical result, about 74% of the box unloading tasks select box areas with relatively small stored quantity and operation number, and the stored quantity and operation number state values of most of the box unloading areas are-8, -6, -4, -2, 0 and 2; about 42% of the box unloading tasks select a box area with a box stacking operation proportion exceeding 90%, about 33% of the box unloading tasks select a box area with a box stacking operation proportion of 100%, and the selection times of the corresponding box areas gradually decrease along with the decrease of the box stacking operation proportion. The results show that in the case of the dock, in the case of the ship unloading box distribution process, box areas with relatively small piled up quantity, relatively small operation number and high box piling operation proportion tend to be selected.

Step four: and extracting a selection rule of a ship unloading box storage box region based on machine learning.

And (3) obtaining 8 types of stock quantity states, job number states, field bridge number states and box stacking proportion states in each hour by using the stock yard state clustering method in the step two, and mining the selection rule of the ship unloading box stock box region in the historical data based on the supervised learning concept in machine learning.

According to the ship unloading operation time, four types of yard states are matched for each ship unloading task in the historical data, when the state value vectors of the stock quantity, the operation number, the field bridge number and the box stacking operation proportion of each box area in the yard respectively belong to a corresponding certain type of typical state, box area allocation features under the state combination in the historical data are counted, wherein the box area allocation features comprise: a state value of the stock quantity, a state value of the operation number, a state value of the field bridge number and a state value of the operation proportion of the stack box in the allocated box area. And counting the number of the tank areas with different storage quantity, operation number, field bridge number and tank stacking operation proportion state values under each state combination, calculating the occurrence probability, and taking the tank area selection probability with each state value under each state combination as a ship unloading tank storage tank area selection rule. Taking a storage yard storage quantity state selection rule as an example, the statistical result is shown in the following table:

TABLE 4 partial store yard stock status selection rules

In table 4, a combination of a stock quantity state of typical state 1, a job number state of typical state 5, a field bridge number state of typical state 3, and a box proportion state of typical state 8 is denoted as a type 1 yard state, and when the yard state is type 1, a box selection probability of 25.00% at a box with a history data stock quantity state value of 1, a box selection probability of 20.83% at a stock quantity state value of 0, a box selection probability of 16.67% at a stock quantity state value of-4, a box selection probability of 10.42% at a stock quantity state value of-6 are respectively denoted as rule 1-1, rule 1-2, rule 1-3, and rule 1-4.

Step five: and (3) designing and verifying an intelligent algorithm for the ship unloading box area distribution based on rules.

In order to determine the influence degree of different influence factors on the allocation of the ship unloading box areas, a random forest algorithm with a disordered sequence is adopted, and the characteristic importance of import and export types, ship unloading berths and four storage yard states is calculated, and the result is shown in figure 8.

The feature importance calculation result shows that in the four storage yard state features, the storage quantity state importance is the operation number state, the field bridge number state is the box stacking operation proportion state. Therefore, in the intelligent algorithm for the ship unloading box area allocation based on the rules, four types of rule sets are sequentially called according to the sequence of C, J, Y, R, and the specific implementation method of the algorithm is as follows:

Storage yard state rule set { storage quantity state rule C, operation number state rule J, field bridge number state rule Y, stacking box operation proportion state rule R }, and alternative box region set B _feasible The minimum number M of the candidate boxes, and in the four types of rule sets, the corresponding types of rules are respectively ordered according to the sequence from high to low of the selection probability in C, J, Y, R;

secondly, sequentially calling rule sets corresponding to the corresponding states according to the sequence from high to low of the importance of the four storage yard states to generate a selectable box area set, wherein the calling sequence is C, J, Y, R;

according to the ship unloading operation sequence, for each ship unloading box i, firstly calculating the storage yard state s in the current storage yard time period _c Selecting the box with the highest probability state in C from high to low, and if the box of the type exists, putting into an alternative set B _c And entering a second layer of screening, if the box region of the type does not exist, continuing to select the box region with the next highest probability state in the C until the alternative set B _c Is not an empty set;

in the second layer screening, the yard state s is calculated _j Sequentially selecting the box regions with higher probability states in J according to the sequence of the selection probability from high to low until the candidate set B _c And B _j The intersection of (1) is not an empty set, and entering a third layer of screening;

in the third layer screening, the storage yard state s is calculated _y Sequentially selecting the box regions with higher probability states in Y according to the sequence of the selection probability from high to low until the candidate set B _c 、B _j And B _y The intersection of (1) is not an empty set, and entering a fourth layer of screening;

in the fourth layer screening, the yard state s is calculated _r Sequentially selecting the box regions with higher probability states in R according to the sequence of the selection probability from high to low until the candidate set B _c 、B _j 、B _y And B _r Is not an empty set and the alternative set B _c 、B _j 、B _y And B _r Adding the intersection of the alternative sets B _feasible ；

If the requirements are not met, starting from the fourth layer of screening, continuing to select the highest probability state box area update candidate set B in the residual rules _r Candidate set B _c 、B _j 、B _y And B _r Adding the intersection of the alternative sets B _feasible Stopping screening until meeting the requirement;

if the requirements are not satisfied yet, returning to the third layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B _y Repeating the fourth layer screening process until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

If the requirements are not satisfied, returning to the second layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B _j Repeating the third and fourth screening processes until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

If the requirements are not satisfied, returning to the first layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B _c Repeating the second, third and fourth screening processes until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

The pseudo code of the ship unloading box area distribution intelligent algorithm based on the rule is as follows:

/>

screening by using a box region selection rule to obtain a ship unloading box alternative box dropping box region set B _feasible After that, the processing unit is configured to,the final box falling areas are distributed by utilizing different heuristic algorithms, including a random principle, a principle of closest berthing to unloading and a principle of minimum owner/stock quantity, and the method is as follows:

(1) The random principle: randomly selecting a box region with the rest box positions from the alternative box region collection as a final box region;

(2) The principle of approximation: selecting a box region which is closest to the berth of the ship unloading and has residual box positions from the alternative box region set to be used as a final box region;

(3) Idle principle: and selecting the box area with the least piled owners, the least piled quantity and the rest box positions from the alternative box area collection as the final box area, wherein the box area piled with the container with the owners is preferentially selected.

Finally, verifying the box area allocation algorithm. In the stacking process, the larger the workload in a single box area is, the larger the operation pressure of a field bridge is, so that the longer the time for the collector card to wait for the field bridge to operate is; meanwhile, in the future suitcase process, the more the number of owners in a single suitcase area is, the more the number of the suitcases is, and the higher the probability of generating traffic jam is. Therefore, the adopted evaluation indexes comprise the maximum value, the minimum value, the average value and the variance of the workload and the owner number of each bin area in the storage yard, and the comparison of the evaluation indexes under different bin area allocation principles is shown in the following table:

TABLE 5 comparison of the case Allocation Effect under different principles

The comparison result of table 5 shows that under the principle of randomly selecting the box areas, various indexes are relatively close to the historical data, which shows that the box area collection obtained by the box area generation algorithm for the ship unloading box alternative box falling is basically consistent with the box area distribution preference in the actual operation process of the wharf of the case, and the effectiveness of the box area selection rule for the ship unloading box falling based on the storage yard state in the fourth step is verified. Among the three box section selection principles, the average cargo owner number of each box section under the principle of closest berthing distance from the ship unloading is the least, and the maximum workload of a single box section and the variance of the workload of each box section under the principle of the least cargo owner number/stocking quantity are smaller, both principles are superior to historical data.

Step six: and (5) designing and verifying a heuristic intelligent algorithm for distributing ship unloading boxes.

And D, according to comparison results of different box region selection principles in the fifth step, selecting the optimal principle to determine the final box region. After the box falling area of the ship unloading box is determined, box position distribution is carried out according to the following principle:

(1) The containers of the same cargo owner are stacked as much as possible with the same or adjacent shellfish positions, and the containers of different cargo owners are stacked as much as possible in a scattered manner;

Based on the principle, on the premise of knowing the box falling area, a ship unloading box position allocation heuristic algorithm is designed, and the main logic flow of the algorithm is as follows:

(1) After the box falling area of any ship unloading box is determined, firstly judging whether the current box area has containers with the same cargo owner; if not, further judging whether the current box area has the bin positions which are not occupied by other cargo owners, if so, selecting the bin position which is farthest from the bin positions occupied by other cargo owners from the feasible bin positions, and selecting a stack with the minimum stacking quantity for stacking; if all the shellfish positions are occupied by other owners, selecting the shellfish position with the lowest owner number from the feasible shellfish positions, and selecting the stack with the largest stacking quantity for stacking;

(2) If the current container area has containers with the same cargo owner, judging whether the container is occupied by the current cargo owner container or not, if so, further judging whether the container is free, and if so, selecting a stack with the least residual container and the free container for stacking;

(3) If the bin space occupied by the cargo owner only has no spare bin space, further judging the cargo owner occupation situation in other bin space, if the cargo owner has other bin space not occupied by other cargo owners, selecting the bin space closest to the bin space occupied by the cargo owner, and selecting a stack with the least number of piled cargo owners and the least number of remaining bin spaces for piling;

(4) If all the shellfish positions are occupied by other cargoes, further judging whether the shellfish positions of the current cargo main container are piled up, and if so, selecting a stack with the least cargo main number and the least residual container positions from the shellfish positions of the current cargo main container to be piled up; if the other bin positions do not have the bin positions of the current cargo main container, selecting the bin position closest to the cargo main bin position from all possible bin positions, and selecting a stack with the minimum cargo main number and the minimum residual bin positions for stacking;

(5) If the current container area has only the bin positions occupied by the current cargo owner and other cargo owners at the same time or the bin positions occupied by other cargo owners at the same time, further judging whether the bin positions of the cargo owner container have spare bin positions, if so, selecting the bin position with the least cargo owner number from all possible bin positions of the cargo owner container, and selecting a stack with the least cargo owner number and the least remaining bin positions for stacking; if only the available bin positions which do not contain the cargo owner container exist, selecting the bin position closest to the cargo owner bin position, and selecting a stack with the minimum cargo owner number and the minimum residual bin position for stacking.

The algorithm logic flow diagram 9 is shown.

And after the box bit is determined, verifying the box bit distribution result. When the boxes are taken in the box area, if the containers are stored together with the cargo owner, the movement of the field bridge in the box area is reduced, and the box taking operation efficiency of the field bridge is improved; if the containers are stacked in a dispersed manner with the cargo owner, empty vehicle movement of a field bridge along the shellfish position direction in the process of carrying the containers can be increased, the operation efficiency is reduced, the queuing time of taking the containers and collecting the cards is further prolonged, and traffic jam between the containers is caused. Considering the distribution condition of the same-cargo main containers in the container area, adopting the stacking density of the same-cargo main containers as the container position distribution evaluation index, and the calculation method comprises the following steps:

wherein n is _i For the bin number that the owner i occupies in the current bin,f _i the forefront bin number, l, occupied by the owner i in the current bin _i The rearmost bin number occupied by the shipper i in the current bin is numbered.

The range of the stacking density of the container of the same cargo owner is (0, 1), the closer the index is to 0, the more scattered the stacking shellfish of the container of the same cargo owner is, the closer the index is to 1, the more dense the stacking shellfish of the container of the same cargo owner is, the more dense the stacking shellfish of the container is, the comparison between the container allocation heuristic algorithm and the stacking density of the container of the same cargo owner in each container area in the historical data is as shown in the following table:

Table 6 heuristic effect comparison for bin allocation

As can be seen from comparison of Table 6, in the history operation process of the wharf of the present embodiment, the stacking density of containers with the same cargo owner reaches 23 at most, which indicates that more different cargo owner containers exist in the container area; in the calculation result of the box position distribution heuristic algorithm of the research design, the container stacking density of all the box areas and all the owners is 1, which indicates that all the containers are stacked intensively according to the owners and the mixed stacking condition does not occur. The result shows that the box position distribution heuristic algorithm designed by the research can effectively avoid mixing and piling of different cargo containers, and is beneficial to improving the field bridge operation efficiency when taking the boxes.

Claims

1. The intelligent algorithm for distributing the ship unloading container areas and the container positions based on the machine learning is characterized by comprising the following steps:

the corresponding state of each box area is the relative size of each state parameter of the current box area in 8 adjacent box areas; the state parameter is the piled quantity/operation number/bridge number/piling box operation proportion;

the calculation method comprises the following steps: comparing the state parameters of any box area with the adjacent box areas in pairs in sequence, and when the parameter value of the box area is larger than that of a certain adjacent box area, adding a corresponding state value of the box area to +1; when the box parameter value is smaller than a certain adjacent box, the corresponding state value of the box is-1; when the box parameter value is equal to a certain adjacent box, keeping the corresponding state value unchanged; after the comparison with all 8 adjacent box areas is finished, the corresponding state value of the accumulated box areas is obtained, and the value range is [ -8,8];

constructing a K-means-based storage yard state value vector clustering model, setting cluster numbers, and clustering four storage yard states of the stored quantity of each box area, the operation number of each box area, the field bridge number of each box area and the stacking operation proportion of each box area respectively; in each type of storage yard state, taking the average value of the storage yard state value vectors of each cluster as the typical storage yard state of the cluster sample, wherein the number of the typical storage yard states is equal to the number of the clusters;

Thirdly, analyzing influence factors selected in a container unloading area;

the container ship unloading area selection influencing factors comprise container attribute characteristics, ship unloading characteristics and yard state characteristics; the container attribute features include container import and export types; the ship unloading characteristics comprise a ship unloading berth number; the storage yard status features comprise a stored quantity type, an operation number type, a field bridge number type and a stacking box operation proportion type of each box area;

(1) Calculating the importance of the box position distribution characteristics of the ship unloading box regions by adopting a random forest algorithm with a disordered sequence, wherein the box position distribution characteristics of the ship unloading box regions comprise an inlet and outlet type, a ship unloading berth, a stockyard storage quantity state, an operation number state, a yard bridge number state and a box stacking operation proportion state of each box region;

the ship unloading box area distribution intelligent algorithm based on the rule comprises the following specific steps:

Storage yard state rule set { storage quantity state rule C, operation number state }Rule J, field bridge number state rule Y, heap operation proportion state rule R, alternative box region set B _feasible The minimum number M of the candidate box regions, and in four types of rule sets C, J, Y, R, the corresponding types of rules are respectively ordered according to the sequence from high to low of the box region selection probability;

According to the ship unloading operation sequence, dividing each ship unloading box i into four layers of screening according to a rule set, and screening in a layered positive sequence, and sequentially selecting box areas with highest probability in various storage yard states into different alternative sets; the intersection of the four different obtained candidate sets is added to the candidate box set B _feasible ；

Judgment B _feasible Whether the minimum number M of the candidate boxes is satisfied, and when the minimum number M of the candidate boxes is satisfied, outputting a candidate set B _feasible The method comprises the steps of carrying out a first treatment on the surface of the When the requirements are not met, starting screening from the fourth layer, continuing to select the highest probability state box area in the residual rules to add the candidate set B _feasible Stopping screening until meeting the requirement; when the requirements are not satisfied yet, performing reverse order screening to obtain the highest probability state box region in the residual rule, adding the highest probability state box region into the alternative set of the corresponding layer, and starting positive order screening from the layer to obtain B _feasible Up to the alternative set B _feasible Stopping screening after meeting the requirements;

(2) Heuristic contrast rules are used to determine the set B of candidate bins _feasible Determining a final box falling area, wherein the final box falling area comprises a random principle, a nearby principle and an idle principle; wherein the random principle is at B _feasible The final box falling area is selected randomly, and the principle nearby is that B is _feasible Selecting the nearest box area to the unloading berth of the container in the current task, wherein the idle principle is that in B _feasible Selecting a box area with the least number of owners or piled quantity;

(3) The algorithm verifies the tank area allocation effect of the intelligent tank area allocation algorithm for verifying the tank area allocation of the ship unloading based on the rule, and the adopted evaluation indexes comprise the maximum value, the minimum value, the average value and the variance of the operation amount and the cargo owner number of each tank area in the storage yard;

the ship unloading box position allocation heuristic intelligent algorithm follows the box position allocation principle which comprises the following steps: (1) The container stacks of the same cargo owners have the same or adjacent shellfish positions, and the containers of different cargo owners are stacked in a scattered way; (2) When selecting a box bit from the shellfish bits, selecting stack bits with less residual box bits; (3) When different cargo owner containers are piled up in the same bin, different stacking positions are selected to respectively pile up the cargo owner containers;

2.1 If there is any bin occupied by the current cargo owner container, further judging whether there is any empty bin; 2.1.1 If the spare box is available, selecting a stack with the least remaining box and the spare box for piling up;

2.1.2 If there is no empty bin, further judging whether there is a situation that the bin is not occupied by other cargo owners;

the verification of the ship unloading box position distribution heuristic algorithm adopts the stacking density of the containers of the same cargo owner as the box position distribution effect evaluation index, and the calculation method is shown in the following formula;

wherein n is _i The bin number f occupied by the owner i in the current bin _i The forefront bin number, l, occupied by the owner i in the current bin _i Rearmost bin occupied by owner i in the current binNumbering;

2. The intelligent algorithm for distributing the ship unloading container boxes and the boxes based on the machine learning according to the claim 1, wherein the four layers of screening are classified according to a rule set;

firstly, calculating a storage yard state s in a current storage yard time period ₁ Selecting RULE in the order of the box selection probability from high to low ₁ The bin having the highest probability state; when there is a box of this type, put into an alternative set B ₁ And entering a second layer of screening; when there is no box area of this type, continuing to select RULE ₁ Bin with next highest probability state in (B) until alternative set B ₁ Is not an empty set;

in the second layer screening, the yard state s is calculated ₂ Sequentially selecting RULE according to the order of the selection probability from high to low ₂ The bin with the highest probability state up to alternative set B ₁ And B ₂ The intersection of (1) is not an empty set, and entering a third layer of screening;

in the fourth layer screening, the yard state s is calculated ₄ Sequentially selecting RULE according to the order of the selection probability from high to low ₄ Bin with higher probability state up to alternative set B ₁ 、B ₂ 、B ₃ And B ₄ Is not an empty set and the alternative set B ₁ 、B ₂ 、B ₃ And B ₄ Adding the intersection of the alternative sets B _feasible 。

3. Machine learning based intelligent algorithm for allocation of containers to areas and locations of ship unloading containers according to claim 1 or 2, characterized in that said B _feasible When the requirement of the minimum number M of the alternative box areas is not met, the specific steps are as follows;

when B is _feasible When the requirement of the minimum number M of the candidate box areas is not met, starting from the fourth layer of screening, continuing to select the box area with the highest probability state in the remaining rules to add into the candidate set B _feasible Stopping screening until meeting the requirement;

if the requirements are not satisfied, returning to the third layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B ₃ Repeating the fourth layer screening process until the requirements are met, stopping screening, and outputting an alternative set B _feasible ；

If the requirements are not satisfied, returning to the first layer screening, continuously selecting the highest probability state box area in the residual rules, and updating the alternative set B ₁ Repeating the second, third and fourth screening processes until the requirements are met, stopping screening, and outputting an alternative set B _feasible 。

4. The intelligent algorithm for distributing ship unloading container boxes and boxes based on machine learning according to claim 1, wherein the clustering number is set to 8.