A container storage location selection method based on storage behavior characteristics

Abstract

The present invention belongs to the technical field of container stacking, and specifically relates to a container storage location selection method based on stacking behavior characteristics, comprising the following steps: step S100: establishing a PSCW++ stacking rule; step S200: establishing a stacking time predictor model; step S300: calculating and obtaining the predicted stacking time of containers to be stacked; step S400: determining a stacking area for the containers to be stacked; step S500: determining a stacking bay position of the containers to be stacked in the stacking area; step S600: determining a location of the containers to be stacked in the stacking bay position; the present invention provides a container storage location selection method based on stacking behavior characteristics, which aims at the problem that the existing container storage location selection technology does not utilize historical data of port operations, predicts the storage time of containers through a neural network, combines a storage location selection method based on rules, and uses the predicted storage time for container area selection and location selection, which can effectively improve the operating efficiency and reduce the container turnover rate.

Description

Technical Field

[0001] The present invention belongs to the technical field of container stacking, and in particular relates to a container storage location selection method based on stacking behavior characteristics. Background Art

[0002] Container storage location selection is to allocate yard storage locations for containers arriving at the port, commonly known as yard planning. The storage location includes container area, bay position and stack layer information. The basis for allocation includes container attributes and container owner attributes. The purpose of allocation is to improve the transfer efficiency of containers and reduce the waiting time of operating container trucks.

[0003] Container storage location selection is the core link in container terminal operations, affecting operating efficiency and operating costs. Its significance can be summarized into the following four aspects. First, through reasonable planning and stacking, the vertical and horizontal space of the yard can be maximized, which plays a role in space optimization and meets the growing throughput demand of container terminals; second, reasonable slot selection can reduce the moving distance of mechanical operating equipment and reduce operating time, thereby improving the operating efficiency of the yard; third, scientific slot selection can optimize the bay layout, reduce the container turnover rate, and reduce operating costs; fourth, by optimizing slot selection, it can ensure the rapid circulation and accurate extraction of containers in the yard, thereby meeting customers' requirements for container processing speed and service quality and improving customer satisfaction.

[0004] Existing container storage location selection technologies include mathematical programming methods, rule-based methods, heuristic search methods, and machine learning methods:

[0005] ① Mathematical programming method is to transform the actual problem into a mathematical model, and use mathematical methods to solve this model to obtain the optimal solution or approximate optimal solution. Mathematical programming is a mathematical modeling method for optimization problems, which involves basic elements such as decision variables, objective functions and constraints. For container storage location selection problems, the objective function is usually to minimize the turnover rate, or minimize the operation time, or minimize the operation cost, or a combination of several. Typical representatives of this type of method are mixed integer programming models and constraint programming models. When the problem scale is small, it is easy to get the optimal solution. Common solution algorithms include simplex method, branch and bound method and iterative calculation method;

[0006] ② The rule-based method is based on the principle that for any container, the process of slot allocation can be regarded as a collection of many judgment activities based on the knowledge and rules of slot allocation. After each judgment or a group of judgments, the range of available slots is reduced accordingly, and the final selected container position is allocated to the container as the optimal position.

[0007] ③ Heuristic search method is a strategy that approaches the optimal solution one by one. It is used to effectively arrange the storage location of containers in a limited storage space. Common heuristic search methods include genetic algorithm, particle swarm algorithm and cargo simulation algorithm. They find the optimal solution by simulating natural processes or human behavior, have global search capabilities and adaptability, and are widely used in solving combinatorial optimization problems.

[0008] ④ The representative of machine learning methods is reinforcement learning, which is a trial-and-error method. It is a method that enables an intelligent agent to optimize its behavior strategy through trial and error in the process of interaction with the environment. It does not require any data in advance, but obtains learning information and updates model parameters by receiving rewards or feedback from the environment for actions. Reinforcement learning has shown good performance in solving traveling salesman problems and scheduling optimization problems, especially when the problem scale is large. Although existing reinforcement learning methods can be used to solve combinatorial optimization problems, they are limited by the effective modeling of the relationship between container stacks. No reinforcement learning algorithm can be effectively applied to the container storage location selection problem.

[0009] The existing container storage location selection methods have their own characteristics. Mathematical programming methods can obtain the optimal solution in theory, but it is difficult to obtain satisfactory location selection results within a limited time. Rule-based methods have the advantage of low computational cost, but the quality of the location selection results obtained is limited and not optimized. Heuristic search methods have a certain ability to solve NP-hard problems, but are slower than rule-based methods, and their solution performance decreases as the problem size increases. Reinforcement learning methods have more advantages in solving medium and large-scale problems, but they are essentially a type of machine learning method with poor interpretability and it is very difficult to model the real environment.

[0010] At the same time, none of the above container storage location selection methods considers the use of historical data of container ports. The historical data of container storage location selection records the spatial position, movement trajectory, storage time and other information of containers in the port yard. By analyzing it, port operators can better understand and optimize the storage and transportation process of containers. By mining the historical data of ports, the storage time can be predicted; by using the storage time, containers with longer storage time can be stored in the container area far from the berth, while containers with shorter storage time can be stored in the container area close to the berth, so as to achieve efficient use of the near-berth container area; by using the storage time, containers with shorter storage time can be stacked on top of containers with longer storage time, so as to reduce the operation of turning over containers. Summary of the invention

[0011] The purpose of the present invention is to provide a container storage location selection method based on storage behavior characteristics, so as to solve the problems existing in the background technology.

[0012] In order to achieve the above technical objectives, the technical solution adopted by the present invention is as follows:

[0013] A container storage location selection method based on storage behavior characteristics comprises the following steps:

[0014] Step S100: Based on the existing PSCW stockpiling rule, a PSCW+ stockpiling rule is formed, and based on the PSCW+ stockpiling rule, a new PSCW++ stockpiling rule is established;

[0015] Step S200: Establishing a storage time predictor model, and using historical data to train the time predictor model;

[0016] Step S300: According to the trained time predictor model, the parameters of the container to be stacked are input to calculate the predicted storage time dt of the container to be stacked;

[0017] Step S400: determining a storage area for the containers to be stacked based on the calculated predicted storage time dt of the containers to be stacked and taking the established PSCW++ storage rules as a reference;

[0018] Step S500: Determine the stacking position of the container to be stacked in the storage area based on the established PSCW++ stacking rule;

[0019] Step S600: According to the calculated predicted stacking time dt of the container to be stacked, the location of the container to be stacked in the stacking bay is determined based on the established PSCW++ stacking rule.

[0020] In step S100, on the basis of the existing PSCW stacking rules, the following stacking rules are added for inland container terminals to form PSCW+ stacking rules, which are as follows:

[0021] Rule A1: Define the export miscellaneous pile area, which is for whole shells and cannot be used to pile import boxes;

[0022] Rule A2: Define the overload container stacking area, which is one or more whole bays, and these bays are associated with the specified yard bridge;

[0023] Rule A3: Different types of containers of the same size cannot be stacked in the same stack. General Purpose Containers (GP containers) and High Cube Containers (HQ containers) cannot be stacked in the same stack.

[0024] Rule A4: Reserve a container transfer space in the import or export mixed pile area, and reserve a container space corresponding to the container area floor height minus 1;

[0025] Rule A5: When containers from multiple owners enter the same bay, containers of the same owner, size, and type shall be stored in a single stack to avoid mixing and causing container overturning;

[0026] Rule A6: The height difference between adjacent stacks should not exceed 3 container heights to avoid isolated stacks.

[0027] On the basis of the formed PSCW+ stockpiling rules, the following stockpiling rules are added to form the PSCW++ stockpiling rules, as follows:

[0028] Rule C1: Containers that have been stored for a long time should be stored in a container area far from the berth, and containers that have been stored for a short time should be stored in a container area close to the berth;

[0029] Rule C2: After the container area is selected, containers with a shorter storage time should be placed on top of containers with a longer storage time, and vice versa.

[0030] The step S200 specifically includes the following sub-steps:

[0031] Step S201: Establish a storage time predictor model, which is specifically as follows:

[0032] Y = f(s);

[0033] s=∑ j W ij *x j ;

[0034] in,

[0035] Y represents the output, corresponding to the storage time of the container;

[0036] W ij Represents the weight parameter from network node j to input node i;

[0037] x j represents the jth input node, all x j The set of represents the input vector;

[0038] f(s) represents the activation function, where It is an S-shaped curve function with an output range between (0,1), which is continuous and differentiable;

[0039] Step S202: Acquire a historical stockpile data set from the TOS system (terminal production operation information system), and divide the historical stockpile data set into a training data set and a test data set in proportion;

[0040] In the stockpiling data set, the data items of the input vector x include: container attributes (including information such as size, status, type, items, arrival time, departure time), container stakeholder attributes (including information such as container holder, carrier ship, consignee, consignor, loading port, unloading port) and container terminal policy attributes (including information such as customs inspection time). The output Y of the model is the storage time of the corresponding container.

[0041] Step S203: training the storage time predictor model, using the training data set in step S202, and adopting the error back propagation (BP) algorithm for training, wherein the learning rate η is set to 0.01;

[0042] Step S204: After the training is completed, the weight parameter W of the storage time predictor model is obtained. ij The model parameter set θ is substituted into the storage time predictor model, and the trained storage time predictor model is verified using the test data set. The weight parameter W is modified. ij The set θ of θ is used to obtain the trained storage time predictor model.

[0043] The step S400 specifically includes the following sub-steps:

[0044] Step S401: Obtain the following parameters from the TOS system, including:

[0045] ①Minimum storage time dt min and the maximum storage time dt max ;

[0046] ②Container area collection B available for container storage k , k=1,2,…,K;

[0047] ③ Available box area B k Distance to berth d k , k=1,2,…,K;

[0048] Step S402: Based on the distance d from the container area to the berth k , from small to large for box area B k Sort by distance d k A set of bins B sorted by size l , l=1,2,…,K;

[0049] Step S403: Based on the calculated predicted storage time dt of the container to be stacked, according to the rule C1 in the established PSCW++ storage rule, the distance d k Small container area, used to store containers with short storage time dt; distance d kThe far container area is used to store containers with a long storage time dt. The container area B to be stacked is determined according to dt. l .

[0050] In step S500, "determining the stacking position of the container to be stacked in the stacking area based on the established PSCW++ stacking rule" is specifically:

[0051] Get the determined container area B to be stacked l The number of bays B, calculate each bay b in turn i ,i∈{1,2,3,...,B}, the number of available slots sn(b i ), after the calculation is completed, sn(b i ) value is the largest and each container to be stacked is allocated to the position.

[0052] The step S600 specifically includes the following sub-steps:

[0053] Step S601: Find available stacks in the allocated bays and construct an available stack set S1;

[0054] Step S602: searching for a matching stack that satisfies rule C2 in the available stack set S1, and constructing a matching stack set S2;

[0055] Step S603: Make the following determination on the matching stack set S2, specifically:

[0056] ① If the matching stack set S2 is not an empty set, then a stack s in the matching stack set S2 i The top bin position of is allocated to the container to be stacked. If there may be multiple stacks that meet rule C2, the stack s with the smallest stack number i is selected. i Assigned to containers to be stacked;

[0057] ② If the matching stack set S2 is an empty set, a stack s in the available stack set S1 will be j The top box slot of is allocated to the container to be stacked. If there may be multiple available stacks in the set S1, the stack s with the smallest stack number j is selected. j Assigned to containers to be stacked.

[0058] In the step S602, "searching for a matching stack that satisfies the rule C2 in the available stack set S1, and constructing a matching stack set S2" is specifically as follows:

[0059] Determine the stacks s in the available stack set S1 one by one j Storage time of the top container dt(s j ), if dt≤dt(s j ), then stack s j To satisfy the stack of rule C2,j Add to the matching stack set S2 that satisfies rule C2; ​​otherwise, the loop control variable is incremented by 1 and the next judgment is performed;

[0060] Among them, dt is the predicted storage time of the container to be stacked, and |S1| represents the number of elements in the set S1.

[0061] The present invention provides a container storage location selection method based on storage behavior characteristics. In order to solve the problem that the existing container storage location selection technology does not utilize the historical data of port operations, a neural network is used to predict the storage time of containers. In combination with a rule-based storage location selection method, the predicted storage time is used for container area selection and container location selection, which can effectively improve the operating efficiency and reduce the container turnover rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention is further illustrated by means of the non-limiting examples given in the accompanying drawings.

[0063] Figure 1 It is a schematic diagram of the main process of the present invention;

[0064] Figure 2 A schematic diagram of establishing a PSCW++ stockpiling rule for the present invention;

[0065] Figure 3 It is the basic framework of the container storage location selection method based on storage behavior characteristics in the present invention;

[0066] Figure 4 This is an example diagram of container priority of the present invention;

[0067] Figure 5 is a flow chart of step S400 of the present invention;

[0068] Figure 6 is a flow chart of step S500 of the present invention;

[0069] Figure 7 is a flow chart of step S600 of the present invention;

[0070] Figure 8 is a flow chart of step S602 of the present invention;

[0071] Fig. 9 This is an example diagram of a bay position structure; DETAILED DESCRIPTION

[0072] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0073] First, the existing PSCW storage rules are as follows:

[0074] Rule B1: P (port of destination), that is, containers of the same port of destination are placed in the same area of ​​the yard (one bay or multiple bays), and containers of different ports of destination cannot be mixed together (i.e. the same bay);

[0075] Rule B2: S (size), that is, containers of the same destination port are stacked in different bays according to different sizes, and containers of different sizes cannot be stacked in the same bay;

[0076] Rule B3: C (category), that is, containers of the same category should be stacked together (one container area or multiple container areas), and containers of different categories should not be stacked together (i.e. in the same container area). Categories refer to ordinary containers, refrigerated containers, dangerous goods containers, etc.

[0077] Rule B4: W (weight), that is, containers of the same weight level are placed together (i.e. in the same stack), and containers of different weight levels are stacked separately (i.e. in different stacks).

[0078] Embodiment 1:

[0079] like Figures 1 to 8 A container storage location selection method based on storage behavior characteristics is shown, characterized in that it includes the following steps:

[0080] Step S100: Based on the existing PSCW stockpiling rules, a PSCW+ stockpiling rule is formed, and based on the PSCW+ stockpiling rule, a new PSCW++ stockpiling rule is established, as follows:

[0081] On the basis of the existing PSCW stacking rules, the following stacking rules are added for inland container terminals to form PSCW+ stacking rules, as follows:

[0082] Rule A1: Define the export miscellaneous pile area, which is for whole shells and cannot be used to pile import boxes;

[0083] Rule A2: Define the overload container stacking area, which is one or more whole bays, and these bays are associated with the specified yard bridge;

[0084] Rule A3: Different types of containers of the same size cannot be stacked in the same stack. General Purpose Containers (GP containers) and High Cube Containers (HQ containers) cannot be stacked in the same stack.

[0085] Rule A4: Reserve a container transfer space in the import or export mixed pile area, and reserve a container space corresponding to the container area floor height minus 1;

[0086] Rule A5: When containers from multiple owners enter the same bay, containers of the same owner, size, and type shall be stored in a single stack to avoid mixing and causing container overturning;

[0087] Rule A6: The height difference between adjacent stacks should not exceed 3 container heights to avoid isolated stacks;

[0088] On the basis of the formed PSCW+ stockpiling rules, the following stockpiling rules are added to form the PSCW++ stockpiling rules, as follows:

[0089] Rule C1: Containers that have been stored for a long time should be stored in a container area far from the berth, and containers that have been stored for a short time should be stored in a container area close to the berth;

[0090] The purpose of rule C1 is to improve operation efficiency. Reduced Instruction Set Computing (RISC) technology in computer architecture shows that high-frequency instructions are executed by faster hardware, thereby improving the overall performance of the processor. For the loading and unloading operations of container terminals, since the distance between the inner container truck and the quay crane is short, the inner container truck takes less time to travel and completes a loading and unloading instruction, which is equivalent to executing high-frequency instructions in RISC. Therefore, the shorter the storage time of the container, the more it should be stored in the container area close to the berth; conversely, the longer the storage time of the container, the more it should be stored in the container area far away from the berth. According to the idea of ​​RISC, rule C1 helps to reduce the time to complete all loading and unloading instructions, thereby improving port operation efficiency;

[0091] Rule C2: After the container area is selected, containers with a shorter storage time should be placed on top of containers with a longer storage time, and vice versa;

[0092] The purpose of rule C2 is to reduce the rate of box turnover. Figure 4 As shown, assuming that the current stack s k The top container has a priority of pr i , if the priority of the container to be stacked is pr j , then rule C2 makes: pr i ≥pr j, where i,j∈{1,2,…,N} represents the container number arriving at the bay, N represents the total number of containers at the bay, and the container priority represents the order of picking up the container. The smaller the value, the earlier the container is picked up. According to Unordered Stacking (US) (Reference: [XinJin, 2023] Container Stacking Optimization Based on Deep Reinforment Learning [J]. Engineering Applications of Artificial Intelligence. 123 (2023): 106508) and Badly Placed Item (BI) (Reference: ① [Kim, 2006] A Heuristic Rule for Relocating Blocks [J]. Computers & Operations Research. 2006, 33 (4): 940–954, ② [Forster, 2012] A Tree Search Procedure for the Container Relocation Problem [J]. Computers & Operations Research,2012,39(2):299–309, ③[Sven Boge,2020]TheParallel Stack Loading Problem Minimizing the Number of Reshuffles in theRetrieval Stage[J].European Journal of Operational Research.280(2020):940–952), satisfying the definition of pr i ≥pr j The stacking process that meets the C2 rule will not cause the increase of the US value and BI value of the bay, while US and BI are two important indicators reflecting the container turnover rate. Therefore, the stacking process that meets the C2 rule will not lead to an increase in the container turnover rate of the bay even though the number of containers in the bay is increasing.

[0093] Specifically, such as Figure 3As shown, container storage location selection includes container area selection and stack layer selection within the container area. Container area selection is also called the big plan of yard planning, which is mainly implemented according to the PSCW principle. In this embodiment, a corresponding container area selection rule C1 is designed according to the storage time to achieve efficient utilization of the near-berth container area; bay position selection is also called the medium plan of yard planning, which mainly matches the stacking operation mode, equipment operation rules and yard status, and excludes bays that do not meet the stacking operation mode and equipment cannot operate. In this embodiment, bay position selection is based on PSCW+; stack layer selection is also called the small plan of yard planning, which is the process of selecting the best container position from the selected bays, mainly considering the yard crane operation sequence and the container placement requirements. The present invention designs a container position selection rule C2 based on the storage time, that is, containers with shorter storage time can be stacked on containers with longer storage time, so as to control the number of container turnovers.

[0094] Step S200: Establishing a storage time predictor model and using historical data to train the time predictor model, specifically including the following sub-steps:

[0095] Step S201: Establish a storage time predictor model, which is specifically as follows:

[0096] Y = f(s);

[0097] s=∑ j W ij *x j ;

[0098] in,

[0099] Y represents the output, corresponding to the storage time of the container;

[0100] W ij Represents the weight parameter from network node j to input node i;

[0101] x j represents the jth input node, all x j The set of represents the input vector;

[0102] f(s) represents the activation function, where It is an S-shaped curve function with an output range between (0,1), which is continuous and differentiable;

[0103] In this step S201, the storage time predictor model is implemented based on a neural network (NN), and one input layer, one hidden layer and one output layer are selected. The basic principle is as follows:

[0104] The input data first enters the input layer and then is passed to the hidden layer. In the hidden layer, each neuron calculates its input (the weighted sum of neurons from the previous layer) and produces an output through the activation function. This output is then passed to the next layer. Finally, the neurons in the output layer calculate their input and produce an output value through the activation function. This value is the prediction result of the neural network.

[0105] Step S202: Obtain a historical stockpile data set from the TOS system (terminal production operation information system), and divide the historical stockpile data set into a training data set and a test data set in proportion. For example, 100,000 data in a certain time interval can be selected for model training and performance testing, of which 80,000 are used as training data sets and 20,000 are used as test data sets.

[0106] In the stockpiling data set, the data items of the input vector x include: container attributes (including information such as size, status, type, items, arrival time, departure time), container stakeholder attributes (including information such as container holder, carrier ship, consignee, consignor, loading port, unloading port) and container terminal policy attributes (including information such as customs inspection time). The output Y of the model is the storage time of the corresponding container.

[0107] Step S203: training the storage time predictor model, using the training data set in step S202, and adopting the error back propagation (BP) algorithm for training, wherein the learning rate η is set to 0.01;

[0108] Step S204: After the training is completed, the weight parameter W of the storage time predictor model is obtained. ij The model parameter set θ is substituted into the storage time predictor model, and the trained storage time predictor model is verified using the test data set. The weight parameter W is modified. ij The set θ of θ is used to obtain the trained storage time predictor model.

[0109] Step S300: According to the trained time predictor model, the parameters of the container to be stacked are input to calculate the predicted storage time dt of the container to be stacked;

[0110] Step S400: According to the calculated predicted storage time dt of the container to be stacked, the storage area of ​​the container to be stacked is determined based on the established PSCW++ storage rule, such as Figure 5 As shown, it specifically includes the following sub-steps:

[0111] Step S401: Obtain the following parameters from the TOS system, including:

[0112] ①Minimum storage time dt minand the maximum storage time dt max ;

[0113] ②Container area collection B available for container storage k , k=1,2,…,K;

[0114] ③ Available box area B k Distance to berth d k , k=1,2,…,K;

[0115] Step S402: Based on the distance d from the container area to the berth k , from small to large for box area B k Sort by distance d k A set of bins B sorted by size l , l=1,2,…,K;

[0116] Step S403: Based on the calculated predicted storage time dt of the container to be stacked, according to the rule C1 in the established PSCW++ storage rule, the distance d k Small container area, used to store containers with short storage time dt; distance d k The far container area is used to store containers with a long storage time dt. The container area B to be stacked is determined according to dt. l .

[0117] Step S500: Based on the established PSCW++ stacking rules, determine the stacking position of the container to be stacked in the stacking area, specifically, Figure 6 The process is shown as follows:

[0118] Get the determined container area B to be stacked l The number of bays B, calculate each bay b in turn i ,i∈{1,2,3,...,B}, the number of available slots sn(b i ), after the calculation is completed, sn(b i ) value is the largest and each container to be stacked is allocated to the position.

[0119] Step S600: According to the calculated predicted storage time dt of the container to be stacked, the location of the container to be stacked in the storage bay is determined based on the established PSCW++ storage rule, such as Figure 7 As shown, it specifically includes the following sub-steps:

[0120] Step S601: Find available stacks in the allocated bays and construct an available stack set S1;

[0121] According to the order of j = 1, 2, ..., S, S represents the total number of stacks in the bay, check the stacks s in turn jIs there any empty box available? If so, stack s j Add to the available stack set S1;

[0122] Step S602: searching for a matching stack that satisfies rule C2 in the available stack set S1, and constructing a matching stack set S2;

[0123] According to the order of j=1,2,…,|S1|, where |S1| represents the size of set S1, check stack s once j Does it meet the requirements of rule C2? If so, stack s j Join to S2;

[0124] The meaning of rule C2 is that containers with a short storage time can be placed on top of containers with a long storage time. Therefore, the inspection process that satisfies rule C2 is as follows: Figure 8 As shown:

[0125] Determine the stacks s in the available stack set S1 one by one j Storage time of the top container dt(s j ), if dt≤dt(s j ), then stack s j To satisfy the stack of rule C2, j Add to the matching stack set S2 that satisfies rule C2; ​​otherwise, the loop control variable is incremented by 1 and the next judgment is performed, where dt is the predicted storage time of the container to be stacked, and |S1| represents the number of elements in the set S1;

[0126] Step S603: Make the following determination on the matching stack set S2, specifically:

[0127] ① If the matching stack set S2 is not an empty set, then a stack s in the matching stack set S2 i The top bin position of is allocated to the container to be stacked. If there may be multiple stacks that meet rule C2, the stack s with the smallest stack number i is selected. i Assigned to containers to be stacked;

[0128] ② If the matching stack set S2 is an empty set, a stack s in the available stack set S1 will be j The top box slot of is allocated to the container to be stacked. If there may be multiple available stacks in the set S1, the stack s with the smallest stack number j is selected. j Assigned to containers to be stacked.

[0129] Embodiment 2:

[0130] 1. Analysis of the effect of rule C1:

[0131] Rule C1: Containers that have been stored for a long time are preferentially stored in the container area far from the berth, while containers that have been stored for a short time are preferentially stored in the container area close to the berth;

[0132] Purpose: To improve the efficiency of loading and unloading ships;

[0133] Metric: TEU / h.

[0134] The theoretical analysis of the performance improvement brought by applying rule C1 is as follows:

[0135] Due to the distance between the container area and the berth, there are differences in the efficiency of loading and unloading operations in different container areas. If Rule C1 is not applied, containers can theoretically be stacked in any container area; if Rule C1 is used, the container area should be selected according to the container storage time.

[0136] The distance between the container area and each berth is known, assuming that d ij To represent the distance between the i-th container area and the j-th berth, ignoring the distance difference between berths, use d i To represent the distance from the i-th box area to the shoreline, the distance is proportional to the time to complete the task, that is, Where v is the travel speed of the inner container truck, which always tends to be stable and can be regarded as a constant.

[0137] Without loss of generality, without the application of rule C1, assuming that the container area can be randomly selected, within a period of time T, there are N containers that need to be loaded and unloaded, and there are M container areas available. Without considering other factors, the probability of each container being randomly assigned to a container area is: The number of containers allocated to each container area is recorded as n i , where i∈{1,2,…,M}, can be viewed as:

[0138] The operation time in time period T is:

[0139]

[0140] remember It represents the sum of the distances from each container area to the berth, so the above formula becomes:

[0141]

[0142] Formula (2) shows that when the number of containers N (i.e., the task volume) is clear, since M and D are fixed values, the larger the v value, the faster the speed, and the shorter the time to complete the task; conversely, the smaller the v value, the longer the time to complete the task.

[0143] Without loss of generality, when rule C1 is applied, assuming that d 1 <d 2 <…d M , then the probability of each container area being assigned to a container is defined as:

[0144]

[0145] Obviously i ∈[0,1], and satisfies: Therefore, formula (3) meets the requirements of probability definition and practical application requirements and is reasonable. When i = 1, At this time p i The value is the largest, which means that the possibility of container area 1 obtaining container allocation is the greatest; on the contrary, when i=M, At this time p i The smallest value means that container zone M has the least possibility of obtaining container allocation.

[0146] According to formula (3), the number of containers allocated to a certain container area is:

[0147]

[0148] The operation time in time period T is:

[0149]

[0150] According to Amdahl's law, the performance improvement brought by applying rule C1 is:

[0151]

[0152] According to Chebyshev's inequality, for a positive number a 1 ,a 2 ,…,a n and b 1 ,b 2 ,…,b n , and there is a relationship: a 1 ≤a 2 ≤…≤a n and b 1 ≤ b 2 ≤…≤b n , then:

[0153]

[0154] The above formula is equivalent to:

[0155]

[0156] Speedup withC1≥1, according to formula (6), we need to prove:

[0157]

[0158] Divide both sides of the above equation by M to get:

[0159]

[0160] Compared with Chebyshev inequality (8), equation (10) holds, that is, Speedup withC1 ≥1, certified.

[0161] In formula (6), Speedup withC1 =1, if and only if d i =d j , Its physical meaning is: when the distances from each container area to the berth are equal, the acceleration ratio is equal to 1, and the performance cannot be improved at this time.

[0162] 2. Analysis of the effect of rule C2:

[0163] Rule C2: Containers with a shorter storage time can be stacked on top of containers with a longer storage time, but not vice versa;

[0164] Purpose: To reduce the rate of box turnover;

[0165] Metric: Number of times boxes are turned over.

[0166] The theoretical analysis of the performance improvement brought by applying rule C2 is as follows:

[0167] Fig. 9 The figure is an example diagram of a bay configuration. The horizontal axis represents the stack number, and the vertical axis represents the layer number. The figure shows a bay configuration with 5 stacks and 4 layers. The blocks in the two-dimensional plane represent container slots. The solid-line blocks with numbers indicate that containers have been stacked, while the dashed-box blocks indicate empty slots, and the dark-colored blocks indicate containers that are about to be stacked. The numbers in the blocks indicate the order of picking up containers, also known as container priority. The smaller the number, the higher the priority, and vice versa. The larger the number, the lower the priority. The priority is proportional to the storage time, that is, the shorter the storage time, the earlier the container is picked up and the higher the priority, and vice versa, the longer the storage time, the lower the corresponding priority. It is worth noting that there are multiple cases where the numbers in the solid-line blocks are repeated, which means that the corresponding container pick-up priorities are the same, which is consistent with the application scenario of inland container terminals.

[0168] According to the following references:

[0169] [[Kim,2006]A Heuristic Rule for Relocating Blocks[J].Computers&Operations Research.2006,33(4):940–954;

[0170] [Forster,2012]A Tree Search Procedure for the Container RelocationProblem[J].Computers&Operations Research,2012,39(2):299–309;

[0171] [Boysen,2016]Boysen N.,Emde S.The parallel stack loading problem tominimize blockages.European Journal of Operational Research.249(2):618–627;

[0172] [Sven Boge,2020]The Parallel Stack Loading Problem Minimizing theNumber of Reshuffles in the Retrieval Stage[J].European Journal ofOperational Research.280(2020):940–952;

[0173] [Fengwei Liu,2023]Dynamic Attention Model-A Deep ReinforcementLearning Approach for Container Relocation Problem[J].36th InternationalConference on Industrial,Engineering and Other Applications of AppliedIntelligent Systems(IEA / AIE 2023),Shanghai,China,July 19–22,2023Proceedings,Part II:273-285;

[0174] [Xin Jin,2023]Container Stacking Optimization Based on DeepReinforment Learning[J].Engineering Applications of ArtificialIntelligence.123(2023):106508;

[0175] The description of the performance measurement of bay structure in the above references, the unordered stacking (US) value US(c) and the badly placed items (BI(c)) of a bay structure c can be used as measurement indicators for the number of box flipping times. The BI value is used as an analysis example below, and the analysis results of the US value can be obtained by the same token.

[0176] Let the position of the dark block before the container is placed be c, and after the container is placed be c'. Without applying rule C2, there is a high possibility that BI(c')>BI(c). For example, in the figure above, if the priority of the dark block is "3", placing it in stack 1, stack 2, and stack 5 will result in the generation of BI items, because the priorities of the top containers in these stacks are all less than 3. In this case, if rule C2 is applied, the dark container can only be placed in stack 3 and stack 4, which will not increase the BI(c') value.

[0177] The possibility of BI(c')>BI(c) is closely related to the storage time of the containers to be stacked. The longer the storage time of the containers to be stacked, the smaller the priority of the dark-colored containers. No matter which stack c they are stacked in, the greater the possibility of BI(c') increasing. For example, if the priority of the dark-colored container is greater than or equal to 6, no matter which stack it is stored in in the above figure, BI(c') will increase by 1. If the priority of the dark-colored container is less than or equal to 1, no matter which stack it is stored in in the above figure, BI(c') will not increase.

[0178] In summary, the container stacking process applying rule C2 will not lead to disordered stacking and improper placement even though the number of containers in the bay increases. Therefore, the number of container turnovers is effectively controlled and non-productive costs are reduced.

[0179] In summary, the present invention provides a container storage location selection method based on storage behavior characteristics. In order to solve the problem that the existing container storage location selection technology does not utilize the historical data of port operations, the storage time of containers is predicted by a neural network, and combined with a rule-based storage location selection method, the predicted storage time is used for container area selection and container location selection, which can effectively improve the operating efficiency and reduce the container turnover rate.

[0180] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present invention. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present invention should still be covered by the claims of the present invention.

Claims

1. A container storage location selection method based on storage behavior characteristics, characterized in that: The following steps are included: Step S100: Based on the existing PSCW stockpiling rules, a PSCW+ stockpiling rule is formed, and based on the PSCW+ stockpiling rule, a new PSCW++ stockpiling rule is established, as follows: On the basis of the existing PSCW stacking rules, the following stacking rules are added for inland container terminals to form PSCW+ stacking rules, as follows: Rule A1: Define the export miscellaneous pile area, which is for whole shells and cannot be used to pile import boxes; Rule A2: Define the overload container stacking area, which is one or more whole bays, and these bays are associated with the specified yard bridge; Rule A3: Different types of containers of the same size cannot be stacked in the same stack, and ordinary standard containers and heightened standard containers cannot be stacked in the same stack; Rule A4: Reserve a container transfer space in the import or export mixed pile area, and reserve a container space corresponding to the container area floor height minus 1; Rule A5: When containers from multiple owners enter the same bay, containers of the same owner, size, and type shall be stored in a single stack to avoid mixing and causing container overturning; Rule A6: The height difference between adjacent stacks should not exceed 3 container heights to avoid isolated stacks; On the basis of the formed PSCW+ stockpiling rules, the following stockpiling rules are added to form the PSCW++ stockpiling rules, as follows: Rule C1: Containers that have been stored for a long time should be stored in a container area far from the berth, and containers that have been stored for a short time should be stored in a container area close to the berth; Rule C2: After the container area is selected, containers with a shorter storage time should be placed on top of containers with a longer storage time, and vice versa; Step S200: Establishing a storage time predictor model and using historical data to train the time predictor model, specifically including the following sub-steps: Step S201: Establish a storage time predictor model, which is specifically as follows: Y = f(s); s=∑ j W ij *x j ; in, Y represents the output, corresponding to the storage time of the container; W ij Represents the weight parameter from network node j to input node i; x j represents the jth input node, all x j The set of represents the input vector; f(s) represents the activation function, where It is an S-shaped curve function with an output range between (0,1), which is continuous and differentiable; Step S202: acquiring a historical stockpile data set from the terminal production operation information system, and dividing the historical stockpile data set into a training data set and a test data set in proportion; In the stockpile data set, the data items of the input vector x include: Container attributes, including information such as size, status, type, contents, arrival time, and departure time; Container stakeholder attributes, including information on the container holder, carrier vessel, consignee, consignor, loading port, and unloading port; Container terminal policy attributes, including information such as customs inspection time; The output Y of the model is the storage time of the corresponding container; Step S203: training the storage time predictor model, using the training data set in step S202, and adopting the error back propagation algorithm for training, wherein the learning rate η is set to 0.01; Step S204: After the training is completed, the weight parameter W of the storage time predictor model is obtained. ij The model parameter set θ is substituted into the storage time predictor model, and the trained storage time predictor model is verified using the test data set. The weight parameter W is modified. ij The set θ of which the trained storage time predictor model is obtained; Step S300: According to the trained time predictor model, the parameters of the container to be stacked are input to calculate the predicted storage time dt of the container to be stacked; Step S400: According to the calculated predicted storage time dt of the container to be stacked, the storage area of ​​the container to be stacked is determined based on the established PSCW++ storage rule, which specifically includes the following sub-steps: Step S401: Obtain the following parameters from the terminal production operation information system, including: ①Minimum storage time dt min and the maximum storage time dt max ; ②Container area collection B available for container storage k , k=1,2,…,K; ③ Available box area B k Distance to berth d k , k=1,2,…,K; Step S402: Based on the distance d from the container area to the berth k , from small to large for box area B k Sort by distance d k A set of bins B sorted by size l , l=1,2,…,K; Step S403: Based on the calculated predicted storage time dt of the container to be stacked, according to the rule C1 in the established PSCW++ storage rule, the distance d k Small container area, used to store containers with short storage time dt; distance d k The far container area is used to store containers that have been stored for a long time. Step S500: Determine the stacking position of the container to be stacked in the storage area based on the established PSCW++ stacking rule; Step S600: According to the calculated predicted stacking time dt of the container to be stacked, the location of the container to be stacked in the stacking bay is determined based on the established PSCW++ stacking rule.

2. A container storage location selection method based on storage behavior characteristics according to claim 1, characterized in that: The step S500 is specifically as follows: Get the determined container area B to be stacked l The number of bays B, calculate each bay b in turn i ,i∈{1,2,3,...,B}, the number of available slots sn(b i ), after the calculation is completed, sn(b i ) value is the largest and each container to be stacked is allocated to the position.

3. A container storage location selection method based on storage behavior characteristics according to claim 2, characterized in that: The step S600 specifically includes the following sub-steps: Step S601: Find available stacks in the allocated bays and construct an available stack set S1; Step S602: searching for a matching stack that satisfies rule C2 in the available stack set S1, and constructing a matching stack set S2; Step S603: Make the following determination on the matching stack set S2, specifically: ① If the matching stack set S2 is not an empty set, then a stack s in the matching stack set S2 i The top bin position of is allocated to the container to be stacked. If there may be multiple stacks that meet rule C2, the stack s with the smallest stack number i is selected. i Assigned to containers to be stacked; ② If the matching stack set S2 is an empty set, a stack s in the available stack set S1 will be j The top box slot of is allocated to the container to be stacked. If there may be multiple available stacks in the set S1, the stack s with the smallest stack number j is selected. j Assigned to containers to be stacked.

4. The method for selecting a container storage location based on storage behavior characteristics according to claim 3 is characterized in that: The step S602 is specifically as follows: Determine the stacks s in the available stack set S1 one by one j Storage time of the top container dt(s j ), if dt≤dt(s j ), then stack s j To satisfy the stack of rule C2, j Add to the matching stack set S2 that satisfies rule C2; ​​otherwise, the loop control variable is incremented by 1 and the next judgment is performed; Among them, dt is the predicted storage time of the container to be stacked, and |S1| represents the number of elements in the set S1.

Images