An automated wharf yard box turning and position selecting random optimization method

Abstract

This invention discloses a stochastic optimization method for container yard repositioning in an automated terminal. By constructing and training a container retrieval probability prediction model, the probability of container i being retrieved in time period t is calculated. This calculation is based on a container retrieval probability matrix formed by the probabilities of all containers in the yard and the entire time period, achieving stochastic optimization of container yard repositioning decisions. The container retrieval probability prediction model is based on a multilayer perceptron (MLP) neural network structure. During the training of this model, historical operation records of the terminal's container yard are acquired, and at least container attribute information, quay crane work path records, and vessel stowage information from these records are used as inputs to the model.

Description

Technical Field

[0001] This invention belongs to the field of logistics technology, and specifically relates to a random optimization method for container repositioning in automated terminal yards. Background Technology

[0002] Ports are fundamental and pivotal facilities, serving as a vital support for economic development. With the advancement of technologies such as big data and automation, automated container terminals offer significant advantages in improving loading and unloading efficiency, enhancing production safety, and improving operating conditions, and have become the future direction of port development.

[0003] like Figure 1 The diagram shows a layout of an automated container terminal. In the terminal yard, containers are stacked vertically in different container areas. During loading operations, containers are first moved from the container area to inner trucks by rail-mounted gantry cranes, then transported by inner trucks to the quay cranes, and finally placed on the ship at their designated locations by the quay cranes. Containers are unique and non-interchangeable; the order in which containers are retrieved is generally determined by the quay cranes based on the ship's stowage plan. Figure 1 In a container yard, a "column" refers to the horizontal arrangement of containers. A "row" typically refers to a unit where containers are arranged perpendicular to the quay edge, and column numbers usually start from the quay edge and proceed sequentially into the yard. A "layer" refers to the number of vertical stacks of containers; in a container yard, a layer represents the height of the stack and is usually counted upwards from the ground.

[0004] Due to the dynamic and uncertain nature of terminal operations, discrepancies inevitably arise between the stacking order of containers within the yard and the order in which they are retrieved. When other containers are stacked on top of a container to be retrieved, obstructing containers need to be moved to other locations; this process is called container repositioning. Determining the location of these obstructing containers is known as container repositioning.

[0005] Therefore, in a container terminal yard, obstructing containers are those located above the target container, hindering its retrieval. To retrieve the target container, these obstructing containers must first be moved to another location; this process is called container repositioning. Container repositioning is the process of determining which obstructing containers need to be moved and where in the yard they should be moved so that the target container can be retrieved smoothly. Since container repositioning significantly reduces the container dispatching efficiency of the terminal yard, an inappropriate container repositioning strategy can lead to repeated container repositioning, further reducing the yard's repositioning efficiency. Summary of the Invention

[0006] One embodiment of the present invention provides a stochastic optimization method for container retrieval in a terminal container yard. This method calculates the probability of container i being retrieved within time period t by constructing and training a container retrieval probability prediction model for the yard. A container pickup probability matrix is ​​formed based on the probability of all containers in the yard and the probability of a complete time period. Achieve stochastic optimization of container yard repositioning decisions.

[0007] Preferably, the container pick-up probability prediction model in the yard is based on a multilayer perceptron (MLP) neural network structure.

[0008] In training the container pick-up probability prediction model for the container yard, historical operation records of the terminal container yard are obtained, and at least container attribute information, quay crane work route records, and ship loading information in the records are used as inputs to the model.

[0009] The container ship operation time corresponding to container pickup is divided into multiple equally spaced time periods, and the time period number containing the container pickup time is the category of the container.

[0010] This embodiment of the disclosure addresses container terminal overturning operations by considering the dynamic uncertainties present during terminal operation, optimizing overturning location selection, determining the placement of obstructing containers, avoiding repeated overturning, and improving terminal yard efficiency. Attached Figure Description

[0011] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0012] Figure 1 A schematic diagram of the layout of an automated container terminal in the prior art.

[0013] Figure 2 A schematic diagram illustrating the training and deployment of a yard container retrieval sequence probability prediction model according to one embodiment of the present invention.

[0014] Figure 3 A schematic diagram comparing the yard turnover rate using the SSF method and the LSGP method according to one embodiment of the present invention.

[0015] Figure 4 A schematic diagram of the probabilistic accuracy sensitivity test results of the LSGP method according to one embodiment of the present invention. 1—Container area, 2—Shell, 3—Column, 4—Layer, 5—Container, 6—Queue crane, 7—Rail-mounted gantry crane, 8—Inner truck, 9—Outer truck. Detailed Implementation

[0016] Regarding the issue of container yard repositioning, existing solutions often employ the Shortest Stack First (SSF) rule to select the best location for container stacking. Obstacle boxMove it to the column with the shortest stack height. This method is simple and fast, but it easily leads to repeated container handling and cannot effectively improve the container dispatching efficiency of the yard.

[0017] Academics have proposed mathematical programming models, heuristics, and tree search methods for the container retrieval and location problem. However, these methods are all based on a given complete container retrieval sequence. Here, the container retrieval sequence refers to the order in which containers are retrieved from the terminal yard. Due to factors such as extreme weather, equipment failure, and ship delays, there are many uncertainties in the actual operation of the terminal, and the complete container retrieval sequence cannot be determined in advance.

[0018] Therefore, to address the problem of high random uncertainty in terminal operations leading to unpredictable container pickup sequences, this disclosure proposes a new concept— Container pick-up probability matrix This paper aims to characterize the uncertainty of container retrieval order in the yard and achieve a true solution by accurately describing this problem. This disclosure uses historical terminal operation data to extract the retrieval time of containers in the yard, along with their corresponding container attribute information, ship stowage, and quay crane work route arrangements, to establish a machine learning prediction model for the probability matrix of container retrieval in the yard.

[0019] Based on the aforementioned container retrieval probability matrix, this disclosure also proposes a rule called LSGP (Local Safety and Global Potential), which comprehensively considers the local safety and global optimization potential of each container relocation to avoid repeated relocations. Finally, this disclosure experimentally verifies the effectiveness and robustness of the proposed method.

[0020] According to one or more embodiments, an automated terminal yard container repositioning and selection stochastic optimization method is provided. Here, stochastic optimization is performed to maximize or minimize the expected value of a target. In cases involving probability, decision-making is used to maximize or minimize the expected value of the target. The automated terminal yard container repositioning and selection stochastic optimization method of this embodiment mainly includes two steps: predicting the probability matrix of the container pick-up sequence in the yard and determining the container repositioning and selection location.

[0021] (1) Prediction Probability matrix of container pick-up sequence at the storage yard.

[0022] In order to capture Extraction timeThe nonlinear relationships between factors, including quay crane work route plans, ship stowage, and container attribute information, are estimated using a Multilayer Perceptron (MLP) to estimate extraction probabilities. A multilayer perceptron is a feedforward artificial neural network containing at least three layers: an input layer, one or more hidden layers, and an output layer. Except for the input nodes, each node is a neuron with a nonlinear activation function. Each connection in an MLP has a weight, and each neuron has a bias; these parameters are continuously updated during training via backpropagation. During training, the network minimizes the difference between the output and the actual value by adjusting the weights and biases; this process is typically quantified using mean squared error or other cost functions. Multilayer perceptrons are used to capture and learn complex mapping relationships between input data. The steps for building a multi-class MLP model are as follows: Figure 2 As shown, the specific description is as follows:

[0023] Step S101: Feature engineering.

[0024] Collect historical terminal operation records. Each record includes three aspects: container attribute information, quay crane work route record, and vessel stowage information. Container attribute information includes container ID, weight, type, dimensions, destination, storage location, and retrieval time. Quay crane work route record includes the quay crane number assigned to each container and its start and end times. Vessel stowage includes the location of each container assigned on the vessel.

[0025] Step S102: Label definition.

[0026] The ship's service time is divided into 12 equal time intervals. The time interval number that includes the container pickup time is used as the category of the container to obtain the label of the multi-classification model.

[0027] Step S103: Model training.

[0028] The MLP model is used. This model has 5 layers of neurons, with 64, 128, 256, 128, and 64 neurons per layer, respectively. The MLP model uses softmax to calculate probabilities, cross-entropy to calculate loss, and backpropagation to update the weights in the fully connected network. The MLP model outputs the probability for each class and selects the class with the highest probability as the prediction.

[0029] Step S104: Extraction of extraction probability.

[0030] Once the model training is complete, the network structure and weights are maintained. When deploying the model, information about the containers to be predicted, excluding extraction time, is collected, and the trained network is used to calculate the probability of each container category. This probability is the container pickup sequence probability in the yard, and the pickup sequence probabilities of all containers form the yard pickup probability matrix. This matrix is ​​a two-dimensional matrix. Its elements This represents the probability that container i is extracted during time period t.

[0031] like Figure 2 The diagram shown illustrates the training and deployment of a container yard container retrieval sequence probability prediction model. The model is trained by analyzing historical operational data and used to predict and optimize container retrieval operations at the terminal. The training steps for this model include:

[0032] S201. First, collect the historical operation data of the terminal, including quay crane work path records, ship loading information, container attribute information, etc. as the feature input of the model, and the container extraction time period sequence number in the historical operation data of the terminal as the output of the model.

[0033] S202 uses the softmax function as the probability calculation function to convert the model's output into a probability distribution for multi-class classification problems. It also defines the input layer, hidden layer, output layer, and the connections between them, and optimizes and adjusts the model structure and parameter weights to minimize the prediction error.

[0034] S203, in the model output layer, uses the argmax function to select the class with the highest prediction probability.

[0035] The deployment steps for the model include:

[0036] S205, Obtain information about the container to be predicted as model input, including quay crane work route plan, ship loading information, and container attribute information;

[0037] S206, the model outputs the container extraction probability, assigning an extraction probability to each container for each time period.

[0038] Therefore, the container retrieval probability matrix is ​​a two-dimensional matrix used to represent the probability of a container being retrieved at various time periods in the terminal yard. This matrix is ​​constructed based on probability data predicted by a multilayer perceptron (MLP) model and is used to optimize container repositioning decisions in the terminal yard.

[0039] (2) Determine the location for container overturning.

[0040] When it is necessary to determine the flipping location of a barrier box, this can be done by calculating the LSGP values ​​of other suitable barrier boxes. The LSGP calculation method is as follows:

[0041] LSGP(k)=P(x→*)+α|N(k)| 2

[0042] Where k represents the column ID where the target is located; P(x→*) represents the probability that the obstructing box x will become the first box to be extracted after being placed in column k; N(k) represents the set of all containers in column k; |N(k)| represents the number of existing containers in column k; α is a coefficient, which is 0.05 in this embodiment.

[0043] The calculation method for P(i→*) is as follows:

[0044] S301, firstly, based on the container pickup probability matrix of the yard, calculate the priority pickup probability matrix M = (m... i,j ), where m i,j Let represent the probability that container i is picked up before container j.

[0045]

[0046] S302, calculate the approximate priority extraction probability of the obstructing container x compared to all existing containers in column k.

[0047]

[0048] S303 normalizes the probability and calculates the final P(x→*) estimate.

[0049]

[0050] The larger the LSGP value, the lower the probability of repeated box flipping caused by obstruction boxes in that column, and the more space is left to prepare for possible box flipping in the future. Therefore, when making a selection, choose the column with the largest LSGP value, and the final box flipping position will be above the current tallest box in that column.

[0051] Therefore, in this embodiment, the LSGP value is an indicator used to evaluate the merits of container repositioning. A higher LSGP value indicates a lower likelihood of future repositioning operations that would otherwise hinder container placement at that location, while also reserving more space for potential future repositioning. In actual repositioning, the column with the highest LSGP value should be selected as the repositioning location.

[0052] Meanwhile, since LSGP computation involves only a small amount of direct computation, the computation time is negligible, while traditional methods of solving this problem often require a long solution search, and the search time increases exponentially as the problem size increases.

[0053] To verify the effectiveness of the proposed method, 56 computational examples of different scales were designed for experimentation. The experimental results are as follows. Figure 3 As shown. Figure 3 The points in the figure represent the container turnover rates under the existing SSF method and the LSGP method proposed in this disclosure, respectively, for the same calculation example. This figure shows that the results obtained by the method proposed in this disclosure are superior to existing methods in all calculation examples. The average container turnover rate of LSGP across 56 calculation examples is 47.27%, a decrease of 6.48% compared to SSF's 53.75%.

[0054] To address the potential inaccuracy of stockpile extraction probabilities, a probability accuracy sensitivity test was conducted on the LSGP method at different prediction accuracies. The results are as follows: Figure 4 . Figure 4 This diagram illustrates the relationship between the accuracy of the predicted probability of container retrieval from the yard and the average container turnover probability. The horizontal axis (X-axis) represents the accuracy of the predicted probability of container retrieval from the yard, ranging from 0.2 to 1.0, reflecting the accuracy in predicting the probability of container retrieval. The vertical axis (Y-axis) represents the average container turnover probability (%), ranging from 47% to 55%, used to measure the model's effectiveness in optimizing the problem; a smaller value indicates that fewer operations are needed to complete the task.

[0055] It can be observed that the method proposed in this disclosure does not require high accuracy in predicting the container pick-up probability. The LSGP method achieves better results when the prediction accuracy is above 0.3. This means that even if the container pick-up probability prediction is not very precise, the LSGP method can still work effectively. This demonstrates that the LSGP method has strong robustness in handling uncertainty and inaccuracy, and can effectively guide container sorting and location selection even when faced with incompletely accurate container pick-up probability predictions, thereby optimizing terminal operations and demonstrating good practicality.

[0056] Therefore, the beneficial effects of this disclosure include:

[0057] 1. The probability matrix of container pickup in the yard is used to characterize the uncertainty of the order of container pickup in the yard. Repeated container picking operations are avoided by reducing the expectation of container picking, which has a good optimization effect.

[0058] 2. The probability matrix of container pickup from the yard is calculated by establishing a machine learning model. This method has strong versatility and can be applied to different scenarios by establishing different machine learning models.

[0059] 3. The proposed method for determining container location based on the container retrieval probability matrix has high computational efficiency and can be solved quickly in scenarios with strict time requirements;

[0060] 4. The proposed method for solving container picking location based on the container picking probability matrix in the yard does not require high accuracy of the container picking probability matrix. It can still achieve good optimization results even when the accuracy of the container picking probability matrix in the yard is not high due to the inaccuracy of the machine learning model.

[0061] It should be understood that in the embodiments of the present invention, the term "and / or" is merely a description of the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, the character " / " in this document generally indicates that the preceding and following associated objects have an "or" relationship.

[0062] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0063] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0064] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for random optimization of container yard rehandling and positioning at a terminal, characterized in that, Through the constructed and trained container yard container picking probability prediction model, the probability that a container i in the container yard is picked at a time period t is calculated a container picking probability matrix formed based on the probabilities of all containers in the container yard and the entire time period realize random optimization of container yard container rehandling location decision In the training of the container yard container pick-up probability prediction model, historical operation records of the container yard of the wharf are obtained, and at least container attribute information, shore crane work path records, and ship stowage in the records are taken as inputs of the model, The container ship operation time corresponding to the pick-up operation is divided into multiple equal-interval time periods, and the time period number containing the container pick-up time is the category of the container, The LSGP value is used as an index for evaluating the pros and cons of the container rehandling position selection, Wherein, k represents the column id where the target position is located. represents the probability that the bin will be the first to be picked up after being placed in column k placed in column k will be the first to be picked up; denotes the set of all containers in column k; represents the number of existing containers in the kth column; is a preset coefficient.

2. The method of claim 1, wherein, The container yard container pick-up probability prediction model is based on a multi-layer perceptron (MLP) neural network structure.

3. The method of claim 1, wherein, The calculation methods include: S301, first, based on the yard container picking probability matrix, calculate the priority picking probability matrix between all containers two by two wherein represents the probability that container i is picked before container j; S302, calculate the blocking box The approximate priority extraction probability of all existing containers in the k column , S303, normalize the probability, calculate the final , 4. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor runs the computer program to implement the method of any one of claims 1 to 3.

5. A storage medium having stored thereon a computer program, characterized in that The computer program is executed by the processor to implement the method of any one of claims 1 to 3.

6. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 3.

Images