Container terminal storage yard management and control rehearsal method and system and storage medium thereof

Abstract

The invention discloses a container terminal storage yard management and control rehearsal method and system and a storage medium thereof. The method comprises the steps of data acquisition and preprocessing; carrying out 3D physical modeling on the layout of the storage yard; the operation process rule logic modeling comprises operation process modeling and resource scheduling logic modeling, and the resource scheduling logic modeling comprises stockpiling prediction, dynamic path planning and reinforcement learning of dynamic optimization scheduling rules; the reinforcement learning dynamically optimizes a scheduling rule, defines three sub-award functions of efficiency award, energy consumption penalty and safety penalty, and dynamically adjusts the weights of the three sub-award functions according to a scene; the state of the storage yard is visually displayed through a 3D dynamic billboard in combination with a thermodynamic diagram; and regularly updating the model and the rule base to adapt to wharf business changes. The storage yard bottleneck is pre-judged in advance through deep learning, and passive response is reduced; meanwhile, the scheduling rule is dynamically optimized in combination with reinforcement learning, so that the model is dynamically adjusted along with data, and the complexity of wharf operation is adapted.

Description

Technical Field

[0001] The present invention belongs to the field of container terminal yard control, and in particular relates to a container terminal yard control preview method system and a storage medium thereof. Background Art

[0002] At present, container terminal yard management and control faces many challenges and problems. First, the yard utilization rate is low. Although the container throughput continues to increase, the yard planning and layout are unreasonable, resulting in the failure to fully utilize space resources. Some terminals are unable to operate efficiently despite tight space. Secondly, the operation efficiency is low, the volume of unloading is large, the operation process is complicated, and the requirements for operators are high. The information on the container area, position, row, and layer in the export heavy container operation is unclear, which further increases the operation time and cost.

[0003] At the same time, the traditional yard information management system is not perfect enough, lacking real-time monitoring and remote operation functions, and unable to timely grasp the distribution of containers in the yard and the status of operating equipment; and the current yard data processing and analysis capabilities are insufficient to support scientific decision-making. In dynamic environments such as traffic flow changes and obstacle movement, traditional yard management methods are difficult to adapt quickly and lack real-time monitoring and rapid response mechanisms to environmental changes.

[0004] Therefore, it is urgent to design a new method system to optimize and solve the above problems and improve the management and control level of container terminal yards. Summary of the invention

[0005] Purpose of the invention: In order to overcome the above shortcomings, the purpose of the present invention is to provide a container terminal yard management and control rehearsal method system and its storage medium, which can predict yard bottlenecks in advance through deep learning and reduce passive responses; at the same time, combined with reinforcement learning, dynamic optimization scheduling rules are realized, so that the model can be dynamically adjusted with the data to adapt to the complexity of terminal operations.

[0006] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides a container terminal yard management and control preview method system and its storage medium, comprising the following steps:

[0007] S1): Data collection and preprocessing;

[0008] S2): 3D physical modeling of yard layout;

[0009] S3): logical modeling of operation process rules, including operation process modeling and resource scheduling logic modeling;

[0010] S301): The resource scheduling logic modeling includes stockpile prediction, dynamic path planning and reinforcement learning (RL) dynamic optimization scheduling rules;

[0011] The reinforcement learning (RL) dynamic optimization scheduling rule defines three sub - reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s), and dynamically adjusts the weights of the three sub - reward functions according to the scenario;

[0012] S4): Visualize and display the yard status through a 3D dynamic dashboard combined with a heat map;

[0013] S5): Regularly update the model and rule base to adapt to changes in terminal operations (such as new shipping lines, equipment upgrades). Anticipate yard bottlenecks in advance through deep learning to achieve dynamic path planning and reduce passive responses; at the same time, combined with reinforcement learning, a dynamic optimization scheduling rule is implemented, quantifying efficiency, energy consumption, and safety performance, enabling the model to dynamically adjust and adapt to the complexity of terminal operations.

[0014] Furthermore, the reinforcement learning (RL) dynamic optimization scheduling rule in S301) includes:

[0015] (a) Define three sub - reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s). The non - congestion weights of the three sub - reward functions are 0.5, 0.3, and 0.2 respectively, and the congestion weights are 0.8, 0.1, and 0.1 respectively;

[0016] The calculation formula of the efficiency reward efficiency(s) sub - reward function is as follows:

[0017]

[0018] where α eff , β eff and γ eff are the weight coefficients of the efficiency score, t avg_op is the average operation time, dev_util is the equipment utilization rate, and t avg_wait is the average waiting time of the vehicle;

[0019] The calculation formula of the energy consumption penalty energy(a) sub - reward function is as follows:

[0020] energy(a) = eng_cons × t op + fuel_cons

[0021] where eng_cons is the equipment energy consumption, fuel_cons is the vehicle fuel consumption, and t op is the operation time;

[0022] The calculation formula of the safety penalty safety(s) sub - reward function is as follows:

[0023] safety(s) = α saf ·acc_rate + β saf ·vio_count + γ saf ·dev_fail_rate

[0024] where α saf , β saf and γ saf are the weight coefficients of the safety risk score, acc_rate is the accident rate, vio_count is the number of violation operations, and dev_fail_rate is the equipment failure rate; two sets of weights for the congestion weight and the non-congestion weight of each sub-reward function are defined, improving the robustness of the calculation preview;

[0025] (b) Calculate the yard congestion level congestion_level by combining the yard utilization rate, equipment utilization rate, and vehicle waiting time. The formula is as follows:

[0026]

[0027] where C is the maximum capacity of the yard, U is the current occupancy of the yard, U equipment is the average utilization rate of the equipment in the terminal, W is the average waiting time of the vehicle in the yard,

[0028] W max is the maximum waiting time threshold of the vehicle in the yard, and α, β, and γ are weight coefficients; calculate the yard congestion level by combining the yard utilization rate, equipment utilization rate, and vehicle waiting time, reflecting the actual operation status of the yard from different angles and providing comprehensive and accurate data support for congestion assessment;

[0029] (c) If the yard congestion level congestion_level is greater than the set threshold, then call the congestion weight of the sub-reward function; otherwise, call the non-congestion weight; set the calling discrimination condition for the congestion weight and the non-congestion weight, improving the robustness of the calculation preview;

[0030] (d) Define the reward function R(s,a). The reward function R(s,a) is obtained by weighted summation of three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s). The formula is as follows:

[0031] R(s, a) = w eff ·efficiency(s) - w eng ·energy(a) - w saf ·sasfety(s)

[0032] where w eff, w eng and w saf are the weight coefficients of the sub - reward functions, namely, the congestion weight or the non - congestion weight; the reward function is obtained by the combined action of three sub - reward functions: efficiency reward, energy consumption penalty, and safety penalty. With the congestion weight or non - congestion weight coefficient, comprehensive calculation of multiple factors is realized, further ensuring the reliability of the calculation;

[0033] (e) Use the SAC (Soft Actor - Critic) algorithm to train in combination with historical data to obtain the scheduling result; the SAC algorithm has a fast convergence speed, can complete training in a short time, and has high training stability, providing guarantee for model training.

[0034] Furthermore, the stack storage prediction in the S301) is specifically to construct an LSTM network, including an input layer, an LSTM layer, a fully - connected layer, and an output layer. Divide the historical stack storage data into a training set, a validation set, and a test set to train the network, and use the trained LSTM network model to predict the container stack storage demand; the LSTM model can process input data in real - time and generate prediction results, providing timely decision - making support for port operation.

[0035] Furthermore, the dynamic path planning in the S301) is specifically to construct a GNN (Graph Neural Network), including normalizing historical data such as traffic network data to the interval of 0 - 1, representing the traffic network as a graph, extracting and updating the feature information of nodes and edges through an MLP (Multi - Layer Perceptron), training using the path cost as the loss function, and using the trained model to dynamically plan the vehicle path; the GNN can effectively process graph - structured data, capture the complex relationships between nodes (such as intersections, distribution points) and edges (such as roads, transportation routes), and better plan transportation.

[0036] Furthermore, the operation process modeling in the S3) is specifically to formulate a priority strategy for allocating the operation sequence according to the ship's departure time and container type, as well as the stack storage rules of centralized stacking of containers on the same ship and separate management of export containers and import containers, improving the operation efficiency of the container terminal yard, optimizing space utilization, reducing operating costs, and enhancing safety.

[0037] Furthermore, the data collection and pre - processing in the S1) include:

[0038] S101): Collect historical operation data and real - time operation data; historical data combined with real - time data provides a reference for historical operation patterns and a basis for real - time decision - making in port operation;

[0039] S102): Clean outliers, fill in missing data, and label key events, improving data quality and the efficiency of model training.

[0040] Further, the specific implementation of the yard layout 3D physical modeling in S2) is to construct a yard 3D grid model based on the terminal design drawings and GIS geographic information data, and model the equipment and containers; the 3D grid model and equipment modeling provide basic data support for the rehearsal; at the same time, the 3D physical model can also be used to simulate the daily operations and emergency responses of the yard, helping to identify potential problems and test solutions.

[0041] The present invention also provides a container terminal yard control rehearsal system for implementing a container terminal yard control rehearsal method, including: a data collection and preprocessing module, a 3D physical modeling module, an operation process rule logic modeling module, a real-time monitoring module, and an update module;

[0042] The data collection and preprocessing module is connected to the 3D physical modeling module and the real-time monitoring module, and is used to collect historical operation data and real-time operation data, and preprocess them; the data collection and preprocessing module provides support for data collection and cleaning, ensuring the smooth training of subsequent models;

[0043] The 3D physical modeling module is connected to the operation process rule logic modeling module, and is used to construct a yard 3D grid model and model the equipment and containers; the 3D physical modeling module ensures the realization of the modeling function and provides basic data support for the training of the model;

[0044] The operation process rule logic modeling module is connected to the real-time monitoring module and the update module, and is used to formulate priority strategies and stacking rules; the operation process rule logic modeling module, as the core calculation module, ensures the realization of the core function of the rehearsal;

[0045] The real-time monitoring module is connected to the update module, and is used to visually display the yard status; the real-time monitoring module ensures the real-time monitoring of various aspects of data in the yard;

[0046] The update module is connected to the operation process rule logic modeling module, and is used to regularly update the model and the rule library; the update module ensures the update of the model and the rules, and maintains the robustness and sustainability of the model.

[0047] The present invention also provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the container terminal yard control rehearsal method are implemented.

[0048] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the container terminal yard control rehearsal method are implemented.

[0049] From the above technical solutions, it can be seen that the present invention has the following beneficial effects:

[0050] 1. A container terminal yard management and control preview method system and its storage medium of the present invention predict the yard bottleneck in advance through deep learning to reduce passive response;

[0051] 2. A container terminal yard management and control preview method system and its storage medium of the present invention realize dynamic optimization scheduling rules in combination with reinforcement learning, so that the model can be dynamically adjusted with data to adapt to the complexity of terminal operations;

[0052] 3. The container terminal yard management and control rehearsal method system and its storage medium of the present invention realize the upgrade from "experience-driven" to "data intelligence-driven", and significantly improve the terminal operation efficiency and risk resistance. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A diagram showing the steps of a container terminal yard management and control preview method according to the present invention;

[0054] Figure 2 It is a schematic diagram of a container terminal yard management and control rehearsal system according to the present invention. DETAILED DESCRIPTION

[0055] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0056] Example

[0057] In this embodiment, Figure 1 The present invention discloses a container terminal yard management and control preview method, comprising the following steps:

[0058] S1): Data collection and preprocessing;

[0059] S2): 3D physical modeling of yard layout;

[0060] S3): logical modeling of operation process rules, including operation process modeling and resource scheduling logic modeling;

[0061] S301): The resource scheduling logic modeling includes stockpile prediction, dynamic path planning and reinforcement learning (RL) dynamic optimization scheduling rules;

[0062] The reinforcement learning (RL) dynamic optimization scheduling rule defines three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s), and dynamically adjusts the weights of the three sub-reward functions according to the scenario.

[0063] S4): Visualize and display the yard status through a 3D dynamic dashboard combined with a heat map.

[0064] S5): Regularly update the model and rule base to adapt to changes in terminal operations (such as adding new shipping lines and equipment upgrades).

[0065] Specifically, use Vue3 to build a dynamic interactive interface, a 3D rendering library to create and render a 3D scene to produce a 3D dynamic dashboard, and combine it with a heat map library to display the real-time heat map in the yard.

[0066] Especially, the resource scheduling logic modeling can also add exception handling logic modeling and a fault response strategy. When a crane fails, the standby equipment will be automatically triggered to take over the task.

[0067] In this embodiment, as Figure 1 , the reinforcement learning (RL) dynamic optimization scheduling rule in S301) includes:

[0068] (a) Define three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s), and the non-congestion weights of the three sub-reward functions are 0.5, 0.3, and 0.2 respectively, and the congestion weights are 0.8, 0.1, and 0.1 respectively;

[0069] The calculation formula of the efficiency reward efficiency(s) sub-reward function is as follows:

[0070]

[0071] Among them, α eff , β eff , and γ eff are the weight coefficients of the efficiency score, t avg_op is the average operation time, dev_util is the equipment utilization rate, and t avg_wait is the average waiting time of the vehicle;

[0072] Specifically, the efficiency reward efficiency(s) refers to an index of the efficiency of completing tasks in a specific state, where s is the specific state, such as the real-time state of the yard (equipment location, container distribution, operation queue, etc.);

[0073] The calculation formula of the energy consumption penalty energy(a) sub-reward function is as follows:

[0074] energy(a) = eng_cons × t op + fuel_cons

[0075] where eng_cons is the energy consumption of the equipment, fuel_cons is the fuel consumption of the vehicle, and t op is the operation time;

[0076] Specifically, the energy consumption penalty energy(a) refers to the energy consumption for performing a specific action under a scheduling instruction, where a is the scheduling instruction, such as "sending the AGV to storage area A";

[0077] The calculation formula for the safety penalty safety(s) sub - reward function is as follows:

[0078] safety(s) = α saf ·acc_rate + β saf ·vio_count + γ saf ·dev_fail_rate

[0079] where α saf , β saf and γ saf are the weight coefficients of the safety risk score, acc_rate is the accident rate, vio_count is the number of violation operations, and dev_fail_rate is the equipment failure rate;

[0080] Specifically, the safety penalty safety(s) refers to the index of the system safety risk under a specific state, where s is the specific state, such as the real - time state of the storage yard (weather, etc.);

[0081] (b) Calculate the yard congestion level congestion_level by combining the yard utilization rate, equipment utilization rate, and vehicle waiting time. The formula is as follows:

[0082]

[0083] where C is the maximum capacity of the yard, U is the current occupancy of the yard, U equipment is the average utilization rate of the equipment at the terminal, W is the average waiting time of the vehicle in the yard, and W max is the maximum waiting time threshold of the vehicle in the yard, and α, β, γ are the weight coefficients;

[0084] Specifically, α, β, γ are set according to experience and can be modified and fine - tuned under special circumstances;

[0085] (c) If the yard congestion level congestion_level is greater than the set threshold, then call the congestion weight of the sub - reward function; otherwise, call the non - congestion weight;

[0086] Specifically, the set threshold can be adjusted according to actual production;

[0087] (d) Define the reward function R(s,a), which is obtained by weighted summation of three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s). The formula is as follows:

[0088] R(s, a) = w eff ·efficiency(s) - w eng ·energy(a) - w saf ·safety(s)

[0089] Where w eff 、w eng and w saf are the weight coefficients of the sub-reward functions, i.e., the congestion weight or non-congestion weight;

[0090] (e) Use the SAC (Soft Actor-Critic) algorithm to train in combination with historical data to obtain the scheduling result.

[0091] Specifically, the operator can be allowed to score the scheduling result and incorporate manual feedback into the reward function. That is, if a certain scheduling is "inefficient", the weight of the efficiency reward is increased.

[0092] Specifically, the reward weight of "task completion speed" can be increased. While increasing the computational amount, the efficiency can be improved to a certain extent.

[0093] In this embodiment, as Figure 1 , the stack storage prediction in S301) is specifically to construct an LSTM network, including an input layer, an LSTM layer, a fully connected layer, and an output layer. And the historical stack storage data is divided into a training set, a validation set, and a test set to train the network, and the trained LSTM network model is used to predict the container stack storage demand.

[0094] Specifically, after constructing the LSTM network, set parameters such as the number of units in the LSTM layer, the learning rate, and the optimizer (such as RMSProp). Use the training set data to train the LSTM model, and adjust the model parameters through the validation set; after training, use the test set to evaluate the model performance, calculate the mean absolute percentage error (MAPE) and the root mean square error (RMSE), and use the trained LSTM model to predict the future container stack storage demand to provide support for the yard planning and operation decision-making of the port.

[0095] In this embodiment, as Figure 1The dynamic path planning in S301) is specifically to construct a GNN graph neural network, including normalizing historical data such as traffic network data to the range of 0 to 1, representing the traffic network as a graph, extracting and updating feature information of nodes and edges through an MLP multi-layer perceptron, using path cost as a loss function for training, and using the trained model to dynamically plan vehicle paths.

[0096] Specifically, the nodes in the GNN graph neural network are intersections, warehouses, distribution centers, etc., and the edges are roads, transportation routes, etc. The neighbor information of each node is aggregated in an iterative manner, and the features of the nodes and edges are updated to achieve optimized path planning; at the same time, it combines real-time data such as real-time traffic flow, obstacle location, vehicle status, etc. to achieve dynamic path planning.

[0097] In this embodiment, Figure 1 The operation process modeling in S3) is specifically to formulate a priority strategy according to the ship departure time and container type to allocate the operation sequence and the storage rules for centralized stacking of containers on the same ship and zoning management of export containers and import containers.

[0098] In this embodiment, Figure 1 , the data collection and preprocessing in S1) include:

[0099] S101): Collect historical operation data and real-time operation data;

[0100] S102): Clean outliers, fill in missing data and mark key events.

[0101] Specifically, historical operation data includes loading and unloading records, equipment operation logs, stockpile distribution, etc.; real-time data includes GPS positioning, sensor data, ship arrival plans, etc.

[0102] In this embodiment, Figure 1 The S2) 3D physical modeling of the yard layout is specifically to build a 3D grid model of the yard according to the terminal design drawings and GIS geographic information data, and model the equipment and containers.

[0103] In particular, the CAD model and geographic information data are integrated in the GIS software to generate a complete 3D yard model.

[0104] In this embodiment, Figure 2 , the present invention also discloses a container terminal yard control preview system, which is used to implement a container terminal yard control preview method, including: a data acquisition and preprocessing module, a 3D physical modeling module, an operation process rule logic modeling module, a real-time monitoring module and an update module;

[0105] The data acquisition and preprocessing module is connected to the 3D physical modeling module and the real-time monitoring module, and is used to collect historical operation data and real-time operation data and preprocess them;

[0106] Specifically, the data acquisition and preprocessing module can integrate sensor devices and collect data through the use of Internet of Things technology, including video streams, environmental parameters, security events, etc.;

[0107] The 3D physical modeling module is connected to the operation process rule logic modeling module and is used to construct a 3D grid model of the yard and model equipment and containers;

[0108] The operation process rule logic modeling module is connected to the real-time monitoring module and the update module and is used to formulate priority strategies and stacking rules;

[0109] The real-time monitoring module is connected to the update module and is used to visually display the yard status;

[0110] Specifically, the real-time monitoring module presents the processed results to the user in an intuitive manner through a display screen;

[0111] The update module is connected to the operation process rule logic modeling module and is used to regularly update the model and the rule base.

[0112] Specifically, this system can be integrated using an equipment control system (ECS) to improve the deployment efficiency.

[0113] In this embodiment, the present invention also discloses a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the container terminal yard control and rehearsal method are implemented.

[0114] In this embodiment, the present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the container terminal yard control and rehearsal method are implemented.

[0115] Specifically, computer-readable media include but are not limited to the following: electrical connection parts with one or more wirings (electronic devices), portable computer disk cartridges (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM).

[0116] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements can still be made, and these improvements should also be regarded as the protection scope of the present invention.

Claims

1. A container terminal yard management and control preview method, characterized in that: It includes the following steps: S1): Data collection and preprocessing; S2): 3D physical modeling of the yard layout; S3): Modeling the logical rules of the operation process, including operation process modeling and resource scheduling logic modeling; S301): The resource scheduling logic modeling includes inventory prediction, dynamic path planning, and reinforcement learning (RL) dynamic optimization of scheduling rules; The reinforcement learning (RL) dynamic optimization of scheduling rules defines three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s), and dynamically adjusts the weights of the three sub-reward functions according to the scenario; S4): Visualize and display the yard status through a 3D dynamic dashboard combined with a heat map; S5): Regularly update the model and rule base to adapt to changes in terminal operations (such as adding new shipping lines, equipment upgrades).

2. The container terminal yard management and control preview method according to claim 1 is characterized by: The reinforcement learning (RL) dynamic optimization of scheduling rules in S301) includes: (a) Define three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s), and the non-congestion weights of the three sub-reward functions are 0.5, 0.3, and 0.2 respectively, and the congestion weights are 0.8, 0.1, and 0.1 respectively; The calculation formula of the efficiency reward efficiency(s) sub-reward function is as follows: Among them, α eff , β eff and γ eff are the weight coefficients of the efficiency score, t avg_op is the average operation time, dev_util is the equipment utilization rate, and t avg_wait is the average waiting time of the vehicle; The calculation formula of the energy consumption penalty energy(a) sub-reward function is as follows: energy(a) = eng_cons × t op + fuel_cons where eng_cons is the energy consumption of the device, fuel_cons is the fuel consumption of the vehicle, and t op is the operation time; The calculation formula of the safety penalty safety(s) sub-reward function is as follows: safety(s) = α saf ·acc_rate + β saf ·vio_count + γ saf ·dev_fail_rate Among them, α saf , β saf and γ saf are the weight coefficients of the safety risk score, acc_rate is the accident rate, vio_count is the number of illegal operations, and dev_fail_rate is the equipment failure rate; (b) Calculate the yard congestion level congestion_level by combining yard utilization rate, equipment utilization rate, and vehicle waiting time. The formula is as follows: Among them, C is the maximum capacity of the storage yard, U is the current occupancy of the storage yard, and U equipment is the average utilization rate of the equipment at the terminal, W is the average waiting time of the vehicle in the storage yard, and W max is the maximum waiting time threshold of the vehicle in the storage yard, and α, β, and γ are weight coefficients; (c) If the yard congestion level congestion_level is greater than the set threshold, then call the congestion weights of the sub-reward functions; otherwise, call the non-congestion weights; (d) Define the reward function R(s,a). The reward function R(s,a) is obtained by weighted summation of the three sub-reward functions: efficiency reward efficiency(s), energy consumption penalty energy(a), and safety penalty safety(s). The formula is as follows: R(s, a) = w eff ·efficiency(s) - w eng ·energy(a) - w saf ·safety(s) Among them, w eff , w eng and w saf are the weight coefficients of the sub-reward function, that is, the congestion weight or the non-congestion weight; (e) Use the SAC (Soft Actor-Critic) algorithm to train with historical data to obtain the scheduling result.

3. The container terminal yard control and rehearsal method according to claim 1, wherein: The inventory prediction in S301) specifically constructs an LSTM network, including an input layer, an LSTM layer, a fully connected layer, and an output layer. Divide the historical inventory data into a training set, a validation set, and a test set to train the network, and use the trained LSTM network model to predict the container inventory demand.

4. The container terminal yard control and rehearsal method according to claim 1, characterized in that: The dynamic path planning in S301) specifically constructs a GNN graph neural network, including normalizing historical data such as traffic network data to the range of 0 to 1, representing the traffic network as a graph, extracting and updating the feature information of nodes and edges through an MLP multi-layer perceptron, training using the path cost as the loss function, and using the trained model to dynamically plan the vehicle path.

5. The container terminal yard control and rehearsal method according to claim 1, wherein: The specific operation process modeling in S3) is to formulate a priority strategy by allocating the operation sequence according to the ship departure time and container type, as well as the stacking rules of centralized stacking of containers on the same ship and separate management of export containers and import containers.

6. The container terminal yard control and rehearsal method according to claim 1, characterized in that: The data collection and preprocessing in S1) include: S101): Collect historical operation data and real-time operation data; S102): Clean outliers, fill in missing data, and label key events.

7. The container terminal yard control and rehearsal method according to claim 6, characterized in that: The specific 3D physical modeling of the yard layout in S2) is to construct a 3D grid model of the yard based on the terminal design drawings and GIS geographic information data, and model the equipment and containers.

8. A container terminal yard control rehearsal system for implementing a container terminal yard control rehearsal method according to claims 1 to 7, characterized in that: It includes: A data collection and preprocessing module, a 3D physical modeling module, an operation process rule logic modeling module, a real-time monitoring module, and an update module; The data collection and preprocessing module is connected to the 3D physical modeling module and the real-time monitoring module, and is used to collect historical operation data and real-time operation data and preprocess them; The 3D physical modeling module is connected to the operation process rule logic modeling module, and is used to construct a 3D grid model of the yard and model the equipment and containers; The operation process rule logic modeling module is connected to the real-time monitoring module and the update module, and is used to formulate a priority strategy and stacking rules; The real-time monitoring module is connected to the update module, and is used to visually display the yard status; The update module is connected to the operation process rule logic modeling module, and is used to regularly update the model and rule library.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the container terminal yard control and rehearsal method described above.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the container terminal yard control and rehearsal method described above.

Images