Optimization of container layout decision policy in the vertical type terminal based on simulation

Terminal Operation Simulation

We would like to introduce a method for optimizing the terminal operation policy based on the simulation of terminal operation. The following contents focus on a particular explanation of how to optimize the container device layout decision policy using block operation simulation, and the introduction focuses on the performance results of the optimization process by OPUS DigiPort

In the case of the terminal operation simulation, it depends on the purpose, whether to implement CHE (Container Handling Equipment) as a statistical model or Physics engine-based model, which considers CHE operation performance (moving speed, variable speed, etc.) and collisions between equipment. The physics engine-based model has higher reliability because it is closer to actual cases. Also, the simulation result can be checked intuitively, as you can see the status of actual terminal operation.

Terminal operation policy is a decision-making method for terminal operation. The work allocation policy for container transport vehicles is to decide on transport vehicles for container transportation, and the container device layout decision policy determines the positioning of the container that comes into the block. YC work allocation policy decides the container, which is the object of YC work, and it also decides AGV’s travel route in case of AGV operating terminals. The stated operation policy is vitally important because its performance level highly affects terminal productivity.

Among these operation policies, we would like to demonstrate the container device layout decision policy in the vertical-type terminal, as shown below.

Container Device Layout Decision Policy 

The feature of a vertical-type yard block is that the flow of the import container and the export container is opposite, as shown below picture. The import container comes in from the seaside and goes out to the landside, and the export container comes in from the landside end loading and goes out to the seaside end loading. Interference problems between cranes must be considered more carefully than horizontal-type terminals due to the circumstances of two-yard cranes included for each block in general and opposite container flows. Besides that, the container device layout needs to be decided by considering re-handling, the crane’s travel range, and other factors.

In the container device layout decision process, there are a few criteria to consider when deciding which stack to stack containers. For example, there are the following evaluation factors such as how much a container need to be moved when leaving the block, how many layers the height of the stack needs to be stacked, how much a 40ft stack shrinks, and how many re-handling occurs. Based on the above-stated criteria, score of the stack to stack up the containers can be calculated. In this way, each criterion can be used to calculate a score for candidate stacks and then determine the device location by the highest overall scored stack.

The importance of each criterion that terminal operators consider may be different. Hence the importance (weighting) could be assigned to each evaluation criterion accordingly. In the terminal where it places a premium on reducing the number of re-handling, for example, a large weighted value will be given to the evaluation criteria whether re-handling has occurred or not. The below picture illustrates the process. 

The policy is a function(s(x)) that receives a candidate stack(x) to be applied and outputs a score for it, and the score can be calculated as weighted sum (Weighted for criteria, C_i,w_i) of multiple criteria(Ci). Among the candidate stacks, the highest scored stack(x*) is chosen as device location. According to the weighted wi value combination, preference of the stack is changed. Hence, the combination of weights is substantially policy.

What is policy optimization, then? It is about finding the combination of weights to achieve the optimal performance. ‘Optimal performance’ is the purpose of policy optimization.

For example, optimized policy targets to minimize the block’s container service could be able to reduce the service time the most compared to other policies. Likewise, the optimized policy targets to minimize the number of re-handling would reduce the number of re-handlings the most compared to other policies.

Genetic Algorithm(GA) is a commonly used artificial intelligence technique when the search space is huge. Due to the weight given to the criteria being generally real numbers, there are infinite weight combinations. Yet, finding the optimized policy is extremely difficult.

OPUS DigiPort's Optimized Device Location Layout Policy

Applying OPUS DigiPort, let’s see how to optimize the device location layout policy and what results come out. GA has been applied as an optimal algorithm, and the below-stated block simulator is used to evaluate chromosomes (candidate policy). The simulator simulates only one block as an object, and emulation including a collision between yard cranes and specification of acceleration/deceleration has been applied.

The following is a diagram illustrating the optimization process using GA. 

The optimization target is “Minimizing of service time delay at the seaside crane.” Hence, the evaluation value for the candidate policy is calculated as the service delay time of the seaside crane from the result of block operation simulation by the candidate policy. The followings are the simulation parameters.  

     3. Block work plan

         - Transshipment ratio 50%

         - % of 20ft and 40ft containers are 50%, each respectively

         - The number of provided services for seaside containers per hour is 13.0, number of provided services for landside containers per hour is 7. 3

         - The average dwell time for the container is 3 days

         - The average block occupancy rate is approximately 56%

Operation performance for the first 168 hours (7 days) has been excluded considering the simulation starts with an empty block circumstance. After approximately 5,000 simulation evaluations are conducted, the following optimization patterns are identified.  

After the rapid growth of performance improvement in the early stage of the optimization process, gradual improvement has been identified. Let’s check how the policies detected during optimization stack up the containers. The red-circled point of the policy in the above graph is figures of stacked containers.  

The left of the picture is the seaside, and the right of the picture is the landside. The 5-layer stack is red-colored, and the single-layer stack is green-colored in the picture. 

In the early stage of policy A, the preference where the container is stacked nearby Transfer Point(TP) is observed. Due to the 50% of Transshipment ratio, it’s is also observed that 5-layer stacks are more distributed on the seaside than landside. Service delay time for policy A is 1,120 seconds. (Refer to ‘A’ in the above table) The average waiting time for transport vehicles at the seaside TP has reached almost 20 minutes. It’s because blocks in this condition have disadvantage regarding delayed time for unloading and shipping the containers. Due to the high number of stacked layers, re-handling occurs more often when the shipment containers are taken out. Also, cranes need be moved further to stack unloading containers because of more highly stacked layers on the seaside. 

Let’s take a look at the most optimized policy E. Contrary to policy A, attempts to keep the stack in the seaside relatively low number of stacked layers are observed. Crane needs to be moved further when stacking the unloading containers to maintain the observed appearance, yet the crane could stack it at a short distance when it is urgent due to the workload. Also, re-handling occurs relatively less when the shipping container is taken out. Service delay time for this policy is 129 seconds, which is approximately 2 minutes. Compared to policy A, the service delay time has been reduced to a tenth. 

For now, let’s have a look at the aspects regarding shipment / importing properties of the containers that make up the stack, not the stack layer aspects. 

Import containers that go out to the landside are red-colored, and they go out to the seaside are green-colored. Other colored ones, excluding red and green, are stacks that mixed containers of both properties. In the case of policy A, the containers have been stacked absurdly. Containers going out to the landside are stacked close to the seaside. Contrariwise, the containers going out to the seaside are stacked close to the landside. In this case, the service delay time will inevitably increase due to crane interference.

Fortunately, starting from policy B, the containers’ outgoing directions are considered to a certain extent. In the case of policy E, almost the only containers going out to the seaside are stacked in the seaside area.

Hitherto, we have looked at a method to optimize terminal operation policies using simulation. Prompt field application of the method, of course, is complex. To apply it to the field, several conditions need to be satisfied. First, real-time constraints must be satisfied. When the field requests a device location, the time required to decide the location must be short enough not to limit field operations. Secondly, the reliability of the simulation which is used for the optimization must be guaranteed. No matter how outstanding an optimization policy is, it would only be useful if the reliability of the simulator that measures its performance is reliable. Lastly, it is required to have a method to update the policy according to the terminal operation circumstance. As the terminal operation circumstance changes, so does the optimized policy. In reality, terminal circumstances change continuously. Hence it is necessary to update the optimized policy accordingly.

<Image Reference>

1. Kim, T.; Yang, Y.; Bae, A.; Ryu, K.R. Optimization of Dispatching Strategies for Stacking Cranes including Remarshaling Jobs. J. Navig. Port Res. 2014, 38, 155–162.

2. Kim, T.; Kim, J.; Ryu, K.R. Deriving Situation-Adaptive Strategy for Stacking Containers in an Automated Container Terminal. In Proceedings of the 2013 International Conference on Logistics and Maritime Systems, Singapore, 12-24 September 2013.

Author: Tae-Kwang Kim, Terminal Business Consultant, CyberLogitec 

The article or images cannot be reproduced, copied, shared or used in any form without the permission of the author and CyberLogitec.