Sie sind auf Seite 1von 9

IOE 591 Midterm Exam 1

Francisco Aldarondo (faldaron) February 17, 2014

Problem 1
Part A
I present a general heuristic procedure (Heuristic 1) to determine a berthing plan assuming vessels have arrival time of zero and equal wieghts. I consider this heuristic for a general case where vessel length does not restrain multiple vessels from being berthed during the same time period. Heuristic 1 is based of heuristic H presented in [2] and heuristic HB in [1]. NOTE: The data provided for this excercise does restrain vessels to be moored one at a time. Under this assumption we present the established solution applying the Weigthed Shortest Processing Time (WSPT) rule. Under the assumption of equal weights the problems becomes a special case of (WSPT). I present an algorithm (SPT algorithm) to solve the problem to optimality. A proof of WSPT optimality is presented in part b. SPT is applied to solve part a and the nal berthing plan is displayed in a time-space diagram in Figure 2. Denitions 1. Jt : set of vessels not yet moored at stage t. 2. Gt : set of vessels to be moored at stage t.

Heuristic 1
1. Sort vessels by ascending order of processing time and by ascending order of vessel length within groups with the same processing time. Set t = 1. 2. Let Jt be the set of ships that are not moored, and set Gt to be the rst vessels in Jt such that their total accumulated length does not exceed S (Note: vessels in Gt need not be consecutive with respect to Jt ). 3. If t is even: Assign the last (largest processing time) vessel in Gt so that it occupies berthing section 1, then assign the remaining vessels in Gt such that no berthing section are left empty in between vessels. Else: Assign the last (largest processing time) vessel in Gt so that it occupies the last berthing section, then assign the remaining vessels in Gt such that no berthing section are left empty in between vessels. 4. Increase t by 1, if there are no more vessels to be moored go to step 5, else go to step 2. 5. Take vessel j occupying berth section l1 through lsj such that these berth sections (or a consecutive subset of these) are empty for x time periods after cj . Take vessel i such that: pi pj + x vessel i and vessel j can be swapped without exceeding maximum number of berth section Check the objective function value after the swap, if the objective value is incresed then discard the swap.

Shortest Processing Time


Additional assumption 1. Vessel lenghts versus berth lenght S constraints vessels to be berthed at most one per time period. 1

SPT Algorithm 1. Set t = 0, sort vessels in ascending order of processing time and store in Jt (See Table 1a and 1b). 2. Assign the rst vessel in Jt (vessel i) to the rst available berthing section at time t (See Figure 1a). 3. Set t = cj (completion time of vessel) and set Jt to be the set of vessels not yet moored in sorted order (See Figure 1b). If set Jt is empty then STOP, else Go to step 2.

Numeric Example

a)

b)

Table 1: a) Original data set. b) Sorted data following step 1 of SPT algorithm. .

a)

b)

Figure 1: a) Step 2 of SPT algorithm applied to data in table 1a. b) Step 3 of SPT algorithm. .

Figure 2: Time-space diagram of berthing plan generated with SPT algorithm.

Part B
We present a general heuristic procedure (Heuristic 2) to determine a berthing plan assuming vessels have arrival time of zero. We consider this a heuristic for a general case where vessel length does not restrain multiple vessels from being moored at the same time. NOTE: The data providad for this excercise does restrain vessels to be moored one at a time. Under this assumption Weigthed Shortest Processing Time (WSPT) rule garantees an optimal schedule. I present the WSPT algortim to solve the problem to optimality. A proof for the WSPT rule is presented, this proof is found in [5]. Part b is solved using the WSPT algoritm and the nal berthing plan is displayed in a time-space diagram in Figure 4. 2

Heuristic 2
1. Apply heuristic 1 keeping track of the order in with vessel are assigned and call it list b. 2. Take the rst vessel with weight greater than 1 in list b. 3. Swap positions with its preceding vessel in list b if vessel size is not restrictive. Evaluate the new solution. 4. If the objective function is larger than the previous solution then discard the last swap and Go to step 6, else update list b (to account for the swap) and repeat step 3. 5. Take the next vessel with weight greater than 1 in list b. If no such vessel exists, then Stop, else Go to step 3.

Weighted Shortest Processing Time


Additional assumption 1. Vessel lenghts versus berth lenght S constraints vessels to be berthed at most one per time period. WSPT Algorithm 1. Set t = 0, sort vessels in ascending order of weigthed processing time (pi /wi ) and store in Jt (See Tables 2a and 2b). 2. Assign the rst vessel in Jt (vessel i) to the rst available berthing section at time t (See Figure 3a). 3. Set t = ci (completion time of vessel i) and set Jt to be the set of vessels not yet moored in sorted order (See Figure 3b). If set Jt is empty then STOP, else Go to step 2. Proof. By constradiction. Suppose a schedule S , that is not WSPT, is optimal. In this schedule there must be at least two adjacent vessels, say vessel i followed by vessel j , such that: pi /wi > pj /wj . Assume that vessel i is moored at time t. Swap vessel i and j in the berthing plan. Call this new schedule S where vessel j is moored at time t and is followed by vessel i. Notice that the total weigthed turnaround time for all other vessel (other that i and j ) remains the same in both schedules. Under S the total weigthed turn around time of vessels i and j is (t + pi )wi + (t + pi + pj )wj , while under S it is (t + pj )wj + (t + pj + pi )wi . It is easy to verify that if pi /wi > pj /wj , the sum of the two weigthed turnaround times under S is strictly less than under S . These contradicts optimality under S .

Numeric Example

a)

b)

Table 2: a) Original data set. b) Sorted data following step 1 of WSPT algorithm.

a)

b)

Figure 3: a) Step 2 of WSPT algorithm applied to data in table 2a. b) Step 3 of WSPT algorithm. .

Figure 4: Time-space diagram of berthing plan generated with WSPT algorithm.

Part C
I present a general heuristic procedure (Heuristic 3) to determine a berthing plan assuming vessels have equal weights. This is a heuristic for a general case where vessel length does not restrain multiple vessels from being moored at the same time. NOTE: The data providad for this excercise does in fact restrain vessels to be moored one at a time. Under this assumption I present the shortest remaining processing time (SRPT) algorithm to obtain an optimal berthing plan. This algorithm applies a special case (equal weights) of the optimal decision rule commonly known as SRPT. I present a proof for SRPT rule in part d that is addapted from [6]. Denitions 1. Vessel with the same arrival time are grouped into batches (indexed by w = 1..W ) 2. Set Aw is comprised of all vessels in batch w that are moored after time of arrival of batch w + 1.

Heuristic 3
1. Sort vessels by ascending order of arrival time. 2. Apply the heuristic 1* for part (a) keeping track of the order in which vessels are assigned and call it list b. Set w = 1. *The heuristic is modied such that arrival times are not violated. At each stage, any vessel going into Gt must have arrival time less than or equal than the rst completion time amongst vessels assigned in the previous stage. 3. If w = W , then STOP. Consider the set Aw of vessels (belonging to batch w) that have been moored after the next batch arrival time (batch w + 1). Take the last vessel (according to list b) in set Aw . 4. Swap vessel from batch w with the rst vessel from batch w + 1 if vessel size is not restrictive. Evaluate the new solution. 5. If the objective function value is larger than the previos solution then discard the last swap and Go to step 6, else update list b (to account for the swap) and repeat step 4. 6. Take the next vessel (following order in list b) from set Aw . If there are no more vessels in set Aw , increase w by 1. Go to step 4.

Shortest Remaining Processing Time


Additional assumption 1. Vessel lenghts versus berth lenght S constraints vessels to be berthed at most one per time period. 4

Denitions 1. A vessels is said to be available at time t if its arraival time is less than or equal to t. 2. Jt : set of available vessels not moored at time t.

SRPT Algorithm
1. Set t to the rst arrival time, sort available vessels in ascending order of processing time and store in Jt (See Tables 3a and 3b). 2. Assign the rst vessel in Jt (vessel i) to the rst available berthing section at time t (See Figure 5a). 3. Set t = ci (completion time of vessel i) and set Jt to be the set of available vessels not yet moored in sorted order (See Figure 5b). If set Jt is empty then STOP, else Go to step 2.

Numeric Example

a)

b)

Table 3: a) Original data set. b) Sorted data following step 1 of SRPT algorithm. .

a)

b)

Figure 5: a) Step 2 of SRPT algorithm applied to data in table 3a. b) Step 3 of SRPT algorithm. .

Figure 6: Time-space diagram for nal solution by SRPT algorithm.

Part D
I present a general heuristic procedure (Heuristic 4) to determine a berthing plan assuming vessels have equal weights. This is a heuristic for a general case where vessel length does not restrain multiple vessels from being moored at the same time. NOTE: The data providad for this excercise does in fact restrain vessels to be moored one at a time. Under this assumption I present the shortest remaining weighted processing time (SRWPT) algorithm to obtain an optimal berthing plan. This algorithm applies the optimal decision rule commonly known as SRPT. I present a proof for SRPT rule that is addapted from [6]. 5

Heuristic 4
1. Sort vessels by ascending order of arrival time. 2. Apply the modied version of heuristic 1 in part (a). Set w = 1. 3. Go to step 2 in part b. 4. Go to step 3 in part c

Shortest Remaining Processing Time


Additional assumption 1. Vessel lenghts versus berth lenght S constraints vessels to be berthed at most one per time period. Denitions 1. A vessels is said to be available at time t if its arraival time is less than or equal to t. 2. Jt : set of available vessels not moored at time t.

SRWPT Algorithm
1. Set t to the rst arrival time, sort available vessels in ascending order of weighted processing time and store in Jt (See Tables 4a and 4b). 2. Assign the rst vessel in Jt (vessel i) to the rst available berthing section at time t (See Figure 7a). 3. Set t = ci (completion time of vessel i) and set Jt to be the set of available vessels not yet moored in sorted order (See Figure 7b). If set Jt is empty then STOP, else Go to step 2. Proof. By constradiction. Suppose there is a optimal schedule S , that is not SRWPT. Hence the exists a time interval (t, t + v ), v > 0, such that: there is a vessel j in Jt with weighted processing time smaller than every vesel in Jx for t x t + v , and either (1) all berth sections are empty during the time interval or (2) there is a unique vessel i moored at time x in the interval. If condition (1) is satised, clearly we can reduce cj without altering the other completion times by mooring vessel j in the time interval (t, t + pj ). Suppose condition (2) is satised swap vessels i and j and call this new berthig plan S . Total weigthed turnaround time for all other vessel (other that i and j ) remains the same in both schedules. Under S the total weigthed turn around time fo vessels i and j is (t + pi )wi + (t + pi + pj )wj , while under S it is (t + pj )wj + (t + pj + pi )wi . It is easy to verify that if pi /wi > pj /wj , the sum of the two weigthed turnaround times under S is strictly less than under S . These contradicts the optimality under S .

Numeric Example

a)

b)

Table 4: a) Original data set. b) Sorted data following step 1 of SRWPT algorithm. .

a)

b)

Figure 7: a) Step 2 of SRWPT algorithm applied to data in table 4a. b) Step 3 of SWRPT algorithm. .

Figure 8: Time-space diagram for nal solution by SRWPT algorithm. Vessels and time period are colored according to arrival time groups (batches).

Problem 2
The berthing plan in Figure 9 is obtained by solving the BAP with maximum number of quay cranes assigned to all vessels. I limited the berthing sections to 20 such that the maximum number of cranes is not violated. The BAP was formulated in AMPL and solved with Gurobi solver available at neos-server.org/neos/.

Figure 9: Time-space diagram for berthing plan of problem 2.

Problem 3
Notation
1. i, j : Minimum amount of time by which completion of vessels i and j , must occur. 2. ij = 1 if cj > ci 0 otherwise 7

Model
Lines (1) through (8) are kept intact as in Slide 58. Constraint number (9) in modied to acount for the new binary variable ij . The following constraints are added. cj ci cj ci > < ij ij T ij i, j, i = j i, j, i = j (10) (11)

Where inequality (10) states that for all pairs of vessels, take vessel i and vessel j , the completion time of vessel i must be at least ij time unit prior to completion time of vessel j whenever i,j = 1, i.e vessel j is completed strictly after vessel i. Inequality (11) forces i,j to take value of 1 whenever cj > ci , i.e. completion time of vessel j occurs strictly after completion of vessel i.

Problem 4
Container are transferred and stored in the container terminal. These two functions can be measured in terms of container handling and space utilization metrics. It is known that these two functions can pull the terminal resources in opposite directions. As an example: ground space utilization and container accessibility are inversely related [4]. Physical characteristics of a port can also impose constraints to terminal operators. Additionally the geographic location and type of trac in a port can result in greater demand towards a particular type of resource. It is foremost important to dene clear service levels and productivity goals in order to invest resources in a productive matter as opposed to reaching for conicting objectives. Given that an increase in throughput is expected, the port might face a capacity constraint issue. For this reason priority should be placed on existing or potential bottle necks in operational capacity. Port resources can be analyzed separately taking into account 5 major components: berth capacity, crane capacity, yard capacity, gate capacity and crew capacity [reference Slide 49]. Conditional on the type of data available several metrics can be utilized or estimated to rank these ve resources by their available capacity. It is suggested to nd an estimate of moves per year (TEU/year or lifts/year) that characterized each component. These number can be directly compared to the predicted yearly throughput. In absence of such a metric, a potential indicator of limited capacity can be current utilization rate (Moves per machine hour) versus average idle times. Resources currently acting as bottle neck will exhibit higher utilization rates where as operations preceding these bottlenecks will have greater idle time due to waiting. Based on what resource or resources are limiting several alternatives can be explored. Limiting Resource Berth area and/or Quay crane Recomended Action Increase the number of quay cranes Enable two container lifts Enable dual command cycles in quay crane (small vessel moored in parallel)

Yard area: Extensive handling times

Reduced area

Spread container over a larger area (reduce stacking height) Increase number of yard trucks Enable dual command truck trips Increase number of gantry cranes Stack container higher Increase berthing priority for loading(export) vessels Enforce penalties on carries to reduce days container sits in the yard

Table 5: Recomended action to take for specic port resource limitations Berthing and crane operations are close related and hence identifying either of these components as a bottle neck will require similar actions. Investing resources in any of these operations will impact vessel turnaround times. When we consider yard operations conicting objectives might come into play. In the case that yard area is identied as a bottleneck resource, then it must be determined whether space or handling times are most limiting. Yard space constraints can be most apparent to the naked eye. Many ports struggle to increase yard space due to geographical, infrastructure and environmental constraints. At a cost of increased handling times, containers can be stacked higher in the yard in order to save yard space. If additionally yard and gantry cranes are scarce, additional measures can be taken to reduce the time a container sits in the yard 8

(presented in table 5). Apart from yard space, handling times can be restrictive given the amount of yard trucks and number and type of gantry cranes available. Idle times and utilization rates can expose if any of these two resources is particularly limited. If trucks represent a bottleneck, dual command cycles can be implemented to increase their productivity. In the case where gantry crane capacity exceeds truck capacity by a signicant amount, it could be feasible to move two containers per truck trip. On the other hand gantry cranes operations can be aected by many factors. Firstly, the spreading of container in the yard can reduce the number of lifts necessary to access a particular container. Comparing the number of lifts of gantry cranes per truck trip will reveal such a behavior. Secondly, dierent gantry cranes congurations can be more exible than others (rubber-tired versus rail-mounted). If gantry cranes exhibit signicant idle times, even though they are identied as a bottleneck resource, then adding exibility would be the solution.

References
[1] Yongpei Guan and Raymond K Cheung. The berth allocation problem: models and solution methods. OR Spectrum, 26(1):7592, 2004. [2] Yongpei Guan, Wen-Qiang Xiao, Raymond K Cheung, and Chung-Lun Li. A multiprocessor task scheduling model for berth allocation: heuristic and worst-case analysis. Operations Research Letters, 30(5):343 350, 2002. [3] Hans-Otto Gunther and Kap-Hwan Kim. Container terminals and terminal operations. OR Spectrum, 28(4):437445, 2006. [4] Hanh Dam Le-Grin and Melissa Murphy. Container terminal productivity: Experiences at the ports of los angeles and long beach. Report, University of Southern California, 2006. [5] Michael Pinedo. Scheduling: theory, algorithms, and systems. Springer, 2012. [6] Linus Schrage. Letter to the editora proof of the optimality of the shortest remaining processing time discipline. Operations Research, 16(3):687690, 1968.

Das könnte Ihnen auch gefallen