Multi-Objective Optimization of Pilots’ FFS Recurrent Training Problem ()
1. Introduction
With the rapid development of the air transportation, aviation safety has become an important issue in Chinese airline companies. FFS (full flight simulator) recurrent training is an important method to ensure airmen’s flight safety. Pilots need to take FFS (full flight simulator) recurrent training every half a year in order to keep the pilot qualification in Chinese airways [1]. It allows two airmen to be trained in an FFS at the same time and the two men had better be a captain and a copilot for the best training effect. Moreover, an airman’s fitting month for the training is from the fifth month to the seventh month after his last training. And the sixth month after his last training is his optimal training month because it is a resource waste for him to be trained in the fifth month and he will not get the best training effect when the training is implemented in the seventh month. To make the plan, we should take these two objectives into account. So PFRT problem is a multi-objective problem.
Pan et al. studied the problem and present a decomposition algorithm [1]. However, the algorithm is not a polynomial time algorithm. Liu [2] described the problem with multiple resource constraints and a Genetic Algorithm was designed to solve it. The algorithm is an approximation algorithm which cannot get the optimal results in general. PFRT problem with resource constraints is an NP-hard problem in theory [3].
Multi-objective programming involves recognition that the decision maker is responding to multiple objectives. Generally, the objectives are conflicting, so that not all objectives can simultaneously arrive at their optimal levels. An assumed utility function is used to choose appropriate solutions. Several fundamentally different utility function form s have been used in multi-objective models. These forms may be divided into three classes: lexicographic, multi-attribute utility and unknown utility. The lexicographic utility function specification assumes that the decision maker has a strictly ordered preemptive preference system among objectives with fixed target levels. Multi-attribute utility approaches allow tradeoffs between objectives in the attainment of maximum utility. The third utility approach involves an unknown utility function assumption. Here the entire Pareto efficient (nondominated) solution set is generated so that every solution is reported wherein one of the multiple objectives is as satisfied as it possibly can be without making any other objective worse off [4]. Many techniques have been developed to solve the multi-objective programming, such as tabu search [5,6], simulated annealing [7], and foremost evolutionary algorithms [8,9]. And other important publications on metaheuristics for multi-objective optimization include the work of Gandibleux et al. [10,11].
PFRT problem is analyzed in this study. In part 2, we describe the problem with mathematical programming models. We generate graphs with which we transform the problem into a longest-route problem on weighted paths. A polynomial time algorithm is developed to solve PFRT problem. In part 3, a case study is given. In last part, we summarize the work.
2. Models and Methodology
2.1. Mathematical Programming Models
The two objectives of PFRT problem are conflicting and the problem can be described as programming (a) and (b).
: the number of the captains whose optimal training month is.
: the number of the copilots whose optimal training month is.
: the number of the captains whose optimal training month is i and who will attend the training in month.
: the number of the captains whose optimal training month is i and who will attend the training in month.
We describe the PFRT problem as the following multi-objective mathematical programming model:
Programming (a):
(1)
(2)
where
Subject to
; (3)
; (4)
; (5)
; (6)
; (7)
; (8)
; (9)
are positive integers and are all nonnegative integers.
The first objective of the programming is to maximize the number of captains who are matched with copilots. The other objective is to minimize the number of the pilots who will not be trained in their optimal training months. The Formulas (3)-(9) assure all pilots will be trained.
The general model for the problem can be described as:
Programming (b):
(10)
(11)
Subject to
; (12)
; (13)
; (14)
; (15)
; (16)
; (17)
(18)
are all positive integer and are all nonnegative integer.
The key work of this research is to find the non-inferior solutions. We will find an initial non-inferior solution by two steps. Firstly, we consider the programming with a single objective as follows. Secondly we will convert get other non-inferior solution by a graphic method in part 2.2.
Programming (c):
(19)
Subject to the constraints (12)-(18).
Here let denote the optimal value for this programming. Then we solve the following programming.
Programming (d):
(20)
Subject to the constraints (12)-(18) and
(21)
Let denote the optimal value for this programming and the optimal solutions is the initialization for the following graph models. We will give a graphic method to find all the other non-inferior solutions sequentially.
2.2. Graph Models
We draw a bipartite graph in Figure 1. The vertex denotes the captains whose optimal training month is i. And the vertex denotes the copilots whose optimal training month is i. Here denotes the module of, i.e. the number of the captains in and denotes the module of and denotes the number of the pilots trained in their optimal training month i. There is an edge weighted by ended by and when there exist pairs of captains and copilots we match between
and.
In graph there may be an edge ended by and where. In other words, there are a captain and a copilot matched where (or). Suppose (or) and () are matched in graph, we can change the training plan by matching to and to (or to and to). Then we suppose there are no ended by and where in the following part of the study.
Let and. What we will do in next parts of this study is to find the largest Q when is decreased by k where k = 1, 2, ....
We assume there are three possible nexus among, , and in graph G shown in Figure 2.
We consider two types to change the nexus among, , and in as follows. One is to break a pair on edge in case 1 in Fiugre 3 or to break a pair on edge and a pair on edge and rebuilt a new pair between and in case 2 and case 3. It is denoted by. The other is to break a pair on edge in case 2 in Figure 3 or to break a pair on edge and a pair on edge and rebuilt a new pair between and in case 1 and 3. It is denoted by.
Then we can conclude that we can find the solution of the problem with by or for all where.
We generate another two weighted graphs and in Figure 3 from the recurrent training graph in Figure 1.
Here denotes month in Figure 3 and when there exist edge in graph, or else,. And when there exists edge, or else,. Let and .
Then we have the follow theorem.
Theorem 1:
Let be the graph with and, we got a non-inferior solution in graph with the largest where when we change the nexus among vertices in graph G by:
(a) for all when and
(b) for all when .
Proof:
In case (a) we find that the nexus among the vertices, , and is case 1 in Figure 3 when. When we change the nexus among, , and by, reduces one and the sum of and increases 1, which equals to, as an result. The nexus among, , and is case 2 or case 3 when, and when we change the nexus by, we will get a new pair between and where and the sum of and is decreased by one and the sum of, and is reduced by one as well, corresponding to. In conclusion, when we change the nexus among, , and by we increase the sum of and by and increase by in other words. Then to find the largest where in graph is to find the longest path in graph when. Case (b) can be proved similarly.
2.3. Algorithm
In this part an algorithm as follows is designed to solve PFRT problem.
Step 1: Initialize: t = 0; for all 1≤ i ≤ n, ai = the number of the captains whose optimal training month is i and bi = the number of the copilots whose optimal training month is i.
Step 2: Find the optimal solutions of programming (c) and (d). denotes the graph with and = the maximum quantity of pilots who will be trained in their optimal training month.
Step 3: Generate the weighted graphs and
Figure 2. Nexus among ui, vi, ui+1 and vi+1 in graph G.
from the graph. Find the longest path in the graphs
and. and
. If, go to step 4.
Else go to step 5.
Step 4: Change the nexus among the vertice, , and by for. Go to step 6.
Step 5: Change the nexus among the vertices, , and by for. Go to step 6.
Step 6: If the non-inferior solution we got in the previous step is not the final solution, let, , go to step 3, else go to step 7.
Step 7: Output the final solution, stop.
3. A Case Study
There are 314 captains and 313 copilots in an airline company. Their FFS training demand is shown in Table 1 and is achieved in Table 2 and Figure 4 where. We paired up 313 pairs of captains and copilots. 609 pilots will attend FFS training in their optimal training months. There are 18 pilots who have to attend the training in the previous month or the succeeding month of their optimal training months.
Table 1. FFS training demand for all pilots
Figure 4. The initial non-inferior solution with P0 = 313 in graph G0.
And there is one captain in month 5 who cannot be trained with copilots because the sum of captains is more than that of copilots.
We draw a couple of new graphs and as follows.
We can find that the longest route in graph is where L = 1 and the longest route in graph is where L = 4. We change the nexus among, , and by. Then will reduce one and will reduce one as well. In the same time will increase one. When we change the nexus among, , and by as well where and rematch the pairs of unmatched captains and copilots who have the same optimal training month, we will get a new graph with the largest where and. Then we get a new non-inferior solution described in Table 3.
Now we paired up 312 pairs of captains and copilots. 613 pilots will attend FFS training in their optimal training months. There are 14 pilots who have to attend the training in the previous month or the succeeding month of their optimal training months. And there are 3 pilots in month 5 and month 1 who cannot be matched with other pilots.
Repeating above steps, we will get the following solutions shown in Tables 4-6.
Now we get all non-inferior solutions.
4. Conclusions
The air transportation developed rapidly in China and it becomes crucially important for airline companies to
training pilots so as to ensure the aviation safety.
PFRT problem is a multi-objective pilots training problem. Mathematic models and graph models about it are built. According to the characteristics of the problem, we convert it into a longest-route problem with weighted paths and design an algorithm to solve it.
This method can effectively generate pilots’ FFS training plans with two kinds of personnel.
Due to the method in this study cannot solve the similar problem involving more than two kinds of personnel yet, further research should be done on it.
5. Acknowledgements
I thank members of the Research Center of Overall Planning Safety and Security Management, Institute of Policy and Management, Chinese Academy of Sciences, for their help in this study. We are also very grateful to the reviewers for their invaluable suggestions to improve the quality of the manuscript.