Professional Documents
Culture Documents
Rough Intervals
Ammar.E.E(1), Emsimir.A.A(2)
Abstract. The goal of this paper is to focus on the solution of fully rough integer linear programming (FRILP)
problems, where all decision parameters and decision variables in the constraints and the objective functions
are rough intervals. In order to solve this problem, we will construct two LP problems with interval coefficients
and variables. One of these problems is an LP where all of its coefficients are upper approximations of rough
intervals and the other is an LP where all of its coefficients are lower approximations of rough intervals. And
then to four crisp LPs. Via these two LPs, two newly solutions (completely and rather satisfactory) are defined.
Some examples are given to demonstrate the results.
Key words: Integer linear programming, Rough set theory, Full rough interval coefficients and variables,
Upper approximation, Lower approximation, Optimal solution, Crisp coefficients.
1. Introduction:
Linear programming (LP) is one of the most popular models used in decision making and optimization
problems. Lots of research, studies and applications of LP models have been reported in numerous books,
monographs, articles and chapters in books, for instance see [3,5]. About rough set theory (RST) was initiated
by Pawlak in (1982) as a method for ambiguity management [10]. RST approach has fundamental importance
in the fields of pattern recognition, data mining, artificial intelligence, machine learning and medical
applications [8]. More widely applied of integer programming can be used to appropriately describe the decision
problems with the management and effective use of resources in engineering technology, business management
and other numerous fields [15]. For a vague concept R, a lower approximation is contained of all objects which
surely belong to the concept R and an upper approximation is contained of all objects which possibly belong to
the concept R. In other words, the lower approximation of the concept is the union of all elementary concepts
which are included in it, whereas the upper approximation is the union of all elementary concepts which have
nonempty intersection with the concept [11]. Ammar and Khalifa in (2014) applied a new method named,
separation method for solving Rough Interval Multi Objective Transportation Problems (RIMOTP), where
transportation cost, supply and demand are rough intervals [1]. Also, discussed the separation method as an
(1)
Prof. of Pure Mathematics and O. R. Faculty of Science - Tanta University, Egypt
(2)
Department of Mathematics, Faculty of science, Tanta University, Egypt.
important tool for the decision makers when they are handling various types of logistic problems having rough
interval parameters of transportation problems. Osman et al, in (2016) presented a solution approach for
RIMOTP. The concept of solving conventional interval programming combined with fuzzy programming is
used to build the solution approach for RIMOTP [9]. Pandian et al in (2016) considered that transportation
problem has all or some parameters as rough integer intervals [12]. Also, proposed a new method named, a
slice-sum method to solve Rough Integer Interval Transportation Problem (RIITP), where transportation cost,
supply and demand are rough integer intervals. In this paper, the focus of our study is improvement a method
to find optimal solutions of integer linear programming problems with rough interval coefficients and rough
interval variables and the optimal values of decision rough variables and rough objective function are rough
intervals. The elements of the paper are organized as follows. In Section 2, some basic about the preliminaries
of RIs are presented. In Section 3, an LP problem with interval coefficients is discussed. Section 4 proposes a
solution method for LP with rough interval coefficients and variable.
2. Basic preliminaries:
The transformation of the linear programming problem decision parameters and decision variables into upper
and lower approximation is usually hard work for many cases, but conversion process needs the following
definitions to be known [7]:
Definition 2.1. The qualitative value A is called a rough interval when one assign two closed intervals A∗ and A∗
on R to it A∗ ⊆ A∗ .Moreover,
(i) If x ∈ A∗ then A surely takes x (denoted by x ∈ A).
(ii) If x ∈ A∗ then A possibly takes x.
(iii) If x ∉ A∗ then A possibly takes x.
where A∗ and A∗ are called the lower approximation interval (LAI) and the upper approximation interval
(UAI) of A, respectively. Moreover, A is denoted by A = ( A∗ , A∗ ). Also, the interval A∗ and A∗ are not the
complement of each other.
Note: We can symbolize to A∗ = [aLL, aUL ] and A∗ = [aLU , aUU ]
Example 2.1. [13] Consider A be the concept of “Hot” as the qualitative value defined on a temperature
variable x in R. One may consider:
A =(A∗ , A∗ ) = ([25∘ , 45∘ ], [15∘ , 65∘ ])
Then, by Definition 2.1 the temperature variable x surely takes a value between 25◦C and 45◦C. Similarly, x
possibly takes a value between 15◦C and 65◦C. Note that, common sense and general knowledge can help
define A∗ and A∗ . For instance:
– The temperature of the human body is about 37° 𝐶 . Any warmer temperature won’t be considered “Cold”.
– In average, a human hand cannot hold an object, whose temperature is over 65◦C, because it is “Hot”.
– If something is colder than the environment (let us say 15° 𝐶 ), it cannot be considered “Warm” anymore.
Definition 2.2. The arithmetic operations on RIs are depending on interval arithmetic. We will state some of
these arithmetic operations as follows [13]:
Let A = ([ aLL , aUL ], [aLU , aUU ]) and B = ([ bLL , bUL ], [bLU , bUU ]) be two RIs if A, B ≥ 0. Then we have:
[Addition:] A+B = ([aLL + bLL , aUL + bUL ], [aLU + bLU , aUU + bUU ]).
[Subtraction:] A – B = ([aLL − bUL , aUL − bLL ], [aLU − bUU , aUU − bLU ]).
[Multiply:] A*B = ([aLL bLL , aUL bUL ], [aLU bLU , aUU bUU]).
[Negation:] - A = ([−aUL , −aLL ] , [−aUU , −aLU ]) .
[Intersection:] A B = ([max{aLL , bLL } , min{aUL , bUL }], [max{aLU , bLU } , min{aUU , bUU }]).
[Union:] A B = ([min{aLL , bLL } , max{aUL , bUL }], [min{aLU , bLU } , max{aUU, bUU}]).
3. Integer linear programming with interval coefficients
The method of more relevance for solving linear programming problems with interval coefficients (LPIC) first
time was proposed by T. Shaocheng [14] in 1994. After that, Chinneck and Ramadan completed Choacheng's
studies [4]. They transformed the original interval LP into two LPs with crisp coefficients. Then linear
programming with rough interval coefficients (LPRIC) by A. Hamzehee, M.A. Yaghoobi and M. Mashinchi [7]
in 2014. In this part, we consider an interval LP problem as follows:
Where for all i , j : [cjL , cjU ], [aLij , aUij ]and [bLi , bUi ] are closed intervals on real numbers R and L , U indicate
to the minimum and the supreme in the interval respectively.
Now, by a previous study to Chinneck and Ramadan [4] we can depend some definitions and theorems for
maximization problem (3,1)-(3,3). Consider the inequality in (3,2) as follows:
Definition 3.1. [7] For inequality (3,2), if there exists one extreme inequality such that its solution set is the
same as SU (SL ) then it is called the maximum value range (minimum value range) inequality.
Theorem 3.1. [7] Consider inequality (3,2), where 𝑥𝑗 ≥ 0 ∀ 𝑗 = 1,2, … … , 𝑛. Then, ∑nj=1 aUij xj ≤ bLi and
∑nj=1 aLij xj ≤ bU
i are the minimum value range and maximum value range inequalities, respectively.
Theorem 3.2. [7] Consider LPIC problem (3,1)-(3,3). Then, for any given feasible solution x ∈ Rn , we have:
Definition 3.2. [7] For a given feasible solution x ∈ Rn of LPIC Problem (3,1)-(3,3), the value ∑nj=1 cjU xj
(∑nj=1 cjL xj ) is called the most suitable value (the least suitable value) of the objective function.
From theorems (3.1 , 3.2) and Definitions (3.1 , 3.2) we can find the best and worst optimal solutions of
the LPIC Problem (3,1)-(3,3). And it by transforming the original LPIC Problem (3,1)-(3,3) into two classical
LPs. We call them LP1 and LP2.
1. The best optimal solution is found by solving:
S. t ∑nj=1 aLij xj ≤ bU
i
xj ≥ 0 and integer j = 1, … … , m , i = 1, … … . , n
2. The worst optimal solution is found by solving:
xj ≥ 0 and integer j = 1, … … , m , i = 1, … … . , n
Summary, LP1 and LP2 uses the greatest and the smallest suitable value of the objective function and the
maximum (minimum) value range inequalities of problem (3,1)-(3,3).
Note that, there are 3 possible outcomes for LPIC problem (3,1)-(3,3) as follows:
1) If LP1 and LP2 have optimal solutions, then LPIC problem (3,1)-(3,3) has a finite bounded optimal
range.
2) If LP2 is unbounded then LPIC problem (3,1)-(3,3) is unbounded.
3) If LP1 is infeasible then LPIC problem (3,1)-(3,3) is infeasible.
Example 3.1. A company produces two sorts of products, A and B, and which need by three departments 1, 2
and 3 to produce them. Each product takes a time to be produced by the three departments, the following table
shows us the details.
product The required production time in department (hr/unit) Seasonal
1 2 3 profit ($/unit)
A [3,5] [2.5,3.5] [1,2] [1.5,3]
B [2,3.5] [1.5,3] [3.5,5] [2.5,3.5]
The time available [550,700] [380,500] [200,300]
x1 , x2 ≥ 0 and integer
According to Section 3. We can be obtained the optimal range of ILPIC for example 3.1. by solving two
classical LPs as follows:
The best optimal solution, ILP U : = 𝐌𝐚𝐱 3x1 + 3.5x2
𝐒. 𝐭 3x1 + 2x2 ≤ 700
x1 , x2 ≥ 0 and integer
x1 , x2 ≥ 0 and integer
The experts' production would like to allocate the production capacity so that not only finds the optimal product
mixture but also uses the three plans together. In order to solve this problem, first the parameters should be
determined. Since the parameters are given by three plans, one way to involve all plans is representing the
parameters by rough interval. In table 4, the upper and lower approximations of rough interval coefficients are
computed based on the union and intersection of the plans. In fact, taking the intersection means consensus on
all the plans, and, taking the union means respecting any of the three plans.
Now let x1 and x2 denoted the respective amounts of products 1 and 2 which should be produced. Then, the
model of this optimization problem can be formulated as follows:
([2.5,3], [1.5,3.5]) ⨂( [ xLL UL LU UU
1 , x1 ], [ x1 , x1 ] )⨁
𝐈𝐋𝐏𝐅𝐑𝐈 = ([ILP LAI
], [ILP UAI
]) = Max { } (4,1)
([2.5,3], [2,3.5] ) ⨂ ( [ xLL
2 , x2 ], [ x2 , x2 ])
UL LU UU
Note, in this study will be limited the variables which are sign-restricted as xiLL , xiLU , xiUL , xiUU ≥ 0
The general formula of the integer linear programming problem with full rough interval(ILPFRI) as:
([ILPLAI ], [ILPUAI ]) = 𝐌𝐚𝐱 {∑nj=1( [cjLL , cjUL ] , [cjLU , cjUU ]) ⨂ ( [ xjLL , xjUL ], [ xjLU , xjUU ])} (4,1,1)
UL LU UU LL UL LU UU
∑m n LL
j=1 ∑i=1([a ij , a ij ] , [a ij , a ij ]) ⨂ ([ xj , xj ] , [ xj , xj ])
𝐒. 𝐭 { UL LU UU
} (4,1,2)
≤ ([bLL
i , bi ] , [bi , bi ])
( [ xjLL , xjUL ], [ xjLU , xjUU ]) (i = 1, … … , m; j = 1, … … , n) are rough interval coefficients and variables of
the objective function and the constraints. Also, let x ∈ Rn denote the vector of all decision variables.
Remark: According to RI prop1erties introduced in Section (2), we have
[cjLL , cjUL ] ⊆ [cjLU , cjUU ] → cjLU ≤ cjLL ≤ cjUL ≤ cjUU ,
UL LU UU LU UL UU
[bLL LL
j , bj ] ⊆ [bj , bj ] → bj ≤ bj ≤ bj ≤ bj ,
1) S LU = {x ∈ Rn | ∑nj=1 aUU LU LU
ij xij ≤ bi , i = 1, … … m; xj ≥ 0, j = 1, … … n: and integer}
2) S LL = {x ∈ Rn | ∑nj=1 aUL LL LL
ij xij ≤ bi , i = 1, … … m; xj ≥ 0, j = 1, … … n: and integer}
UL UL
3) S UL = {x ∈ Rn | ∑nj=1 aLL
ij xij ≤ bi , i = 1, … … m; xj ≥ 0, j = 1, … … n: and integer}
4) S UU = {x ∈ Rn | ∑nj=1 aLU UU UU
ij xij ≤ bi , i = 1, … … m; xj ≥ 0, j = 1, … … n: and integer}
Note: for the sets SLU , SLL , SUL , and SUU we have: S LU ⊆ S LL ⊆ S UL ⊆ S UU
Definition 4.1.2. [7],[2] consider all of the corresponding ILPFRI problems and LP of problem (4,1,1)-(4,1,3).
(a) The interval [ILPLL , ILP UL ]([ILPLU , ILPUU ]) is called the surely (possibly) optimal range symbolized
[ILPLAI ] ([ILPUAI ]) of problem (4,1,1)-(4,1,3), if the optimal range of each (ILPFRI) is a superset (subset)
of [ILPLL , ILPUL ]([ILPLU , ILPUU ]).
(b) Let [ILPLL , ILPUL ]([ILP LU , ILPUU ]) be surely optimal (possibly) optimal range of the problem (4,1,1)-
(4,1,3). Then the rough interval ([ILPLL , ILPUL ], [ILPLU , ILPUU ])is called the rough optimal range of
problem (4,1,1)-(4,1,3), also any point, optimal value belongs to [ILPLL , ILPUL ], ([ILPLU , ILPUU ]) is
called a completely (rather) satisfactory solution of the problem (4,1,1)-(4,1,3).
(c) A solution x is surely-feasible, iff it belongs to the lower approximation of the feasible set.
(d) A solution x is possibly -feasible, iff it belongs to the upper approximation of the feasible set.
(e) A solution x is surely-not feasible iff it does not belong to the upper approximation of the feasible set.
4.2. Solution procedures: In order to solve Problem (4,1,1)-(4,1,3), follow the following steps.
Step 1: Find the possibly optimal range [ILPLU , ILPUU ] by solving the following ILPFI problem via section 3
Step 2: Find the surly optimal range [ILPLL , ILPUL ] by solving the following ILPFI problem via section 3.
𝐒. 𝐭 { ∑nj=1 aLU UU
ij xj ≤ bUU
i } (4,2,8)
𝐒. 𝐭 { ∑nj=1 aUU LU LU
ij xj ≤ bi } (4,2,11)
Step 4. The optimal range of ILPFI problem (4,2,7)-(4,2,9) can be obtained by solving two classical LPs as
follows:
𝐒. 𝐭 { ∑nj=1 aUL 𝐿𝐿 LL
ij xj ≤ bi } (4,2,14)
In the ILPFI Problem (ILPUAI ) is transformed to LP problems ILPLU and ILP UU , where their feasible sets
are SLU and SUU , respectively.
In the ILPFI Problem (ILPLAI ) is transformed to LP problems ILPLL and ILPUL , where their feasible sets are
S LL and S UL , respectively.
and rough integer intervals variabls. and roug integer intervals variabls.
and rough integer intervals variabls. and rough integer intervals variabls.
Optimal values and optimal solutions of Example 4.2.1. as following results:
([1,2], [1,3]⨂([ x1LL , x1UL ], [ x1LU , x1UU ])⨁ ([2,3], [1,4]) ⨂([ x2LL , x2UL ], [ x2LU , x2UU ])⨁
([1,3], [1,4]) ⨂([ x3LL , x3UL ], [ x3LU , x3UU ])⨁ ([1,3], [1,5]) ⨂([ x4LL , x4UL ], [ x4LU , x4UU ])
𝐒. 𝐭
([2,3], [1,5])⨂ ([ x5LL , x5UL ], [ x5LU , x5UU ])⨁([1,2], [1,4])⨂([ x6LL , x6UL ], [ x6LU , x6UU ])
{ ≤ ([30,35], [25,60]) }
([2,3], [1,5]) ⨂([ x1LL , x1UL ], [ x1LU , x1UU ])⨁([1,2], [1,3]) ⨂( [ x2LL , x2UL ], [ x2LU , x2UU ])
{ }
≤ ([8,16], [5,30])
([1,3], [1,5]) ⨂( [ x3LL , x3UL ], [ x3LU , x3UU ])⨁([1,2], [1,4]) ⨂( [ x4LL , x4UL ], [ x4LU , x4UU ]
{ }
≤ ([8,10], [5,15])
([1,2], [1,3]) ⨂([ x5LL , x5UL ], [ x5LU , x5UU ])⨁([2,3], [1,4]) ⨂([ x6LL , x6UL ], [ x6UL , x6UU ])
{ }
≤ ([5,15], [4,20])
{xiLL , xiLU , xiUL , xiUU ≥ 0 , where i = 1,2,3,4,5,6 and rough integer intervals variables }
In order to solve problem (e.g. 4.2.2.), we have to solve two ILPFI problems as follows:
UAI [12,30]⨂[xLU
1 , x1 ]⨁ [4,20]⨂[ x2 , x2 ] ⨁ [1,1] ⨂[ x3 , x3 ]⨁
UU LU UU LU UU
ILP = 𝐦𝐚𝐱 { }
[1,6]⨂[ xLU UU LU UU LU UU
4 , x4 ]⨁ [1,5]⨂[ x5 , x5 ] ⨁ [1,1]⨂[ x6 , x6 ]
𝐋𝐀𝐈
[18,25]⨂[x1LL , x1UL ]⨁[12,18]⨂[ x2LL , x2UL ] ⨁[1,1] ⨂[ x3LL , x3UL ]⨁
𝐈𝐋𝐏 = 𝐌𝐚𝐱 { }
[2,3]⨂[ x4LL , x4UL ] ⨁ [1,3]⨂[ x5LL , x5UL ]⨁[1,1] ⨂ [ x6LL , x6UL ]
x1UU + x2UU ≤ 30
x3UU + x4UU ≤ 15
x5UU + x6UU ≤ 20
5x1LU + 3x2LU ≤ 5
5x3LU + 4x4LU ≤ 5
3x5LU + 4x6LU ≤ 4
3x1LL + 2x2LL ≤ 8
3x3LL + 2x4LL ≤ 8
2x5LL + 3x6LL ≤ 5
2x1UL + x2UL ≤ 16
x3UL + x4UL ≤ 10
x5UL + 2x6UL ≤ 15
ILP UU = 1065, where x1UU = 30, x2UU = 0, x3UU = 0, x4UU = 15, x5UU = 15 , x6UU = 0
ILP UL = 328, where x1UL = 8, x2UL = 0, x3UL = 0, x4UL = 10, x5UL = 8.500, x6UL = 0
ILP LL = 58.500, where x1LL = 2.666, x2LL = 0, x3LL = 0, x4LL = 4, x5LL = 2.500, x6LL = 0
ILP LU = 18.25, where x1LU = 1, x2LU = 0, x3LU = 0, x4LU = 1.25, x5LU = 1.333, x6LU = 0
When we apply the branch and bound Example 4.2.2. to find integer optimal solutions
ILP UU = 1065, where x1UU = 30, x2UU = 0, x3UU = 0, x4UU = 15, x5UU = 15 , x6UU = 0
ILP UL = 328, where x1UL = 8, x2UL = 0, x3UL = 0, x4UL = 10, x5UL = 8, x6UL = 0