1. Introduction
In actual industrial production, with the increasingly fierce competition among enterprises, jobs with similar characteristics are divided into the same group for processing to adapt to the efficient production rhythm (Potts et al. [
1], Pei et al. [
2], Lee et al. [
3]). The group technology (expressed by
) arises, corresponding to the group technology scheduling problem in scheduling theory. The essence of enterprise production is a huge consumption of energy, for example, in the case of the input of steel; such cases can be collectively referred to as resources. The input of resources can effectively reduce the processing process of the job and then shorten the processing time; that is, the greater the input, the shorter the processing time. The corresponding cost input also increases accordingly. Therefore, striking a balance between resource consumption and cost growth has become one of the most popular research topics. In the processing operation, in addition to resources affecting the processing time, machine/manual learning also shorten its duration, which corresponds to the scheduling problem with learning effect. On the other hand, when the customer submits an order, there is a delivery date, also known as the due date, which is fined for advance/delay to the customer. Therefore, each group of jobs has a common due date of reasonable arrangement of job production so as to reduce fines, which also plays an important role in scheduling. According to the above, in the group environment of resource allocation, learning effect and common due date reasonably determine the sequences of the job and the group so as to reduce costs, which is main goal of this study. The relevant research results of the above production environment (resource allocation, learning effect, common due date, and groups) will be elaborated in the future.
In reality, the processing times of jobs are alterable according to resource allocation (Shabtay and Steiner [
4], Wang and Cheng [
5], Kayvanfar et al. [
6], Tian [
7]) and learning effect (Wu et al. [
8], Azzouz et al. [
9], Zhao [
10], Liu et al. [
11]). The learning effect is the efficiency gained through repetitive production. Yin [
12] studied single-machine resource allocation scheduling with the learning effect. Under common due window assignment, the researchers showed that some non-regular objectives are polynomially solvable. Zhao [
13] studied the problems of no-wait flow shop scheduling with learning effect and convex resource allocation. Wang et al. [
14] investigated single-machine scheduling with truncated learning effect and resource allocation. Jiang et al. [
15] considered Seru production scheduling problems with resource allocation. For the total processing cost plus scheduling measure minimizations, they proved that the problems are polynomially solvable. Wang and Wang [
16] addressed the single-machine scheduling with convex resource allocation and a time-dependent learning effect. For three versions stemming from the total resource consumption cost and the scheduling cost, they proposed some solution algorithms.
In order to achieve higher productivity, the
is often used in production and processing. For example, Shabtay et al. [
17] studied these two types of resource allocation functions separately. Zhu et al. [
18] studied the single-machine scheduling with learning effect, resource allocation, and
. Li et al. [
19] considered a single-machine scheduling problem involving the due date assignment under
environment. They provided an
time complexity for the above three due date assignment methods. Lu et al. [
20] investigated a single-machine scheduling problem with decreasing time-dependent processing times and group technology assumption. For makespan minimization, they showed that the problem can be solved in polynomial time. Sun et al. [
21] introduced single-machine group scheduling with learning effect and resource allocation. Chen et al. [
22] considered a single-machine group scheduling problem with due date assignment and resource allocation. Liang et al. [
23] dealt with a single-machine resource allocation scheduling problem with deteriorating jobs and
. Yan et al. [
24] studied a single-machine problem with resource allocation and deteriorating effect in group technology environment. Liu and Wang [
25] and Wang and Liu [
26] considered single-machine
scheduling with resource allocation and due date assignment.
In view of the importance of due date assignment (Shabtay [
27], Shabtay et al. [
28], Cheng [
29], Li et al. [
30], Zhang et al. [
31], Lv and Wang [
32]), most of the above literature studies the special polynomial-time solvable cases of
problems. In this paper, we integrate the case where processing time follows a convex function of resource allocation under common due date allocation as well as learning effects, i.e., the job number of each group is different. We conduct a specific study on the objective of minimizing the weighted sum of earliness, tardiness, common due date, resource consumption, and makespan. For the general case of the problem, we propose the heuristic, simulated annealing, and branch-and-bound algorithms. The rest of this paper is organized as follows. In
Section 2, the problem statement is presented.
Section 3 offers basic properties. A special case is discussed in
Section 4.
Section 5 presents the specific solving algorithms of the problem. In
Section 6, computational experiments are proposed. Finally, the conclusions are given in
Section 7.
2. Problem Statement
There are n independent jobs grouped into r groups (i.e., ) to be processed by a single machine. Each has an independent setup time , where the number of jobs allocated to is (i.e., ).
If the job
(
,
) is scheduled at the
th (
) position of group
(
), the actual processing time of job
is
where
is the amount of resource of
in
,
is a positive parameter which represents the workload of
in
,
denotes the learning index of the job position,
denotes the group learning index, and
is a given constant.
We define the earliness and tardiness of the job
as
and
for
,
, where
is the common due date of group
and
is the completion time of
. The objective is to find the optimal group sequence
∈
, where
is set of all group sequences, the optimal internal job sequence
∈
within
for
, where
is set of all internal job sequences, the optimal set of due dates
, and the optimal resource allocation matrix
for
,
, which together minimize an objective function that includes penalties due to earliness, tardiness, due date assignment, resource consumption, and makespan as given below:
where
,
,
,
, and
represent the per unit cost of the due date cost, earliness penalties, tardiness penalties, and processing time,
is the cost of resources allocated to
. Using the three-parameter representation, the problem can be denoted as
where
denotes common due date. The literature review related to the problems with
is given below (see
Table 1).
As can be seen in
Table 1, most of the studies in the literature on the group sequencing problem are limited to considering only the common due date or only the convex function of the processing time with respect to the resources. This study integrates the case where processing time follows a convex function of resource allocation under common due date allocation as well as learning effects.
4. Special Case
For
the complexity remains an open problem, so a special case is discussed, i.e.,
and
for
.
Theorem 4. Forif and for , the optimal group sequence is solvable by solving the assignment problem (denoted by ). Proof. Apparently, the optimal sequence
of
can be determined by Theorem 3, and
is a constant for
and
,
. We let the binary variables
and
where
the optimal group sequence is obtained by the
:
□
Therefore, the special case (i.e.,
and
for
) of
can be solved optimally by the algorithm described below (Algorithm 1).
Algorithm 1 Special Case |
Input: n, r, , , , , , , , , First step: Calculate by using Lemma 1; Second step: Calculate by using Lemma 2; Third step: Determine the jobs sequence in each group by Theorem 2; Fourth step: Calculate values by Equation (11); Fifth step: Determine the optimal group sequence by the (13)–(16); Sixth step: Calculate the optimal resource allocation by Equation (7). Output: optimal resource allocation, optimal sequence
|
Theorem 5. For the special case and ofit is solvable by Algorithm 1 in time. Proof. The correctness of the algorithm follows directly from Theorems 2–4. Steps 1, 2 and 4 need time , respectively. In Step 3, sorting internal job sequence needs time . The in Step 5 needs time . Thus, the total complexity of Algorithm 1 is . □
The following example illustrates the application of Algorithm 1 for
Example 1. There are jobs, groups, , , where . The of each job are shown in Table 2. Solution:
Step 1. From Lemma 1, we have
Step 2. From Lemma 2, we have
Step 3. According to Theorem 3, the optimal internal job sequences are
Step 4. The values of
(by Equation (11)) are given in
Table 3,
.
Step 5. Solving the (13)–(16), it follows that .
Step 6. From Equation (7),
can be calculated as follows:
5. General Case
If the optimal job sequence
is given by Theorem 3, the term
cannot be minimized by the non-increasing (LPT) order or the non-decreasing (SPT) order of
.
Example 2. There are three jobs belonging to two groups, , , where , . The of each job are shown in Table 4. According to Lemmas 1 and 2, we have
Then, the optimal internal job sequences are .
According to the SPT order of
, the term can be calculated as follows:
According to the LPT order of
, the term can be calculated as follows:
Therefore, the LPT order of
is not an optimal group sequence.
Example 3. There are 3 jobs belonging to 2 groups, , , where , . The of each job are shown in Table 5. According to Lemmas 1 and 2, we have
Then, the optimal internal job sequences are .
According to the SPT order of
, the term can be calculated as follows:
According to the LPT order of
, the term can be calculated as follows:
Therefore, the SPT order of
is not an optimal group sequence.
5.1. Heuristic Algorithm
According to Theorem 3, the optimal schedule of jobs within each group can be obtained and the optimal resource allocation for a given schedule can be obtained by Theorem 1. From
Section 5, the following heuristic algorithm (HA) can be proposed (Algorithm 2).
Algorithm 2 HA |
Input: , r, , , , , , , , , First step: Calculate by using Lemma 1; Second step: Calculate by using Lemma 2; Third step: Determine the jobs sequence in each group by Theorem 3; Fourth step: Schedule groups by the non-increasing of and calculate objective function value by Equation (6); Fifth step: Schedule groups by the non-increasing of and calculate objective function value by Equation (6); Sixth step: Schedule groups by the non-increasing of and calculate objective function value by Equation (6); Seventh step: Calculate the optimal resource allocation by Equation (7). Output: resource allocation, local optimal sequence
|
5.2. Simulated Annealing Algorithm
As in Lai et al. [
34], simulated annealing (SA) (Algorithm 3) is actually a greedy algorithm, but it introduces a stochastic element to the search process. This paper uses this classical algorithm to compare it with the proposed HA algorithm (Algorithm 2). Simulated annealing accepts with some probability a solution that is worse than the current one, so it is possible to jump out of this local optimum and research for the global optimum, SA is a good choice for
Algorithm 3 SA |
Input: , r, , , , , , , , , First step: Set the internal job sequence by the Second step of Algorithm 1; Second step: Use the pairwise interchange (PI) neighborhood generation method; Third step: Calculate the objective value of the original schedule ; Fourth step: Calculate the objective value of the new schedule . If the is less than , it is accepted. Nevertheless, if the is higher, it might still be accepted with a decreasing probability as the process proceeds. This acceptance probability is determined by the following exponential distribution function: where is a parameter and is the change in the objective function. In addition, the method is used to change in the kth iteration as follows: where is a constant. After preliminary trials, is used; Fifth step: If increases, the new sequence is accepted when , where is randomly sampled from the uniform distribution; Sixth step: The schedule is stable after iterations. Output: resource allocation, local optimal sequence
|
5.3. Lower Bound
We let
be a group sequence, where
(
) is the scheduled (unscheduled) part, and there are
f groups in
. We have
From the above equation, it can be seen that the terms
,
, and
are known, which can be seen as constants. And the term
can be minimized by Theorem 2, and
. Then, we have the following lower bound:
where
, and
. We let
; then, the second lower bound can be written as
it can be known that
by Theorem 2. And
where
and
. Note that
and
(
) do not necessarily correspond to the same group. To make the lower bound tighter, we select the largest of the three as the lower bound, that is,
5.4. Branch-and-Bound Algorithm
Based on the above-mentioned lower bounds, the branch-and-bound (B&B) algorithm with enumeration as the central idea is proposed, and this enumeration algorithm can be used as an optimal program to solve the problem posed in this study in the steps described below (Algorithm 4).
Algorithm 4
|
Input: , r, , , , , , , , , First step: Use Algorithm 1 to obtain an initial solution for the problem; Second step: Calculate the lower bound for the node; Third step: If the lower bound for an unfathomed partial schedule of groups is larger than or equal to the value of the objective function of the initial solution, eliminate the node and all the nodes following it in the branch. Calculate the objective function value of the completed schedule; if it is less than the initial solution, replace it as the new solution; otherwise, eliminate it. Fourth step: Continue until all nodes are explored. Output: resource allocation, optimal sequence
|
Example 4. We consider an example in which there are 14 jobs belonging to five groups, , where . The of each job, job number , and setup time of each group are shown in Table 6 and Table 7. Solution:
From Algorithm 2, we have
. According to Steps 3-5, the initial group sequences are:
and their corresponding objective values are
, and
We select the smallest
as the upper bound. According to the B&B algorithm, the following search tree can be obtained (see
Figure 1). The numbers in
Figure 1 represent the lower values, and the
is defined as Level 0. From
Figure 1, the optimal group sequence is
, where at Level 1, for group
, from Equations (18)–(20), the lower bounds are