A new probability model for insuring critical path problem
with heuristic algorithm
Zhenhong Li, Yankui Liu
n
, Guoqing Yang
College of Mathematics & Computer Science, Hebei University, Baoding 071002, Hebei, China
article info
Article history:
Received 4 April 2012
Received in revised form
1 June 2012
Accepted 4 July 2012
Available online 30 July 2014
Keywords:
Critical path
Probability model
Stochastic optimization
Minimum risk criterion
Hybrid algorithm
abstract
In order to obtain an adequate description of risk aversion for insuring critical path problem, this paper
develops a new class of two-stage minimum risk problems. The first-stage objective function is to
minimize the probability of total costs exceeding a predetermined threshold value, while the second-
stage objective function is to maximize the insured task durations. For general task duration
distributions, we adapt sample average approximation (SAA) method to probability objective function.
The resulting SAA problem is a two-stage integer programming model, in which the analytical
expression of second-stage value function is unavailable, we cannot solve it by conventional optimiza-
tion algorithms. To avoid this difficulty, we design a new hybrid algorithm by combining dynamic
programming method (DPM) and genotype-phenotype-neighborhood based binary particle swarm
optimization (GPN-BPSO), where the DPM is employed to find the critical path in the second-stage
programming problem. We conduct some numerical experiments via a critical path problem with 30
nodes and 42 arcs, and discuss the proposed risk averse model and the experimental results obtained by
hybrid GPN-BPSO, hybrid genetic algorithm (GA) and hybrid BPSO. The computational results show that
hybrid GPN-BPSO achieves the better performance than hybrid GA and hybrid BPSO, and the proposed
critical path model is important for risk averse decision makers.
& 2014 Elsevier B.V. All rights reserved.
1. Introduction
In a complex project management problem, we often use a
directed network graph to describe various task s and the relation-
ships among the tasks. In this framework, the arcs represent
dependent tasks and the arc weights serve as associated task
durations. Also, there exists a node 0 representing the start of the
project and a node n representing its termination. A project can be
considered completed if all its activities have been finished. An
important theoretical result is that the minimum time to complete
all the activities in the activity network equals to the length of the
longest path from the source node to the destination node [1].
Thus, this path, called critical path, represents the sequence of
activities, which will take the longest time to complete. Chen et al.
[2] developed a polynomial time algorithm to find the critical path
and analyzed the float of each arc in a time-constrained activity
network. Guerriero and Talarico [3] proposed a general method to
find the critical path in a deterministic activity-on-the-arc net-
work, considering three different types of time constraints.
Another area of research dealt with the stochastic nature of
activity time. For example, Kelley [4,5] and Moehring [6] estimated
the probability that a project would be completed by a given
deadline if the duration for each activity is not known with
certainty; Burt and Garman [7], Bowman [8] and Mitchell and
Klastorin [9] treated mass uncertain information by heuristic-
based and Monte Carlo simulation-based techniques, and Shen
et al. [10] proposed expectation and chance-constrained models
for insuring critical path problems and designed decomposition
strategies to solve these models.
In this paper, we approach the insuring critical path problem
from a new viewpoint. It is known that the appropriateness of
expectation criterion for insuring critical path problem depends on
the assumption that the insuring process can be repeated a great
number of times, this implies by the law of large numbers that in
the long run the average cost will be equal to the expected cost.
But, this assumption will often not be justified and thus the
expected cost may not be of much interest to risk averse decision
makers. On the other hand, the optimal solution of the expected
value problem may only assure the achievement of the corre-
sponding expected cost with a relatively small probability. Conse-
quently, the risk averse decision maker will not consider the
solution of the expected value problem to be optimal. Instead,
what may be desired is a solution ensuring a low probability of
very large costs. These considerations lead us to adopt minimum
risk criterion in insuring critical path problems. In the proposed
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/neucom
Neurocomputing
http://dx.doi.org/10.1016/j.neucom.2012.07.061
0925-2312/& 2014 Elsevier B.V. All rights reserved.
n
Corresponding author. Tel./fax: þ86 312 5066629.
E-mail addresses: lizhenhong6688@163.com (Z. Li), yliu@hbu.edu.cn (Y. Liu),
ygqfq100@gmail.com (G. Yang).
Neurocomputing 148 (2015) 129–135