Thank you so much for your reply Prof. Rubin.
I agree the problem have a relatively significant number of variables. It is because I want to determine routing and scheduling decisions. A binary variable x
v,ui,j,u is equal to 1 if the u
th task of vehicle v consists of going from i to j. And another variable y
v,u determines the time at which task u for vehicle v starts. These variables are primarily needed to implement operation synchronizations among vehicles. Also, a given node along the network may be visited multiple times.
It is correct, when integrality constraints are relaxed, the objective function is zero and this is due to the nature of the objective function which does not minimize travel costs but service durations.
Do you mean there should be an error in the formulation? or do you think another formulation with different variables should be used?
------------------------------
GJ Hn
------------------------------
Original Message:
Sent: Wed October 21, 2020 01:30 PM
From: Paul Rubin
Subject: CPLEX taking too long to solve an MILP
I suspect your model contains an error. The log says that the lower bound is stuck on zero, meaning that in the absence of integrality restrictions you could satisfy all constraints without doing anything that would contribute to the objective function. (The model also seems rather large for an instance with three customers and two variables -- over 17,000 variables! -- but perhaps you are scheduling hourly for several months?)
Paul
------------------------------
Paul Rubin
Professor Emeritus
Michigan State University
Original Message:
Sent: Tue October 20, 2020 05:38 PM
From: GJ Hn
Subject: CPLEX taking too long to solve an MILP
Hello,
I have developed a deterministic combined vehicle routing and scheduling problem that includes time precedence and operation synchronization constraints. I am trying to solve a very small problem instance with 3 customers and 2 vehicles but it is taking days to solve.
Before switching to the use of a solution algorithm or heuristic, I am wondering whether there is any parameters to adjust in CPLEX to improve computation time? Also is there a way to detect whether the model is weak? and what about the use of an HPC server?
Note that the problem is formulated in java using AMPL API. I have attached the node log.
Thanks!
------------------------------
G Hn
------------------------------
#DecisionOptimization