Computational Intelligence - February 2017 - 31

designed to make better use of the information
Location management in LTE networks aims at quickly
about both feasible and infeasible solutions. In
general, feasibility rules [34] based constraint
tracking where the users are, and this tracking makes
handling methods place a higher value on feasiit possible to deliver calls, short message services
ble solutions, which tend to ignore the imporand other mobile phone services to the users in a
tant function that infeasible solutions perform in
searching. However, study with Deb [35] has
timely way.
shown that, if properly used, the information
about infeasible solutions can effectively
enhance the search efficiency. Due to this fact, we apply the
linear combination of the two costs. For instance, the paging
M2M decomposition strategy [30] to make better use of infeasicost was ignored [4] or considered as a constraint [5], [6] by the
ble solutions. The population is decomposed into a number of
authors. They claimed that minimizing the location update cost
subpopulations in M2M framework, and every individual merewas advantageous since paging capacity can be easier to quantily competes with its counterparts belonging to the same subfy as a constraint. The authors in [7] treated bandwidth as a
population so that the infeasible solutions are more likely to
scarce resource and tried to minimize the paging cost. A weight
survive than they would be in a single-population EMO algofactor should be provided when combining paging cost and
rithm, because of less selection pressure. In this way, a certain
location update cost together using the weighted sum approach
number of promising infeasible solutions will be kept in the
[8], [9], [22]. The solutions obtained in this way are sensitive to
evolutionary process to guide the population search. The M2M
the weight used in forming the objective function. Tcha et al.
decomposition strategy can also be beneficial in maintaining the
proposed a cutting plane algorithm for an integer programpopulation diversity [30].
ming model of LA planning problem [10]. Since LA planning is
Three generated networks are used to simulate the real situan NP-hard problem [10], many heuristic algorithms, such as
ation of user movements and geographic information. ComArtificial Neural Network (ANN) [11], Simulated Annealing
parison experiments are conducted to investigate the
(SA) [5], [6], [12], Evolutionary Algorithm (EA) [13]-[16] and
effectiveness and efficiency of the proposed multi-objective TA
Greed Search (GS) [17] have been used to address this problem.
planning model and the M2M decomposition strategy for solvComputational complexity of heuristic algorithms for LA
ing this model. The results obtained by solving this new model
planning is also concerned by researchers. Gondim et al. introare compared with those of a single-objective TA planning
duced the elitist individuals preserving based crossover and
model proposed by Subrata and Zomaya [8]. The simulation
edge-based mutation to accelerate the convergence of their
results show that the proposed multi-objective model can
new EA for LA partitioning [16]. The computational efficiency
achieve better trade-offs between the location update cost and
of SA, taboo search, and genetic algorithm for the LA planning
paging cost. We also compare the results obtained by optimizproblem was studied and compared in [18].
ing the proposed model using M2M and MOEA/D. The main
Meanwhile, some researchers studied the TA/LA planning
difference between the two algorithms is that M2M utilizes the
problem from different aspects, and a lot of beneficial achievepopulation decomposition strategy, while MOEA/D does not.
ments have been made. Toril et al. proposed a automatic methSimulation results show that M2M achieves both better conod for TA replanning by analyzing the frequency of user
vergence and diversity than MOEA/D.
movements and change of traffic trends in an LTE network
The remainder of this paper is organized as follows: Sec[19]. A new TAL configuration method was introduced by
tion II explains the related work. Section III describes the
Ikeda et al. to detect the burst of location updates from the
formulation of the TA planning problem and the multilocation update records [20]. Lee et al. [21] proposed an integer
objective model proposed in this paper. Section IV presents
programming model for graph partitioning. Cayirci and Akythe design of the EMO algorithm based on the M2M
ildiz used the number of LA boundary crossings as the measure
decomposition to solve the proposed model. In Section V,
of location update cost, and proved that the location update
simulation experiments of three generated networks are concost can be minimized by minimizing the inter-LA traffic flow
ducted, and experimental results are shown and analyzed.
in the network [22]. Krichen et al. extended the classical LA
Finally, we conclude the paper in Section VI.
planning problem by including additional objectives and constraints [23]. An integer programming model was developed to
achieve better trade-offs between the network performance
II. Related Work
and TA reconfiguration cost in [24]. A multi-layer LA design
A TA is similar to the concept of a Location Area (LA) in the
model, which allows an LA paging several areas, was studied
GSM network, and LA planning has been extensively studied.
and analyzed by Park and Soni [25].
These models and methods for LA planning apply to TA planEMO algorithms are powerful tools for solving complex
ning in LTE networks as well [24]. Thus we also review some
optimization problems. The applications of EMO algorithms
of the relevant work about LA planning.
in engineering have been becoming increasingly popular and
Most LA planning models consider only one objective,
perfect. Dorn et al. applied NSGA-II (a fast and elitist
either in the form of paging cost, location update cost or a

FEbruary 2017 | IEEE ComputatIonal IntEllIgEnCE magazInE

Table of Contents for the Digital Edition of Computational Intelligence - February 2017