11 Parameter Control in Evolutionary Optimisation
377
result in optimisation problems with over a billion variables [60]. Large-scale
optimisation is also present with many data analytic and learning problems [61]. The
problems range from shape design optimisation for aircraft wings and turbine blades
[62], satellite layout design [63] and parameter calibration of water distribution
system [64] to seismic waveform inversion [65]. A major challenge of large-scale
optimisation is the exponential growth in the size of the search space with respect
to the number of decision variables [56].
In recent years, LSGO gained attention and attracted wide interest from
researchers and practitioners as well as mathematicians and engineers. The
challenges motivated the design of many kinds of efficient, effective and robust
kinds of metaheuristic algorithms to solve LSGO problems with high-quality
solutions and high convergence performance as well as with low computational cost
[66].
To achieve acceptable results even for the same problem, different parameter settings along with different reproduction schemes at different stages of optimisation
process are needed. Therefore, several techniques (e.g. [67, 68]) have been designed
to adjust control parameters in an adaptive or self-adaptive manner instead of a trialand-error procedure.
11.5.2 Dynamic Optimisation
Real-world optimisation problems are usually subject to changing conditions over
time. The effects of these changes could influence several aspects of the problem,
such as the objective function, the problem instance, its constraints etc. Therefore,
the optimal solution to the problem might change over time. These problems, when
solved by an optimisation algorithm on-the-fly, are called dynamic optimisation
problems (DOPs) [69].
One can understand that optimising dynamic problems is not a simple task. The
algorithm is expected to be able to track the current optimal solution as well as the
changing optimal solution over time. Therefore, the optimisation procedure has to
be able to detect these changes and react quick enough. This also requires dynamic
change of the ratio for exploration and exploitation parts of the search. Both adaptive
[70] and self-adaptive [71] parameter control can be used.
Based on a comprehensive survey [72], four different strategies can be used to
help population-based algorithms to adapt in dynamical environments:
• Increasing diversity of the population after a change is detected, e.g. by increasing mutation rate every N generations
• Maintaining diversity throughout the run, to avoid convergence of the population
on one point
• Memory-based approaches, taking into consideration older solutions and sometimes making predictions based on historical data
377
result in optimisation problems with over a billion variables [60]. Large-scale
optimisation is also present with many data analytic and learning problems [61]. The
problems range from shape design optimisation for aircraft wings and turbine blades
[62], satellite layout design [63] and parameter calibration of water distribution
system [64] to seismic waveform inversion [65]. A major challenge of large-scale
optimisation is the exponential growth in the size of the search space with respect
to the number of decision variables [56].
In recent years, LSGO gained attention and attracted wide interest from
researchers and practitioners as well as mathematicians and engineers. The
challenges motivated the design of many kinds of efficient, effective and robust
kinds of metaheuristic algorithms to solve LSGO problems with high-quality
solutions and high convergence performance as well as with low computational cost
[66].
To achieve acceptable results even for the same problem, different parameter settings along with different reproduction schemes at different stages of optimisation
process are needed. Therefore, several techniques (e.g. [67, 68]) have been designed
to adjust control parameters in an adaptive or self-adaptive manner instead of a trialand-error procedure.
11.5.2 Dynamic Optimisation
Real-world optimisation problems are usually subject to changing conditions over
time. The effects of these changes could influence several aspects of the problem,
such as the objective function, the problem instance, its constraints etc. Therefore,
the optimal solution to the problem might change over time. These problems, when
solved by an optimisation algorithm on-the-fly, are called dynamic optimisation
problems (DOPs) [69].
One can understand that optimising dynamic problems is not a simple task. The
algorithm is expected to be able to track the current optimal solution as well as the
changing optimal solution over time. Therefore, the optimisation procedure has to
be able to detect these changes and react quick enough. This also requires dynamic
change of the ratio for exploration and exploitation parts of the search. Both adaptive
[70] and self-adaptive [71] parameter control can be used.
Based on a comprehensive survey [72], four different strategies can be used to
help population-based algorithms to adapt in dynamical environments:
• Increasing diversity of the population after a change is detected, e.g. by increasing mutation rate every N generations
• Maintaining diversity throughout the run, to avoid convergence of the population
on one point
• Memory-based approaches, taking into consideration older solutions and sometimes making predictions based on historical data