In the last decades, thanks to the growing of the available computational capabilities given by modern personal computers, optimization techniques based on evolutionary algorithms (EAs) have received great attention and they have been effectively applied to several problems in engineering and sciences.
EAs are based on stochastic iterative procedures which consider a pool of trial solutions at each iteration thus enabling an efficient sampling of the solution space as compared to single-agent stochastic optimization algorithms (e.g., Simulated Annealing). The pool of solutions iteratively updates through the use of proper operators/rules until a convergence criterion is reached.
EAs have shown many attractive features suitable for dealing with large, complex, and nonlinear problems. More specifically, they are hill-climbing algorithms which not require the differentiation of the cost function, which is a “must” for gradient-based methods. Moreover, a-priori information can be easily introduced, usually in terms of additional constraints on the actual solution, and they can directly deal with real values as well as with a coded representation of the unknowns (e.g., binary coding). As regards to the architecture of their implementation, EAs can be effectively hybridized with deterministic procedures and are suitable for parallel computing.
Despite several positive advantages offered by the EAs, further researches are required in this framework to overcome the well-known drawbacks. As a matter of fact, the high computational load and the low convergence rate should be reduced especially when dealing with either high-dimension or highly nonlinear problems.
In this talk, a review of EA-based approaches for engineering is presented focusing on the theoretical framework and the state-of-the-art, but also pointing out potentialities, limitations, and current trends of the research in such a field.
Co-Author: P. Rocca