Region based memetic algorithm for real-parameter optimisation

Memetic algorithms with an appropriate trade-off between the exploration and exploitation can obtain very good results in continuous optimisation. That implies the evolutionary algorithm should be focused in exploring the search space while the local search method exploits the achieved solutions. To tackle this issue, we propose to maintain a higher diversity in the evolutionary algorithm's population by including a niching strategy in the memetic algorithm framework. In this work, we design a novel niching strategy where the niches divide the search space into hypercubes of equal size called regions forbidding the presence of two solutions in each region. The objective is to avoid the competition between the local search and the evolutionary algorithm. We tested this niching strategy in a memetic algorithm with local search chaining and obtained significant improvements. The resulting model also appeared to be very competitive with state-of-the-art algorithms. One of the main issues when designing an evolutionary algorithm (EA) [3] for real-coded parameter optimisation problems is to offer a good exploration of the search space and, at the same time, to exploit the most promising regions to obtain high quality solutions. Memetic algorithms (MA) were proposed [34] to manage these competing objectives. They are a hybridisation between EA and local search (LS) algorithms, mixing in one model the exploration power of EA and the exploitative power of the LS. MAs are characterised by the combination of an exploration algorithm and a local improvement algorithm. MAs with an appropriate trade-off between the exploration and exploitation can obtain accurate solutions, improving the search [14,21]. Therefore, the key issue when designing a MA is to organise both efforts in the most cooperative way. Niching strategies have been used in EA to either identify various optima in a fitness landscape or to maintain a strong diversity in the EA's population [13]. In our study, we consider that using niching strategy to maintain a higher diversity in the population leads to a better separation of the effort between the EA and the LS. Thus, we design a niching strategy to limit the competition between the EA and the LS in a MA. The purpose of this method is to let the EA focus on the exploration task by limiting its exploitation power, this task being more efficiently performed by the LS method. Contrarily to most niching strategies where the niches are defined around the solutions of the population, the niches …