Publications

Abstract : Differential Evolution (DE) is a recent addition to the repository of Evolutionary Algorithms (EAs) under Evolutionary Computing Techniques. As similar to other EAs, DE also used for optimization based on population of members. During population optimization the classical DE faces major problem of premature convergence, which causes the sample members to converge early to a local optimum though there is a global optimum in the search space. This paper presents methods to reduce the effect premature convergence and achieve better optimal solutions. There are few works in the literature for same reason by altering the control parameters of DE viz., scaling factor (F) and crossover rate (CR). However, we propose methods to make suitable amendments in the population level directly after detecting premature convergence during the search of DE. These methods can be added as an additional component to DE algorithm. The proposed components are to replace the population members with distinct highest objective function values with random members, to replace the population members with distinct lowest objective function values with random members, to replace members in random fashion and to increase the population size dynamically to counter the early convergence of the population. The above mentioned techniques are implemented and added with classical DE. The performance efficacy of DE with added components is verified on implementing it over a set of Benchmarking Function Suite with functions of different characteristics. The experimental results proved that DE with above components added is able to achieve better optimum values than the classical DE.