Multi-objective evolutionary algorithm with prediction in the objective space

Multi-objective optimization problem resolution using Evolutionary Algorithms (EAs) has not yet been completely addressed. Issues such as the population diversity loss and the EA sensitivity to the Pareto front shape affect the algorithm performance. Various EAs include knowledge acquisition for the evolutionary process to deal with such problems. Several issues are crucial for the addition of knowledge using a probabilistic model; among these, we mention the time at which the algorithm should update the model, and which information is suitable for constructing the model. To handle these issues, we propose the Non-dominated Sorting Differential Evolution improvement with Prediction in the Objective Space (OSP-NSDE). When several premises based on the Approximated Hypervolume metric are achieved, the OSP-NSDE triggers the Objective Space Prediction (OSP) strategy. The OSP identifies trends in the movements of non-dominated individuals in the objective space, and then rapidly determines promising regions in the search space and generates a new population considering such regions. Regular variation operators are used to produce the offspring whenever the OSP condition is not satisfied. The OSP-NSDE effectiveness was verified using 31 well-known functions and three real-world problems, and compared with EA-based algorithms and others with collective intelligence. (C) 2019 Elsevier Inc. All rights reserved.