RiverOpt: A Multiobjective Optimization Framework Based on Modified River Formation Dynamics Heuristic
Satyabrata Dash, Sukanta Dey, Anish Augustine, Rudra Sankar Dhar, Jan Pidanič, Zdeněk Němec, Gaurav Trivedi · 2019
In river formation dynamics (RFD) method, water drops pursue a probable path to flow from high altitudes to flat surface. This geographical metaphor adopts a decreasing gradient principle supported by sedimentation and erosion mechanisms to reach for a feasible solution. In this paper, a new multi-objective optimization framework, RiverOpt is presented based on a modified RFD method. In this method, the probability of selecting the next path in RFD method is modified to exploit both transverse and longitudinal slopes. Further, the sedimentation parameter in RFD method is improved by introducing a sediment coefficient. Later, an external archive is integrated with RiverOpt framework to keep track of nondominated solutions in each generation. For benchmarking the performance of the proposed framework, a set of standard multiobjective test problems is employed. The results are compared with peer multiobjective optimization algorithms using two performance indicators (i.e., generational distance and hypervolume). Experimental results show that the proposed RiverOpt framework demonstrates competitive results in terms of convergence and diversity of Pareto optimal solutions. Finally, a case study of low noise amplifier circuit is analyzed to showcase effectiveness of the proposed framework.