Krill herd algorithm (KHA) is a novel nature inspired (NI) optimization technique that mimics the herding behavior of krill, which is a kind of fish found in nature. The mathematical model of KHA is based on three phenomena observed in the behavior of a herd of krills, which are, moment induced by other krill, foraging motion and random physical diffusion. These three key features of the algorithm provide a good balance between global and local search capability, which makes the algorithm very powerful. The objective is to minimize the distance of each krill from the food source and also from the point of highest density of the herd, which helps in convergence of population around the food source. Improvisation has been made by introducing neighborhood distance concept along with genetic reproduction mechanism in basic KH Algorithm. KHA with mutation and crossover is called as (KHAMC) and KHA with neighborhood distance concept is referred here as (KHAMCD). This paper compares the performance of the KHA with its two improved variants KHAMC and KHAMCD. The performance of the three algorithms is compared on eight benchmark functions and also on two real world economic load dispatch (ELD) problems of power system. Results are also compared with recently reported methods to establish robustness, validity and superiority of the KHA and its variant algorithms.