Grasshopper optimization algorithm (GOA) is a newly proposed meta-heuristic algorithm that simulates the biological habits of grasshopper seeking for food sources. Nonetheless, some shortcomings exist in the basic version of GOA. It may quickly drop into local optima and show slow convergence rates when facing some complex basins. In this work, an improved GOA is proposed to alleviate the core shortcomings of GOA and handle continuous optimization problems more efficiently. For this purpose, two strategies, including orthogonal learning and chaotic exploitation, are introduced into the conventional GOA to find a more stable trade-offbetween the exploration and exploitation cores. Adding orthogonal learning to GOA can enhance the diversity of agents, whereas a chaotic exploitation strategy can update the position of grasshoppers within a limited local region. To confirm the efficacy of GOA, we compared it with a variety of famous classical meta-heuristic algorithms performed on 30 IEEE CEC2017 benchmark functions. Also, it is applied to feature selection cases, and three structural design problems are employed to validate its efficacy in terms of different metrics. The experimental results illustrate that the above tactics can mitigate the deficiencies of GOA, and the improved variant can reach high-quality solutions for different problems. (C) 2020 Elsevier Ltd. All rights reserved.