GPU-based variation of parallel invasive weed optimization algorithm for 1000D functions

Abstract

Considering the problems of slow convergence and easily getting into local optimum of intelligent optimization algorithms in finding the optimal solution to complex high-dimensional functions, we have proposed an improved invasive weed optimization (IIWO). Concrete adjustments include setting the newborn seeds per plant to a fixed number, changing the initial step and final step to adaptive one, and re-initializing the solution which exceeds the boundary value. Meanwhile, through applying the algorithm to the GPU platform, a parallel IIWO (PIIWO) based on GPU is obtained. The algorithm not only improves the convergence, but also strikes a balance between the global and local search capabilities. The simulation results of solving on the CEC' 2010 1000-dimensional (1000D) functions, have shown that, compared with other algorithms, our designed IIWO can yield better performance, faster convergence, higher accuracy and stronger robustness; whilst the PIIWO has significant speedup than the IIWO.