Population based optimization via differential evolution and adaptive fractional gradient descent

Abstract

We propose a differential evolution algorithm based on adaptive fractional gradient descent (DE-FGD) to address the defects of existing bio-inspired algorithms, such as slow convergence speed and local optimum. The crossover and selection processes of the differential evolution algorithm are discarded and the adaptive fractional gradients are adopted to enhance the global searching capability. For the benchmark functions, our proposed algorithm Specifically, our method has higher searching accuracy than several state of the art bio-inspired algorithms. Furthermore, we apply our method to specific tasks - parameters estimation of system response functions and approximate value functions. Experiment results validate that our proposed algorithm produces accurate estimations and improves searching efficiency