A hybrid differential evolution algorithm with invasive weed optimization and its application to modeling of carbon content

This paper aims to the prediction of carbon content in spent catalyst in a continuous catalytic reforming (CCR) plant based on least squares vector machines (LSSVM). When modeling by LSSVM, the problem of optimizing the hyper-parameters draws many researchers' attention. In this paper, a novel hybrid algorithm named IWODE is proposed to deal with it. The algorithm embeds invasive weed optimization (IWO) as a local refinement procedure into differential evolution with adaptive crossover rate. New competitive exclusion and adaptive step length of spatial dispersal based on individuals' distance are introduced to make IWO more suitable as a local search algorithm. Simulation results and comparisons based on some well-known benchmarks indicate the efficiency of IWODE. And the predicted results of carbon content using the proposed method agree with the actual values well. The method is compared with five other techniques, including LSSVM optimized by DE, IWO, other two modified versions of DE and back propagation neural network (BPNN). The obtained results demonstrate that the proposed IWODE-LSSVM is superior to others in generalization performance and prediction ability.