Mercury is an important pollutant emitted from coal-fired power plants. Elemental mercury (Hg-0) is harder to be removed than oxidized mercury (Hg2+) and particulate bound mercury (Hg-p) in the flue gas at back-end of furnace. In this study, a method based on computational intelligence was proposed to enhance Hg-0 removal efficiency. It was realized by improving the transformation efficiency of Hg-0 into Hg2+ and Hg-p and then removing them by air pollution control devices. First, relationships between Hg-0 concentrations at the stack and variables like open values of secondary air, open values of over fire air, oxygen at the exit of economizer, load, coal qualities and so on were modeled with aid of tuned PCA-support vector machine. Then, manipulated variables and regulated variables were optimized by particle swarm optimization algorithm to enhance transformation efficiency of He. A field thermal adjustment test was carried out on some 600 MW unit and the proposed method was applied to that unit and compared with ACO. Results showed that removal efficiencies were enhanced greatly in general. The increment of removal efficiency can reach up to 14.71%. Besides, optimal strategies can be found in few iterations, making it suitable for online applications. (C) 2018 Elsevier Ltd. All rights reserved.