It is significant to evaluate accurately energy efficiency of ethylene cracking furnace as the highest energy consumption device in ethylene production. However, previous energy efficiency evaluation methods mainly concentrate on system-level evaluation of ethylene production and fail to consider effects of production load on operation conditions, which are unsuitable for detailed evaluation online. Considering single energy efficiency content and impractical improvement scheme of DEA and unreasonable static baselines in TOPSIS, a modified TOPSIS energy efficiency evaluation method is proposed. A set of new energy efficiency indicators is designed through matter conversion and energy transformation together with matter and energy interaction. To acquire real-time energy efficiency a simulation model of cracking furnace is established by employing radial basis function neural network. To improve evaluation accuracy, the relations among energy efficiency and operation conditions and production load are quantified by calculation formulas of energy efficiency indicators and functions extracted from simulation model. The sequential quadratic programming algorithm is suggested to solve dynamic baselines according to real-time production load by adjusting operation conditions within constraints. Furthermore, optimal operation conditions are provided by searching for maximum comprehensive energy efficiency. Finally, validity of proposed evaluation method is illustrated by applying in a practical cracking furnace. (C) 2018 Elsevier Ltd. All rights reserved.