Improved artificial bee colony-based optimization of boiler combustion considering NOX emissions, heat rate and fly ash recycling for on-line applications

A combustion optimization framework for on-line applications is proposed based on improved Artificial Bee Colony (ABC) algorithm. First, an enhanced General Regression Neural Network (Enhanced-GRNN) is designed with Gaussian Adaptive Resonance Theory (GART) learning and polynomial extrapolation to get better on-line performance and extreme value extraction. Then two improvements to classical ABC algorithm are proposed: a multi-segment method for evaluating the quality of food sources in employed bee phase based on an analysis of probability distribution of foods in different iteration segments; and a memory-based strategy for onlooker bees to find new foods and evaluate their quality considering better moving directions and steps. A cost function was also designed in this study, considering the factors of coal consumption, NOX emissions and the recycling potential of fly ash. Experiments with data samples from a 600 MW utility boiler show that the proposed framework is fast and flexible, and that the on-line optimization results can provide reasonable optimal advices to operating engineers at coal-fired power plants.