Predictive models of coal consumption rate based on ant colony optimization algorithm and hybrid modeling strategies

Coal Consumption Rate (CCR) is one of the most important economic indicators of thermal power generating units, which, to some extent, affects the market competitiveness of the thermal power plants and determines whether the emissions are up to standard. However, due to the influence of many coupling factors on CCR, how to reduce coal consumption has become an outstanding technical problem. This paper presents generalized predictive models of CCR based on the ant colony optimization algorithm and hybrid modeling strategies. The data collected from two thermal power generating units rated 300[Formula: see text]MW and 600[Formula: see text]MW, are used to build the forecasting models and the Pearson correlation coefficient is used to determine the input parameters. 4 neural networks are optimized adaptively by the optimization algorithm and used to build the CCR Single Models (SMs). Compared with the neural networks that are not optimized and other optimized algorithms and models, the ACO-CCR models have higher forecasting accuracy. To further improve the prediction accuracy, two hybrid modeling strategies are proposed to build the CCR Hybrid Models (HMs). The predictive results show that the two hybrid modeling strategies could effectively improve the prediction accuracy of the single models. So, it gives a way out for building the CCR HM, especially when the forecasting accuracy of the CCR SM is generally very low.