In view of large range change of operating conditions of the coal-fired power units and difficulties in building global model of turbine heat rate, this paper presents data-driven dynamic energy consumption model using the property that the just-in-time learning local modeling approach can ensure the model accuracy in the large range change of operation conditions. Firstly, based on the technology mechanism analysis of operation process, the instrumental variable of energy consumption model is determined by using grey relational analysis. Secondly, distance, direction, and the direction trend between samples are all considered to design the similarity measurement criterion and obtain the modeling neighborhood dataset. Thirdly, the weighting factor based on the similarity of modeling data and the parameter optimization based on the niche differential evolution algorithm are introduced to the weighted least square support vector machine, a dynamic energy consumption model is build. Finally, the active update strategy is adopted to realize the selective preservation of the new data in model database. Simulation results show that the proposed method can more accurately predict the turbine heat rate, and provide the basic data for operation optimization and energy saving diagnosis of coal-fired power units.
