Mechanistic data-driven combined modelling and predictive control method for coal-fired power unit

With the increase of the proportion of new energy sources connected to the grid, challenges are also emerged for modeling and control of the conventional coal-fired power units to increase their operational flexibility of deep peak-regulation. Traditional data-driven modelling (DDM) requires a large amount of training data containing a variety of operating conditions, which are hard to be fully satisfied in engineering practice. This paper investigates the mechanics and dynamic characteristics of boiler-turbine systems in the coal-fired power unit, and proposes a mechanistic data-driven combined modelling (M-DDM) method. It not only requires less amount of data comparing with the DDM based method, but also covers a wide range of operating conditions by considering the dynamic characteristics of the coal-fired power unit. A model predictive control (MPC) based method was subsequently designed as the control vehicle. In order to improve the computing efficiency and reduce the probability of model mismatch considering the involved ordinary differential equations (ODEs) and nonlinearities. reinforcement learning is embedded into the MPC. Finally, simulations are conducted for demonstrating the superior performance of the proposed methods comparing with other commonly deployed modeling and control methods.