Estimating Fuel Characteristics from Simulated Circulating Fluidized Bed Furnace Data

This paper proposes a soft sensor to estimate the elementary fuel characteristics in combustion-thermal power plants. The proposed approach is data-driven. The input-output data is generated by a digital twin. Application targets circulating fluidized bed boiler, where furnace (combustion) side is considered only. First, the nonlinear dynamics of the furnace is approximated with a linear time-invariant dynamic model. Then two separate methods, Kalman filter and internal governor, are applied for state estimation. Results show that the approach is viable and has low computational complexity, but the weakly observable modes are difficult to predict accurately.