Online prediction of biomass moisture content in a fluidized bed dryer using electrostatic sensor arrays and the Random Forest method

The inherent moisture content in biomass needs to be dried before it is used for energy production. Fluidized bed dryers (FBD) are widely applied in drying biomass and the moisture content should be monitored continuously to maximise the efficiency of the drying process. In this paper, the moisture content of biomass in a FBD is predicted using electrostatic sensor arrays and a random forest (RF) based ensemble learning method. The features of electrostatic signals in the time and frequency domains, correlation velocity and the outlet temperature and humidity of exhaust air are chosen to be the input of the RF model. Model training is accomplished using the data taken from a lab-scale experimental platform and the hyper-parameters of the RF model are tuned based on the Bayesian optimization algorithm. Finally, comparisons between the online predicted and sampled values of biomass moisture content are conducted. The maximum relative error between the online predicted and reference values is less than 13%, indicating that the RF model provides a viable solution to the online monitoring of the fluidized bed drying process.