Prediction of the biomass particles through the physics informed neural network.

Woody biomass energy is a kind of renewable energy that contributes to the reduction of
greenhouse gas emissions, the creation of healthier forests, and the reduction of wildfire
danger.
Simulations of biomass combustion, in general, are time-consuming simulations with a large
number of input particles. We use a deep hidden physics-based neural network model to
predict the behavior of particles throughout the simulation based on the equations of motion
to achieve an efficient simulation and reduce the processing effort. We replace discrete
element methods with inverse methods, which have the advantage of simulating velocity
fields without knowing the simulation's boundary and initial conditions. Reconstruction of the
velocity fields is done using a recurrent neural network in conjunction with a physics-based
loss function. The proposed model is suitable for modeling problems that involve moving
particles in a fixed bed. The number of neurons and activation functions in the artificial neural
network are optimized, and the effect of the sampling method and the number of outputs are
studied.