Using LSTM and PSO techniques for predicting moisture content of poplar fibers by Impulse-cyclone Drying

Abstract

Impulse-cyclone drying (ICD) is a new type of pretreatment method to remove the excess moisture of wood fibers (WFs) with high speed and low energy consumption. However, the process parameters are often determined by the experience of the process operators, thus the quality of WF drying lacks an objective basis and cannot be ensured. To address this issue, this study adopted the long short-term memory (LSTM) neural network, backpropagation neural network, and Central-Composite response surface method to establish a moisture content (MC) prediction model and a process parameter optimization model based on single-factor experiments. The initial MC, inlet air temperature, feed rate, and inlet air velocity were taken as the experimental factors, and the final MC was taken as the inspection index. The parameters of LSTM were optimized by particle swarm optimization (PSO) algorithm, and the predicted value of MC was fitted to the model. The PSO-optimized LSTM had higher prediction accuracy than did the typical prediction models. The optimal process for the targeted MC, which was obtained by PSO, was featured with an initial MC of 10.3%, inlet air temperature of 242°C, feed rate of 90 kg/h, and inlet air velocity of 8 m/s. PSO-LSTM could be a new approach for predicting the MC of WFs, which, in turn, could provide a theoretical basis for the application of ICD technology in the biomass composite industry.