A tobacco moisture prediction approach based on VAE and PSO-BiLSTM

The tobacco moisture is one of the key indexes in the process of making cut tobacco, which directly affects the aroma and overall quality of tobacco, thereby deciding the quality of cigarette. This paper takes the tobacco moisture as subject investigated, the process parameters data of the influencing factors of tobacco moisture are collected, and the Variational Autoencoder is adopted to extract the key potential variables of these data. A moisture prediction method based on Particle Swarm Optimization Bidirectional Long Short-Term Memory network is proposed. The results of MAPE=0.02 and RMSE=0.03 showed that the method can achieve the accurate prediction of tobacco moisture.