Paddy Moisture Content and Drying Time Prediction Using Neural Networks

Abstract

The current paddy drying practice in the rice mill is heating and blowing air continuously to reduce the paddy moisture content to a certain level (i.e., 12%-14%). The stopping time of the drying process is determined via manual checking of the paddy moisture content (MC) from time to time. Several problems (i.e., inconsistent dried paddy quality and excess energy consumption) arise in the usual practice due to human error. Hence, the MC time-series data collected from an actual rice mill can be related to the environment and drying factors to build a moisture content profile of the drying paddy using Long Short-Term Memory (LSTM) neural networks. The prediction is carried out every 2 hours up to 20 hours from the beginning of the drying process. Based on the prediction of the MC level, the optimum drying time can be determined when the MC level reaches the desired level. When evaluated against the validation dataset, our system successfully predicted the MC from 2 hours to 10 hours ahead of time, with a Root Mean Square Error (RMSE) of just 1.5 units.