Predictive analysis in solar kiln drying of wood using recurrent neural networks

Abstract

Prediction in data mining, is a technique used in predicting results or outcomes of future occurrence in reference to existing information. Several predictive models have been developed for different fields of study. In solar kiln drying experiment, as a result of dependence on nature for its operation, outcomes of drying process is unstable and varies with weather variability. Although predictive models have been developed for wood drying experiments, there is very limited information on the use of Neural Networks for predicting outcomes in solar kiln drying of wood. In this work, Long Short-term Memory model, a special type of Recurrent Neural Network was adopted for prediction in solar kiln drying of wood. Data collected on external (atmospheric) and internal conditions of a solar kiln sited at from Forestry Research Institute of Nigeria was used for this study. Daily ambient and internal temperature and relative humidity were used as input data. The closeness of relationship between the experimental and predicted values (Mean Square Error, MSE = 0.97; 30.4) and (Squared Correlation, R2=0.68, 0.85) for Temperature and Relative Humidity respectively revealed that the model had a good agreement with data. The Equilibrium Moisture Content (EMC) of internal solar kiln environment which influences the outcome of drying was considered. The EMC of internal solar kiln environment was predicted for the next 730 days and suitability of the model for prediction was examined giving an MSE value of 0.2 and r2 value of 0.87. The findings of this study suggest a viable model for predicting drying outcomes under varying weather conditions.