Time-series prediction of organomineral fertilizer moisture using machine learning

This study aims to model and forecast the drying process of three new types of commercial organomineral fertilizers: gold sulfur, 25.5.5, and 5x10 at elevated temperatures. They absorb and release moisture, depending on the conditions. Accurate prediction of drying behaviour is essential. Drying was carried out at temperatures of 70, 75, and 80 degrees C through natural convection. The data are unimodal time series of the moisture rate (MR). MR ). Supervised machine/deep learning techniques such as nonlinear autoregressive (NAR) network, adaptive neural fuzzy inference systems (ANFIS), long short-term memory (LSTM) network, gated recurrent unit (GRU), and hybrid of convolutional neural network (CNN) and recurrent neural network (RNN) are used in addition to wellknown regression-based formulas. The models can predict 30 minutes ahead. An error analysis was performed for performance comparison using the metrics, root mean square error (RMSE), RMSE ), and coefficient of determinant, R 2 . Two types of validation were performed by monitoring error convergence and using prediction curves. The most effective forecasting was obtained at an air temperature of 80 degrees C for all materials with machine learning. While RMSE =0.0030 having R 2 = 0.98843 by the LSTM network for gold sulfur (80 degrees C), ANFIS was the best for 25.5.5 and 5x10 with RMSE =0.0056, R 2 = 0.75585 and RMSE =0.0060, R 2 = 0.81938, respectively, in the MR prediction. GRU was remarkable for both its speed of 13.77 sec and RMSE =0.009. CNN-RNN has a more complex structure but lower performance. The results demonstrate that machine learning techniques are better than regressions. Among them, ANFIS provides the most reliable results. Regressions, including exponential terms, were good at providing a general curve shape but not peaks and drops. In addition, regressions are not good at forecasting.