Akademik Veri Yönetim Sistemi
Journal of Tekirdag Agricultural Faculty, cilt.21, sa.2, ss.482-500, 2024 (ESCI)
Although regression is a traditional data processing method, machine and deep learning methods have been widely used in the literature in recent years for both modelling and prediction. However, in order to use these methods efficiently, it is important to perform a preliminary evaluation to understand the data type. Therefore, pre-evaluation procedures are described in this study. Experimental uncertainty analysis was performed to determine the measurement uncertainties in the measurement devices and sensors used in the drying experimental setup. Significant and insignificant relationships between variables in the data set were determined by Pearson correlation matrix. Autocorrelation and partial autocorrelation functions were used to determine the time series lag in the drying data and an AR(5) series with 5 lags was determined. The data were found to have variable variance due to peaks and troughs in the raw data resulting from the natural behaviour of the drying process. Modelling success was achieved with the normalisation pre-evaluation process performed without distorting the raw data. Thus, it has been shown that better models can be obtained compared to traditional models. In order to avoid unnecessary time and computational costs in the trial and error method used to determine the number of hidden layers and neurons in the machine learning method, various formulas proposed in the literature were compared. It is shown that the correlation coefficient alone is not sufficient to determine the goodness of the model. In modelling the data in this study, the NARX model was found to converge to the desired value faster and with less error than ANFIS and LSTM models. In the simulation of a rotary drum dryer, the optimum number of mesh elements was determined as 1137 by mesh independence analysis. In this way, unnecessary over-calculations were also prevented. Of course, all these methods are already available in statistical science. However, in this study, the methods to be used for modelling and prediction purposes are carefully selected and explained with examples, especially for young researchers who are outside this field to gain speed and easy comprehension.