ANN model for predicting operating parameters of a variable rate applicator

The suitable operating parameters of fluted roller metering mechanism need to be selected to address variability of application of inputs in a variable rate applicator. At present, the selection of operating parameters depends mainly on empirical rules and experimental trials. This paper presents the results of development and evaluation of multiple linear regression (MLR) and artificial neural network (ANN) models for predicting operating parameters of fluted roller metering mechanism of a variable rate applicator. The MLR and ANN models were developed to predict operating parameters viz. application rate, particle damage and particle distribution per unit area based on the data collected from experimental trials conducted under laboratory condition using fluted roller metering mechanism. The MLR models simulated the fluted roller exposed length with coefficient of determination (R²) values of 072, 0.65, 0.74 for urea, SSP and MOP fertilizers, respectively during training and 0.62, 0.54 and 0.59 for urea, SSP and MOP fertilizers, respectively during testing. The ANN model was optimized for 3–1–4 configuration with Levenberg–Marquardt (LM) algorithm, which indicated good performance during testing with the coefficient of determination (R²) of 0.60–0.84, 0.71–0.91, and 0.59–0.87 for granular SSP, urea and MOP fertilizer, respectively. The Nash–Sutcliffe coefficient (E) for ANN training data set ranged 0.66–0.85, 0.71–0.92 and 0.61–0.85 for granular SSP, urea and MOP fertilizer, respectively. It was concluded that the ANN model predicted the operating parameters of the variable rate applicator better than MLR model with r² value close 1.