Performance of fuzzy inference systems to predict the surface temperature of broiler chickens

Abstract

This study aimed to compare fuzzy systems with different configurations to predict the surface temperature (ts) of broiler chickens subjected to different intensities and durations of thermal challenges in the second week of life. Data on the ts of broiler chickens aged 8 to 11 days were acquired by infrared thermography and subjected to combinations of four dry-bulb temperatures (tdb) (24, 27, 30, and 33 °C) and four durations of thermal challenges (DTC) (1, 2, 3, or 4 days). The input variables of the fuzzy systems were tdb and DTC, and the output variable was ts. The Mamdani inference method involving five defuzzification methods [center of gravity (centroid), bisector of the area (bisector), largest of maximum (lom), middle of maximum (mom), and smallest of maximum (som)], and Sugeno inference with two defuzzification methods [weighted average (wtaver) and weighted sum (wtsum)] were evaluated. For both inference methods, triangular and Gaussian pertinence functions were tested for input and output variables, except for Sugeno inference, which used singletons functions as output variables. While developing fuzzy systems, different configurations must be compared, and the system with smaller simulation errors should be selected.