Combining real-time data and InraPorc® simulations to perform precision feeding in pigs

International audienceApplying precision feeding in growing pigs requires methods and models for real-time analysis of performance and prediction of nutrient requirements. This study compares two methods for these calculations and evaluates their potential to reduce nutrient input in pig feeding. A dataset of individual daily bodyweight (BW) and feed intake (FI) kinetics of 285 growing pigs, reared from 81 to 156 days of age in an experimental station in ad libitum feeding conditions, was used. The first approach (PF1) was developed during the Feed-a-Gene project, and uses Holt-Winters and MARS methods to daily forecast individual FI and BW, respectively. Standardised digestible lysine (dLys) requirements were computed daily according to the factorial method based on performance forecasting. The second method (PF2) relies on a set of 2,200 virtual pigs which ad libitum performances where simulated using InraPorc. Based on the comparisons of past FI and BW kinetics between real and virtual pigs, a set of up to 10 virtual nearest neighbours was determined daily for each real pig. The expected individual performances and dLys requirements were then obtained by averaging InraPorc data of these 10 virtual pigs. The precision feeding for each pig was then simulated. For each approach, the proportions of two premix diets (A and B, 9.7 MJ NE/kg, crude protein content of 16.9 and 9.3% and dLys content of 1.0 and 0.4 g/MJ NE, respectively) were computed daily to reach calculated requirements. Nitrogen (N) and dLys intakes and N excretion were computed individually using daily real performance and A:B ratio, considering equal performances independently of feeding method. A classical two-phase feeding (2-P) was also simulated (A:B ratio=83:17 before 65 kg mean BW, 50:50 afterwards). Compared to 2-P, N and Lys intakes and N excretion were respectively reduced by 6.6%, 9.6% and 11.9% with PF1 and 9.1%, 13.1% and 16.2% with PF2. Performance predictions were similar for PF1 and PF2 compared to real performance, but PF2 allows a better day-by-day stability, allowing a smoother decrease in N and dLys supply along growing period. The potential of the new method has now to be tested in vivo