Application of an artificial neural network to optimise energy inputs: An energy- and cost-saving strategy for commercial poultry farms

The current study estimates target values of energy inputs along with an assessment of energy- and cost-saving strategies for poultry farms. In 2019, cross-sectional data were collected from 192 farmers at environmentally controlled poultry farms in Pakistan. A well-structured questionnaire was used to conduct face-to-face interviews with respondents. The results reveal that 1 MJ energy input at poultry farms produced 1.9 MJ of energy output. The Levenberg–Marquardt algorithm found the best topology of the ANN model at a hidden layer consisting of 10 neurons, including the lowest mean absolute percentage error (14.42) and the highest R2 (0.83) and model efficiency (0.79). The training model confirmed the inefficient use of energy inputs in the farms and a 3.37% overuse of energy inputs at a given amount of energy output. Particularly, fuel energy was overused by 51.02%. For each flock of chickens (1000 birds), the use of energy inputs at a set target level saved 318.32 MJ of energy input and 5.59 USD in costs. Moreover, at the targeted energy inputs, every year, the cost savings per farm could be 958.84 USD. The parametric analysis reported that the energy inputs of electricity, maize, soybean, and minerals and vitamins significantly increased energy output by 0.80, 0.05, 0.41, and 0.09 units, respectively. Overuse of energy inputs was confirmed because IE and BE showed a decreasing return to scale (RTS< 1). The promising ability of such a training model suggests that using the recommended energy inputs can maximise energy efficiency, and minimise the cost of production on poultry farms.