Volume 21: Sustainable Energy Solutions for a Post-COVID Recovery towards a Better Future: Part IV

Predicting milk cooling, milk harvesting and water heating electricity use on dairy farms using artificial neural networks Philip Shine, Michael D. Murphy

Abstract

Artificial neural network (ANN) models were developed to predict milk cooling, milk harvesting and water heating electricity consumption using data collected from 56 pasture-based Irish dairy farms. The methodology employed variable selection, outlier detection, hyper-parameter tuning and nested cross-validation. The ANN models were developed to predict monthly electricity use, while monthly predicted values were also aggregated and assessed at farm- and catchment-levels. Model input variables were constrained to stock and milk production, infrastructural equipment and farm management details. The ANN algorithm predicted monthly electricity consumption for milk harvesting with an error of 22% (relative prediction error), milk cooling to within 24% and water heating to within 31%. Prediction errors reduced to 16%, 12% and 9%, respectively when predicted values were aggregated at the farm-level. In addition, significant reductions in prediction errors were calculated when milk harvesting (0.8%), milk cooling (1.8%), and water heating (1.9%) predictions were aggregated at the catchment-level. This demonstrates the potential effectiveness of the developed ANN models as tools for macro-level simulations.

Keywords energy, modelling, neural network, machine learning, dairy, agricultural sustainability