Dynamic Linear Models for multivariate hierarchical pig body weight modelling: first steps using the python pyBATS package

Increasing productivity has been the main focus of the pig farming sector, however often at the cost of pig welfare. Recent technological developments allowed for the rise of a range of precision livestock farming (PLF) technologies that can automatically monitor the pigs and assist the farmer. These PLF technologies are usually isolated solutions. By combining multiple PLF technologies into an integrated system, such as a digital twin, the collected data could improve the pig monitoring, enhance the insight in pig health, welfare and productivity and could optimize modelling results. Dynamic linear models (DLMs) can be used to integrate the measurements of multiple PLF technologies into an integrated model for monitoring and prediction. In this research, the goal is to create a multivariate hierarchical DLM used for weight monitoring and prediction based on data from multiple PLF technologies (RFID antennas at feeder and drinker, water flow sensors, weight scales, climate sensors and cameras for computer vision based behaviour measurements). The model will be built in python, using the pyBATS package, in multiple steps. Starting from a relatively simple univariate DLM for weight, then introducing the hierarchy and finally adding other sensor information in order to come to the multivariate hierarchical DLM model.