Spatial modeling of pigs’ drinking patterns as an alarm reducing method I. Developing a multivariate dynamic linear model

•A spatial dynamic linear model for modeling of pigs’ drinking patterns is developed.•The model relies on simultaneously monitored water consumption from multiple pens.•Seven different correlation structures between drinking patterns are defined.•A weaner herd and a finisher herd are modeled separat...

Full description

Saved in:
Bibliographic Details
Published in:Computers and electronics in agriculture 2019-06, Vol.161, p.79-91
Main Authors: Dominiak, K.N., Pedersen, L.J., Kristensen, A.R.
Format: Article
Language:English
Subjects:
Correlation
Early warning
Error analysis
Flowmeters
Growing pigs
Measuring instruments
Modelling
Multivariate analysis
Pens
Simultaneous monitoring
Spatial model
Water consumption
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A spatial dynamic linear model for modeling of pigs’ drinking patterns is developed.•The model relies on simultaneously monitored water consumption from multiple pens.•Seven different correlation structures between drinking patterns are defined.•A weaner herd and a finisher herd are modeled separately.•Model fit (MSE) indicate drinking patterns are correlated. The overall objective of this paper is to present the development of a spatial multivariate dynamic linear model (DLM) modeling the water consumption of growing pigs throughout the entire growth periods. The water consumption from multiple pens in multiple sections are monitored simultaneously by flow meters in both a commercial herd of finisher pigs (30–110 kg) and a research facility herd of weaner pigs (7–30 kg). The diurnal drinking patterns are modeled by a multivariate DLM, which is superpositioned by four sub-models describing three harmonic waves and a growth trend. The overall hypothesis of this paper is that pens and sections in a herd of growing pigs are correlated, and that this correlation can be modeled using model parameters defined at different spatial levels. Therefore seven model versions are defined to reflect a variety of temporal correlation structures between the monitored drinking patterns. The model versions were trained on learning data of the two herds, and run on separate test data sets from the herds. Their ability to fit the test data is measured as mean square error (MSE). Results for the finisher herd indicate that drinking patterns from pens within the same section are correlated (MSE  = 13.850). For the weaner herd, results indicate an inverse relation between the degree of correlation and the model fit. Thus, the best fit (MSE  = 1.446) is found for the model version expressing least correlation in data from pens across the herd.
ISSN:0168-1699
1872-7107
DOI:10.1016/j.compag.2018.06.032