Hostname: page-component-7b7567c8c-7pwfn Total loading time: 0 Render date: 2026-11-16T20:39:19.620Z Has data issue: false hasContentIssue false

Adaptive models for online estimation of individual milk yield response to concentrate intake and milking interval length of dairy cows

Published online by Cambridge University Press:  08 April 2011

G. ANDRÉ*
Affiliation:
Livestock Research, Wageningen University and Research Centre, P.O. Box 65, 8200 AB Lelystad, The Netherlands
B. ENGEL
Affiliation:
Biometris, Wageningen University and Research Centre, P.O. Box 100, 6700 AC Wageningen, The Netherlands
P. B. M. BERENTSEN
Affiliation:
Business Economics Group, Wageningen University and Research Centre, P.O. Box 8130, 6700 EW Wageningen, The Netherlands
G. VAN DUINKERKEN
Affiliation:
Livestock Research, Wageningen University and Research Centre, P.O. Box 65, 8200 AB Lelystad, The Netherlands
A. G. J. M. OUDE LANSINK
Affiliation:
Business Economics Group, Wageningen University and Research Centre, P.O. Box 8130, 6700 EW Wageningen, The Netherlands
*
*To whom all correspondence should be addressed. Email: geert.andre@wur.nl
Rights & Permissions [Opens in a new window]

Summary

Automated feeding and milking of dairy cows enables the application of individual cow settings for concentrate supply and milking frequency. Currently, general settings are used, based on knowledge about energy and nutrient requirements in relation to milk production at the group level. Individual settings, based on the actual individual response in milk yield, have the potential for a marked increase in economic profits. In the present study, adaptive dynamic models for online estimation of milk yield response to concentrate intake and length of milking interval are evaluated. The parameters in these models may change over time and are updated through a Bayesian approach for online analysis of time series. The main use of dynamic models lies in their ability to determine economically optimal settings for concentrate intake and milking interval length for individual cows at any day in lactation. Three adaptive dynamic models are evaluated, a model with linear terms for concentrate intake and length of milking interval, a model that also comprises quadratic terms, and an enhanced model (EM) in order to obtain more stable parameter estimates. The linear model is useful only for forecasting milk production and the estimated parameters of the quadratic model were found to be unstable. The parsimony of the EM leads to far more stable parameter estimates. It is shown that the EM is suitable for control and monitoring, and therefore promises to be a valuable tool for application within precision livestock farming.

Information

Type
Modelling Animal Systems
Copyright
Copyright © Cambridge University Press 2011
Figure 0

Fig. 1.Daily concentrate intake (kg/day) during lactation (DIM) for the randomly selected cow.

Figure 1

Fig. 2.Milk yield (kg/day) during lactation (DIM) for models T1, T2 and EM. Observations (points), forecasts (centre line) and 90% probability interval (upper and lower lines), for the randomly selected cow.

Figure 2

Fig. 3.Standardized forecast errors v. DIM for models T1, T2 and EM. Normal error (.), potential outlier (×) and signal for deterioration (+).

Figure 3

Table 1.Average goodness-of-fit statistics for the different models based on the results of 15 cows. The range over the 15 cows is given in parentheses

Figure 4

Fig. 4.Estimated observation variance during lactation (DIM) for models T1, T2 and EM, for the randomly selected cow.

Figure 5

Fig. 5.Development of the estimated intercept during lactation (DIM), parameter α2 in models T1, T2 and EM, for the randomly selected cow. Online (points), retrospective (centre line) and 90% confidence interval of the retrospective parameter estimates (lower and upper lines).

Figure 6

Fig. 6.Development of parameter estimates α4, α5 and α6 during lactation (DIM) in model EM, for the randomly selected cow. Online (points), retrospective (centre line) and 90% confidence interval of the retrospective parameter estimates (lower and upper lines).

Figure 7

Table 2.Averaged condition numbers of the correlation matrix of the parameter estimates for models T1, T2 and EM, including the range for the 15 cows in parentheses

Figure 8

Fig. 7.Estimated response curve of daily milk yield (kg/day, Y-axes) for one cow milked 2·85 times/day on concentrate intake (kg, X-axes) at 50, 100, 150, 200 and 250 DIM for models T1 (solid line), T2 (dashed line) and EM (dotted line).

Figure 9

Fig. 8.Estimated response curve of daily milk yield (kg/day, Y-axes) on number of milkings per day (X-axes) at 50, 100, 150, 200 and 250 DIM for models T1 (solid line), T2 (dashed line) and EM (dotted line).

Figure 10

Table 3.Parameter estimates of model EM for 15 cows at 100 and 200 DIM. Standard errors in parentheses are given for the first primi- and multiparous cow

Figure 11

Fig. 9.Histogram of forecast errors (kg milk/day) classified as normal (white), signal for deterioration (grey) or potential outlier (black) for all data of the 15 cows together.