Piglet mortality is a current issue in the pig production and the large variability between herds suggests a management component to the mortality. Studies show that the mortality may be reduced by the supervision of farrowing or through climate regulation in the farrowing pens. However, this is possible only if the farrowing time is known and thus provides sufficient time for the management to make and execute the decisions. The gestation period of a sow is approximately 115 (SD=2) days. However, an initial cost-benefit analysis recommended increased precision in the prediction of farrowing to make the increased management efforts cost-effective for the pig producer. Recently, a wide range of sensor technology have become available to monitor the behavioural and physiological changes of sow. Evidence show that appropriate utilization of sensor technology may increase the precision of prediction of onset of farrowing. Prediction is feasible only if the prediction is online and automated. The thesis is focused on constructing a system that can give predictions about the expected time to farrowing of individual sows based on automatic sensor recordings such as water consumption, video based activity measurements, and photo-cells based activity measurements. The warnings could serve to activate the floor heating system to ensure a sufficiently high temperature for the new born piglets, as well as to help the farmer to organize extra surveillance around farrowing. The thesis is based on three submitted manuscripts, describing the system at different stages. The kernel in the thesis is a Markov process with four subsequent states Before Nest-Building, Nest-Building, Resting and the absorbing state Farrowing; the states were selected based on ethological knowledge about sow behaviour. However, the sojourn time distribution in each state is not exponential. Therefore a continuous time discrete state semi-Markov process based on a Phase-Type distribution (in this case Erlang distributed) was formulated. Finally, the Markov process was transformed to a discrete time process. A Hidden Markov Model (HMM) was used for this process. The model is called Hidden Phase-type Markov Model (HPMM), and the time steps corresponded to each updating with sensor information at which the time to farrowing was predicted. The POMDP model for optimizing the floor-heat regulation system was used to demonstrate the decision support tool as an extension to the prediction algorithm. The tools for the prediction of onset of farrowing, estimation of HPMM and optimal decision making provides a framework for handling a large amount of sensor data available and gives an overview of how to integrate information from several sensors on the pen level. The complexity of the models imply that the prediction algorithm and decision tool may be run on the herd level computer; whereas the parameters were estimated on the central level computers.

1. Prediction Estimation Decision Tool Final Remarks
Methods for Sensor Based Farrowing Prediction and
Floor-heat Regulation
The Intelligent Farrowing Pen
Ph.D. Dissertation
Aparna U.
Dept. of Animal Science
Aarhus University
Denmark
18 March, 2014
Aparna U. The Intelligent Farrowing Pen 1 / 42

2. Prediction Estimation Decision Tool Final Remarks
Objective
To develop and validate an automated system that
monitors the pre-parturition behaviour of the sow in the
farrowing pen,
predicts the onset of farrowing
would further help the farm manager to optimize the decisions
related to parturition and post-parturition - e.g. optimal
oor-heat regulation system
by integrating several sensor information
purpose
to reduce the piglet mortality
*Study was supported by The Danish National Advanced Technology
Foundation
Aparna U. The Intelligent Farrowing Pen 2 / 42

3. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm

4. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy

5. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters

6. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters

7. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm

8. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm

9. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm Estimation Algorithm

10. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm Estimation Algorithm
Optimal Floor-Heat Regulation System

11. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm Estimation Algorithm
Optimal Floor-Heat Regulation SystemAparna U. The Intelligent Farrowing Pen 3 / 42

12. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm
Aparna U. The Intelligent Farrowing Pen 4 / 42

13. Prediction Estimation Decision Tool Final Remarks
Gestation Period
Behavioural States
Mating Farrowing
115±2 days
Aparna U. The Intelligent Farrowing Pen 5 / 42

14. Prediction Estimation Decision Tool Final Remarks
Gestation Period
Behavioural States
Mating Farrowing
115±2 days
Nest-Building
Aparna U. The Intelligent Farrowing Pen 5 / 42

15. Prediction Estimation Decision Tool Final Remarks
Gestation Period
Behavioural States
Mating Farrowing
115±2 days
Nest-Building
Resting
Aparna U. The Intelligent Farrowing Pen 5 / 42

16. Prediction Estimation Decision Tool Final Remarks
Gestation Period
Behavioural States
Mating Farrowing
115±2 days
Nest-Building
RestingBefore Nest-Building
Aparna U. The Intelligent Farrowing Pen 5 / 42

17. Prediction Estimation Decision Tool Final Remarks
Gestation Period
Behavioural States
Mating Farrowing
115±2 days
Nest-Building
RestingBefore Nest-Building
110 days Into Farrowing
Pen
Aparna U. The Intelligent Farrowing Pen 5 / 42

18. Prediction Estimation Decision Tool Final Remarks
Farrowing Pen
Number of sows observed: 64
Aparna U. The Intelligent Farrowing Pen 6 / 42

19. Prediction Estimation Decision Tool Final Remarks
Sensor set-up at the pen level
Number of sows observed: 64
Aparna U. The Intelligent Farrowing Pen 7 / 42

20. Prediction Estimation Decision Tool Final Remarks
Online Recording of Sensor Observation for a Sow - An idea
Water consumption
Aparna U. The Intelligent Farrowing Pen 8 / 42
qq
q
q
q
q
qqqq
q
qqq
q
q
q
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
q
q
q
qqq
q
qqq
q
q
q
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
qq
q
q
q
qqqqqq
q
qqq
q
q
q
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
q
qq
q
q
q
qqqqq
q
q
q
qqqq
114 115 116 117
02468
days since mating
waterconsumption(log)
17:30:00

21. Prediction Estimation Decision Tool Final Remarks
Sensor observation for a sow - meanActivity
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
qqq
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
qqq
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
q
q
qq
q
qq
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
qq
q
qqq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
qq
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
qq
q
q
q
q
q
q
q
q
q
q
q
q
q
q
q
qq
q
q
q
q
q
q
345678910
Day of Farrowing:2009−01−15 03:47:09
Time Of Day
meanActivity(logtransformed)
Jan 06 Jan 08 Jan 10 Jan 12 Jan 14
case−1
Aparna U. The Intelligent Farrowing Pen 9 / 42

22. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
time line:-
t0
t
Tt: remaining time to onset of farrowing
Stochastic process

23. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
time line:-
t0
t
Tt: remaining time to onset of farrowing
Stochastic process
?Markov process

24. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
time line:-
t0
t
Tt: remaining time to onset of farrowing
Stochastic process
?Markov process
(ω1, ς1) (ω2, ς2) (ω3, ς3)

25. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
time line:-
t0
t
Tt: remaining time to onset of farrowing
Stochastic process
(ω1, ς1) (ω2, ς2) (ω3, ς3)
Markov process semi-Markov process

26. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
time line:-
t0
t
Tt: remaining time to onset of farrowing
Stochastic process
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
Markov process semi-Markov process

27. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

28. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

29. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

30. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

31. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

32. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

33. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

34. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

35. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

36. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

37. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

38. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

39. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

40. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

41. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

42. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
States into smaller divisions - Behavioural Phases
'phases' reect the time since mating

43. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
(m1, λ1) (m2, λ2) (m3, λ3)
'semi' Markov Process over phases
Aparna U. The Intelligent Farrowing Pen 10 / 42

44. Prediction Estimation Decision Tool Final Remarks
Farrowing process as a Markov Model
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
(ω1, ς1) (ω2, ς2) (ω3, ς3)Erlang
(m1, λ1) (m2, λ2) (m3, λ3)
'semi' Markov Process over phases
Phase-type
Convolution of three Phase-type distributions
Aparna U. The Intelligent Farrowing Pen 10 / 42

45. Prediction Estimation Decision Tool Final Remarks
Farrowing and Prediction Process
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
'Markov Model'
Aparna U. The Intelligent Farrowing Pen 11 / 42

46. Prediction Estimation Decision Tool Final Remarks
Farrowing and Prediction Process
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
'Markov Model'
Prediction process
t0 tI t2 t3
- discrete time points
Aparna U. The Intelligent Farrowing Pen 11 / 42

47. Prediction Estimation Decision Tool Final Remarks
Farrowing and Prediction Process
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
'Hidden' + 'Markov Model'αt : distribution of phases
Tt :time to onset of farrowing ∼ PH(αt, S)
Prediction process
t0 tI t2 t3
- discrete time points
Aparna U. The Intelligent Farrowing Pen 11 / 42

48. Prediction Estimation Decision Tool Final Remarks
Farrowing and Prediction Process
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
'Hidden' + 'Markov Model'αt : distribution of phases
Tt :time to onset of farrowing ∼ PH(αt, S)
Prediction process
t0 tI t2 t3
- discrete time points
Sensor
Pr(Yt | Phaset)
YI Y2 Y3
Aparna U. The Intelligent Farrowing Pen 11 / 42

49. Prediction Estimation Decision Tool Final Remarks
Farrowing and Prediction Process
Farrowing process - continuous time
Before Nest-Building Nest-Building Resting F
'Hidden' + 'Markov Model'αt : distribution of phases
Tt :time to onset of farrowing ∼ PH(αt, S)
Prediction process
t0 tI t2 t3 tN
- discrete time points
Sensor
Pr(Yt | Phaset)
YI Y2 Y3
Aparna U. The Intelligent Farrowing Pen 11 / 42

50. Prediction Estimation Decision Tool Final Remarks
Prediction of Onset of Farrowing
t0 tI t t + δ tI + Nδ
?
Yt
αt+δ: distribution of phases in the next prediction point
Prediction of αt+δ
At each prediction point,
1 calculates αt+δ using time transition (Markov chain)
2 updates αt+δ using the sensor information
Aparna U. The Intelligent Farrowing Pen 12 / 42

51. Prediction Estimation Decision Tool Final Remarks
Prediction of Distribution of Phases αt+δ
Aparna U. The Intelligent Farrowing Pen 13 / 42
400 600 800 1000
0.000.020.040.060.080.10
Behavioural Phases
probabilityofphases
Before
Nest−Building
Nest−Building
day−109.4

52. Prediction Estimation Decision Tool Final Remarks
Prediction of Onset of Farrowing
Prediction of αt+δ
At each prediction point,
1 calculates αt+δ using time transition (Markov chain)
2 updates αt+δ using the sensor information
Tt+δ: time to onset of farrowing ∼ PH(αt+δ, S)
Statistical measures...
Expected time to onset of farrowing
Aparna U. The Intelligent Farrowing Pen 14 / 42

53. Prediction Estimation Decision Tool Final Remarks
Prediction of Onset of Farrowing
Prediction of αt+δ
At each prediction point,
1 calculates αt+δ using time transition (Markov chain)
2 updates αt+δ using the sensor information
Tt+δ: time to onset of farrowing ∼ PH(αt+δ, S)
Statistical measures...
Expected time to onset of farrowing
Probability of onset of farrowing in 12 hours
Aparna U. The Intelligent Farrowing Pen 14 / 42

54. Prediction Estimation Decision Tool Final Remarks
Validating the Prediction - how???
Aparna U. The Intelligent Farrowing Pen 15 / 42

55. Prediction Estimation Decision Tool Final Remarks
Validating the Prediction - how???
Aparna U. The Intelligent Farrowing Pen 15 / 42

56. Prediction Estimation Decision Tool Final Remarks
Online Prediction Curve - an example
Aparna U. The Intelligent Farrowing Pen 16 / 42
q
107 108 109 110 111
050100150
days since mating
ETF(hours)

57. Prediction Estimation Decision Tool Final Remarks
Warning periods - success Vs failure
Aparna U. The Intelligent Farrowing Pen 17 / 42

58. Prediction Estimation Decision Tool Final Remarks
Example of False-warning Period
Aparna U. The Intelligent Farrowing Pen 18 / 42
qqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqq
q
q
q
q
q
q
qq
q
qq
q
q
q
q
q
q
qqq
qqqqqqqqqq
112 113 114 115
050100150200250
days since mating
ETF(hours)

59. Prediction Estimation Decision Tool Final Remarks
Validating the Prediction Algorithm
Sensor
Sample True Warning Dur.
Error
size Warnings (hours)
% Mean SD (hours)
Water 38 21 11.7 2.2 3.4
Video 55 98 14.4 12.5 1.6
Water-Video 35 97 11.5 4.6 0.7
*threshold set at 12 hours
Aparna U. The Intelligent Farrowing Pen 19 / 42

60. Prediction Estimation Decision Tool Final Remarks
Validating the Prediction Algorithm
Sensor
Sample True Warning Dur.
Error
size Warnings (hours)
% Mean SD (hours)
Water 38 21 11.7 2.2 3.4
Video 55 98 14.4 12.5 1.6
Water-Video 35 97 11.5 4.6 0.7
*threshold set at 12 hours
Aparna U. The Intelligent Farrowing Pen 19 / 42

61. Prediction Estimation Decision Tool Final Remarks
Validating the Prediction Algorithm
Sensor
Sample True Warning Dur.
Error
size Warnings (hours)
% Mean SD (hours)
Water 38 21 11.7 2.2 3.4
Video 55 98 14.4 12.5 1.6
Water-Video 35 97 11.5 4.6 0.7
*threshold set at 12 hours
Aparna U. The Intelligent Farrowing Pen 19 / 42

62. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm

63. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Estimation Algorithm
Aparna U. The Intelligent Farrowing Pen 20 / 42

64. Prediction Estimation Decision Tool Final Remarks
Estimation of HPMM Parameters
by maximizing the likelihood function
Aparna U. The Intelligent Farrowing Pen 21 / 42

65. Prediction Estimation Decision Tool Final Remarks
Estimation of HPMM Parameters
by maximizing the likelihood function
Stochastic Expectation-Maximization algorithm (SEM algorithm)
-iterative method
Uses time of mating and farrowing information in addition to
sensor information
Phases are allocated by weighted sampling - Stochastic
Aparna U. The Intelligent Farrowing Pen 21 / 42

66. Prediction Estimation Decision Tool Final Remarks
Estimated Sojourn time distribution
Behavioural State
Duration Number Rate
(hours) of (per hour)
Mean SD Phases
Before Nest-Building 751.20∗ 29.58 645 0.86
Nest-Building 17.02 0.80 458 26.91
Resting 0.53 0.22 6 11.40
Gestation period, days 117 1.2 1109 -
*in addition to 85 days
Aparna U. The Intelligent Farrowing Pen 22 / 42

67. Prediction Estimation Decision Tool Final Remarks
Estimated Sojourn time distribution
Behavioural State
Duration Number Rate
(hours) of (per hour)
Mean SD Phases
Before Nest-Building 751.20∗ 29.58 645 0.86
Nest-Building 17.02 0.80 458 26.91
Resting 0.53 0.22 6 11.40
Gestation period, days 117 1.2 1109 -
*in addition to 85 days
Aparna U. The Intelligent Farrowing Pen 22 / 42

68. Prediction Estimation Decision Tool Final Remarks
Estimated Sojourn time distribution
Behavioural State
Duration Number Rate
(hours) of (per hour)
Mean SD Phases
Before Nest-Building 751.20∗ 29.58 645 0.86
Nest-Building 17.02 0.80 458 26.91
Resting 0.53 0.22 6 11.40
Gestation period, days 117 1.2 1109 -
*in addition to 85 days
Aparna U. The Intelligent Farrowing Pen 22 / 42

69. Prediction Estimation Decision Tool Final Remarks
Estimated Sojourn time distribution
Behavioural State
Duration Number Rate
(hours) of (per hour)
Mean SD Phases
Before Nest-Building 751.20∗ 29.58 645 0.86
Nest-Building 17.02 0.80 458 26.91
Resting 0.53 0.22 6 11.40
Gestation period, days 117 1.2 1109 -
*in addition to 85 days
Aparna U. The Intelligent Farrowing Pen 22 / 42

70. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
Aparna U. The Intelligent Farrowing Pen 23 / 42

71. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
diurnal rhythm conditioned on the Phase/State
Aparna U. The Intelligent Farrowing Pen 23 / 42

72. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
diurnal rhythm conditioned on the Phase/State
model selection
Aparna U. The Intelligent Farrowing Pen 23 / 42

73. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
diurnal rhythm conditioned on the Phase/State
model selection
dependency of the variables
Aparna U. The Intelligent Farrowing Pen 23 / 42

74. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
diurnal rhythm conditioned on the Phase/State
model selection
dependency of the variables
dependency on the phases
Aparna U. The Intelligent Farrowing Pen 23 / 42

75. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observation
meanActivity - sdActivity - grid activity
Pr(Yt | Si ) ∼ N(µ
(Y )
i , σ2
i
(Y )
)ζ
Simple linear model
sine-cosine functions - harmonic variables
Aparna U. The Intelligent Farrowing Pen 24 / 42

76. Prediction Estimation Decision Tool Final Remarks
Conditional Distribution of Sensor Observations
0 5 10 15 20
56789
Mean Level of meanActivity
Time (hrs)
meanActivity
q q
q
q
q
q q q q
q
q
q
q
q
q q q
q
q
q
q
q q q
q
q
q
q
q q
q
q
q
q
q q q
q
q
q q q
q
q
q
q
q
q
q
state−1
state−2
state−3
Aparna U. The Intelligent Farrowing Pen 25 / 42

77. Prediction Estimation Decision Tool Final Remarks
Pattern of Water Consumption
*
*
*
*
*
*
*
*
*
*****
*
***
**
***********
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
**
*
*
*
*
*
*
*
***
*
*****
*
**********
*
**
*
**
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
********
*
****
*
********
*
**
*
*
*
*
*
*
**
*
**
**
*
**
*
*
*
*
*
*
*
*
*
*
*
**
*
*****
*
****
*
*
********
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
**
*
**
*
*
*
*
*
*
******
*
******
*
***
*
*
*
***
**
*
*
*
*
*
*
*
*
**
*
*
*
**
**
*
**
*
****
*
*********
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
**
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
**
*
*
*
*
**
**
**
*
*
*
*
*
**
*
***
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
***********
**
****
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
******
**
*
*
**********
*
*
*
*
***
*
*
*
***
*
*
*
*
*
*
*
*
*
*
**************
*
*
******
*
*
**
*
*
*
*
*
*
**
*
**
*
*
*
*
*
*
*
*
*
*
***
106 108 110 112 114 116 118
02468
sinceMating
LWater
*
*
**
*
*
*
*
*
*****
*
***
**
***********
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
**
*
*
*
*
*
*
*
***
*
*****
*
**********
*
**
*
**
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
********
*
****
*
********
*
**
*
*
*
*
**
**
*
**
**
*
**
*
*
*
*
**
*
*
*
*
*
**
*
*****
*
****
*
*
********
*
*
*
*
*
*
*
*
*
*
*
*
*
**
**
*
**
*
**
*
*
*
*
*
*
******
*
******
*
***
*
*
*
***
**
*
*
*
*
*
*
*
*
**
*
*
*
**
**
*
**
*
****
*
*********
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
**
*
**
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
**
*
*
*
*
**
**
**
*
*
*
*
*
**
*
***
*
*
*
*
*
***
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
***********
**
****
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
******
**
*
*
**********
*
*
*
*
***
*
*
*
***
*
*
*
*
*
*
**
*
*
**************
*
Aparna U. The Intelligent Farrowing Pen 26 / 42

78. Prediction Estimation Decision Tool Final Remarks
Pattern of Water Consumption
*
*
*
*
*
*
*
*
*
*****
*
***
**
***********
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
**
*
*
*
*
*
*
*
***
*
*****
*
**********
*
**
*
**
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
********
*
****
*
********
*
**
*
*
*
*
*
*
**
*
**
**
*
**
*
*
*
*
*
*
*
*
*
*
*
**
*
*****
*
****
*
*
********
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
**
*
**
*
*
*
*
*
*
******
*
******
*
***
*
*
*
***
**
*
*
*
*
*
*
*
*
**
*
*
*
**
**
*
**
*
****
*
*********
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
**
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
**
*
*
*
*
**
**
**
*
*
*
*
*
**
*
***
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
***********
**
****
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
******
**
*
*
**********
*
*
*
*
***
*
*
*
***
*
*
*
*
*
*
*
*
*
*
**************
*
*
******
*
*
**
*
*
*
*
*
*
**
*
**
*
*
*
*
*
*
*
*
*
*
***
106 108 110 112 114 116 118
02468
sinceMating
LWater
*
*
**
*
*
*
*
*
*****
*
***
**
***********
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
**
*
*
*
*
*
*
*
***
*
*****
*
**********
*
**
*
**
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
*
*
*
********
*
****
*
********
*
**
*
*
*
*
**
**
*
**
**
*
**
*
*
*
*
**
*
*
*
*
*
**
*
*****
*
****
*
*
********
*
*
*
*
*
*
*
*
*
*
*
*
*
**
**
*
**
*
**
*
*
*
*
*
*
******
*
******
*
***
*
*
*
***
**
*
*
*
*
*
*
*
*
**
*
*
*
**
**
*
**
*
****
*
*********
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
**
*
**
*
**
*
*
*
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
**
*
**
*
*
*
*
**
**
**
*
*
*
*
*
**
*
***
*
*
*
*
*
***
*
*
*
*
*
*
*
*
***
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
*
*
*
*
*
***********
**
****
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
**
*
******
**
*
*
**********
*
*
*
*
***
*
*
*
***
*
*
*
*
*
*
**
*
*
**************
*
Aparna U. The Intelligent Farrowing Pen 26 / 42

79. Prediction Estimation Decision Tool Final Remarks
Probability of Water Consumption
at given time of day and state
0 5 10 15 20
0.00.20.40.60.8
time of day (hours)
probabilityofdrinking
q q
q
q
q
q
q
q q q q q q
q
q
q
q
q
q q
q
q
q
q q
q
q
q
q
q
q
q
q
q
q
q
q
q
q q q q q q q q q q
state−1
state−2
Aparna U. The Intelligent Farrowing Pen 27 / 42

80. Prediction Estimation Decision Tool Final Remarks
Estimation of HPMM Parameters
Challenges with conditional model
changing pattern with calender time - Behavioural
Phases/States
diurnal rhythm conditioned on the Phase/State
model selection
dependency of the variables ?
dependency on the phases ?
Conclusion
Duration of the Nest-Building state is similar to other studies
Computational time - 26mins per iteration
Aparna U. The Intelligent Farrowing Pen 28 / 42

81. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm

82. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Optimal Floor-Heat Regulation System

83. Prediction Estimation Decision Tool Final Remarks
Overview of the Study
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Optimal Floor-Heat Regulation System
Aparna U. The Intelligent Farrowing Pen 29 / 42

84. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Malmkvist et. al. (2006)→ survival of one extra piglet per litter

85. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Malmkvist et. al. (2006)→ survival of one extra piglet per litter

86. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Malmkvist et. al. (2006)→ survival of one extra piglet per litter

87. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Malmkvist et. al. (2006)→ survival of one extra piglet per litter

88. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Malmkvist et. al. (2006)→ survival of one extra piglet per litter

89. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Phase-(A)
Phase-(B)
Phase-(C)Malmkvist et. al. (2006)→ survival of one extra piglet per litter
Aparna U. The Intelligent Farrowing Pen 30 / 42

90. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
HeatOn
HeatO
Farrowing
Phase-(A)
Phase-(B)
Phase-(C)Malmkvist et. al. (2006)→ survival of one extra piglet per litter
Aparna U. The Intelligent Farrowing Pen 30 / 42

91. Prediction Estimation Decision Tool Final Remarks
Floor-heat Regulation on Pen Level
Time since onset of farrowing
Floor-temperature
−20 0 20 40 60 80
10
15
20
25
30
35
40
Farrowing
HeatOn
HeatO
Phase-(A)
Phase-(B)
Phase-(C)Aparna U. The Intelligent Farrowing Pen 30 / 42

92. Prediction Estimation Decision Tool Final Remarks
How to choose threshold???
Aparna U. The Intelligent Farrowing Pen 31 / 42

93. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
Aparna U. The Intelligent Farrowing Pen 32 / 42

94. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
Aparna U. The Intelligent Farrowing Pen 32 / 42

95. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2
U1 U2
Y1
C1
d1
H1
Aparna U. The Intelligent Farrowing Pen 32 / 42

96. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2
U1 U2
Y1
C1 C2
d1
H1
Aparna U. The Intelligent Farrowing Pen 32 / 42

97. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2
U1 U2
Y1
C1 C2
d1 d2
H1
Aparna U. The Intelligent Farrowing Pen 32 / 42

98. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2
U1 U2
Y1
C1 C2
d1 d2
H1 H2
Aparna U. The Intelligent Farrowing Pen 32 / 42

99. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2 t3
U1 U2 U3
Y1 Y2
C1 C2
d1 d2
H1 H2
Aparna U. The Intelligent Farrowing Pen 32 / 42

100. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2 t3
U1 U2 U3
Y1 Y2
C1 C2 C3
d1 d2
H1 H2
Aparna U. The Intelligent Farrowing Pen 32 / 42

101. Prediction Estimation Decision Tool Final Remarks
Floor-heat regulation System
Partially Observable Markov Decision Process (POMDP)
belief
time
phase
sensor obs.
temperature
decision
heating cost
t1 t2 tF−1 tF tF+1
U1 U2 UF−1 UF UF+1
Y1 Y2 YF−1
C1 C2 CF−1 CF
d1 d2 dF−1
H1 H2 HF−1
I
Aparna U. The Intelligent Farrowing Pen 32 / 42

102. Prediction Estimation Decision Tool Final Remarks
Approximate Solution
Approximate POMDP solution
Markov Decision Process - optimization for known phases
Value Iteration Method
Optimize the total expected utility
w.r.t. phase number and oor-temperature
Greedy Strategies
Aparna U. The Intelligent Farrowing Pen 33 / 42

103. Prediction Estimation Decision Tool Final Remarks
MDP look-up decision table
Approximate POMDP Solution
Floor-temperature Behavioural Phase ∆Q(opt) Decision(opt)
21 Phase-1
21 Phase-2
...
...
21 Phase-1000
22 Phase-1
22 Phase-2
...
...
22 Phase-1000
...
...
*∆Q(opt)
: reward of heating
Aparna U. The Intelligent Farrowing Pen 34 / 42

104. Prediction Estimation Decision Tool Final Remarks
Approximating to POMDP
Decision Vs Belief state for the given oor-temperature
400 600 800 1000
0.000.010.020.030.040.050.060.07
phases
αt
t=112
t=114
t=117.6
t=118.1
t=118.3
Aparna U. The Intelligent Farrowing Pen 35 / 42

105. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Aparna U. The Intelligent Farrowing Pen 36 / 42

106. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Most likely phase
Aparna U. The Intelligent Farrowing Pen 36 / 42

107. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Most likely phase
Random phase
Aparna U. The Intelligent Farrowing Pen 36 / 42

108. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Most likely phase
Random phase
Voting
Aparna U. The Intelligent Farrowing Pen 36 / 42

109. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Most likely phase
Random phase
Voting
Random action
Aparna U. The Intelligent Farrowing Pen 36 / 42

110. Prediction Estimation Decision Tool Final Remarks
POMDP Greedy Strategies
QMDP - expectation over all the phases
Most likely phase
Random phase
Voting
Random action
*Above greedy strategies return almost identical rewards.
Aparna U. The Intelligent Farrowing Pen 36 / 42

111. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
Aparna U. The Intelligent Farrowing Pen 37 / 42

112. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
energy supply,
Aparna U. The Intelligent Farrowing Pen 37 / 42

113. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
energy supply, mortality model
Aparna U. The Intelligent Farrowing Pen 37 / 42

114. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
energy supply, mortality model and price of a piglet
Aparna U. The Intelligent Farrowing Pen 37 / 42

115. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
energy supply, mortality model and price of a piglet
POMDP strategy performed better than simple heuristic
strategy especially when the room-temperature and energy
supply was varied
Aparna U. The Intelligent Farrowing Pen 37 / 42

116. Prediction Estimation Decision Tool Final Remarks
Robustness
POMDP for Floor-heat Regulation Strategy
Decision tool is robust to the changes in room temperature,
energy supply, mortality model and price of a piglet
POMDP strategy performed better than simple heuristic
strategy especially when the room-temperature and energy
supply was varied
POMDP does not suggest to turn on the heater if it is not
benecial
Aparna U. The Intelligent Farrowing Pen 37 / 42

117. Prediction Estimation Decision Tool Final Remarks
Final Remarks...
Aparna U. The Intelligent Farrowing Pen 38 / 42

118. Prediction Estimation Decision Tool Final Remarks
Desired System - data management to decision
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm Estimation Algorithm
Optimal Floor-Heat Regulation Strategy

119. Prediction Estimation Decision Tool Final Remarks
Desired System - data management to decision
HerdLevelComputer
at the pen level
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Prediction Algorithm Estimation Algorithm
Optimal Floor-Heat Regulation StrategyAparna U. The Intelligent Farrowing Pen 39 / 42

120. Prediction Estimation Decision Tool Final Remarks
Future works...
Studying the corners of the desired system
Aparna U. The Intelligent Farrowing Pen 40 / 42

121. Prediction Estimation Decision Tool Final Remarks
Future works...
Studying the corners of the desired system
Including farmer's observation on farrowing into model
Aparna U. The Intelligent Farrowing Pen 40 / 42

122. Prediction Estimation Decision Tool Final Remarks
Future works...
Studying the corners of the desired system
Including farmer's observation on farrowing into model
Improving the methods and the calculation speed of
Estimation algorithm
Aparna U. The Intelligent Farrowing Pen 40 / 42

123. Prediction Estimation Decision Tool Final Remarks
Future works...
Studying the corners of the desired system
Including farmer's observation on farrowing into model
Improving the methods and the calculation speed of
Estimation algorithm
Similar decision support system for other applications
Aparna U. The Intelligent Farrowing Pen 40 / 42

124. Prediction Estimation Decision Tool Final Remarks
Take home message...
Contributions
Framework of an automated system starting from data
management to decision The Intelligent Farrowing Pen
Model for monitoring the pre-parturition behaviour of an
individual sow
Prediction of onset of farrowing - can directly calculate the
expected time to farrowing for an individual sow
Prediction model as the kernel of a decision support system
Modelling sows diurnal rhythm in sensor observations
Integrating information from several sensors
Aparna U. The Intelligent Farrowing Pen 41 / 42

125. Prediction Estimation Decision Tool Final Remarks
The Intelligent Farrowing Pen
HerdLevelComputer
Sensor
Observations
Herd
Database
Prediction
Algorithm
Warning
Strategy
CentralLevelComputer
Historical Data
Estimation
Algorithm
Herd Specic
Parameters
Heat Parameters
Mortality
Model
Costs
Optimization
Algorithm
Aparna U. The Intelligent Farrowing Pen 42 / 42