philip techniques powerpoint

Upload: larn-nina-raymundo

Post on 19-Feb-2018

220 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/23/2019 Philip Techniques Powerpoint

    1/15

    Comparison of modeling techniquemilk-production forecasting

    M. D. Murphy ,* M. J. OMahony , L. Shalloo ,* P. French ,* and

    J. Upton * 1 *

  • 7/23/2019 Philip Techniques Powerpoint

    2/15

    ABSTRACTThe objective of this study was to assess the

    suitability of 3 different modeling techniques for tprediction of total daily herd milk yield from a her140 lactating pasture-based dairy cows over varyingforecast horions!

    " nonlinear auto-regressive model with e#ogenousa static artificial neural network$ and a multiple lregression model were developed using 3 yr of histormilk-production data!

  • 7/23/2019 Philip Techniques Powerpoint

    3/15

    INTRODCTION%ilk production from pasture-based dairy cows

    susceptible to variation due to seasonality ofpasture production &"dediran et al!$ '01'($ graiconditions &)audracco et al!$ '01'($ disease&*ollard et al!$ '000($ nutritional interventions

    &+olver and %uller$ 1,,($ and other disturbances&.lori et al!$ 1,,,/ Tekerli et al!$ '000(!

  • 7/23/2019 Philip Techniques Powerpoint

    4/15

    INTRODCTION

    The aim of this study was to assess thesuitability of a static neural network &SANN), a model, and a nonlinear auto regressive model withe#ogenous input &NARX) for the prediction of totaldaily herd milk yield &DHMY) over varying forecas

    horions!The most successful model was selected accordi

    to its abilities to generate the most accurateforecast using very limited training data in lowvolumes over a long- &30 d($ medium- &30 to 0 dand short-term &10 d( horion!

  • 7/23/2019 Philip Techniques Powerpoint

    5/15

    !aterials and methodata

    *ollection 1! ata were collected from a research farm in thesouth of 2reland for a period of 4 yrs &'00'010(!

    '! aily herd milk yield &liters( and number of cowsmilked on that corresponding herd 2% was collected!

    3! %ilk yields were recorded from a conventionalherringbone swingover milking parlor using 2nternation*ommittee for "nimal 5ecording approved milk meters!

    4! The model was set at herd level and evaluated bycomparing daily milk yields across a herd of 140 pastubased 6olstein-7riesian cattle!

    ! The milking season of '010 was selected as thetarget prediction horion and the previous 3 yr of dat

    were used to train the model!

  • 7/23/2019 Philip Techniques Powerpoint

    6/15

    !aterials and method

    %odel 2nputs2n previous studies certain variables were found to

    have an influence on milk production8 season of calvingclimatic conditions$ number of 2%$ and stocking rate!

    2n this study the farm graing area remained statiwhereas the number of cows graing varied throughout t

    year! 9imilarly$ the season of calving &spring( was keptconstant in the herd over several years!

    6ence the total herd milk production behaves in acyclical pattern!

  • 7/23/2019 Philip Techniques Powerpoint

    7/15

    !aterials and method

    :eural:etworks"n ":: is a mathematical model whose operating

    principle is based on biological neural networks!

    The ":: architecture comprises a series ofinterconnected layered neurons through which inputs arprocessed!

    These inputs values are multiplied by the synaptiweights$ which represent the strength of the neuralconnections!

  • 7/23/2019 Philip Techniques Powerpoint

    8/15

    !aterials and method

    %ultiple ;inear 5egression %odel

    ;inear regression models characterie theassociation between a dependent and independentvariable!

    The relationship between these variables can bee#pressed in a '-dimensional space!

    6owever$ few outputs can be accurately profiledusing just one input!

    %ost real world systems are controlled by multipleinputs!

  • 7/23/2019 Philip Techniques Powerpoint

    9/15

    Result

    Totalied forecast errors of the 3 models over thecycle with 4 different piecewise horions ranging fromd!

  • 7/23/2019 Philip Techniques Powerpoint

    10/15

    Discussion

    %odel model to correct its projected trajectory over thne#t horion based on information from the previoushorion!

  • 7/23/2019 Philip Techniques Powerpoint

    11/15

    Discussion

    %odel

  • 7/23/2019 Philip Techniques Powerpoint

    12/15

    Discussion

    %odel

  • 7/23/2019 Philip Techniques Powerpoint

    13/15

    Discussion

    %odel

  • 7/23/2019 Philip Techniques Powerpoint

    14/15

    Conclusion

    The :"5> was shown to be the most effective milk-production model! it was moreaccurate than the 9":: anfor each moving horion and over the majority of indivprediction ranges &3!? of ranges(!

    7rom this research$ the :"5> model appears to be aalternative to conventional regression models!

  • 7/23/2019 Philip Techniques Powerpoint

    15/15

    thanks"B#$ Carl %hilip Bumanlag