So our our third and final final
if you know the theme of work with

the energy and Dr Doherty over
here is there practical advice

giving leaving behind.

OK so I know this is the last presentation
for just a couple words as I had the great

pleasure of working with this group they
were hard workers very responsive and

besides had a lot of fun which makes
the semester go a lot faster for

the advisor as well.

The client is here.

Thank you Mike for

everything that you've done I saw him
somewhere there you are right in front me.

So they'll tell you more about the project
but they were asked to provide a forecast

for delivery of wind turbine parts
to different wind farms across

the country and I think they decided
to go a little bit further and

ended up giving them two
deliverables instead and

they'll tell you about the second
deliverable long with the forecast

will do no one figured

out in my name is I'm going to look into
their Who thought of the when drilling off

the decision strategy that were
conducted with jet engines by semester.

We're going to make the members
Osgar Marial Jackie some devil and

you know what I think I got that my
governor for his time and dedication to

a project and also welfare to do a better
doctor the whole picture of much for

software project due to
the process of delivering and

offering when there are components
across North America.

So we decided to do by just Boston to
do parts because the on road part and

the also part for the auto part which
is when the trucks carrying those

MacDermott components are in shipment or
are being shipped in delivery.

We've created a corpus into all
this focused into this information

from the window in most
implication provided

by gender to which Oscar is
going to talk about later and

also information from the World Wide Web
to create a group estimated to have a.

Liable for the shipments using this
improve the estimate the rival the crane

allocation sequencing tool of the way
the nets cast thinks this is an input and

also coming from information from G.M.
or G.M. to come up with a plan of the day

currently for the time of day the G.'s
A receiver is a person in charge of

allocating the different grades
to different jobs right they

thought because the math of
mathematical rigor of our panel.

There was that the savings of one
point I moved dollars from Gene or

G.E. is one of the pioneers in
the alternative energy sector

has installed the resistant
thousand with terminal weights and

thousand two there you can
see in the bar right there.

Sure they're very big components
they require specialized deliveries.

So this is a component.

Then they cover typical winter but
indulge eleven Janet directed thirty

one of this or thirty one wind farms and
each of those went

on foot approximately between eighty and
one hundred wind turbines and I want to

go into more of the thought about these
are all part of the process of the above.

Thank you so this is a high level view
about our own process it all begins with

G. energy in Atlanta Georgia where
the headquarters is the bid out contracts

to the carriers as well as negotiate
contracts with the customer wind farm

sites they relay that information to the
Elysee or the logistics execution Center

in New Greenville South Carolina the
Elysee outfit components the carriers and

then issues buildings to the carrier.

They also send a forecast two weeks
prior to G.E. field manager at

the before the G.E. you'll enjoy the
presence of forecasts and actual delivery

to the customer and the carrier then
deliver the component to the wind farm.

Is the G.E. field managers responsibility
however to communicate any discrepancies

to back their receipt for them to rectify.

So I think a moment right now to the final
tension because we'll be talking about

pensions rather presentation potential
is any time a little different.

Since Idol and unloading and loading
dock after delivery is due and that.

Expose the energy and the customer and
the main driver of the tension is the is

the forecast that said two weeks prior
because it serves as a rough estimate.

So this is the with so
many supplication the G.

energy is currently spearheading it's
right now a tool to use internally

on the top left you can see project
description right next to that you can

see the current weather conditions on the
truck right below that you can see G.P.S.

information component types and serial
numbers and right next to that you can see

a live map of all these trucks traveling
to their respective wind farms.

So this Tools very useful explain presents
to the user a lot of information sort

of very large information
in a very elegant way but

it's really up to the user to come
up with his own up plan of the day

through this experience as I'm
going you previously mentioned.

So that's where we step in to
provide that mathematical analysis.

So here's the rest of model we
constructed in order to improve E.T.A.'s

specifically look at these variables
The first is actual distance.

So right now the window
man except location you so

you click in distance from
the origin to the destination.

We use greater circle distances
between the historical G.P.S.

data plans to calculate that we
then look at the current weather

we look at the weather conditions
along the route as well when stations.

Another reason why we will get away
stations because these trucks go into

a category known as oversized loads and by
law if a truck has an oversized load they

have to stop at each and every win of each
and every one station along the route.

Now I want to replay talk about the
forecasting tool that I'll be mentioning

in a second the forecasting tool actually
calculates that actual distance and

weather variables using Google A.P.I.

is imposed why data from Google Maps and
Google web.

So here's the regression model built.

It's a component specific model.

Specifically we constructed out it for
planes in towers

because that's what we had the most
data on for the blades component.

We can say that the estimated
mean drive time is with them.

Forty six minutes and we still have
a ninety five percent confidence and

we did a similar analysis for the towers
where we estimated the mean drive time to

be plus or minus forty minutes with
ninety five percent confidence

we package this tool and
in the forecasting we back it all this up

all this map into the forecasting tool
with the forecasting tool basically

provide the user two functionality so the
first is the binder specific beyond well

it's all user to simply click
the button fill out the user form and

hit the run forecast by then and within
a few seconds that has to be the time of

arrival shows up in this case it's six
hours in two minutes or the user can

probably do is going to be using more out
than we just bought all the be all right.

Bills for a specific date.

So again like record books that
he filled out the user for and

they had to run forecast by him and within
a few minutes of whole list of the rivals

for that day populate So with that I'll
be passing in our two yards and mentally.

Talking about the ones I
process I thank you very much.

I ask.
So

essentially after we've calculated the
estimated times of arrival we begin what's

known as the onsite process in which each
of the individual trucks drive up to

their respective this nation that in
specialized cranes come in are full

of the oversized components from these
trucks as you can see in our little

animation now I'm going to talk about
this process and slightly more detail.

So they are suits different
members of our float

each are executed by three different but
one of three different types of cranes and

these are the heavy grains like grains and
shovel crews which aren't actually

grains are specialized crews they
deal with so after loading trucks.

Now each one of these different
types of articles will require one

crane to complete to bring suit complete
and there is even a single type of part of

the requires two different types
of Crane crews to complete and

basically this forms what we like
to call the crane allocation

problem in which you must allocate
all these shipments there.

I bring the day across the available
cranes to minimize attention cause for

that you can do as with

any mathematical formulation there are
several assumptions that come into play.

These are all his own views
are the main sanctions present for him.

I'll go over the three key ones.

So the first yourself into the greens
are operated independently and

as I mentioned this is due to
the fact there's only one article and

then requires two different types
of our crane crews to complete.

So we can actually decompose
the first of the large

problem into two smaller problems.

The second one suction is
that the travel times inside

only to find the buffalo times all
these screens are deterministic.

And if I was reasonable to make after
interviewing G.E. field managers and

the site's A.B.C. over the last
important for gas is the forecast.

More or other assumption is the forecasts
are hundred percent accurate.

No This may seem unreasonable but the fact
of the many areas that are forecasting

to all that are here previously discussed
can be recalculated multiple times during

the day and you can get more accurate
forecasts as the day progresses and

I'll touch more on that as we go
further into the presentation.

So the first form of
solution that we propose

to solve this problem is a network flow
model approach has a long network for

models of begins with the Source
know which represents the beginning

STATE OF THE with for for the day were
in this particular example the starting

position of a single character from there
of the crank an hour followed any one

of the shipments arriving during the day
here we have shipments one two and

three with shipment zero
representing the idle shipment so

that would indicate the crane
is just sitting at the pad and

isn't doing anything after completing
especially when the crane must process

progress through the remaining shipments
over there and finishes are flows and this

gives a second decision dimension which is
the position in the sequence dimension.

So essentially what this would
indicate is the second interest in

the sea that we use for
our versus here and if.

Slower to go through vertex two three.

For example is really integrate that
are praying of loaded shipment over three

second in particular sequence for
the day to actualize the sequences.

We've created arcs to go between these
various uses no there are no arcs

between verses with like shipment in the
season that's because the same shipment

cannot be offloaded more than
once during the given day.

Unless of course it is the idle shipment
in which case the crimper remain idle as

are all positions during the sequence and
finally getting position for the day for

the grains revenuers represented
by the signal we can use linear

constraints to create certain feasible
paths through this network which

will correspond to possible flooding
schedules for there for the grain and

we can cycle through these different paths
to find the one with the least attention

cost and there will be optimal Now of
course this is a very simplified view

of what this problem may be since each
wind farm possesses multiple cranes of

each of the different types
of the previously mentioned.

So we have a third decision
dimension to our problem

namely the crane decision to measure.

So now we can take into account
factors such as you're going to

require two grounds to complain are flown
on one painted complete are followed

by limiting of about of float on the stern
vertices across these are different cuts.

So basically we can cycle through all
these different allocations and if either

one of the least attention cost and again
there will be after to solve this problem.

We created the following
text in a true program and

as you can see this formulation
is fairly intricate and

unfortunately it's problematic it's do
computation to complex to be executed

well during the course of the work bit so
we have to explore different solution

method to make it more affordable for
the user to be for us or without including

the sequencing the rest of the way the
U.S.B. begins as we saw all the shipments.

But so
the way the US that begins is we saw a.

Of all the shipments they received during
the day in early struggle order and

then elevated them evenly across
all the brands are available for

us that we improve in this innocent
initial solution using two different types

of operations for
cost wops The first is the lateral swap in

which we swap two shipments between
their two respective allocated grids

the second type of swap is
the horizontal swap in order for

going to scale up to shipments internally
within the sequence of a given price

the way the sponsor executed it
is we perform a number of swabs

equal to the number of shipments
arriving during the day but

since the value of performing
a single spark can be determined

that list will perform subsequent survives
we perform another round of swap and

we actually repeat the structure four
times after repeating this four times

we generate several candidates schedules
which really can cycle through and

finally one of the least attention passed
and selected as the best candidate to go

parrot against our current solution if the
candidate has a small attention cast and

we repeat that you respect until
the schedule is fun which no other

candidate has a lesser
detention cost then and

then we will put that
as our best answer so

to validate our interest Dick we ran it
against our algorithms are all runs out to

us and essentially we can see here in
the smaller problems are interesting

to manage to get the optimum value
as soon as we got to the tension and

problem or the time difference became
more evidence and sorry the risk manager

solved the problem in seven seconds while
the elderly took over thirty minutes and

the twenty shipment problem
when we ran it using our

algorithm used express
to execute the mix and

if your program and actually ran the five
hour mark it ran out of memory and

could complete solving the problem while
our interested measure resolve the problem

in less than four minutes and
as you can see receive the.

Solution than the one
that was output by our

zero zero two package of
this very restimulated G.E.

we created the crane allocation sequencing
to all forecast for short and the way

via your manager of the same receiver
would use this is big in condition

information for the dead in this Starbucks
next to improve the distances between

their pads which really we can compute
using their coronets in the second box.

There's a good information for
the cranes which will be using for the day

they'll hear the rumble and all generate
three individual schedules one for

each of the different types of cranes
there are built which will help

build the information
cation information and

are flowing schedule and expected in
terms of cost for that now to account for

variability we can actually account for
it by the several factors the first

is we notice that thirty percent of
the shipments arrive in the early morning

between seven and eight am which gives us
a time of year to rerun our forecast and

also due to the slower more run
time of our view is that it

allows us to rerun the optimization to
figure out different allocations for

the dip in the second key observation in
this is that there are multiple solutions

to this problem with
the same potential cost.

So we can actually retain
the same potential cost for

the date using a difficultly
different allocation

now to see how this further
translates into value.

I'll give authority my teammate Mario
is going to discuss our valuation and

you don't have him.

So it wasn't a monetary
value to a project or

a client product providing us with
a buffer to live from wind farms.

When primary which was invented by the US
One of the most efficient with firms in

terms of loading and wind for me which
was having to fight as one of the least

efficient when promising terms of loading
so then we got a bit of attention because

the incurred of each of this one
comes from our people read here and

then we can deliver the pressure
because that would have been occurring.

For green.

Occasion toll was run to
create a schedule for

each of this one of which
are Mark Mark and Luke here.

So this provided us with a richer series
of between thirty and eighty percent and

then I weighted average of
forty six percent white.

Forty six percent because we
noted that around two thirds of

the wind farms across the of America
behaved like one from me and around one

third of the wind farms behaved like win
for me this forty six percentage points

around one point eight million dollars for
the year two thousand and twelve.

However it is important to note that the
savings are only going to G. energy and

of the customers which are the way
firms also include the terms of cost so

that their customers can expect
a result of this amount as well.

So some additional benefits that
cannot be answered by our for

the carriers it is there faster
turnover is one of the trucks for

the veggie energy side receiver and the
recent problems are better organization

due to our mathematical
model invite only for

the customer they can expect a breach in
the with telematics knowledge go first.

As Oscar mentioned previously the
Wintel America application is that kind

of the release of the customer or
the wind farms.

Therefore we will provide them with
with better information throughout

the Haitians I would also like to
mention that I replied has breath

expressed a lot of interest in linking our
tours to the winter medics application and

that they expect to have the power
of this news in the near future.

So with that said I would like
to summarize uprising by saying

that we started with the problem of drugs
agreeing that the tension cost of owning

dogs to fester for
gassing up with ration and

I pray now the vision sequence from
top to reverse this attention cost and

we calculate a series of around
one point nine million dollars for

four thousand eleven without alike
to complete the presentation and

open the door for question.

Attn

fine now voyeuristic

isn't better than our algorithm right all
the runtime tests are able to show that

our your wrist is able to get the optimal
solution and small his problems and

in a large piece from which currently we
didn't have the resources to solve fairly

well our view is that it is able to get
a very good solution which is better than

the other musicians often do we
were using in a smaller month or so

in terms of running time in
the sense of us to get better.

So you're about your cost
your preallocation where

you put information about your theory.

Well how does the didn't you tell him this
and well we've included his part of our

deliverables the source good of the panel
of patients it was until two G. energy.

So they'll be able to change
whatever parameters are necessary

when they're implementing it for
new types of rates

and this interaction with a palm or
something like

that probably has something like it or

not with help with your home or from

your own story this report on another

visit with friends very in size and
they're all very very large one from so

even if your provider we provide them with
good forecasts they still have to figure

out where to allocate the cranes
efficiently to to receive these disorders.

So our good forecast is not enough to
solve the tire your home over problem.

Sounds for your house to give you

a scream like to three mph my

friends are like really getting more and
more and more like a forty

minute drive to have my car from one
end to the other into these wind farms.

So out of patience he

likes Arran at.