Jason has a master's degree

in your modeling and

performance art or about where he's going.

Thank you.

OK thanks like Steve said.

If you look at the announcement the I
think the title was systems modeling and

performance in architecture but

when I was working over it last night I
thought I needed those commas in there so

systems comma modeling and
performance in architecture and

I'm going to talk about three themes that
I'm interested in and summarize them and

show how they lead into some of the work
that I've been doing and maybe talk

a little bit briefly about some of
the things thinking about going from here.

So.

Let's hope this doesn't give
anybody vertigo like a TED talk.

So the three big themes that I
have here one being an engineer.

I think in terms of
quantifiable performance.

Decisions and then the models and the
systems that help support those decisions.

OK.

And I'll talk about what my kind
of thoughts on what models and

systems are in a little bit.

But first it was Doug
said I'm I'm an engineer.

And one of the it's kind of two quotes I
like to keep in the front of my mind this

first one is that engineering is done only
with numbers analysis with numbers is

without numbers is only an opinion says
one of Aiken's laws of spacecraft design.

The University of Maryland has got several
of them they're good for keeping you.

Kind of keeping you grounded.

When it comes to technical
decision making but

before anybody starts to misunderstand me.

This is architecture.

And I also realize that not everything
that can be counted counts.

And not everything that
counts can be counted.

OK I'm going to kind of use
these words as we go along.

So what I'm interested in
is the things that count.

That can be counted.

OK.

So what can be counted that counts.

If you may have seen some of these
diagrams before these are Sankey

diagrams from Lawrence Livermore
there's two of them.

And the first one here.

Is just to put things in
context is the estimated

US energy use in two thousand and eleven.

This just came out last
twenty twelve it takes them

a while to put the data together.

But.

I have a laser pointers.

I rarely use these so

I had to remember to do it this
morning over here on the side.

It's kind of hard to see but

these are basic feed stocks of energy
that goes into the U.S. economy.

Up here at the top we have.

Solar nuclear hydro wind and

geothermal this geothermal by the way
is not a geothermal heat pump.

It's using energy from inside
the earth to create electricity.

Followed by natural gas coal biomass and
petroleum and

this just shows where they all were
all these different flows of energy go

a lot of it goes into
electricity generation.

Most of petroleum Here goes into
transportation and industrial purposes.

One thing that.

Can be kind of frightening is I
hear that in just the basic laws of

thermodynamics tell us that nothing is one
hundred percent efficient in pretty much

everything is quite a deal less than
one hundred percent efficient so

all this energy that goes in
most of it is just wasted and

usually goes into heating up the air.

OK About as good as we can get is about
a sixty percent conversion efficiency

there's always going to be
something that's wasted.

Now what we're interested in
here is this small part of it.

These are basically represent buildings.

OK so let's zoom in a little bit.

Let me put it when we come back out.

Take a look at this about one
hundred quads of energy went

through the U.S.
economy in two thousand and eleven and

what is a quad a quad is
a quadrillion B.T.U..

Quadrillion is ten to the fifteenth.

The one with fifteen zeros after it and
a B.T.U. is a unit of energy.

You see this on your air conditioners.

It's a British thermal unit it's the
amount of energy that it takes to raise

one pound of water by
one degree Fahrenheit or

to put it into something that maybe
is a little bit more every day.

It's about the energy that you get from
burning one wouldn't kitchen match.

So about one hundred quadrillion kitchen
matches are what go through the U.S.

economy in a given year.

So that's a lot of a lot
of kitchen matches.

And so are our interest is
right in here with buildings.

And where does all of this
come from residential and

commercial construction
electricity coming in and

natural gas a little bit of petroleum for
heating in the north little

bit of biofuels and
a few other smaller kind of niche players.

So what is it that causes all of this and
I'm going to keep coming back to this.

Diagram is this is kind of a model
of notional schematic but a.

A model of some of the energy
processes in a building

this is very generic and
you can see it's fairly complicated.

I come will it complicated
in a certain way.

I come from mechanical engineering where
our our outlook is to take one problem and

look at it in great detail.

OK here we look at it
a little bit differently.

We look at a problem composed
of many smaller problems

that are all interlinked
in complex ways but

each of those smaller problems we usually
treat simply OK so the complexity doesn't

come from the individual problem it comes
from how things are interlinked together.

This is a this is a in general
a fairly dynamic system.

Where we have occupants exchanging
heat in mass with different parts of

the building on the low interior
air the environment in or

directly through
the building on follow up.

We want to stay comfortable so
we have active systems to game the thermal

processes off our bodies
to make us comfortable.

We also use appliances which
interact with everything else and

we have some utilities that provide
some sort of energy feedstock to us

either in the form of energy electricity.

Natural gas or some other things.

So it's a complicated.

It's a complicated situation.

Well what do we want.

Well of course we want.

Thermal comfort in a healthy space and

this is something that's
somewhat countable.

If we're directly looking at thermal
comfort there are ways to put a number on

it but it gets a little complicated
when you deal with things like social

cultural expectation and an adaptation to
climate it gets a little bit difficult and

from a practical point of view we don't
really have thermal comfort sensors what

we have are air temperature sensors so
we use that as a proxy measure for

thermal com for and
it's some form of interior temperature

could be interior temperature or
there are some others as well like.

With temperature that are more
appropriate for certain circumstances and

that is definitely countable.

Is other things electricity we need this
energy too in order to keep ourselves

comfortable and have a healthy environment
and of course that's countable as well and

a lot of times that's when we
talk about what's important.

We kind of come down to a lot about this
inner Gene a lot about this energy but

I think it was really something
else that again itself.

Depending on what you're asking is
also really a proxy measure as well.

OK Well energy is one thing we're
worried about also the environmental.

Unintended side effects of using energy
are also what we're interested in and

so this is our energy related carbon
dioxide emissions in two thousand and ten.

This is the latest data we have also from
Lawrence Livermore same kind of diagram.

It's a little little
this kind of this split

makes it seem like residential and
commercial or buildings don't have don't

seem to have as much of a contribution but
that's because a lot of what

these lines come from the feedstock
that's directly burned at the site.

OK So this is your furnace or heating.

At the site but we still use electricity.

Here.

And this is one of the things that
we're also really interested in and

obviously can be counted.

And if we're looking at that.

Another the things that we have
to worry about is this pollution

arrow that goes from our utilities
back into our environment.

OK.

And this is countable as well.

And this is countable as well.

And this is these are things
that are important to us.

So what can be counted that counts.

Is some of the well
that's the first step but

we can NEVER little quote
that I like to keep in.

Mind is remember why you're doing it.

It's always a good thing to have
in the front of your mind so.

Decisions decisions is
the other big theme.

And while I don't get directly
involved in rigorous decision making.

It's always something that
kind of forms a landscape

that we're always talking about because
everything that we do when we're

designing buildings is making some kind
of decision making some kind of decision.

In the first thing that we have to do and
we think about this is

what are we really after so
Lewis Carroll was a mathematician.

In addition to writing books and

he's actually one of the kind of
the godfathers of decision theory.

And a lot of his writings here and
with Alice.

Have a lot to do with decisions.

And so this exchange with
the Cheshire Cat which way should I go.

That depends on where you want to get to.

I don't care.

So it doesn't matter which way you go.

And so these are these are some kind
of a lot to kind of to be said there.

If we want to make a decision we first
have to know what's really important.

So what really counts.

Well one thing that you can do is ask.

Why do I want for more comfort.

Why do we want there more comfort.

And why do you want to use less energy.

Why do you want to use less energy and
a lot of this can be if you structure it

can be structured in kind of
a hierarchy of objectives.

This is one that kind of
a generic hierarchy and

these colored boxes are actually
a mapping of the NAB

accreditation criteria kind
of overlaid on top of this.

Credit taste.

It is one of my committee duties.

So I'm thinking about that.

It also comes into my teaching because
one of the things I'm trying to do in my

teaching is bring more of an explicit
notion of what we really want to do and

not just teach these little parts but

how it all relates together
in a larger picture.

So why do we want to use less energy like
kind of come down to these two things and

I'm apologizing this is
a little bit small but

we want to minimize
the emission of pollutants and

we can go up a tree if we keep asking
ourselves why do we want to do that.

And if we keep asking ourselves why
do we want to do that sooner or

later we end up with these things
that are NAB criteria related

that all come to sense what you could
say as a definition of sustainability.

We don't want to leave things worse for
future generations to do now.

Also we want to minimize costs.

We want to use less energy
because it costs money.

And we don't want to use money
if we don't have to have.

Why do we want thermal comfort and is
a little bit more related to one another.

Well you could say it's we want to do it
just because we want to be comfortable.

A boss might say we want
to maximize comfort.

So that you're productive.

So that's another thing.

That could be important as well.

So why are we bothering with
going through all of this because

you may not get what you want unless
you understand what you really want.

And so for example.

This is kind of rare but
it can come up that if you design for

low energy consumption.

You may not that may not necessarily
correspond to low consumption.

It often does.

But you can cook Arco system that would
have a lower energy consumption than

an alternative that would have
higher energy consumption but.

Lower emissions.

And it can happen.

And also another kind of little
practical measures if you can get

to what really counts and
you can move up that hierarchy.

It tends to reduce the number of criteria.

If you can do it which makes
tradeoffs a little bit easier.

There is a school of thought
that's here on campus is very much

espoused by Chris Prius and mechanical
engineering and program officer at N.S.F.

named George Hazel rig that all our
objective should just be a single

objective and in the context of mechanical
engineering it's maximizing profit.

OK.

Not sure how we can do that with buildings
but it does make things quite a bit easier

that I'll allude to a little bit later but
I'm not going to get into that much.

OK.

So does it really count.

Now how do you count that can
be counted that really counts.

So we have to if we're interested
in technical decision making.

We need to be able to count things and.

If there's a little bit of
judgment involved here.

But for each one of these objectives
that we have we can associate

a quantitative measure of performance or
a performance indicator with it.

Just some examples dollars per year for
minimizing costs for thermal comfort.

There's a variety of things
called percent mean vote

percentage of persons dissatisfied
percentage of on net hours.

One thing that's kind
of interesting is that

the metric that you choose can actually
drive certain design decisions.

You may have heard of daylight factor for
day lighting.

It's kind of a crude measure and
it actually leads if that's all you're

thinking about to some design decisions
that you may not necessarily intend.

So there are some others
metric tons of carbon and

each one of these processes can be fed
into these performance indicators.

So.

Next question what's
the source of our counting.

How do we feed these.

Well it's models and systems.

You know kind of get a little bit into the
meanings of these words because it's not

there's not one hundred percent
agreement about what a model and

what a system is just going to
give you a few of those not to.

Get bogged down in in you
know some semantic debate but

to kind of bring out a point
to bring out a point.

So the first thing I look at is systems.

And well.

What are systems and a lot of times
what people say is a system is.

Reality.

Some part of reality that
we're interested in and

you know that's there's a valid way thing
to say that for instance a pendulum.

Why do I bring up a pendulum because
there's a really long quote here

by an old time.

General Systems theorist and.

Ross Ashby.

I know this is a lot of text but.

In one of his books he mentions that he
wants to be very clear about how system

is defined in your first impulse is to
say it's that thing they're the pendulum.

But even when we think about that
pendulum we don't think about it

in its totality right.

We don't really even if
we're only subconsciously.

We are always extracting some
subset some small aspect of it or

set of aspects about it.

The real pendulum for instance has
a link not only linked in position but

has temperature electric conductivity.

Several other things.

Shape specific gravity and on and on and
nobody ever even when you're looking

at that reality ever thinks about
that entire totality of that reality.

No matter what you're always thinking
about some limited part of it.

So the truth according to Ashby is that

only certain sets of facts
are important to us.

And so
the system is not the thing itself but

a list of variables and I would add to
that not just the list of variables but

the relations between the variables as
well and I put those together myself.

OK.

So that's my personal take on
systems some other things models.

What is a model.

Is some abstraction of a limited
part of reality that's useful for

thinking with it's not a thing.

It's a thing for thinking.

I still remember Bill Hiller's
quote on these things.

Or from Marvin Minsky a model for
a system.

Where system is some part of reality and
an experiment is anything to which he can

be applied in order to
answer questions about S..

Again this is taking
system to mean reality.

OK So like I said I take models and
systems to be near synonymous.

They're both abstractions of reality.

Sometimes I use systems when I'm the
system Hood the relations are important.

The model when it's more the abstraction
that's kind of important but

the point is not about who says what
definition goes with that word.

But it's to notice two things even if they

have these different
conceptions of this is that.

Models and

systems are in can always always
incomplete representations of reality.

That depend not only on the models
purpose but also on the observer.

So the observer is essentially
part of the system.

Also that implicit especially in this
description is the separation of the model

itself from the actions performed on the
models so the experiment that you do on

the model is separate from the model
itself although they might be related

different experiments might be more
appropriate for different kinds of models.

So if we look down a little bit.

Co.

How do we think with them.

Well we use them as before to feed to
run experiments on them and to come

up with estimates of these performance
indicators things that really count.

Well how well can we count with them.

How well can we count with them.

And it depends on a few many factors but
you kind of boil them down into

into three things One is the choice
of variables and by variables I mean

things that not only change but might
also be parameters so design parameters.

These are always choices that
any observer is going to make.

The choice of the relations
between the variables.

You may have the same set of variables but

depending on the purpose we may choose
one set of relations over another

the relations may be more or
less complicated depending on the purpose.

And also our state of information
about those variables.

We don't know everything.

And specially when it comes to buildings.

We know very very little.

A lot of times about these
variables as we go along.

So putting all that
together means that our and

estimates of countable performance
is always going to be uncertain.

If you examples of this but

one thing that I say to all of my
students is whenever they see some.

Quantifiable performance estimate.

That's claimed to represent that is
actually claimed to represent reality.

Their first reaction should be skepticism.

Right.

First reaction should be skepticism it's
very easy to fall in love with a model and

the results it gives you because you put
a lot of time and effort into it and

they've actually service quite
well in a lot of domains so.

The good thing is that this
uncertainty can also be counted.

We can also count that and
quantify it using probability theory.

So we can put estimates
on these sorts of things.

Again is the implicit separation
between the model and

the actions performed on the model.

So we have reality in this
case my house last spring.

Some reality.

But is then abstracted in society
model for some purpose and

then translated into some sort of
experiment for which we get some count

of performance that might be interesting
for us and if we take a look at this.

There's quite a difference
between not only reality but

between two different
models of that reality.

Now little being a little unfair here
because the point of this exercise was not

necessarily to model reality but
to be serve as a sanity check but

nevertheless the fact is that even
though the big trend is the same

These things can be very
different can be very different.

OK.

So models and systems are used
to help us with our decisions.

So how do we count.

What can be counted that really counts.

I'll confess to never seeing this movie
but I have seen the last scene of it.

This is the candidate where
Robert Redford's character somehow

finds himself elected to senator
which he is never really intended and

in the last line of the movie
is what do we do now.

OK so what what do we do now.

And actually I mean for
now what do I do now.

Not me right here.

But what does an individual do.

OK There's a difference between
what we do now and what I do now.

So.

Aggregation we have all these
performance indicators.

That if we sit back and
just look at them without everything else.

These are quantitative expressions
of what we believe counts.

But these conflict in their
values are uncertain.

They conflict because if we want
to minimize the massive amount

of carbon emission that's easy.

We don't use energy which
will also lower our bills but

then we won't be comfortable and so these.

Every decision most
every decision we make.

At some level has multiple criteria.

Evolved and
in general they're going to conflict.

So how do we handle that not only that but

we're not very certain about
what the numbers are OK.

But we seem to have this intuition that

what we should do is some function
of these numbers at least

when we're talking about things
that count that can be counted.

So it's some function of these and
it turns out that that's true that's true.

We can combine these into a single number
indicating our degree of satisfaction

to those preferences under uncertainty.

Now this is just going to be a Cliff Notes
overview of decision theory because

it's underlying things and it's in the
background but it's very difficult to do.

So if we choose to be rational.

If we want to choose a now if we
want to choose a course of action

that is consistent with our
preferences as we've stated them.

For example we accept certain propositions
that encode reality rationality.

There's about six of them that
are that are kind of the standard set.

Most of them are kind of hard
to put into a bumper sticker.

Except for the first one which says that
preferences on outcomes are transitive

that says that if you prefer outcome and
outcome B. And how can beat out can see

then you would prefer outcome eight
outcomes see which makes sense right.

A few of these others are also in
code those sorts of things but

they deal with lotteries.

And that's so that they bring
probability into the into the effect.

And you can go through and
create an actual mathematical proof

a rigorous formal logical proof that
there is this quantitative measure that

is called utility and the outcome with
the highest expected utility uncertain and

under uncertainty is the most preferred.

That is the most consistent with
your preferences as you stated them.

OK So
that's the one that you should decide.

And that's what that's what this
normative decision theory tells you.

Now it's really expressed only for one.

Preference.

Or one objective but
we have multiple criteria but

they can be combined
into a single utility.

But it's very difficult and
it involves some assumptions.

To go on.

So it's.

It's pretty tough and time consuming.

Trust me.

So.

Things are getting kind of hard putting
this together is pretty difficult.

It's time consuming to do to
build a model of how a building

operates even with very nice software.

It's hard to do it takes a lot of time.

That's just as hard.

That's just as hard and the way I kind
of look at this area putting all these

together is in terms of utility being an
engineer I think of this is kind of like.

The the law that expresses the theoretical
efficiency of heat pumps and heat engines.

There is another logical argument that

derive that shows you what the maximum
efficiency of an engine or a heat pump is.

But if you were ever to try approach that.

The device you would end up building would
be enormous and would operate so slowly.

You wouldn't get any use out of it.

It's kind of how this decision theory is
but it's still just like this efficiency

serves a role as the ideal
that you want to strive to but

we're probably never going to get there.

OK so this underlies some of our thinking.

But we don't actually do this because.

It's not really practical.

In a real designed setting.

It's just hard to do.

One of the things that makes this
difficult is you have is the elicitation

of your preferences and you have to ask
strange questions involving buying into

lotteries and all this kind of thing and
if you have to do that a lot of times

it just gets complicated and and
there's a few assumptions involved.

And it gets better now.

Go from what do I want to do now
to what the we want to do now.

OK this is another layer that I'm only
going to touch on because I'm not that.

Certainly not an expert in it.

But one of the tenets of decision theory
is that only individuals can make

decisions because only
an individual can have preferences.

Even if you think you're
agreeing with somebody else.

It's really only that individuals
can make decisions but

individuals don't design buildings.

In their totality teams do or
groups do but

they aren't engaging games and
by games I mean game theory.

And this is interactions between people
which each may be operating with some

utility and this is an understatement.

Everybody that I've been around who works
in this field for practical purposes.

Waved me off.

Don't go there.

It's a rabbit hole.

Game theory describes the games
that people engage in

the inverse of game theory is
something called mechanism design

where you try to design a game where that
everybody in control by playing the game

ends up contributing to
a desired social result but

from everything I've seen the mathematics
of that is intractable and

actually can be put down in math and
it's very difficult to deal with.

So let's.

Leave it alone.

With leave it alone but
it forms forms a background.

These are the ideals that will never get
there and ideals will never get there so

to kind of sum up.

Making decisions is
the fundamental activity and

here we're talking about
making technical decisions.

It's difficult in a practical.

But we can still count.

We can count the value
we place on performance.

One of the things in this idea of utility
is that you're making a decision not so

much on that number but

on the value you place on that number
of that performance indicator.

And this is kind of interesting.

Just as a side.

Slight tangent people who study

professional gamblers people
who actually make money.

Gambling.

In the back of their mind actually
are thinking in terms of utility really or

thinking in terms of utility and that's
how they do it they just have an innate

ability to think about value and
not so much money.

Of course the whole system works because
everybody else gambles and loses.

So performance counts we count
using performance indicators and

models which have a system quality and
experiments on those models

are used to feed our performance
indicators all models are imperfect.

Because there's observational biases
from the observer the choice of

variables the choice of the relations
between those variables information about

the variables so
their outcomes are uncertain and

models in a fundamental way are distinct
from the experiments that run on them.

OK main things that I think
about are this and these.

All of this is in the background.

It's part of the thinking it's the ideal.

We want to get there we try
to get there but we probably.

OK.

So to back up.

Going to take a few.

A little tour through some of the work
I've done around these things.

OK.

So the first one is a separation of model
simul from simulation or declarative

modeling kind this is some work I've
done with Ritchie over in Oxford.

And this is one where I wanted to look at

what was going on in this princess of
Wales conservatory A.Q. Gardens in London.

Now this is this is a kind of building for

which the normal simulation
programs out there don't have

the embedded models in them to be able
to determine what is going on inside

of here is a lot of moisture transport
is a lot of transpiration in the leaves.

Here's a simple one dimensional model
of a greenhouse showing different.

Pathways of energy through
the soil vegetation

vapor heating everything
through the cover.

And it's not something that you can
do with a program that already exists

is something that you have to encode for
yourself and so

I put it together in a language
called Modelica and would Elica is

a model description language but
it's not a programming language.

Most of the tools out there with one or
two exceptions embed their

models in a programming language that
in codes computational assignments.

What to do.

This separates what to do from
what the behavior is OK and

has some benefits and
I'm coming to see some downfalls as well.

But it has a lot of benefits being able
to make the separation between the act

of modeling in the act of
simulating because it takes.

The burden off of the model are of having
to figure out how to implement the model.

So you can kind of think of
this these model description

languages these declarative
model description languages as

kind of the same sort of evolution that we
have in going from programming computer.

A computer in assembly language or
machine language up to

using a high level language at first
like Fortran and then some systems

programming languages computer systems
programming languages like C. and C. plus

plus that take the level of abstraction
that remove the human from the machine but

have a well defined transformation
from the human language or

closer to human language to the machine
through a compiler Mandela takes it one

step further where you get rid of
the Constitution computational execution

altogether and focus on what the problem
is not on how to solve the problem and

disentangling those how some benefits for

modeling systems especially those that
are fairly nonstandard in art and

in some of the standard tools so
delicate it's an object oriented component

based language that allows
hierarchical modeling.

And here we have it has a graphical lair.

But it's fundamentally textual.

It has this graphical layer we take
different zones that are just instances of

a generic object that you can instantiate
with different parameter values and

each one of these if you crack it open.

Has a larger model and so this is
starting to get kind of complicated.

One of the difficulties with this language
is if you're actually working with it.

If you're not careful.

All this business can get out of hand.

So a big problem is how
do you cluster these

models together in a way
that's been official.

OK So this is that.

Same basically the same model as
I showed you in a graphic before.

And if you keep zooming in you can
see one of the features of the model.

It's in capsulated with some section
called equations here in the thing to

notice is this line if you were doing
this as a in a programming language and

this would be an illegal statement.

You could say that because I
have one two three four five

things on the left side of the equal sign.

So if you're doing this in C.

or Fortran you have to have one thing
on there because that's an assignment

statement it says take the operation
on the right hand side of the equation.

Do it and then put the result into
the variable pointed to by that name.

OK.

You know so I haven't done that here.

None of these many of these
parameters are left undefined.

What happens is all you're doing
is expressing in a row a relation

a mathematical relation between
these different variables.

And then it's up to the translator that
goes from this higher level language down

to the computation to figure out what
to do and there's ways to do that.

Of that that work very well.

So this separates them
on the act of modeling.

From the act of simulating on the model.

So it's very nice for
things that are fairly nonstandard.

In principle.

And this particular
language is intended for

systems that are that
are dynamically changing in time.

It's not always appropriate.

Cost that it comes with as you
have a lot of other machinery.

And we have to expose some of
these internal variables and

things get complicated

if you look at some of the libraries that
people are putting together for buildings.

This very quickly gets out of hand
in my opinion it can get kind of

get kind of complicated and hard to use so
it's useful in some situations.

The next thing I'll talk about.

Touches on what kind of those
situations are more useful.

But if we look at it there's a another
model another version of the model that

was encoded in Matlab which is
a computational environment built around

a programming language or
Simon state that's one difference in just

kind of the general productivity between
only having to deal with the model

versus having to deal with the model and
the operations on that model is that this

model was unstable when it was
first made because the new you

the model lawyer had to worry about
the numerics and they had to use a fairly

In fact they had to take parts of
the model out of consideration.

That actually had an effect
on the on the end result.

Just in order to get it to run.

Now they went back and fix that with some
more sophisticated numerical techniques

but that was another layer
of work that they had to do.

Another thing is
the results are similar but

this again shows that there's
differences between models.

One thing that's kind of nice about
when Delic is if you look at these

temperatures these red lines.

These are the internal temperatures.

One thing that's easy to do with Modelica
is is to do multi domain modeling.

And he's you can model the physics from.

Not just thermodynamics and
heat transfer but

also electronics discrete events and
controls.

And so my model differed from this
model in the controls that it used.

Using proportional integral
derivative controllers that enabled

fairly bit more tight tight control of
the wintertime interior temperatures.

I'll touch on that a little bit later.

And again we've seen this before.

Some of the differences between
aggregated results monthly gas use for

heating you see the difference between
meter data and output from the do.

Models.

Once again models are not reality for

a variety of reasons and two models don't
necessarily agree with one another.

Depends on how the observer
decides to put these together.

OK So I mentioned multi domain modeling
in addition to declarative modeling and

this is some work that
I did with my thesis.

In which this language was used to
incorporate a high level of physical

fidelity combined with a low level of
physical fidelity so I mentioned that.

And when we're talking about buildings.

Usually we're worried the complexity comes
in relations among simple sub models

where as an engineering we're looking
at one model that's in full complexity.

This is one that kind of mixes
those two ideas we look at a simple

sub model combined with
a complex Southern model.

In one.

The complex model is computational fluid
dynamics where centrally the full.

Equations of motion of a fluid are solved.

I did it in a fairly non-trivial.

And established but

non traditional way that's a little
bit more suited for this language.

And the thing to note.

Excuse me but I had some things here.

Some of the things to note is that here
we have a one dimensional problem.

Combined with essential a two
dimensional problem in one.

We also have other domains.

Involving controls to run a simple problem
that happens to have an exact solution.

And we see that the two match
up pretty well they're so

different different domains and

different Fidelity's of domains
are mixed together with this this model.

See if T. is useful in some cases
although it's a little bit of a niche.

Application when it comes to buildings for
applications like natural ventilation or

if you're worried about contaminate
transport and hospitals and

the things like.

Some other things that one of my students.

Roy raises a is working on is what
we've looked at is all of these

a lot of these models work in a direction
from a design to a performance.

And.

So that's why these arrows are going this
direction we go we flow from a model

to a performance indicator and

Roy is very interested in decisions
being made earlier stages of design.

And so one thing she wants a look at is
what can we can we invert that problem.

Can we say we want some statement
of a performance that we want and

what does that tell us about how
we should choose our design.

OK this is actually
a very hard problem and

people have looked at it in different
ways and some of the difficulties that

you get if you want to reverse
these arrows and go the other way.

Is that you're really are thinking about
preferences on the probabilities of

performance to create probabilistic
bounds on a design space.

OK And you can think of a design space as
being a collection of all the parameters

that would define a design and
those parameters.

Kind of going back to my C.F.D.
days if you think of those as

each parameter being on an axis and
all of these axes in some high dimensional

space being you know forming
a orthogonal coordinate system.

If you can wrap your head around that you
can think of lots of different designs

forming a cloud.

In that space.

It's kind of like face space.

So her interest was in
forming probabilistic now.

On that phase space where
the problems is is if you have quote.

One performance and

by performance one performance I mean one
probability distribution on a performing.

Well many designs could
have that same performance.

There's an infinite number of designs and
if there's an infinite number of designs.

Some set of those could map
to a single performance

which she was trying to do is take
one probabilistic performance.

And come up with some estimate of how many
designs correspond to that performance.

OK so it can only be probabilistic.

This inverse cannot tell
you what to design but

can maybe provide some boundaries on what
to design in a probabilistic fashion.

So some of the things that she
did as a toy problem using

something called a Basie and
belief network uses Basine probabilities

to just look at what
the if you have a certain.

Preference.

If you want to minimize let's say
the amount of heat after artificial

heating you have to do in a space where
is that probably slickly tell you about

what the area of south window should be
if you came up with some probabilities.

She's still working on it that
will probably put together for

a conference paper so it's an interesting.

It's an interesting thing to think about
but it's actually very difficult because.

I would imagine that a lot of what
people actually want to do is

we want to perform something.

How do we do it rather than let's try
something come up with a design and

then test it and then iterate back.

OK So this is a little bit
more of a brighter flashlight.

Hopefully.

As we go along.

Also with Roy and freed.

We're just starting on
a project with D.B.L. on early

design decision making and this is this
is also related to this probabilistic.

Notion and

the fact that we don't have a whole lot
of information about our variables.

So if we take a look at two
different design options be an A.

In time if you look at any one time.

IME each one has some uncertainty
associated with it or

some probability distribution that I'm
representing here with error bars.

And in this case.

It shows that option V.
dominates option A At this time.

So if this is all you're
thinking about and

you have a low set of information
these error bars are rather large

it would seem that you would
want to choose Option B. right.

But at an early time we don't
have much of a building at all.

And we don't know what other decisions
that we make that we go along.

If in some parallel universe we
carried both of these designs through.

That could make B. get worse and
a get better in addition to having these

probabilities these uncertainties drop
as we more and more define the system.

So the question is and this this.

This sums up the problem with
early design decision making is

our state of information is very low and
no tool out there.

Explicitly recognizes this problem.

OK.

We don't there's nothing that says
everything just looks at a snapshot and

doesn't take into account domination.

Through time as opposed to
domination within time.

So this is one of the things
that we're looking at.

In the questions and some here.

Can you make a design decision in early
design phase with the confidence that

the performance related outcome will
be preferred at the final design.

And what are those decisions.

OK We want to find what those are because
it's probably not going to be

every decision out there.

Also what models can support those
decisions with that confidence.

So this is combining both decision
making and the choice of model.

So what are the variables and where.

The relations of the variables and we've
come up with a probabilistic index for

this where we take a model
as being valid if we're.

Have full confidence that it's
leading us in the right direction.

This is different than another concept
of validity which says a model is valid

if it represents reality.

Our concept of model validity is if it
gives us the right decision to sum it up.

Another thing I've actually neglected to
put here is how do we integrate these

models and these decisions with
representations of what a building is so

all the models I've been talking about
are models of what a building does.

What a building does is some
function of what a building is.

And a large part of what a building
is can be represented in the building

information.

Another thing we want to do is look
at how to combine these two together

into something a little bit
more unified is a big question.

So these are some of the the work
that I'm working on.

I have a few other.

I don't have slides on them but
on some more speculative directions.

One thing that you kind of unavoidable.

If you start thinking about this
declarative modeling where you think of

sub models and relations among the models
as you start having to think about how do

you organize what those models are and
how to relate to one another so

you start bumping into the field
of knowledge representation and

ontology and so
there's a there's a field out there on

this very active community then just
starting to not so much interact with but.

Come into contact with there's
rigorous logical ways of putting

these things together.

Also a lot of the early systems theorist
dealt with the language of said.

It's.

Mathematical sets objects and
binary relations between objects.

Another field of mathematics that is
potentially useful is that of category

theory which contains sets and

the primary objective in
category theory is to look at

how you can go from one category of
things to another category of things.

So the focus is not necessarily
on the objects themselves but on

mappings between objects and app mappings
between different categories of objects.

It's starting to show some applications
in computer science as well but

it's something that I'm I'm.

Right now teaching trying to teach myself.

So I'd like to thank you.

But one of the thing before I do is just
kind of point out some things I've done

with teaching just very
briefly I found just

in teaching that if you learned
to think about things coming from

engineering the whole focus is
on learning to solve problems.

And not structuring models.

Although that's in the background but

they're they're fairly distinct
activities and I've come in.

From engineering to me to teach
most of our undergraduates

sort of in the way that I was taught us
in engineers how to work problems and

that's important but
it's a little bit different than modeling.

Some aspect of reality.

And this is all of this and
a cup of coffee and

five cents won't even get you a cup of
coffee because it's not scientific.

But it's just.

I'd like to just show a little bit that
this is a project that I did in class last

spring where I asked students
not to solve a problem.

But to break down a situation
into objects and relations

between the objects so they weren't asked
to do any math at all but just concern.

Actually put things together in

the topologies of the way things
behave and not actually solve it.

So it's very conceptual and
I did have a before and after survey.

This is not scientific.

It's just anecdotal.

And these questions are not the exact
questions but condense down so

that they fit as a pile.

Where black is the answer
they gave before and.

Read is the answer that
they gave after and

most of them found that they were
able to identify systems and

the processes in those systems
be able to put them together and

support decisions they found value in it.

This doesn't say whether they
actually did it correctly.

That's a whole nother thing.

But it does say that they found
value in thinking this way.

Most of them found it quite useful and

it was a quite useful thing to do so
with that.

I'll open it up to questions and
discussions.

Thank you so.

Yeah you are right.

Yeah I thought you were going to.

Yeah yeah.

Basically there prior probabilities and
you go through it and

come up with post your ears so basis
theorem is worked into this so bipolar.

Yeah.

Well they can.

It's possible yes.

It's we can't say in general.

Either way this is an example of perhaps
an extreme case to illustrate the problem.

And those These distributions this

one slice in time in general
will not dominate like this.

They may overlap.

But just as a an example they're
shown as dominating those

dominating Yeah there is that there's a so
I don't mean to point the laser right.

Yeah yeah

well.

Let me.

That's actually for me.

Zoom back here.

This is one of the things
that we're looking at here

probably one zero zero zero.

Right.

In fact all this decision theory
business and utility is often

criticized because some people at
least in the past have have claimed or

wanted to use it as a model for
how people do make decisions but

that assumes people are rational and
people are very much not rational.

And yeah.

People are very much not
rational All that does

is to say that this is what you
should do if you want to be rational.

But but come back to it all right well.

It's.

The underlying this is why I
said when it comes to decisions

all of that complication is the ideal but

we don't want to do all that complication
we want to get closer to it.

OK so some of this decision making funding
principles are within these questions

in the way that we frame the problem and
how we're going to work on them.

They're not for utility but
they insure us they're a little bit.

Mentioned there a little bit now and
to your question on complexity.

I want to go back.

Is going to I don't know how to.

I usually have a side bar for this thing.

To help me.

See.

OK.

This quote.

I chose this person.

Because he's very much.

His whole line of work was
about making things simple.

It was about making things simple
everything he did was about trying to

break things down into a smaller problem.

He very much recognized that.

His initial interest was in
how does the brain work and

the brain is made up of all of these
enormous numbers of complex of.

Connections.

And there's a very much less there's
a limit to how much you can compute

with these with these
types of complex systems.

So his whole program was about making
things as simple as possible and

I think Freed is also in
that same spirit as well.

One of the one of the advantages
of saying that a model is valid

by tying it to decisions
is that a complex model.

That leads to the same decision as
a simple model means that those two models

are practically equivalent from
a practical point of view.

And so if you get the same
decision from the simpler model.

Then if you use the simpler
model it has the advantage that

you might be able to mentally
process it a little bit more easily.

Right now.

And so what's the smart way.

And so what's the smart way to simplify.

What's the right way to simplify and
that's what that's why I like

this this definition of model
validity that ties it to a decision.

Is it tells you kind of a threshold for
what a valid model

is it doesn't tie it to what reality
is it ties it to what you want to do.

And so that's a test case for if you've
made it through simple are in principle.

Now.

The math of it really only works for
discrete options

if you have continuous options it
gets very difficult but in spirit.

That's where we want to go.

Is that OK OK we're not quite there yet.

Yet.

As in time to make a decision or
sequential decision making.

It just time to do it.

OK.

Well one of the things that makes
are you talking about rigorous

decision making with all this
machinery that I alluded to or.

Right.

And you can even build a model perhaps.

Well one thing that we would
want to do is if you can.

What if you can use simpler models
simpler models take less time to build.

It's another reason for
using a simple model.

If you can get away with it.

Tom Butler at SES did modeling for
the Henman it took him a week

just to put together the occupancy
schedule for a model of that building.

Now.

Perhaps for what purpose.

He was using it for that may have been
you know the effort that he put into it

may have been worth it.

So another question is
does the question that

you're asking justify spending
the time on it with a complex model.

So yes simpler models.

One of the I would mention that.

Does this modeling and simulation
actually help the Does it drive design or

does it do we use it to rate
the design for certification like lead.

Doing a lead certification is a time
consuming process to build that

model the way they want you to do it.

And it's so time consuming that.

It's almost not worth it to do it
especially given the state of uncertainty.

Now another thing that some people
try to do in making these models is

try to automate them using
things like graph grammars or

triple graph grammars where you
can set up these rules where you.

Create models based on
a on a on a grammar.

It's a graphical implementation
of the same thing as a as

a like it's once a grammar.

Chris previous I mentioned it works a lot
in that field trying to automate models or

certainly some of the creation of
models using transformations that

can be automated.

So you don't want to transform
between one model to another model.

They could be different kinds
of model at the same time or

they could be the same kind of
model at a different time or

two different models at the same time
there's different ways of doing it.

And there's there are tools
available out there but they're not.

There they tend to be more targeted
towards software engineering.

For the moment so model creation.

Yes Is A Big is a big deal.

And when it comes to
this Modelica language.

It's a double edged sword in
some ways it's faster for

simple things but
when you try to get complex.

Especially for one off kinds of problems.

It actually gets a little bit more
complicated if you ever want to see

you can look at the buildings library that
they're creating at Lawrence Berkeley.

Crack that open and it's it's hard.

It actually gets harder to understand the
more complicated it gets in my opinion.

Yes.

Right.

Yeah that's so I'll be honest I something
I don't really have an answer for.

But I think it's one thing
that we're trying to find out.

Usually the I think the assumption
has kind of underlying been is the.

If you can make a model that
has everything in it then that

works everywhere and all the time.

But as you said they're hard to work with
and they're and they're difficult so

yeah I think that's a I'm afraid don't
have an answer for the question but

I think it's a very good point.

Right.

Sure.

Sure.

Well maybe some of the ROI
is sort of inverse work.

Maybe maybe help speak to that because if
you want to instead of trying something

trying a design and then testing it
based on no experience or what have you.

Roy type of work might help to shine a
larger flashlight things that you may not

have seen before.

So it's.

Again what are the right
models to use with that.

Is is a is an important thing and
her work used a statistical model and

as opposed to a physics model.

She used a statistical model.

I think because it was easy but
there is this kind of ties back into this.

Model description language of Modelica
that if you build models declaratively

instead of as a in a programming language.

Depending on the typology of the model.

You can operate that model from multiple
inputs are from instead of going from an.

In other words you can switch what your
inputs in your outputs are within some

limits.

That depend on the topology of the model.

Thank you.