PLEASE LIKE, SHARE, SUBCRIBE video! Thanks you very much!
For more infomation >> The search for Baylee Despot continues - Duration: 3:25. 
For more infomation >> Watch Live: The National for Monday, Aug 6, 2018 — Saudi Arabia, Earthquake, Rick Gates - Duration: 1:04:18. 
For more infomation >> Lake Powell News Network LIVE: News and Entertainment For Page and Lake Powell - Duration: 15:41. 
Today I am going to present about PLEXOS product.
PLEXOS is unique in its class,
and like PLEXOS there are companies that produce products that are unique in its class
which have different vectors of differentiation.
For example, take the case of Apple which is characterized
by producing products that are simple and easy to use.
Uber did what nobody have done, mixing networking with technology.
And Tesla mixed cars with luxury.
These three companies have in common that
they did not invent these products. They reinvented and made them better.
Regarding what concern us here, I'm going to tell you
about the differentiation vector of our Software PLEXOS.
the first is about its architecture which is unique.
and the second is that the features it has are unique
and the 20 minutes presentation that I will show today is focused in these 2 points.
About the architecture, I want to show you three properties.
first is that it is transparent,
the second is that it is customizable
and the third is that it is integrated
It is transparent regarding the problem formulation.
It is a commercial software that has graphical interface.
but all the equations i.e. objective function, constraints,
and all complex equations can be printed out, and the user can look at them.
If the user wants to go further, he can get inside and see everything ...
we do not show the source code but we display the equations
and our more technical users love this. There are users who
do not bother with this, but if you want, you can do it.
Thus, the first characteristic is that PLEXOS is auditable
and you can see everything that's inside the software.
The second is that it is customizable,
if you can write a generic constraint in a piece
of paper then you can place it in the PLEXOS.
There are two classes that we have - well PLEXOS
is based on classes, memberships and properties -
then there are two classes that are keys to this:
the Constraint class, where the user can model
generic constraint using Graphical User Interface
and the Decision Variable class where the user
can create the decision variable he wants which can be integer or continuous.
So it's not that we're handing over a product with
which we say that this is how the modelling has to be
but the user can customize this and can do it to his own taste.
And if you link this with what I previously showed, that it is
transparent, you can have complete control of the software.
and third is that it is integrated,
in a single database there are the modules of long term, medium and short term
all the complex connections that can be found between modelling phases,
where even you can find different names in PLEXOS is just one
Then it connects long, medium and short term automatically.
One can see the bridge between different phases.
what each phase passes to another, as I said
everything is transparent and you can see it.
Let's pass on to the features now, today I'm going to talk about three.
The first, about time.
The second, on stochastic optimization.
And the third, about technologies that manage risk in the electricity market.
With respect to time, I split it into three:
In the long term I'll tell you what we offer, which is
different from the rest in regards to long term modelling.
The granularity that the software has.
And the third characteristic which we named Interleave
that is to analyse deviations in the day ahead scheduling.
Then, typically we always separate the models and say
this is long-term so we do not include short term detail.
and the model is configured as simple as possible so it can run
and one does the separation between the two phases as shown in the image here.
Long-term models were characterized when they
started as they only minimized capital costs
and one then had to connect it to a model that simulated the operation to inspect
more or less how it would work, iterating between them.
Then, PLEXOS includes the co-optimization of
generation and transmission around 2000 - 2002
in other words it analyses investment and operation
simultaneously obviously simplifying the resolution
you can't run these models per hour because they are
too big, so you have to simplify it using blocks instead.
And now what we're seeing with our customers is that
investment plans generated with long-term models
don't answer the questions that are being posed
on the flexibility that the sector needs.
They are things like this. This is a weekly dispatch of
a combined cycle CCGT, a case study we have of Peru,
x-axis shows seven days then you can see that during
nights there is a drop in CCGT power generation
There are things that the classic models of long-term do not consider
which are: ramps, minimum stable levels, minimum down and up time, etc.
that is everything related to the unit commitment.
Then the software evolved, took that and offers
what is called the LT plus Unit Commitment.
in PLEXOS you can model this and can capture what is the value of flexibility
in a long term problem or short term problem.
We have risk constraints in long term.
When you have multiple scenarios you always wants to minimize the expected cost
here we have the y-axis and the solution at the far right hand
is the solution that one would get at minimum expected cost.
there is a generation and transmission investment
plan associated to each one of these points.
but if you put the other axis which is the x-axis, and look
at the tenth percentile scenario of one of the solutions
one can see that this tenth percentile scenario
deviates a lot from the expected cost solution.
but one can control risk, and even though the expected
cost may go a bit up, all scenarios have similar costs
This is also included in our software.
Degradation, well, typically one can think of solar panels
that start with a given efficiency and as time progresses lose efficiency
it can also happen with thermal plants. Now
you can tell me that is super easy to model
you only change the value of the property, but is not so easy when they are candidates
because when you are doing a generation investment plan and you have
degradation as variable, the model has to choose what to invest
and this can be used to evaluate Demand Response programs
that are effective in the early years but then eventually they decline in efficiency
In Granularity, the software has a menu where
you can choose between modelling by blocks
you can choose to model a block per year, or modelling hourly or as
Milorad said you can model down to the second, only with one button
There is a single database, obviously you have to input hourly data,
and if not what the software does is to interpolate
them to find the required per minute values
if you want to do simulations using a per minute resolution.
so the user can choose a resolution option and then run the model.
It was acceptable before to run by blocks, but with
everything we're seeing nowadays it is not so acceptable
Thermal plants begin to suffer, they do cycling and operate at minimum stable levels.
and you start seeing this when renewables have high penetration in the system.
And the next is Interleave. Here I have a graph showing that what
is planned the day before is not the same than actually happened.
So now it is impossible to model and predict,
even in day ahead, what is going to happen.
This is a real case of May 21st in Chile where there was an outage
and then you can see in blue curve that the generator was
supposed to generate and then after a few hours stop
and what happened? they didn't generate when was supposed
and after the unit was repaired it came online with a ramp.
Then how can we model these deviations in a software?
And here comes the interleave module, what it does?
With interleave you can model what happens in real time fixing
the resources that don't have the flexibility to respond.
there are certain machines that are on and some off
which are slow to start because you have to move people
prepare the equipment for starting, these resources are not available when
such an outage occurs, or when a cloud passes, or when wind blows over.
My colleague Fiac is going to show you some case
studies in Europe where this has been applied
What I wanted to show you with this is that what you plan,
even during the day before, is not what is going to happen
because too many things happens in real time
that cannot be anticipated and then modelled.
Regarding stochastic optimization I'll talk about two things today
The first one is respect to stochastic unit commitment,
and the second about the stochastic hydro.
Regarding stochastic unit commitment, the problem is as follows:
This is the dilemma the operator faces which is reflected in this super simple system
of two thermal machines that are inflexible, and wind
so if you put this model in PLEXOS you can find various
solutions depending on how much wind is assumed,
then it comes the question: how can I find the commit on off of
these machines minimizing the expected cost of the entire system?
and that is the concept of stochastic unit commitment which the software offers.
In fact, there are few reports on the internet where this is
modelled in the US and you can see this concept applied there.
I want to show you a video, which in fact is on our website,
that illustrates the concept of stochastic unit commitment
What about the Stochastic unit commitment? You got
compromised with a solution and then anything can happen
and you can say you know? maybe I would have done
something else, but you got compromised with the solution
you say I turn on these generators and then see what happens, but
you make sure that the expected cost is minimum when you planned.
Regarding Stochastic Hydro, we have investigated this
topic for several years taking the Chilean system as test case.
In the literature you can see that there are two ways to solve this type of problems
one is by decomposition, that has been the standard
in the countries of the Latin America region
and the second, by a methodology called rolling.
We implemented the rolling methodology in PLEXOS about a year
ago because it integrates better with all the existing features
everything I have shown you here is coupled with this.
We have done many tests with the Chilean market,
why? because the Chilean system is quite particular
it has all the hydro stochastic modelling required in the countries of the region,
and on top the transmission system is required
to be modelled at a fairly high level of detail
then that's why it is a tremendous case study, and all
our tests have been conducted with the Chilean case.
Obviously, this worked well with a tiny model
then when the size increased to the Chilean case
we encountered memory problems among other things but all that is solved now
What are the benefits of this technique?
First is that it doesn't have convergence problems
and the second is that you can use integer variables with this methodology.
Regarding risk management, this is from the point of view of the electrical system
I will talk about storage technologies that PLEXOS can model as supply.
demand and reliability, what can PLEXOS do for reliability?
Storage technologies
Really PLEXOS can model all of them, the most
popular are those shown in this list here
which are concentrated solar power (CSP) the 24/7 solar plant,
CAES, batteries, pump storage and electric vehicles.
I'll go one by one to show you more detail
You can say, Concentrated Solar Power is super easy to model.
It is solar 24/7 and you can say, you know what? I will model a
plant that generates a constant block at zero variable cost.
or you can reduce electricity demand and model to reflect this CSP.
but there are many other things that are involved in this model that I show here
first you have to model the solar light reaching the mirrors; then you have to
model the chimney, it has maximum heat, minimum heat, it has heat loss as well;
and everything that has to do with the electrical part,
which are all unit commitment properties that are there:
min stable level, ramps, operational minimums.
There are studies available on the internet, you search CSP PLEXOS and you can
find studies that has been undertaken precisely incorporating all these things.
CAES is similar, but here what is done is that
the air is compressed and then you expand it,
and you can put more properties, for example: the
efficiency of the cycle, the compressor load
also one can model the variable heat rate that this generator has
and you can model non convex as it is also possible to model this with non convexity.
Batteries, the concept is the same as the previous.
They charge and discharge when the system needs it
but there are also more properties that are specific to batteries: Maximum power
required to load, maximum ramp-up, ramp down, charge and discharge efficiency,
maximum cycles per hour and per day, and also degradation
capacity - the batteries as they are used they degrade.
The pump storage follows the same concept as
the previous, but there are two reservoirs,
and generate when prices are high and when prices are low they pump water up.
Do they lose efficiency? Yes, I mean it is not the same energy
that is generated than what is consumed from the network
but what matters here is the arbitrage between periods where
there is maximum demand, or say a high price and low price
And electric vehicles is the last one I'll show. They also charge
and discharge, and the important here is ancillary services.
I did not tell you this about the others but all
of them can participate in ancillary services
For example when they are charging they can participate as load
in ancillary services market offering a potential disconnection.
and obviously one can simulate and can analyse whether there may be a policy for that
and when charging they can also participate in ancillary services.
In electric vehicles, we can raise this question:
are we going to have an uncontrolled policy?
this is not in the current debate but it will be in a few years,
we will have an uncontrolled policy where
cars are charged with no control of anything?
maybe it does not create problems in the system. or perhaps some charging
policies to control the system load where price signals have to be given
the system operator can't control this so it has
to be controlled by a signal , such as price.
All of this can be analysed with this software.
Regarding the topic of demand response you can model smart loads
the same properties of generation can be modelled on the load
one can model loads having maximum response times, maximum
generation, so you have a generator behaving as a load
and obviously if you optimize this you can get something super optimal 24/7
but this is modelled and the results are used to generate the required public policy.
Regarding reliability, this is the example of CAISO. CAISO
uses PLEXOS to do these studies to analyse the next 4 months.
it does this using Monte Carlo simulations and selecting 2,000 samples.
takes 161 samples of load, 10 samples of wind and 5 samples of solar
and uses PLEXOS as Montecarlo engine to generate a probability function,
and evaluate which are the risks of the system in the next 4 months
The latest study was released in May and is available on the internet.
An important point regarding Monte Carlo is when you do the
analysis of system reliability using the convolution technique
you only consider the max capacity and forced outage of
each unit then you do the convolution to find the LOLP
but when all these things come together i.e. intermittency, gas restrictions
or even the reservoirs you cannot model it with the convolution
So, what is needed here? One must use Montecarlo and ensure convergence
In the figure the y-axis is the calculated LOLP and in x-axis several
outage samples and you see that close to 400 the value LOLP is converging.
Then PLEXOS can also be used as Monte Carlo engine.
That was my presentation, in summary there are
three important characteristics about the product.
I talked about time, stochastic optimization and risk management.
Now, I will leave you with my colleague Fiac who will illustrate
some applications that PLEXOS has in other countries.
Thank you
Đăng ký:
Đăng Nhận xét (Atom)
Không có nhận xét nào:
Đăng nhận xét