Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Explorers

Data Science & Analytics


Case C: Slashing Energy Costs with Data Science Powered by IBM Watson Data Platform

Transcript

Buildings account for 40% of U.S. energy consumption. Older buildings spend up to 30% of their annual
budget on utility bills. BlocPower helps nonprofits, churches, and small businesses reduce their energy
costs, while making their buildings healthier and safer. Powered by IBM Watson Data Platform, BlocPower
makes data rich recommendations, so building managers or superintendents can make smarter cases for
investments in energy efficiency improvements.

Tooraj Arvajeh
Chief Engineering Officer, BlocPower

Hi my name is Tooraj, I lead the engineering operation at BlocPower. We use technology to transform
inner city buildings, low-income neighborhoods, and small businesses to become energy efficient,
environmentally friendly, and healthy for the residents, owners, and surrounding neighborhoods.

John J Thomas
Distinguished Engineer and Director, IBM Competitive Project Office

Hi I'm John Thomas my team and I got involved in this very interesting project with BlocPower. When
Tooraj came to us and described his business problem we thought this is a perfect solution to build out
with Watson Data Platform and Watson cognitive services. So I learned a thing or two about buildings, and
saw that there is a relationship between the thermal mass of the building and how the building is used.
What is thermal mass? It is the amount of heat that is held by the structure. Stone has got high thermal
mass which means that there is a lot of energy that has to be spent to heat up that structure. The
structure holds that heat before people can use the insides in a comfortable way.

Now think about what is the least efficient way to use a building with high thermal mass? To use it
sporadically? If you have a church with high thermal mass and you open up the doors of the church only
on Sunday mornings, you have to heat up that whole structure just for the two or three hours that the
building is being used. This isn’t very efficient, however with a church that holds heat in its structure, if the
utilization or the use of that building is much higher, for example they may be running a soup kitchen or
may be opening up the doors for social activities, the energy efficiency actually tends to be better.

I am in the data science experience tool of the Watson data platform. It allows a data scientist and
BlocPower to bring in historical data and do some exploratory analysis with the data. So in my
notebook I am looking at what factors affect the energy efficiency of buildings. The metric I'm working
with is EUI it is a common matrix that describes the energy efficiency of a building. The higher the EUI
value the less efficient that building is and you can see that in the graph here and looking at buildings
across the five boroughs of New York City and the higher EU I values are shaded in darker colors. Here you
can see that Bronx has got a lot of high EUI or low efficiency buildings. This isn’t surprising because Bronx
has got a lot of old stone buildings. Let’s scroll down and look at the distribution of these buildings, looks
like about 50% of them are bricks, 30% are stone and about 20% are concrete. Now look at how I have
done this visualization, a single line of code. The Brunel visualization library lets me create these very
compelling visualizations without having to do a lot of programming. These are all
Explorers
Data Science & Analytics
Case C: Slashing Energy Costs with Data Science Powered by IBM Watson Data Platform

interesting observations but there are no real insights yet till you look deeper at the relationship between
different factors, different attributes and EUI.

Let’s take a look at some other observations this one shows me the relationship between heating systems
and heat distribution systems. Here I have the relationship between high ceilings and big windows. Let's
move on to this final graph here. Seeing the power of unsupervised machine learning at play, my analysis
is telling me that there are three distinct clusters of church buildings: low EUI, medium and high EUI.
When I click on the low EUI one, it tells me the factors that are significant to that particular cluster: not
that old, gas furnaces and so on. With high EUI ones I see other factors that contribute to the behavior of
that church. As a data scientist my analysis has shown me that there are different clusters, types of
buildings in terms of energy efficiency. So if I can build a machine learning model and deploy that when I
have a new church come in I can score that church against machine learning model to understand how
efficient that church is going to be. We will look at how to create that machine learning model you can take
a programmatic approach or you can take the more democratized approach that IBM provides. So let's
take a look at that.

So here I have a guided, a visual way to create my machine learning model. I am in the pipeline artifact of
DSX and I can bring in my data, see a preview of that data. There are transformers that I can work with to
change or work with that initial data. I can decide what features from my data set should be used in the
machine learning model. Let's create a few models and look at how they compare in terms of
performance. I'm going to build three different logistic regression models, they have different features
being used from my data set. When I compare the performance of them I can see visually that one of my
models has got very good performance. The area under the ROC curve is 0.98 extremely good ,when I do
that with my untrained data I can see that the model still behaves very good, the performance is very high
and I am now satisfied that this model will behave well enough for me to deploy and operationalize.
With a single click I can deploy this it gives me an Endpoint and an API key. These are the only two
pieces of information that my app developer needs to know. So I'm going to pass this information on to
the app developer team.

Tooraj Arvajeh
Chief Engineering Officer, BlocPower

This is where the power of technology and innovation comes in. So we said what if we make the
knowledge and expertise our engineers accessible to everyone via their mobile phones? In other words
every building owner, manager and tenant can have an engineer friend who can guide them on how to
conduct a technical survey of their buildings, identify replacement measures to reduce their bills and
evaluate their cost and benefits. We also wanted this virtual engineer to be interactive and intelligent, so
we decided to talk to our friends at IBM and to see what solutions they can offer us.

John J Thomas
Distinguished Engineer and Director, IBM Competitive Project Office

BlocPower has partnered with a company that builds mobile applications using a platform called harmony
built on compose. The app development team uses these APIs to invoke the machine learning model that
has been built and deployed. What if you could engage the end user in a more compelling way? We are
going to use Watson conversation services and Watson image recognition and extract information from
Explorers
Data Science & Analytics
Case C: Slashing Energy Costs with Data Science Powered by IBM Watson Data Platform

the superintendent about the church, enough information to build a profile of that church and then invoke
the machine learning model and get a score back in terms of how energy efficient it is.

We took the best of machine learning and deep learning and brought them together. The best of Watson
data platform and Watson cognitive services working together to build a compelling solution for
BlocPower.

You might also like