MSP-SEC

For predicting our energy prices, we built a model in Azure Machine Learning Studio and used two regression models, to compare the results
Our first model is a Neural Network Regression which is far more auccurate than our second model a Linear Regression.
Our Machine Learning model is acessable by a REST API.
Furthermore our machine learning model has a web service.
For testing our function, we used a raspberry pi simulator in the beginning and built our own pi afterwards.
A httpsTrigger allowed us to send a messega via an eventhub to an iot hub to the raspberry pi itself.
Through this whole process we had to create some services.

Inspiration

In times where energy and more accurately electricity is such a desirable commodity it is very important to have efficient mechanisms to store, provide and consume electricity. The biggest challenge is the great variance (also vary in demand and supply making energy providing very inefficient and expensive. We wanted to solve this problem, because it has a great impact and can affect the world we live in in a positive manner.

What it does

End devices that consume significant ammount of electricity (like a server farm, Cooling/heating System of the TUM Informatics department....) are equipped with an IoT device that is connected to the SPOT Market trading Platform and Azure IoT Hub. Based on the data and the predictive artificial functions provided by Azure the system predicts the prices for the next time interval. Futhermore, the end device has a defined performance intensity interval. This means that if in the next time interval the price for electricity is very high the End device will run on 80% performance intensity - if possible.

To clarify this mechanism we provide the following example: Let us assume that the Informatics department can be heated to a tempreature between 20°C and 24°C. If now we have predicted the price to be high for the next 2 hours then we will only heat up to 20°C to save energy.

The second feature is that our intelligent system is equipped with a smart scheduler. This means if a huge server farm has to run updates anytime tomorrow between 1 PM and 11 PM and the update time is 2h, our system will determine the time interval where this can be done with the minimum price and by doing that contributing to decrease the inefficient usage of grid capacity, because it creates an demand when there is a higher suppy than demand (reason for cheaper prices).

How we built it

Rasperry Pi connected to Azure IoT Hub, where the coordination of the Pi (The Pi is a simulated end device) is maintained and the predictions are run in Azure Machine Learning using a Neural Network regression model. Furthermore we have an R environment to analyze the data further and all sub systems are connected via Rest APis and controlled by a web server.

Challenges we ran into

We ran into some challenges with setting up and wiring up the Raspberry Pi and setting up R on a Ubuntu Server. But the bigger challenge was the communication and combination of all our different parts, that are summed up and pooled on our website.

However nothing that could stop us.

Accomplishments that we're proud of

All in all, were are proud to have such a big variety of techs connected to each other and communicating with each other. The range is from C#, Node.js, LEDs and resitors on a breadboard to R and a Apache server in the Azure Cloud.

For example our Raspberry Pi that is connected to an Azure IoT-Hub and is controllable via a URL. This is possible, because we connected a httpTrigger to an EventHub that is connected to the IoT-Hub. Therefore we are able to give the Pi commands in any web-browser or program.

What we learned

We learned a lot about wiring the Raspberry Pi, creating and cleaning Data to get a usable DataSet for machine learning models and about connecting everything together with for example Azure Functions and Rest.

What's next for MSP-SEC

The revolution in energy distribution!

Code and Presentation

Most parts of the code and the date we used, are accessible through the links in "Try it out". We tried to make our code, which runs in the cloud in most parts, accessible to you by loading it into a public git repository. Furthermore you can visit our website as the control center for the user and slides of our presentation are available in the stated git repo.