Location visible to members
5:15 onwards - arrive, check in at reception, beer and pizza on upper floor balcony
6:00 - 6:45 - first 2 talks
6:45 - 7:00 - break
7:00 -7:45 - 2nd 2 talks
Join us at AGL's new offices above Southern Cross station for some energy related data science talks.
To access the building, go up the escalator on the far right, accessed from the footbridge as seen in the photo below, where you will arrive in a lobby with a cafe. You will need to sign in to get through security - there will be someone there to arrange this.
Electric Load Forecasting with Neural Networks
Forecasting short term electricity demand is necessary in order to schedule supply to match demand so there are no power cuts. This talk will show how looking at model errors was used as a technique to determine the factors that contribute to electricity demand.
Saving lines - electricity conductor fault detection
Arman Tajbakhsh & Peter Barnes (Geomatic Technologies)
There are many kilometres of overhead electricity lines around the world. Conductor faults, ranging from corrosion to breaks, affect the quality and continuity of power supply and can have catastrophic consequences for communities. We’ll show how utilities including AusNet acquire high resolution imagery then analyse the data to identify potential faults. We’ll discuss considerations and rationale around the technical solution, challenges, results, and where to next.
Green Algorithms : Saving Energy
February 2016 was the hottest month on record, ever. Server farms currently consume 3% of the world’s electricity and now have the same carbon footprint as the airline industry. Data centre energy consumption is doubling every four years. Hence it is essential for us to be aware of the energy consumed during data processing and also look at ways to reduce this energy consumption. At CEET, we are currently analyzing the energy consumed by some of the top data mining algorithms and also looking at ways to reduce algorithm energy consumption. kmeans clustering algorithm is a popular data mining algorithm. In this talk, using the example of kmeans clustering algorithm, I will illustrate various factors affecting energy consumption of algorithms. Implementation of distributed clustering by splitting a large data sets into smaller chunks reduce processing memory requirement and energy consumption. However, there are limits to the gains in processing speed and savings in energy consumption obtained through the “divide and conquer” approach of processing large data sets.
Solving fault and phase detection use cases with smart meter data
Peter McTaggart (Powercor), Lyndon Maydwell & Adel Foda (Silverpond)
The introduction of AMI smart meters has created a wealth of data useful for network diagnostics. We will discuss how smart meter data can be used to perform neutral fault detection and phase identification, and some of the applications these enable.
Additionally, the presentation will cover some of the challenges and solutions involved with constructing a data-pipeliine for scaled data-science operations. Architecture and technologies significantly impact development, testing, and operations. As such, they should be considered carefully when building a data-science platform.