blog

MSc project proposal: An online competition for comparing energy disaggregation algorithms

A sizeable challenge in the energy disaggregation community is that of comparing NILM algorithms from different researchers. In other words, if we have two papers, and one paper reports an accuracy of 80%, and another reports an accuracy of 85% then we cannot infer that the second paper is better because the authors used different datasets, different pre-processing etc. Hence we are working on a project proposal for the consideration of Imperial Computer Science MSc students. If a group of students selects the project then they'll work on it for the duration of next term. Here's the full, draft project specification. Comments most welcome!

NILMTK v0.2 wins best demo at BuildSys 2014; and demo is online

Oli Parson and Nipun Batra (my two awesome collaborators on the core of our open source energy disaggregation framework NILMTK) have just attended BuildSys 2014 in Memphis. Oli presented NILMTK v0.2 as a demo and it won best demo! Hurray! Well done Oli for doing an excellent presentation. The demo is an IPython notebook and is available to view online: NILMTK v0.2 BuildSys demo. Also, you can read more about energy disaggregation at BuildSys on Oli's blog.

Make machine learning easy enough for kids to use in their creations (MSc group project proposal)

The aim of this project is to make sophisticated machine learning tools so easy to use that kids (and adults with little knowledge of computer science) can use machine learning algorithms in their creations. The approach is to wrap machine learning tools as simple ‘building blocks’ which can be bolted together with existing rapid prototyping tools to allow users to quickly throw together sophisticated inventions. e.g. Lego robots which can classify objects using a video camera, or can understand natural speech, or learn some task through trial and error (e.g. balancing on one leg). Full (draft) spec here

16kHz recordings of voltage and current for whole home for >1 year available

Our 'UK Domestic Appliance-Level Electricity (UK-DALE)' dataset was released earlier this year (here's our paper on the dataset). Up until now, the data available consisted of data recorded every second (for the whole house power demand of two houses) or recorded every six seconds (for all appliance-level data and the whole-house demand for all four houses). But I also recorded voltage and current at 16kHz for two houses. I have finally gotten round to figuring out how to put almost 4 TBytes of data online.

The 16kHz signal is compressed as FLAC and is available in 1 hour chunks (each chunk is about 200 MBytes in size). You can download it using anonymous FTP. Details here under the section 'The full 16 kHz dataset via FTP'.

And, while we're on the topic of updates to UK-DALE, a little while ago I updated the metadata for UK-DALE to bring it into line with the new NILM Metadata v0.2 schema, and I also updated the NILMTK converter for UK-DALE.

Site now requires login to post comments

My blog has been getting a lot of comment spam which has been tedious to try to resolve. I am using several anti-spam Drupal modules and these catch a lot of spam but about 10 spam messages were still getting through per day. It got so bad that I completely disabled comments on my site but this was clearly not a good solution. So I have changed the permissions so users must log into my site before you can post comments. You can register a new account on this site or you can log in using Facebook, Google or Twitter (achieved using the Drupal HybridAuth plugin).

Report from NILM2014@London on comparing NILM algorithms

The first "NILM in London" workshop was held on Wednesday 3rd September. In this blog post, I'd like to try to summarise the discussion around comparing NILM algorithms.

Report from NILM2014@London on building an online discussion venue and wiki for NILM

P1060562.jpgP1060562.jpg

The first "NILM in London" workshop was held on Wednesday 3rd September. It was a lot of fun and we had some great conversations. In this blog post, I'd like to try to summarise the discussion around building an online discussion forum for NILM.

At IBM until end of October

Hi,

Sorry I haven't posted many blog posts recently. I'm currently about half way through a 3-month placement at IBM in Hursley, UK.

NILMTK v0.2 released!!

Hurray! We've released NILMTK v0.2! See this list of improvements on the docs page.

Introducing NILM Metadata: a schema for energy datasets and prior knowledge about appliances

One of our aims with the open-source energy disaggregation toolkit NILMTK is to make it easy to import any of the 10+ NILM datasets currently available. One of the pain points when writing a NILMTK importer for a new dataset is that each dataset uses a different metadata schema and, sometimes, there simply is no metadata associated with some datasets. At best, this means that we have to manually map from the dataset's appliance names to NILMTK's standard appliance names. At worse, it means that it's impossible to unambiguously import the dataset (did that channel really only include the fridge? It sure looks like there are other appliances on there. What's the wiring hierarchy between the mains meter, the circuit meters and the appliance meters? Does that channel record active power or apparent power? What pre-processing has already been applied? etc. etc.)

Pages

Subscribe to RSS - blogs