tag:blogger.com,1999:blog-3236722478096264150.post9034797163351750646..comments2020-01-15T11:22:11.686+00:00Comments on Disaggregated Homes: Why NIALM shouldn't be modelled as the knapsack or subset-sum problemUnknownnoreply@blogger.comBlogger5125tag:blogger.com,1999:blog-3236722478096264150.post-67637948983015647502013-04-06T16:46:11.173+01:002013-04-06T16:46:11.173+01:00Hi Oliver, for sure we are planning to evaluate ou...Hi Oliver, for sure we are planning to evaluate our approach using real data, but up to now we have not decided which data set to use.<br /><br />Concerning your question, we did not observed this behavior but we also did not evaluated in detail if the algorithm is dependent on the size of the appliance power. But we are planning to evaluate this in future work. In the current work the used appliances are generated randomly in predetermined limits. If we have presentable news I'll let you know!Dominik Egarterhttps://www.blogger.com/profile/02309482391767313613noreply@blogger.comtag:blogger.com,1999:blog-3236722478096264150.post-85586240368819617612013-04-06T16:45:13.598+01:002013-04-06T16:45:13.598+01:00This comment has been removed by the author.Dominik Egarterhttps://www.blogger.com/profile/02309482391767313613noreply@blogger.comtag:blogger.com,1999:blog-3236722478096264150.post-56244460706261058402013-04-05T22:08:32.918+01:002013-04-05T22:08:32.918+01:00Hi Dominik, that sounds really interesting, thanks...Hi Dominik, that sounds really interesting, thanks for the update! Are you planning to evaluate your approach using real data? I'd love to see a discussion of the advantages/disadvantages using the REDD data set.<br /><br />Also, I had a thought while reading your paper which I'd be interested to hear your opinion on. You model disaggregation as a knapsack problem, in which the profit of each appliance is 1. My understanding is that the optimisation will therefore favour solutions of large numbers of low energy appliances rather than fewer high energy consuming appliances. Did you observe this behaviour in your experiments?Oliver Parsonhttps://www.blogger.com/profile/13718839510983451443noreply@blogger.comtag:blogger.com,1999:blog-3236722478096264150.post-83517997222964803072013-04-05T21:56:33.770+01:002013-04-05T21:56:33.770+01:00Thanks for discussing our paper in your blog :)
At...Thanks for discussing our paper in your blog :)<br />At the current state we have, as discussed, a strong abstraction of real appliances, but we are already preparing a sophisticated model, where improvements of the model and optimization approach are shown. If I have news I'll let you know.<br /><br />Dominik Egarterhttps://www.blogger.com/profile/02309482391767313613noreply@blogger.comtag:blogger.com,1999:blog-3236722478096264150.post-26309704023210452712013-04-05T10:13:24.085+01:002013-04-05T10:13:24.085+01:00Very interesting, thank you for highlighting this ...Very interesting, thank you for highlighting this paper (and their blog). It's an interesting approach. I have a lot of sympathy with the idea that we should try to take advantage of state duration information <em>where appropriate</em>. But I completely agree with your critiques of this paper.<br /><br />As you say, appliance durations are highly variable. Even for appliances whose state durations fall within a relatively tight range (like toasters and kettles) the duration can change by around +/- 50%. And appliances which, at first glance, appear to follow a strict program (like a washing machine) show variation in their state durations (for example, a washing machine's heater only remains on long enough to bring the water temperature up to a set point... and that duration depends on a number of external factors). And, then, of course there are all the appliances whose state durations vary massively (lights, TVs, computers etc etc)<br /><br />They say in their "Future work" section that <br /><br /><em><br />"The current version of the algorithm was tested for on/off appliances with constant time duration. In future work, we plan to extend our algorithm for arbitrary shapes of power pro les and to evaluate the approach using real appliances"<br /></em><br /><br />So it'll be interesting to see how well their system works on real data. TBH, it's a shame they didn't evaluate their approach using <em>any</em> real data in this first paper. Granted, gathering your own real data is a PITA, but - as you've suggested in the past - it's easy to synthesise very realistic data using TraceBase<br /><br />(by the way, I really value your blog posts on recent NIALM papers. It's like an on-line NILM reading group ;) )Jack Kellyhttps://www.blogger.com/profile/14542966078686299001noreply@blogger.com