Model-Free Iterative Temporal Appliance Discovery for Unsupervised Electricity Disaggregation
Authors: Mark Valovage, Akshay Shekhawat, Maria Gini
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We test iterative appliance discovery against an existing competitive unsupervised method using two publicly available datasets. Results using different sampling rates show iterative discovery has faster runtimes and produces better accuracy. |
| Researcher Affiliation | Academia | Mark Valovage, Akshay Shekhawat, Maria Gini Computer Science and Engineering University of Minnesota, Minneapolis, MN {valov002,shekh027,gini}@umn.edu |
| Pseudocode | Yes | Algorithm 1: Iterative Episode Discovery |
| Open Source Code | No | The paper does not provide any statement or link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | BLUED: The Building-Level f Ully-labeled dataset for Electricity Disaggregation (BLUED) contains power measurements sampled at 60 Hz over a period of 1 week for a single house with 43 appliances (Anderson et al. 2012a). REDD: The Reference Energy Disaggregation Dataset (REDD) contains real power measurements for six houses (Kolter and Johnson 2011). |
| Dataset Splits | No | The paper uses publicly available datasets and discusses performance across different sampling rates and event detection methods, but it does not specify explicit training, validation, or test dataset splits for its own evaluation. |
| Hardware Specification | No | For each house, we allowed each method to run for 24 hours on a conventional laptop. This description is too vague and does not provide specific hardware details (e.g., CPU, GPU models, memory). |
| Software Dependencies | No | The paper mentions methods like 'Bayesian change detection' and 'genetic k-means' but does not list specific software libraries or their version numbers that are necessary to replicate the experiment. |
| Experiment Setup | Yes | For the maximum episode length, Lmax, we found diminishing returns for Lmax > 5... The max horizon window, Wmax... we experimented in the range Wmax [5, 30], and set Wmax = 20 in our results below. For the power threshold, we used a grid search to explore ν [5, 100] watts in increments of 5 watts. |