An Open Problem: Energy Data Super-Resolution
An Open Problem: Energy Data Super-Resolution
复制标题
一个悬而未决的问题：能源数据超分辨率
DOI：
10.1145/3427771.3429995
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Wang, Hongning
中科院分区：
文献类型：
--
作者：
Kukunuri, Rithwik;Batra, Nipun;Wang, Hongning
在这篇笔记论文中，我们向Buildsys社区提出了一个开放的问题：能源数据超分辨率，指的是在给定低分辨率功耗的情况下以更高分辨率估计家庭功耗的任务。当智能电表由于带宽、定价、旧硬件等过多问题而以非常低的采样率收集数据时，超分辨率特别有用。这个问题是由计算机视觉界图像超分辨率的成功所激发的。在本文中，我们正式介绍的问题，目前的基线方法和算法，我们用来“解决”这个问题。我们评估的算法在现实世界的数据集上的性能，并讨论结果。我们还讨论了是什么使这个问题很难，为什么一个微不足道的基线是很难击败。
In this notes paper, we present an open problem to the Buildsys community: energy data super-resolution, referring to the task of estimating the power consumption of a home at a higher resolution given the low-resolution power consumption. Super-resolution is especially useful when the smart meters collect data at a very low-sampling rate owing to a plethora of issues such as bandwidth, pricing, old hardware, among others. The problem is motivated by the success of image super resolution in the computer vision community. In this paper, we formally introduce the problem and present baseline methods and the algorithms we used to "solve" this problem. We evaluate the performance of the algorithms on a real-world dataset and discuss the results. We also discuss what makes this problem hard and why a trivial baseline is hard to beat.
使用神经网络进行序列到点学习，用于非侵入式负载监控
DOI：
10.48550/arxiv.1612.09106
发表时间：
2016
期刊：
arXiv e-prints
影响因子：
--
作者：
Zhang Chaoyun
通讯作者：
Zhang Chaoyun