Electricity Demand Activation Extraction
| Autor: | Xueqi Dai, Bérénice Huquet, Pauline Laviron, Themis Palpanas |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
Measure (data warehouse)
Computer science Smart meter Nearest neighbor search 02 engineering and technology Energy consumption 010501 environmental sciences computer.software_genre 01 natural sciences Robustness (computer science) Outlier Scalability 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Data mining computer Energy (signal processing) 0105 earth and related environmental sciences |
| Zdroj: | e-Energy |
| DOI: | 10.1145/3447555.3464865 |
| Popis: | A powerful tool for reducing energy consumption is energy disaggregation (also called NILM Non-Intrusive Load Monitoring), where the goal is to disaggregate the smart meter readings of a household's total electricity consumption to the consumption of that household's individual appliances. State-of-the-art machine learning methods are widely used to solve the NILM problem, but in order to generalize well they require a large amount of data, which are not readily available. We thus need labeled electricity consumption readings from individual appliance activations. Though, manually annotating the start and end of single-appliance activations is extremely laborious and time consuming. Therefore, automated activation extraction methods are needed. Earlier approaches to solve this problem suffer from limitations, such as incomplete signatures, double signatures, and outliers. In this work, we introduce three scalable methods based on techniques that use time series similarity search. The first method is Cartesio that (improves on earlier work that relies on known features of the appliance) and separately detects the start and end times of an appliance activation. The second method is ValmA, a method for identifying previously unknown candidate signatures of variable length, which is essentially parameter-free. The third method is SimBA, a similarity search based method for efficient detection of known signatures in large datasets. These signatures can be computed from the activations extracted using the previous methods. Our experimental results with real 6 and 10 seconds-sampling data demonstrate that, compared to a state-of-the-art solution, our methods improve the accuracy and robustness of appliance activation extraction in very large time series collections. To compare these methods, we also describe a new accuracy measure that takes into account the special characteristics of subsequences, leading to more precise performance evaluation results. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|