Clustering time series of smart meter measurements

undergraduate thesis

Matija Teršek (Author), Erik Štrumbelj (Mentor)

Abstract

In this thesis we provide a compact review of 8 time series representations in combination with 2 clustering algorithms and 2 indices for internal clustering validation. We analyse time series measured by smart meter devices and check how their representations affect clustering. We conclude that no representation can be directly used for the task and that more focus should be put on preprocessing. Additionally, we compare representations and 4 similarity measures on simulated time series. We find out that similarity measures outperform representations in most cases and that a variational autoencoder-based representation works the best for simulated time series.

Keywords

time series;similarity measures;representations;clustering;recurrent neural networks;variational autoencoders;computer science;computer and information science;computer science and mathematics;interdisciplinary studies;diploma;

Data

Language:	English
Year of publishing:	2020
Typology:	2.11 - Undergraduate Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[M. Teršek]
UDC:	004(043.2)
COBISS:	28477443
Views:	451
Downloads:	115
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	Slovenian
Secondary title:	Gručenje časovnih vrst meritev pametnih merilnih naprav
Secondary abstract:	V diplomskem delu povzamemo 8 predstavitev časovnih vrst v kombinaciji z 2 algoritmoma za gručenje in 2 indeksoma za interno validacijo gručenja. Eksperimentalno preverimo vpliv predstavitev časovnih vrst na gručenje podatkov, ki so jih izmerile pametne merilne naprave. Ugotovimo, da nobena izmed predstavitev ni takoj in neposredno uporabna, in da se je bolj pomembno osredotočiti na predprocesiranje. Uporabnost predstavitev časovnih vrst v gručenju preverimo tudi na umetnih podatkih. Rezultate primerjamo z gručenjem celih časovnih vrst, kjer uporabimo 4 različne mere podobnosti. Ugotovimo, da so mere podobnosti v večini primerov boljše, najbolje pa se obnese predstavitev, ki temelji na variacijskem avtokodirniku.
Secondary keywords:	časovne vrste;predstavitve;gručenje;rekurzivne nevronske mreže;variacijski avtokodirnik;mere podobnosti;računalništvo;računalništvo in informatika;računalništvo in matematika;interdisciplinarni študij;univerzitetni študij;diplomske naloge;
Type (COBISS):	Bachelor thesis/paper
Study programme:	1000407
Embargo end date (OpenAIRE):	1970-01-01
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	48 str.
ID:	12027631