Exportar Publicação

Martins, A. A. A. F., Lagarto, J., Canacsinh, H. , Reis, F. & Cardoso, M. G. M. S. (2021). Short-term load forecasting using time series clustering . 16th Conference on Sustainable Development of Energy, Water and Environment Systems (SDEWES) .

A. A. Martins et al.,  "Short-term load forecasting using time series clustering ", in 16th Conf. on Sustainable Development of Energy, Water and Environment Systems (SDEWES) , 2021

@misc{martins2021_1734878556965,
	author = "Martins, A. A. A. F. and Lagarto, J. and Canacsinh, H.  and Reis, F. and Cardoso, M. G. M. S.",
	title = "Short-term load forecasting using time series clustering ",
	year = "2021",
	url = "https://www.dubrovnik2021.sdewes.org/"
}

TY  - CPAPER
TI  - Short-term load forecasting using time series clustering 
T2  - 16th Conference on Sustainable Development of Energy, Water and Environment Systems (SDEWES) 
AU  - Martins, A. A. A. F.
AU  - Lagarto, J.
AU  - Canacsinh, H. 
AU  - Reis, F.
AU  - Cardoso, M. G. M. S.
PY  - 2021
UR  - https://www.dubrovnik2021.sdewes.org/
AB  - Short-term load forecasting has a major role in energy planning. The accuracy of load forecasting
has a direct impact on the way the power system is operated and managed and it is essential for power markets strategy. For systems operators short-term load forecasting can be used to perform fundamental operations such as economic dispatch and unit commitment, as well as to coordinate generation between hydro and thermal units. It can also be used to establish spinning
reserve values in advance of any critical situation that might arise in the power system and it is
based on load time series.
Load time series are volatile, non-linear and non-stationary and depend on multiple factors,
namely, meteorological (e.g. temperature), calendar (e.g. holidays, weekends, working days),
network topology (e.g. load shifting) and random noise.
We propose a new approach to deal with short-term load forecasting. It resorts to a clustering
algorithm, using K-Medoids with a combination of different dissimilarity measures to deal with
the complex nature of load data, capturing differences in time series trends, values, cyclical
behaviors and autocorrelation patterns. A summated indicator of several (normalized) cohesion-separation
indices is used to determine the number of clusters. Since load data depend heavily on
meteorological factors, the temperature time series is also considered. Finally, we resort to
similarity pattern sequence searching in the historical data set.
The proposed approach is applied to 2014-2017 time series data of a system operator including
load (at the power system level) and temperature data in 15-minutes intervals. They are used to
obtain the load forecast for the 96 periods of the day-ahead which is an important input for
internal processes of the system operator such as operational planning. The results obtained are
promising, when compared with alternative similarity pattern sequence approaches
ER  -