Univariate Methods for Short‐Term Load Forecasting

Parsimonious Short-Term Load Forecasting for Optimal Operation Planning of Electrical Distribution Systems

“…and on-site by taking advantage of the limited storage and processing capabilities of smart meters in the field. Using a small resolution (e.g., 5 minutes per sample), the short-term memory of regression models can capture slow consumption patterns affected by exogenous variables, such as temperature and humidity [1]. Moreover, control actions that affect the collective consumption in a seasonal fashion (e.g., demand response at certain hours of the day, scheduled charging of EVs, etc.)…”

“…where N is the interval of the data series used to fit the model and K is the interval of the lags under study. If the model is appropriate the statisticQ is approximately distributed as χ 2 (K − p − q), where p = 0 and q = 2 in the ARIMA model (1). Thus, considering K = 4, 032 (i.e., two weeks) and fitting the model using N = 34, 944 samples (i.e., one year) the value ofQ was approximately 3,032 using the frequentist approach, and 3,472 for the Bayesian approach.…”

“…S HORT-TERM load forecasting is of great importance in the operation planning of bulk energy systems. Regulation bids, energy arbitrage, and market-clearing mechanisms are conducted on hourly bases, which puts a lot of pressure on forecasting techniques to provide accurate and fast estimations of future demands [1]. With the advent of smart meters and advanced distributed automation, electrical distribution system (EDS) operators are also becoming active users of short-term operation planning [2].…”

“…As shown in previous empirical studies, multiplicative autoregressive integrated moving average (ARIMA) models are simplistic and accurate analytical methods for short-term forecasting of aggregated demands [1]. All forecasting techniques are subject to error but, when dealing with univariate seasonal stochastic time series, regression models are regarded as the best option to provide fast and reliable forecasts [21].…”

Bayesian Inference of Triple Seasonal Autoregressive Models

“…Since the traditional seasonal autoregressive moving average (SARMA) models can not capture multiple seasonalities, some researchers have extended these models to model and forecast time-series with multiple seasonalities, see for example Amin (2018) in the Bayesian framework, and see also De Livera et al (2011) and Sulandari et al (2021) in the non-Bayesian framework. Modeling time-series with two seasonality layers in the non-Bayesian framework has been the interest of several researchers, see for example Taylor (2008bTaylor ( , 2008a, Ryu et al (2017), Deb et al (2017), Taylor and McSharry (2017) and Lago et al (2018). However, few work have been introduced for the analysis of time-series with three seasonality layers, see for example Taylor (2010bTaylor ( , 2010a De Livera et al (2011), Taylor and Snyder (2012) and Dumas and Cornélusse (2018).…”