Short-term electricity demand forecasting using double seasonal exponential smoothing
Summary (2 min read)
Introduction
- Online electricity demand prediction is required for the control and scheduling of power systems.
- The authors consider online, univariate forecasting of half-hourly data.
- A within-week seasonal cycle, of duration 336 half-hour periods, is evident when one compares the demand on the corresponding day of adjacent weeks.
- The authors adapt the Holt-Winters method so that it can accommodate two seasonalities.
- In the next section, the authors describe how ARIMA models have been adapted for online electricity demand forecasting, in order to capture multiple seasonalities in the demand series.
Multiplicative Double Seasonal ARIMA Models
- The one short-term forecasting method that has remained popular over the years, and appears in many papers as a benchmark approach, is multiplicative seasonal ARIMA modelling.
- The model is often expressed as ARIMA(p,d,q)×(P,D,Q)s.
- The multiplicative seasonal ARIMA model can easily be extended to take care of three or more seasonalities by the introduction of additional polynomial functions of the lag operator and additional difference operators in expression (1).
Double Seasonal Holt-Winters
- The method is only able to accommodate one seasonal pattern.
- This is evident from the recent taxonomies of Hyndman et al.15 and Taylor16.
- The formulation involves separate seasonal indices, Dt and Wt, for the two seasonalities.
- The local s1-period seasonal index, Dt, is estimated by smoothing the ratio of observed value, Xt, to the product of the local level, St, and local s2-period seasonal index, .
- A double additive seasonality method can be developed in a similar way from the standard Holt-Winters method for additive seasonality.
Empirical Comparison of Methods
- The authors carried out empirical analysis in order to address two main issues.
- Secondly, the authors wanted to compare forecasting performance of the new formulation with a well-specified multiplicative double seasonal ARIMA model.
- The data used was 12 weeks of half-hourly electricity demand in England and Wales from Monday 5 June 2000 to Sunday 27 August 2000.
- In practice, interactive facilities tend to be used for special days, which allow operator experience to supplement or override the system offline.
- If a forecasting method is unable to tolerate gaps in the historical series, the special days can be smoothed over, leaving the natural periodicities of the data intact7.
Multiplicative Double Seasonal ARIMA
- The authors used the Box-Jenkins modelling methodology to identify the most suitable ARIMA model based on the 2,688 observations in the estimation sample.
- The autocorrelation function and partial autocorrelation function were used to select the order of the model, which was then estimated by maximum likelihood.
- The residuals were inspected for any remaining autocorrelation.
- Laing and Smith7 explain that, in the multiplicative double seasonal ARIMA formulation in expression (1), polynomials of order greater than two are rarely necessary when fitting a model to half-hourly data for England and Wales.
- The authors investigated differencing and a logarithmic transformation for demand but found neither to improve the SBC.
Holt-Winters Exponential Smoothing
- The authors produced forecasts using the following three Holt-Winters methods: Holt-Winters for Within-Day Seasonality -.
- This is the new Holt-Winters for double multiplicative seasonality, given in expressions (6)-(10), using both a 48-period cycle for the within-day seasonality and a 336-period seasonal cycle for the within-week seasonality.
- 9, the authors also calculated the mean absolute error, root mean square error and root mean square percentage error, but they do not report these results here because the relative performances of the methods for these measures were very similar to those for the MAPE.
- Double Seasonal Holt-Winters outperforms Holt-Winters for Within-Week Seasonality for 38 of the 48 lead times, indicating that there is benefit in using a method that is able to pick up both seasonalities.
Adjusting for Error Autocorrelation in the Holt-Winters Methods
- Inspection of the 1-step-ahead errors, in the estimation sample of 2,688 periods, revealed sizeable first-order autocorrelation for all three Holt-Winters methods, indicating that the forecasts were suboptimal.
- The k-step-ahead forecasts from forecast origin τ are then modified by adding the term λkeτ.
- Chatfield22 found that the modification resulted in improvements in accuracy when applied to the autocorrelated errors from Holt-Winters for multiplicative seasonality.
- This led to far greater improvements in post-sample accuracy than were found using the two-stage estimation approach.
- It is less clear why the new method with residual autocorrelation adjustment outperforms the ARIMA model.
Summary and Conclusions
- Online short-term electricity demand forecasting requires a robust, univariate procedure.
- The authors have shown how the method can be adapted for time series with two seasonalities.
- The Holt-Winters methods were improved by the inclusion of an AR(1) model for the residuals.
- We, therefore, conclude that there is strong potential for the use of the new double seasonal HoltWinters formulation in online short-term electricity demand forecasting.
- Rather than recommending the new method in preference to all others, the authors feel that a more useful approach would be to use several different methods.
Did you find this useful? Give us your feedback
Citations
890 citations
823 citations
Cites background or methods or result from "Short-term electricity demand forec..."
...Taylor (2003a) also obtained somewhat disconcerting results with the Gardner-McKenzie procedure....
[...]
...For time series containing two seasonal cycles, Taylor (2003b) adds one more seasonal component to the A-M method....
[...]
...The second part of Table 4 gives symmetric APEs for the M3 competition data as reported in Makridakis and Hibon (2000), Hyndman et al. (2002), and Taylor (2003a)....
[...]
...Hyndman et al.’s (2002) taxonomy, as extended by Taylor (2003a), is helpful in describing the methods....
[...]
...As Taylor (2003a) observed, generalized Holt is a clumsy way to model a multiplicative trend because the local slope is estimated by smoothing successive differences of the local level....
[...]
776 citations
Cites methods from "Short-term electricity demand forec..."
...…includes an exponential smoothing model for double seasonality that was originally developed for forecasting shortterm electric utility demand (Taylor 2003); a periodic Auto Regressive model; and a model based on robust exponential smoothing based on exponentially weighted least absolute…...
[...]
761 citations
Cites background or methods from "Short-term electricity demand forec..."
...Taylor (2003) extended the linear version of the Holt–Winters method to incorporate a second seasonal component as follows: yt = t−1 + bt−1 + s(1)t + s(2)t + dt, (1a) t = t−1 + bt−1 + αdt, (1b) bt = bt−1 + βdt, (1c) s(1)t = s(1)t−m1 + γ1dt, (1d) s(2)t = s(2)t−m2 + γ2dt, (1e) where m1 and m2 are the…...
[...]
...To avoid the potentially large optimization problem, the initial states are usually approximated with various heuristics (Taylor 2003, 2010b; Gould et al. 2008), a practice that does not lead to optimized seed states....
[...]
...Important exceptions (Harvey and Koopman 1993; Harvey, Koopman, and Riani 1997; Taylor 2003, 2010b; Pedregal and Young 2006; Gould et al. 2008; Taylor and Snyder 2009) handle some but not all of the above complexities....
[...]
...Similarly, in modeling complex seasonality, the existing exponential smoothing models (e.g., Taylor 2003, 2010b; Gould et al. 2008; Taylor and Snyder 2009) suffer from various weaknesses such as overparameterization, and the inability to accommodate both non-integer period and dualcalendar effects....
[...]
...The double seasonal Holt–Winters additive seasonal model described by Taylor (2003) is given by BATS(1,1,0,0,m1,m2), and that with the residual AR(1) adjustment in the model of Taylor (2003, 2008) is given by BATS(1,1,1,0,m1,m2)....
[...]
532 citations
Cites methods from "Short-term electricity demand forec..."
...The seasonal Holt-Winters method has been adapted in order to accommodate the two seasonal cycles in electricity demand series [18]....
[...]
References
19,748 citations
12,650 citations
10,118 citations
5,341 citations
2,029 citations