Australian state-level demand

The rawdata for other states are also stored in the tsibbledata github repository (under the data-raw folder), but these are not included in the package to satisfy CRAN space constraints. However, anyone can still load and use the data with the following code.

library(tidyverse)
library(lubridate)
library(tsibble)
repo <- "https://raw.githubusercontent.com/tidyverts/tsibbledata/master/data-raw/vic_elec/"
states <- c("NSW","QLD","SA","TAS","VIC")
dirs <- paste0(repo, states, "2015")

# Read holidays data
holidays <- paste0(dirs,"/holidays.txt") %>%
  as.list() %>%
  map_dfr(read_csv, col_names=FALSE, .id="State") %>%
  transmute(
    State = states[as.numeric(State)],
    Date = dmy(X1), 
    Holiday = TRUE
  )
# Read temperature data
temperatures <- paste0(dirs,"/temperature.csv") %>%
  as.list() %>%
  map_dfr(read_csv, .id = "State") %>%
  mutate(
    State = states[as.numeric(State)],
    Date = as_date(Date, origin = ymd("1899-12-30"))
  )
# Read demand data
demands <- paste0(dirs,"/demand.csv") %>%
  as.list() %>%
  map_dfr(read_csv, .id = "State") %>%
  mutate(
    State = states[as.numeric(State)],
    Date = as_date(Date, origin = ymd("1899-12-30"))
  )
# Join demand, temperatures and holidays
aus_elec <- demands %>%
  left_join(temperatures, by = c("State", "Date", "Period")) %>%
  transmute(
    State,
    Time = as.POSIXct(Date + minutes((Period-1) * 30)),
    Period,
    Date = as_date(Time),
    DOW = wday(Date, label=TRUE),
    Demand = OperationalLessIndustrial, 
    Temperature = Temp,
  ) %>%
  left_join(holidays, by = c("State", "Date")) %>%
  replace_na(list(Holiday = FALSE))
# Remove duplicates and create a tsibble
aus_elec <- aus_elec %>%
  filter(!are_duplicated(aus_elec, index=Time, key=State)) %>%
  as_tsibble(index = Time, key=State)

This block of code reads in raw data files containing holiday information, temperatures and electricity demand for each state, and then joins them into a single tsibble. For some reason, there are duplicated rows from South Australia, so the last few lines removes the duplicates before forming a tsibble, keyed by State.

aus_elec
## # A tsibble: 1,155,408 x 8 [30m] <UTC>
## # Key:       State [5]
##    State Time                Period Date       DOW   Demand Temperature
##    <chr> <dttm>               <dbl> <date>     <ord>  <dbl>       <dbl>
##  1 NSW   2002-01-01 00:00:00      1 2002-01-01 Tue    5714.        26.3
##  2 NSW   2002-01-01 00:30:00      2 2002-01-01 Tue    5360.        26.3
##  3 NSW   2002-01-01 01:00:00      3 2002-01-01 Tue    5015.        26.3
##  4 NSW   2002-01-01 01:30:00      4 2002-01-01 Tue    4603.        26.3
##  5 NSW   2002-01-01 02:00:00      5 2002-01-01 Tue    4285.        26.3
##  6 NSW   2002-01-01 02:30:00      6 2002-01-01 Tue    4075.        26.3
##  7 NSW   2002-01-01 03:00:00      7 2002-01-01 Tue    3943.        26.3
##  8 NSW   2002-01-01 03:30:00      8 2002-01-01 Tue    3884.        26.3
##  9 NSW   2002-01-01 04:00:00      9 2002-01-01 Tue    3878.        26.3
## 10 NSW   2002-01-01 04:30:00     10 2002-01-01 Tue    3838.        26.3
## # … with 1,155,398 more rows, and 1 more variable: Holiday <lgl>

This data set contains half-hourly data from all states from 1 January 2002 – 1 March 2015 (and in the case of Queensland to 1 April 2015). The temperature variable is from a weather station in the capital city of each state.

GEFCOM 2017

The Global Energy Forecasting Competition in 2017 involved data on hourly zonal loads of ISO New England from March 2003 to April 2017. The data have already been packaged into tibble format by Cameron Roach in the gefcom2017data Github repository. So it is relatively easy to convert this to a tsibble.

devtools::install_github("camroach87/gefcom2017data")
library(gefcom2017data)
gefcom2017 <- gefcom %>% 
  ungroup() %>%
  as_tsibble(key=zone, index=ts)
gefcom2017
## # A tsibble: 1,241,710 x 15 [1h] <UTC>
## # Key:       zone [10]
##    ts                  zone  demand drybulb dewpnt date        year month
##    <dttm>              <chr>  <dbl>   <dbl>  <dbl> <date>     <dbl> <fct>
##  1 2003-03-01 00:00:00 CT      3386      25     19 2003-03-01  2003 Mar  
##  2 2003-03-01 01:00:00 CT      3258      23     18 2003-03-01  2003 Mar  
##  3 2003-03-01 02:00:00 CT      3189      22     18 2003-03-01  2003 Mar  
##  4 2003-03-01 03:00:00 CT      3157      22     19 2003-03-01  2003 Mar  
##  5 2003-03-01 04:00:00 CT      3166      23     19 2003-03-01  2003 Mar  
##  6 2003-03-01 05:00:00 CT      3255      23     20 2003-03-01  2003 Mar  
##  7 2003-03-01 06:00:00 CT      3430      24     20 2003-03-01  2003 Mar  
##  8 2003-03-01 07:00:00 CT      3684      24     20 2003-03-01  2003 Mar  
##  9 2003-03-01 08:00:00 CT      3977      25     21 2003-03-01  2003 Mar  
## 10 2003-03-01 09:00:00 CT      4129      27     22 2003-03-01  2003 Mar  
## # … with 1,241,700 more rows, and 7 more variables: hour <dbl>,
## #   day_of_week <fct>, day_of_year <dbl>, weekend <lgl>,
## #   holiday_name <chr>, holiday <lgl>, trend <dbl>

Details of the data (and the competition) are available on Tao Hong’s website.