Title: Electricity Load Curves Forecasting at Individual Level
Version: 0.4.0
Date: 2026-04-07
Depends: R (≥ 3.5.0)
Imports: stats, utils, mgcv, earth, keras3, Rcpp, tensorflow, data.table, wavelets, jsonlite, xml2, DBI, RSQLite
LinkingTo: Rcpp
Suggests: knitr, reticulate, rmarkdown, testthat (≥ 3.0.0)
Author: Frederic Bertrand ORCID iD [cre, aut], Fatima Fahs [aut], Myriam Maumy-Bertrand ORCID iD [aut]
Maintainer: Frederic Bertrand <frederic.bertrand@lecnam.net>
Description: Implements forecasting methods for individual electricity load curves, including Kernel Wavelet Functional (KWF), clustered KWF, Generalized Additive Models (GAM), Multivariate Adaptive Regression Splines (MARS), and Long Short-Term Memory (LSTM) models. Provides normalized dataset adapters for iFlex, StoreNet, Low Carbon London, and REFIT; download and read support for IDEAL and GX; explicit Python backend selection for TensorFlow-based LSTM fits; helpers for daily segmentation and rolling-origin benchmarking; and compact shipped example panels and benchmark-result datasets.
LazyLoad: yes
LazyData: yes
VignetteBuilder: knitr
License: GPL-3
Encoding: UTF-8
URL: https://fbertran.github.io/elcf4R/, https://github.com/fbertran/elcf4R
BugReports: https://github.com/fbertran/elcf4R/issues
RoxygenNote: 7.3.3
NeedsCompilation: yes
Packaged: 2026-04-17 06:43:24 UTC; bertran7
Repository: CRAN
Date/Publication: 2026-04-21 20:52:21 UTC

Forecasting Individual Electricity Load Curves

Description

elcf4R provides methods and supporting workflows for day-ahead forecasting of individual electricity load curves. The current package surface includes Kernel Wavelet Functional models, clustered KWF, GAM, MARS and LSTM estimators, an explicit helper to configure the Python backend used by the LSTM path, dataset adapters for iFlex, StoreNet, Low Carbon London and REFIT, scaffolded download/read support for IDEAL and GX, helpers to build daily segments, and rolling-origin benchmarking utilities.

Author(s)

Maintainer: Frederic Bertrand frederic.bertrand@lecnam.net (ORCID)

Authors:

See Also

Useful links:

Assign segments to a fitted KWF clustering model

Description

Assign segments to a fitted KWF clustering model

Usage

elcf4r_assign_kwf_clusters(object, segments)

Arguments

object

An elcf4r_kwf_clusters object.

segments

Matrix or data frame of daily segments.

Value

A character vector of cluster labels.

Run a rolling-origin benchmark on a normalized panel

Description

Evaluate the package forecasting methods on a normalized panel using a deterministic rolling-origin design. The runner supports the current temperature-aware gam, mars, kwf, kwf_clustered and lstm wrappers and returns both aggregate scores and, optionally, saved point forecasts.

Usage

elcf4r_benchmark(
  panel,
  benchmark_index = NULL,
  methods = NULL,
  entity_ids = NULL,
  cohort_size = NULL,
  train_days = 28L,
  test_days = 5L,
  benchmark_name = NULL,
  dataset = NULL,
  use_temperature = TRUE,
  method_args = NULL,
  include_predictions = TRUE,
  thermosensitivity_panel = NULL,
  benchmark_index_carry_cols = NULL,
  seed = NULL,
  tz = "UTC"
)

Arguments

panel

Normalized panel data, typically returned by one of the ⁠elcf4r_read_*()⁠ adapters.

benchmark_index

Optional day-level index. If NULL, it is created with elcf4r_build_benchmark_index().

methods

Character vector of method names to evaluate. Supported values are "gam", "mars", "kwf", "kwf_clustered" and "lstm". If NULL, the runner uses gam, mars, kwf, kwf_clustered and adds lstm only when its backend is available.

entity_ids

Optional character vector of entity IDs to benchmark.

cohort_size

Optional maximum number of eligible entities to keep after sorting by entity_id.

train_days

Number of days in each training window.

test_days

Number of one-day rolling test origins per entity.

benchmark_name

Optional benchmark identifier. If NULL, one is derived from the dataset label and benchmark design.

dataset

Optional dataset label overriding unique(panel$dataset).

use_temperature

Logical; if TRUE, methods that support temperature will use it when non-missing temperature information is available for the current window.

method_args

Optional named list of per-method argument overrides.

include_predictions

Logical; if TRUE, return a long table of saved point forecasts and naive forecasts.

thermosensitivity_panel

Optional normalized panel used for thermosensitivity classification. Defaults to panel.

benchmark_index_carry_cols

Optional carry_cols passed to elcf4r_build_benchmark_index() when benchmark_index is not supplied.

seed

Optional integer seed forwarded to methods that support user-supplied seeding, such as LSTM, unless overridden in method_args.

tz

Time zone used to derive dates and within-day positions.

Value

An object of class elcf4r_benchmark with elements results, predictions, cohort_index, spec and backend.

Examples

id1 <- subset(
  elcf4r_iflex_example,
  entity_id == unique(elcf4r_iflex_example$entity_id)[1]
)
keep_dates <- sort(unique(id1$date))[1:6]
panel_small <- subset(id1, date %in% keep_dates)

bench <- elcf4r_benchmark(
  panel = panel_small,
  methods = "gam",
  cohort_size = 1,
  train_days = 4,
  test_days = 1,
  include_predictions = TRUE
)
head(bench$results)

Build a day-level benchmark index from a normalized panel

Description

Create a compact day-level index from a normalized panel. The returned object contains one row per complete entity-day and can be reused to define deterministic benchmark cohorts without shipping the full panel.

Usage

elcf4r_build_benchmark_index(
  data,
  carry_cols = NULL,
  id_col = "entity_id",
  timestamp_col = "timestamp",
  value_col = "y",
  temp_col = "temp",
  resolution_minutes = NULL,
  complete_days_only = TRUE,
  drop_na_value = TRUE,
  tz = "UTC"
)

Arguments

data

Normalized panel data, typically returned by one of the ⁠elcf4r_read_*()⁠ adapters.

carry_cols

Optional character vector of additional day-level columns to propagate into the benchmark index. If NULL, all non-core columns are carried.

id_col

Name of the entity identifier column.

timestamp_col

Name of the timestamp column.

value_col

Name of the load column.

temp_col

Name of the temperature column.

resolution_minutes

Sampling resolution in minutes. If NULL, it is inferred from the data.

complete_days_only

Passed to elcf4r_build_daily_segments().

drop_na_value

Passed to elcf4r_build_daily_segments().

tz

Time zone used to derive dates and within-day positions.

Value

A day-level data frame suitable for elcf4r_benchmark().

Examples

idx <- elcf4r_build_benchmark_index(
  elcf4r_iflex_example,
  carry_cols = c("dataset", "participation_phase", "price_signal")
)
head(idx)

Build daily load-curve segments from a normalized panel

Description

Convert a long-format load table into one row per entity-day and one column per within-day time index. This is the matrix representation required by functional load-curve models and rolling benchmark scripts.

Usage

elcf4r_build_daily_segments(
  data,
  id_col = "entity_id",
  timestamp_col = "timestamp",
  value_col = "y",
  temp_col = "temp",
  carry_cols = NULL,
  expected_points_per_day = NULL,
  resolution_minutes = NULL,
  complete_days_only = TRUE,
  drop_na_value = TRUE,
  tz = "UTC"
)

Arguments

data

Data frame containing at least entity id, timestamp and load.

id_col

Name of the entity identifier column.

timestamp_col

Name of the timestamp column.

value_col

Name of the load column.

temp_col

Optional name of a temperature column used to derive day summaries.

carry_cols

Optional day-level columns to propagate into the returned covariate table. Their first non-missing value within each day is kept.

expected_points_per_day

Expected number of samples per day. If NULL, it is derived from resolution_minutes.

resolution_minutes

Sampling resolution in minutes. If NULL, it is inferred from timestamps or from a resolution_minutes column. Fractional minute values are allowed.

complete_days_only

If TRUE, incomplete or duplicated days are dropped from the output.

drop_na_value

If TRUE, days with missing load values are dropped.

tz

Time zone used to derive dates and within-day positions.

Value

A list with components segments, covariates, resolution_minutes and points_per_day.

Examples

id1 <- subset(
  elcf4r_iflex_example,
  entity_id == unique(elcf4r_iflex_example$entity_id)[1]
)
daily <- elcf4r_build_daily_segments(id1, carry_cols = "participation_phase")
dim(daily$segments)
names(daily$covariates)

Derive deterministic KWF calendar groups

Description

Build the deterministic day groups used by the residential KWF workflow: weekdays, pre_holiday, and holiday.

Usage

elcf4r_calendar_groups(dates, holidays = NULL)

Arguments

dates

Vector coercible to Date.

holidays

Optional vector of holiday dates. If supplied, holiday dates are labelled "holiday" and the dates immediately before them are labelled "pre_holiday".

Value

An ordered factor with levels monday, tuesday, wednesday, thursday, friday, saturday, sunday, pre_holiday, holiday.

Examples

elcf4r_calendar_groups(
  as.Date(c("2024-12-24", "2024-12-25", "2024-12-26")),
  holidays = as.Date("2024-12-25")
)

Classify thermosensitivity from daily load data

Description

Estimate thermosensitivity using the residential rule based on the ratio between mean winter load and mean summer load.

Usage

elcf4r_classify_thermosensitivity(
  data,
  id_col = "entity_id",
  date_col = "date",
  value_col = "y",
  threshold = 1.5,
  winter_months = c(12L, 1L, 2L),
  summer_months = c(6L, 7L, 8L)
)

Arguments

data

Data frame containing at least an identifier, a date and a load column. Long-format panels are accepted and are aggregated to mean daily load before classification.

id_col

Name of the entity identifier column.

date_col

Name of the date column.

value_col

Name of the load column.

threshold

Ratio threshold above which the series is classified as thermosensitive. Defaults to 1.5.

winter_months

Integer vector of winter months.

summer_months

Integer vector of summer months.

Value

A data frame with one row per entity and columns winter_mean, summer_mean, ratio, thermosensitive, and status.

Examples

example_ts <- data.frame(
  entity_id = rep("home_1", 4),
  date = as.Date(c("2024-01-10", "2024-01-11", "2024-07-10", "2024-07-11")),
  y = c(12, 11, 6, 5)
)
elcf4r_classify_thermosensitivity(example_ts)

Download the ELMAS dataset from figshare

Description

This function downloads the original ELMAS archive from its public figshare URL and unpacks it to a local directory.

Usage

elcf4r_download_elmas(dest_dir)

Arguments

dest_dir

Directory where the files should be unpacked.

Value

A character vector with the paths of the extracted files.

Download selected GX dataset components

Description

Download selected assets from the official GX figshare dataset record. The helper only uses the dataset record itself and does not rely on the authors' code repository.

Usage

elcf4r_download_gx(dest_dir, components = "shapefile", overwrite = FALSE)

Arguments

dest_dir

Directory where the downloaded files should be stored.

components

Character vector of GX components to fetch. Supported values are "shapefile" and "database".

overwrite

Logical; if TRUE, existing local files are replaced.

Value

A character vector with the downloaded local file paths. Zip assets are extracted into dest_dir and the extracted paths are returned.

Download selected IDEAL dataset components

Description

Download selected assets from the IDEAL Household Energy Dataset record on Edinburgh DataShare. The helper is docs-first: it always retrieves the licence/readme files and documentation.zip, while heavy raw-data archives must be requested explicitly through components.

Usage

elcf4r_download_ideal(
  dest_dir,
  components = "documentation",
  overwrite = FALSE
)

Arguments

dest_dir

Directory where the downloaded files should be stored.

components

Character vector of IDEAL components to fetch. Supported values are "documentation", "metadata_and_surveys", "coding", "auxiliary", "household_sensors" and "room_and_appliance_sensors".

overwrite

Logical; if TRUE, existing local files are replaced.

Value

A character vector with the downloaded local file paths.

Download one or more StoreNet household files from figshare

Description

Download one or more StoreNet household files such as H6_W.csv into a local directory. The helper uses the figshare article API to resolve the actual file download URL when household-level article IDs are available. Otherwise it falls back to the public StoreNet archive and extracts the requested household files into dest_dir.

Usage

elcf4r_download_storenet(
  dest_dir,
  ids = "H6_W",
  article_ids = NULL,
  overwrite = FALSE,
  archive_url = "https://figshare.com/ndownloader/files/45123456"
)

Arguments

dest_dir

Directory where the downloaded files should be stored.

ids

Character vector of StoreNet household identifiers, for example "H6_W". Use NULL to extract every ⁠H*_W.csv⁠ file from the archive.

article_ids

Optional named integer vector that maps each requested household identifier to a figshare article ID. When NULL, the built-in mapping is used.

overwrite

Logical; if TRUE, existing local files are replaced.

archive_url

Optional figshare archive download URL used when a requested identifier is not present in the article-ID mapping.

Details

The default mapping currently covers the H6_W household file used by the package examples. Additional households can be downloaded either by providing a named article_ids vector or by relying on the public archive fallback.

Value

A character vector with the downloaded local file paths.

Toy subset of ELMAS hourly cluster profiles

Description

A compact subset of the public ELMAS dataset containing hourly load profiles for 3 commercial or industrial load clusters over 70 days. The object is intended for lightweight examples and tests that demonstrate time-series or segment-based workflows without shipping the full source archive.

Format

A tibble with 5,040 rows and 3 variables:

time

Hourly timestamp.

cluster_id

Cluster identifier, one of 3 retained ELMAS clusters.

load_mwh

Cluster load in MWh.

Source

Public ELMAS dataset, reduced with package data-raw scripts for examples and tests.

Fit a GAM model for load curves

Description

Fit a GAM model for load curves

Usage

elcf4r_fit_gam(data, use_temperature = FALSE)

Arguments

data

Data frame with columns y (load), time_index (numeric or factor for within day position), dow, month, optional temp and other covariates.

use_temperature

Logical. If TRUE, include temperature as smooth effect and interactions.

Value

An object of class elcf4r_model with method = "gam".

Examples

id1 <- subset(
  elcf4r_iflex_example,
  entity_id == unique(elcf4r_iflex_example$entity_id)[1]
)
train_data <- subset(id1, date < sort(unique(id1$date))[11])
test_data <- subset(id1, date == sort(unique(id1$date))[11])
fit <- elcf4r_fit_gam(train_data[, c("y", "time_index", "dow", "month", "temp")], TRUE)
pred <- predict(fit, newdata = test_data[, c("y", "time_index", "dow", "month", "temp")])
length(pred)

Fit a Kernel Wavelet Functional model for daily load curves

Description

Fit a day-ahead Kernel Wavelet Functional (KWF) model on ordered daily load curves. The implementation computes wavelet-detail distances on the historical context days, applies Gaussian kernel weights, restricts those weights to matching calendar groups when available, and can apply the approximation/detail correction used for mean-level non-stationarity.

Usage

elcf4r_fit_kwf(
  segments,
  covariates = NULL,
  target_covariates = NULL,
  use_temperature = FALSE,
  wavelet = "la12",
  bandwidth = NULL,
  use_mean_correction = TRUE,
  group_col = NULL,
  holidays = NULL,
  weights = NULL,
  recency_decay = NULL,
  temperature_bandwidth = NULL
)

Arguments

segments

Matrix or data frame of past daily load curves (rows are days, columns are within-day time points) in chronological order.

covariates

Optional data frame with one row per training segment. When present, the function looks for deterministic grouping information in context_group, kwf_group, calendar_group, or the column named by group_col. If no explicit group column is present, groups are derived from date and holidays, or from dow as a fallback.

target_covariates

Optional one-row data frame describing the day to forecast. When it contains date, the previous day is used as the context day for calendar grouping, which matches the residential KWF protocol for pre-holiday handling.

use_temperature

Deprecated and ignored. Kept for backward compatibility with earlier package examples.

wavelet

Wavelet filter name passed to wavelets::dwt(). Defaults to "la12", the least-asymmetric filter. If the series is too short for the requested filter, the function falls back to "haar".

bandwidth

Optional positive bandwidth for the Gaussian kernel on wavelet distances. If NULL, it is inferred from the distances to the last observed segment.

use_mean_correction

Logical; if TRUE, apply the approximation/detail correction used for mean-level non-stationarity.

group_col

Optional column name containing precomputed KWF groups in covariates and target_covariates.

holidays

Optional vector of holiday dates used by elcf4r_calendar_groups() when deterministic groups are derived from date.

weights

Optional numeric prior weights of length nrow(segments). Only the first nrow(segments) - 1 values are used in the historical pairing step.

recency_decay

Optional non-negative recency coefficient applied as an exponential prior on the historical context days.

temperature_bandwidth

Deprecated and ignored. Kept only for backward compatibility with older examples.

Value

An object of class elcf4r_model with method = "kwf".

Examples

id1 <- subset(
  elcf4r_iflex_example,
  entity_id == unique(elcf4r_iflex_example$entity_id)[1]
)
daily <- elcf4r_build_daily_segments(id1, carry_cols = "participation_phase")
fit <- elcf4r_fit_kwf(
  segments = daily$segments[1:10, ],
  covariates = daily$covariates[1:10, ],
  target_covariates = daily$covariates[11, , drop = FALSE]
)
length(predict(fit))

Fit a clustered KWF model for daily load curves

Description

Cluster dyadically resampled daily curves in a wavelet-energy feature space and use the resulting cluster labels as the grouping structure inside the KWF forecast.

Usage

elcf4r_fit_kwf_clustered(
  segments,
  covariates = NULL,
  target_covariates = NULL,
  wavelet = "la12",
  bandwidth = NULL,
  use_mean_correction = TRUE,
  max_clusters = 10L,
  nstart = 30L,
  cluster_seed = NULL,
  weights = NULL,
  recency_decay = NULL,
  clustering = NULL
)

Arguments

segments

Matrix or data frame of past daily load curves in chronological order.

covariates

Optional data frame with one row per segment.

target_covariates

Optional one-row data frame for the target day.

wavelet

Wavelet filter name passed to wavelets::dwt(). Defaults to "la12".

bandwidth

Optional positive bandwidth for the Gaussian kernel in the underlying KWF fit.

use_mean_correction

Logical; if TRUE, apply the approximation/detail correction in the underlying KWF fit.

max_clusters

Maximum number of candidate clusters considered by the Sugar jump heuristic.

nstart

Number of random starts for kmeans.

cluster_seed

Deprecated and ignored. Clustered KWF now uses deterministic non-random starts.

weights

Optional prior weights passed through to the base KWF fit.

recency_decay

Optional recency prior passed through to the base KWF fit.

clustering

Optional elcf4r_kwf_clusters object. When supplied, the stored clustering model is reused instead of being refit.

Value

An object of class elcf4r_model with method = "kwf_clustered".

Fit an LSTM model for daily load curves

Description

The LSTM implementation uses one or more previous daily curves to predict the next daily curve. When use_temperature = TRUE and temp_mean is available in covariates, the daily mean temperature is added as a second input feature repeated across the within-day time steps.

Usage

elcf4r_fit_lstm(
  segments,
  covariates = NULL,
  use_temperature = FALSE,
  lookback_days = 1L,
  units = 16L,
  epochs = 10L,
  batch_size = 8L,
  validation_split = 0,
  seed = NULL,
  verbose = 0L
)

Arguments

segments

Matrix or data frame of past daily load curves (rows are days, columns are time points).

covariates

Optional data frame with one row per training day.

use_temperature

Logical. If TRUE, use temp_mean from covariates as an additional input feature when available.

lookback_days

Number of past daily curves used as one training input.

units

Number of hidden units in the LSTM layer.

epochs

Number of training epochs.

batch_size

Batch size used in keras3::fit().

validation_split

Validation split passed to keras3::fit().

seed

Optional integer seed passed to TensorFlow. When NULL, the current backend RNG state is used.

verbose

Verbosity level passed to keras3::fit() and predict().

Value

An object of class elcf4r_model with method = "lstm".

Examples

if (interactive() &&
    requireNamespace("reticulate", quietly = TRUE) &&
    reticulate::virtualenv_exists("r-tensorflow")) {
  elcf4r_use_tensorflow_env(virtualenv = "r-tensorflow")
  if (isTRUE(getFromNamespace(".elcf4r_lstm_backend_available", "elcf4R")())) {
    id1 <- subset(
      elcf4r_iflex_example,
      entity_id == unique(elcf4r_iflex_example$entity_id)[1]
    )
    daily <- elcf4r_build_daily_segments(id1)
    fit <- elcf4r_fit_lstm(
      segments = daily$segments[1:10, ],
      covariates = daily$covariates[1:10, ],
      use_temperature = TRUE,
      epochs = 1,
      units = 4,
      batch_size = 2,
      verbose = 0
    )
    length(predict(fit))
  }
}

Fit a MARS model for load curves

Description

Fit a MARS model for load curves

Usage

elcf4r_fit_mars(data, use_temperature = FALSE)

Arguments

data

Data frame with columns y (load), time_index (numeric or factor for within day position), dow, month, optional temp and other covariates.

use_temperature

Logical. If TRUE, include temperature as smooth effect and interactions.

Value

An elcf4r_model object with method = "mars".

Examples

id1 <- subset(
  elcf4r_iflex_example,
  entity_id == unique(elcf4r_iflex_example$entity_id)[1]
)
train_data <- subset(id1, date < sort(unique(id1$date))[11])
test_data <- subset(id1, date == sort(unique(id1$date))[11])
fit <- elcf4r_fit_mars(train_data[, c("y", "time_index", "dow", "month", "temp")], TRUE)
pred <- predict(fit, newdata = test_data[, c("y", "time_index", "dow", "month", "temp")])
length(pred)

iFlex benchmark index of complete participant-days

Description

A compact index of complete days derived from the public iFlex hourly panel. Each row represents one participant-day with enough metadata to define deterministic benchmark cohorts without shipping the full raw panel.

Format

A data frame with 563,150 rows and 11 variables:

day_key

Unique key built as entity_id__date.

entity_id

Participant identifier.

date

Calendar date.

dow

Day of week.

month

Month as a two-digit factor.

temp_mean

Mean daily outdoor temperature in degrees Celsius.

temp_min

Minimum daily outdoor temperature in degrees Celsius.

temp_max

Maximum daily outdoor temperature in degrees Celsius.

participation_phase

Experiment phase from the source dataset.

price_signal

Experimental price-signal label, when available.

n_points

Number of hourly samples retained for the day.

Source

Public iFlex raw file data_hourly.csv, reduced with data-raw/elcf4r_iflex_subsets.R.

iFlex benchmark results for shipped forecasting methods

Description

Saved benchmark results for a deterministic rolling-origin evaluation on a subset of the iFlex data. The shipped results use a fixed participant cohort, a 28-day training window and multiple one-day rolling test forecasts per participant. The current shipped benchmark includes the operational gam, mars, kwf, kwf_clustered and lstm wrappers.

Format

A data frame with 20 variables:

benchmark_name

Identifier of the benchmark design.

dataset

Dataset label, always "iflex".

entity_id

Participant identifier.

method

Forecasting method: gam, mars, kwf, kwf_clustered or lstm.

test_date

Date of the forecast target day.

train_start

First day in the training window.

train_end

Last day in the training window.

train_days

Number of training days.

test_points

Number of hourly points in the target day.

use_temperature

Logical flag for temperature-aware fitting.

thermosensitive

Thermosensitivity flag when seasonal coverage is sufficient, otherwise NA.

thermosensitivity_status

Status of the winter/summer ratio classification step.

thermosensitivity_ratio

Estimated winter/summer mean-load ratio when available.

fit_seconds

Elapsed fit-and-predict time in seconds.

status

Benchmark execution status.

error_message

Error message when a fit failed.

nmae

Normalized mean absolute error.

nrmse

Normalized root mean squared error.

smape

Symmetric mean absolute percentage error.

mase

Mean absolute scaled error.

Source

Derived from elcf4r_iflex_benchmark_index and the public iFlex raw file with data-raw/elcf4r_iflex_benchmark_results.R.

iFlex example panel for package examples

Description

A compact hourly electricity-demand panel extracted from the public iFlex dataset. The object contains 14 complete days for each of 3 participants and is intended for examples, tests and lightweight vignettes.

Format

A data frame with 1,008 rows and 16 variables:

dataset

Dataset label, always "iflex".

entity_id

Participant identifier.

timestamp

Hourly UTC timestamp.

date

Calendar date of the observation.

time_index

Within-day hourly index from 1 to 24.

y

Hourly electricity demand in kWh.

temp

Outdoor temperature in degrees Celsius.

dow

Day of week.

month

Month as a two-digit factor.

resolution_minutes

Sampling resolution in minutes.

participation_phase

Experiment phase from the source dataset.

price_signal

Experimental price-signal label, when available.

price_nok_kwh

Experimental electricity price in NOK per kWh.

temp24

Lagged 24-hour temperature feature from the source file.

temp48

Lagged 48-hour temperature feature from the source file.

temp72

Lagged 72-hour temperature feature from the source file.

Source

Public iFlex raw file data_hourly.csv, reduced with data-raw/elcf4r_iflex_subsets.R.

Cluster daily segments for clustered KWF

Description

Build a reusable clustering model for daily load-curve segments in the wavelet-energy feature space used by the clustered KWF workflow.

Usage

elcf4r_kwf_cluster_days(
  segments,
  wavelet = "la12",
  max_clusters = 10L,
  nstart = 30L,
  cluster_seed = NULL
)

Arguments

segments

Matrix or data frame of daily load curves in chronological order.

wavelet

Wavelet filter name passed to wavelets::dwt(). Defaults to "la12".

max_clusters

Maximum number of candidate clusters considered by the Sugar jump heuristic.

nstart

Number of random starts for kmeans.

cluster_seed

Deprecated and ignored. Clustering now uses deterministic non-random starts.

Value

An object of class elcf4r_kwf_clusters.

Low Carbon London benchmark results for shipped forecasting methods

Description

Saved rolling-origin benchmark results for the shipped methods on a fixed Low Carbon London cohort of households. The benchmark is based on 30-minute load curves and reports NMAE, NRMSE, sMAPE and MASE.

Format

A data frame with the same benchmark-result schema as elcf4r_iflex_benchmark_results.

Source

Derived from the local LCL raw file with data-raw/elcf4r_lcl_artifacts.R.

Low Carbon London example panel for package examples

Description

A compact normalized panel extracted from a small group of households in the Low Carbon London dataset. The object contains complete 30-minute days and is intended for examples and lightweight benchmarking workflows.

Format

A data frame with normalized panel fields:

dataset

Dataset label, always "lcl".

entity_id

Low Carbon London household identifier.

timestamp,date,time_index,y,temp,dow,month,resolution_minutes

Common normalized panel fields.

Source

Public LCL raw file LCL_2013.csv, reduced with data-raw/elcf4r_lcl_artifacts.R.

Forecast accuracy metrics for load curves

Description

Compute NMAE, NRMSE, sMAPE and MASE between observed and predicted load curves, as in the posters.

Usage

elcf4r_metrics(truth, pred, seasonal_period = NULL, naive_pred = NULL)

Arguments

truth

Numeric vector or matrix of observed values.

pred

Numeric vector or matrix of predicted values, same shape.

seasonal_period

Seasonal period for the naive benchmark in the MASE denominator (for daily curves with half hourly sampling, a value of 48 is appropriate).

naive_pred

Optional numeric vector or matrix of naive benchmark predictions with the same shape as truth. When supplied, MASE is computed against this explicit naive forecast instead of inferring the denominator from seasonal_period within truth.

Value

A named list with elements nmae, nrmse, smape, mase.

Normalize a load panel to the elcf4R schema

Description

Convert a raw long-format load table into a normalized panel that uses the column names expected by the package examples and model wrappers.

Usage

elcf4r_normalize_panel(
  data,
  id_col,
  timestamp_col,
  load_col,
  temp_col = NULL,
  dataset = NA_character_,
  resolution_minutes = NULL,
  tz = "UTC",
  keep_cols = NULL
)

Arguments

data

Data frame containing at least an entity identifier, a time stamp and a load column.

id_col

Name of the entity identifier column.

timestamp_col

Name of the timestamp column.

load_col

Name of the load column.

temp_col

Optional name of the temperature column.

dataset

Short dataset label stored in the normalized output.

resolution_minutes

Sampling resolution in minutes. If NULL, it is inferred from the timestamps. Fractional minute values are allowed for high-frequency data.

tz

Time zone used to parse timestamps.

keep_cols

Optional character vector of extra source columns to keep.

Value

A data frame with normalized columns dataset, entity_id, timestamp, date, time_index, y, temp, dow, month and resolution_minutes, plus any requested keep_cols.

Read and normalize the GX residential transformer-level scaffold

Description

Read the GX dataset from either the official SQLite database or a flat export and return a normalized long-format panel. GX is treated as a transformer/community-level dataset rather than an individual-household dataset.

Usage

elcf4r_read_gx(
  path = "data-raw",
  ids = NULL,
  start = NULL,
  end = NULL,
  tz = "Asia/Shanghai",
  n_max = NULL,
  drop_na_load = TRUE
)

Arguments

path

Path to a GX SQLite database, a flat export file, or a directory containing one of them.

ids

Optional vector of GX community/profile identifiers to keep.

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps. Defaults to "Asia/Shanghai".

n_max

Optional maximum number of rows to read.

drop_na_load

Logical; if TRUE, rows with missing load values are dropped.

Value

A normalized data frame with GX transformer-level data.

Read and normalize the IDEAL hourly aggregate-electricity scaffold

Description

Read a direct IDEAL hourly aggregate-electricity file or search an extracted auxiliarydata.zip directory for a matching hourly summary file, then return a normalized long-format panel.

Usage

elcf4r_read_ideal(
  path = "data-raw",
  ids = NULL,
  start = NULL,
  end = NULL,
  tz = "Europe/London",
  n_max = NULL,
  source = "auxiliary_hourly",
  drop_na_load = TRUE
)

Arguments

path

Path to an IDEAL hourly summary file or to an extracted IDEAL auxiliary-data directory.

ids

Optional vector of IDEAL household identifiers to keep.

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps. Defaults to "Europe/London".

n_max

Optional maximum number of rows to read.

source

IDEAL source flavor. Currently only "auxiliary_hourly" is supported.

drop_na_load

Logical; if TRUE, rows with missing load values are dropped.

Value

A normalized data frame with IDEAL household data.

Read and normalize the iFlex hourly dataset

Description

Read the iFlex hourly consumption table and return a normalized long-format panel ready for feature engineering, segmentation and benchmarking.

Usage

elcf4r_read_iflex(
  path = file.path("data-raw", "iFlex"),
  ids = NULL,
  start = NULL,
  end = NULL,
  tz = "UTC",
  n_max = NULL
)

Arguments

path

Path to data_hourly.csv or to the directory that contains it.

ids

Optional vector of participant identifiers to keep.

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps. Defaults to "UTC" because the iFlex timestamps are stored with a trailing Z.

n_max

Optional maximum number of rows to read. Intended for quick prototyping on a small subset of the raw file.

Value

A normalized data frame with load, temperature and calendar fields. The output also keeps participation_phase, price_signal, price_nok_kwh, temp24, temp48 and temp72.

Read and normalize the Low Carbon London dataset

Description

Read a wide Low Carbon London (LCL) smart-meter file and reshape it into a normalized long-format panel with one row per household timestamp.

Usage

elcf4r_read_lcl(
  path = file.path("data-raw", "LCL_2013.csv"),
  ids = NULL,
  start = NULL,
  end = NULL,
  tz = "UTC",
  n_max = NULL,
  drop_na_load = TRUE
)

Arguments

path

Path to an LCL CSV file or to a directory containing one.

ids

Optional vector of LCL household identifiers to keep, for example "MAC000002".

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps.

n_max

Optional maximum number of timestamp rows to read.

drop_na_load

Logical; if TRUE, rows with missing load values are dropped after reshaping.

Value

A normalized data frame with LCL household data.

Read and normalize the REFIT cleaned household dataset

Description

Read one or more CLEAN_House*.csv files from the REFIT dataset, optionally select appliance channels, resample them to a regular time grid, and return a normalized long-format panel.

Usage

elcf4r_read_refit(
  path = "data-raw",
  house_ids = NULL,
  channels = "Aggregate",
  start = NULL,
  end = NULL,
  tz = "UTC",
  resolution_minutes = 1L,
  agg_fun = c("mean", "sum", "last"),
  n_max = NULL,
  drop_na_load = TRUE
)

Arguments

path

Path to a REFIT file or to a directory containing CLEAN_House*.csv files.

house_ids

Optional vector of house identifiers to keep. These are matched against file stems such as "CLEAN_House1".

channels

Character vector of load channels to extract. Defaults to "Aggregate".

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps.

resolution_minutes

Target regular resolution in minutes for the normalized output. Defaults to 1.

agg_fun

Aggregation used when resampling to the target grid. One of "mean", "sum" or "last".

n_max

Optional maximum number of raw rows to read per file.

drop_na_load

Logical; if TRUE, rows with missing load values are dropped after resampling.

Value

A normalized data frame with REFIT household data.

Read and normalize the StoreNet household dataset

Description

Read one or more StoreNet-style household CSV files such as H6_W.csv, derive the household identifier from the file name, and return a normalized long-format panel.

Usage

elcf4r_read_storenet(
  path = file.path("data-raw", "H6_W.csv"),
  ids = NULL,
  start = NULL,
  end = NULL,
  tz = "UTC",
  n_max = NULL,
  load_col = "Consumption(W)",
  keep_cols = c("Discharge(W)", "Charge(W)", "Production(W)", "State of Charge(%)")
)

Arguments

path

Path to a StoreNet CSV file or to a directory containing files named like H6_W.csv.

ids

Optional vector of household identifiers to keep. Identifiers are matched against the file stem, for example "H6_W".

start

Optional inclusive lower time bound.

end

Optional inclusive upper time bound.

tz

Time zone used to parse timestamps.

n_max

Optional maximum number of rows to read per file.

load_col

Name of the load column to normalize. Defaults to "Consumption(W)".

keep_cols

Optional extra source columns to keep. Defaults to the main battery and production fields when present.

Value

A normalized data frame with StoreNet household data.

REFIT benchmark results for shipped forecasting methods

Description

Saved rolling-origin benchmark results for the shipped methods on the REFIT example cohort after resampling to 15-minute resolution. The benchmark reports NMAE, NRMSE, sMAPE and MASE.

Format

A data frame with the same benchmark-result schema as elcf4r_iflex_benchmark_results.

Source

Derived from the local REFIT raw files with data-raw/elcf4r_refit_artifacts.R.

REFIT example panel for package examples

Description

A compact normalized panel extracted from the REFIT cleaned dataset after resampling to 15-minute resolution. The object contains complete days for one house and is intended for examples and lightweight benchmarking workflows.

Format

A data frame with normalized panel columns plus REFIT-specific fields:

dataset

Dataset label, always "refit".

entity_id

Entity identifier, here the aggregate household channel.

timestamp,date,time_index,y,temp,dow,month,resolution_minutes

Common normalized panel fields.

house_id

REFIT house identifier derived from the file name.

channel

Load channel name, for example "Aggregate".

unix

Minimum Unix timestamp within the resampling bucket.

issues

Maximum issues flag within the resampling bucket.

Source

Public REFIT cleaned raw files, reduced with data-raw/elcf4r_refit_artifacts.R.

StoreNet benchmark results for shipped forecasting methods

Description

Saved rolling-origin benchmark results for the shipped methods on the local StoreNet household example. The benchmark is derived from complete 1-minute household days and reports NMAE, NRMSE, sMAPE and MASE for every shipped row. The clustered KWF variant is only included when the shipped StoreNet cohort is classified as thermosensitive.

Format

A data frame with the same benchmark-result schema as elcf4r_iflex_benchmark_results.

Source

Derived from the local StoreNet raw file with data-raw/elcf4r_storenet_artifacts.R.

StoreNet example panel for package examples

Description

A compact normalized panel extracted from the local StoreNet household file H6_W.csv. The object contains a small set of complete 1-minute household days and is intended for examples and lightweight benchmarking workflows.

Format

A data frame with normalized panel columns plus StoreNet-specific fields:

dataset

Dataset label, always "storenet".

entity_id

Household identifier derived from the file name.

timestamp,date,time_index,y,temp,dow,month,resolution_minutes

Common normalized panel fields.

discharge_w,charge_w,production_w

Battery and production fields from the source file in watts.

state_of_charge_pct

Battery state of charge in percent.

source_file

Source CSV file name.

Source

Public StoreNet raw file H6_W.csv, reduced with data-raw/elcf4r_storenet_artifacts.R.

Select the Python environment used for TensorFlow-backed LSTM fits

Description

This helper provides an explicit, user-invoked way to bind the Python environment used by reticulate before calling elcf4r_fit_lstm().

Usage

elcf4r_use_tensorflow_env(python = NULL, virtualenv = NULL, required = TRUE)

Arguments

python

Optional path to a Python interpreter passed to reticulate::use_python().

virtualenv

Optional virtualenv name or path passed to reticulate::use_virtualenv().

required

Logical passed to the corresponding reticulate selector.

Value

Invisibly returns the selected Python interpreter path when it can be determined.

Examples

if (interactive() &&
    requireNamespace("reticulate", quietly = TRUE) &&
    reticulate::virtualenv_exists("r-tensorflow")) {
  elcf4r_use_tensorflow_env(virtualenv = "r-tensorflow")
}

Assign new segments to a fitted KWF clustering model

Description

Assign new segments to a fitted KWF clustering model

Usage

## S3 method for class 'elcf4r_kwf_clusters'
predict(object, segments, ...)

Arguments

object

An elcf4r_kwf_clusters object returned by elcf4r_kwf_cluster_days().

segments

Matrix or data frame of new daily segments.

...

Unused, present for method compatibility.

Value

A character vector of cluster labels.

Predict from an elcf4r_model

Description

Predict from an elcf4r_model

Usage

## S3 method for class 'elcf4r_model'
predict(object, newdata = NULL, ...)

Arguments

object

An elcf4r_model created by one of the package fit functions.

newdata

Optional new data for methods that need it. For gam and mars, this should be a long-format data frame with the same predictor columns used for fitting. For lstm, newdata may be a matrix or data frame of recent daily segments, or a list with elements segments and optional covariates.

...

Unused, present for method compatibility.

Value

Numeric predictions. For KWF and LSTM this is a forecast daily curve.