README.md

metadata

license: openrail
task_categories:
  - time-series-forecasting
size_categories:
  - 1K<n<10K

Download the Dataset:

from datasets import load_dataset

dataset = load_dataset("LeoTungAnh/electricity_hourly")

Dataset Card for Electricity Consumption

This dataset encompasses hourly electricity consumption in kilowatts (kW) across a span of three years (2012-2014), involving 370 individual clients in Portugal.

Preprocessing information:

• Grouped by hour (frequency: "1H").
• Applied Standardization as preprocessing technique ("Std").

Dataset information:

• Number of time series: 370
• Number of training samples: 26208
• Number of validation samples: 26256 (number_of_training_samples + 48)
• Number of testing samples: 26304 (number_of_validation_samples + 48)

Dataset format:

  Dataset({
  
      features: ['start', 'target', 'feat_static_cat', 'feat_dynamic_real', 'item_id'],
      
      num_rows: 370
      
  })

Data format for a sample:

• 'start': datetime.datetime

• 'target': list of a time series data

• 'feat_static_cat': time series index

• 'feat_dynamic_real': None

• 'item_id': name of time series

Data example:

{'start': datetime.datetime(2012, 1, 1, 1, 0),

 'target': [-0.19363673541224083, -0.08851588245610625, -0.19363673541224083, ... -0.5615597207587115,...],
 
 'feat_static_cat': [0],
 
 'feat_dynamic_real': None,
 
 'item_id': 'MT_001'
}

Usage:

• The dataset can be used by available Transformer, Autoformer, Informer of Huggingface.
• Other algorithms can extract data directly by making use of 'target' feature.