Structured Dataset of Daily Electricity Demand, Generation, Load Shedding, and Supply Constraints in Bangladesh (2019–2024).

This dataset presents a structured, multi-version compilation of daily electricity system records for Bangladesh, spanning the period from November 21, 2019, to December 30, 2024. It was developed by programmatically extracting 1,867 daily PDF reports from the publicly accessible archive of the Bangladesh Power Development Board (BPDB): https://misc.bpdb.gov.bd/daily-generation-archive.

The dataset is organized into five progressive versions, each contained in a separate folder. These versions reflect successive enhancements—ranging from initial raw extraction to final preprocessing suitable for machine learning workflows. The dataset supports granular investigation at both national and divisional levels.

Version 1 comprises raw records parsed directly from the BPDB reports. It contains unprocessed inconsistencies, missing entries, and formatting noise. This version is preserved to support traceability.

Version 2 offers a cleaned and verified dataset where duplicate entries were removed, and missing values were recovered using source files. Daily national and divisional records were reconciled and validated.

Version 3 adds temporal and calendar-based features. National and religious holidays were annotated manually and categorized by type. These features are intended to help capture behavioral variations in electricity consumption related to festive or reduced-activity periods.

Version 4 applies robust data curation techniques, including forward and backward interpolation for missing values, logical imputation for invalid demand records, and column-wise consistency checks. Two temporal attributes—year and month—were also added to facilitate seasonal analysis.

Version 5 is optimized for time series modeling. It incorporates smoothed corrections for outlier values using centered rolling medians and scales key numerical features for modeling readiness. The tabular structure remains unchanged from earlier versions, ensuring continuity for comparative analysis.

Each version is supplied as a single .xlsx file with flat headers. A README.txt file explains the processing logic and provides a full breakdown of the steps followed in dataset construction. Although the source code is not included, the process is thoroughly documented to enable reproducibility. A separate column_descriptions.xlsx file describes all variables in the final version in detail.

The dataset exhibits multiple seasonal and temporal trends that reflect operational rhythms of Bangladesh’s energy system. Distinct patterns emerge across weekdays, months, and holidays, and regional variation is evident across administrative divisions. These characteristics make the dataset well-suited for predictive modeling, policy analysis, and infrastructure planning under varying demand scenarios.