Dataset for "The impact of Building Management System data pre-processing on Physics-based Digital Twin calibration and energy prediction accuracy"

This dataset supports the study of how data preprocessing affects the calibration performance of building-scale digital twins (DTs) using Building Management System (BMS) data.

The dataset consists of four main components:

(1) **Sample raw BMS data**
A one-year (2024) electricity consumption time series from a single anonymised meter (Meter_A), recorded at 15-minute intervals. The data is provided as a representative sample to illustrate the structure and characteristics of real BMS data, while avoiding disclosure of sensitive building-level information.

(2) **Processed data samples**
The corresponding preprocessed version of the raw data using the selected optimal pipeline (GESD-based anomaly detection, Transformer-based imputation, and no smoothing). This allows direct comparison between raw and processed signals.

(3) **Digital Twin calibration results**
Comprehensive results for 24 data processing combinations (DPCs), including both annual and monthly calibration outputs. The dataset reports key performance metrics (e.g., NMBE, CVRMSE, and dynamic consistency indicators), enabling systematic evaluation of preprocessing effects.

(4) **Model and experiment configuration records**
Parameters and settings for the Transformer-based imputation model, along with relevant experimental configurations to ensure reproducibility of the modelling pipeline.

The dataset is designed to facilitate replication of the study’s findings and to support further research on the interaction between data preprocessing and digital twin performance in building energy modelling.

Due to data confidentiality constraints, only a representative subset of raw BMS data is provided, and all identifying information has been removed.