Analysis of Virtual Smart Meter Data Sets for Designing Demand Response Models Based on Building Stock Energy Model

This study examined the possibility of developing a time-series-based electricity demand response model by applying a VSM data set model generated from 30,000 household on BSEM in Quebec, Canada, to the demand response of the domestic electricity market. The CBL method, which has been adopted as a conventional demand forecasting method, has limitations in sufficiently reflecting the nonlinearity of maximum power demand and demand response because it calculates based on the average of past power consumption data. Therefore, in this study, the structure of the VSM data set which generated BSEM and the TRNSYS-CREST simulation were analyzed, and the applicability of the demand response was evaluated through the verification of the LSTM, GRU, and CNN-LSTM models using the VSM data set. As the results , the LSTM and GRU models showed slightly higher error index than the CBL, while the CNN-LSTM structure effectively predicted short-term fluctuations through convolutional features combined with time dependency. This confirmed the possibility of more precise DR if the conventional CBL method and the AI prediction model were combined. In conclusion, this study confirmed the possibility of constructing a Korean VSM data set that reflects institutional factors such as BSEM-based energy prediction, TOU rate system, and DR, focusing on apartment complexes in Korea, and the possibility of follow-up research on simulation and empirical models based on RSM data.