시간 단위의 M&V 베이스라인 구축을 위한머신러닝 알고리즘 기반 건물에너지 예측 모델의 성능 비교

As an alternative to existing simple regression monthly baseline method, we developed an hourly baseline model for M&V based on prediction models with machine learning techniques. This paper evaluated three data-driven energy models used to predict building electricity energy consumption: K-nearest neighbor (KNN) model, Random Forest (RF) model, and Artificial Neural Network (ANN) Model. As a result, CVRMSE is about 10% in all three models. In addition, it was confirmed that the ANN is superior to the KNN or RF in terms of the prediction accuracy of the energy consumption pattern in which the energy consumption is rapidly fluctuated with time.