A review of artificial intelligence based building energy use prediction:Contrasting the capabilities of single and ensemble prediction models
2. Current trends: AI-based building energy use prediction
A total number of 35 representative journal articles were identified to comprehend the current research status and trends of AI-based building energy use prediction
The selection criteria for narrowing recent work included the building types, prediction approaches, energy output types predicted, time scale of the prediction, and input data types used for prediction.
The current research trends of AI-based building energy use prediction based on the investigation results of each aspect are discussed in the following part of this section.
2.1. Building type
- the tested buildings may be classified into four categories,
e.g., commercial, residential, educational and research, and other building types.
- the AI-based prediction models were largely used for energy prediction of the educational and research, and commercial building types, i.e., 42% and 33% respectively
the reason is due to data availability and, potentially, easier access to the available data
2.2. Prediction method
- the single prediction method is widely used for AI-based , 91% of all AI-based predictions used one prediction algorithm, while the studies on ensemble prediction method for building energy use prediction are limited (9%).
- Because the research on ensemble prediction method for building energy use prediction is still at the initial stage because of its complexit
- results of these studies demonstrated the superiority of ensemble models over single prediction models
- In this paper, we classified the learning algorithms into four categories: regression, ANN, SVR, and all others.
- regression (26%), ANN (41%), SVR (12%), and all others (21%).
- ANN is the most widely used algorithm among these four categories
- Various ANNs including MLP [9], BPNN [14,25], FFNN [12], and RBFN [35], were used in these recent studies
- However, SVR has shown its superiority in terms of prediction accuracy in building energy use prediction compared with other learning algorithms
- In addition, many other algorithms such as ARMAX [17], CHAID [25], and CR [27,30] were used for building energy use prediction.