Comparative Study of Building Energy Use Prediction Based on Three Artificial Neural Network Algorithms

With the advent of the era of big data, artificial neural network (ANN) algorithms have been widely used in the field of building energy data analysis. In order to effectively use ANN algorithms to predict building energy consumption, the data-driven building energy consumption prediction with three typical ANNs: Backpropagation neural network (BPNN), generalized regression neural network (GRNN), and fuzzy neural network (FNN) were studied. The simulated data of an office building model setup by EnergyPlus is presented for a case study. The BPNN algorithm with different hidden layer numbers, GRNN algorithm with different scatter constants, and FNN algorithm with different evolution times were investigated, and the optimal parameters of each neural network algorithm for building energy consumption prediction were finally obtained. The results show that the MSEs of all ANN-based models are almost the same with very small values. But the operation time is very different, which of GRNN has the smallest value. So, the GRNN is highly recommended for building energy consumption prediction due to its both good prediction accuracy and short operation time. This study helps to guide the selections of ANNs and the determinations of related parameters of their algorithms in engineering application.