Multi-objective hyperparameter optimization of artificial neural network in emulating building energy simulation

Artificial Neural Networks (ANNs) play a significant role in emulating Building Energy Simulation (BES), forecasting building energy consumption, and optimizing energy retrofit measures. Determining the appropriate architecture for an ANN that can manage multiple predictions simultaneously is a complex task that requires extensive experimentation and validation to achieve optimal performance. To address this challenge, a novel methodology referred to as Multi-Objective Hyperparameter Optimization of ANN (MOHO-ANN) is introduced. This approach involves aligning ANN prediction results with data to achieve optimal performance by tuning the ANN’s hyperparameters. The methodology consists of calibrating the BES model, creating data using a model sampling step for ANN training , and formulating multi-objective optimization using hyperparameter tuning to obtain the set of optimal ANN architectures. Lastly, a Multi-Criteria Decision-Making (MCDM) step is employed to select the optimal ANN. This method is applied to retrofit an existing building by incorporating weather data, passive, and renewable retrofit measures. The ANN is used to predict hourly energy consumption, energy generation, and thermal comfort in the retrofit scenario. 1.75 million hourly data points have been used to train, validate, and test the ANN model. The results underscore the practicality of MOHO-ANN procedure in achieving predictions, demonstrating a coefficient of determination (R2) exceeding 0.98. Further research directions should consider employing the optimal architectural model to perform retrofit optimization by incorporating climate change.