Toward Digital Twin-Enabled Smart Buildings: An Evolutionary Neural Network Approach for Energy Prediction
TL;DR
An Evolutionary Field Optimization (EFO)-based multi-input multi-output artificial neural network (MIMO-ANN) model to simultaneously predict the heating load, cooling load, CO2 emissions, and lighting energy consumption of smart buildings is proposed.
Abstract
The increasing pace of urbanization and climate change necessitate a holistic assessment of building energy performance during the early design phase. This study proposes an Evolutionary Field Optimization (EFO)-based multi-input multi-output artificial neural network (MIMO-ANN) model to simultaneously predict the heating load, cooling load, CO2 emissions, and lighting energy consumption of smart buildings. The model’s dataset consists of 7963 observations generated via EnergyPlus building energy simulations of standardized TOKİ residential units constructed post-earthquake in Türkiye. No operational or physically measured building energy consumption data were used in the model development process. For the validation setting, the simulation-generated dataset was split into training (60%), validation (10%), and test (30%) subsets. The EFO algorithm was employed to automatically optimize the ANN architecture by dynamically determining the optimal number of hidden layers and neurons. The optimization process demonstrated strong global search capability and fast convergence, reducing the objective function by approximately 86% within 10 iterations. Experimental results on the test subset showed exceptional predictive accuracy for simulation data, with test R2 values ranging from 0.9996 to 0.9998 across all four outputs, indicating that the optimized network topology effectively avoided overfitting. While the model’s performance under real-world operational uncertainties and varying occupant behaviors remains to be fully investigated, the proposed EFO-ANN framework provides a computationally efficient and highly accurate analytical core for early-stage design. It serves as a strategic decision-support tool intended for architects and engineers designing post-disaster housing, public authorities forming national energy efficiency policies, and developers building predictive engines for digital twin-enabled smart building systems.