Kentsel bina enerji modellemesi: Zaman serisi analizi için dinamik ve veriye dayalı modellerin entegre edilmesi

This study addresses a critical gap in Urban Building Energy Modeling (UBEM) by introducing an innovative hybrid approach that combines dynamic and statistical models. Previous hybrid studies faced challenges in result validation, model calibration, and delivering predictions at high temporal resolution. To fill this gap, the study presents an integrated UBEM for time-series analysis. The methodology involves creating a dynamic model for 13 buildings on a university campus, with key parameters calibrated using Bayesian Optimization. This calibration reduces the simulation's annual Mean Absolute Percentage Error (MAPE) from 21.56\% to 4.60\% and the monthly MAPE from 22.14\% to 9.90\%. The calibrated dynamic model's hourly energy consumption data is then incorporated into a statistical model, a Long Short-Term Memory (LSTM) network utilized in time-series regression. The statistical model predicts the hourly energy consumption with an average R-squared value of 0.915. The proposed hybrid model facilitates the creation of synthetic hourly energy consumption data for urban building stocks. In this sense, the energy use patterns derived from the hourly consumption can be combined with the building characteristics to identify parameters that shape the building energy demand. The hybrid model can also optimize the building energy efficiency design by exploring various configurations in the building envelope and operational schedules and minimizing the energy consumption and the resultant environmental impact. When adapted to the urban scale, this hybrid model can provide valuable insights for urban planners in identifying high-demand areas and implementing energy-efficient interventions based on high-resolution temporal energy consumption data. Instead of testing energy and cost-efficiency scenarios using dynamic simulations, the proposed hybrid model can effectively monitor and control the hourly building energy use over time-series analysis once calibrated. This study underscores the potential of hybrid modeling in UBEM despite facing challenges, like complexities in generating the validation data from a limited number of metered energy consumption and computational constraints. In summary, this research introduces a robust hybrid UBEM and draws a roadmap for future research to comprehend urban building energy demand accurately. Future research includes reliability improvements for the validation and input data, efficient and precise dynamic modeling approaches, and utilizing thermal interactions between buildings within the statistical model.