Automatic Optimization-Based Calibration Using Genetic Algorithms: A Case Study of a School Energy Model

Abstract

Simulation software allows detailed modeling of building behavior. Nevertheless, they require a large number of input parameters, which can be affected by uncertainty leading to differences between actual and simulated results and making a calibration process required. The present study investigates an automatic optimization-based calibration built by coupling the simulation engine EnergyPlus and Python as simulation and optimization manager. Given the numerical feature of the building energy model, the elitist genetic algorithm was selected to execute a scalarized multi-objective optimization aiming to identify the most effective combinations of seven uncertain input parameters, which affect the space heating consumption. The optimization problem is solved scalarizing two objective functions—normalized mean bias error (NMBE) and coefficient of variation of the RMSE (CV(RMSE))—into a utility function, to minimize the difference between simulated and monitored space heating consumption based on a five-month control period. The methodology was thus tested on a school building located in southern Italy. The model accuracy was evaluated using the ASHRAE Guideline 14–2014 metrics: NMBE and CV(RMSE). An NMBE of 0.11% and a CV(RMSE) of 13.9% were reached after the optimization process, allowing the model to be considered calibrated.