Building energy model automated calibration using Pymoo

Abstract

This paper describes a two-stage calibration method, starting with detailed modeling and followed by optimization-based calibration. To put the proposed calibration method to the test, a real-life existing building was modeled using TRNSYS (Transient System Simulation Tool). The first stage approach emphasizes the significance of a detailed description of the building. In this regard, we compared the results of two different simulations: one that considers the dynamic behavior of infiltration and convective heat exchange, and the other that uses average values for the aforementioned quantities. The detailed model helped reduce the gap between simulation and measurement, where one of the simulated zones' mean absolute error (MAE) decreased from 1.11°C to 0.63°C. The calibration problem is viewed as an optimization problem in the second-stage approach, and the MAE between the simulated and measured values is regarded as an objective function to be minimized. In this work, we demonstrate the implementation of Pymoo (Multi-Objective Optimization in Python) as a calibration tool for building energy simulation software. The detailed energy simulation was fine-tuned using the genetic algorithm. The optimization results produced an MAE of around 0.3°C for both studied zones.