Evaluating the impact of data quality on the accuracy of the predicted energy performance for a fixed building design using probabilistic energy performance simulations and uncertainty analysis

Probabilistic building performance simulations enables a method for evaluating how the data quality used to create input models affects the accuracy of the predicted energy performance. The method described in this paper explores an uncertainty analysis of a fixed building design and system by including the discrepancy between the declared property of a product, material, or behaviour quantified under specific conditions, and the actual performance in real-world use. The method was tested using a case study, a multi-family building, using two datasets based on different data quality to quantify the discrepancies. The models' outcome was validated against field measurements from 28 buildings built using the fixed building design. The main findings were that data quality significantly shifted the probability density curves and consequently impacted the predictions and accuracy of the predictive models, showing the importance of high-quality input models and a validation process based on a probability distribution. The study also indicated that higher quality data does not equal narrower distributions in the input models. In this case, the case study showed that, despite well-known building properties, narrowing the performance gap can only occur through larger variation in the input data models, resulting in a larger predicted energy performance interval. (c) 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Automated summary

Probabilistic building performance simulations are used to evaluate the impact of data quality on the accuracy of predicted energy performance. The study showed that data quality significantly affects the accuracy of predictive models, highlighting the importance of high-quality input models and validation based on probability distributions. Despite well-known building properties, narrowing the performance gap requires larger variation in input data models, resulting in a larger predicted energy performance interval.