Evaluating the suitability of regression-based emulators of building performance in practice: a test suite

Abstract

Building Performance Simulation (BPS), a 2 useful tool to assess the operational perform-3 ance of buildings and systems, can often be 4 computationally expensive. The use of BPS is 5 cumbersome for problems where the speed of 6 response is important, e.g., real-time control, 7 uncertainty quantification, parametric explor-8 ation, or stock modelling. Emulators, such as 9 those based on regression, offer a faster substi-10 tute, but their reliability can be questionable. 11 This paper proposes seven tests to check if an 12 emulator is a suitable replacement for simula-13 tion in practice. The tests are categorized us-14 ing four criteria: accuracy, speed, generalisab-15 ility, and ease of use. The tests can be included 16 in the process of setting up an emulator-based 17 workflow. A use case is provided for emulat-18 ors based on linear and non-linear regression 19

Principal component analysis of day‐ahead electricity price forecasting in CAISO and its implications for highly integrated renewable energy markets

“…This is accomplished by identifying which variables contribute the most to the total variability of the original vector of random variables, denoted as x . The covariance matrix, which is defined as the expected value of the product of xi and xj , represents the relationships among the variables and is not a zero vector (Hong & Wu, 2012; Li et al, 2018; Naderi et al, 2023; Rastogi et al, 2021). By finding the eigenvectors and eigenvalues of the covariance matrix, we can determine which PCs (Hopwood et al, 2020; Nikkhah et al, 2019) account for the largest variances and subsequently explain most of the total variability of the vector.…”

Replacing energy simulations with surrogate models for design space exploration

“…Seyedzadeh et al [10] use Gradient Boosted Regression Trees to accurately estimate building energy consumption to calculate Building Emission Rates and to then find optimal solutions for retrofitting. Rastogi et al [11] present testing strategies to evaluate the feasibility of regression based SMs to substitute BPS. Besides presenting a testing methodology, they conclude that non-linear models are better suitable for SMs replacing BPS than linear models.…”