TY - JOUR

T1 - Development and comparison of local solar split models on the example of Central Europe

AU - Schlager, Elke

AU - Feichtinger, Gerald

AU - Gursch, Heimo

N1 - Publisher Copyright: © 2022 The Author(s)

PY - 2023/4

Y1 - 2023/4

N2 - Solar radiation influences many and diverse fields like energy generation, agriculture and building operation. Hence, simulation models in these fields often rely on precise information about solar radiation. Measurements are often restricted to global irradiance, whereby measurements of its single components, direct and diffuse irradiance, are sparse. However, information on both, the direct and diffuse irradiance, is necessary for simulation models to work reliably. In this study, solar separation models are developed using 10-min training data from two different locations in Austria. Direct horizontal irradiance is predicted via regressing the direct fraction using several objective functions. The models are first trained on a data set including data from both locations, and evaluated regarding root mean squared deviation (RMSD), mean bias deviation (MBD), and coefficient of determination (R2) on measured and estimated direct normal irradiance. The two best performing models are then selected for further analysis. This analysis comprises of an evaluation of the models per season, transferability of trained modes between two locations in Austria, a transferability and generalisability study conducted for four more locations in Central Europe, and a comparison with the trusted Engerer model. The solar separation model with polynomial terms up to order three and Ridge regularisation outperforms the second model based a logistic term in combination with mixed quadratic terms as well as the Engerer model.

AB - Solar radiation influences many and diverse fields like energy generation, agriculture and building operation. Hence, simulation models in these fields often rely on precise information about solar radiation. Measurements are often restricted to global irradiance, whereby measurements of its single components, direct and diffuse irradiance, are sparse. However, information on both, the direct and diffuse irradiance, is necessary for simulation models to work reliably. In this study, solar separation models are developed using 10-min training data from two different locations in Austria. Direct horizontal irradiance is predicted via regressing the direct fraction using several objective functions. The models are first trained on a data set including data from both locations, and evaluated regarding root mean squared deviation (RMSD), mean bias deviation (MBD), and coefficient of determination (R2) on measured and estimated direct normal irradiance. The two best performing models are then selected for further analysis. This analysis comprises of an evaluation of the models per season, transferability of trained modes between two locations in Austria, a transferability and generalisability study conducted for four more locations in Central Europe, and a comparison with the trusted Engerer model. The solar separation model with polynomial terms up to order three and Ridge regularisation outperforms the second model based a logistic term in combination with mixed quadratic terms as well as the Engerer model.

KW - Solar irradiance

KW - Direct normal irradiance

KW - Solar separation model

KW - Solar regression

KW - Solar model transferability

KW - Seasonality

UR - http://www.scopus.com/inward/record.url?scp=85144497328&partnerID=8YFLogxK

U2 - 10.1016/j.egyai.2022.100226

DO - 10.1016/j.egyai.2022.100226

M3 - Article

VL - 12.2023

JO - Energy and AI

JF - Energy and AI

IS - April

M1 - 100226

ER -