Predictive modelling for solar thermal energy systems: A comparison of support vector regression, random forest, extra trees and regression trees

TLDR

Predictive analytics are essential for managing decentralized energy systems, requiring accurate models of renewable generation and building consumption to optimize grids and support fault detection. The study compares tree‑based ensemble models (random forest and extra trees), decision trees, and support vector regression for predicting useful hourly energy from a solar thermal collector system. The authors developed these models and evaluated them on generalisation ability, accuracy, and computational cost. Random forest and extra trees achieve comparable predictive accuracy (RMSE ≈ 6.9–7.1 kWh), decision trees are faster but less accurate (RMSE ≈ 8.8 kWh), and support vector regression has the highest training time (~1.3 s).

Abstract

Predictive analytics play an important role in the management of decentralised energy systems. Prediction models of uncontrolled variables (e.g., renewable energy sources generation, building energy consumption) are required to optimally manage electrical and thermal grids, making informed decisions and for fault detection and diagnosis. The paper presents a comprehensive study to compare tree-based ensemble machine learning models (random forest – RF and extra trees – ET), decision trees (DT) and support vector regression (SVR) to predict the useful hourly energy from a solar thermal collector system. The developed models were compared based on their generalisation ability (stability), accuracy and computational cost. It was found that RF and ET have comparable predictive power and are equally applicable for predicting useful solar thermal energy (USTE), with root mean square error (RMSE) values of 6.86 and 7.12 on the testing dataset, respectively. Amongst the studied algorithms, DT is the most computationally efficient method as it requires significantly less training time. However, it is less accurate (RMSE = 8.76) than RF and ET. The training time of SVR was 1287.80 ms, which was approximately three times higher than the ET training time.