Data-Driven Predictive Modeling Techniques for High-Dimensional Solar–Thermal Performance Analysis

Abstract

The operating conditions of solarthermal energy systems are very dynamic and complex and performance is dictated by nonlinear relationships among the environmental variables, material properties, system geometry, and operational controls. Proper modeling of high-dimensional behavior of this nature is still an important issue of conventional physics-based and reduced-order thermal models especially when scalability and real-time analysis are of interest. This paper suggests an elaborate evidence-based predictive modelling framework to general purpose high-dimensional solar-thermal performance forecasting, using scalable data processing pipelines and smart methodologies of machine learning. The framework combines systematic preprocessing of data, feature dispensing and dimensionality reduction to effectively handle large sized and heterogeneous data that is a product of real-world measurements and high-fidelity simulations. Several predictive models, such as ensemble learning and deep neural architecture, are tested to learn and provide the complex nonlinear correlations in the feature space, whereas model interpretability methods are utilized to detect key aspects of performance that affect the results. Results of experimental studies also confirm that the proposed framework is better than the typical regression-based algorithms or the simplified thermal models in terms of overall accuracy in prediction, resistance to noise, and scalability. The results also suggest that the fidelity of prediction is not greatly impaired by intelligent feature compression as this significantly increases model efficiency. On the whole, this study demonstrates that scalable data-driven methods can play a potent role as a supplement to conventional thermal analysis by achieving precise performance predictions, system optimization, and decision-making with regard to next-generation solar-thermal energy systems that will be functioning in a highly dimensional and data intensive environment.