Transient heat transfer performance prediction using a machine learning approach for sensible heat storage in parabolic trough solar thermal power generation cycles

Although solar thermal energy is one of the most promising renewable energy sources, intermittency often limits its potential. Thermal storage is the key technology for adjusting the time gap between the supply and demand of renewable energy. Even if the performance of sensible heat storage can be analyzed using numerical simulations or experiments in conventional technology, any method is time-consuming due to its time-dependent nature. In this work, a methodology to predict the transient heat transfer performance of sensible heat storage, which is used with direct steam generation parabolic trough solar thermal power generation cycles, based on a machine learning technique is developed. This methodology can handle varying fluid flow rates and is, therefore, ad-vantageous for temperature control problems. To demonstrate the capabilities of the proposed method, regres-sion models are trained using transient heat transfer analysis results based on the weather data of Bawean Island, Indonesia. During the regression model training process, Gaussian process regression shows the best prediction accuracy among 26 regression models. The prediction errors of 14-day consecutive operation turbine shaft work are 1.45 % or less, compared to the transient simulation results. The trained regression model is also applied to Kupang, about 1300 km apart from Bawean. The turbine shaft work prediction errors of 14-day consecutive operations are 3.24 % or less. Because of a significant computational time reduction, the proposed methodology is suited for solar thermal power generation site selection and plant concept design.

Automated summary

A machine learning technique is developed to predict the transient heat transfer performance of sensible heat storage in solar thermal systems. Regression models trained using weather data from different locations demonstrate high prediction accuracy and applicability.