Intelligent modeling and analysis of hybrid organic Rankine plants: Data-driven insights into thermodynamic efficiency and economic viability

Hybrid organic Rankine cycle (HORC) is a hydrodynamic plant used from industrial processes for low-temperature heat sources, such as geothermal, solar, and waste heat. Intelligent models were developed to predict the first and the second thermodynamic efficiencies and the levelized energy cost to optimize the overall thermal and economic efficiency of hybrid organic Rankine cycle-powered plants. Deep learning, gradient-boosting framework, and Kernel models such as Long Short-Term Memory (LSTM), Light Gradient Boosting Machine (LGBM), and Kernel Ridge Regression (KRR) models were developed to predict the three outputs of HORC according to five subsets: subset-1: design variables, subset-2: temperature variables, subset-3: power variables, subset-4: heat exchanger variables, and subset-5: all previous variables. The LSTM model generally achieved superior performance across the multiple input variables used in predicting the three model outputs. The LSTM model attained the lowest mean absolute percentage error (MAPE) (4.8%–13%), the highest coefficient of determination (R²) (up to 0.994), and the lowest root mean square error (RMSE) as low as 0.002. This demonstrated superior predictive accuracy across the various model input subsets. The LGBM model, however, showed moderate performance, with the MAPE reaching up to 25.6% and the R² ranging from 0.624 to 0.986. In contrast, the KRR model struggled to demonstrate exceptional performance, especially with the heat exchanger dataset, thus exhibiting a MAPE up to 54.4% and an R² value as low as 0.408. Therefore, we advocate that the LSTM model could be the most reliable model for predicting the system efficiency of HORC.