Bayesian-optimized-CNN-LSTM-based performance prediction for organic Rankine cycle system with cyclopentane

The organic Rankine cycle (ORC) has emerged as a highly efficient technology for waste heat recovery, with rapidly expanding applications. In this study, a waste heat recovery test bench utilizing cyclopentane as the working fluid was established. During experimentation, the mass flow rate and turbine inlet pressure significantly influenced the system performance but varied synchronously, preventing the isolated examination of their individual effects on system performance. However, in large-scale applications – where stable turbine expansion ratios and smooth start-stop operations are critical – independent control of these parameters is often required. To address this challenge, a Bayesian-optimized convolutional neural network-long short-term memory (BO-CNN-LSTM) model, trained on experimental data, was developed to predict system performance, decoupled the coupled system’s key parameters and analyzed their individual impacts. Results demonstrate that this method accurately predicts system performance, with an R2 value exceeding 0.99, and successfully decouple the coupled parameters. Their independent influences on the system performance were analyzed respectively. This approach effectively simplifies experimentation and reduces potential experimental costs.