Abstract:
Waste heat recovery (WHR) technology is gaining popularity in recent years to address global energy crisis and provide an environment friendly robust alternative for utilizing medium and low grade heat. Among the available technologies for WHR purpose, Organic Rankine Cycle (ORC) is a reliable method in respect of the working fluid property, cycle configuration and efficiency, economy, and robustness.Therefore, optimization of the ORC for WHR purpose is becoming a promising research topic in the energy sector. However, most of the work on ORC optimization have focused on the traditional approach of acquiring the cycle state points from the direct thermodynamic modelling, which can be complex and time consuming. In this study, an efficient method of ORC optimization for WHR is presented based on artificial intelligence (AI) technique. Based on four ORC configurations namely Basic ORC, Reheating ORC, Internal Regenerative ORC, and Combined Reheating Internal Regenerative ORC, a database is constructed with nine pre-selected organic working fluid for training an AI framework. The prediction performances of four machine learning (ML) algorithms, namely Linear Regression (LR), Support Vector Regression (SVR), Decision Tree Regression (DTR), and Random Forest Regression (RFR), are compared and RFR is observed to produce the best prediction results. After that, a Back Propagation Neural Network (BPNN) is trained with the formulated dataset and the prediction results are compared with that of RFR. It is found that, BPNN is faster but more data sensitive whereas RFR can produce fairly accurate prediction result with less amount of data than BPNN. From the sensitivity analysis, two parameters namely heat source temperature and reheating pressure ratio are found to be the most influential in ORC performance. Based on this analysis, an optimization framework is constructed keeping the trained AI algorithms as proxy model to generate the objective function data. The optimized thermal and exergy efficiency are found to be close to one another for traditional thermodynamic modelling, RFR, and BPNN which indicates that the proposed ANN scheme can effectively be used as a proxy model for thermodynamic analysis during the optimization process. Furthermore, the RFR and BPNN work nearly 7 times and 20 times faster than the traditional approach, indicating that they can successfully be used as a less complex and less time consuming alternative for ORC modelling.
