Enhancing energo-exergo-economic performance of Kalina cycle for low- to high-grade waste heat recovery: Design and optimization through deep learning methods

The Kalina cycle has proven to be a reliable bottoming cycle for low-grade waste heat recovery. However, compared with other recovery cycles (e.g. the Rankine cycle), it is characterized by a lower efficiency rate and constraints on the heating medium inlet temperature. This study proposes a systematic method for configuring and optimizing three novel Kalina-trilateral-based systems to overcome those disadvantages. This accurate technique integrates thermodynamics with deep learning to accelerate the computation process. First, the actual thermodynamic-economic features of the alternative systems are modeled through the analyses of energy, exergy, and economy. Second, the surrogate models of the systems are developed through a long short-term memory (LSTM) network. Third, the direct and hybrid optimization algorithms are applied separately to the actual and surrogate models. The objective functions yield thermal efficiency, exergy efficiency, and distributed payback time. Moreover, the strength Pareto evolutionary algorithm (SPEA-II) is employed to solve the multiobjective optimization problem. According to the results, the LSTM network had considerable power to simulate and predict the energo-exergo-economic performance of an energy system. Furthermore, the computation time was much shorter for a hybrid algorithm with the maintained accuracy. Consequently, the heating source temperature constraint was eliminated in the final alternative cycle (KTS-36). Compared with the base system (KCS-34), the thermodynamic and economic objective functions were improved by 74.3% and 34%, respectively.