Deep Q-Value Neural Network (DQN) Reinforcement Learning for the Techno-Economic Optimization of a Solar-Driven Nanofluid-Assisted Desalination Technology

Abstract

A solar-driven desalination system, featuring a single-slope solar still is studied here. For this design, Al2O3 nanofluid is utilized, and the condition achieving the highest efficiency and cost-effectiveness is found using a reinforcement learning called a deep Q-value neural network (DQN). The results of optimization are implemented for the built experimental setup. Experimental data obtained under the climatic conditions of Tehran, Iran, are employed to compare the enhancement potential of the optimized solar still system with nanofluid (OSTSWNF) with the solar still system with water (STSWWA). The hourly fluid temperatures in the basin as well as the hourly and cumulative freshwater production (HFWP and CFWP) are discussed. A number of other parameters, including daily water production and efficiency in addition to the cost per liter (CPL) of the resulting desalinated water, are also taken into account. The results reveal that annual water production increases from 1326.8 L to 1652.4 L, representing ~25% growth. Moreover, the annual average efficiency improves by ~32%, rising from 41.6% to 54.7%. A great economic enhancement is seen as well, with the CPL decreasing by ~8%, i.e., from USD 0.0258/L to USD 0.0237/L.

Artificial intelligence applications in solar energy

“…These optimization methods entail generating and developing a desired solution to the problem throughout multiple generations. Mutation, cross-over, selection, and other genetic algorithms are commonly used for candidates in difficult regions, preferred solutions or closest approach assertions, genetics, and algorithms [105]- [108]. Fig.…”