Hourly performance forecast of a dew point cooler using explainable Artificial Intelligence and evolutionary optimisations by 2050

Abstract

The empirical success of the Artificial Intelligence (AI), has enhanced importance of the transparency in black box Machine Learning (ML) models. This study pioneers in developing an explainable and interpretable Deep Neural Network (DNN) model for a Guideless Irregular Dew Point Cooler (GIDPC). The game theory based SHapley Additive exPlanations (SHAP) method is used to interpret contribution of the operating conditions on performance parameters. Furthermore, in a response to the endeavours in developing more efficient metaheuristic optimisation algorithms for the energy systems, two Evolutionary Optimisation (EO) algorithms including a novel bio-inspired algorithm i.e., Slime Mould Algorithm (SMA), and Particle Swarm Optimization (PSO), are employed to simultaneously maximise the cooling efficiency and minimise the construction cost of the GIDPC. Additionally, performance of the optimised GIDPCs are compared in both statistical and deterministic way. The comparisons are carried out in diverse climates in 2020 and 2050 in which the hourly future weather data are projected using a high-emission scenario defined by Intergovernmental Panel for Climate Change (IPCC). The results revealed that the hourly COP of the optimised systems outperform the base design. Although power consumption of all systems increases from 2020 to 2050, owing to more operating hours as a result of global warming, but power savings of up to 72%, 69.49%, 63.24%, and 69.21% in hot summer continental, Arid, tropical rainforest and Mediterranean hot summer climates respectively, can be achieved when the systems run optimally.