A Machine Learning Based Approach for Selection of SF₆ Alternatives

Finding a suitable replacement for sulphur hexafluoride (SF₆) in the gas insulated equipment is a major challenge facing the energy industry due to its incredibly high global warming potential (GWP), which is 25,200 times greater than that of CO₂ with an atmospheric lifetime of 3,200 years. There are around 114 million unique compounds in the PubChem database, and it is physically impossible to test all existing chemical compounds through laboratory scaled
investigation. An iterative approach of mixture optimisation with a genetic algorithm can help narrow down the search space to a more feasible number of candidates. Besides a vast number of compounds and mixture combinations, there are also a range of parameters such as dielectric strength, boiling point, toxicity and GWP that must be collectively considered, which points to the application of advanced multi-objective optimisation techniques for balancing all the required properties. The benefit of the computational approach is clearly evident as the generated mixtures contain researched solutions reported in the literature such as C₃F₇CN and CF₃I. Furthermore, the developed approach is effective at identifying an optimal set of mixtures from a large space of possible mixture combinations and ratios.