Accurate energy system modelling of chemical separations is a critical component of technology selection to minimize operating costs, energy consumption and emissions. Here we report a hybrid modelling approach based on data-driven and mechanistic models to holistically compare chemical separation performance. Our model can be used to select the most suitable technology for a given chemical separation, such as membrane separation, evaporation, extraction or hybrid configurations, by training a machine learning model to predict solute rejection using an open-access membrane dataset. We estimated an average 40% reduction in energy consumption and carbon dioxide emissions for industrially relevant separations using our methodology. We predicted and analysed 7.1 million solute rejections across several industrial sectors. Pharmaceutical purification could realize carbon dioxide emissions reductions of up to 90% by selecting the most efficient technology. We mapped the reduction in carbon dioxide emissions and the reduction in operating costs globally, establishing parameter thresholds to facilitate corporate and governmental decision-making.

化学分离的准确能源系统建模是技术选择的关键组成部分，可最大限度地降低运营成本、能耗和排放。在这里，我们报告了一种基于数据驱动和机理模型的混合建模方法，以全面比较化学分离性能。我们的模型可用于为给定的化学分离选择最合适的技术，例如膜分离、蒸发、萃取或混合配置，方法是使用开放访问膜数据集训练机器学习模型来预测溶质排斥。我们估计，使用我们的方法，工业相关分离的能耗和二氧化碳排放量平均减少了 40%。我们预测并分析了多个工业部门的 710 万次溶质拒绝。通过选择最有效的技术，药物纯化可以减少高达 90% 的二氧化碳排放。我们绘制了全球二氧化碳排放量减少和运营成本降低的地图，建立了参数阈值以促进公司和政府决策。