The incorporation of phase change materials (PCM) into building envelopes has proven to reduce carbon emissions and energy consumption to combat climate change. However, the energy performance of PCM-enhanced building depends on several factors and the optimization of such factors using empirical building design is more difficult and sometimes even impossible. Therefore, this paper proposes a multi-objective optimization method to consider multi-objectives, including building energy consumption, economic benefit, and carbon-saving. In this study, a non-dominated Sorting Genetic Algorithm III (NSGA III) is coupled with a Stacking model to minimize building operational energy consumption (BOEC) and maximize life cycle economic benefit (LCEB) and life cycle carbon reduction (LCCR) simultaneously by finding the optimum configuration of PCM thickness, window-to-wall ratio, exterior glazing U-value and solar heat gain coefficient. The results show that the Stacking model combined with 8 heterogeneous machine learning models has the best performance for predicting energy consumption with a high correlation efficiency (R2 = 0.97). In addition, the building optimized with the Stacking-NSGA III framework shows a reduction of BOEC by 45.38 % and an increase of LCCR by 10.75 kg·CO2.e/m2. Moreover, the LCEB over a 50-year service life is 452.21 CNY/m2. It is believed that the proposed multi-objective optimization method can help stakeholders to find the most suitable PCM-building design strategies for their specific needs. •The effects of various decision variables of building envelope on energy consumption were investigated.•A novel Stacking model for predicting building energy consumption was proposed.•NSGA-III algorithm was used to address the multi-objective trade-off of PCM-enhanced building envelopes.•The design strategies for PCM-enhanced building envelopes were discussed.