Model predictive control of a phase change material–enhanced switchable insulation system using long short-term memory networks and genetic optimization

This paper presents a computational framework for transforming building facades into active climate-responsive systems through an integrated automation architecture. The study's primary goal is to maximize energy efficiency by using real-time Model Predictive Control to organize the synergy between phase change materials (PCMs) and switchable insulation systems (SIS). A key contribution is an autonomous hybrid envelope, driven by a decision support system, implemented using a framework that bridges EnergyPlus with a long short-term memory neural network (LSTM) and a genetic algorithm. The methods involved numerical modeling validated against experimental data with high precision to automate SIS actuation. Key results demonstrate that an intelligent control system achieved an 84.08% total energy reduction over uninsulated baselines and a 53.16% improvement over static insulation. With a 25 ms inference speed and a 5-s optimization loop, the framework provides a computationally efficient and scalable solution for the automated management of constructed facilities in diverse climates.