Research on multi-objective energy optimization design for multi-story residential buildings in Suzhou region based on artificial neural networks

To address the issues of high energy consumption, low thermal comfort, and excessive greenhouse gas emissions in residential buildings, this study optimizes multi-story residential buildings in the Suzhou region using a multi-objective Non-dominated Sorting Genetic Algorithm III (NSGA-III) coupled with an Artificial Neural Network (ANN). First, a representative multi-story residential building in Suzhou was selected as the research object, and a simulation model was developed in Design Builder. JePlus was used for batch simulations and dataset generation to train and validate the ANN metamodel. Before optimization, sensitivity analysis was conducted on input parameters to determine their influence. The NSGA-III algorithm was then dynamically linked with the ANN, iterating based on predefined objectives and adjusting inputs to produce 200 Pareto front solutions. The VIKOR method identified the final retrofit strategy. Compared with the original design, energy consumption, thermal comfort, and life-cycle CO 2 emissions improved by 45.8%, 12.2%, and 28.0%, respectively. This study supports energy-saving upgrades for cold-region residential buildings, highlights multi-objective optimization potential, and serves as a reference for Suzhou’s residential design.