Surrogate-Assisted Many-Objective Optimization of Building Energy Management

Building energy management usually involves a number of objectives, such as investment costs, thermal comfort, system resilience, battery life, and many others. However, most existing studies merely consider optimizing less than three objectives since it becomes increasingly difficult as the number of objectives increases. In addition, the optimization of building energy management relies heavily on time-consuming energy component simulators, posing great challenges for conventional evolutionary algorithms that typically require a large number of real function evaluations. To address the above-mentioned issues, this paper formulates a building energy management scenario as a 10-objective optimization problem, aiming to find optimal configurations of power supply components. To solve this expensive many-objective optimization problem, six state-of-the-art multi-objective evolutionary algorithms, five of which are assisted by surrogate models, are compared. The experimental results show that the adaptive reference vector assisted algorithm is proven to be the most competitive one among the six compared algorithms; the five evolutionary algorithms with surrogate assistance always outperform their counterpart without the surrogate, although the kriging-assisted reference vector assisted evolutionary algorithm only performs slightly better than the algorithm without surrogate assistance in dealing with the 10-objective building energy management problem. By analyzing the non-dominated solutions obtained by the six algorithms, an optimal configuration of power supply components can be obtained within an affordable period of time, providing decision makers with new insights into the building energy management problem.