Performance indices and evaluation of algorithms in building energy efficient design optimization
CHALLENGE - Building energy efficient design optimization is an emerging technique that is increasingly being used to design buildings with better overall performance and a particular emphasis on energy efficiency. To achieve building energy efficient design optimization, algorithms are vital to generate new designs and thus drive the design optimization process. Therefore, the performance of algorithms is crucial to achieving effective energy efficient design techniques.
SOLUTION - To establish a common ground for evaluation, a benchmark building, following the model of the DOE medium office building, is defined and a benchmark energy efficient design optimization problem is generated. The optimization problem contains six design variables, including four upper positions of the windows on each elevation, orientation, and thermal conductivity of the opaque building envelope. A set of performance indices, namely, stability, robustness, validity, speed, coverage, and locality, is proposed to evaluate the overall performance of algorithms. The Hooke-Jeeves algorithm, MultiObjective Genetic Algorithm II, and Multi-Objective Particle Swarm Optimization algorithm are evaluated by using the proposed performance indices and benchmark design problem. The Hooke-Jeeves algorithm is implemented in GenOpt, while the MOGA-II and the MOPSO algorithm are applied in modeFRONTIER.
BENEFITS - Results show that when facing an energy efficient design problem, the algorithm must be carefully selected based on the nature of the problem and the performance indices that matter the most. Here MOGA-II and the MOPSO algorithm generally perform better than the Hooke-Jeeves algorithm except for the locality. Table 7 above lists the algorithm performance under the six indices.