Spectral clustering based demand-oriented representative days selection method for power system expansion planning

•The scheduling demand for adequacy and flexibility are accurately captured.•Enhance the adaptability to non-hyperspherical data distribution of clustering.•Reduce the relative error of total cost in system expansion planning.•The proposed method is effective in LP/MILP models and GEP/GTEP problems.•Investigate the influence of new energy and model parameters on selection methods. Due to the computational burden of the system expansion planning (SEP) models, it is a common approach to select representative days by clustering to make the SEP models tractable. However, the performance of the selection of representative days is unsatisfactory, as the conventional methods fail to properly capture the demand for adequacy and flexibility in system scheduling and to effectively cluster the feature vectors that are commonly high-dimensional and highly non-hyperspherical distributed. This paper proposes a spectral clustering based demand-oriented (SCDOR) representative days selection method, which extracts the daily net load duration curve (NLDC) and net load ramp duration curve (NLRDC) as a feature vector to accurately capture demand for the adequacy and flexibility. Additionally, the feature vectors are regarded as connected vertices in a graph, and the representative days are obtained through the graph partition by spectral clustering that is adaptable to different data distribution. The superior performance of SCDOR is demonstrated through the case studies with different levels of modeling complexity on the Texas power system and IEEE RTS-79 system.