Validity index based improvisation in reproducibility of load profiling outcome

Abstract

Under competitive market environment, the knowledge of load variation and its behavioural aspect are essential for the survival of electric supply utility. Load profiling is a useful tool to acquire it, and k-means clustering is a crisp, robust and widely proposed method. However, the multiple runs of a k-means algorithm for the same problem do not result in a deterministic outcome because of its dependency on initialisation. This study proposes an improved k-means algorithm to reduce the stochasticity in the outcome. The methodology combines clustering and qualitative validation processes and extracts the best outcome based on several validity indices that take into account cohesion and separation. The external cluster validity indices are innovatively used for comparison of the proposed method with the k-means algorithm. If the same problem is repeatedly executed, the variation in the outcome is lesser as compared to the standard k-means algorithm, thus improving the repeatability and overall quality of the clustering. The 24 real-life data-sets of load patterns of distribution feeders of the Central Indian state discom have been used for the study. The results show that proposed methodology reduces the stochasticity in the outcome significantly.