Robust design of ambient-air vaporizer based on time-series clustering

A methodology for the robust design of an ambient-air vaporizer under time-series weather conditions is proposed. Two techniques are used to extract representative features in the time-series data. (i) The major trend of a day is rapidly identified by the discrete wavelet transform (DWT), in which a high level of Haar function reflects the trend of a day and drastically reduces the data size. (ii) The k-means clustering method groups the similar features of a year, and the reconstructed time-series dataset extracted by the centroids of clusters represents the weather conditions of a year. The results of the multi-feature-based optimization were compared with non-wavelet based and multi-period optimization by simulation under a year of data. The design structure from the feature extraction shows 22.92% better performance than the original case and is 12 times more robust in different weather conditions than clustering with raw data.