I have multiple hourly time series measurements from different measurement points, for multiple weeks. My goal is to eventually cluster the measurement points into clusters, but to reduce dimensionality and improve clustering, I would like to attempt to find representative daily pattern for each time series first. So comparing the daily patterns from each time series, finding if there is some clear pattern repeating in multiple days and then averaging them into one representative daily pattern. If repeating daily patterns are not found, then this measurement point could be labeled as outlier. What would be good and efficient method for this? Normal euclidean distance between each day is one option but maybe not the most optimal because of small shifts. DTW would maybe work better, but being too computationally heavy for a large dataset.
