The paper proposes an approach to perform clustering on time series data. The data is fetched from temperature sensors, that are located in a public building, which defines certain features of the data as an origin of it. We present preliminary descriptive statistics of the data, which shows that it cannot serve as a feature in our case. The approach addresses to the problem of sufficient subsequence length for a time series, which allows to form stable groups of specific data. We have considered several distance metrics. The method is the iterative process of construction a solution, during which the window in a time series is the subject to be changed. The results might be used to build more precise yet general prediction model for groups of similar objects.