In time-of-use tariff period partition, clustering algorithms are
commonly used. However, as load demands become more diverse in this big
data era, large amount of non-linear data makes conventional clustering
algorithms methods no longer be applicable in this field alone. Facing
high-time-resolution daily load data with strong non-linearity, we
propose a new method to partition periods. It consists of an improved
fuzzy c-means clustering algorithm and a correction method for abnormal
periods. Firstly, we propose modified fuzzy membership functions to
improve the initialization of clustering for operation efficiency.
Secondly, the method for calculating the fuzzy parameters based on the
loss function is given. Thirdly, the initial period partition is
obtained by the improved clustering. Next, the recognition model and
fuzzy subsethood-based correction model for abnormal periods are
structed, then the corrected period partition is confirmed. Finally, the
effectiveness of the proposed methods is verified by two daily load data
with a time resolution of 5 minutes.