The analysis of the short-term power system load forecasting model based on the fuzzy clustering

A model of the short-term power system load forecasting based on fuzzy clustering is presented. It can be classified as similarity-based models relying on the assumption that if patterns of the time series sequences are similar, then the forecast patterns associated with them are also similar. This means that the patterns of the neighboring sequences are in some to each other, which does not change significantly with time. Depending the character and properties of the time series, this relation as well as similarity degree can be shaped with the help of definitions of patterns, membership and defuzzification functions, and the distance measure between patterns. Four types of the membership functions with optimized parameters were used in the model construction. The model performance with the various distance measures between patterns was empirically examined. The model sensitivity to the membership function width was analyzed. The model resistance to the blurred noisy and missing data and the model performance with the various definitions of the reference sets were analyzed. The tests allow us to formulate some conclusions on the model quality and resistance.