Short-term load forecasting based on parallel frameworks

In this paper, a short-term load forecasting method which considers the different load characteristics in different periods is proposed. Firstly, we use parallel K-Means algorithm to cluster the daily load curves with 96 points of electricity customer to obtain some date groups with different load characteristics. Then for each date group, we use the daily load curves in the group to build load forecasting model of every point by the random vector functional-link net. Finally, we find the similar historical day of the forecast day by dynamic time warping method and use the forecasting models of the date group which contains the similar date to forecast the customer's load in the forecast day. Empirical study shows that the method is suitable for the short-term load forecasting of massive customers and has satisfactory forecasting accuracy.