Study of Short-term Load Forecasting Based on FCM and Multiple WOA-LSSVM Combination under Electricity Market Environment

Affected by many factors such as the development of the electricity market and extreme weather, it is very difficult to obtain high-performance short-term load forecasting methods. The day-ahead load forecasting of power system is deeply studied by dividing 24-hour period into several time sub-periods, clustering similar days separately, and building multiple sets of models for collaborative optimization. As a result, an improved short-term load forecasting method is proposed by considering similar days and time division. First, the total data set is divided into several data subsets based on the time division result. Then, Fuzzy C-Means (FCM) algorithm is applied to each data subset and several classes are grouped according to the characteristics of load segments. Next, multi-groups of Whale Optimization Algorithm (WOA) and least squares support vector machine (LSSVM) models are established to train the data subsets filtered by the category of the day before the prediction date. Finally, the improved model which considering similar day clustering and period partitioning, the FCM and WOA-LSSVM combined model which does not consider period partitioning and the traditional WOA-LSSVM model are used to predict the day-ahead load data of Jiangxi Province. The result of the experiment shows that the improved method can combine the advantages of similar day clustering and optimization based on load characteristics in different periods, which has high prediction accuracy and practical application value.