An adaptive forecasting method for the aggregated load with pattern matching

Electrical load forecasting plays a vital role in the operation of power system. In this paper, novel adaptive short-term method for aggregated is built. The proposed consists two stages: forecast model preparation stage and selection stage. first stage, based on historical data all consumers, typical monthly patterns are firstly identified an optimal fashion with aid cosine similarity. Then, each pattern, stacking ensemble learning to train model. second according similarity between individual latest month consumers classified into different groups where group corresponds particular pattern. group, corresponding trained employed final calculated as simple aggregation produced consumers. Case studies conducted open dataset show that, compared single model, can effectively improve accuracy.