Improving Aggregated Load Forecasting Using Evidence Accumulation k-Shape Clustering

Aggregated load forecasting provides a basis for load aggregators to take part in the power market. By separating time series into several groups according to shape characteristic, k-Shape clustering-based approach is an effective way to implement aggregated forecasting. In this paper, we aim at further improving its performance on both probabilistic and deterministic aggregated forecasts by using ensemble technique. By transforming the partitions of load profiles into the similarity matrix, an evidence accumulation k-Shape clustering method is proposed to establish the hierarchical similarity structure of customers. Then, through varying the number of clusters, multiple probabilistic or deterministic aggregated forecast results are obtained. Determination of the optimal weights for combining the results is formulated as linear programming (LP) problems, with the objective of minimizing the pinball loss or mean absolute percent error (MAPE) respectively. Case study on the open dataset demonstrates the superiority of the proposed method.