Greedy Clustering-based Monthly Electricity Consumption Forecasting Model

Abstract

Accurate monthly electricity consumption forecasting is indispensable for electricity retailers to mitigate trading risks in the electricity market. Clustering-based forecasting method are commonly used to generate accurate monthly electricity consumption forecasting results. This paper focuses on the problem that the existing clustering-based monthly electricity consumption forecasting methods perform clustering and forecasting independently, causing that the joint optimization of two steps cannot be achieved. The reason for this situation is that the target of current clustering algorithms, maximizing individual similarity in a group, is not consistent with the final target of improving the forecasting accuracy. To solve the above problem, the greedy clustering-based monthly electricity consumption forecasting model (GCMECF) is proposed in this paper. Its clustering step takes improving the overall predictability as the optimization target, which is closely related to the forecasting target. In this way, with matching targets, the joint optimization of clustering and forecasting can be achieved. Meanwhile, the selection of the optimal number of clusters is decided based on the forecasting performance under multiple clustering scenarios. The case study verifies the effectiveness and superiority of the proposed method via a realworld dataset.