A spatial load forecasting method based on DBSCAN clustering and NAR neural network

Abstract

Abstract In order to improve the accuracy of spatial load forecasting in power grid planning stage, a spatial load forecasting method based on density-based spatial clustering of applications with noise (DBSCAN) clustering algorithm and nonlinear auto regressive (NAR) neural network is proposed. This method consists of three stages: cell division, clustering, and forecasting. At first, zones are divided into cellules that are taken as the basic unit of spatial load forecasting. Historical yearly load profiles, along with geographic information and land use types, are extracted from cells as features. Furthermore, similar cells are classified into several clusters according to these features. Finally, a NAR neural network is established to forecasting load one year ahead for each cluster, where the historical load profiles are taken as input. Experiments reveal that our proposed model decreases MAE by 45.95%, 42.04% and 47.49% respectively compared with linear regression, grey theory and exponential smoothing, showing great improvements in accuracy.