Cross-substation short-term load forecasting based on types of customer usage characteristics

Abstract

This paper presents a short-term load forecasting scheme based on usage characteristics of customers. Four types of customers including industrial, commercial, high density residential, and low density residential sectors are considered. The days of week including special holidays are also taken into account. To be more specific, previous loads and forecasted temperature are used as the input to support vector machines to predict load in the next 24 hours. A new normalization method based on temporal segments is also proposed. Rather than testing only on the training substations, the cross-substation test is also experimented. The good performances with the mean absolute error (MAE) of 1.45 MW and the mean absolute percentage error (MAPE) of 4.58% are achieved on average when testing on the same substations. The average MAE and MAPE for the cross-substation test are 1.46 MW and 7.66%, respectively. This demonstrates that the proposed forecasting scheme can be applied in new substations without retraining the system.