Key points are not available for this paper at this time.
A novel method for short-term electrical load forecasting using back propagation neural networks (BPNNs) is proposed for reducing the forecasting error. Conventionally, BPNN for load forecasting will have a single network structure trained by either similar day (SD) or day ahead (DA) approach. A model trained using either similar day or day ahead can only learn the characteristics of either approach. Also, a single BPNN model that incorporates both will have high complexity in its structure. The proposed sequential hybrid neural network method employs BPNNs in two stages, utilizing both similar day and day ahead. The proposed method is compared against similar day and day ahead approaches. The models are tested using hourly electrical load data from the Electric Reliability Council of Texas, Texas in USA and the Global Energy Forecasting Competition of 2012. It is observed that the proposed method showed an improvement in forecasting accuracy over the BPNN and artificial neural network-particle swarm optimization models available in literature.
Eapen et al. (Tue,) studied this question.