An Ensemble Model based on Deep Learning and Data Preprocessing for Short-Term Electrical Load Forecasting

Electricity load forecasting is one of the hot concerns current electricity market, and many models are proposed to satisfy market participants’ needs. Most have shortcomings large computation or low precision. To address this problem, a novel deep learning data processing ensemble model called SELNet proposed. We performed an experiment with model; consisted two parts: forecasting. In part, autocorrelation function (ACF) was used analyze raw on determine be input into model. The variational mode decomposition (VMD) algorithm decompose raw-data set relatively stable modes named intrinsic functions (IMFs). According time distribution lag determined using ACF, reshaped 24 × 7 8 matrix M, where 24, 7, represent h, days, IMFs, respectively. two-dimensional convolutional neural network (2D-CNN) extract features from M. improved layer reshape extracted according order. A temporal then employed learn time-series combined fully connected complete prediction. Finally, performance verified in Eastern Market Texas. demonstrate effectiveness forecasting, we compared it gated recurrent unit (GRU), TCN, VMD-TCN, VMD-CNN models. TCN exhibited better than GRU mean absolute percentage error (MAPE) which over 5%, less that GRU. Following addition VMD basic 2–3%. comparison between VMD-TCN indicated application 2D-CNN improves forecast performance, only few samples having MAPE 4%. model’s prediction effect each season discussed, found can achieve high-precision season.