A Stacking Learning Model Based on Multiple Similar Days for Short-Term Load Forecasting

It is challenging to obtain accurate and efficient predictions in short-term load forecasting (STLF) systems due the complexity nonlinearity of electric signals. To address these problems, we propose a hybrid predictive model that includes sliding-window algorithm, stacking ensemble neural network, similar-days method. First, leverage algorithm process time-series data with high non-stationarity. Second, an learning scheme networks improve performance. Specifically, contain two types networks: base-layer meta-layer networks. During pre-training process, network integrates radial basis function (RBF), random vector functional link (RVFL), backpropagation (BPNN) provide robust model. The utilizes deep belief (DBN) improved broad system (BLS) enhance accuracy. Finally, prediction method developed extract relationship different time dimensions, further enhancing robustness accuracy demonstrate effectiveness our model, it evaluated using real from five regions United States three consecutive years. We compare several state-of-the-art conventional neural-network-based models. Our proposed improves by 16.08%, 16.83%, 22.64% compared DWT-EMD-RVFL, SWT-LSTM, EMD-BLS, respectively. Empirical results achieves better baselines.