Electric Load Forecasting Based on SSA-GRU

To increase the reliability of electric load forecasts, This paper proposes a two-stage electric load forecasting model based on Singular Spectral Analysis pattern decomposition (SSA) and Gate Recursive Unit network (GRU). First, By using SSA, the electric load sequence is broken down, and then the GRU model is built to predict the subseries separately, and the forecast results of the subseries are superimposed to achieve highly accurate electric load forecasting. Compared with the traditional model, The SSA-GRU model suggested in this study has extremely broad application and is not dependent on either the smoothness requirement or the assumption of a parametric model. The experimental results show that the MAE of this hybrid model is 2774.5293, RMSE is 4111.5517, R2 is 0.9017 and MAPE is 0.0126 on the test set. It is proved that the errors of the prediction results of the designed method for wind power are all within the optimal error range and the predicted values have higher values. Compared with the comparison models, This paper's SSA-GRU model has a stronger predictive impact.