Short-Term Power Load Forecasting Based on Genetic Algorithm Improved VMD-BP

Initially, the Variational Mode Decomposition (VMD) technique is applied to break down the original load data into multiple Intrinsic Mode Functions (IMFs) along with a residual component. This step effectively tackles the non-stationary and intricate nature of the load data. Next, a Back Propagation (BP) model is utilized to forecast the various segmented components resulting from the decomposition. To further refine the VMD-BP neural network, a genetic algorithm is employed to optimize parameters such as thresholds, initial weights, and smoothing factors. By aggregating the predictions of the individual components, the final load forecast is produced. Experimental results indicate that the R2 value achieved by the GA-VMD-BP model is 31.71% and 1.46% higher than those of the BP model and VMD-BP model, respectively. Furthermore, the GA-VMD-BP model shows a decrease in MAE by 205.91 MW and 48.51 MW, RMSE by 383.06 MW and 51.64 MW, and MAPE by 2.95% and 0.62%, compared to the other two models.