Short-Term Power Load Prediction Based On SSA-VMD-Iterated Function System

Load forecasting plays a critical role in the power system,as accurate and timely load predictions enable effective grid dispatch and generator control.Currently,various methods are used for load forecasting,such as artificial neural network models,time series models,and regression analysis models.However,these models require a large amount of historical data for training and may suffer from convergence issues.Therefore,this paper proposes a fractal theory-based SSA-VMD and Iterative Function System(IFS)model.The research methodology of this paper begins by selecting a reference day that is similar to the actual measurement day.The relative entropy is used as the fitness value in the Sparrow Search Algorithm.By comparing the correlation between the subsequence signals and the original signal sequence under different modal numbers and penalty factors in the Variational Mode Decomposition,the value of relative entropy is determined.A smaller difference between the two sequences corresponds to a smaller relative entropy,indicating a greater presence of true components in the subsequence signals,and vice versa.Then,optimal parameters are used for mode decomposition,dividing the subsequence into sequences that require fractal iteration and sequences that do not.The sequences requiring fractal iteration undergo range-scaling analysis using the method of interpolation interval determination based on waveform extremum points,and the iterative compression factor is calculated.The interpolated data and iterative compression factor are then input into the iterative function system to obtain the prediction results in terms of average absolute percentage error.Finally,a comparative analysis is conducted between the proposed SSA-VMD and IFS model,LSTM neural network model,and time series prediction model.The results demonstrate that the SSA-VMD and IFS model provides more accurate load predictions compared to the other two models.Furthermore,real-time data is used for validation,confirming that the SSAVMD and IFS short-term load forecasting model outperforms the long short-term memory neural network model when faced with limited dataset availability.This research method is helpful to improve the accuracy and rapidity of power load prediction and provide a more reliable reference for power grid dispatching and generator set regulation..