An Optimized Power Load Forecasting Algorithm Based on VMD-SMA-LSTM

Abstract

Accurate load forecasting can scientifically guide the optimal operation and scheduling of urban power grids. This study introduces an enhanced power load forecasting algorithm, integrating slime mould algorithm (SMA) and long short-time memory (LSTM) to effectively address the hyperparameter challenges associated with LSTM, while also applying variational modal decomposition (VMD) to load forecasting. In the data processing stage, the Bisecting Kmeans algorithm (Bi-Kmeans) is used to identify the outliers of the measured load data, then the random forest (RF) is used to correct them, which determines reasonable load data. In the data analysis stage, the processed load data undergoes VMD, yielding components with distinct central frequencies, and the components of different frequencies are determined according to their energy values. In the prediction stage, an optimized LSTM using SMA is proposed to predict different frequency components separately, and the prediction results of multiple components are inversely reconfigured to obtain the load prediction results. Case studies demonstrate that the proposed algorithm outperforms other power load forecasting methods in prediction accuracy.