VMD-LSTM prediction model based on the multi-strategy Antlion optimization algorithm

Abstract To more accurately predict the trend of time series changes, a VMD-LSTM prediction model based on the Multi-strategy Antlion Optimization Algorithm (MALO) is proposed. First, the efficiency of the ant search is enhanced by a dimension-cross random walk strategy. A two-stage weighted selection strategy is employed to increase population diversity and strengthen the algorithm’s ability to escape local optima. Then, MALO is used to optimize the decomposition level and penalty factor of VMD. The optimized VMD is applied to decompose the time series into multiple subsequences. Finally, LSTM prediction models are established for each subsequence. The prediction results are reconstructed by aggregation. Experimental results on six benchmark functions demonstrate that MALO significantly outperforms other compared algorithms in convergence speed and accuracy. Additionally, the MALO-VMD-LSTM model achieves normalized RMSE and MAE values of 1% and 0.8%, respectively, in the prediction of underground water inflow in mining, demonstrating superior prediction accuracy and generalization ability compared to other models.