GRU Networks with improved Local Attention for Nonlinear Time Series Analysis

This research explores a novel computational approach for time series analysis based on LSTM’s lightweight variant GRU, combined with local attention mechanisms. The proposed model integrates innovative feature selection techniques and evolutionary algorithms to optimize model performance, addressing complex nonlinear problems in sequential data processing. Experimental results on financial datasets show that the optimized model improves prediction accuracy and generalization capability, offering important application value in data-driven decision-making and risk management. The model achieves a 0.9% improvement in R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, 5.0% reduction in MAE, and 9.4% decrease in MSE compared to standard baseline models, with potential applications for various time-dependent forecasting tasks.