Research and Application of a Time Series Data Prediction Model Based on Dense Residual Networks and Long Short-Term Memory Networks

This study proposes a hybrid deep learning model, termed Dense Residual Long Short-Term Memory (DRLSTM), for financial time series forecasting. The architecture integrates residual connections to stabilize gradient flow and dense connectivity to enhance feature reuse, combined with the memory and gating mechanisms of LSTM to improve long-term dependency modeling. Real-world datasets from the SSE Composite Index, NASDAQ, and Hang Seng Index are employed, where training samples are constructed using a sliding window approach, normalized with Z-score, and hyperparameters optimized via grid search. Experimental evaluations show that DRLSTM provides more accurate and robust predictions than representative benchmark models. The results suggest that DRLSTM can serve as a practical component of financial decision support systems and other expert applications requiring reliable time series forecasting.