Bayesian Optimisation Based LSTM Model for Time Series Forecasting Problems

While Long Short-Term Memory (LSTM) networks excel in handling time series data, Bayesian optimisation techniques offer significant advantages in tuning model parameters to adapt to variable matching conditions and data characteristics. In this study, by combining Bayesian optimisation with LSTM model, the prediction performance and model robustness are significantly improved. Experimental results show that the Bayesian optimised LSTM model improves the R2 metric by 23.3% compared with the traditional LSTM model. Bayesian optimisation approximates the global optimal solution step by step by constructing a probabilistic model of the objective function and using the model to intelligently select evaluation points in the unexplored parameter space. The results of this paper show that the Bayesian optimised LSTM model outperforms the traditional LSTM model in terms of prediction accuracy, model stability and generalisation ability. This research provides a novel and effective approach for future applications in time series data analysis, especially in practical scenarios that require rapid adaptation to complex dynamic changes, which is of great theoretical significance and application value.