A hybrid intelligent time-series framework for predicting short-term LNG sendout rate

Accurate forecasting of the liquefied natural gas (LNG) sendout rate is essential for stabilizing downstream natural gas supply and shaving emergency peak effectively. However, the volatility of LNG sendout poses challenges for traditional methods in capturing fluctuation patterns, affecting prediction accuracy and decision-making in pipeline and LNG terminals. To overcome the drawbacks, this study introduces a hybrid intelligent time series framework for day-ahead LNG sendout rate forecasting. Firstly, the LNG sendout rate time series is decomposed into multiple intrinsic mode functions (IMFs) using the optimized variational mode decomposition (OVMD) method, thereby reducing non-stationarity and data complexity. Since the performance of time series models is significantly affected by parameter selection, and manual tuning is inefficient for obtaining optimal parameter combinations, a sparrow search algorithm (SSA)-optimized Long Short-Term Memory (LSTM) model is developed. The nonlinear temporal features in each intrinsic mode function are captured by the SSA-LSTM model, and the results are subsequently aggregated to generate the final LNG sendout rate forecast. The proposed intelligent framework is validated by taking an LNG receiving terminal in China as an example. The results demonstrate that the hybrid model outperforms several advanced machine learning models in accuracy and stability, achieving R ² values of 0.929 and 0.970 for gaseous and liquid sendout rates, respectively. Compared to other models, the proposed hybrid framework effectively extracts temporal information from the load series and mitigates the influence of nonlinear factors, significantly enhancing prediction accuracy and offering insights for scheduling and operational optimization of LNG terminals.