Do multisource data matter for NGP prediction? Evidence from the G-LSTM model

Area of Science:

• Energy Economics
• Computational Intelligence
• Time Series Forecasting

Background:

• Natural gas prices (NGPs) are vital for energy scheduling and planning.
• NGP prediction is challenging due to inherent nonlinearity and randomness.
• Existing models often struggle with the complexity of influencing factors.

Purpose of the Study:

• To develop a multifactor-driven hybrid forecasting model for natural gas prices.
• To leverage multisource data and advanced optimization techniques for improved prediction.
• To enhance decision-making in energy scheduling, planning, and control.

Main Methods:

• Extracted sentiment and readability from news text using VADER and textstat.
• Integrated news and search index data using correlation coefficient and CRITIC methods.
• Employed Gray Wolf Optimization to fine-tune Long Short-Term Memory (LSTM) model parameters.
• Validated the model using Henry Hub spot prices from 2012-2022.

Main Results:

• Multisource data significantly improved the predictive power for natural gas prices.
• The hybrid model demonstrated superior performance across various scenarios.
• Ablation experiments confirmed the value of incorporating diverse data sources.

Conclusions:

• The proposed hybrid model offers a promising approach for accurate NGP forecasting.
• Integrating information gain from multisource data enhances prediction effectiveness.
• The model provides a robust tool for energy market analysis and decision support.