Empirical Mode Decomposition with Swarm-Optimized Support Vector Regression for Natural Gas Price Forecasting
Abstract
Natural gas is a strategic energy commodity exhibiting nonlinear and highly volatile price movements due to supply-demand fluctuations, market dynamics, and geopolitical influences. These factors complicate accurate forecasting and necessitate advanced methods capable of modeling complex data patterns. This study proposes a hybrid forecasting model that integrates Empirical Mode Decomposition (EMD), Support Vector Regression (SVR), and Particle Swarm Optimization (PSO) to predict natural gas prices and assess predictive performance. The analysis utilizes a dataset of 1,575 daily closing prices from January 2020 to December 2025. EMD decomposes the original time series into seven Intrinsic Mode Functions (IMFs) and one residual component. Each component is modeled using SVR with a Radial Basis Function (RBF) kernel, and PSO is used to optimize model parameters. Forecasting performance is evaluated using Mean Absolute Percentage Error (MAPE) across three data partitioning schemes. Results indicate that the 70:15:15 partition yields the most accurate model, achieving a MAPE of 2.2641%. The 90-day forecast projects a gradual decline in natural gas prices after a peak in mid-January 2026, followed by relative price stability through March 2026. These findings suggest that the hybrid EMD-SVR-PSO model effectively captures the nonlinear dynamics of natural gas price data and delivers accurate forecasts, positioning it as a valuable decision-support tool for policymakers, industry stakeholders, and investors.