Non-Ferrous Metal Price Point and Interval Prediction Based on Variational Mode Decomposition and Optimized LSTM Network

Abstract

The accurate forecasting of metal prices is of great importance to industrial producers as the supply of metal raw materials is a very important part of industrial production. The futures market is subject to many factors, and metal prices are highly volatile. In the past, most of the relevant research has focused only on deterministic point forecasting, with less research performed on interval uncertainty forecasting. Therefore, this paper proposes a novel forecasting model that combines point forecasting and interval forecasting. First, a novel hybrid price point forecasting model was established using Variational Modal Decomposition (VMD) and a Long Short-Term Memory Neural Network (LSTM) based on Sparrow Search Algorithm (SSA) optimization. Then, five distribution functions based on the optimization algorithm were used to fit the time series data patterns and analyze the metal price characteristics, Finally, based on the optimal distribution function and point forecasting results, the forecasting range and confidence level were set to determine the interval forecasting model. The interval forecasting model was validated by inputting the price data of copper and aluminum into the model and obtaining the interval forecasting results. The validation results show that the proposed hybrid forecasting model not only outperforms other comparative models in terms of forecasting accuracy, but also has a better performance in forecasting sharp fluctuations and data peaks, which can provide a more valuable reference for producers and investors.