Prediction of China’s Polysilicon Prices: A Combination Model Based on Variational Mode Decomposition, Sparrow Search Algorithm and Long Short-Term Memory

Abstract

Given the non-stationarity, nonlinearity, and high complexity of polysilicon prices in the photovoltaic (PV) industry chain, this paper introduces upstream and downstream material prices of the PV industry chain and macroeconomic indicators as influencing factors. The VMD–SSA–LSTM combination model is constructed to predict polysilicon prices, which is based on Variational Mode Decomposition (VMD) and utilizes the Sparrow Search Algorithm (SSA) to optimize the Long Short-Term Memory (LSTM) network. The research shows that decomposing the original polysilicon time series using the VMD algorithm effectively extracts the main features of polysilicon price data, reducing data instability. By optimizing the learning rate, hidden layer nodes, and regularization coefficients of the LSTM model using the Sparrow Search Algorithm, the model achieves higher convergence accuracy. Compared to the traditional LSTM model and VMD–LSTM model, the VMD–SSA–LSTM model exhibits the smallest error and the highest goodness-of-fit on the polysilicon dataset, demonstrating higher predictive accuracy for polysilicon prices, which provides more accurate reference data for market analysis and pricing decisions of the polysilicon industry.

Keywords

• polysilicon price prediction
• Variational Mode Decomposition
• Sparrow Search Algorithm
• Long Short-Term Memory Network