Harnessing the Power of Improved Deep Learning for Precise Building Material Price Predictions

Accurate forecasting of construction material prices is essential for effective cost control and risk management in construction projects. However, due to the influence of various complex factors, building material prices exhibit high nonlinearity and instability, making traditional prediction methods often inadequate for achieving optimal results. This study introduces an innovative prediction model, CEEMDAN-VMD-GRU-ARIMA, specifically designed for forecasting the price of prestressed steel bars. This model uniquely combines CEEMDAN and VMD to address nonlinear characteristics, and it innovatively incorporates sample entropy for the adaptive selection of either GRU or ARIMA for prediction. Additionally, a VMD decomposition mode number K value optimization method based on a sparse index is proposed. Experimental results demonstrate that the model performs exceptionally well, achieving an Adjusted R-squared value of 0.81, with various error indicators significantly surpassing those of the baseline model. This approach offers new insights for short-term price prediction of building materials and contributes to enhancing the economic benefits and management efficiency of construction projects.