Research on Distribution Network Material Price Forecasting Based on Machine Learning

Abstract

To tackle price volatility in distribution network material procurement and improve precision in setting reference prices, this study introduces two hybrid machine learning methods for price forecasting. Materials are first classified using the ABC method. CNN-SVM and TCN-GRU hybrid models are then developed, incorporating historical procurement data and market factors. Prediction is optimized through error analysis and a model selection mechanism. Results show that the models effectively capture nonlinear and temporal patterns, with a mean absolute deviation consistently below 3.5% across material categories. The proposed "classification and optimized selection" framework supports cost-effective procurement and digital transformation.