To address the challenging nonlinear forecasting problem caused by multi-source disturbances in the load of Hebei Province’s ferrous metal smelting industry, this paper proposes a hybrid QRCNN-BiGRU prediction model incorporating quantile regression. The framework consists of three key components: (1) a multi-scale dilated convolutional module that extracts local mutation features and periodic components from load sequences, effectively capturing short-term fluctuations; (2) a bidirectional gated recurrent unit (BiGRU) that analyzes energy transmission in production processes and policy retroactive effects; and (3) a quantile regression layer that generates probabilistic prediction intervals. Validation using actual 2022 power grid data from Hebei Province demonstrates that the model achieves accurate load curve forecasting with excellent probabilistic prediction performance.