As a cornerstone of the national economy, the energy sector plays a critical role in ensuring energy security and advancing the Dual Carbon goals. However, existing forecasting models often fail to capture systematic information in regression residuals and neglect the stochastic nature of optimization algorithms, raising concerns about reproducibility. To fill these gaps, this study develops a novel hybrid modeling framework that integrates Kernel Ridge Regression (KRLS) with Support Vector Regression (SVR) optimized by the Fruit Fly Optimization Algorithm (FOA) and its quantum-enhanced variant (QFOA). A two-stage “baseline fitting plus residual correction” paradigm is established to extract implicit information from residuals. All experiments are implemented under a rigorous reproducibility protocol with a fixed random seed and 30 repeated runs. Based on 300 provincial observations from 2014 to 2023, the empirical results show that QFOA-SVR + KRLS delivers the best performance for electricity consumption (Y1), while FOA-SVR + KRLS achieves a higher R2 for energy intensity (Y2). The models also capture significant provincial heterogeneity, with eastern regions showing lower prediction errors than western and northern areas. This study provides a reproducible and regionally adaptive foundation for formulating differentiated energy policies in China.