Adaptive Convolutional Residual Network for Dual-Task Forecasting in Energy Market Planning

Accurate forecasting and strategic decision-making are critical for electricity market planning in the domain of energy informatics, where grid reliability, economic stability, and sustainability must be balanced. This paper introduces the Adaptive Convolutional Residual Network (ACRN), a dual-task deep learning architecture designed to jointly perform project risk classification and electricity price regression. The model is trained and evaluated on a real-world dataset from Frankfurt, Germany, comprising hourly records from 2018 to 2024 and including 12 core variables such as electricity price, market demand, renewable generation, and regulatory policies. ACRN integrates novel preprocessing techniques including weighted temporal interpolation, dynamic thresholding, and hybrid normalization, along with adaptive feature refinement and context-aware feature derivation. The framework achieves a classification accuracy of 98.5%, F1- score of 98.1%, AUC of 99.0%, and regression MAPE of 2.33% while reducing computational cost by 10% compared to baseline models. In comparison with state-of-the-art methods such as EfficientNet, WideResNet, and Gradient Boosting, ACRN outperforms all in both predictive accuracy and time efficiency. These results demonstrate ACRN’s robustness and scalability in addressing the multidimensional forecasting requirements of modern power markets. The proposed model offers a self-contained, high-performance solution for energy market planning with practical relevance for both policy and industry applications.