FOPF Neural Network: A Flexible Neural Network Framework Addressing Renewable Energy Uncertainty

This paper introduces a Flexible Optimal Power Flow (FOPF) neural network to address renewable energy uncertainties in power systems. Unlike conventional OPF neural networks that ignore renewable fluctuations, FOPF innovatively embeds either stochastic programming or robust optimization constraints within its architecture. The framework combines physical power system constraints with deep learning model, enabling simultaneous optimization of operational feasibility and computational efficiency. This neural network-based constraint encoding approach establishes a new paradigm for renewable-integrated power flow optimization, effectively bridging theoretical rigor with practical implementation needs. Experiments on IEEE 14-bus and 118-bus systems show the robust FOPF variant maintains 100% feasibility across all uncertainty levels, while the stochastic version achieves 81.1–99.3% feasibility with controlled performance degradation. Results demonstrate the robust FOPF's superiority in high-uncertainty scenarios and the stochastic FOPF's efficiency in moderate conditions.