Carbon-Aware Optimal Power Flow via Neural Network with Emission Alignment

This paper develops a carbon-aware optimal power flow (CA-OPF) framework that integrates neural network-predicted carbon emission flows as mixed-integer linear programming (MILP) constraints. We first train a neural network to capture the carbon allocation process. This model is then transformed into MILP constraints and integrated into the OPF framework. Additionally, we introduce a system-wide emission alignment to maintain physical consistency between predicted and actual emissions. The resulting framework provides accurate spatial carbon intensity predictions while enabling direct operational control for emission reduction. Simulation results on the IEEE 30-bus system demonstrate the approach’s effectiveness in maintaining carbon intensity prediction accuracy across diverse operating conditions. By bridging carbon accounting with actionable control, this work offers a practical tool for power system decarbonization that preserves both computational efficiency and operational reliability.