Machine Learning-Driven Peer-to-Peer Energy Trading Within Active Distribution Networks

Peer-to-peer (P2P) energy trading within active distribution networks offers a promising opportunity for prosumers and consumers to trade energy locally and gain financial benefits. This market-driven approach improves local energy flexibility and encourages the use of small-scale distributed energy resources. However, real-time operation and scalability of P2P markets are among the challenges due to the large number of participants as well as the uncertainty of peers' generation and consumption. In this paper, machine learning (ML)-driven approach is proposed to address these issues while addressing the P2P market and networks' constraints. Various ML methods are applied to the P2P integrated distribution system optimization model, and the method with the lowest error is selected as the candidate algorithm, and then hyperparameters are tuned to effectively balance bias and variance, particularly important for larger datasets where overfitting risks can escalate without proper regularization. The trained ML model is computationally efficient and fast enough to support real-time optimal P2P market operations while considering network physical constraints and additional network usage charges like power loss costs, voltage regulation, and congestion costs. The results demonstrate the effectiveness of the proposed approach in achieving reliable, scalable, and efficient energy trading within active distribution networks.