Recurrent neural networks in distribution network planning: enhanced decision-making under uncertainty

The electrical distribution networks are rapidly growing due to the integration of new consumer profiles and distributed generation, creating challenges in predicting the fluctuating demand. This paper presents a distribution network planning method to address these challenges, which considers a Recurrent Neural Network (RNN) model, combined with a confidence interval threshold, to forecast time-series demand scenarios while considering the aggregated PV generation from smart meters. A case study in a radial medium voltage distribution network in Spain was used to test the proposed method, ensuring the operational thermal limits of transformers. This validation confirms its practical applicability in real-world scenarios, essential for maintaining network integrity and efficiency.