Artificial Load Profiles and PV Generation in Renewable Energy Communities Using Generative Adversarial Networks

Understanding the electrical behavior of consumption and generation in residential areas featuring distributed photovoltaic generation is an important asset for both distribution companies and communities. This analysis is often in statistical terms. However, load and generation data can be noisy, sparse and irregular, with resulting difficulties in the following statistical analysis. A generative adversarial machine learning approach can be used to create artificial data with meaningful stochastic properties. This work focuses on the use of DCWGAN to create different profiles taken from multiple data clusters of residential sectors. This approach successfully extracts the statistical properties of the dataset and creates profiles based each on cluster with a given randomness. The implemented neural system is a powerful tool that can be used to create datasets useful for power flow analysis, optimal grid management and economic revenue simulation.

Automated summary

The study utilizes a generative adversarial machine learning approach to generate artificial data with meaningful stochastic properties, tackling the challenges in statistical analysis. The method successfully extracts statistical properties of the dataset and creates diverse profiles based on given randomness for each cluster.