Energy Management in Grid Connected Photovoltaic Systems using Neural Networks

Abstract

The increasing use of photovoltaic generation and battery storage systems brings challenges in controlling the energy flow variables, in particular the reference current (𝐼𝑟𝑒𝑓). This study is focused on developing a Feedforward Artificial Neural Network (ANN) model to estimate 𝐼𝑟𝑒𝑓 based on input variables such as State of Charge (SOC) and Photovoltaic (PV) power. The methodological approach includes creating a dataset of 5,000 samples within operational limits and dividing it into three subsets: 70% for training, 15% for validation, and 15% for testing. The ANN training process features two hidden layers with 10 neurons each and uses the algorithm to minimize the Mean Squared Error (MSE). The results show high accuracy with a correlation coefficient close to 1 (𝑅 ≈ 0.9974) and low errors, with an MSE of approximately 0.054479. The error distribution, which shows a concentration around 0 with low dispersion, also corroborates the model’s robustness. These findings confirm its effectiveness in energy management systems.