Predicting Prosumer Energy Patterns and Minimizing Imbalance Cost

Prosumers are the individual households and businesses that produce and consume energy. Due to the unpredictive nature of prosumers, it can be challenging to predict their energy patterns. This can lead to energy imbalance. Energy imbalance is a situation where the amount of energy that is expected to be consumed or produced does not align with the actual amount of energy used or generated. Energy imbalance is a critical issue for power grid operators as it can lead to operational challenges, inefficiencies, and increased costs. To tackle this problem, this study explores possibilities of predicting energy consumption and production of prosumers using machine learning models. Several models were developed to predict the amount of electricity produced and consumed by households and businesses in Estonia that have working solar panels. Data about weather conditions, relevant energy prices and past energy consumptions is available from October 2021 to April 2023. Prediction is made for the period between January 2023 to April 2023. Mean absolute error was used as a metric for this problem. Three models were tested: persistence algorithm, random forest and XGBoost. Naive solution in the form of persistence algorithm provides baseline result of 164.86. Random forest with a random search algorithm for hyperparametrs provides result of 105.31 for the test dataset. However, XGBoost showed the best performance among these three models. It achieved a MAE score of 76.45 on the test dataset. It was demonstrated that gradient tree boosting models, specifically XGBoost, are valuable for predicting energy patterns of prosumers.