Comparison of Different Forecasting Techniques for Microgrid Load Based on Historical Load and Meteorological Data

Microgrids (MGs) are structures that provide electrical energy to the loads with Distributed Energy Resources (DERs), energy storage systems and control mechanisms. An MG can be operated either grid-tied or islanded mode. However, DERs, which are generally consist of renewable resources, may have difficulty in providing uninterrupted energy due to environmental dependencies in energy production. Considering the necessity of voltage and frequency synchronization for the systems connected to the grid and the need for uninterrupted energy to the loads for systems separate from the grid, it is seen that the energy production in the MG should be done in a planned manner. Therefore, load and renewable energy sources forecasting is especially important for MGs. It can be used to plan the operation of generation units with short-term load forecast, as well as adding extra generation units or determine the contract details with medium and long-term forecast. In this study, the load profile of an MG was forecasted using historical load, weather and calendar parameters. Forecast outputs for Linear Regression (LR), Regression Tree (RT), Support Vector Regression (SVR), Gaussian Process Regression (GPR) and Artificial Neural Network (ANN) methods were evaluated by performance metrics and the most suitable algorithm for this data set was tried to be found. As a result of the estimations made for the next year in the model trained on seven years of previous data, it was observed that the Mean Absolute Percent Error (MAPE) value of the ANN method fell below 3% for this model.