Forecasting Uncertainty Parameters of Virtual Power Plants Using Decision Tree Algorithm

Abstract

Renewable energy sources (RESs) are increasingly used to meet the world’s growing electrical needs, especially for the economic benefits and environmental problems associated with fossil fuel use. Small-scale renewable energy sources, controllable loads, energy storage devices, and other nonrenewable sources are effectively integrated to form a virtual power plant (VPP). Uncertainty in forecasting renewable energy generation due to the intermittent nature of renewable energy sources is one of the biggest challenges in VPPs. Power generation by RESs changes with the day of the week, season, location, climate, and resource availability. Also, load demand and utility price vary with time and need to be forecasted for proper energy management of VPPs. However, the dispatching and planning of VPPs are significantly impacted by the volatile nature of RESs, load demand, and utility price. Predicting these uncertainties with high accuracy is essential to balance the electrical power generation and the load demand. In this article, a decision tree (DT) algorithm is proposed, to predict the uncertainty parameters, such as the day-ahead power from the RES, load demand, and utility prices of VPPs. The efficiency of the proposed model and the predicted results are compared with other complex models, such as the artificial neural network (ANN) and auto-regressive integrated moving average (ARIMA) algorithms. Root-mean square error (RMSE), mean square error (MSE), coefficient of determination (R 2), and mean absolute error (MAE) are the statistical metrics used to evaluate the accuracy of the prediction. One-year meteorological data of the Chennai zone in India is considered for predicting the uncertainty parameters. IEEE 16-bus and 33-bus test systems are used to validate the forecasting model. It is evident from the results that the proposed DT algorithm can predict the uncertainty parameters more accurately and use lesser time than the ANN and ARIMA algorithms.