Smart Reserve Planning Using Machine Learning Methods in Power Systems with Renewable Energy Sources

Abstract

Estimation of the power obtained from intermittent renewable energy sources (IRESs) is an important issue for the integration of these power plants into the power system. In this study, the expected power not served (EPNS) formula, a reliability criterion for power systems, is developed with a new method that takes into consideration the power generated from IRESs and the consumed power (CP) estimation errors. In the proposed method, CP, generated wind power (GWP), and generated solar power (GSP) predictions made with machine learning methods are included in the EPNS formulation. The most accurate prediction results were obtained with the Multi Layer Perceptron (MLP), Long-Short Term Memory (LSTM), and Convolutional Neural Network (CNN) algorithms used for prediction, and these results were compared. Using different forecasting methods, the relation between forecast accuracy, reserve requirement, and total cost was examined. Reliability, smart reserve planning (SRP), and total cost analysis for power systems were carried out with the CNN algorithm, which provides the most successful prediction result among the prediction algorithms used. The effect of increasing the limit EPNS value allowed by the power system operator, that is, reducing the system reliability, on the reserve requirement and total cost has been revealed. This study provides a useful proposal for the integration of IRESs, such as solar and wind power plants, into power systems.