Probabilistic Assessment of Power Systems with Renewable Energy Sources based on an Improved Analytical Approach

Abstract

The increasing penetration of renewable energy sources has introduced great uncertainties and challenges into computation and analysis of electric power systems. To deal with uncertainties, probabilistic approaches need to be used. In this paper, we propose a new framework for probabilistic assessment of power systems taking into account uncertainties from input random variables such as load demands and renewable energy sources. It is based on the cumulant-based Probabilistic Power Flow (PPF) in combination with an improved clustering technique. The improved clustering technique is also developed in this study by making use of Principal Component Analysis (PCA) and Particle Swarm Optimization (PSO) to reduce the range of variation in the input data, thus increasing the accuracy of the traditional cumulant-based PPF (TCPPF) method. In addition, thanks to adopting PCA for dimensionality reduction, the improved clustering technique can be effectively dealt with a large number of input random variables so that the proposed framework for probabilistic assessment can be applied for large power systems. The IEEE-118 bus test system is modified by adding five wind and eight solar photovoltaic power plants to examine the proposed method. Uncertainties from input random variables are represented by appropriate probabilistic models. Extensive testing on the test system indicates good performance of the proposed approach in comparison to the traditional cumulant-based PPF and Monte Carlo simulation. The IEEE-118 bus test system is modified by adding five wind and eight solar photovoltaic power plants to examine the proposed method. Extensive testing on the test system, using Matlab (R2015a) on an Intel Core i5 CPU 2.53 GHz/4.00 GB RAM PC, indicates good performance of the proposed approach (PPPF) in comparison to the TCPPF and Monte Carlo simulation (MCS): In terms of computation time, PPPF needs 4.54 seconds, while TCPPF and MCS require 2.63 and 251 seconds, respectively; ARMS errors are calculated for methods using benchmark MCS and their values clearly demonstrate the higher accuracy of PPPF in estimating probability distributions compared to TCPPF, i.e., the maximum (Max) and mean (Mean) values of ARMS errors of all output random variables are: ARMSPPPFmax = 0.11%, ARMSTCPPFmax = 0.55%, and ARMSPPPFmean = 0.06%, ARMSTCPPFmean = 0.35%.