A hybrid machine learning and meta-heuristic algorithm based service restoration scheme for radial power distribution system

摘要

In-service Restoration (SR), the healthy section of the feeder can be re-energized by finding the optimal path for power flow. Through conventional methods which are mainly deterministic in nature, the computational burden is very high. Therefore, researchers have proposed various meta-heuristic based methods to solve the SR problem. But since, these methods are probability based; one single algorithm cannot guarantee optimal solution for all scenarios. Hence, the authors have proposed a Machine Learning (ML) based framework, which can predict the best SR scheme for a particular fault scenario among the SR solutions obtained through various meta-heuristic algorithms. The supervised ML model is developed using the fault features as input values and the best performing meta-heuristic algorithm as the target value. To check the validity of the ML framework, the authors have taken four different meta-heuristic algorithms, which are, Enhanced Integer Coded Particle Swarm Optimization (EICPSO), Shuffled Frog Leaping Algorithm (SFLA), Non-Dominated Sorting Genetic Algorithm-II (NSGA-II), and Ant Colony Optimization (ACO) algorithm. The ML model can be extended for any number of algorithms. Multi-class Support Vector Machine (SVM) algorithm is used as the estimator in the current work to train and test the developed model. The results obtained through SVM are 95.95% accuracy, 94% precision, and 94% f1-score. The performance of SVM is better when compared with other state-of-the-art ML estimators namely K-Nearest Neighbor (KNN), Random Forest (RF), and Logistic Regression (LR). The experiment is conducted for IEEE 33 bus and IEEE 69 bus test systems. A total of 320 fault instances are created using Power System Computer-Aided Design (PSCAD) software. The features of these fault points are used as input for the ML model which is extracted using a discrete wavelet transform. This study is conducted for the balanced radial test system.