Advanced Neural Network with Optimized Training Parameters Based on the Cuckoo Search for Detecting and Classifying Faults in PV Power Systems

When PV power systems become primary instead of backup or complementary systems, fault detection is crucial to avoid system degradation and failure. The main challenges in current methods for detecting faults in PV power systems are accuracy of the results, speed of detection and classification of the fault. This paper presents a novel approach to optimizing training parameters for Artificial Neural Networks. Using the Cuckoo Search algorithm (CSA). CSA has been selected from dozens of optimization algorithms based on a deep survey and comparisons which showed definite advantages for CSA, including robustness, platform independence, fewer parameters, and ease of implementation. Optimizing the training parameters of the ANN accelerated the convergence and boosted the accuracy of the neural network. As a result, significantly faster convergence is realized with increased accuracy in fault detection and classification. Improvements were obtained in all the test scenarios including electrical and shading related faults. The paper also presents a comparison between the proposed method and existing optimization technique, the genetic algorithm (GA), for validation. Resulting comparisons demonstrate the advances achieved by the proposed methodology highlighting significant improvements in both accuracy and speed of convergence.