부산대학교 도서관

Predicting the output power and detecting the faults on the solar photovoltaic (PV) systems are the challenging tasks in the modern decades. The conventional works are highly focused on developing an optimization and classification methodologies for accomplishing PV output power forecasting and fault prediction separately. But, this paper focused on solving a multi-objective research problem by implementing an innovative optimization and classification methodologies. For obtaining the maximum power yield from the solar PV panels, the maximum peak point tracking controlling technique is used in this system. In the first stage of work, the PV output power prediction is performed by deploying an endowed multi-verse optimization (EMVO) methodology. This technique intends to improve the convergence rate of optimization by estimating the best fitness value. For computing the best fitness value, the nonlinear inertial weight has been estimated with respect to the distance near the optimal universe. Also, the selection operation is performed to obtain the survival of the fittest value based on the crossover probability and random dimension value. Based on this, the power prediction can be performed with the operations of crossover, mutation and selection. In the second stage of work, the fault detection is mainly concentrated to ensure the normal operations on the PV system. For this purpose, an augmented radial basis functional network (ARBFN) technique is implemented for exact fault detection. In the proposed work, the EMVO and ARBFN methodologies are mainly used for accomplishing the objectives of PV output power forecasting and fault prediction at earlier stage. Here, the set of wavelet patterns have been extracted and used for training the classifier, which improves the overall accuracy rate of fault classification. Moreover, the experimental results evaluate the performance of both power prediction and fault detection techniques by using various measures. By using the proposed EMVO–ARBFN mechanisms, the absolute error value is reduced up to 0.2%, relative error is reduced to 3%, accuracy of prediction is increased to 98%, and fault mis-classification rate is minimized to 2.5%. Also, the obtained results have been compared with the existing approaches for proving the superiority of the proposed system, which includes the pre-turn method, back propagated neural network, artificial neural network, support vector machine, support vector regression, and other integrated machine learning models.