No abstract

Predicting students at risk of academic failure is valuable for higher education institutions to improve student performance. During the pandemic, with the transition to compulsory distance learning in higher education, it has become even more important to identify these students and make instructional interventions to avoid leaving them behind. This goal can be achieved by new data mining techniques and machine learning methods. This study took both the synchronous and asynchronous activity characteristics of students into account to identify students at risk of academic failure during the pandemic. Additionally, this study proposes an optimal ensemble model predicting students at risk using a combination of relevant machine learning algorithms. Performances of over two thousand university students were predicted with an ensemble model in terms of gender, degree, number of downloaded lecture notes and course materials, total time spent in online sessions, number of attendances, and quiz score. Asynchronous learning activities were found more determinant than synchronous ones. The proposed ensemble model made a good prediction with a specificity of 90.34%. Thus, practitioners are suggested to monitor and organize training activities accordingly.

Özet-Fotovoltaik (FV) sistemler için doğru bir hata tespit yeteneği, işletme maliyetlerini ve bir arıza nedeniyle oluşabilecek devre dışı kalma sürelerini azaltarak FV sistemin verimliliğini artırabilir. Bu çalışmada, FV sistemler için bir hata tespit yöntemi önerilmiştir. Önerilen yöntem, topluluk öğrenmesi temelli bir modelin FV sistemlerdeki hataları sınıflandırmak amacıyla kullanılmasına dayanmaktadır. Topluluk öğrenmesi yöntemi, tek bir öğrenme algoritmasının genelleme yeteneğinin ve sağlamlığının üstüne çıkabilmek için farklı algoritmaların tahminlerini birleştirir. Bu çalışmada, sınıflandırma problemlerinde yaygın olarak kullanılan bazı öğrenme algoritmalarından bir topluluk öğrenmesi modeli oluşturulmuştur. Topluluk modeli, daha sonra parametre optimizasyonu uygulanarak geliştirilmiştir. Öğrenme algoritmalarının her biri ve bunları birleştiren topluluk modeli tahmin doğrulukları açısından karşılaştırılmıştır. Önerilen yöntem, Scikit-learn makine öğrenme kütüphanesi ile Python kullanılarak gerçekleştirilmiştir. Yöntemin deneysel geçerliliği Muğla'da (Türkiye) kurulu bir konut tipi FV sistemden elektriksel ve meteorolojik ölçüm verileri kullanılarak yapılmıştır. Sonuçlar, optimize edilmiş bir topluluk öğrenmesi modeliyle, önerilen yöntemin yalnızca sınıflandırma doğruluğunu geliştirmediğini, aynı zamanda fotovoltaik sistem hata tespiti için güçlü bir genelleme yeteneğine de sahip olduğunu göstermektedir. Anahtar Kelimeler-hata tespiti, sınıflandırma, topluluk öğrenmesiAbstract-An accurate fault detection capability for photovoltaic (PV) systems can improve PV system productivity by reducing operational costs and possible downtimes caused by a failure. In this paper, a fault detection method for PV systems is proposed. The proposed method is based on the use of an ensemble learning based model for classifying faults in PV systems. Ensemble learning combines the predictions of different algorithms in order to improve generalizability and robustness over a single learning algorithm. In this study, an ensemble learning model is built from some learning algorithms that commonly used in the classification problems. The ensemble model is then improved via parameter optimization. Each learning algorithms and the ensemble model that combines them are compared in terms of their prediction accuracy. The proposed method was implemented using Python with Scikit-learn machine learning library. The experimental validation of the method has been performed using electrical and meteorological measurements data from a residential PV system installed in Muğla (Turkey). Results show that, with an optimized ensemble learning model, the proposed method not only improves the classification accuracy but also has a strong generalization ability for PV system fault diagnosis.

Manufacturers generally share datasheet values of photovoltaic (PV) modules at only standard test conditions (STC). These conditions enable PV modules to generate high power but are rarely encountered in the real environment. Therefore, accurate modeling of PV modules is very important in terms of estimating the energy that can be obtained under all operating conditions. Many studies have been conducted in this field in the literature. In this study, a new method is proposed for the implementation of the commonly used five-parameter model. This new method uses a bisection search algorithm for calculating the value of the series resistance, which is one of the five parameters, and thus extracting the other parameters. The datasheet values provided by the manufacturers are sufficient for obtaining the series resistance and therefore other parameters. The accuracy of the method was first tested by comparing the datasheet values of the three different PV modules with the outputs of the proposed method. Finally, the simulation accuracy of the proposed method for different operating conditions was tested by comparing the real measurement data collected by the National Renewable Energy Laboratory (NREL) with the outputs of the method under the same operating conditions. The results show that the proposed method demonstrates good agreement with both datasheet values and real measurement data. The method offers a good balance of simplicity-accuracy.