Fault detection and classification in solar based distribution systems in the presence of deep learning and social spider method

• Published in: Solar energy 2023-09, Vol.262, p.111868, Article 111868
• Main Authors: Cao, Hanhua, Zhang, Huanping, Gu, Changle, Zhou, Yuhuai, He, Xiu
• Format: Article
• Language: English
• Subjects: Digital twin; Fault classification and detection; Generative adversarial networks (GANs); Social spider optimization algorithm; Solar based distribution systems
• ISSN: 0038-092X; 1471-1257
• DOI: 10.1016/j.solener.2023.111868

•Introducing a novel deep learning based model for fault detection and classification (FDC) in solar-based distribution systems, which utilizes the generative adversarial networks (GANs) for accurate and precise classification of faults.•Utilizing GANs in FDC due to their ability to generate realistic data from a given input, detect and classify faults that may not be visible to the naked eye, and learn from the data they generate, allowing for continuous improvement in accuracy over time.•Suggesting the use of the digital twin of the distribution system for recording synchronous data, as it allows for more efficient and effective data collection.•Proposing the social spider optimization (SSO) algorithm to optimize the search space of the generator and discriminator networks in GANs, improving the accuracy and performance of the FDC system.•Demonstrating the effectiveness of the proposed method using real-world datasets, showcasing its potential as a reliable and efficient solution for FDC in solar-based distribution systems. This research proposes an intelligent method for fault detection and classification (FDC) in solar based distribution systems using Generative Adversarial Networks (GANs) and Social Spider method. The suggested method is constructed using the combination of GANs and Social Spider method, which is a hybrid system to increase its capability for the classification purposes. The GANs model is used to detect fault signatures in the system, while the Improved Social Spider method is used to reinforce its training process. The proposed method is evaluated on the big data gathered using the digital twin of a solar based distribution system for different situations of operation. The results show that GANs can detect fault signatures with high accuracy and the Improved Social Spider method can classify the fault types with high accuracy. The proposed method compared with other existing methods and the results show that the suggested method outperforms the existing methods considering accuracy, recall, precision and speed. The proposed method can be used for FDC in solar based distribution systems, and can be developed to other distribution systems.