(A) study on the energy analysis based intelligent arc-fault detection method in photovoltaic system

Abstract

Global energy consumption based on fossil fuels has increased considerably with rapid economic growth over the past decade. To mitigate accompanying problems related to climate change, new technologies and renewable energy resources that can replace existing fossil fuels are emerging. In particular, solar power and energy storage systems (ESSs) are being rapidly developed as renewable energy resources. Moreover, with technological improvements and the dissemination of renewable energy resources, technology related to the health management of such energy systems has garnered significant academic attention. In this context, this study focuses on the arc-fault phenomenon induced by fires, which are frequently occurrences in renewable energy architectures. The analysis of this phenomenon is important as it reduces the lifetime of system components. First, a novel proactive arc-fault detection method is proposed based on artificial intelligence (AI). Moreover, the arc-fault phenomena are discussed, and energy analysis is conducted via arc-fault experiments on commercial devices in a photovoltaic (PV) system using series DC arc-fault generators based on UL 1699 B. A transfer learning-based arc-fault detection method using a two-stage training technique is proposed to detect series DC arc-faults proactively by considering low-energy arc-fault characteristics. A one-layer long short-term memory network is combined with a lightweight one-dimensional convolutional neural network to detect arc-faults based on solely the measured current information. The results of offline and online experiments conducted using a commercial grid-connected PV inverter indicate that the proposed method outperforms alternative real-time methods operating on a single-board computer. Subsequently, a model transfer learning-based arc-fault energy estimation methods are proposed based on the aforementioned analysis. Moreover, a risk assessment method is proposed using AI-based arc-fault energy estimation to analyze and interpret the arc-fault level directly. Further, the superiority of the proposed method is verified by comparing its performance with that of existing general deep learning methods. Offline experiments are performed by applying the proposed method to PV systems based on an arc-fault estimation platform developed using commercial devices. By directly estimating arc-fault energy levels, the proposed methods provide useful information on risk assessment, enabling users to develop an efficient health management system. Finally, the representative proposed method is utilized to assess risks in ESS, and its performance is verified—it was successfully generalized for arc-fault diagnosis in various distributed energy resources. In summary, an energy analysis-based intelligent detection and estimation methods for arc-fault risk assessment in PV systems are proposed in this study. In addition, AI-based arc-fault energy estimation and risk assessment are performed on an ESS, a representative renewable energy source, to evaluate the performance of the proposed methods. The experimental results corroborate the significant contribution of the proposed methods to the fault diagnosis and health management of distributed renewable energy resources.