Wenjun Yan's research while affiliated with Zhejiang University and other places

The ‘mismatch losses’ problem is commonly encountered in distributed photovoltaic (PV) power generation systems. It can directly reduce power generation. Hence, PV array reconfiguration techniques have become highly popular to minimize the mismatch losses. In this paper, a dynamical array reconfiguration method for Total-Cross-Ties (TCT) and Series–Parallel (SP) interconnected PV arrays is proposed. The method aims to improve the maximum power output generation of a distributed PV array in different mismatch conditions through a set of inverters and a switching matrix that is controlled by a dynamic and scalable reconfiguration optimization algorithm. The structures of the switching matrix for both TCT-based and SP-based PV arrays are designed to enable flexible alteration of the electrical connections between PV strings and inverters. Also, the proposed reconfiguration solution is scalable, because the size of the switching matrix deployed in the proposed solution is only determined by the numbers of the PV strings and the inverters, and is not related to the number of PV modules in a string. The performance of the proposed method is assessed for PV arrays with both TCT and SP interconnections in different mismatch conditions, including different partial shading and random PV module failure. The average optimization time for TCT and SP interconnected PV arrays is 0.02 and 3 s, respectively. The effectiveness of the proposed dynamical reconfiguration is confirmed, with the average maximum power generation improved by 8.56% for the TCT-based PV array and 6.43% for the SP-based PV array compared to a fixed topology scheme.

Electroluminescence (EL) is considered an efficient technique for the quality assessment of photovoltaic (PV) modules through observing the cell internal characteristics. Most cell defects exhibit the linear characteristics in the EL images that can gradually develop into fatal defects. However, such linear features can be hardly identified in the EL images of the polycrystalline silicon cells due to complex backgrounds. Numerous algorithmic solutions have been developed for module defect detection and analysis, but few results are reported for defects diagnosis of EL images. This paper proposes a deep learning-based automatic linear defects diagnosis solution for polycrystalline silicon photovoltaic cells based on EL images. The Hessian matrix-based defects feature extraction and a multi-scale line detector-based defect feature enhancement are adopted to achieve improved performance. A deep learning-based model for defects diagnosis is also proposed. The proposed solution is extensively evaluated through experiments against the existing machine learning models, i.e. Vgg16, ResNet50 and InceptionV3. The numerical results demonstrate the effectiveness and superiority of the proposed solution.

Linear defects detection of photovoltaic (PV) modules plays a key role in the health assessment in PV plants. However, the conventional defects diagnosis is mainly carried out manually and hence is inefficient or even infeasible in practice. With the recent advances in computing and communication technologies, different forms of edge computing facilities are designed with less human intervention and a non-destructive inspection in the industry. However, the assessment of PV plants using electroluminescence images faces some fundamental challenges. This paper proposed an automatic linear defects detection system for large-scale PV plants based on an edge-cloud computing framework. A novel deep learning-based PV defects detection algorithmic solution is developed considering the trade-off between detection performance and computational complexity through allocating the computing tasks across the edge devices, edge server and cloud server. The proposed solution is evaluated through experiments using real-world data and the numerical results demonstrate its effectiveness and accuracy of PV defect detection as well as the reduction of communication overhead by decreasing the size of EL images.

Given the huge installed capacity of photovoltaic (PV) worldwide, the traditional defect detection system for PV plants is infeasible, especially for large-scale plants. In this paper, unmanned aerial vehicles mounted with several sensors and a computer in the cloud are used cooperatively to establish an internet-of-things-based cloud-edge computing infrastructure, which can automatically detect defects with low latency, low cost, and high accuracy. The pre-trained models trained in the cloud server are embedded into the processor in unmanned aerial vehicles to implement online detection. Specifically, given the characteristic of defects in electroluminescence images, a two-stage algorithm is proposed to identify the defects with high performance. In the first stage, cells, the basic unit in the PV module, are extracted using an encoder-decoder network. Then, a vision-based incremental defect classification algorithm is proposed for defect detection that integrates deep learning with prior knowledge to maximize computing efficiency. The performance of the proposed system is evaluated through extensive experiments.

Defects detection with Electroluminescence (EL) image for photovoltaic (PV) module has become a standard test procedure during the process of production, installation, and operation of solar modules. There are some typical defects types, such as crack, finger interruption, that can be recognized with high accuracy. However, due to the complexity of EL images and the limitation of the dataset, it is hard to label all types of defects during the inspection process. The unknown or unlabeled create significant difficulties in the practical application of the automatic defects detection technique. To address the problem, we proposed an evolutionary algorithm combined with traditional image processing technology, deep learning, transfer learning, and deep clustering, which can recognize the unknown or unlabeled in the original dataset defects automatically along with the increasing of the dataset size. Specifically, we first propose a deep learning-based features extractor and defects classifier. Then, the unlabeled defects can be classified by the deep clustering algorithm and stored separately to update the original database without human intervention. When the number of unknown images reaches the preset values, transfer learning is introduced to train the classifier with the updated database. The fine-tuned model can detect new defects with high accuracy. Finally, numerical results confirm that the proposed solution can carry out efficient and accurate defect detection automatically using electroluminescence images.