Although deep learning methods have demonstrated promising performance in photovoltaic (PV) fault diagnosis, the black-box nature of deep learning models poses challenges. This means that conventional models cannot provide trustworthy information for reasonable explanations in diagnosing PV faults. At the same time, when faced with the pattern characteristics of unknown faults, the model is prone to confusion, which can lead to misdiagnosis. To address these challenges, this paper proposes a PV fault diagnosis model that combines an interpretable uncertainty quantification with class incremental learning, named the Class Incremental Uncertainty Fusion Network (CIUFN). This model not only achieves precise diagnosis for PV fault types but also quantifies and analyzes the uncertainty in diagnostic results, thereby providing measurable indicators for the credibility of the model. Meanwhile, taking the uncertainty information as an entry point, we further studied the incremental learning mechanism for unknown faults to achieve accurate identification of unknown fault types. Experimental results demonstrate that the CIUFN model exhibits excellent diagnostic accuracy and interpretability across various PV fault scenarios and maintains outstanding accuracy even when the number of PV fault categories increases.

Soiling can reduce the output power and work efficiency of photovoltaic (PV) modules, causing serious economic losses to PV systems. The cleaning schedules can be optimized to save economic expenses through the methods capable of estimating the power loss of PV modules resulting from soiling. This paper proposes a deep learning framework that combines visible light and infrared image information with dual branch cross-modality feature fusion. Initially, the MobileNetV2 is applied as the backbone of the dual branch framework to enhance the training efficiency and reduce the computational complexity. Subsequently, a cross-modality differential aware fusion module based on the channel attention mechanism (CA-CMDAF) is introduced to improve the crossmodality feature fusion capability of the model. Moreover, a multi-cascade and cross-modality fusion network and a multi-scale fusion network are integrated to further facilitate the effectiveness of feature fusion and reduce the loss of visual details during the feature extraction. Lastly, extensive experiments are carried out on the multimodality dataset. The comparison results demonstrate the superior performance of the proposed dual branch network framework with the average accuracy of 88.27 %, which is higher than that of the single-modality models trained on either visible light or infrared images alone.

Recently, promising progresses have been made in photovoltaic (PV) arrays fault diagnosis (FD) due to the importance of operation and maintenance of PV power plants. However, PV arrays inevitably experience gradual degradation due to the complexity of operating conditions, resulting in domain shift of output data, which has a significant negative impact on the performance of FD. To address these problems, this study proposes a two-stage cross-domain, i.e., adaptive generative adversarial network deep learning approach for PV arrays FD under different degradation levels. In the first stage, the Normal data from the source domain (PV arrays without performance degradation) is utilized for training. Then, the Maximum Mean Discrepancy (MMD) loss is introduced to the fault generators in adversarial training to produce high-level feature representations of source domain fault data. In the second stage, identical training steps are used to guide the fault generators. Specifically, Normal data from the target domain i.e., PV arrays with performance degradation, is utilized to generate fault data features that are consistent with the target domain features. Then, the cross-domain adaptive FD model can be trained by using generated fault data features. The proposed model can not only learn the relationship from the different types of data, but also utilize target domain PV array data under healthy conditions to manually generate fake samples for cross-domain adaptive FD. Experimental results show that the Precision of the proposed model in the two tasks is 98.34% and 92.93 %, with Recall is 98.23 % and 94.13%, F1-Score is 0.9823 and 0.9274, all of which are better than those of the comparison models.

Photovoltaic (PV) power generation is widely utilized to satisfy the increasing energy demand due to its cleanness and inexhaustibility. Accurate PV power forecasting can improve the penetration of PV power in the grid. However, it is pretty challenging to predict PV power in short-term under precious future meteorological information absence conditions. To address this problem, this study proposes the hybrid Contrastive Learning and Temporal Convolutional Network (CL-TCN), and this forecasting approach consists of two parts, including model training and adaptive processes of forecasting models. In the model training stage, this forecasting method firstly trains 18 TCN models for 18 time points from 9:00 a.m. to 17:30 p.m. These TCN models are trained by only using historical PV power data samples, and each model is used to predict the next half-hour power output. The adaptive process of models means that, in a practical forecasting stage, PV power samples from historical data are firstly evaluated and scored by a CL based data scoring mechanism to search for the most similar data samples to current measured samples. Then these similar samples are further applied to training a single above-mentioned well-trained TCN model to improve its performance in forecasting the next half-hour PV power. The experimental results tested at the time resolution of 30 min demonstrate that the proposed approach has superior performance in forecasting accuracy not only in smooth PV power samples but also in fluctuating PV power samples. Moreover, the proposed CL based data scoring mechanism can filter useless data samples effectively accelerating the forecasting process.

针对光伏组件热斑若未及时发现处理,会严重影响光伏组件及阵列正常运行的问题,为了有效检测光伏阵列热斑,提出一种基于YOLOv5框架的深度学习热斑检测方法.首先,采用像素加权平均法融合红外和可见光图像作为检测对象,实现同时对光伏组件热斑和遮挡物的检测,并初步分析热斑成因.其次,改进模型框架,在轻量级网络MobileNetV3-large的基础上,融合坐标注意力机制,设计更轻量、更高效的MobileNetCA作为特征提取网络.然后,针对训练中正负样本数量极不平衡的情况,更换损失函数为变焦距损失函数,达到训练中突出正例的效果.同时,改进模型anchor box目标框生成算法,使生成的目标框与实际标注框更一致.实验结果表明,改进后的模型mAP为88.9％,较原YOLOv5s模型提升了3.8％,且模型参数量仅为原模型的48.6％.

How to find the abnormal points in data effectively and quickly and give a reasonable explanation is the main content of anomaly detection. The development of deep learning technology provides a new idea for the abnormal analysis and processing of voltage test data. This paper applies deep learning theory to the abnormal analysis and processing of voltage test data, and puts forward a model for the abnormal analysis and processing of voltage test data based on deep learning. Based on 3D CNN (convolutional neural network), the constructed time series voltage test data are classified, evaluated and analyzed. In this paper, the output of 3D CNN is as close as possible to the input, the voltage test data is taken as the input, the minimum reconstruction error is taken as the tuning standard in the training stage, and the network output is the voltage reconstruction data corresponding to the input. The research results show that the hidden levels 1, 2, 3 and 4 all show good classification accuracy, all reaching more than 90%. The proposed algorithm does perform well in different outlier ratios and has good robustness.

As photovoltaic (PV) arrays are exposed to the outdoors year-round, they are susceptible to various faults. The shading condition, degradation or dust coverage can make fault signals more complex, forming compound faults. These faults can lead to a large loss of power generation or irreversible damage to the PV modules, and even fires in severe cases. Moreover, unknown fault types that have never been seen in the training set may occur at actual working conditions. Therefore, accurate diagnosis of various types of single and compound faults (closed-set faults) by considering the identification of unknown faults, namely open-set faults diagnosis, is crucial to improve the efficiency of operation and maintenance. A 1D VoVNet-SVDD based open-set fault diagnosis model for PV arrays is proposed. The model is a two-stage network model consisting of a 1D VoVNet network and a multi-classification Support Vector Data Description (SVDD) in series. The 1D VoVNet network automatically extracts fault features from the input original I-V curve data. These extracted fault features are then combined with environmental parameters to construct the SVDD model. The SVDD identifies known fault types by con-structing a hypersphere for each fault type. Fault types that are not classified into any of the hyperspheres are considered as unknown faults, enabling open-set diagnosis. The experimental results show that the proposed model can accurately classify the closed-set faults among the three designed testing tasks while identify unknown type faults. The comparison demonstrates that the proposed algorithm is superior to the compared models.

Accurate faults diagnosis for photovoltaic (PV) array is one of the vital factors that guarantee the reliable operation of PV power plant. Artificial intelligence (AI) based fault detection and diagnosis (FDD) models are promising techniques. In order to automatically extract the faults features from the raw electrical data of PV array and create efficient FDD model with small dataset, a FDD scheme using Wasserstein generative adversarial network (WGAN) and convolutional neural network (CNN) is designed. The proposed FDD model is consisting of three modules, a discriminator, a generator and a classifier for fault diagnosis. By analyzing sequential PV data in a 2-Dimension way, the proposed discriminator and generator learn the distribution of PV data under various PV system operations. Then they are utilized to generate more labeled samples to improve the performance of the CNN based classifier. Thus, the proposed FDD model can be trained only requiring minor labeled samples. A laboratory grid-connected PV system is established to experimentally investigate the performance of the developed method. The results demonstrate that the designed FDD model can accurately diagnose line-line and open circuit faults.

The relationship between environmental factors with the power output of photovoltaic (PV) stations is unclear due to the non-linear characteristics of PV systems, which is challenging for PV power forecasting technology. To cope with these challenges, a hybrid forecasting approach called hybrid K-Means++ and Deep Neural Network with input and output adjusting structures (K-IAOA-DNN) is proposed to accurately predict PV power output. The proposed forecasting approach designs several features for K-Means++ to search for similar samples, reducing the complexity of the following forecasting model. Due to changeable environmental conditions and characteristics of PV systems like degradation, the prediction result of a forecasting model may deviate from the expected one. Therefore, IAOA-DNN model is developed by using a DNN adopting the two proposed structures e. g., the input adjusting structure and the output adjusting structure. The input adjusting structure uses several trainable parameters to determine the importance of each meteorological input for further feature extracting by DNN. The output structure in the prediction model is used for analyzing features extracted from DNN to generate a scalar factor fine-tuning the output of DNN. Additionally, the proposed method provides more accurate forecasting results with average RMSE, NRMSE, and MAE values of 14.43, 0.048, and 9.53 kW, respectively.

提出一种基于伪标签-1D DenseNet-KNN的光伏阵列故障诊断方法,实现在少标签样本下的光伏阵列复合故障开集识别.首先,分析各种常见单一故障和灰尘覆盖下复合故障的I-V特性曲线;然后,为克服常规半监督机器学习算法需手动提取数据特征的问题,采用一种伪标签与1D DenseNet相结合的半监督方法自动提取特征;最后,将从训练数据中提取的特征、训练数据预测的标签和测试样本提取的特征输入KNN算法并进行开集复合故障诊断.实验表明,该方法不仅能准确分类各种已知类别样本,还能识别出未知类别故障,且模型训练只需要少量的标签数据.

针对新建光伏发电站在光伏功率预测过程中因缺少训练数据导致预测精度较低和光伏发电功率的不稳定等问题,提出一种结合改进的深度卷积生成对抗网络(DCGAN)、注意力机制(Attention)和LSTM网络组合的光伏功率预测方法.首先,将DCGAN中生成器的特征提取网络由二维卷积改为一维卷积,对光伏数据进行扩充.其次,将Attention模块加入LSTM模块中,生成新的输入特征.最后,对新生成的LSTM模型进行功率预测,并采用澳大利亚沙漠知识太阳能中心(DKASC)Alice Springs电站的数据进行验证.实验结果表明,结合深层卷积生成的对抗网络与Attention-LSTM混合预测方法能有效提升预测精度.

Photovoltaic (PV) systems working outdoors are susceptible to various faults. The deposition of dust on the PV array may make these faults more complicated, resulting in a kind of compound faults. The similarity between compound faults and single faults leads to their misclassification. Therefore, accurately detection of potential PV array compound faults is essential to improve the operating efficiency and safety of PV systems. Addressing the above situation, this paper proposes a fault diagnosis (FD) scheme for PV array using a multi-scale SE-ResNet network. In addition, a multi-scale receptive field fusion module (MRFF) is designed to improve the diagnostic performance of the model. This model can automatically extract multi-scale fault features from input raw current–voltage curves data and environmental parameters. The single faults, partial shading conditions (PSCs), and compound faults under different degrees of dust covering can be diagnosed. In addition, the dust coverage degree is estimated simultaneously, which can provide a basis for developing a cleaning schedule. Simulation and experiment results demonstrate the superior performance of this method compared with other approaches, and also indicate the proposed model can be applied to the condition of dust accumulate on the bottom of the PV panels through transfer learning.

光伏阵列的局部遮阴会导致光伏功率损失.光伏阵列表面的遮挡物是造成局部阴影的因素之一,因此提出了一种基于CenterNet的光伏阵列遮盖物检测方法.通过迁移学习预训练模型,分步训练获得检测模型.该模型在不同光照方向、不同遮挡程度以及不同远近距离等条件下,能够较精确地识别遮盖物种类并定位.实验结果表明,该遮盖物检测模型的mAP为0.81,检测速度为16 ms,优于所选的对比算法,具有较强的实时性.

The effective fault diagnosis algorithm for the DC side photovoltaic (PV) array of a PV system (PVS) plays an important role in the operation efficiency and safety for PV power plants. But for fault diagnosis models it may fail to diagnose PV array (PVA) faults without detailed and quite fine fault features, especially line-line faults (LLF) occurring in the PVS that works under complex working conditions like low irradiance conditions and LLF with fault impedance. To address these challenges, this paper proposes a fault diagnosis scheme to diagnose different PVA faults using a proposed Dual-channel Convolutional Neural Network (DcCNN), which is able to automatically extract features and weight these features for fault classification. The important and fine features from the current and voltage electrical time series graph (ETSG) are extracted respectively by DcCNN in a double input way. Then, a proposed feature selection structure (FSS) is designed to improve the proposed fault diagnosis model capacity for diagnosing PVA faults under various conditions, including LLF, partial shading condition (PSC) and open circuit faults (OCF). Comparing to manually designed features, FSS not only helps DcCNN extract important features from PVA current and voltage automatically but also evaluates extracted features for further classification of DcCNN. Moreover, in the training stage, a proposed penalty is applied on DcCNN to constrain FSS, resulting in its sparse weight distribution. A comprehensive experiment based on a laboratory roof grid connected PVS is conducted. The results demonstrate the superior performance of the proposed approach compared with other algorithms as it can extract high-discriminative features from PVA current and voltage for different PVA faults, which is also effective on diagnosing LLF under low irradiance conditions and LLF with fault impedance.

Fault diagnosis of photovoltaic array plays an important role in operation and maintenance of PV power plant. The nonlinear characteristics of photovoltaic array and the Maximum Power Point Tracking technology in the inverter prevent conventional protection devices to trip under certain faults which reduces the system’s efficiency and increases the risks of fire hazards. In order to better diagnose photovoltaic array faults under Maximum Power Point Tracking conditions, the sequential data of transient in time domain under faults are analyzed and then applied as the input fault features in this work. Firstly, the sequential current and voltage of the photovoltaic array are transformed into a 2-Dimension electrical time series graph to visually represent the characteristics of sequential data. Secondly, a Convolutional Neural Network structure comprising nine convolutional layers, nine max-pooling layers, and a fully connected layer is proposed for the photovoltaic array fault diagnosis. The proposed model for photovoltaic array fault diagnosis integrates two main parts, namely the feature extraction and the classification. Thirdly, this model automatically extracts suitable features representation from raw electrical time series graph, which eliminates the need of using artificially established features of data and then employs for photovoltaic fault diagnosis. Moreover, the proposed Convolutional Neural Network based photovoltaic array fault diagnosis method only takes the array of voltage and current of the photovoltaic array as the input features and the reference panels used for normalization. The proposed approach of photovoltaic array fault diagnosis achieved over 99% average accuracy when applied to the case studies. The comparisons of the experimental results demonstrate that the proposed method is both effective and reliable.