DEEP CONVOLUTIONAL NEURAL NETWORK FOR AUTOMATIC DETECTION OF DAMAGED PHOTOVOLTAIC CELLS
DEEP CONVOLUTIONAL NEURAL NETWORK FOR AUTOMATIC DETECTION OF DAMAGED PHOTOVOLTAIC CELLS
R. Pierdicca1, E.S. Malinverni1, F. Piccinini1, M. Paolanti2, A. Felicetti2, P. Zingaretti2
1Universita´ Politecnica delle Marche, Dipartimento di Ingegneria Civile, Edile e dell’Architettura, 60100 Ancona, Italy (x.x. xxxxxxxxxx, x.xxxxxxxxx)@xxxxx.xxxxxx.xx
2Universita´ Politecnica delle Marche, Dipartimento di Ingegneria dell’Informazione, 60100 Ancona, Italy (x.xxxxxxxx, x.xxxxxxxxx)@xx.xxxxxx.xx, x.xxxxxxxxxx@xxxxx.xxxxxx.xx
Commission II, XX XX/4
KEY WORDS: UAV, monitoring/inspection, Anomaly Detection, Deep Learning, Photovoltaic Panels, Thermography, Classification
ABSTRACT:
The number of distributed Photovoltaic (PV) plants that produce electricity has been significantly increased, and issue of monitoring and maintaining a PV plant has become of great importance and involves many challenges as efficiency, reliability, safety, and stability. This paper presents the novel approach to estimate the PV cells degradations with DCNNs. While many studies have performed images classification, to the best of our knowledge, this is the first exploitation of data acquired with a drone equipped with a thermal infrared sensor. The experiments on “Photovoltaic images Dataset”, a collected dataset, are presented to show the degradation problem and comprehensively evaluate the method presented in this research. Results in terms of precision, recall and F1-score show the effectiveness and the suitability of the proposed approach.
1. INTRODUCTION
In the near future, a key challenge will be represented by the use of renewable energy sources (RES), whose application is in- creasing in demand due to the environmental and climatic chal- lenges and the recovery of global economy. Renewable energy sources will represent the only alternative to limit fossil fuel us- age and pollution. World leading energy outlooks from Interna- tional Energy Agency (IEA), U.S. Energy Information adminis- tration (EIA) and BP are showing the trends in the energy sector with significant increase in renewable energy conversion and uti- lization for the future (IEA, 2016), (EIA, 2016), (BPp.l.c, 2016).
The number of distributed Photovoltaic (PV) plants that produce electricity has been significantly increased, and most of the instal- lations are becoming decentralized. In this sector, in the last five years, Europe itself has experienced one of the largest growths: the electricity generation by RES in general, and PV in partic- ular. Thus, issue of monitoring and maintaining a PV plant has become of great importance and involves many challenges as ef- ficiency, reliability, safety, and stability (Xxxxxxxxx et al., 2015), (Xxxxxxxxx et al., 2015).
Monitoring the state of health of a system is crucial; in fact, detecting the degradation of solar panels is the only way to en- sure good performances over long periods of time (Xxxxxx et al., 2018). Beside avoiding waste of energy, the reason for maintain- ing a correct functional status of a plant is also economic. Indeed, the degradation of long-term performance and overall reliability of PV plants can drastically reduce expected revenues. PV plants are more and more extensive, composed by thousands of mod- ules, potentially affected by the following fault types at:
1. Module level: mismatches between modules, shading, glass breakage, busbar failure, diode failures, delamination, bro- xxx interconnects and hot-spots (Xxxxx et al., 2017).
2. Array level: optical degradation or (fault), electrical mis- matches, etc.
There are many methods proposed in the literature in the last decade regarding the identification of the damaged panel, and spans from electrical diagnostic, statistical inference from mon- itored control units, shading detection and so on (Xxxxx et al., 2007), (Xxxxxxxxx et al., 2014), (Xxxxxxx and Xxxxxxxxx, 2010). Al- though standard monitoring approaches proved to be effective, they are able only to ensure power losses detection in a portion of the PV field, while the accurate localization of faulty modules re- quires strings disassembling, visual and/or electrical inspection. The above-mentioned techniques, and in particular electrical di- agnosis, are time demanding, cause stops in the energy genera- tion, and often require laboratory instrumentation, thus resulting not cost effective for frequent inspections. Moreover, it should be noted that PV plants are often located in inaccessible places, making any intervention dangerous. Therefore, the safety of op- eration deeply impacts on the maintenance costs.
In this regard, a strong contribution was given by the recent diffu- sion of unmanned aerial vehicles (UAV), equipped with a thermal infrared sensor, making this technique widely accessible and af- fordable and becoming a de-facto standard for data acquisition to inspect the PV system to detect faults (Quater et al., 2014). Regardless the reason of the fault, malfunctioning PV cell has a higher temperature than adjacent normal cells; it is hence quite easy to detect them with a drone equipped with a thermal infrared sensor by drone. The challenge here is to expedite the process of detecting these anomalies, which could be very time consuming and depending on the visual interpretation of the operator. A cur- rent practice in fact, is to inspect frame by frame the video stream of the thermal camera.
The recent literature presents different approaches to inspect PV fields, as well as automatic tools for image processing and fault
detection and classification (Xxxxx et al., 2016). The main chal- lenges of these approaches are: positioning, individual module identification, defect detection and defect classification. Regard- ing these approaches, the problem of data interpretation still needs to be solved. One of the biggest obstacle is represented by the large number of pictures to be analyzed. Automatic data man- aging intelligent systems, exploiting advanced signal processing algorithms for computer vision, able to both identify single panel frames are needed, and might represent the winning solution in order to make the monitoring task more affordable and reliable. When dealing with large image collections analysis, Deep Con- volutional Neural Networks (DCNNs) are proved to be of the high-level task, compared to the widely used machine learning approach (e.g. Support Vector Machine (SVM), Neural Networks (NN), kNearest Neighbor (KNN), Decision Tree (DT), Random Forest (RF), Xxxx Xxxxx), and DCNN-based approaches outper- formed state-of-the-art methods in image classification. They pave the way towards the complete image understanding and they are surpassing other approaches by a large margin in terms of ac- curacy and sometimes even efficiency, discovering multiple levels of representation which can represent more abstract semantics of the data. In particular, the use of convolutional architectures is one key aspect of the deep learning success in image classifica- tion. The convolutional layers learn the feature representations of their input images, and this makes them the core block of DCNN. DCNNs have led to a series of breakthroughs for image classifi- cation by integrating low/mid/high-level features and classifiers in an end-to-end multi-layer. DCNNs can extract image features and classify them from a large number of image data automati- cally. Compared with feature extraction techniques (e.g., XXXX, HOG, GIST), these can make better performance and does not need hand designed image features.
Given the above and in line with recent research trends, this pa- per outlines the novel approach to estimate the PV cells degra- dations with DCNNs. While many researchs in literature have performed images classification, to the best of our knowledge, this is one of the first exploitation of data acquired with a drone equipped with a thermal infrared sensor. The experiments on “Photovoltaic images Dataset”, a newly collected dataset, are pre- sented to show the degradation problem and comprehensively evaluate the method presented in this research. The DCNN that we select is based on the VGG16 network architecture (Xxxxxxxx and Xxxxxxxxx, 2014), because it is the suitable solutions to im- age classification (Paolanti et al., 2017a), (Xxxxxxxx et al., 2017b), (Xxxxxxx et al., 2017). The dataset is comprised of a total amount
of 3336 images, captured with a Skyrobotic SR-SF6 drone, equipped with a Flir Tau 2 640 thermo camera. Images were acquired with nadir configuration, keeping an average flight height of 50 m. To obtain the ground truth of the collected pictures, the true cell degradation has been manually estimated by human annotators, thus providing a more precise and less noisy dataset.
To briefly summarize, the main contributions of this work are: (i) a demonstration that Deep Learning architectures can be applied for detection of PV cells damaged; (ii) a challenging new dataset of thermal images collected by drone equipped with a thermal in- frared sensor, hand-labelled with ground truth; (ii) performance comparison of different data collection for thermal image classi- fication.
Paper is organized as follows: Section 2 gives the description of the related work, Section 3 gives the insight to the method used in this analysis, Section4 gives the experimental results. Future works with the conclusions are presented in Section5.
2. RELATED WORK
In recent years, several authors have proposed the use of UAVs for inspection activities (Xxxxxxxx et al., 1999), (Xxxxxx and Xxxxxx, 2004). The current practice adopted by the majority of PV plants owner is to perform inspection sporadically, with random cri- teria and without controlling the overall health of the installa- tion. Visual inspection and output measurement methods can be used for fault diagnosis in PV panels with reduced output effi- ciency (Xxxxxx et al., 2014). Recently, methods for automated fault detection by combining aerial photogrammetry with com- puter vision technology have been developed and two technolo- xxxx are mainly involved in this process: the automatic extraction of the ROI (region of interest) of a PV array field from the given images and the automatic diagnosis of defective panels based on extracted PV panel areas (Xxx et al., 2017).
In (Xxxxxxxx et al., 2015) and in (Xxxxxxx et al., 2014),the authors have presented methods for the ROI extraction from terrestrial thermal infrared image sequences, applying image segmentation techniques (Xxxxxxxx et al., 2004).
The Canny edge operator (Canny, 1986) and image segmentation techniques are also used for the development of an algorithm for panel area extraction from thermal infrared images captured with a UAV in (Xxx et al., 2017). However, the area extraction method using the Canny edge operator did not lend itself well to creating a single polygon for the panel area due to noises within and out- side the panels. Instead, the image segmentation-based area ex- traction method has created polygons of individual panel areas, albeit limited by imperfect linearity. As well Xxxxxxxx et al. have adopted the Canny edge operator for the design of a method to identify the location of hot spot cells on a PV panel (Xxxxxxxx et al., 2015). In this study the authors have employed a method for comparing the intensity characteristics of individual panel area polygons for the algorithm development. Another study for the panel fault diagnosis was performed by Xxx et al. (Xxx et al., 2017). They have proposed an algorithm capable of automated PV panel fault diagnosis using intensity-related statistical values of each panel based on extracted panel area polygons.
An additional fault detection and diagnostic method is based on hardware redundancy, in which several similar subsystems un- dertake the same task. The detection of anomalies is done by collecting and analysing each subsystems data (Daliento et al., 2017). Another possibility in monitoring and diagnosis is based on electrical parameters such as Performance Ratio considering comparison between arrays (Bizzarri et al., 2015), or strings by comparing currents (Xxxx et al., 2013), inferential tool, returning information about the operation of the PV field (Xxxxxxxxx et al., 2015), dataset of observed string currents and voltages and their respective low-pass-filtered time derivatives (Xxx-Xxxxxxx and Xxxx, 2012), or I-V curve of a single string using information from the inverter (Davarifar et al., 2014), etc. A common prob- lem of all these monitoring techniques is the large amount of data to be analyzed.
More sophistication algorithms and methods for fault detection and diagnostic are required,since PV array characteristics are highly nonlinear. Nowadays, the research is moving towards the direc- tion of the use of artificial intelligence and data mining. Basically, these methods can be split into three categories: signal process- ing methods, classification methods, and inference methods. Sig- nal processing methods are extracting some features of the mea- sured signals, important for a particular state of health of the PV
system. Two most commonly used methods are wavelet trans- form techniques (il Song Xxx, 2016) and Fast Fourier Trans-
Table 1. Thermographic Payload: thermocamera Flir, Tau640 model
form (FFT) (Xxxxx and Xxxxxx, 2003). The classification meth-
ods instead are artificial intelligence based,and the knowledge is built from an available dataset. Supervised algorithms are first be- ing trained and then on the basis of large labeled data supervised learning algorithms can learn the characteristics of the system and make the prediction. Artificial neural networks (XXX) have been proposed for PV systems: working under partial shading xxxxx- xxxxx (Xxxxxx et al., 2009); for the monitoring and supervision of health status of a PV system in (Xxxxx and Xxxxxxx, 2012); and for short-circuit fault detection of PV arrays in (Xxxxxxxxxxx et al., 2011). In other works, Bayesian networks (Xxxxxxx and Xxxxxxxx, 2011), and fuzzy logic (XxxxxXxxx et al., 2004) have also been successful in estimating PV output or perform fault di- agnoses. Data mining methods for fault detection and isolation in PV systems have also been proposed in the literature, such as decision-tree method (Xxxx et al., 2012), K-nearest neighbor, support vector machine (SVM) (Yi and Xxxxxxx, 2016), etc.
Artificial neural network (XXX) models were shown to closely match performance of array models in PV systems (Xxxxx and Xxxxxxxxxxxxxxx, 2011). This learning technique is requiring only basic system monitoring hardware. Advantages of using the XXX in PV health monitoring system are: no need for a priori information of the system components or topology to accurately model the output power, it can monitor the degradation of the system over its lifetime, it can indicate catastrophic failures by monitoring the degradation rate.
3. METHODS
In this section, we introduce the framework as well as the dataset used for evaluation. We use especially trained DCNN for cells degradations. The framework is depicted in Figure 1. Further details are given in the following subsections. The framework is comprehensively evaluated on the “Photovoltaic images Dataset”, a proprietary dataset collected for this work. The details of the data collection and ground truth labeling are discussed in subsec- tion 3.2.
3.1 Network Architecture
×
The DCNN aims at providing information about the cells degra- dation and is therefore trained with image labels indicating the anomaly in the images. The training is performed with a VGG- 16 net (Xxxxxxxx and Xxxxxxxxx, 2014). The VGG-16 network is chosen for its relative ease of implementation and its success in the ILSVRC-2014 competition where it placed first in the 2a challenge. The VGG-16 is a very deep, 16-convolutional-layer network that is originally trained on the ImageNet database con- sisting of millions of labeled images in 1000 classes (Krizhevsky et al., 2012). The model was developed in Keras, a high-level neural networks library, written in Python and capable of running on top of either TensorFlow1. The VGG-16 network consists of 5 convolutional blocks with corresponding output filter sizes [54, 128, 256, 512, 512] and then a fully-connected classifier. Our im- plementation for VGG16 net follows the practice in (Xxxxxxxx and Xxxxxxxxx, 2014). The image is resized to 224 224 pixel. Since there is not batch normalization layer in VGG16, input im- ages are been normalized. We use the stochastic gradient descent (SGD) optimizer with a mini-batch size of 2. After preliminary experiment the learning rate is fixed to 0.001. We use a weight
1xxxxx://xxx.xxxxxxxxxx.xxx/
Overview Tau 640
Thermal Imager Uncooled VOx Microbolometer FPA / Digital Video Display Formats 640 512
Analog Video Display Formats 640 480 (NTSC); 640 512 (PAL)1 Pixel Pitch 17µm
Spectral Band 7.513.5µm
Full Frame Rates 30 Hz (NTSC) – 25 Hz (PAL)
Exportable Frame Rates 7.5 Hz NTSC; 8.3 Hz PAL Sensitivity (NEdT) ¡50 mK at f/1.0
Scene Range (High Gain) 25oC to +135oC
Scene Range (Low Gain) 40oC to +550oC Time to Image ¡5.0 sec
decay of 0.0001 and a momentum of 0.9. A small learning rate is used to prevent wrecking the previously learned feature weights. The network is initially trained with a binary cross entropy (BCE) for 10 epochs.
3.2 Photovoltaic images Dataset
A thermographic inspection of a ground-based PV system has carried out with a power of approximately 66 MW in South Africa. The thermographic acquisitions were made in 7 working days, from 21 to 27 January 2017 with sky predominantly clear and with maximum irradiation. This situation is optimal to enhance any abnormal behaviour of the entire panels or portion of them. The analysis was carried out with a constant flight altitude of 50 meters with respect to the surface of the panels. A Skyrobotic SR-SF6 drone equipped with Thermographic Payload (Table 1) with FLIR TAU 2 infrared camera.
The “Photovoltaic images Dataset” is composed of 3336 thermal images collected as follows:
• 811 images with dameged PV cells;
• 2525 images with normal PV cells.
×
−
The ground truth of the collected pictures has been manually es- timated by human annotators, thus providing a more precise and less noisy dataset. Figure 2 shows two examples of pictures of “Photovoltaic images Dataset”. The images have a dimension of 640 512 and contain temperature values, therefore these images have single band. Temperature accuracy in files is 0.01 Celsius. The temperature of the images range from 19.27o to 103.33o with a mean of 45.71o and a standard deviation of 8.96.
It has been shown in (Xxxxxxx et al., 2013), (Xxxxxxx et al., 2014) for neighbor embedding methods and anchored regression meth- ods and in (Xxxx et al., 2014), (Xxxx et al., 2016) for the convo- lutional neural networks-based methods that more training data results in an increase in performance up to a point where expo- nentially more data is necessary for any further improvement. Re- xxxxxxx the image classification literature (Xxxxxxxxx et al., 2014), we consider also the flipped and rotated versions of the training images. If we rotate the original images by 90, 180, 270 and flip them left-to-right and top-to-bottom (see Figure 3 and (Xxxxxxx et al., 2000)), we get more images without altered content. In particular, in case of flipped images we obtain a new unpdated dataset of 4866 images (2433 normal and 2433 damaged). In case of rotated images we obtain 6488 (3244 normal and 3244 damaged).
Figure 1. DCNN image classification pipeline.
(a) (b)
Figure 2. Figure 2a is an image with damaged photovoltaic cells. Figure 2b represents an image shown a normal photovoltaic cells.
3.3 Performance Evaluation Metrics
To evaluate the performance of the algorithms the following met- rics are employed:
•
Accuracy: approximates the effectiveness of the algorithm by showing the probability of the true value of the class la- bel (Xxxxxxxx et al., 2006):
can be interpreted as a weighted average of the precision and recall.
•
Support: is the number of occurrences of each class in ground truth (correct) target values.
The use of a confusion matrix can also be insightful for xxx- lyzing the results of the model. Infact, the information about actual and predicted classifications done by system is de-
Accuracy
tp + fn
= tp + tn + fp + fn (1)
picted by confusion matrix (Xxxxxx, 1998). Confusion ma- trix is a specific table layout that allows visualization of the performance of an algorithm, where each column of the ma-
where tp is the number of true positives and fn the number of false negatives.
•
Recall: is a function of its correctly classified examples (true positives) and its incorrectly classified examples (false neg- atives).
recall = tp (2)
tp + fn
•
Precision: is a function of true positives and examples in- correctly classified as positives (false positives).
trix represents the instances in a predicted class, and each row represents the instances in an actual class.
4. RESULTS AND DISCUSSION
In this section, the results of the experiments conducted on “Pho- tovoltaic images Dataset” are reported. This dataset presents un- balanced data, i.e. the classes are not represented equally. This is a scenario where the number of observations belonging to one class is significantly lower than those belonging to the other classes. Since classification algorithms tend to produce unsatisfactory clas- sifiers when faced with unbalanced datasets (Chawla, 2009), in
precision = tp
tp + fp
• F1-score: is a measure of a test’s accuracy.
(3)
addition to the performance of the DCNN for automatic inspec- tion, we also present the performance with the flipped and rotated (by 90, 180, 270) versions of the training images. In this way, we increase the quality of the dataset and ensure the validity of the experiments.
F1 −
score =
(β 2 + 1) ∗ precision ∗ recall
β 2 ∗ precision + recall
(4)
In fact, dealing with unbalanced datasets entails strategies such as improving classification algorithms or balancing classes in the
The F1-score is evenly balanced when β = 1. It favours precision when β > 1, and recall otherwise. The F1-score
training data (data preprocessing) before providing the data as in- put to the DCNN. The main objective of balancing classes is to
(a) resized (b) up-down flipped (c) left-right flipped
(d) rotated 90 (e) rotated 180 (f) rotated 270
Figure 3. Figure 3a is image seen by input net after resizing of original image data. Figure 3b-3c are image rispectively up-down and left-right flipped from resized image. Figure 3d-3e-3f are image rotated rispectively of 90, 180 and 270 degree from resized image.
either increasing the frequency of the minority class or decreas-
Table 2. Unbalanced Dataset
ing the frequency of the majority class. This is done in order
to obtain approximately the same number of instances for both the classes. Five dataset are evaluated: unbalanced “Photovoltaic images Dataset”, balanced “Photovoltaic images Dataset”, “Pho- tovoltaic images Dataset” with flipped images left-to-right and top-to-bottom, “Photovoltaic images Dataset” with rotated im- ages by 90, 180, 270, “Photovoltaic images Dataset” with both flipped and rotated images.
The dataset unbalanced is comprised of a total amount of 3336 pictures, including 811 pictures with anomaly, 2525 pictures nor- mal. The dataset balanced comprises 1622 images, where 811 are images with anomaly and 811 normal. The dataset with flipped images collects 4866, where 2433 are images with anomaly and 2433 normal. The dataset in which the images are rotated is com- posed by 6488 images (3244 with anomaly and 3244 normal). Finally, the dataset with both flipped and rotated images includes 9732 pictures (4866 with anomaly, 4866 pictures normal).
We perform the experiments by splitting the labeled dataset into a training set and a test set. Each classifier will only be trained based on the training set. Likewise, the test set is also fixed in the beginning and used for all test purposes. The dataset is split into 80% training and 20% test images,.
The performance of the VGG-16 network for the unbalanced dataset is reported in Table 2. As expected, the values of precision and recall are not very high because we deal with unbalanced data. In fact, the metrics reflect the underlying class distribution and for the normal images we obtain high values of precision, recall (such as 100%) and F1-score.
Table 3 depicts precision, recall and F1 score of the balanced dataset. The performance of the network is good and comparing
Category Precision Recall F1-Score Support
Damaged 0.00 0.00 0.00 163
Normal 0.76 1.00 0.86 505
MEAN() 0.57 0.76 0.65 668
Table 3. Balanced Dataset
Category Precision Recall F1-Score Support
Damaged 0.83 0.50 0.63 163
Normal 0.64 0.90 0.75 505
MEAN() 0.74 0.70 0.69 326
the different classes reveals that both normal and anomaly can be recognized.
Table 4, Table 5 and Table 6 report the results of the VGG-16 performance with data augmentation. As can be seen, the per- formance of all classifiers is good, thus demonstrating the effec- tiveness and the suitability of the proposed approach. The per- formance of the classification with images rotated and flipped is much higher than the performance of the flipped images classi- fication but slightly lower than the classification performance of the dataset with images rotated. This comparison shows that rec- ognizing the anomaly in images with unbalanced dataset is chal- lenging.
The following Tables show the confusion matrices of the applied VGG-16 network.
5. CONCLUSIONS
Monitoring and maintaining a PV plant represent a challenging but rewarding task and detecting the degradation of solar panels
Table 4. Balanced Dataset with data augmentation (flipped images)
Table 6. Balanced Dataset with data augmentation (flipped and rotated images)
Category | Precision | Recall | F1-Score | Support | Category | Precision | Recall | F1-Score | Support | |
Damaged | 0.83 | 0.50 | 0.63 | 163 | Damaged | 0.67 | 0.78 | 0.72 | 978 | |
Normal | 0.64 | 0.90 | 0.75 | 505 | Normal | 0.73 | 0.61 | 0.67 | 978 | |
MEAN() | 0.74 | 0.70 | 0.69 | 326 | MEAN() | 0.70 | 0.69 | 0.69 | 1956 |
Table 5. Balanced Dataset with data augmentation (rotated images)
Damaged
Damaged Normal
Category | Precision | Recall | F1-Score | Support |
Damaged | 0.79 | 0.67 | 0.73 | 652 |
Normal | 0.72 | 0.83 | 0.77 | 652 |
MEAN() | 0.76 | 0.75 | 0.75 | 1304 |
is the only way to ensure good performances over long periods of time. In this paper, we introduce a deep learning approach for estimating the PV cells degradations. The DCNN aims at pro- viding information about the cells degradation and is therefore trained with image labels indicating the anomaly in the images. The training has been performed with a VGG-16 net, which has been chosen for its relative ease of implementation and its effec- tiveness in case of image classification. The approach is able to achieve high precision and recall for damage classification. The experiments on the “Photovoltaic images Dataset” yield high ac- curacies and demonstrate the effectiveness and suitability of our approach. Further investigation will be devoted to improve our approach by employing a larger dataset and extracting additional informative features such as conductibility. Moreover, other DC- NNs are trained and their performances are compared.
ACKNOWLEDGEMENTS
This work was funded by Fly Engineering (www.flyengineering. it). The authors would like to thank Xxxxxxx Xxxxxxx for his support to the work.
REFERENCES
XxxxxXxxx, M., Xx-Xxxxxxx, A. M. and Xxxx, X. X., 2004. Neuro- fuzzy-based solar cell model. IEEE Transactions on Energy Con- version 19(3), pp. 619–624.
Xxxxxxxxx, X., Xxxxxxx, X. X., Xxxxx, B., Xxxxxx, F., Xxxxx, A., X’Xxxxxxxx, X., xx xx Xxxxx Xxxxxxxxx, M. and Xxxxx, V., 2015. Integrating variable renewables in europe : Current status and recent extreme events. IEEE Power and Energy Magazine 13(6), pp. 67–77.
Xxxx, M., Xxxxxxxxx, X. and Xxxxxxxx, N., 2013. Examination of fault detection technique in pv systems. pp. 1–4.
Xxx-Xxxxxxx, X. and Xxxx, S. C., 2012. Online photovoltaic array hot-spot bayesian diagnostics from streaming string-level electric data. pp. 002432–002437.
Xxxxxxxx, X., Xxxxxxxxx, A., Xxxxxxx, L. and Xxxxxxxxx, C., 2015. Monitoring performance and efficiency of photovoltaic parks. Renewable Energy 78, pp. 314 – 321.
BPp.l.c, 2016. Bp energy outlook 2016. London, UK.
Xxxxx, X., 1986. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-8(6), pp. 679–698.
Normal
114
342
375
147
Table 7. Confusion matrix of the Unbalanced Dataset.
17
82
146
81
Balanced Dataset Damaged Normal
Damaged
Normal
Table 8. Confusion matrix of the Balanced Dataset.
114
342
375
147
Damaged Normal
Damaged
Normal
Table 9. Confusion matrix of the Balanced Dataset with data augmentation (flipped images).
Xxxxxxxxx, X., Xxxxxxxx, X., Xxxxxxx, X. and Xxxxxxxxx, X., 2014. Return of the devil in the details: Delving deep into con- volutional nets. arXiv preprint arXiv:1405.3531.
Damaged
Normal
Damaged Normal
113
438
549
214
Xxxx, X., Xxx, X. X., He, X. and Xxxx, X., 2016. Image super- resolution using deep convolutional networks. IEEE transactions on pattern analysis and machine intelligence 38(2), pp. 295–307.
Xxxxx, X., xx Xxxxxx, X., Xxxxx, X., Xxxxxx, X., Xxxxxx, J., xxx Xxxx, X., Xxxxxxxxxxx, X., Xxxxxxx, X., Stettler, S., Xxxxxxxxxx, P., Bofinger, X., Xxxxxxxxx, M., Xxxxxxxxx, X. and Xxxxxxxxx, D., 2007. Monitoring and remote failure detection of grid-connected pv systems based on satellite observations. Solar Energy 81(4), pp. 548 – 564.
EIA, 2016. International energy outlook 2016. U.S. Energy Information Administration, Washington, DC 20585, DOE/EIA- 0484(2016).
Xxxxxxx, X. X., Xxxxxxx, X. X. and Xxxxxxxxxx, O. T., 2000.
Table 10. Confusion matrix of the Balanced Dataset with data augmentation (rotated images).
383
763
595
215
Damaged Normal
Damaged
Normal
Table 11. Confusion matrix of the Balanced Dataset with data augmentation (flipped and rotated images).
Xxxxxx, N. V., 2009. Data mining for imbalanced datasets: An overview. In: Data mining and knowledge discovery handbook, Springer, pp. 875–886.
Xxxxxxx, X. and Xxxxxxxxx, S., 2010. Automatic supervision and fault detection of pv systems based on power losses analysis. En- ergy Conversion and Management 51(10), pp. 1929 – 1937.
Xxxxxxx, X. and Xxxxxxxx, J., 2011. Intelligent fault detection and diagnostics in solar plants. 2, pp. 948–953.
Xxxxxxxxx, X., Faifer, M., Xxxxx, X. and Xxxxxxx, S., 2015. Refer- ence strings for statistical monitoring of the energy performance of photovoltaic fields. pp. 591–596.
Xxxxxxxx, S., Xxxxxxx, X., Xxxxxxxxx, P., Xxxxx, A. M., Xxxxxx, X., Xxxxxx, R. and Xxxxxxx, P., 2017. Monitoring, diagnosis, and power forecasting for photovoltaic fields: a review. International Journal of Photoenergy.
Xxxxxxxxx, M., Xxxxx, X., Xxxxxxx, X., Xxxxx, E. and Xxxxxxxxxx, Z., 2014. Partial shading fault diagnosis in pv system with dis- crete wavelet transform (dwt). pp. 810–814.
Xxxx, X., Xxx, X. X., Xx, K. and Xxxx, X., 2014. Learning a deep convolutional network for image super-resolution. In: European Conference on Computer Vision, Springer, pp. 184–199.
Learning low-level vision. International journal of computer vi- sion 40(1), pp. 25–47.
Xxxxxxxx, X. X., Xxxxx, X. X. and Xxxxxx, S., 2004. Digital image processing using matlab: Xxxxxxx xxxxxxxx xxxx. Upper Saddle River, New Jersey.
Xxxxxxxxx, F., Xxxxxx, M., Xxxxxxxx, M., Xxxx, S. and Xxxxxx,
P. B., 2015. Planning for pv plant performance monitoring by means of unmanned aerial systems (uas). International Journal of Energy and Environmental Engineering 6(1), pp. 47–54.
Xxxxxxxx, X., Xxxxxxx, I. and Xxxxxxx, A., 1999. Development of a flight test system for unmanned air vehicles. IEEE Control Systems 19(1), pp. 55–65.
IEA, 2016. World energy outlook 2016. International Energy Agency, Paris, France, 10.1787/weo-2016-en.
il Song Xxx, 2016. On-line fault detection algorithm of a pho- tovoltaic system using wavelet transform. Solar Energy 126, pp. 137 – 145.
Xxxxxx, D. C., Xxxxxx, C., Xxxxx, S. R., Xxxxxxx, X. X. and Xx- xxxxxx, M., 2018. Robust pv degradation methodology and ap- plication. IEEE Journal of Photovoltaics 8(2), pp. 525–531.
Xxx, X., Xxxx, X. and Xxx, C., 2017. Automatic fault recognition of photovoltaic modules based on statistical analysis of uav ther- mography. ISPRS-International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences pp. 179–182.
Xxxxxx, X., 1998. Glossary of terms. Machine Learning 30, pp. 271–274.
Xxxxxxxxxx, X., Xxxxxxxxx, X. and Xxxxxx, X. X., 2012. Imagenet classification with deep convolutional neural networks. In: Xx- xxxxxx in neural information processing systems, pp. 1097–1105.
Xxxxx, X. X., Xxxxxxx, X. X., Xxxxxxxx, X. and Xxxxxx, X. X., 2016. A deep convolutional neural network and a random forest classifier for solar photovoltaic array detection in aerial imagery. pp. 650–654.
Xxxxx, X. X. and Xxxxxx, R., 2003. Design and analysis of aerospace dc arcing faults using fast fourier transformation and artificial neural network. 2, pp. 793 Vol. 2.
Xxxxxx, X. X., Xxxxxx, X. and Xxxxxxxx, S., 2009. Performance evaluation of solar photovoltaic arrays including shadow effects using neural network. pp. 3357–3362.
Xxxxxx, X. and Xxxxxx, L., 2004. Control and perception tech- niques for aerial robotics. Annual reviews in Control 28(2), pp. 167–178.
Xxxxxxxx, M., Kaiser, C., Xxxxxxxxx, R., Xxxxxxxx, E. and Xxx- xxxxxxx, P., 2017a. Visual and textual sentiment analysis of brand- related social media pictures using deep convolutional neural net- works. In: International Conference on Image Analysis and Pro- cessing, Springer, pp. 402–413.
Xxxxxxxx, M., Xxxxxxx, M., Xxxxxxx, X., Xxxxxxxxxx, P. and Xxxxxxxx, E., 2017b. Mobile robot for retail surveying and inventory using visual and textual analysis of monocular pictures based on deep learning. In: Mobile Robots (ECMR), 2017 European Conference on, IEEE, pp. 1–6.
Xxxxx, X., Xxxxxxxxx, X., Xxxxxxxxxxxx, X., Xxxxxxxx, G. and Xxxxxxxxx, X. X., 2017. Forecasting degradation rates of different photovoltaic systems using robust principal component analysis and arima. IET Renewable Power Generation 11(10), pp. 1245– 1252.
Xxxxxx, P. B., Xxxxxxxxx, F., Xxxx, S., Xxxxxxxx, X. and Xxxxxx, M., 2014. Light unmanned aerial vehicles (uavs) for coopera- tive inspection of pv plants. IEEE Journal of Photovoltaics 4(4), pp. 1107–1113.
Xxxxx, X. and Xxxxxxx, X., 2012. Photovoltaic prognostics and heath management using learning algorithms. pp. 001535– 001539.
Xxxxx, X. X. and Xxxxxxxxxxxxxxx, X. X., 2011. Comparison of a recurrent neural network pv system model with a traditional component-based pv system model. pp. 002426–002431.
Xxxxxxx, X., Xxxxxxxxx, D., Xxxxxxxx, X., Xxxxxxxx, D. and Xxxxxxxxxxxxx, S., 2014. Early defect diagnosis in installed pv modules exploiting spatio-temporal information from thermal images. In: Proceedings of the 12th quantitative infrared ther- mography conference (QIRT). Bordeaux, France.
Xxxxxxxxx, X., xx Xxxxx, M. A., Xxxxxxx, X., Xxxxxx, D. and Karatepe, E., 2014. New procedure for fault detection in grid connected pv systems based on the evaluation of current and volt- age indicators. Energy Conversion and Management 86, pp. 241
– 249.
Xxxxxxxx, X. and Xxxxxxxxx, A., 2014. Very deep convolu- tional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
Xxxxxxxx, X., Xxxxxxxxx, N. and Xxxxxxxxxx, S., 2006. Beyond accuracy, f-score and roc: a family of discriminant measures for performance evaluation. In: Australasian joint conference on ar- tificial intelligence, Springer, pp. 1015–1021.
Xxxxxxx, M., Xxxxxxxx, M., Xxxxxxxx, X., Xxxxxxx, A. and Xxx- xxxxxxx, P., 2017. Robotic platform for deep change detection for rail safety and security. In: Mobile Robots (ECMR), 2017 Euro- pean Conference on, IEEE, pp. 1–6.
Xxxxxxxxxxx, Karatepe, X. and Xxxxxx, T., 2011. Controlling of artificial neural network for fault diagnosis of photovoltaic array. pp. 1–6.
Xxxxxxx, X., Xx Xxxx, V. and Xxx Xxxx, L., 2014. A+: Adjusted anchored neighborhood regression for fast super-resolution. In: Asian Conference on Computer Vision, Springer, pp. 111–126.
Xxxxxxx, X., Xx, V. and Xxx Xxxx, L., 2013. Anchored neigh- borhood regression for fast example-based super-resolution. In: Computer Vision (ICCV), 2013 IEEE International Conference on, IEEE, pp. 1920–1927.
Xxxxxxxx, X., Xxxxxxxxxxxx, D., Xxxxxxxx, P. and Xxxxxxxxxx, X., 2015. Fault diagnosis of photovoltaic modules through image processing and canny edge detection on field thermographic mea- surements. International Journal of Sustainable Energy 34(6), pp. 351–372.
Xx, X. and Xxxxxxx, X. X., 2016. A novel detection algorithm for line-to-line faults in photovoltaic (pv) arrays based on support vector machine (svm). pp. 1–4.
Xxxx, X., Xxxx, X., Xxxxxx, B., xx Xxxxx, X. X., Xxxxxxxx, X. and Xxxxx, R., 2012. Decision tree-based fault detection and classifi- cation in solar photovoltaic arrays. pp. 93–99.