Conference PaperPDF Available


Timo Huuhtanen, Alexander Jung
Department of Computer Science,
Aalto University, Espoo, Finland
We apply convolutional neural networks (CNN) for monitoring the
operation of photovoltaic panels. In particular, we predict the daily
electrical power curve of a photovoltaic panel based on the power
curves of neighboring panels. An exceptionally large deviation be-
tween predicted and actual (observed) power curve can be used to
indicate a malfunctioning panel. The problem is quite challenging
because the power curve depends on many factors such as weather
conditions and the surrounding objects (causing shadows with a reg-
ular time pattern). We demonstrate, by means of numerical exper-
iments, that the proposed method is able to predict accurately the
power curve of a functioning panel. Moreover, the proposed ap-
proach outperforms the existing approaches that are based on simple
interpolation filters.
Index Termsmachine learning, convolutional neural net-
works, photovoltaic panels, wireless sensor networks, predictive
The deployment of photovoltaic (PV) power plants has increased
significantly in recent years. The growth of number and size of PV
power plants also raises the importance of predictive maintenance.
Optimal power production requires monitoring of each individual
PV panel. A typical monitoring system consists of sensors connected
to each PV panels measuring the electrical power production of each
panel as function of time, i.e. the power curve. The power mea-
surements are collected by a wireless sensor network (WSN) and
then analyzed using data mining tools. PV monitoring systems have
been studied during the past decade [1], [2], [3] and also commercial
systems exist [4], [5], [6]. The ultimate goal of predictive mainte-
nance is to identify malfunctioning PV panels. However, the cur-
rent monitoring systems focus merely on collecting and visualizing
operational data. We introduce an algorithm which detects a mal-
functioning PV panel based on the power measurement history data
(time-series of the power measurements) of the target panel and the
neighboring panels.
The problem is challenging because of the large dynamic varia-
tion in the power measurements of functioning panels. These varia-
tions throughout ”normal” or functioning operations are due to sev-
eral factors, e.g.
F1 The power generated at a certain time is proportional to the
solar irradiance received by the panel. The irradiance varies
due to seasons and daytime depending on the geographical
location and precise position (orientation) of the panel. How-
ever, these variations can in principle be calculated explicitly.
Moreover, these factors are strongly correlated for panels of
the same type and which are located close to each other.
Fig. 1. Samples of measured solar panel power curves of the target
panel and two neighboring panels.
F2 The local weather (e.g., cloud intensity) has a strong influ-
ence on the amount of power generated by the panel. The
maximum power on a sunny day can be close to the maxi-
mum nominal power of the panel while the maximum power
on cloudy day may be less than 20% of the nominal power.
Weather predictions are not accurate enough to predict the
output power of a single PV panel in each moment of time
[7]. This causes large and somewhat unpredictable variations
especially in geographical areas where partly cloudy days are
common [8].
F3 Normally the power curves of close-by panels are very close
to each other in a fully cloudy weather, or in sunshine with-
out shadows. However, nearby objects (trees, buildings, etc.)
may cause shadows to fall on the panels individually. This
causes large differences between adjacent panels and it oc-
curs regularly at certain daytimes (due to the position of the
F4 Malfunctioning panels may cause gradual or fast drops for the
amount of generated power.
In Fig. 1, we illustrate the power curves obtained for three simi-
lar adjacent panels where factors F1-F3 are present. During the first
part of the curve (until time index t265) the weather has been
cloudy and the measurements of the three panels follow each other
very closely. For times t > 265, the weather has become sunny and
shadows of nearby objects falling individually on each panel cause
the power curves to differ from each other significantly. Our main
interest is in accurate differentiation between changes due to factors
F1-F3 or factor F4, i.e. identification of malfunctioning panels.
Anomaly detection using WSN has been studied extensively [9],
[10], [11], [12], [13]. Artificial neural networks (ANN) have also
been used in processing the data from WSNs [14]. However, pre-
dictive maintenance of PV based power plants sets some special re-
quirements for these methods and this application area has received
little attention so far. ANNs have been used in the context of PV but
the aim has been to create forecasts for produced PV power - for the
purpose of power grid balancing [7], [15], [16], [17]. In this paper,
we study whether deep neural networks can be applied in predictive
maintenance of PV panels. To the best of our knowledge, the prob-
lem of predictive maintenance for PV systems has not been earlier
tackled using deep learning methods.
We consider predictive maintenance as a classification problem, i.e.,
we classify each panel as ’functioning’ or ’malfunctioning’. The
classification is based on the measured power curves. What hinders
a naive application of standard classifiers (e.g., naive Bayes or lo-
gistic regression) is the lack of large amounts of training data from
malfunctioning panels since panel faults are rare. Instead, we follow
the approach described in [10], [11], [12] and do anomaly detection
by comparing the actual power curve measured at a particular panel
with the predicted power curve based on the neighboring panels. A
large deviation between actual and predicted power curves indicates
a malfunctioning panel. Our problem then becomes to find the best
predictor for the power of the target panel using the power values of
the neighboring panels. In this study, all our data is from fully func-
tional panels and we aim at constructing a reliable predictor. The
actual measurements of the target panel are known and can be used
as training and test data for the predictor. In this paper we focus on
neural network based estimators [18].
The power measurements of one panel form a time series
P(t, d, s), where tis time of the day (in minutes), dcorresponds to
the date and sis the spatial location of the panel. Such sequential
data is usually processed with Recurrent Neural Networks (RNN)
or Long Short-Term Memory (LSTM) Networks [18]. However,
those methods have challenges with long-term dependencies [18]:
the gradient values tend to disappear or explode. This is relevant
also in our case, because we have long time-dependencies (due to
the regular daily pattern of the power curves). Additionally, these
methods do not consider absolute time but only treat time relative to
the current instance of time. In our application, however, the signal
is very much dependent on absolute time: the daily power curve
has a quite regular daily shape - from sunrise to sunset - as referred
in factor F1 mentioned in Section 1. Thus, instead of single power
measurements, we take the daily power curve as our input. The input
data can be considered as 2-dimensional signal, one dimension rep-
resenting the time of the day tand a second dimension representing
the spatial location of the panel s.
Our task in this paper is then to find a predictor ˆ
Pfor the power
curve of the target panel in location starget based on the panels in
its spatial neighborhood s∈ V,
P(t, d, starget ) = f(P(t, d, s)); s∈ V (1)
which minimizes the mean square error
P(t, d, starget )P(t, d, starget ))2(2)
In the spatial domain, we limit the input to our interpolation to
power measurements of the neighboring panels of our target panel.
On one hand we want to keep the algorithm simple, on the other hand
physical reasons suggest that this small neighborhood is sufficient
- the closest neighbors yield all the relevant information about the
illumination of our target panel.
Convolutional Neural Networks (CNN) have been successful in find-
ing patterns in 2-dimensional signals. Since our input data (cf. (1))
is actually 2-dimensional, we focus in this paper in the derivatives of
CNNs in solving our target problem.
The next step is to select the architecture to be used, i.e. the num-
ber and type of the layers to be used in the CNN. Because we only
have a limited amount of available real-life data, we try to keep the
amount of model parameters reasonable to avoid overfitting. On the
other hand, we need to have an architecture which has enough power
to model the complexity of the function to be estimated. As a com-
promise between these requirements, we limit our considerations in
this paper to two-layer CNN architectures. The design choice for the
types of layers can be split into two parts:
1. How to calculate the local interpolators (based on the input
values in the close time-space neighborhood) for the target
panel? The problem suggests the use of convolutional layer
for the first layer as also used in [19] and [20].
2. How to combine the local interpolators in the second layer?
A standard method is to use convolutional layer with parame-
ter sharing, which means that the combining is done the same
way at all times of the day - i.e. our system is time-invariant.
However, in this application we have phenomena depending
on absolute time of the day; e.g. regular shadows falling on
a panel e.g. every day between 10 am and 12 am. A con-
volutional layer without parameter sharing can learn to take
this into account; i.e. to do the combination differently on
different time of the day. This approach is sometimes called
unshared convolution [21].
In what follows we consider two options for the overall archi-
tecture are
Two fully convolutional layers (CC)
Fully convolutional first layer, unshared convolution for the
second layer (CUC)
The equation for fully convolutional layer is
y(t, s;l) = σ(
x(ti, s j, k)·h(i, j, k;l)) (3)
Where xis the input, yis the output, his the convolution kernel
and σis the activation function. Index k[0, K 1] represents
the different feature maps (inputs) from the previous layer, index
l[0, L 1] represents the different feature maps (outputs) in this
layer. The size of the convolution kernel is I·J·Kand there are L
different kernels. The equation for unshared convolutional layer is
y(t, s;l) = σ(
x(ti, s j, k)·h(i, j, k;l, t)) (4)
In contrast to (3), here the kernel his also a function of t.
The used hyperparameters for our methods are learning rate,
sizes of convolution kernels, batch size and number of epochs used
in the training.
As a benchmark to our methods we use a simple predictor con-
structed by calculating the average of the two adjacent panels, s0and
s1, for each time instance according to
P(t, d, starget ) = P(t, d, s0) + P(t, d, s1)
In what follows, we refer to the algorithm (5) as AVE. Even
though AVE is very simple, it is a reasonable interpolator when all
the panels are illuminated similarly (e.g. a fully cloudy day or a
fully sunny day with no shadows). In contrast to the simple AVE
method, the proposed CC and CUC methods are be able to model
the dynamics of the regular shadows to some extent.
(a) CC
(b) CUC
Fig. 2. The neural network architectures used in this paper.
As a proof of concept, we applied the proposed method to a synthetic
dataset and to a dataset containing actual measurements collected by
sensors. Both the datasets included daily power curves of the target
panel and the two neighboring panels. The number of parameters for
CUC algorithm is proportional to the length of the daily power curve,
T- so, the length of the curve was limited to T= 400 samples at 1
minute sample interval to keep the complexity reasonable. Around
80% of the data was used as training data and the rest as test data.
The synthetic dataset has data for 1000 days where power curves
for the target panel for odd days, do, are generated with (6) and for
even days, de, with (7).
P(t, do, starget ) = 0.5·P(t, do, s0) + 0.5·P(t, do, s1)(6)
P(t, de, starget ) = 0.9·P(t, de, s0) + 0.1·P(t, de, s1)(7)
The real-life dataset consists of measurements from three adja-
cent panels during 157 days from a solar monitoring pilot system
provided by SOLA Sense Ltd.
The two variants of the algorithm (CC, CUC) were implemented
and simulated with the two datasets. The algorithm architectures are
illustrated in Figure 2 and their key hyperparameters are summarized
in Table 1.
Table 1. Key hyperparameters for CC and CUC algorithms
I J K L Calculation
1st layer 3 2 1 9 Eq. (3)
2nd layer (CC) 3 1 9 1 Eq. (3)
2nd layer (CUC) 1 1 9 1 Eq. (4)
The mean square errors of the test data set for the two algo-
rithms as well as the simple average interpolator (5) are summarized
in Table 2. Examples of power curves of one day for real data are
illustrated in Figure 3 and for synthetic data in Figure 4. Here solid
line (’target’) is the true measurement of our target panel; i.e. we aim
at getting as close to that curve as possible. Dotted line represents
the proposed algorithm (’CC’ or ’CUC’) and the ’+’ signs represent
our benchmark (’AVE’).
Table 2. Mean Square Error of different algorithms and test sets
Synthetic test set 0.000037 0.000010 0.000421
Real test set 0.002589 0.002346 0.003561
Both the proposed algorithms, CC and CUC, outperformed the
benchmark significantly with the synthetic data. For real data the
difference was also clear but no longer that significant. Unshared
convolutional algorithm got the smallest test error with real data as
well as with the synthetic data. The example curve from the real
data in Figure 3 also reveals the different nature of the algorithms.
There are notches around time 180...250 where the power of the tar-
get panel drops to zero (probably caused by a shadow falling on the
target panel). The unshared convolutional algorithm could follow
this better compared to the fully convolutional algorithm and got
better result with the synthetic test set for the same reason.
The amount of real training data was relatively small for this
study. However, even with the available data we could do some inter-
esting findings regarding the behavior and performance of different
(a) CC
(b) CUC
Fig. 3. Performance of the proposed algorithms on a sample day of
the real-life test set.
algorithms. A larger training dataset would probably help to confirm
our findings. A more extensive study of different hyperparameter
values and algorithm options would probably also resulted perfor-
mance improvements.
According to our results, the application of CNN based methods has
potential for predictive maintenance for PV systems. Our results also
open several possibilities for further study; e.g. considering larger
(a) CC
(b) CUC
Fig. 4. Performance of the proposed algorithms on synthetic test set.
real-life datasets. Probably the performance of both the algorithms
can be improved by fine-tuning the hyperparameter values. CUC al-
gorithm can clearly predict the regular shadows and get performance
gain from that. It was also the overall winner with the real-life data.
On the other hand also CC algorithm seemed to be able to learn some
patterns of the training data with a smaller amount of parameters.
More research would be needed to understand, where this perfor-
mance gain comes from; i.e. what are the features of the data it can
learn and utilize in the prediction. An interesting option would also
be to try to combine CC and CUC algorithms into a hybrid model
having best properties of both approaches.
[1] C. Ranhotigamage and S. C. Mukhopadhyay, “Field Trials and
Performance Monitoring of Distributed Solar Panels Using a
Low-Cost Wireless Sensors Network for Domestic Applica-
tions,” IEEE Sensors Journal, vol. 11, no. 10, pp. 2583–2590,
Oct. 2011.
[2] P. Papageorgas, D. Piromalis, K. Antonakoglou, G. Vokas,
D. Tseles, and K. G. Arvanitis, “Smart Solar Panels: In-situ
Monitoring of Photovoltaic Panels based on Wired and Wire-
less Sensor Networks,” Energy Procedia, vol. 36, pp. 535–545,
Jan. 2013.
[3] P. Guerriero, F. Di Napoli, G. Vallone, V. d’Alessandro, and
S. Daliento, “Monitoring and diagnostics of PV plants by
a wireless self-powered sensor for individual panels, IEEE
Journal of Photovoltaics, vol. 6, no. 1, pp. 286–294, 2016.
[4] “,” .
[5] “,” .
[6] “,” .
[7] J. Antonanzas, N. Osorio, R. Escobar, R. Urraca, F. J.
Martinez-de Pison, and F. Antonanzas-Torres, “Review of pho-
tovoltaic power forecasting, Solar Energy, vol. 136, pp. 78–
111, Oct. 2016.
[8] K. Lappalainen and S. Valkealahti, “Photovoltaic mismatch
losses caused by moving clouds, Solar Energy, vol. 158, pp.
455–461, 2017.
[9] M. A. Alsheikh, S. Lin, D. Niyato, and H. P. Tan, “Machine
Learning in Wireless Sensor Networks: Algorithms, Strategies,
and Applications,” IEEE Communications Surveys Tutorials,
vol. 16, no. 4, pp. 1996–2018, 2014.
[10] Y. Zhang, N. Meratnia, and P. Havinga, “Outlier Detection
Techniques for Wireless Sensor Networks: A Survey,” IEEE
Communications Surveys Tutorials, vol. 12, no. 2, pp. 159–
170, 2010.
[11] M. Markou and S. Singh, “Novelty detection: a reviewpart 1:
statistical approaches,” Signal Processing, vol. 83, no. 12, pp.
2481–2497, Dec. 2003.
[12] M. Markou and S. Singh, “Novelty detection: a reviewpart 2:
neural network based approaches,” Signal Processing, vol. 83,
no. 12, pp. 2499–2521, Dec. 2003.
[13] H.H.W.J. Bosman, G. Iacca, A. Tejada, H. J. Wrtche, and
A. Liotta, “Spatial anomaly detection in sensor networks using
neighborhood information,” Information Fusion, vol. 33, pp.
41–56, Jan. 2017.
[14] F. Oldewurtel and P. Mahonen, “Neural Wireless Sensor Net-
works,” in International Conference on Systems and Networks
Communications, 2006. ICSNC ’06, Oct. 2006, pp. 28–28.
[15] J. Shi, W. J. Lee, Y. Liu, Y. Yang, and P. Wang, “Forecast-
ing Power Output of Photovoltaic Systems Based on Weather
Classification and Support Vector Machines,” IEEE Transac-
tions on Industry Applications, vol. 48, no. 3, pp. 1064–1069,
May 2012.
[16] Muhammad Qamar Raza, Mithulananthan Nadarajah, and
Chandima Ekanayake, “On recent advances in PV output
power forecast, Solar Energy, vol. 136, pp. 125–144, Oct.
[17] Florian Barbieri, Sumedha Rajakaruna, and Arindam Ghosh,
“Very short-term photovoltaic power forecasting with cloud
modeling: A review, Renewable and Sustainable Energy Re-
views, vol. 75, pp. 242–263, Aug. 2017.
[18] I. Goodfellow, Y. Bengio, and A. Courville, Deep Learning,
MIT Press, 2016.
[19] J. Yang, M. N. Nguyen, P. P. San, X. Li, and S. Krishnaswamy,
“Deep Convolutional Neural Networks on Multichannel Time
Series for Human Activity Recognition., in IJCAI, 2015, pp.
[20] F. J. Ordez and D. Roggen, “Deep Convolutional and LSTM
Recurrent Neural Networks for Multimodal Wearable Activity
Recognition,” Sensors, vol. 16, no. 1, pp. 115, Jan. 2016.
[21] Yann LeCun, “Generalization and network design strategies,
Connectionism in perspective, pp. 143–155, 1989.
... The proposed solution allowed self-checking to reduce washing machine damage and unnecessary maintenance and increased productivity through automatic problem identification using acoustic sensor data and precise part preparation. Similarly, CNNbased PdM has been applied to photovoltaic panels (Huuhtanen and Jung, 2018) and wind turbines (Canizo et al., 2017). The applications reviewed in this section are listed in Table 9. Predictive maintenance CNN Acoustic sensors (Pan et al., 2017(Pan et al., ) 2017 Predictive maintenance CNN Wind turbines (Canizo et al., 2017(Canizo et al., ) 2018 Predictive maintenance CNN Photovoltaic panels (Huuhtanen andJung, 2018) 2013 Service-life prediction ANN Facade (Dias et al., 2014) ...
... Similarly, CNNbased PdM has been applied to photovoltaic panels (Huuhtanen and Jung, 2018) and wind turbines (Canizo et al., 2017). The applications reviewed in this section are listed in Table 9. Predictive maintenance CNN Acoustic sensors (Pan et al., 2017(Pan et al., ) 2017 Predictive maintenance CNN Wind turbines (Canizo et al., 2017(Canizo et al., ) 2018 Predictive maintenance CNN Photovoltaic panels (Huuhtanen andJung, 2018) 2013 Service-life prediction ANN Facade (Dias et al., 2014) ...
Full-text available
The built environment is responsible for roughly 40% of global greenhouse emissions, making the sector a crucial factor for climate change and sustainability. Meanwhile, other sectors (like manufacturing) adopted Artificial Intelligence (AI) to solve complex, non-linear problems to reduce waste, inefficiency, and pollution. Therefore, many research efforts in the Architecture, Engineering, and Construction community have recently tried introducing AI into building asset management (AM) processes. Since AM encompasses a broad set of disciplines, an overview of several AI applications, current research gaps, and trends is needed. In this context, this study conducted the first state-of-the-art research on AI for building asset management. A total of 578 papers were analyzed with bibliometric tools to identify prominent institutions, topics, and journals. The quantitative analysis helped determine the most researched areas of AM and which AI techniques are applied. The areas were furtherly investigated by reading in-depth the 83 most relevant studies selected by screening the articles’ abstracts identified in the bibliometric analysis. The results reveal many applications for Energy Management, Condition assessment, Risk management, and Project management areas. Finally, the literature review identified three main trends that can be a reference point for future studies made by practitioners or researchers: Digital Twin, Generative Adversarial Networks (with synthetic images) for data augmentation, and Deep Reinforcement Learning.
... The wireless sensor readings are indicators of the health of the engine and are transmitted by a wireless network for central engine control, where the Remaining Useful Life (RUL) of the engine is predicted by models and guides the engine manufacture to maintain the engine if RUL is approaching to a critical value. To accurately implement the PdM systems, many models were employed, such as fuzzy models [9], big data frameworks [10], and deep learning-based models [11][12][13][14]. ...
... The constant increase in global energy consumption has led to a worsening shortage of primary energy resources such as fossil fuels, coal, and natural gas globally, with South Africa experiencing electricity load shedding since 2007 due to energy generation and maintenance difficulties [1]. These, together with an increase in fossil fuel prices, have compelled improved developments in such renewable energy aspects as forecasting for the improvement of the operational performance of a solar energy system and for scheduling the maintenance of the solar power [2][3][4]. The growth of solar renewable energy has been increasing rapidly, with the total installed PVs reaching over 67.4 gigawatts and 627 gigawatts globally at the end of 2011 and 2019, respectively. ...
Full-text available
Challenges in utilising fossil fuels for generating energy call for the adoption of renewable energy sources. This study focuses on modelling and nowcasting optimal tilt angle(s) of solar energy harnessing using historical time series data collected from one of South Africa’s radiometric stations, USAid Venda station in Limpopo Province. In the study, we compared random forest (RF), K-nearest neighbours (KNN), and long short-term memory (LSTM) in nowcasting of optimum tilt angle. Gradient boosting (GB) is used as the benchmark model to compare the model’s predictive accuracy. The performance measures of mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE) and R2 were used, and the results showed LSTM to have the best performance in nowcasting optimum tilt angle compared to other models, followed by the RF and GB, whereas KNN was the worst-performing model.
... Another key issue with using ML algorithms for discovering solar cells is the inaccurate predictions and outcomes from the ML models. [118] In most cases, ML algorithms give the confidence interval of the forecasted and predicted values of the solar cells. However, the predicted values for the discovery of solar cells seem to approach up to a maximum of 95% using the GPR and Bayesian optimization using the probability distribution, which sometimes proves to result in the poor fabrication of solar cells. ...
Full-text available
Solar photovoltaic (PV) technology has merged as an efficient and versatile method for converting the Sun's vast energy into electricity. Innovation in developing new materials and solar cell architectures is required to ensure lightweight, portable, and flexible miniaturized electronic devices operate for long periods with reduced battery demand. Recent advances in biomedical implantable and wearable devices have coincided with a growing interest in efficient energy-harvesting solutions. Such devices primarily rely on rechargeable batteries to satisfy their energy needs. Moreover, Artificial Intelligence (AI) and Machine Learning (ML) techniques are touted as game changers in energy harvesting, especially in solar energy materials. In this article, we systematically review a range of ML techniques for optimizing the performance of low-cost solar cells for miniaturized electronic devices. Our systematic review reveals that these ML techniques can expedite the discovery of new solar cell materials and architectures. In particular, this review covers a broad range of ML techniques targeted at producing low-cost solar cells. Moreover, we present a new method of classifying the literature according to data synthesis, ML algorithms, optimization, and fabrication process. In addition, our review reveals that the Gaussian Process Regression (GPR) ML technique with Bayesian Optimization (BO) enables the design of the most promising low-solar cell architecture. Therefore, our review is a critical evaluation of existing ML techniques and is presented to guide researchers in discovering the next generation of low-cost solar cells using ML techniques.
... Deep learning (DL) is a subset of machine learning built on artificial neural networks. DL methods such as convolutional neural networks (CNN), recurrent neural networks (RNN), deep reinforcement learning, autoencoders, etc., are emerging as important approaches in the maintenance of various equipment and system (Amihai et al. 2018;Huuhtanen and Jung 2018;Janssens et al. 2017). ...
This article aims to review the literature on condition-based maintenance (CBM) by analyzing various terms, applications, and challenges. CBM is a maintenance technique that monitors the existing condition of an industrial asset to determine what maintenance needs to be performed. This article enlightens the readers with research in condition-based maintenance using machine learning and artificial intelligence techniques and related literature. A bibliometric analysis is performed on the data collected from the Scopus database. The foundation of a CBM is accurate anomaly detection and diagnosis. Several machine-learning approaches have produced excellent results for anomaly detection and diagnosis. However, due to the black-box nature of the machine learning models, the level of their interpretability is limited. This article provides insight into the existing maintenance strategies, anomaly detection techniques, interpretable models, and model-agnostic methods that are being applied. It has been found that significant work has been done towards condition based-maintenance using machine learning, but explainable artificial intelligence approaches got less attention in CBM. Based on the review, we contend that explainable artificial intelligence can provide unique insights and opportunities for addressing critical difficulties in maintenance leading to more informed decision-making. The analysis put forward encouraging research directions in this area.
... Therefore, each module can be monitored by comparing its output to the output of adjacent modules. For example, a convolutional neural network has been proposed in [23] to detect faulty modules in this manner. Similarly, a random forest classifier proposed in [24] is able to identify open circuit, shading and snow cover. ...
The energy losses and costs associated with faults in photovoltaic (PV) systems significantly limit the efficiency and reliability of solar power. Since existing methods for automatic fault diagnosis require expensive sensors, they are only cost-effective for large-scale systems. To address these drawbacks, we propose a fault diagnosis model based on graph neural networks (GNNs), which monitors a group of PV systems by comparing their current and voltage production over the last 24 h. This methodology allows for monitoring PV systems without sensors, as hourly measurements of the produced current and voltage are obtained via the PV systems' inverters. Comprehensive experiments are conducted by simulating 6 different PV systems in Colorado using 6 years of real weather measurements. Despite large variations in number of modules, module type, orientation, location, etc., the GNN can accurately detect and identify early occurrences of 6 common faults. Specifically, the GNN reaches 84.6% +/- 2.1% accuracy without weather data and 87.5% +/- 1.6% when satellite weather estimates are provided, significantly outperforming two state-of-the-art PV fault diagnosis models. Moreover, the results suggest that GNN can generalize to PV systems it was not trained on and retains high accuracy when multiple PV systems are simultaneously affected by faults.
... Another successful example is the use of CNN for bearing fault diagnosis and faults classification in the work [4] in which 97,3% accuracy of fault predicting was achieved. According to [6] CNN has the potential in monitoring the operation and predictive maintenance of photovoltaic panels. Janessens et al. [7] successfully used a feature extraction CNN-based model on raw data for rotating machinery fault diagnosis and classification. ...
Conference Paper
Full-text available
This study will review common-used machine learning techniques for predictive analytics and maintenance techniques. Some popular deep learning methods and their features and limitations will be discussed. The paper will open the scope for the possibility of further detailed development of new predictive maintenance techniques using deep neural networks.
Predictive maintenance systems face a rich set of constraints along dimensions such as latency, throughput, physical size, monetary cost, as well as energy and power consumption. To meet performance requirements, predictive maintenance systems require specialized compute units (i.e., accelerators) in addition to conventional processor cores. Unfortunately, size and cost constraints commonly result in developers being forced into selecting System-on-Chip (SoC) platforms that do not have sufficient resources to fully accelerate all performance-critical functions — in essence raising the challenging question of how to optimally distribute the available resources across accelerators. This work introduces the Resource-Constrained Accelerator Selection (RCS) methodology, which identifies near-optimal multi-accelerator configurations for predictive maintenance applications. RCS takes a library of resource-scalable accelerator architectures as input and then selects the combination of accelerator configurations that minimizes end-to-end latency. We find that enabling RCS for typical predictive maintenance applications requires a resource-scalable Fast Fourier Transform (FFT) accelerator and propose ScaleFFT to fill this gap. We apply RCS and ScaleFFT to a collection of edge computing applications with different sensor bandwidths and find that they reduce end-to-end latency by 2.4× on average for a 256K-point FFT compared to a state-of-the-art configuration that only accelerates the machine learning algorithm. Moreover, we demonstrate that RCS enables real-world gains in oil well and train track monitoring systems.
The recent proliferation of Internet of Things (IoT) sensors has driven a myriad of industrial and urban applications. Through analyzing massive data collected by these sensors, the proactive maintenance management can be achieved such that the maintenance schedule of the installed equipment can be optimized. Despite recent progress in proactive maintenance management in industrial scenarios, there are few studies on proactive maintenance management in urban informatics. In this paper, we present an integrated framework of IoT and cloud computing platform for proactive maintenance management in smart city. Our framework consists of (1) an IoT monitoring system for collecting time-series data of operating and ambient conditions of the equipment and (2) a hybrid deep learning model, namely convolutional bidirectional long short-term memory model (CBLM) for forecasting the operating and ambient conditions based on the collected time-series data. In addition, we also develop a naïve Bayes classifier to detect abnormal operating and ambient conditions and assist management personnel in scheduling maintenance tasks. To evaluate our framework, we deployed the IoT system in a Hong Kong public toilet, which is the first application of proactive maintenance management for a public hygiene and sanitary facility to the best of our knowledge. We collected the sensed data more than 33 days (808 hours) in this real system. Extensive experiments on the collected data demonstrated that our proposed CBLM outperformed six traditional machine learning algorithms.
Full-text available
Mismatch losses is a major issue in the photovoltaic (PV) system and are mainly caused by partial shading; largest mismatch losses are caused by sharp shadows. These shadows are a typical problem for rooftop and residential installations. In large-scale PV plants, partial shading is mostly caused by moving clouds which produce gentle irradiance transitions causing typically only minor irradiance differences between adjacent PV modules. This paper presents a study of the mismatch losses of PV arrays with various layouts and electrical configurations during around 27,000 irradiance transitions identified in measured irradiance data. The overall effect of the mismatch losses caused by moving clouds on the energy production of PV plants was also studied. The study was conducted using a mathematical model of irradiance transitions and an experimentally verified MATLAB/ Simulink model of a PV module. The relative mismatch losses during the identified irradiance transitions ranged from 1.4% to 4.0% depending on the electrical configuration and layout of the PV array. The overall effect of the mismatch losses caused by moving clouds on the total electricity production of PV arrays was about 0.5% for the PV array with strings of 28 PV modules and substantially smaller for arrays with shorter strings. The proportions of the total mismatch losses caused by very dark or highly transparent clouds were small. About 70% of the total mismatch losses were caused by shadow edges with shading strengths ranging between 40% and 80%. These results indicate that the mismatch losses caused by moving clouds are not a major problem for large-scale PV plants. An interesting finding from a practical point of view is that the mismatch losses increase the rate of power fluctuations compared to the rate of irradiance fluctuations.
Full-text available
The field of wireless sensor networks (WSNs), embedded systems with sensing and networking capability, has now matured after a decade-long research effort and technological advances in electronics and networked systems. An important remaining challenge now is to extract meaningful information from the ever-increasing amount of sensor data collected by WSNs. In particular, there is strong interest in algorithms capable of automatic detection of patterns, events or other out-of-the order, anomalous system behavior. Data anomalies may indicate states of the system that require further analysis or prompt actions. Traditionally, anomaly detection techniques are executed in a central processing facility, which requires the collection of all measurement data at a central location, an obvious limitation for WSNs due to the high data communication costs involved. In this paper we explore the extent by which one may depart from this classical centralized paradigm, looking at decentralized anomaly detection based on unsupervised machine learning. Our aim is to detect anomalies at the sensor nodes, as opposed to centrally, to reduce energy and spectrum consumption. We study the information gain coming from aggregate neighborhood data, in comparison to performing simple, in-node anomaly detection. We evaluate the effects of neighborhood size and spatio-temporal correlation on the performance of our new neighborhood-based approach using a range of real-world network deployments and datasets. We find the conditions that make neighborhood data fusion advantageous, identifying also the cases in which this approach does not lead to detectable improvements. Improvements are linked to the diffusive properties of data (spatio-temporal correlations) but also to the type of sensors, anomalies and network topological features. Overall, when a dataset stems from a similar mixture of diffusive processes precision tends to benefit, particularly in terms of recall. Our work paves the way towards understanding how distributed data fusion methods may help managing the complexity of wireless sensor networks, for instance in massive Internet of Things scenarios.
Full-text available
Human activity recognition (HAR) tasks have traditionally been solved using engineered features obtained by heuristic processes. Current research suggests that deep convolutional neural networks are suited to automate feature extraction from raw sensor inputs. However, human activities are made of complex sequences of motor movements, and capturing this temporal dynamics is fundamental for successful HAR. Based on the recent success of recurrent neural networks for time series domains, we propose a generic deep framework for activity recognition based on convolutional and LSTM recurrent units, which: (i) is suitable for multimodal wearable sensors; (ii) can perform sensor fusion naturally; (iii) does not require expert knowledge in designing features; and (iv) explicitly models the temporal dynamics of feature activations. We evaluate our framework on two datasets, one of which has been used in a public activity recognition challenge. Our results show that our framework outperforms competing deep non-recurrent networks on the challenge dataset by 4% on average; outperforming some of the previous reported results by up to 9%. Our results show that the framework can be applied to homogeneous sensor modalities, but can also fuse multimodal sensors to improve performance. We characterise key architectural hyperparameters’ influence on performance to provide insights about their optimisation.
Full-text available
This paper focuses on human activity recognition (HAR) problem, in which inputs are multichannel time series signals acquired from a set of body-worn inertial sensors and outputs are predefined human activities. In this problem, extracting effective features for identifying activities is a critical but challenging task. Most existing work relies on heuristic hand-crafted feature design and shallow feature learning architectures, which cannot find those distinguishing features to accurately classify different activities. In this paper, we propose a systematic feature learning method for HAR problem. This method adopts a deep convolutional neural networks (CNN) to automate feature learning from the raw inputs in a systematic way. Through the deep architecture, the learned features are deemed as the higher level abstract representation of low level raw time series signals. By leveraging the labelled information via supervised learning, the learned features are endowed with more discrimi-native power. Unified in one model, feature learning and classification are mutually enhanced. All these unique advantages of the CNN make it out-perform other HAR algorithms, as verified in the experiments on the Opportunity Activity Recognition Challenge and other benchmark datasets.
Full-text available
Wireless sensor networks monitor dynamic environments that change rapidly over time. This dynamic behavior is either caused by external factors or initiated by the system designers themselves. To adapt to such conditions, sensor networks often adopt machine learning techniques to eliminate the need for unnecessary redesign. Machine learning also inspires many practical solutions that maximize resource utilization and prolong the lifespan of the network. In this paper, we present an extensive literature review over the period 2002-2013 of machine learning methods that were used to address common issues in wireless sensor networks (WSNs). The advantages and disadvantages of each proposed algorithm are evaluated against the corresponding problem. We also provide a comparative guide to aid WSN designers in developing suitable machine learning solutions for their specific application challenges.
Conference Paper
This paper focuses on human activity recognition (HAR) problem, in which inputs are multichannel time series signals acquired from a set of body-worn inertial sensors and outputs are predefined hu-man activities. In this problem, extracting effec-tive features for identifying activities is a critical but challenging task. Most existing work relies on heuristic hand-crafted feature design and shallow feature learning architectures, which cannot find those distinguishing features to accurately classify different activities. In this paper, we propose a sys-tematic feature learning method for HAR problem. This method adopts a deep convolutional neural networks (CNN) to automate feature learning from the raw inputs in a systematic way. Through the deep architecture, the learned features are deemed as the higher level abstract representation of low level raw time series signals. By leveraging the labelled information via supervised learning, the learned features are endowed with more discrimi-native power. Unified in one model, feature learn-ing and classification are mutually enhanced. All these unique advantages of the CNN make it out-perform other HAR algorithms, as verified in the experiments on the Opportunity Activity Recogni-tion Challenge and other benchmark datasets.
This paper endeavors to provide the reader with an overview of the various tools needed to forecast photovoltaic (PV) power within a very short-term horizon. The study focuses on the specific application of a large scale grid-connected PV farm. Solar resource is largely underexploited worldwide whereas it exceeds by far humans’ energy needs. In the current context of global warming, PV energy could potentially play a major role to substitute fossil fuels within the main grid in the future. Indeed, the number of utility-scale PV farms is currently fast increasing globally, with planned capacities in excess of several hundred megawatts. This makes the cost of PV-generated electricity quickly plummet and reach parity with non-renewable resources. However, like many other renewable energy sources, PV power depends highly on weather conditions. This particularity makes PV energy difficult to dispatch unless a properly sized and controlled energy storage system (ESU) is used. An accurate power forecasting method is then required to ensure power continuity but also to manage the ramp rates of the overall power system. In order to perform these actions, the forecasting timeframe also called horizon must be first defined according to the grid operation that is considered. This leads to define both spatial and temporal resolutions. As a second step, an adequate source of input data must be selected. As a third step, the input data must be processed with statistical methods. Finally, the processed data are fed to a precise PV model. It is found that forecasting the irradiance and the cell temperature are the best approaches to forecast precisely swift PV power fluctuations due to the cloud cover. A combination of several sources of input data like satellite and land-based sky imaging also lead to the best results for very-short term forecasting.
Variability of solar resource poses difficulties in grid management as solar penetration rates rise continuously. Thus, the task of solar power forecasting becomes crucial to ensure grid stability and to enable an optimal unit commitment and economical dispatch. Several forecast horizons can be identified, spanning from a few seconds to days or weeks ahead, as well as spatial horizons, from single site to regional forecasts. New techniques and approaches arise worldwide each year to improve accuracy of models with the ultimate goal of reducing uncertainty in the predictions. This paper appears with the aim of compiling a large part of the knowledge about solar power forecasting, focusing on the latest advancements and future trends. Firstly, the motivation to achieve an accurate forecast is presented with the analysis of the economic implications it may have. It is followed by a summary of the main techniques used to issue the predictions. Then, the benefits of point/regional forecasts and deterministic/probabilistic forecasts are discussed. It has been observed that most recent papers highlight the importance of probabilistic predictions and they incorporate an economic assessment of the impact of the accuracy of the forecasts on the grid. Later on, a classification of authors according to forecast horizons and origin of inputs is presented, which represents the most up-to-date compilation of solar power forecasting studies. Finally, all the different metrics used by the researchers have been collected and some remarks for enabling a fair comparison among studies have been stated.
In this paper, an innovative sensor suited to perform real-time measurements of operating voltage and current, open-circuit voltage, and short-circuit current of string-connected photovoltaic (PV) panels is presented. An effective disconnection system ensures that the sensor does not affect the behavior of the string during the measurement phase and offers many benefits like the automatic detection of bypass events; moreover, the sensor does not require additional cables thanks to a wireless communication and a power supply section based on energy harvesting. An extensive experimental campaign is performed to prove the reliability and usefulness of the sensor for continuous monitoring of PV plants. The capability to detect faults and accurately localize malfunctioning panels in a PV string is highlighted.