Review of electric vehicles’ charging data anomaly

detection based on deep learning

Qi Feng Hengjie Li* Yun Zhou* Donghan Feng
School of Electrical and School of Electrical and Dept. of Electrical Dept. of Electrical
Information Engineering, Information Engineering, Engineering, Engineering,
Lanzhou University of Lanzhou University of Shanghai Jiao Shanghai Jiao
Technology Technology Tong University Tong University
Lanzhou, China Lanzhou, China Shanghai, China Shanghai, China
[email protected] [email protected] [email protected] [email protected]

Yun Wang Yun Su

Electric Power Electric Power
2022 Power System and Green Energy Conference (PSGEC) | 978-1-6654-9993-4/22/$31.00 ©2022 IEEE | DOI: 10.1109/PSGEC54663.2022.9881073

Research Institute, Research Institute,

State Grid Shanghai State Grid Shanghai
Electric Power Company Electric Power Company
Shanghai, China Shanghai, China
[email protected] [email protected]

Abstract—Anomalies in data may be caused by errors in the anomalous data in a data set that is different from the rest of the
data, but sometimes the appearance of anomalous data often also data. However, sometimes anomalies in data also suggest the
suggests the emergence of a new domain or a new underlying existence of a new, previously unknown domain. The field of
process. Anomaly detection has therefore been an active area of anomaly data detection has undergone decades of development
research for decades. This paper is based on what anomaly since 1960, but its specificity has led to a very active field of
detection is, the main areas of application for anomaly detection, anomaly detection for decades [2]. As a result, anomaly
and the challenges of anomaly detection at this stage. Deep detection systems are often considered an important component
learning-based anomaly detection is proposed based on the of various systems.
unpredictability, diversity, and sparsity that exist in anomaly
detection, and deep learning anomaly detection algorithms are Anomaly detection systems are used in many areas of the
classified into unsupervised deep anomaly detection algorithms daily life of their data-driven approach based on the detection of
and semi-supervised deep anomaly detection algorithms based on anomalous data. The literature [3] applies it to anti-fraud in the
the use of anomaly labels. This paper reviews and analyses the financial industries and credit cards for fraud detection by
research on these two algorithms, and elaborates on the treating fraud or financial risks as anomalies; The literature [4]
advantages and disadvantages of each type of anomaly detection has applied it to the detection of rare disease abnormalities such
algorithm when applied. Deep anomaly detection is also applied to as early-onset Alzheimer's disease as well as chronic cerebral
the anomaly detection of charging data and is classified according
infarction, cancer, etc. by treating rare diseases as abnormalities;
to the different types of charging data, into deep anomaly detection
The literature [5] by treating the intrusions in network traffic as
for electric vehicles, deep anomaly detection for charging stations,
and deep anomaly detection for charging poles. Finally, we
anomalous data applying them to the daily maintenance of
provide an outlook on various algorithms for anomaly detection network security; In the literature [6], anomaly detection
and future applications of anomaly detection in charging data. systems are used to detect or predict the faults that may occur
during the operation of machinery in real-time monitoring; In
Keywords—deep learning, anomaly detection, unsupervised, the literature [7], performs graph structure detection or
semi-supervised, charging data anomaly population detection by anomaly detection systems to detect
outbreak points of epidemics, etc.
I. INTRODUCTION
However, there are certain challenges in the development
Anomaly detection is also commonly referred to as outlier and daily application of anomaly detection. Such as [8]: Since
detection or novelty detection. Anomaly detection generally anomalous data or frequent frauds are constantly iterated as the
refers to the stage of identifying anomalous data that system is updated, it is difficult for the anomaly detection system
significantly deviate from the majority of the data [1]. Anomaly to predict future anomalies based on the anomalous data already
detection is mainly a data-driven approach to identify found and the frauds that occurred in the past, which shows the
This work was sponsored in part by the National Natural Science unpredictability of anomaly detection; Moreover, there is a
Foundation of China (52167014), in part by the Science and Technology certain degree of diversity in anomaly detection, as the causes of
Commission of Shanghai Municipality (21DZ2204800), and in part by the
anomalous data are various, and multiple causes affect each
Science and Technology Project of State Grid Shanghai Municipal Electric
Power Company of China (52094022000G). other and are implicated in the other, so it is difficult to judge

978-1-6654-9993-4/22/$31.00 ©2022 IEEE

337
Authorized licensed use limited to: ULAKBIM UASL - YILDIZ TEKNIK UNIVERSITESI. Downloaded on October 13,2024 at 17:35:32 UTC from IEEE Xplore. Restrictions apply.
 the causes of anomalous data through simple anomalous data; obtaining anomalous data and the imbalance of anomalous data
There is also sparsity in anomaly detection, and it is difficult to [10]. As a result, machine learning usually defines a scoring
get better training results for anomaly detection models because function to rank the degree of anomalies of the data and ranks
of the presence of the extreme imbalance in the anomaly data the data in the scoring function from the highest to the lowest
detected by the anomaly detection system. anomaly scores, and the data with high anomaly scores are
defined as anomalous data. Unsupervised anomaly detection is
Based on deep learning, it can perform data dimensionality a very important research area, in particular in the field of
reduction for high-dimensional anomalous data and data fundamental research in machine learning and the identification
expansion for unbalanced anomalous data, etc [9]. Deep of anomalous data. Unsupervised anomaly detection is therefore
learning can effectively target and solve the problems of divided into three classifications according to algorithmic ideas
unpredictability, variety, and sparsity in anomaly detection. as follows:
Most existing anomaly detection systems are therefore deep
anomaly detection systems built on deep learning. Deep A. Clustering-based algorithm
anomaly detection refers to anomaly detection of anomalous The essence of a clustering based algorithm is that data
data by learning features through the neural network or by a points are divided and defined in such a way that aggregations
direct output of outlier scores. In the field of solving very of normal data are defined as a cluster and anomalous data exists
complex anomaly data detection, deep anomaly detection in isolation. It is therefore particularly important for clustering
performs better and has higher performance than the general algorithms how the number of clusters is chosen, otherwise it
common anomaly detection. will result in more normal data being classified as abnormal data
sets or abnormal data in the form of small clusters being
considered as normal data. So in an anomaly detection system,
it is essential to analyze each dataset and set a specific parameter
according to the characteristics of that dataset to ensure the
effect of clustering.
The literature [11] uses a SOM (Self-Organizing Map)
algorithm based on the clustering idea to treat the da-ta further
away from the cluster center as anomalous data and applies the
SOM algorithm to anomaly detection in cloud infrastructure to
achieve good anomaly detection results. The literature [12]
applies the K-means algorithm to anomaly detection in the
weblogs, which detects anomalies in large-scale datasets
constructed from logs from heterogeneous sources, treating data
far from the cluster centers as anomalous data thereby achieving
the goal of detecting anomalies on a large number of unlabeled
data logs.
B. Depth-based algorithms
The depth-based algorithm essentially involves mapping all
the data in the dataset into the hierarchical structure of the k-
dimensional space, and assuming that the abnormal data are
Fig. 1. Deep anomaly detection framework diagram
distributed at the periphery of the hierarchical structure of the k-
dimensional space, while the normal data gradually approach the
As shown in Figure 1, this paper classifies depth anomaly
center of the hierarchical structure as the depth deepens. The
detection into unsupervised depth anomaly detection, semi-
literature [13] used an MVE (minimum volume ellipsoid)
supervised depth anomaly detection, and supervised depth
estimator to perform anomaly detection. MVE is based on fitting
anomaly detection according to the presence or absence of
a model of the distribution of probabilities for fifty percent or
supervision. But supervised deep anomaly detection trains
more points of data to the boundary of the smallest ellipsoidal
binary or multiclass classification discriminators by using labels
sphere shown by the solid ellipse and defining data that are not
for both normal and anomalous data. While supervised deep
within this boundary as anomalous data.
anomaly detection methods have improved performance,
because of the lack of labeled training samples, these methods Unsupervised deep anomaly detection based on depth idea
are not as popular as semi-supervised deep anomaly detection or mostly uses isolated forest algorithm in anomaly detection of
unsupervised deep anomaly detection. In addition, the high-dimensional data. The basic idea of the isolated forest
performance of using supervised deep anomaly detection is sub- algorithm is to randomly cut the sample space by the tree model
optimal due to the imbalance of the anomaly data. Therefore method to determine anomalous data, and because of its special
supervised deep anomaly detection will not be considered in this characteristics to the extent that the isolated forest algorithm is
study. highly accurate and fast when dealing with big data [14]. The
literature [15] addresses the poor performance of the isolated
II. UNSUPERVISED DEEP ANOMALY DETECTION forest algorithm that is prone to local anomaly detection as well
In many applications, labeled training sets for anomalous as high-dimensional anomaly detection based on the feature that
data are not easily available because of the difficulty of

338
Authorized licensed use limited to: ULAKBIM UASL - YILDIZ TEKNIK UNIVERSITESI. Downloaded on October 13,2024 at 17:35:32 UTC from IEEE Xplore. Restrictions apply.
 anomalous pixels are more easily detected than general The algorithm for the semi-supervised anomaly detection
background pixels and are more easily isolated after detection. model generated based on the idea of linear regression is the
PCA (principal component analysis) based on the unsupervised
C. Proximity-based algorithms anomaly detection model [19]. The main idea of PCA is to map
The neighborhood-based algorithm mainly discriminates n-dimensional features to k-dimensions, which are new
based on the fact that normal data are located in the dense orthogonal features also known as principal components, and are
neighborhood, while abnormal data are adjacent to more sparse k-dimensional features reconstructed from the original n-
data, so the neighborhood-based algorithm is also defined as a dimensional features. The PCA algorithm has been used for
density-based algorithm. semi-supervised deep anomaly detection in detecting anomalous
The discrimination method of the density-based anomaly data using the anomaly data detection model classifier.
detection system is mainly by calculating the density of each B. Proximity-based approach
data area in the dataset and defining the areas with a lower For routine research, the discovery of abnormal data is given
density as outlier areas. At this stage of research, the more weight than the discovery of normal data. In such cases
discrimination method of density-based anomaly detection where anomalous data are more important, a semi-supervised
systems mainly uses LOF (local outlier factors) [16]. The local deep anomaly detection algorithm based on the idea of
outlier is defined differently from the traditional anomaly proximity is presented, which usually depends on the degree of
detection system. The local outlier assigns a value to each data isolation of the anomalous data from the surrounding normal
that indicates a local outlier value relative to its neighborhood, data [20].
and the larger the local outlier value, the lower the neighborhood
density and the more likely the data is to be anomalous. The OCC and ASAD anomaly detection classifiers first used
However, it is difficult to determine the minimum neighborhood a local outlier based on the idea of proximity. LOF identify
in LOF, and the computational complexity and time complexity anomalous data by comparison of the target data with its
increase with the increase of data dimensionality. neighboring data, and these data belonging to the normal
category and those falling into the anomalous category are
III. SEMI-SUPERVISED DEEP ANOMALY DETECTION discriminated against according to their neighboring data. As a
The semi-supervised deep anomaly detection model defines result, when the data contains few data features but also has
that only a single class label exists for all training data. Semi- many anomaly detection objects, the anomaly detection
supervised deep anomaly detection models judge anomalous performance of LOF is more accurate, the model performance
data in general based on the scoring criterion for scoring will be more stable, and the classification results will be more
anomalous data, which has the following properties [17]: data in accurate. In the literature [21], a novel outlier detection
the input space and learning feature space that is close to each algorithm, called kNN-LOF, is proposed to address the
other are more likely to be sharing the same label of proximity limitations of existing outlier detection methods, and
and continuity; robustness of discriminative attributes that experimental results show that the algorithm has good
preserve the distinction between normal and anomalous data. performance in improving outlier detection accuracy.
However, the semi-supervised deep anomaly detection model C. Probability-based algorithms
that also uses labeled data from one category offers considerable
performance improvements compared to the unsupervised deep The probability-based algorithms are also known as
anomaly detection model. The key to semi-supervised deep stochastic models, and the random variables that exist in these
anomaly detection is how to make full use of a large amount of models are variables that can take on different values at random
unlabeled diverse data to improve the detection ability of the and can be either discrete or continuous. Discrete random
training model on computationally large unknown datasets. The variables have a finite or countably infinite number of states,
current semi-supervised deep anomaly detection algorithms are which do not have to be integers but may be named states with
divided into three main categories as follows. no numerical value. Stochastic models have therefore been
applied to anomaly detection, and have shown good
A. Linear regression classification results through practical applications.
Linear regression is an important application in statistics. ABOD (angle-based outlier detection) is based on measuring
Linear regression usually refers to the procedure of predicting a the angular difference between an object, which is usually
particular dependent variable through a series of independent abstracted as a vector, and other objects for anomaly data
variables [18]. In linear regression, there are special not easily detection. Although ABOD provides a good improvement in
detectable associations between variables that define different dimensionality compared to other anomaly detection methods,
dimensions, and these different dimensions can be related by a ABOD has some problems in terms of time complexity.
matrix of correlation coefficients. Under this definition of linear Consequently, Kriegel et al. proposed another alternative to the
regression, the anomalous data can therefore be defined database samples based on ABOD, FastABOD, to improve the
according to the magnitude of the effect that the other time complexity of ABOD [22].
independent variables have on the dependent variable. So linear
regression is used for anomalous data detection mainly for high- IV. DEEP ANOMALY DETECTION OF CHARGING DATA
dimensional data, and the impact of the anomalous data on the
model performance is avoided by data dimensionality reduction As the scale of EVs (electric vehicles) continues to expand,
of the high-dimensional data. the proper operation of EVs is particularly important [23]. The
normal operation of EVs is closely related to the daily operation

339
Authorized licensed use limited to: ULAKBIM UASL - YILDIZ TEKNIK UNIVERSITESI. Downloaded on October 13,2024 at 17:35:32 UTC from IEEE Xplore. Restrictions apply.
 of charging stations and charging posts. The normal operation electrical equipment abnormalities manifest as lights not
of EVs relates to people's daily travel as well as transportation lighting, dim lights, flickering, fuse ringing, etc. The abnormal
development. Whether the EVs charging station, as the causes of the above are usually filament burned, loose wire, poor
connection point between EVs users and the power grid, can grounding, broken or short circuit, charging voltage adjustment
work properly or not is the key to the normal operation of EVs, is too high, and a variety of switch failures.
and the normal operation of the charging station is related to the
normal operation and daily profit of the charging station. So For the above-mentioned anomalies of EVs, anomaly
ensuring the proper operation of the triad of EVs, charging detection of possible anomalies of EVs will be performed by
stations, and charging posts is crucial to support EVs combining deep learning. The literature [26] obtained ideas from
development, promote efficient EVs use, and ensure the statistics to detect anomalies in battery data of EVs and obtained
operational efficiency and safety of the grid [24]. a more desirable result. In the literature [27], two steps of
improved PCA and density-based depth anomaly detection were
used to achieve thermal runaway prediction.
B. Abnormality detection of charging stations
According to the survey, it is found that the abnormalities
that appear in the charging station at this stage are mainly
reflected in the abnormalities of external components divided
into [28]: the inability to coin in the charging station; no voltage
when the charging station is plugged into the electric vehicle;
the charging station display does not light up; the charging
station connection cable is disconnected; the charging station
cannot be charged; the charging station voice messy report and
other external abnormalities.
The charging station can therefore be considered as a whole
and holistic anomaly detection can be done based on the whole
charging station. The literature [29] proposes a framework for
collecting EVs battery-related data from several benefit
recipients with different characteristics, based on ideas derived
from an isolated forest with two characteristics. The literature
[30] detects the charging data of EVs and the anomalies of the
charging system in charging stations by building a concerned
model.
C. Abnormal detection of charging piles
Fig. 2. Charging data anomaly framework diagram
Charging pile abnormalities are divided into internal and
However, the normal operation and maintenance of these external abnormalities. Most of the external anomalies of the
three systems are mostly dependent on the anomaly detection charging post are physical damage to the display, charging gun,
system. In this article, charging data anomalies are classified as emergency stop button, indicators, and other external
shown in Figure 2, and divided into three categories: EVs components of the charging post. Internal anomalies in charging
charging data anomalies, charging station data anomalies, and stations are generally attacks against EVs charging stations,
charging pile data anomalies. users, and the grid [31]. The three attacks mentioned above lead
to charging data leakage inside the charging pile, leakage of
A. Deep anomaly detection for electric vehicle personal information of EVs users, corruption of charging and
The abnormalities of EVs can be divided into three main scheduling configuration parameters, and destruction of battery
categories: abnormal power system, abnormal vehicle chassis conditions.
components, and abnormal electrical equipment [25]. The literature [32] proposed an SVM (support vector
Power system abnormalities are mainly caused by the machine) based method for detecting power theft at charging
abnormal power battery system and the abnormal electric motor piles. The literature [33] developed a fault diagnosis system for
drive system. The power battery system abnormalities can be electric vehicle charging units based on deep learning to
divided into three categories according to the part where the accurately locate and quickly resolve charging unit faults.
abnormality occurs, which are single battery abnormality,
V. CONCLUSION
battery management system abnormality, and line or connection
abnormality. The motor drive system abnormalities include both This paper details the significance of deep learning
motor abnormalities and motor controller abnormalities. The applications in anomaly detection, daily data detection, and
majority of abnormalities in the chassis parts are caused by the anomaly detection model operation problems. Accurate
transmission, which is responsible for shifting, torque, modeling of anomaly detection systems is an important
reversing, and using neutral to temporarily cut off the power, as foundation for their evaluation and operational optimization
well as the steering system, braking system, driving system, and studies, but anomaly detection models are unpredictable,
other systems that control the chassis form of the car. Most diverse, and sparse, and deep learning provides a new direction

340
Authorized licensed use limited to: ULAKBIM UASL - YILDIZ TEKNIK UNIVERSITESI. Downloaded on October 13,2024 at 17:35:32 UTC from IEEE Xplore. Restrictions apply.
 for solving these problems. At present, the research related to [14] Barbariol T, Chiara F D, Marcato D, et al. A Review of Tree-Based
anomaly detection based on deep learning is more mature. Approaches for Anomaly Detection[J]. Control Charts and Machine
Learning for Anomaly Detection in Manufacturing, 2022: 149-185.
Accordingly, this paper classifies deep anomaly detection
[15] Song X, Aryal S, Ting K M, et al. Spectral-Spatial Anomaly Detection of
learning into unsupervised deep anomaly learning and semi- Hyperspectral Data Based on Improved Isolation Forest[J]. IEEE
supervised deep anomaly learning based on the use of deep Transactions on Geoscience and Remote Sensing, 2021.
learning labels. And based on the different algorithmic ideas, [16] Alsini R, Almakrab A, Ibrahim A, et al. Improving the outlier detection
they are then presented in a detailed classification. method in concrete mix design by combining the isolation forest and local
outlier factor[J]. Construction and Building Materials, 2021, 270: 121396.
Due to the increasing scale of EVs, the proper operation of [17] Ruff L, Vandermeulen R A, Görnitz N, et al. Deep semi-supervised
EVs is particularly important. So this paper applies deep anomaly detection[J]. arXiv preprint arXiv:1906.02694, 2019.
learning-based anomaly detection to the anomaly detection of [18] Rustam F, Reshi A A, Mehmood A, et al. COVID-19 future forecasting
charging data, divides the anomalies of charging data into using supervised machine learning models[J]. IEEE access, 2020, 8:
anomalies of EVs, anomalies of charging stations, and 101489-101499.
anomalies of charging piles, and introduces in detail the causes [19] Merrill N, Olson C C. Unsupervised ensemble-kernel principal
of each anomaly and the process of anomaly detection by deep component analysis for hyperspectral anomaly detection[C]. Proceedings
anomaly detection system at the present stage. of the IEEE/CVF conference on computer vision and pattern recognition
workshops, 2020: 112-113.
REFERENCES [20] Tsukada M, Kondo M, Matsutani H. A neural network-based on-device
learning anomaly detector for edge devices[J]. IEEE Transactions on
[1] Yahaya S W, Lotfi A, Mahmud M. A consensus novelty detection Computers, 2020, 69(7): 1027-1044.
ensemble approach for anomaly detection in activities of daily living[J].
[21] Xu H, Zhang L, Li P, et al. Outlier detection algorithm based on k-nearest
Applied Soft Computing, 2019, 83: 105613.
neighbors-local outlier factor[J]. Journal of Algorithms & Computational
[2] Ni F, Zhang J, Noori M N. Deep learning for data anomaly detection and Technology, 2022, 16: 17483026221078111.
data compression of a long‐ span suspension bridge[J]. Computer‐
[22] Ye H, Kitagawa H, Xiao J. Continuous angle-based outlier detection on
Aided Civil and Infrastructure Engineering, 2020, 35(7): 685-700.
high-dimensional data streams[C]. Proceedings of the 19th International
[3] Hilal W, Gadsden S A, Yawney J. A Review of Anomaly Detection Database Engineering & Applications Symposium, 2015: 162-167.
Techniques and Applications in Financial Fraud[J]. Expert Systems with
[23] Sun X, Li Z, Wang X, et al. Technology development of electric vehicles:
Applications, 2021: 116429.
A review[J]. Energies, 2019, 13(1): 90.
[4] Alvi A M, Siuly S, Wang H. Neurological abnormality detection from
[24] Sanguesa J A, Torres-Sanz V, Garrido P, et al. A review on electric
electroencephalography data: a review[J]. Artificial Intelligence Review,
vehicles: Technologies and challenges[J]. Smart Cities, 2021, 4(1): 372-
2021: 1-38.
404.
[5] Ahmed M, Mahmood A N, Hu J. A survey of network anomaly detection
[25] Tian Liu, Analysis of electric vehicle charging anomaly detection based
techniques[J]. Journal of Network and Computer Applications, 2016, 60: on BP neural network[J]. Inner Mongolia Coal Economy, 2019.
19-31.
[26] Ma M, Wang Y, Duan Q, et al. Fault detection of the connection of
[6] He H, Zhang J, Wang Y, et al. Detecting Electric Vehicle Battery Failure
lithium-ion power batteries in series for electric vehicles based on
via Dynamic-VAE[J]. arXiv preprint arXiv:2201.12358, 2022.
statistical analysis[J]. Energy, 2018, 164: 745-756.
[7] Ibrahim M R, Youssef S M, Fathalla K M. Abnormality detection and
[27] Li D, Zhang Z, Wang Z, et al. Timely Thermal Runaway Prognosis for
intelligent severity assessment of human chest computed tomography
Battery Systems in Real-world Electric Vehicles Based on Temperature
scans using deep learning: a case study on SARS-COV-2 assessment[J].
Abnormality[J]. IEEE Journal of Emerging and Selected Topics in Power
Journal of Ambient Intelligence and Humanized Computing, 2021: 1-24.
Electronics, 2022.
[8] Deng L, Lian D, Huang Z, et al. Graph Convolutional Adversarial
[28] Elma O. A dynamic charging strategy with hybrid fast charging station
Networks for Spatiotemporal Anomaly Detection[J]. IEEE Transactions
for electric vehicles[J]. Energy, 2020, 202: 117680.
on Neural Networks and Learning Systems, 2022.
[29] Jin R, Wei B, Luo Y, et al. Blockchain-based data collection with efficient
[9] Janiesch C, Zschech P, Heinrich K. Machine learning and deep
anomaly detection for estimating battery state-of-health[J]. IEEE Sensors
learning[J]. Electronic Markets, 2021, 31(3): 685-695.
Journal, 2021, 21(12): 13455-13465.
[10] Luo M, Wang K, Cai Z, et al. Using imbalanced triangle synthetic data
[30] Li Y, Zhang L, Lv Z, et al. Detecting anomalies in intelligent vehicle
for machine learning anomaly detection[J]. Comput., Mater. Continua,
charging and station power supply systems with multi-head attention
2019, 58(1): 15-26.
models[J]. IEEE Transactions on Intelligent Transportation Systems,
[11] Stephanakis I M, Chochliouros I P, Sfakianakis E, et al. Hybrid self- 2020, 22(1): 555-564.
organizing feature map (SOM) for anomaly detection in cloud
[31] Yong X, Ji W. Research on Detection and Fault Diagnosis Technology of
infrastructures using granular clustering based upon value-difference
Electric Vehicle Charging Facilities[C]. Journal of Physics: Conference
metrics[J]. Information Sciences, 2019, 494: 247-277.
Series, 2020: 022100.
[12] Henriques J, Caldeira F, Cruz T, et al. Combining k-means and xgboost
[32] Hong H, Su Y, Zheng P, et al. A SVM-based detection method for
models for anomaly detection using log datasets[J]. Electronics, 2020,
electricity stealing behavior of charging pile[J]. Procedia Computer
9(7): 1164.
Science, 2021, 183: 295-302.
[13] Mahmoudzadeh A, Azimi I, Rahmani A M, et al. Lightweight
[33] Gao D-X, Hou J-J, Liang K, et al. Fault diagnosis system for electric
photoplethysmography quality assessment for real-time IoT-based health
vehicle charging devices based on fault tree analysis[C]. 2018 37th
monitoring using unsupervised anomaly detection[J]. Procedia Computer
Chinese Control Conference (CCC), 2018: 5055-5059.
Science, 2021, 184: 140-147.

341
Authorized licensed use limited to: ULAKBIM UASL - YILDIZ TEKNIK UNIVERSITESI. Downloaded on October 13,2024 at 17:35:32 UTC from IEEE Xplore. Restrictions apply.