Exploration of the Potential of Process Mining for Intrusion Detection in
Smart Metering
G
¨
unther Eibl
1
, Cornelia Ferner
1
, Tobias Hildebrandt
2
, Florian Stertz
2
, Sebastian Burkhart
1
,
Stefanie Rinderle-Ma
2
and Dominik Engel
1
1
Josef Ressel Center for User-Centric Smart Grid Privacy, Security and Control,
Salzburg University of Applied Sciences, Urstein S
¨
ud 1, 5412 Puch/Salzburg, Austria
2
Workflow Systems and Technology, University of Vienna, W
¨
ahringer Straße 29, 1090 Vienna, Austria
{guenther.eibl, cornelia.ferner, sebastian.burkhart, dominik.engel}@en-trust.at,
{tobias.hildebrandt, florian.stertz, stefanie.rinderle-ma}@univie.ac.at
Keywords:
Process Mining, Intrusion Detection, Smart Grids, Smart Metering.
Abstract:
Process mining is a set of data mining techniques that learn and analyze processes based on event logs. While
process mining has recently been proposed for intrusion detection in business processes, it has never been
applied to smart metering processes. The goal of this paper is to explore the potential of process mining for
the detection of intrusions into smart metering systems. As a case study the remote shutdown process has been
modeled and a threat analysis was conducted leading to an extensive attack tree. It is shown that currently
proposed process mining techniques based on conformance checking do not suffice to find all attacks of the
attack tree; an inclusion of additional perspectives is necessary. Consequences for the design of a realistic
testing environment based on simulations are discussed.
1 INTRODUCTION
In an effort to modernize traditional energy grids and
move towards smart grids, information and commu-
nication technologies are used to enable communica-
tion and real-time interaction between producers, con-
sumers and other stakeholders. In this transition smart
meters play an important role, as they enable the in-
clusion of end-user homes.
While smart grids have the potential to achieve
substantial benefits, also new threats arise (Berthier
et al., 2010). In particular, the new possibility for
a remote turnoff can also be seen as a vulnerability
that could be exploited by an attacker to accomplish a
large scale turnoff. Especially for critical infrastruc-
tures, there is a need to investigate possible weak-
nesses and attacks before being put into operation.
This implies that envisaged processes should be mod-
eled and checked for weaknesses. This paper explores
the potential of process mining for intrusion detection
in smart metering with particular emphasis on detect-
ing intrusions that may lead to a large-scale turnoff.
Process mining with subsequent conformance
checking for anomaly detection has been sug-
gested for security enhancement (Van der Aalst and
de Medeiros, 2005). However, anomaly detection is
not specific enough. Anomalies are created using
simple heuristics such as by interchanging of events
(Bezerra et al., 2009), (Bezerra and Wainer, 2013)
which is likely not describing anomalies arising from
intrusions. Additionally, since conformance check-
ing only considers the control flow the benefit of this
technique is still unclear. Anomaly detection using
process mining was combined with misuse detection
considering four types of attacks related to the organi-
zational perspective (Jalali and Baraani, 2012). Con-
formance checking can be applied for the detection of
violations of security requirements like e.g. Separa-
tion or Binding of Duties requirements (Accorsi and
Stocker, 2012).
This paper has two main contributions: instead of
only creating anomalies, attacks on the smart meter
shutdown process are more realistically and system-
atically derived and consequences for future enhance-
ment of the evaluation methodology are discussed.
Shortcomings of the current process mining method-
ology using conformance checking for intrusion are
shown yielding consequences for future enhancement
of the intrusion detection methodology.
The paper is structured as follows. Background
about process mining, attack-defense trees and evalu-
ation by simulations are provided in Section 2. In Sec-
38
Eibl, G., Ferner, C., Hildebrandt, T., Stertz, F., Burkhart, S., Rinderle-Ma, S. and Engel, D.
Exploration of the Potential of Process Mining for Intrusion Detection in Smar t Metering.
DOI: 10.5220/0006103900380046
In Proceedings of the 3rd International Conference on Information Systems Security and Privacy (ICISSP 2017), pages 38-46
ISBN: 978-989-758-209-7
Copyright
c
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
tion 3 the shutdown use case is modeled, attacks are
systematically derived and detection methods based
on process mining are proposed. Several general find-
ings, benefits and limitations of the method are dis-
cussed in Section 4. Finally, Section 5 contains con-
clusion and outlook.
2 PRELIMINARIES
2.1 Process Mining
The goal of process discovery is to automatically dis-
cover a process model by analyzing logs of recorded
process events (Van der Aalst, 2011). The process
model that is found can then be analyzed in order to
improve or correct the process.
A case (or instance) is a specific execution of a
process. Each case basically consists of a sequence
of activities (or events) which are typically identified
by their name, e.g. “Breaker interrupts power supply”
(Figure 1). The process model describes the decisions
that influence the process path and the ordering of ac-
tivities (control-flow perspective).
Process instances can be described by additional
attributes. The organizational perspective focuses on
users and their roles and how they influence the pro-
cess, e.g., it specifies which person performed the ac-
tivity. The case perspective looks at properties or data
of cases like a customer’s energy consumption over a
certain time interval. The time perspective describes,
e.g., durations of activities and is particularly suited
to detect bottlenecks.
In conformance checking, executed process in-
stances are compared to a process model. A case is
considered valid, if it can be created by the process.
By trying to replay the case it is checked, if the or-
dering of its activities is compatible with the process.
Conformance checking can be used in two different
ways: on one hand it can be used to rate the process
by the proportion of valid traces of the event log. On
the other hand, for anomaly detection a trace is con-
sidered an anomaly if it is not a valid instance of the
process.
2.2 Attack-defense Trees
An attack tree (Salter et al., 1998) is an intuitive vi-
sual representation of attacks combined with formal
semantics and algorithms that enable qualitative and
quantitative analysis. The root of an attack tree is the
goal of the attack. The tree is then iteratively con-
structed by dividing each goal into sub-goals where
either all sub-goals must be reached to reach the goal
(AND) or any sub-goal suffices (OR). The refinement
is repeated until all sub-goals can be reached by ba-
sic actions. Attack trees have been extended in many
different directions (Kordy et al., 2014).
Since in this paper not only attack but also detec-
tion methods should be modeled, an attack tree is not
sufficient as a modeling structure. Based on the sur-
vey (Kordy et al., 2014), two different approaches can
be identified as prime candidates for modeling both
the attack and the detection method. An attack coun-
termeasure tree (ACT) (Roy et al., 2012) can model
attack, detection and mitigation events at any part of
the tree and calculate probabilities of events. Attack
defense trees (ADT) (Kordy et al., 2012) can model
interleaving attacker and defender actions and pro-
vides several general methods for attribute computa-
tion like probability of success. While attack defense
trees have the advantage that defensive actions can be
refined, attack countermeasure trees have the advan-
tage that both detection and mitigation events can be
modeled.
2.3 Evaluation of Process Mining-based
Intrusion Detection by Simulations
Similar to existing approaches we propose to study
and evaluate intrusion detection using simulations.
There are several reasons for using simulations: (i)
in contrast to a real system ground truth is avail-
able which permits a comparison of different detec-
tion methods, (ii) an assessment is possible prior to a
roll-out which is important especially for highly criti-
cal infrastructures and (iii) variants of the system and
its processes can be compared beforehand.
A simulation environment for intrusion detection
needs to simulate (i) the process with its variations
and (ii) effects of intrusions. For both normal and
anomalous process instances, execution logs are be-
ing created. While the former can be achieved in a
typical process mining environment, the latter needs
an additional effort.
As a first step in that direction, both the model and
the resulting traces can be transformed (Stocker and
Accorsi, 2013). Model transformations persistently
change the model by transforming AND-gateways to
XOR-gateways and vice versa or swapping of two ac-
tivities. Trace transformations are then performed on
valid traces of the transformed model. Some trace
transformers work by simply changing delays in the
process execution or skipping of activities. Other
trace transformations create policy violations related
to the organizational perspective like e.g. authentica-
tion or binding-of-duties violations.
In this paper, the effect of intrusions aiming at a
Exploration of the Potential of Process Mining for Intrusion Detection in Smart Metering
39
Control System
Control System
start 2.1
Send
shutdown
order
Display
status
end 2.1
Smart Meter
Smart Meter
Breaker
interrupts
power
supply
Display
status
Smart
Meter
Log entry
Confirm
status
start SM
end SM
Backoffice
Backoffice
Message
registration
and
processing
Move-out without
new tenant
No payment after
dunning process (-> 9.1)
Deregistration of
existing customer
start
backoffice
end
backoffice
Figure 1: Use case 2.1 Shutdown after dunning or customer change.
smart meter shutdown on traces of the smart meter
shutdown process is investigated. These traces should
be more realistic than traces that are created by trans-
formations of valid traces that e.g. perform a skip-
ping or swapping of arbitrary activities. On one hand
the smart meter shutdown process (Figure 1) is mod-
eled as realistically as possible on the basis of a tex-
tual description of smart metering use cases that the
major energy providers in Austria have agreed upon
(Oesterreichs-Energie, 2015). On the other hand pos-
sible attacks aiming at a shutdown are derived in a
systematic way and the consequences of these attacks
on the resulting traces are determined. Thus, the re-
sulting traces should be highly realistic and could be
further used for intrusion detection testing scenarios.
3 DERIVATION AND DETECTION
OF SHUTDOWN ATTACKS
In this section attacks on a rather simple, but very im-
portant process, the shutdown process, are derived.
Then their effects on the process are explored which
determines how they could be detected. Both at-
tacks and detection methods are modeled as an attack-
defense trees. The consequences of the attacks on the
traces give indications, how attacks could be simu-
lated more realistically.
In order to simplify the analysis, the intrusion is
assumed to happen at only a single location. In or-
der to properly perform conformance checking we as-
sume a dedicated intrusion detection system that col-
lects all events and stores them in a log. Each event
is stored together with the id of its process instance.
Additionally, it is assumed that the creation and stor-
age of the logs is unaffected by the intruder and that
the information about the process is available for the
analysis.
3.1 Smart Meter Shutdown Process
The process model of the remote smart meter shut-
down use case was created based on a smart meter-
ing use case document (Oesterreichs-Energie, 2015).
This document contains textual descriptions of smart
metering use cases that the major energy providers in
Austria have agreed upon. Use case 2.1 was identified
as the use case that models the remote turn-off due to
a customer deregistration or a moving out without a
new tenant. Based on the textual description of the
use case the shutdown process was modeled using the
software tool Signavio.
The resulting process is shown in Figure 1. It
starts with the reception of an incoming message that,
e.g., states that the customer moved out. The mes-
sage is registered in the backoffice which then sends
a shutdown order to the control system which in turn
sends a shutdown order to the smart meter. After in-
terrupting the power supply, the smart meter sends its
(new) status information back to the control system.
3.2 Choice of Analysis Methodology
Both attack countermeasure and attack-defense trees
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
40
Shutdown
Shutdown message
CS->SM
Changes at
CS->SM
Changes
at CS
Triggered by
shutdown message
BO->CS
Interrupt at SM
Detect
shutdown
Handling of
status messages
Figure 2: Top levels of the attack-defense tree.
(ADT) are general enough for modeling the trees of
this paper. Attack defense trees were chosen since a
software tool, called ADTools (Kordy et al., 2013),
exists that allows easy creation and editing of the
trees. Attacks are represented by red, elliptic nodes,
detection measures are shown as green rectangles
(Figure 2). All figures were created from the same
attack-defense tree by hiding the parts above or be-
low several nodes, which is indicated by black trian-
gles above or below the corresponding nodes.
The considered attacks aim at forcing an irregular
shutdown of smart meters. In the following, attacks
achieving a shutdown are systematically constructed
as an attack tree by starting with the activity “Breaker
interrupts power supply” and going back along the
process.
3.3 Attack-defense Tree Derivation
In order to realistically describe effects of attacks, in
this section attacks on the shutdown process are de-
rived. Then their effects on the process are explored
which determine (i) how they could be detected and
(ii) how attacks could be simulated more realistically.
3.3.1 Top of the Tree
The first level of the tree (Figure 2) corresponds to
two different final attacks that can lead to a shutdown
at the smart meter: (i) a shutdown message is sent
from the control system (CS) to the smart meter (SM)
and (ii) direct interruption at the smart meter (Figure
1).
The direct interruption at the smart meter (after
an intrusion to the smart meter) can not be detected
before the shutdown occurs by process mining which
is indicated by the fact that no countermeasure node
is below the node Interrupt at SM.
However, besides the two attacks, the first level
also shows the detection method Detect shutdown
which works independently from the attacks by de-
tecting the shutdown itself through consideration of
the status message that is sent after the shutdown took
place. This detection method is described in more de-
tail in Section 3.3.5 and depends on the way the at-
tacker handles status messages.
Going one step back in the process model (Fig-
ure 1) a shutdown message from the control system
(CS) to the smart meter (SM) can be achieved in three
different ways: either by changes (i) at the communi-
cation line CSSM, (ii) its origin CS or (iii) the part
before (Triggered shutdown message). In the follow-
ing, each of these 3 subgoals and its detection meth-
ods are described in more detail.
In order to execute some of the following detec-
tion methods, it is assumed that each message is ex-
panded to include also the ID of the underlying pro-
cess and the ID of the smart meter that should be
turned off.
3.3.2 Shutdown Message CSSM
A shutdown message from the control system to the
smart meter can be achieved by replaying an old shut-
down message or by creating a new shutdown mes-
sage (Figure 3). A replay attack can be directly
achieved by intercepting, possibly redirecting and
sending a shutdown message. It can be detected by
checking, if the received process ID is the ID of a cur-
rently active shutdown process. A redirection could
additionally be detected when the receiving smart me-
ter checks, if the shutdown message is including its
ID. A replay to the same receiver can be additionally
detected by comparing the received process ID with
all previously obtained process IDs.
The second possibility is to create an entirely new
shutdown message and send it to the smart meter. An
intruder unaware of the underlying processes will not
include a process ID, which can be detected immedi-
ately by checking the existence of a process ID.
Exploration of the Potential of Process Mining for Intrusion Detection in Smart Metering
41
Changes at
CS->SM
Replay
Check if
Pr-ID is active
New message
CS->SM
No process ID
Check existence
of a process ID
Create random
process ID
Conformance
checking
Figure 3: Attacks aiming at creating a shutdown message
from the control system (CS) to the smart meter (SM) at
the channel between them and the corresponding detection
methods.
A process-aware intruder could try to overcome
the problem by creating a random process ID. If the
created ID is completely new, conformance checking
would detect the situation by recognizing a process
which can not be a valid instance of the process model
because it starts with the interrupt. If the created ID
corresponds to a finished process or an active process
in the wrong state (or more generally with an incom-
patible marking) the shutdown activity is not allowed
(enabled) and the situation can be detected by confor-
mance checking.
A general property of the methodology can al-
ready be seen at this stage of the analysis: the intru-
sion itself is not detected. Instead, the consequences
of the intrusion when aiming at the shutdown such as
missing or wrong IDs or wrong “states” of the system
are detected. As such this approach works at a higher
(process) level and is complementary to methods that
detect the intrusion into the system.
It should be noted that the tree can be expanded at
the attack leaves by intrusion goals. For example, a
direct shutdown message from CS to the SM can be
sent after either an intrusion to the CS or an intrusion
to the communication channel between the CS and the
SM. This was in fact done in order to test the analysis
capability of ADT. By choosing the domain “satisfia-
bility of the scenario” in ADT, it can be evaluated, if
an undetected shutdown occurs or not given that dif-
ferent intrusions, attacks and detection methods take
place. Since the intrusions needed are rather obvious,
this expansion is not done for sake of readability of
the tree.
Changes
at CS
SM-IP
mapping
Receiving SM
checks target SM-ID
SM-customer
mapping
Receiving SM
checks target SM-ID
Figure 4: Attacks aiming at creating a shutdown message
from the control system (CS) to the smart meter (SM) by
changes at the CS and the corresponding detection methods.
3.3.3 Changes at the Control Center
Many malicious changes at the control center are pos-
sible. It is plausible that the information about the
linkage between smart meters, smart meter IDs and
customers is stored at the control system and not in
the backoffice. A modified mapping between smart
meters and their IDs results in shutdown messages
that are sent at the wrong smart meters (Figure 4).
This can trivially be detected if the receiving smart
meter compares the ID appended at the message with
its own.
A modified mapping between smart meters and
customers also results in shutdown messages that are
sent at the wrong smart meters. Consequently it can
be detected by the same method.
3.3.4 Shutdown Message BOCS
After an intrusion in the corresponding communica-
tion channel, the shutdown message from the back-
office to the control system can be achieved either
through a replay or through the creation of a new mes-
sage (Figure 5).
The replay has already been discussed in Sec-
tion 3.3.2 and can be detected with the same method.
The new message can be achieved (i) by creating and
sending a new message, (ii) by two changes at the
backoffice of (iii) by an external message to the BO.
The first possibility and its detection method has also
already been discussed in Section 3.3.2.
The next two attacks assume an intrusion in the
backoffice. If the registration of, e.g., a move-out
message is faked, a process ID is created and a shut-
down message is sent to the CS. Since the incoming
message event is not contained in the log, the cor-
responding trace is invalid. Therefore conformance
checking would recognize this attack.
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
42
Triggered by
shutdown message
BO->CS
Replay
Check if
Pr-ID is active
New message
BO->CS
Create and send
new message
No process ID
Check existence
of a process ID
Create random
process ID
Conformance
checking
Fake message
registration at BO
Conformance
checking
Fake move-out
flag at BO
Check if a
shutdown process
exists
Fake external
move-out message
to BO
Analyze
other processes
Figure 5: Attacks aiming at creating a shutdown message from the backoffice (BO) to the control system (CS) and the
corresponding detection methods.
An even simpler attack could try to start the shut-
down process by declaring that the customer has
moved out. In the simplest case this attack could con-
sist of changing a single boolean entry in a database.
Note, however, that the modeled process needs to start
with one of three incoming messages from the cus-
tomer (denoted as double-circled events in Figure 1).
Therefore, due to the absence of an incoming start
message the shutdown process does not start. Conse-
quently, the change of a boolean entry does not suffice
for triggering a shutdown since no shutdown process
exists. It should, however, be noted that in less strict
processes such a change may suffice to trigger a shut-
down.
Finally, the next attack considers the first step of
the process and of sending a move-out message from
outside to the backoffice. Although this attack is out
of scope of this paper since no intrusion is necessary,
it is interesting to study the implications of this attack.
Since the reception of the message at the backoffice
starts the shutdown process, it clearly can not be de-
tected with process mining of the shutdown process.
It turns out that the activity message registration and
processing is itself more a process than a simple ac-
tivity. Therefore, it should be modeled as a business
process which could enable the detection of this non-
intrusive attack.
3.3.5 Handling of Status Messages
So far, the part of the process before the shutdown has
been considered. Except the direct shutdown after in-
trusion into the smart meter, in principle any of the
detection methods described above could detect the
attack before shutdown takes place. Now the part of
the process before the shutdown has been considered.
While the study of this process can not prevent a shut-
down, it could still be used to detect the intrusion in
order to e.g. prevent a large-scale shutdown.
Corresponding to the modeled process, after in-
terrupting the power supply the smart meter sends its
new status information back to the control system.
While the attacker has reached its shutdown goal, he
could try to remain undetected. The attacker has three
possibilities (Figure 6).
If the attacker does nothing, the status message
will be sent to the control center. Various ways to
detect the illegitimate shutdown are possible. In the
simplest case, a comparison with a simple list of
smart meters that should have received a remote turn-
Exploration of the Potential of Process Mining for Intrusion Detection in Smart Metering
43
Handling of
status messages
No special
action
Compare with
egitimate shutdown list
Inhibit
status messages
Time perspective
Fake status
message
No process ID
Check existence
of a process ID
Create random
process ID
Conformance
checking
Figure 6: Different ways the attacker can deal with status messages and corresponding detection methods.
off command will reveal the attack. For example,
this happens for the so far undetected shutdown per-
formed directly at the smart meter.
The simplest countermeasure of the attacker
would consist of preventing the smart meter from
sending status messages. Depending on the details
of the system, this could be detected by the absence
of regular status messages, i.e., by the consideration
of the time perspective. The resulting never-ending
process could for example be detected by consider-
ing the duration between the sending of the shutdown
message from the CS to the smart meter and the re-
ception of the status message.
Finally, the attacker could try to create and send a
new status message. This possibility is analogous to
the situation in Section 3.3.2 and can be detected as
described there.
4 DISCUSSION
In this section, several general properties of the dis-
cussed detection approach are discussed. The discus-
sion should clarify the benefits but also the limits of
intrusion detection by process mining when done in
the proposed way.
4.1 Consequences for Simulations of
Attacks
This analysis above has consequences for testing the
detection of intrusions by simulations: intrusions can
lead to traces where the first activities are skipped,
i.e., tasks from the beginning are missing (e.g., when
the attacker creates and sends a new shutdown mes-
sage). Moreover, the inhibition of sending the log
messages will result in process instances that do not
end. Therefore, while simulations that skip single
tasks or exchange two tasks may be adequate for
anomaly detection (Bezerra et al., 2009; Bezerra and
Wainer, 2013), they are unrealistic and thus not suited
for simulations of intrusions.
4.2 Consequences for Preprocessing
While never-ending traces also need to be considered
in the testing simulations, they also need to be taken
into account by the detection method. Some anomaly
detection methods apply a scoping step in the prepro-
cessing phase, where only finished processes remain
(Bezerra et al., 2009). While scoping can be benefi-
cial for anomaly detection, in the intrusion detection
scenario, scoping will throw away the never-ending
process instances which consequently lead to reduced
intrusion detection rates.
4.3 Need for other Perspectives
The same attack also reveals the need to take other
perspectives like the time perspective into account
considering, e.g., the duration between events. It is
likely that the organizational view plays an important
role especially in business processes like, e.g., the reg-
istration and processing of incoming messages in the
backoffice. The organizational perspective could also
assist in the detection of the wrong connection be-
tween customers and smart meters. The data perspec-
tive could help in clarifying the smart meter status by,
e.g., analyzing the energy consumption. As already
stated above, intrusion detection based on the control-
flow is more likely able to detect intrusions when the
intrusion is exploited by a malicious attacker. In con-
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
44
trast, by definition, an eavesdropping attack does not
change the way the system works. Therefore, the
control-flow is unchanged and other perspectives are
necessary for the detection of eavesdropping attacks.
4.4 Prohibition of the Shutdown
In order to prevent the shutdown, conformance check-
ing would have to be done before the breaker in-
terrupts. This requirement would also result in the
need to analyze incomplete traces. In order to pro-
hibit a large-scale turnoff due to an intrusion, con-
formance checking should be applied early enough,
i.e., before the shutdown is performed, asking for a
check by the intrusion detection system. Performing
all these checks in time is suspected to require big ef-
forts. However, even in this case the intrusion into the
smart meter and its exploitation can not be detected
before the shutdown by the consideration of the con-
trol flow.
4.5 Strongly Controlled Processes
The detection methods above either look at the pro-
cess ID, the smart meter ID or perform conformance
checking. While the check of the smart meter ID is
clearly beneficial, the check of the process ID has a
significant drawback: (shutdown) messages without
a valid process-ID automatically need to be consid-
ered as dangerous. Therefore, the process must be
strongly modeled, controlled and monitored not al-
lowing unattributed messaged, i.e., messages that are
not assigned to a certain process. Otherwise a lot
of false positive alarms would occur. This property
certainly limits the practical application. In order to
work in practice, a way to distinguish messages that
are not assigned to a process due to intrusions from
unassigned messages due to loose process control will
need to be found.
4.6 Limitation of the Analysis
The analysis above focuses on exploitations of intru-
sions that lead to a shutdown. Other attacks are also
possible. As an example, an attack could try to con-
fuse the system by inhibiting the smart meter from
sending the confirmation of the status, send wrong
status or consumption messages. While these attacks
do not lead to a turnoff it could still result in consider-
able damage. For example wrong consumption values
sent in another use case could result to a wrong sys-
tem status which in turn could trigger wrong activities
by the network operator.
5 CONCLUSION AND OUTLOOK
In this paper the potential of process mining for the
detection of attacks on the smart metering shutdown
use case is explored. This is done by systematically
deriving attacks on the modeled shutdown process.
For each attack, its detection using process mining
is analytically explored and modeled by an attack-
defense tree. It could be shown that process mining
has the potential to detect exploitations of intrusions
aiming at an illegitimate shutdown of smart meters.
Based on this analysis several benefits and limitations
of the method are discussed: from the methodical
view, the control-flow oriented analysis should be ac-
companied by considering other views of the process.
From a practical view, the fact that attacks can lead to
unattributed messages is supposed to be problematic
for loosely controlled processes.
Subsequent work will explore ways to combine
the control flow analysis with other views. In addi-
tion to this analytic study, a comparative validation
study with real data will be performed. Besides find-
ing solutions for dealing with unattributed messages,
the method will be compared with other approaches in
terms of detection and false positive rates. This will
be achieved by using an envisaged toolchain which is
suited for an automated large-scale evaluation of the
approaches.
REFERENCES
Accorsi, R. and Stocker, T. (2012). On the exploitation of
process mining for security audits: the conformance
checking case. In Proceedings of the 27th Annual
ACM Symposium on Applied Computing, pages 1709–
1716. ACM.
Berthier, R., Sanders, W. H., and Khurana, H. (2010). In-
trusion Detection for Advanced Metering Infrastruc-
tures: Requirements and Architectural Directions. In
2010 First IEEE International Conference on Smart
Grid Communications, pages 350–355. IEEE.
Bezerra, F. and Wainer, J. (2013). Algorithms for anomaly
detection of traces in logs of process aware informa-
tion systems. Information Systems, 38(1):33–44.
Bezerra, F., Wainer, J., and Van Der Aalst, W. M. P. (2009).
Anomaly Detection using Process Mining. In 10th In-
ternational Workshop, Enterprise, Business-Process
and Information Systems Modeling, volume 29, pages
149–161.
Jalali, H. and Baraani, A. (2012). Process aware host-based
intrusion detection model. International Journal of
Communication Networks and Information Security,
4(2):117–124.
Kordy, B., Kordy, P., Mauw, S., and Schweitzer, P.
(2013). ADTool: Security Analysis with Attack-
Exploration of the Potential of Process Mining for Intrusion Detection in Smart Metering
45
Defense Trees. In International Conference on Quan-
titative Evaluation of Systems.
Kordy, B., Mauw, S., Radomirovi
´
c, S., and Schweitzer, P.
(2012). Attack-Defense Trees. Journal of Logic and
Computation, page exs029.
Kordy, B., Pi
`
etre-cambac
´
ed
`
es, L., and Schweitzer, P.
(2014). DAG-Based Attack and Defense Modeling :
Don’ t Miss the Forest for the Attack Trees. Computer
science review, 13:1–38.
Oesterreichs-Energie (2015).
¨
Osterreich Use-Cases f
¨
ur
das Smart Metering Advanced Meter Communication
System (AMCS).
Roy, A., Kim, D. S., and Trivedi, K. S. (2012). ACT:
Towards unifying the constructs of attack and de-
fense trees. Security and Communication Networks,
5(8):929–943.
Salter, C., Saydjari, O. S., Schneier, B., and Wallner,
J. (1998). Toward A Secure System Engineering
Methodology. In Proceedings of the 1998 Workshop
on New Security Paradigms (NSPW ’98), pages 2–10.
Stocker, T. and Accorsi, R. (2013). SecSy: Security-
aware Synthesis of Process Event Logs. In Proceed-
ings of the 5th International Workshop on Enterprise
Modelling and Information Systems Architectures, St.
Gallen, Switzerland.
Van der Aalst, W. M. (2011). Process Mining: Discov-
ery, Conformance and Enhancement of Business Pro-
cesses. Springer.
Van der Aalst, W. M. and de Medeiros, A. K. A. (2005).
Process mining and security: Detecting anomalous
process executions and checking process confor-
mance. Electronic Notes in Theoretical Computer Sci-
ence, 121:3–21.
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
46